From b48671b59838181400a366745a354709cfae612d Mon Sep 17 00:00:00 2001 From: Dario Date: Thu, 17 Aug 2023 14:02:34 -0300 Subject: [PATCH] Add FidelityFX Super Resolution 2.2 (FSR 2.2.1) support. Introduces support for FSR2 as a new upscaler option available from the project settings. --- doc/classes/RenderingServer.xml | 8 +- doc/classes/Viewport.xml | 8 +- drivers/vulkan/rendering_device_vulkan.cpp | 12 +- drivers/vulkan/vulkan_context.cpp | 23 + drivers/vulkan/vulkan_context.h | 3 + editor/plugins/node_3d_editor_plugin.cpp | 6 +- editor/plugins/node_3d_editor_plugin.h | 1 + scene/main/viewport.cpp | 4 +- scene/main/viewport.h | 2 + servers/rendering/renderer_rd/effects/SCsub | 31 +- .../renderer_rd/effects/debug_effects.cpp | 23 +- .../renderer_rd/effects/debug_effects.h | 8 +- .../rendering/renderer_rd/effects/fsr2.cpp | 890 +++++ servers/rendering/renderer_rd/effects/fsr2.h | 202 ++ servers/rendering/renderer_rd/effects/taa.cpp | 14 - servers/rendering/renderer_rd/effects/taa.h | 1 - .../rendering/renderer_rd/environment/sky.cpp | 11 +- .../rendering/renderer_rd/environment/sky.h | 2 +- .../render_forward_clustered.cpp | 124 +- .../render_forward_clustered.h | 9 +- .../forward_mobile/render_forward_mobile.cpp | 6 +- .../renderer_rd/renderer_scene_render_rd.cpp | 31 +- .../shaders/effects/fsr2_accumulate_pass.glsl | 8 + .../effects/fsr2_autogen_reactive_pass.glsl | 8 + .../fsr2_compute_luminance_pyramid_pass.glsl | 7 + .../shaders/effects/fsr2_depth_clip_pass.glsl | 8 + .../shaders/effects/fsr2_lock_pass.glsl | 7 + .../shaders/effects/fsr2_rcas_pass.glsl | 7 + .../fsr2_reconstruct_previous_depth_pass.glsl | 8 + .../effects/fsr2_tcr_autogen_pass.glsl | 8 + .../shaders/effects/motion_vector_inc.glsl | 6 + .../shaders/effects/motion_vectors.glsl | 19 +- .../scene_forward_clustered.glsl | 2 + .../renderer_rd/shaders/scene_data_inc.glsl | 2 +- .../storage_rd/render_scene_buffers_rd.cpp | 33 +- .../storage_rd/render_scene_buffers_rd.h | 35 +- .../storage_rd/render_scene_data_rd.cpp | 3 +- .../storage_rd/render_scene_data_rd.h | 4 +- servers/rendering/renderer_scene_cull.cpp | 47 +- servers/rendering/renderer_scene_cull.h | 5 +- servers/rendering/renderer_viewport.cpp | 53 +- servers/rendering/renderer_viewport.h | 2 + servers/rendering/rendering_device.h | 18 + servers/rendering/rendering_method.h | 2 +- .../storage/render_scene_buffers.cpp | 2 +- servers/rendering_server.cpp | 4 +- servers/rendering_server.h | 2 + thirdparty/README.md | 13 + thirdparty/amd-fsr2/LICENSE.txt | 21 + thirdparty/amd-fsr2/ffx_assert.cpp | 81 + thirdparty/amd-fsr2/ffx_assert.h | 132 + thirdparty/amd-fsr2/ffx_error.h | 59 + thirdparty/amd-fsr2/ffx_fsr2.cpp | 1373 ++++++++ thirdparty/amd-fsr2/ffx_fsr2.h | 458 +++ thirdparty/amd-fsr2/ffx_fsr2_interface.h | 395 +++ thirdparty/amd-fsr2/ffx_fsr2_maximum_bias.h | 46 + thirdparty/amd-fsr2/ffx_fsr2_private.h | 86 + thirdparty/amd-fsr2/ffx_types.h | 367 ++ thirdparty/amd-fsr2/ffx_util.h | 78 + .../amd-fsr2/patches/godot-changes.patch | Bin 0 -> 21350 bytes .../amd-fsr2/shaders/ffx_common_types.h | 429 +++ thirdparty/amd-fsr2/shaders/ffx_core.h | 52 + thirdparty/amd-fsr2/shaders/ffx_core_cpu.h | 332 ++ thirdparty/amd-fsr2/shaders/ffx_core_glsl.h | 1669 +++++++++ .../amd-fsr2/shaders/ffx_core_gpu_common.h | 2784 +++++++++++++++ .../shaders/ffx_core_gpu_common_half.h | 2978 +++++++++++++++++ thirdparty/amd-fsr2/shaders/ffx_core_hlsl.h | 1502 +++++++++ .../amd-fsr2/shaders/ffx_core_portability.h | 50 + thirdparty/amd-fsr2/shaders/ffx_fsr1.h | 1250 +++++++ .../amd-fsr2/shaders/ffx_fsr2_accumulate.h | 295 ++ .../shaders/ffx_fsr2_accumulate_pass.glsl | 92 + .../ffx_fsr2_autogen_reactive_pass.glsl | 93 + .../shaders/ffx_fsr2_callbacks_glsl.h | 704 ++++ .../shaders/ffx_fsr2_callbacks_hlsl.h | 799 +++++ thirdparty/amd-fsr2/shaders/ffx_fsr2_common.h | 565 ++++ .../ffx_fsr2_compute_luminance_pyramid.h | 189 ++ ...x_fsr2_compute_luminance_pyramid_pass.glsl | 134 + .../amd-fsr2/shaders/ffx_fsr2_depth_clip.h | 258 ++ .../shaders/ffx_fsr2_depth_clip_pass.glsl | 67 + .../amd-fsr2/shaders/ffx_fsr2_force16_begin.h | 1 + .../amd-fsr2/shaders/ffx_fsr2_force16_end.h | 1 + thirdparty/amd-fsr2/shaders/ffx_fsr2_lock.h | 115 + .../amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl | 56 + .../ffx_fsr2_postprocess_lock_status.h | 106 + thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas.h | 67 + .../amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl | 80 + ...ruct_dilated_velocity_and_previous_depth.h | 145 + ..._fsr2_reconstruct_previous_depth_pass.glsl | 65 + .../amd-fsr2/shaders/ffx_fsr2_reproject.h | 136 + .../amd-fsr2/shaders/ffx_fsr2_resources.h | 105 + thirdparty/amd-fsr2/shaders/ffx_fsr2_sample.h | 605 ++++ .../amd-fsr2/shaders/ffx_fsr2_tcr_autogen.h | 250 ++ .../shaders/ffx_fsr2_tcr_autogen_pass.glsl | 122 + .../amd-fsr2/shaders/ffx_fsr2_upsample.h | 194 ++ thirdparty/amd-fsr2/shaders/ffx_spd.h | 936 ++++++ 95 files changed, 21941 insertions(+), 113 deletions(-) create mode 100644 servers/rendering/renderer_rd/effects/fsr2.cpp create mode 100644 servers/rendering/renderer_rd/effects/fsr2.h create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_accumulate_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_autogen_reactive_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_compute_luminance_pyramid_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_depth_clip_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_lock_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_rcas_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_reconstruct_previous_depth_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/fsr2_tcr_autogen_pass.glsl create mode 100644 servers/rendering/renderer_rd/shaders/effects/motion_vector_inc.glsl create mode 100644 thirdparty/amd-fsr2/LICENSE.txt create mode 100644 thirdparty/amd-fsr2/ffx_assert.cpp create mode 100644 thirdparty/amd-fsr2/ffx_assert.h create mode 100644 thirdparty/amd-fsr2/ffx_error.h create mode 100644 thirdparty/amd-fsr2/ffx_fsr2.cpp create mode 100644 thirdparty/amd-fsr2/ffx_fsr2.h create mode 100644 thirdparty/amd-fsr2/ffx_fsr2_interface.h create mode 100644 thirdparty/amd-fsr2/ffx_fsr2_maximum_bias.h create mode 100644 thirdparty/amd-fsr2/ffx_fsr2_private.h create mode 100644 thirdparty/amd-fsr2/ffx_types.h create mode 100644 thirdparty/amd-fsr2/ffx_util.h create mode 100644 thirdparty/amd-fsr2/patches/godot-changes.patch create mode 100644 thirdparty/amd-fsr2/shaders/ffx_common_types.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core_cpu.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core_glsl.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core_gpu_common.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core_gpu_common_half.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core_hlsl.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_core_portability.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr1.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_glsl.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_hlsl.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_common.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_begin.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_end.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_lock.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_postprocess_lock_status.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_reproject.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_resources.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_sample.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl create mode 100644 thirdparty/amd-fsr2/shaders/ffx_fsr2_upsample.h create mode 100644 thirdparty/amd-fsr2/shaders/ffx_spd.h diff --git a/doc/classes/RenderingServer.xml b/doc/classes/RenderingServer.xml index 7331a7fe32c3..9f64cbf6103a 100644 --- a/doc/classes/RenderingServer.xml +++ b/doc/classes/RenderingServer.xml @@ -4507,7 +4507,10 @@ Use AMD FidelityFX Super Resolution 1.0 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member Viewport.scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] disables scaling. - + + Use AMD FidelityFX Super Resolution 2.2 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member Viewport.scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR2. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] will use FSR2 at native resolution as a TAA solution. + + Represents the size of the [enum ViewportScaling3DMode] enum. @@ -4708,6 +4711,9 @@ Draws the motion vectors buffer. This is used by temporal antialiasing to correct for motion that occurs during gameplay. + + Internal buffer is drawn instead of regular scene so you can see the per-pixel output that will be used by post-processing effects. + Variable rate shading is disabled. diff --git a/doc/classes/Viewport.xml b/doc/classes/Viewport.xml index 43640916f442..94765fdcbe56 100644 --- a/doc/classes/Viewport.xml +++ b/doc/classes/Viewport.xml @@ -442,7 +442,10 @@ Use AMD FidelityFX Super Resolution 1.0 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] disables scaling. - + + Use AMD FidelityFX Super Resolution 2.2 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member Viewport.scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR2. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] will use FSR2 at native resolution as a TAA solution. + + Represents the size of the [enum Scaling3DMode] enum. @@ -552,6 +555,9 @@ + + Draws the internal resolution buffer of the scene before post-processing is applied. + The texture filter reads from the nearest pixel only. The simplest and fastest method of filtering, but the texture will look pixelized. diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp index 803a109863be..c0bba079e6ab 100644 --- a/drivers/vulkan/rendering_device_vulkan.cpp +++ b/drivers/vulkan/rendering_device_vulkan.cpp @@ -5750,8 +5750,8 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector &p_uniforms, Buffer *buffer = uniform_buffer_owner.get_or_null(uniform.get_id(0)); ERR_FAIL_COND_V_MSG(!buffer, RID(), "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") is invalid."); - ERR_FAIL_COND_V_MSG(buffer->size != (uint32_t)set_uniform.length, RID(), - "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") size (" + itos(buffer->size) + " does not match size of shader uniform: (" + itos(set_uniform.length) + ")."); + ERR_FAIL_COND_V_MSG(buffer->size < (uint32_t)set_uniform.length, RID(), + "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") size (" + itos(buffer->size) + " is smaller than size of shader uniform: (" + itos(set_uniform.length) + ")."); write.dstArrayElement = 0; write.descriptorCount = 1; @@ -9550,6 +9550,14 @@ uint64_t RenderingDeviceVulkan::limit_get(Limit p_limit) const { VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); return subgroup_capabilities.size; } + case LIMIT_SUBGROUP_MIN_SIZE: { + VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); + return subgroup_capabilities.min_size; + } + case LIMIT_SUBGROUP_MAX_SIZE: { + VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); + return subgroup_capabilities.max_size; + } case LIMIT_SUBGROUP_IN_SHADERS: { VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities(); return subgroup_capabilities.supported_stages_flags_rd(); diff --git a/drivers/vulkan/vulkan_context.cpp b/drivers/vulkan/vulkan_context.cpp index b55eb586de8b..b99eb93680a1 100644 --- a/drivers/vulkan/vulkan_context.cpp +++ b/drivers/vulkan/vulkan_context.cpp @@ -503,6 +503,7 @@ Error VulkanContext::_initialize_device_extensions() { register_requested_device_extension(VK_KHR_16BIT_STORAGE_EXTENSION_NAME, false); register_requested_device_extension(VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME, false); register_requested_device_extension(VK_KHR_MAINTENANCE_2_EXTENSION_NAME, false); + register_requested_device_extension(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false); if (Engine::get_singleton()->is_generate_spirv_debug_info_enabled()) { register_requested_device_extension(VK_KHR_SHADER_NON_SEMANTIC_INFO_EXTENSION_NAME, true); @@ -738,9 +739,12 @@ Error VulkanContext::_check_capabilities() { multiview_capabilities.max_view_count = 0; multiview_capabilities.max_instance_count = 0; subgroup_capabilities.size = 0; + subgroup_capabilities.min_size = 0; + subgroup_capabilities.max_size = 0; subgroup_capabilities.supportedStages = 0; subgroup_capabilities.supportedOperations = 0; subgroup_capabilities.quadOperationsInAllStages = false; + subgroup_capabilities.size_control_is_supported = false; shader_capabilities.shader_float16_is_supported = false; shader_capabilities.shader_int8_is_supported = false; storage_buffer_capabilities.storage_buffer_16_bit_access_is_supported = false; @@ -871,6 +875,7 @@ Error VulkanContext::_check_capabilities() { VkPhysicalDeviceFragmentShadingRatePropertiesKHR vrsProperties{}; VkPhysicalDeviceMultiviewProperties multiviewProperties{}; VkPhysicalDeviceSubgroupProperties subgroupProperties{}; + VkPhysicalDeviceSubgroupSizeControlProperties subgroupSizeControlProperties = {}; VkPhysicalDeviceProperties2 physicalDeviceProperties{}; void *nextptr = nullptr; @@ -879,6 +884,15 @@ Error VulkanContext::_check_capabilities() { subgroupProperties.pNext = nextptr; nextptr = &subgroupProperties; + + subgroup_capabilities.size_control_is_supported = is_device_extension_enabled(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME); + + if (subgroup_capabilities.size_control_is_supported) { + subgroupSizeControlProperties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES; + subgroupSizeControlProperties.pNext = nextptr; + + nextptr = &subgroupSizeControlProperties; + } } if (multiview_capabilities.is_supported) { @@ -901,6 +915,8 @@ Error VulkanContext::_check_capabilities() { device_properties_func(gpu, &physicalDeviceProperties); subgroup_capabilities.size = subgroupProperties.subgroupSize; + subgroup_capabilities.min_size = subgroupProperties.subgroupSize; + subgroup_capabilities.max_size = subgroupProperties.subgroupSize; subgroup_capabilities.supportedStages = subgroupProperties.supportedStages; subgroup_capabilities.supportedOperations = subgroupProperties.supportedOperations; // Note: quadOperationsInAllStages will be true if: @@ -908,6 +924,11 @@ Error VulkanContext::_check_capabilities() { // - supportedOperations has VK_SUBGROUP_FEATURE_QUAD_BIT. subgroup_capabilities.quadOperationsInAllStages = subgroupProperties.quadOperationsInAllStages; + if (subgroup_capabilities.size_control_is_supported && (subgroupSizeControlProperties.requiredSubgroupSizeStages & VK_SHADER_STAGE_COMPUTE_BIT)) { + subgroup_capabilities.min_size = subgroupSizeControlProperties.minSubgroupSize; + subgroup_capabilities.max_size = subgroupSizeControlProperties.maxSubgroupSize; + } + if (vrs_capabilities.pipeline_vrs_supported || vrs_capabilities.primitive_vrs_supported || vrs_capabilities.attachment_vrs_supported) { print_verbose("- Vulkan Variable Rate Shading supported:"); if (vrs_capabilities.pipeline_vrs_supported) { @@ -947,6 +968,8 @@ Error VulkanContext::_check_capabilities() { print_verbose("- Vulkan subgroup:"); print_verbose(" size: " + itos(subgroup_capabilities.size)); + print_verbose(" min size: " + itos(subgroup_capabilities.min_size)); + print_verbose(" max size: " + itos(subgroup_capabilities.max_size)); print_verbose(" stages: " + subgroup_capabilities.supported_stages_desc()); print_verbose(" supported ops: " + subgroup_capabilities.supported_operations_desc()); if (subgroup_capabilities.quadOperationsInAllStages) { diff --git a/drivers/vulkan/vulkan_context.h b/drivers/vulkan/vulkan_context.h index 9fd2c40a06d5..af3f9777750f 100644 --- a/drivers/vulkan/vulkan_context.h +++ b/drivers/vulkan/vulkan_context.h @@ -52,9 +52,12 @@ class VulkanContext { public: struct SubgroupCapabilities { uint32_t size; + uint32_t min_size; + uint32_t max_size; VkShaderStageFlags supportedStages; VkSubgroupFeatureFlags supportedOperations; VkBool32 quadOperationsInAllStages; + bool size_control_is_supported; uint32_t supported_stages_flags_rd() const; String supported_stages_desc() const; diff --git a/editor/plugins/node_3d_editor_plugin.cpp b/editor/plugins/node_3d_editor_plugin.cpp index 4284fd8ae21d..70e7f679f94b 100644 --- a/editor/plugins/node_3d_editor_plugin.cpp +++ b/editor/plugins/node_3d_editor_plugin.cpp @@ -3475,7 +3475,8 @@ void Node3DEditorViewport::_menu_option(int p_option) { case VIEW_DISPLAY_DEBUG_CLUSTER_DECALS: case VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES: case VIEW_DISPLAY_DEBUG_OCCLUDERS: - case VIEW_DISPLAY_MOTION_VECTORS: { + case VIEW_DISPLAY_MOTION_VECTORS: + case VIEW_DISPLAY_INTERNAL_BUFFER: { static const int display_options[] = { VIEW_DISPLAY_NORMAL, VIEW_DISPLAY_WIREFRAME, @@ -3503,6 +3504,7 @@ void Node3DEditorViewport::_menu_option(int p_option) { VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES, VIEW_DISPLAY_DEBUG_OCCLUDERS, VIEW_DISPLAY_MOTION_VECTORS, + VIEW_DISPLAY_INTERNAL_BUFFER, VIEW_MAX }; static const Viewport::DebugDraw debug_draw_modes[] = { @@ -3532,6 +3534,7 @@ void Node3DEditorViewport::_menu_option(int p_option) { Viewport::DEBUG_DRAW_CLUSTER_REFLECTION_PROBES, Viewport::DEBUG_DRAW_OCCLUDERS, Viewport::DEBUG_DRAW_MOTION_VECTORS, + Viewport::DEBUG_DRAW_INTERNAL_BUFFER, }; for (int idx = 0; display_options[idx] != VIEW_MAX; idx++) { @@ -5115,6 +5118,7 @@ Node3DEditorViewport::Node3DEditorViewport(Node3DEditor *p_spatial_editor, int p display_submenu->add_radio_check_item(TTR("ReflectionProbe Cluster"), VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES); display_submenu->add_radio_check_item(TTR("Occlusion Culling Buffer"), VIEW_DISPLAY_DEBUG_OCCLUDERS); display_submenu->add_radio_check_item(TTR("Motion Vectors"), VIEW_DISPLAY_MOTION_VECTORS); + display_submenu->add_radio_check_item(TTR("Internal Buffer"), VIEW_DISPLAY_INTERNAL_BUFFER); display_submenu->set_name("display_advanced"); view_menu->get_popup()->add_submenu_item(TTR("Display Advanced..."), "display_advanced", VIEW_DISPLAY_ADVANCED); diff --git a/editor/plugins/node_3d_editor_plugin.h b/editor/plugins/node_3d_editor_plugin.h index e58e224ff4d4..851138961e47 100644 --- a/editor/plugins/node_3d_editor_plugin.h +++ b/editor/plugins/node_3d_editor_plugin.h @@ -155,6 +155,7 @@ class Node3DEditorViewport : public Control { VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES, VIEW_DISPLAY_DEBUG_OCCLUDERS, VIEW_DISPLAY_MOTION_VECTORS, + VIEW_DISPLAY_INTERNAL_BUFFER, VIEW_DISPLAY_MAX, // > Keep in sync with menu. diff --git a/scene/main/viewport.cpp b/scene/main/viewport.cpp index 5f8f5225c91c..971c20f172df 100644 --- a/scene/main/viewport.cpp +++ b/scene/main/viewport.cpp @@ -4502,7 +4502,7 @@ void Viewport::_bind_methods() { #ifndef _3D_DISABLED ADD_GROUP("Scaling 3D", ""); - ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast)"), "set_scaling_3d_mode", "get_scaling_3d_mode"); + ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast),FSR 2.2 (Slow)"), "set_scaling_3d_mode", "get_scaling_3d_mode"); ADD_PROPERTY(PropertyInfo(Variant::FLOAT, "scaling_3d_scale", PROPERTY_HINT_RANGE, "0.25,2.0,0.01"), "set_scaling_3d_scale", "get_scaling_3d_scale"); ADD_PROPERTY(PropertyInfo(Variant::FLOAT, "texture_mipmap_bias", PROPERTY_HINT_RANGE, "-2,2,0.001"), "set_texture_mipmap_bias", "get_texture_mipmap_bias"); ADD_PROPERTY(PropertyInfo(Variant::FLOAT, "fsr_sharpness", PROPERTY_HINT_RANGE, "0,2,0.1"), "set_fsr_sharpness", "get_fsr_sharpness"); @@ -4553,6 +4553,7 @@ void Viewport::_bind_methods() { BIND_ENUM_CONSTANT(SCALING_3D_MODE_BILINEAR); BIND_ENUM_CONSTANT(SCALING_3D_MODE_FSR); + BIND_ENUM_CONSTANT(SCALING_3D_MODE_FSR2); BIND_ENUM_CONSTANT(SCALING_3D_MODE_MAX); BIND_ENUM_CONSTANT(MSAA_DISABLED); @@ -4600,6 +4601,7 @@ void Viewport::_bind_methods() { BIND_ENUM_CONSTANT(DEBUG_DRAW_CLUSTER_REFLECTION_PROBES); BIND_ENUM_CONSTANT(DEBUG_DRAW_OCCLUDERS) BIND_ENUM_CONSTANT(DEBUG_DRAW_MOTION_VECTORS) + BIND_ENUM_CONSTANT(DEBUG_DRAW_INTERNAL_BUFFER); BIND_ENUM_CONSTANT(DEFAULT_CANVAS_ITEM_TEXTURE_FILTER_NEAREST); BIND_ENUM_CONSTANT(DEFAULT_CANVAS_ITEM_TEXTURE_FILTER_LINEAR); diff --git a/scene/main/viewport.h b/scene/main/viewport.h index 68084fc35f2f..a32077a489e8 100644 --- a/scene/main/viewport.h +++ b/scene/main/viewport.h @@ -98,6 +98,7 @@ class Viewport : public Node { enum Scaling3DMode { SCALING_3D_MODE_BILINEAR, SCALING_3D_MODE_FSR, + SCALING_3D_MODE_FSR2, SCALING_3D_MODE_MAX }; @@ -167,6 +168,7 @@ class Viewport : public Node { DEBUG_DRAW_CLUSTER_REFLECTION_PROBES, DEBUG_DRAW_OCCLUDERS, DEBUG_DRAW_MOTION_VECTORS, + DEBUG_DRAW_INTERNAL_BUFFER, }; enum DefaultCanvasItemTextureFilter { diff --git a/servers/rendering/renderer_rd/effects/SCsub b/servers/rendering/renderer_rd/effects/SCsub index 86681f9c74d3..8e13715447cc 100644 --- a/servers/rendering/renderer_rd/effects/SCsub +++ b/servers/rendering/renderer_rd/effects/SCsub @@ -2,4 +2,33 @@ Import("env") -env.add_source_files(env.servers_sources, "*.cpp") +env_effects = env.Clone() + +# Thirdparty source files + +thirdparty_obj = [] + +thirdparty_dir = "#thirdparty/amd-fsr2/" +thirdparty_sources = ["ffx_assert.cpp", "ffx_fsr2.cpp"] +thirdparty_sources = [thirdparty_dir + file for file in thirdparty_sources] + +env_effects.Prepend(CPPPATH=[thirdparty_dir]) + +# This flag doesn't actually control anything GCC specific in FSR2. It determines +# if symbols should be exported, which is not required for Godot. +env_effects.Append(CPPDEFINES=["FFX_GCC"]) + +env_thirdparty = env_effects.Clone() +env_thirdparty.disable_warnings() +env_thirdparty.add_source_files(thirdparty_obj, thirdparty_sources) +env.servers_sources += thirdparty_obj + +# Godot source files + +module_obj = [] + +env_effects.add_source_files(module_obj, "*.cpp") +env.servers_sources += module_obj + +# Needed to force rebuilding the module files when the thirdparty library is updated. +env.Depends(module_obj, thirdparty_obj) diff --git a/servers/rendering/renderer_rd/effects/debug_effects.cpp b/servers/rendering/renderer_rd/effects/debug_effects.cpp index 3d26a9a8df79..589e9e46f659 100644 --- a/servers/rendering/renderer_rd/effects/debug_effects.cpp +++ b/servers/rendering/renderer_rd/effects/debug_effects.cpp @@ -340,25 +340,38 @@ void DebugEffects::draw_shadow_frustum(RID p_light, const Projection &p_cam_proj } } -void DebugEffects::draw_motion_vectors(RID p_velocity, RID p_dest_fb, Size2i p_velocity_size) { +void DebugEffects::draw_motion_vectors(RID p_velocity, RID p_depth, RID p_dest_fb, const Projection &p_current_projection, const Transform3D &p_current_transform, const Projection &p_previous_projection, const Transform3D &p_previous_transform, Size2i p_resolution) { MaterialStorage *material_storage = MaterialStorage::get_singleton(); ERR_FAIL_NULL(material_storage); UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton(); ERR_FAIL_NULL(uniform_set_cache); - RID default_sampler = material_storage->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_LINEAR, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED); + RID default_sampler = material_storage->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_NEAREST, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED); RD::Uniform u_source_velocity(RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE, 0, Vector({ default_sampler, p_velocity })); + RD::Uniform u_source_depth(RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE, 1, Vector({ default_sampler, p_depth })); RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_dest_fb, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_DROP, RD::FINAL_ACTION_DISCARD); RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, motion_vectors.pipeline.get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_dest_fb), false, RD::get_singleton()->draw_list_get_current_pass())); - motion_vectors.push_constant.velocity_resolution[0] = p_velocity_size.width; - motion_vectors.push_constant.velocity_resolution[1] = p_velocity_size.height; + Projection reprojection = p_previous_projection.flipped_y() * p_previous_transform.affine_inverse() * p_current_transform * p_current_projection.flipped_y().inverse(); + RendererRD::MaterialStorage::store_camera(reprojection, motion_vectors.push_constant.reprojection_matrix); + + motion_vectors.push_constant.resolution[0] = p_resolution.width; + motion_vectors.push_constant.resolution[1] = p_resolution.height; + motion_vectors.push_constant.force_derive_from_depth = false; RID shader = motion_vectors.shader.version_get_shader(motion_vectors.shader_version, 0); - RD::get_singleton()->draw_list_bind_uniform_set(draw_list, uniform_set_cache->get_cache(shader, 0, u_source_velocity), 0); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, uniform_set_cache->get_cache(shader, 0, u_source_velocity, u_source_depth), 0); + RD::get_singleton()->draw_list_set_push_constant(draw_list, &motion_vectors.push_constant, sizeof(MotionVectorsPushConstant)); + RD::get_singleton()->draw_list_draw(draw_list, false, 1u, 3u); + +#ifdef DRAW_DERIVATION_FROM_DEPTH_ON_TOP + motion_vectors.push_constant.force_derive_from_depth = true; + RD::get_singleton()->draw_list_set_push_constant(draw_list, &motion_vectors.push_constant, sizeof(MotionVectorsPushConstant)); RD::get_singleton()->draw_list_draw(draw_list, false, 1u, 3u); +#endif + RD::get_singleton()->draw_list_end(); } diff --git a/servers/rendering/renderer_rd/effects/debug_effects.h b/servers/rendering/renderer_rd/effects/debug_effects.h index ae32d9491209..b813d577e4c0 100644 --- a/servers/rendering/renderer_rd/effects/debug_effects.h +++ b/servers/rendering/renderer_rd/effects/debug_effects.h @@ -72,8 +72,10 @@ class DebugEffects { } shadow_frustum; struct MotionVectorsPushConstant { - float velocity_resolution[2]; - float pad[2]; + float reprojection_matrix[16]; + float resolution[2]; + uint32_t force_derive_from_depth; + float pad; }; struct { @@ -91,7 +93,7 @@ class DebugEffects { ~DebugEffects(); void draw_shadow_frustum(RID p_light, const Projection &p_cam_projection, const Transform3D &p_cam_transform, RID p_dest_fb, const Rect2 p_rect); - void draw_motion_vectors(RID p_velocity, RID p_dest_fb, Size2i p_velocity_size); + void draw_motion_vectors(RID p_velocity, RID p_depth, RID p_dest_fb, const Projection &p_current_projection, const Transform3D &p_current_transform, const Projection &p_previous_projection, const Transform3D &p_previous_transform, Size2i p_resolution); }; } // namespace RendererRD diff --git a/servers/rendering/renderer_rd/effects/fsr2.cpp b/servers/rendering/renderer_rd/effects/fsr2.cpp new file mode 100644 index 000000000000..6129393589ae --- /dev/null +++ b/servers/rendering/renderer_rd/effects/fsr2.cpp @@ -0,0 +1,890 @@ +/**************************************************************************/ +/* fsr2.cpp */ +/**************************************************************************/ +/* This file is part of: */ +/* GODOT ENGINE */ +/* https://godotengine.org */ +/**************************************************************************/ +/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */ +/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */ +/* */ +/* Permission is hereby granted, free of charge, to any person obtaining */ +/* a copy of this software and associated documentation files (the */ +/* "Software"), to deal in the Software without restriction, including */ +/* without limitation the rights to use, copy, modify, merge, publish, */ +/* distribute, sublicense, and/or sell copies of the Software, and to */ +/* permit persons to whom the Software is furnished to do so, subject to */ +/* the following conditions: */ +/* */ +/* The above copyright notice and this permission notice shall be */ +/* included in all copies or substantial portions of the Software. */ +/* */ +/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */ +/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */ +/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */ +/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */ +/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */ +/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */ +/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ +/**************************************************************************/ + +#include "fsr2.h" + +#include "../storage_rd/material_storage.h" +#include "../uniform_set_cache_rd.h" + +using namespace RendererRD; + +#ifndef _MSC_VER +#include +#define wcscpy_s wcscpy +#endif + +RD::TextureType ffx_resource_type_to_rd_texture_type(FfxResourceType t) { + switch (t) { + case FFX_RESOURCE_TYPE_TEXTURE1D: + return RD::TEXTURE_TYPE_1D; + case FFX_RESOURCE_TYPE_TEXTURE2D: + return RD::TEXTURE_TYPE_2D; + case FFX_RESOURCE_TYPE_TEXTURE3D: + return RD::TEXTURE_TYPE_3D; + default: + return RD::TEXTURE_TYPE_MAX; + } +} + +FfxResourceType rd_texture_type_to_ffx_resource_type(RD::TextureType t) { + switch (t) { + case RD::TEXTURE_TYPE_1D: + return FFX_RESOURCE_TYPE_TEXTURE1D; + case RD::TEXTURE_TYPE_2D: + return FFX_RESOURCE_TYPE_TEXTURE2D; + case RD::TEXTURE_TYPE_3D: + return FFX_RESOURCE_TYPE_TEXTURE3D; + default: + return FFX_RESOURCE_TYPE_BUFFER; + } +} + +RD::DataFormat ffx_surface_format_to_rd_format(FfxSurfaceFormat f) { + switch (f) { + case FFX_SURFACE_FORMAT_R32G32B32A32_TYPELESS: + return RD::DATA_FORMAT_R32G32B32A32_SFLOAT; + case FFX_SURFACE_FORMAT_R32G32B32A32_FLOAT: + return RD::DATA_FORMAT_R32G32B32A32_SFLOAT; + case FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT: + return RD::DATA_FORMAT_R16G16B16A16_SFLOAT; + case FFX_SURFACE_FORMAT_R16G16B16A16_UNORM: + return RD::DATA_FORMAT_R16G16B16A16_UNORM; + case FFX_SURFACE_FORMAT_R32G32_FLOAT: + return RD::DATA_FORMAT_R32G32_SFLOAT; + case FFX_SURFACE_FORMAT_R32_UINT: + return RD::DATA_FORMAT_R32_UINT; + case FFX_SURFACE_FORMAT_R8G8B8A8_TYPELESS: + return RD::DATA_FORMAT_R8G8B8A8_UNORM; + case FFX_SURFACE_FORMAT_R8G8B8A8_UNORM: + return RD::DATA_FORMAT_R8G8B8A8_UNORM; + case FFX_SURFACE_FORMAT_R11G11B10_FLOAT: + return RD::DATA_FORMAT_B10G11R11_UFLOAT_PACK32; + case FFX_SURFACE_FORMAT_R16G16_FLOAT: + return RD::DATA_FORMAT_R16G16_SFLOAT; + case FFX_SURFACE_FORMAT_R16G16_UINT: + return RD::DATA_FORMAT_R16G16_UINT; + case FFX_SURFACE_FORMAT_R16_FLOAT: + return RD::DATA_FORMAT_R16_SFLOAT; + case FFX_SURFACE_FORMAT_R16_UINT: + return RD::DATA_FORMAT_R16_UINT; + case FFX_SURFACE_FORMAT_R16_UNORM: + return RD::DATA_FORMAT_R16_UNORM; + case FFX_SURFACE_FORMAT_R16_SNORM: + return RD::DATA_FORMAT_R16_SNORM; + case FFX_SURFACE_FORMAT_R8_UNORM: + return RD::DATA_FORMAT_R8_UNORM; + case FFX_SURFACE_FORMAT_R8_UINT: + return RD::DATA_FORMAT_R8_UINT; + case FFX_SURFACE_FORMAT_R8G8_UNORM: + return RD::DATA_FORMAT_R8G8_UNORM; + case FFX_SURFACE_FORMAT_R32_FLOAT: + return RD::DATA_FORMAT_R32_SFLOAT; + default: + return RD::DATA_FORMAT_MAX; + } +} + +FfxSurfaceFormat rd_format_to_ffx_surface_format(RD::DataFormat f) { + switch (f) { + case RD::DATA_FORMAT_R32G32B32A32_SFLOAT: + return FFX_SURFACE_FORMAT_R32G32B32A32_FLOAT; + case RD::DATA_FORMAT_R16G16B16A16_SFLOAT: + return FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT; + case RD::DATA_FORMAT_R16G16B16A16_UNORM: + return FFX_SURFACE_FORMAT_R16G16B16A16_UNORM; + case RD::DATA_FORMAT_R32G32_SFLOAT: + return FFX_SURFACE_FORMAT_R32G32_FLOAT; + case RD::DATA_FORMAT_R32_UINT: + return FFX_SURFACE_FORMAT_R32_UINT; + case RD::DATA_FORMAT_R8G8B8A8_UNORM: + return FFX_SURFACE_FORMAT_R8G8B8A8_UNORM; + case RD::DATA_FORMAT_B10G11R11_UFLOAT_PACK32: + return FFX_SURFACE_FORMAT_R11G11B10_FLOAT; + case RD::DATA_FORMAT_R16G16_SFLOAT: + return FFX_SURFACE_FORMAT_R16G16_FLOAT; + case RD::DATA_FORMAT_R16G16_UINT: + return FFX_SURFACE_FORMAT_R16G16_UINT; + case RD::DATA_FORMAT_R16_SFLOAT: + return FFX_SURFACE_FORMAT_R16_FLOAT; + case RD::DATA_FORMAT_R16_UINT: + return FFX_SURFACE_FORMAT_R16_UINT; + case RD::DATA_FORMAT_R16_UNORM: + return FFX_SURFACE_FORMAT_R16_UNORM; + case RD::DATA_FORMAT_R16_SNORM: + return FFX_SURFACE_FORMAT_R16_SNORM; + case RD::DATA_FORMAT_R8_UNORM: + return FFX_SURFACE_FORMAT_R8_UNORM; + case RD::DATA_FORMAT_R8_UINT: + return FFX_SURFACE_FORMAT_R8_UINT; + case RD::DATA_FORMAT_R8G8_UNORM: + return FFX_SURFACE_FORMAT_R8G8_UNORM; + case RD::DATA_FORMAT_R32_SFLOAT: + return FFX_SURFACE_FORMAT_R32_FLOAT; + default: + return FFX_SURFACE_FORMAT_UNKNOWN; + } +} + +uint32_t ffx_usage_to_rd_usage_flags(uint32_t flags) { + uint32_t ret = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT; + + if (flags & FFX_RESOURCE_USAGE_RENDERTARGET) { + ret |= RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT; + } + + if (flags & FFX_RESOURCE_USAGE_UAV) { + ret |= RD::TEXTURE_USAGE_STORAGE_BIT; + ret |= RD::TEXTURE_USAGE_CAN_COPY_FROM_BIT; + ret |= RD::TEXTURE_USAGE_CAN_COPY_TO_BIT; + } + + return ret; +} + +FfxErrorCode create_backend_context_rd(FfxFsr2Interface *backend_interface, FfxDevice device) { + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + + // Store pointer to the device common to all contexts. + scratch.device = device; + + // Create a ring buffer of uniform buffers. + // FIXME: This could be optimized to be a single memory block if it was possible for RD to create views into a particular memory range of a UBO. + for (uint32_t i = 0; i < FSR2_UBO_RING_BUFFER_SIZE; i++) { + scratch.ubo_ring_buffer[i] = RD::get_singleton()->uniform_buffer_create(FFX_MAX_CONST_SIZE * sizeof(uint32_t)); + ERR_FAIL_COND_V(scratch.ubo_ring_buffer[i].is_null(), FFX_ERROR_BACKEND_API_ERROR); + } + + return FFX_OK; +} + +FfxErrorCode get_device_capabilities_rd(FfxFsr2Interface *backend_interface, FfxDeviceCapabilities *out_device_capabilities, FfxDevice device) { + FSR2Effect::Device &effect_device = *reinterpret_cast(device); + + *out_device_capabilities = effect_device.capabilities; + + return FFX_OK; +} + +FfxErrorCode destroy_backend_context_rd(FfxFsr2Interface *backend_interface) { + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + + for (uint32_t i = 0; i < FSR2_UBO_RING_BUFFER_SIZE; i++) { + RD::get_singleton()->free(scratch.ubo_ring_buffer[i]); + } + + return FFX_OK; +} + +FfxErrorCode create_resource_rd(FfxFsr2Interface *backend_interface, const FfxCreateResourceDescription *create_resource_description, FfxResourceInternal *out_resource) { + // FSR2's base implementation won't issue a call to create a heap type that isn't just default on its own, + // so we can safely ignore it as RD does not expose this concept. + ERR_FAIL_COND_V(create_resource_description->heapType != FFX_HEAP_TYPE_DEFAULT, FFX_ERROR_INVALID_ARGUMENT); + + RenderingDevice *rd = RD::get_singleton(); + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + FfxResourceDescription res_desc = create_resource_description->resourceDescription; + + // FSR2's base implementation never requests buffer creation. + ERR_FAIL_COND_V(res_desc.type != FFX_RESOURCE_TYPE_TEXTURE1D && res_desc.type != FFX_RESOURCE_TYPE_TEXTURE2D && res_desc.type != FFX_RESOURCE_TYPE_TEXTURE3D, FFX_ERROR_INVALID_ARGUMENT); + + if (res_desc.mipCount == 0) { + // Mipmap count must be derived from the resource's dimensions. + res_desc.mipCount = uint32_t(1 + floor(log2(MAX(MAX(res_desc.width, res_desc.height), res_desc.depth)))); + } + + Vector initial_data; + if (create_resource_description->initDataSize) { + PackedByteArray byte_array; + byte_array.resize(create_resource_description->initDataSize); + memcpy(byte_array.ptrw(), create_resource_description->initData, create_resource_description->initDataSize); + initial_data.push_back(byte_array); + } + + RD::TextureFormat texture_format; + texture_format.texture_type = ffx_resource_type_to_rd_texture_type(res_desc.type); + texture_format.format = ffx_surface_format_to_rd_format(res_desc.format); + texture_format.usage_bits = ffx_usage_to_rd_usage_flags(create_resource_description->usage); + texture_format.width = res_desc.width; + texture_format.height = res_desc.height; + texture_format.depth = res_desc.depth; + texture_format.mipmaps = res_desc.mipCount; + + RID texture = rd->texture_create(texture_format, RD::TextureView(), initial_data); + ERR_FAIL_COND_V(texture.is_null(), FFX_ERROR_BACKEND_API_ERROR); + + rd->set_resource_name(texture, String(create_resource_description->name)); + + // Add the resource to the storage and use the internal index to reference it. + out_resource->internalIndex = scratch.resources.add(texture, false, create_resource_description->id, res_desc); + + return FFX_OK; +} + +FfxErrorCode register_resource_rd(FfxFsr2Interface *backend_interface, const FfxResource *in_resource, FfxResourceInternal *out_resource) { + if (in_resource->resource == nullptr) { + // Null resource case. + out_resource->internalIndex = -1; + return FFX_OK; + } + + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + const RID &rid = *reinterpret_cast(in_resource->resource); + ERR_FAIL_COND_V(rid.is_null(), FFX_ERROR_INVALID_ARGUMENT); + + // Add the resource to the storage and use the internal index to reference it. + out_resource->internalIndex = scratch.resources.add(rid, true, FSR2Context::RESOURCE_ID_DYNAMIC, in_resource->description); + + return FFX_OK; +} + +FfxErrorCode unregister_resources_rd(FfxFsr2Interface *backend_interface) { + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + LocalVector dynamic_list_copy = scratch.resources.dynamic_list; + for (uint32_t i : dynamic_list_copy) { + scratch.resources.remove(i); + } + + return FFX_OK; +} + +FfxResourceDescription get_resource_description_rd(FfxFsr2Interface *backend_interface, FfxResourceInternal resource) { + if (resource.internalIndex != -1) { + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + return scratch.resources.descriptions[resource.internalIndex]; + } else { + return {}; + } +} + +FfxErrorCode destroy_resource_rd(FfxFsr2Interface *backend_interface, FfxResourceInternal resource) { + if (resource.internalIndex != -1) { + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + if (scratch.resources.rids[resource.internalIndex].is_valid()) { + RD::get_singleton()->free(scratch.resources.rids[resource.internalIndex]); + scratch.resources.remove(resource.internalIndex); + } + } + + return FFX_OK; +} + +FfxErrorCode create_pipeline_rd(FfxFsr2Interface *backend_interface, FfxFsr2Pass pass, const FfxPipelineDescription *pipeline_description, FfxPipelineState *out_pipeline) { + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + FSR2Effect::Device &device = *reinterpret_cast(scratch.device); + FSR2Effect::Pass &effect_pass = device.passes[pass]; + + if (effect_pass.pipeline.pipeline_rid.is_null()) { + // Create pipeline for the device if it hasn't been created yet. + effect_pass.root_signature.shader_rid = effect_pass.shader->version_get_shader(effect_pass.shader_version, effect_pass.shader_variant); + ERR_FAIL_COND_V(effect_pass.root_signature.shader_rid.is_null(), FFX_ERROR_BACKEND_API_ERROR); + + effect_pass.pipeline.pipeline_rid = RD::get_singleton()->compute_pipeline_create(effect_pass.root_signature.shader_rid); + ERR_FAIL_COND_V(effect_pass.pipeline.pipeline_rid.is_null(), FFX_ERROR_BACKEND_API_ERROR); + } + + // While this is not their intended use, we use the pipeline and root signature pointers to store the + // RIDs to the pipeline and shader that RD needs for the compute pipeline. + out_pipeline->pipeline = reinterpret_cast(&effect_pass.pipeline); + out_pipeline->rootSignature = reinterpret_cast(&effect_pass.root_signature); + + out_pipeline->srvCount = effect_pass.sampled_bindings.size(); + ERR_FAIL_COND_V(out_pipeline->srvCount > FFX_MAX_NUM_SRVS, FFX_ERROR_OUT_OF_RANGE); + memcpy(out_pipeline->srvResourceBindings, effect_pass.sampled_bindings.ptr(), sizeof(FfxResourceBinding) * out_pipeline->srvCount); + + out_pipeline->uavCount = effect_pass.storage_bindings.size(); + ERR_FAIL_COND_V(out_pipeline->uavCount > FFX_MAX_NUM_UAVS, FFX_ERROR_OUT_OF_RANGE); + memcpy(out_pipeline->uavResourceBindings, effect_pass.storage_bindings.ptr(), sizeof(FfxResourceBinding) * out_pipeline->uavCount); + + out_pipeline->constCount = effect_pass.uniform_bindings.size(); + ERR_FAIL_COND_V(out_pipeline->constCount > FFX_MAX_NUM_CONST_BUFFERS, FFX_ERROR_OUT_OF_RANGE); + memcpy(out_pipeline->cbResourceBindings, effect_pass.uniform_bindings.ptr(), sizeof(FfxResourceBinding) * out_pipeline->constCount); + + bool low_resolution_mvs = (pipeline_description->contextFlags & FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS) == 0; + + if (pass == FFX_FSR2_PASS_ACCUMULATE || pass == FFX_FSR2_PASS_ACCUMULATE_SHARPEN) { + // Change the binding for motion vectors in this particular pass if low resolution MVs are used. + if (low_resolution_mvs) { + FfxResourceBinding &binding = out_pipeline->srvResourceBindings[2]; + wcscpy_s(binding.name, L"r_dilated_motion_vectors"); + } + } + + return FFX_OK; +} + +FfxErrorCode destroy_pipeline_rd(FfxFsr2Interface *backend_interface, FfxPipelineState *pipeline) { + // We don't want to destroy pipelines when the FSR2 API deems it necessary as it'll do so whenever the context is destroyed. + + return FFX_OK; +} + +FfxErrorCode schedule_gpu_job_rd(FfxFsr2Interface *backend_interface, const FfxGpuJobDescription *job) { + ERR_FAIL_COND_V(!backend_interface, FFX_ERROR_INVALID_ARGUMENT); + ERR_FAIL_COND_V(!job, FFX_ERROR_INVALID_ARGUMENT); + + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + scratch.gpu_jobs.push_back(*job); + + return FFX_OK; +} + +FfxErrorCode execute_gpu_job_clear_float_rd(FSR2Context::Scratch &scratch, const FfxClearFloatJobDescription &job) { + RID resource = scratch.resources.rids[job.target.internalIndex]; + FfxResourceDescription &desc = scratch.resources.descriptions[job.target.internalIndex]; + + ERR_FAIL_COND_V(desc.type == FFX_RESOURCE_TYPE_BUFFER, FFX_ERROR_INVALID_ARGUMENT); + + Color color(job.color[0], job.color[1], job.color[2], job.color[3]); + RD::get_singleton()->texture_clear(resource, color, 0, desc.mipCount, 0, 1); + + return FFX_OK; +} + +FfxErrorCode execute_gpu_job_copy_rd(FSR2Context::Scratch &scratch, const FfxCopyJobDescription &job) { + RID src = scratch.resources.rids[job.src.internalIndex]; + RID dst = scratch.resources.rids[job.dst.internalIndex]; + FfxResourceDescription &src_desc = scratch.resources.descriptions[job.src.internalIndex]; + FfxResourceDescription &dst_desc = scratch.resources.descriptions[job.dst.internalIndex]; + + ERR_FAIL_COND_V(src_desc.type == FFX_RESOURCE_TYPE_BUFFER, FFX_ERROR_INVALID_ARGUMENT); + ERR_FAIL_COND_V(dst_desc.type == FFX_RESOURCE_TYPE_BUFFER, FFX_ERROR_INVALID_ARGUMENT); + + for (uint32_t mip_level = 0; mip_level < src_desc.mipCount; mip_level++) { + // Only push the barriers on the last copy. + // FIXME: This could be optimized if RenderingDevice was able to copy multiple mip levels in a single command. + BitField post_barrier = (mip_level == (src_desc.mipCount - 1)) ? RD::BARRIER_MASK_ALL_BARRIERS : RD::BARRIER_MASK_NO_BARRIER; + RD::get_singleton()->texture_copy(src, dst, Vector3(0, 0, 0), Vector3(0, 0, 0), Vector3(src_desc.width, src_desc.height, src_desc.depth), mip_level, mip_level, 0, 0, post_barrier); + } + + return FFX_OK; +} + +FfxErrorCode execute_gpu_job_compute_rd(FSR2Context::Scratch &scratch, const FfxComputeJobDescription &job) { + UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton(); + ERR_FAIL_NULL_V(uniform_set_cache, FFX_ERROR_BACKEND_API_ERROR); + + FSR2Effect::RootSignature &root_signature = *reinterpret_cast(job.pipeline.rootSignature); + ERR_FAIL_COND_V(root_signature.shader_rid.is_null(), FFX_ERROR_INVALID_ARGUMENT); + + FSR2Effect::Pipeline &backend_pipeline = *reinterpret_cast(job.pipeline.pipeline); + ERR_FAIL_COND_V(backend_pipeline.pipeline_rid.is_null(), FFX_ERROR_INVALID_ARGUMENT); + + Vector compute_uniforms; + for (uint32_t i = 0; i < job.pipeline.srvCount; i++) { + RID texture_rid = scratch.resources.rids[job.srvs[i].internalIndex]; + RD::Uniform texture_uniform(RD::UNIFORM_TYPE_TEXTURE, job.pipeline.srvResourceBindings[i].slotIndex, texture_rid); + compute_uniforms.push_back(texture_uniform); + } + + for (uint32_t i = 0; i < job.pipeline.uavCount; i++) { + RID image_rid = scratch.resources.rids[job.uavs[i].internalIndex]; + RD::Uniform storage_uniform; + storage_uniform.uniform_type = RD::UNIFORM_TYPE_IMAGE; + storage_uniform.binding = job.pipeline.uavResourceBindings[i].slotIndex; + + if (job.uavMip[i] > 0) { + LocalVector &mip_slice_rids = scratch.resources.mip_slice_rids[job.uavs[i].internalIndex]; + if (mip_slice_rids.is_empty()) { + mip_slice_rids.resize(scratch.resources.descriptions[job.uavs[i].internalIndex].mipCount); + } + + ERR_FAIL_COND_V(job.uavMip[i] >= mip_slice_rids.size(), FFX_ERROR_INVALID_ARGUMENT); + + if (mip_slice_rids[job.uavMip[i]].is_null()) { + mip_slice_rids[job.uavMip[i]] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), image_rid, 0, job.uavMip[i]); + } + + ERR_FAIL_COND_V(mip_slice_rids[job.uavMip[i]].is_null(), FFX_ERROR_BACKEND_API_ERROR); + + storage_uniform.append_id(mip_slice_rids[job.uavMip[i]]); + } else { + storage_uniform.append_id(image_rid); + } + + compute_uniforms.push_back(storage_uniform); + } + + for (uint32_t i = 0; i < job.pipeline.constCount; i++) { + RID buffer_rid = scratch.ubo_ring_buffer[scratch.ubo_ring_buffer_index]; + scratch.ubo_ring_buffer_index = (scratch.ubo_ring_buffer_index + 1) % FSR2_UBO_RING_BUFFER_SIZE; + + BitField post_barrier = (i == (job.pipeline.constCount - 1)) ? RD::BARRIER_MASK_ALL_BARRIERS : RD::BARRIER_MASK_NO_BARRIER; + RD::get_singleton()->buffer_update(buffer_rid, 0, job.cbs[i].uint32Size * sizeof(uint32_t), job.cbs[i].data, post_barrier); + + RD::Uniform buffer_uniform(RD::UNIFORM_TYPE_UNIFORM_BUFFER, job.pipeline.cbResourceBindings[i].slotIndex, buffer_rid); + compute_uniforms.push_back(buffer_uniform); + } + + FSR2Effect::Device &device = *reinterpret_cast(scratch.device); + RD::Uniform u_point_clamp_sampler(RD::UniformType::UNIFORM_TYPE_SAMPLER, 0, device.point_clamp_sampler); + RD::Uniform u_linear_clamp_sampler(RD::UniformType::UNIFORM_TYPE_SAMPLER, 1, device.linear_clamp_sampler); + + RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin(); + RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, backend_pipeline.pipeline_rid); + RD::get_singleton()->compute_list_bind_uniform_set(compute_list, uniform_set_cache->get_cache(root_signature.shader_rid, 0, u_point_clamp_sampler, u_linear_clamp_sampler), 0); + RD::get_singleton()->compute_list_bind_uniform_set(compute_list, uniform_set_cache->get_cache_vec(root_signature.shader_rid, 1, compute_uniforms), 1); + RD::get_singleton()->compute_list_dispatch(compute_list, job.dimensions[0], job.dimensions[1], job.dimensions[2]); + RD::get_singleton()->compute_list_end(); + + return FFX_OK; +} + +FfxErrorCode execute_gpu_jobs_rd(FfxFsr2Interface *backend_interface, FfxCommandList command_list) { + ERR_FAIL_COND_V(!backend_interface, FFX_ERROR_INVALID_ARGUMENT); + + FSR2Context::Scratch &scratch = *reinterpret_cast(backend_interface->scratchBuffer); + FfxErrorCode error_code = FFX_OK; + for (const FfxGpuJobDescription &job : scratch.gpu_jobs) { + switch (job.jobType) { + case FFX_GPU_JOB_CLEAR_FLOAT: { + error_code = execute_gpu_job_clear_float_rd(scratch, job.clearJobDescriptor); + } break; + case FFX_GPU_JOB_COPY: { + error_code = execute_gpu_job_copy_rd(scratch, job.copyJobDescriptor); + } break; + case FFX_GPU_JOB_COMPUTE: { + error_code = execute_gpu_job_compute_rd(scratch, job.computeJobDescriptor); + } break; + default: { + error_code = FFX_ERROR_INVALID_ARGUMENT; + } break; + } + + if (error_code != FFX_OK) { + scratch.gpu_jobs.clear(); + return error_code; + } + } + + scratch.gpu_jobs.clear(); + + return FFX_OK; +} + +FfxResource get_resource_rd(RID *rid, const wchar_t *name) { + FfxResource res = {}; + if (rid->is_null()) { + return res; + } + + wcscpy_s(res.name, name); + + RD::TextureFormat texture_format = RD::get_singleton()->texture_get_format(*rid); + res.description.type = rd_texture_type_to_ffx_resource_type(texture_format.texture_type); + res.description.format = rd_format_to_ffx_surface_format(texture_format.format); + res.description.width = texture_format.width; + res.description.height = texture_format.height; + res.description.depth = texture_format.depth; + res.description.mipCount = texture_format.mipmaps; + res.description.flags = FFX_RESOURCE_FLAGS_NONE; + res.resource = reinterpret_cast(rid); + res.isDepth = texture_format.usage_bits & RD::TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; + + return res; +} + +FSR2Context::~FSR2Context() { + ffxFsr2ContextDestroy(&fsr_context); +} + +FSR2Effect::FSR2Effect() { + FfxDeviceCapabilities &capabilities = device.capabilities; + uint64_t default_subgroup_size = RD::get_singleton()->limit_get(RD::LIMIT_SUBGROUP_SIZE); + capabilities.minimumSupportedShaderModel = FFX_SHADER_MODEL_5_1; + capabilities.waveLaneCountMin = RD::get_singleton()->limit_get(RD::LIMIT_SUBGROUP_MIN_SIZE); + capabilities.waveLaneCountMax = RD::get_singleton()->limit_get(RD::LIMIT_SUBGROUP_MAX_SIZE); + capabilities.fp16Supported = RD::get_singleton()->has_feature(RD::Features::SUPPORTS_FSR_HALF_FLOAT); + capabilities.raytracingSupported = false; + + bool force_wave_64 = default_subgroup_size == 32 && capabilities.waveLaneCountMax == 64; + bool use_lut = force_wave_64 || default_subgroup_size == 64; + + String general_defines_base = + "\n#define FFX_GPU\n" + "\n#define FFX_GLSL 1\n" + "\n#define FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS 1\n" + "\n#define FFX_FSR2_OPTION_HDR_COLOR_INPUT 1\n" + "\n#define FFX_FSR2_OPTION_GODOT_REACTIVE_MASK_CLAMP 1\n" + "\n#define FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS 1\n"; + + if (use_lut) { + general_defines_base += "\n#define FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE 1\n"; + } + + String general_defines = general_defines_base; + if (capabilities.fp16Supported) { + general_defines += "\n#define FFX_HALF 1\n"; + } + + Vector modes; + modes.push_back(""); + + // Since Godot currently lacks a shader reflection mechanism to persist the name of the bindings in the shader cache and + // there's also no mechanism to compile the shaders offline, the bindings are created manually by looking at the GLSL + // files included in FSR2 and mapping the macro bindings (#define FSR2_BIND_*) to their respective implementation names. + // + // It is not guaranteed these will remain consistent at all between versions of FSR2, so it'll be necessary to keep these + // bindings up to date whenever the library is updated. In such cases, it is very likely the validation layer will throw an + // error if the bindings do not match. + + { + Pass &pass = device.passes[FFX_FSR2_PASS_DEPTH_CLIP]; + pass.shader = &shaders.depth_clip; + pass.shader->initialize(modes, general_defines); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_reconstructed_previous_nearest_depth" }, + FfxResourceBinding{ 1, 0, L"r_dilated_motion_vectors" }, + FfxResourceBinding{ 2, 0, L"r_dilatedDepth" }, + FfxResourceBinding{ 3, 0, L"r_reactive_mask" }, + FfxResourceBinding{ 4, 0, L"r_transparency_and_composition_mask" }, + FfxResourceBinding{ 5, 0, L"r_prepared_input_color" }, + FfxResourceBinding{ 6, 0, L"r_previous_dilated_motion_vectors" }, + FfxResourceBinding{ 7, 0, L"r_input_motion_vectors" }, + FfxResourceBinding{ 8, 0, L"r_input_color_jittered" }, + FfxResourceBinding{ 9, 0, L"r_input_depth" }, + FfxResourceBinding{ 10, 0, L"r_input_exposure" } + }; + + pass.storage_bindings = { + // FSR2_BIND_UAV_DEPTH_CLIP (11) does not point to anything. + FfxResourceBinding{ 12, 0, L"rw_dilated_reactive_masks" }, + FfxResourceBinding{ 13, 0, L"rw_prepared_input_color" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 14, 0, L"cbFSR2" } + }; + } + + { + Pass &pass = device.passes[FFX_FSR2_PASS_RECONSTRUCT_PREVIOUS_DEPTH]; + pass.shader = &shaders.reconstruct_previous_depth; + pass.shader->initialize(modes, general_defines); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_input_motion_vectors" }, + FfxResourceBinding{ 1, 0, L"r_input_depth" }, + FfxResourceBinding{ 2, 0, L"r_input_color_jittered" }, + FfxResourceBinding{ 3, 0, L"r_input_exposure" }, + FfxResourceBinding{ 4, 0, L"r_luma_history" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 5, 0, L"rw_reconstructed_previous_nearest_depth" }, + FfxResourceBinding{ 6, 0, L"rw_dilated_motion_vectors" }, + FfxResourceBinding{ 7, 0, L"rw_dilatedDepth" }, + FfxResourceBinding{ 8, 0, L"rw_prepared_input_color" }, + FfxResourceBinding{ 9, 0, L"rw_luma_history" }, + // FSR2_BIND_UAV_LUMA_INSTABILITY (10) does not point to anything. + FfxResourceBinding{ 11, 0, L"rw_lock_input_luma" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 12, 0, L"cbFSR2" } + }; + } + + { + Pass &pass = device.passes[FFX_FSR2_PASS_LOCK]; + pass.shader = &shaders.lock; + pass.shader->initialize(modes, general_defines); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_lock_input_luma" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 1, 0, L"rw_new_locks" }, + FfxResourceBinding{ 2, 0, L"rw_reconstructed_previous_nearest_depth" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 3, 0, L"cbFSR2" } + }; + } + + { + Vector accumulate_modes; + accumulate_modes.push_back("\n"); + accumulate_modes.push_back("\n#define FFX_FSR2_OPTION_APPLY_SHARPENING 1\n"); + + String general_defines_accumulate; + if (RD::get_singleton()->get_device_vendor_name() == "NVIDIA") { + // Workaround: Disable FP16 path for the accumulate pass on NVIDIA due to reduced occupancy and high VRAM throughput. + general_defines_accumulate = general_defines_base; + } else { + general_defines_accumulate = general_defines; + } + + Pass &pass = device.passes[FFX_FSR2_PASS_ACCUMULATE]; + pass.shader = &shaders.accumulate; + pass.shader->initialize(accumulate_modes, general_defines_accumulate); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_input_exposure" }, + FfxResourceBinding{ 1, 0, L"r_dilated_reactive_masks" }, + FfxResourceBinding{ 2, 0, L"r_input_motion_vectors" }, + FfxResourceBinding{ 3, 0, L"r_internal_upscaled_color" }, + FfxResourceBinding{ 4, 0, L"r_lock_status" }, + FfxResourceBinding{ 5, 0, L"r_input_depth" }, + FfxResourceBinding{ 6, 0, L"r_prepared_input_color" }, + // FSR2_BIND_SRV_LUMA_INSTABILITY(7) does not point to anything. + FfxResourceBinding{ 8, 0, L"r_lanczos_lut" }, + FfxResourceBinding{ 9, 0, L"r_upsample_maximum_bias_lut" }, + FfxResourceBinding{ 10, 0, L"r_imgMips" }, + FfxResourceBinding{ 11, 0, L"r_auto_exposure" }, + FfxResourceBinding{ 12, 0, L"r_luma_history" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 13, 0, L"rw_internal_upscaled_color" }, + FfxResourceBinding{ 14, 0, L"rw_lock_status" }, + FfxResourceBinding{ 15, 0, L"rw_upscaled_output" }, + FfxResourceBinding{ 16, 0, L"rw_new_locks" }, + FfxResourceBinding{ 17, 0, L"rw_luma_history" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 18, 0, L"cbFSR2" } + }; + + // Sharpen pass is a clone of the accumulate pass. + Pass &sharpen_pass = device.passes[FFX_FSR2_PASS_ACCUMULATE_SHARPEN]; + sharpen_pass = pass; + sharpen_pass.shader_variant = 1; + } + + { + Pass &pass = device.passes[FFX_FSR2_PASS_RCAS]; + pass.shader = &shaders.rcas; + pass.shader->initialize(modes, general_defines_base); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_input_exposure" }, + FfxResourceBinding{ 1, 0, L"r_rcas_input" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 2, 0, L"rw_upscaled_output" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 3, 0, L"cbFSR2" }, + FfxResourceBinding{ 4, 0, L"cbRCAS" } + }; + } + + { + Pass &pass = device.passes[FFX_FSR2_PASS_COMPUTE_LUMINANCE_PYRAMID]; + pass.shader = &shaders.compute_luminance_pyramid; + pass.shader->initialize(modes, general_defines_base); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_input_color_jittered" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 1, 0, L"rw_spd_global_atomic" }, + FfxResourceBinding{ 2, 0, L"rw_img_mip_shading_change" }, + FfxResourceBinding{ 3, 0, L"rw_img_mip_5" }, + FfxResourceBinding{ 4, 0, L"rw_auto_exposure" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 5, 0, L"cbFSR2" }, + FfxResourceBinding{ 6, 0, L"cbSPD" } + }; + } + + { + Pass &pass = device.passes[FFX_FSR2_PASS_GENERATE_REACTIVE]; + pass.shader = &shaders.autogen_reactive; + pass.shader->initialize(modes, general_defines); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_input_opaque_only" }, + FfxResourceBinding{ 1, 0, L"r_input_color_jittered" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 2, 0, L"rw_output_autoreactive" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 3, 0, L"cbGenerateReactive" }, + FfxResourceBinding{ 4, 0, L"cbFSR2" } + }; + } + + { + Pass &pass = device.passes[FFX_FSR2_PASS_TCR_AUTOGENERATE]; + pass.shader = &shaders.tcr_autogen; + pass.shader->initialize(modes, general_defines); + pass.shader_version = pass.shader->version_create(); + + pass.sampled_bindings = { + FfxResourceBinding{ 0, 0, L"r_input_opaque_only" }, + FfxResourceBinding{ 1, 0, L"r_input_color_jittered" }, + FfxResourceBinding{ 2, 0, L"r_input_motion_vectors" }, + FfxResourceBinding{ 3, 0, L"r_input_prev_color_pre_alpha" }, + FfxResourceBinding{ 4, 0, L"r_input_prev_color_post_alpha" }, + FfxResourceBinding{ 5, 0, L"r_reactive_mask" }, + FfxResourceBinding{ 6, 0, L"r_transparency_and_composition_mask" }, + FfxResourceBinding{ 13, 0, L"r_input_depth" } + }; + + pass.storage_bindings = { + FfxResourceBinding{ 7, 0, L"rw_output_autoreactive" }, + FfxResourceBinding{ 8, 0, L"rw_output_autocomposition" }, + FfxResourceBinding{ 9, 0, L"rw_output_prev_color_pre_alpha" }, + FfxResourceBinding{ 10, 0, L"rw_output_prev_color_post_alpha" } + }; + + pass.uniform_bindings = { + FfxResourceBinding{ 11, 0, L"cbFSR2" }, + FfxResourceBinding{ 12, 0, L"cbGenerateReactive" } + }; + } + + RD::SamplerState state; + state.mag_filter = RD::SAMPLER_FILTER_NEAREST; + state.min_filter = RD::SAMPLER_FILTER_NEAREST; + state.repeat_u = RD::SAMPLER_REPEAT_MODE_CLAMP_TO_EDGE; + state.repeat_v = RD::SAMPLER_REPEAT_MODE_CLAMP_TO_EDGE; + state.repeat_w = RD::SAMPLER_REPEAT_MODE_CLAMP_TO_EDGE; + state.min_lod = -1000.0f; + state.max_lod = 1000.0f; + state.anisotropy_max = 1.0; + device.point_clamp_sampler = RD::get_singleton()->sampler_create(state); + ERR_FAIL_COND(device.point_clamp_sampler.is_null()); + + state.mag_filter = RD::SAMPLER_FILTER_LINEAR; + state.min_filter = RD::SAMPLER_FILTER_LINEAR; + device.linear_clamp_sampler = RD::get_singleton()->sampler_create(state); + ERR_FAIL_COND(device.linear_clamp_sampler.is_null()); +} + +FSR2Effect::~FSR2Effect() { + RD::get_singleton()->free(device.point_clamp_sampler); + RD::get_singleton()->free(device.linear_clamp_sampler); + + for (uint32_t i = 0; i < FFX_FSR2_PASS_COUNT; i++) { + RD::get_singleton()->free(device.passes[i].pipeline.pipeline_rid); + device.passes[i].shader->version_free(device.passes[i].shader_version); + } +} + +FSR2Context *FSR2Effect::create_context(Size2i p_internal_size, Size2i p_target_size) { + FSR2Context *context = memnew(RendererRD::FSR2Context); + context->fsr_desc.flags = FFX_FSR2_ENABLE_HIGH_DYNAMIC_RANGE; + context->fsr_desc.maxRenderSize.width = p_internal_size.x; + context->fsr_desc.maxRenderSize.height = p_internal_size.y; + context->fsr_desc.displaySize.width = p_target_size.x; + context->fsr_desc.displaySize.height = p_target_size.y; + context->fsr_desc.device = &device; + + FfxFsr2Interface &functions = context->fsr_desc.callbacks; + functions.fpCreateBackendContext = create_backend_context_rd; + functions.fpGetDeviceCapabilities = get_device_capabilities_rd; + functions.fpDestroyBackendContext = destroy_backend_context_rd; + functions.fpCreateResource = create_resource_rd; + functions.fpRegisterResource = register_resource_rd; + functions.fpUnregisterResources = unregister_resources_rd; + functions.fpGetResourceDescription = get_resource_description_rd; + functions.fpDestroyResource = destroy_resource_rd; + functions.fpCreatePipeline = create_pipeline_rd; + functions.fpDestroyPipeline = destroy_pipeline_rd; + functions.fpScheduleGpuJob = schedule_gpu_job_rd; + functions.fpExecuteGpuJobs = execute_gpu_jobs_rd; + functions.scratchBuffer = &context->scratch; + functions.scratchBufferSize = sizeof(context->scratch); + + FfxErrorCode result = ffxFsr2ContextCreate(&context->fsr_context, &context->fsr_desc); + if (result == FFX_OK) { + return context; + } else { + memdelete(context); + return nullptr; + } +} + +void FSR2Effect::upscale(const Parameters &p_params) { + // TODO: Transparency & Composition mask is not implemented. + FfxFsr2DispatchDescription dispatch_desc = {}; + RID color = p_params.color; + RID depth = p_params.depth; + RID velocity = p_params.velocity; + RID reactive = p_params.reactive; + RID exposure = p_params.exposure; + RID output = p_params.output; + dispatch_desc.commandList = nullptr; + dispatch_desc.color = get_resource_rd(&color, L"color"); + dispatch_desc.depth = get_resource_rd(&depth, L"depth"); + dispatch_desc.motionVectors = get_resource_rd(&velocity, L"velocity"); + dispatch_desc.reactive = get_resource_rd(&reactive, L"reactive"); + dispatch_desc.exposure = get_resource_rd(&exposure, L"exposure"); + dispatch_desc.transparencyAndComposition = {}; + dispatch_desc.output = get_resource_rd(&output, L"output"); + dispatch_desc.colorOpaqueOnly = {}; + dispatch_desc.jitterOffset.x = p_params.jitter.x; + dispatch_desc.jitterOffset.y = p_params.jitter.y; + dispatch_desc.motionVectorScale.x = float(p_params.internal_size.width); + dispatch_desc.motionVectorScale.y = float(p_params.internal_size.height); + dispatch_desc.reset = p_params.reset_accumulation; + dispatch_desc.renderSize.width = p_params.internal_size.width; + dispatch_desc.renderSize.height = p_params.internal_size.height; + dispatch_desc.enableSharpening = (p_params.sharpness > 1e-6f); + dispatch_desc.sharpness = p_params.sharpness; + dispatch_desc.frameTimeDelta = p_params.delta_time; + dispatch_desc.preExposure = 1.0f; + dispatch_desc.cameraNear = p_params.z_near; + dispatch_desc.cameraFar = p_params.z_far; + dispatch_desc.cameraFovAngleVertical = p_params.fovy; + dispatch_desc.viewSpaceToMetersFactor = 1.0f; + dispatch_desc.enableAutoReactive = false; + dispatch_desc.autoTcThreshold = 1.0f; + dispatch_desc.autoTcScale = 1.0f; + dispatch_desc.autoReactiveScale = 1.0f; + dispatch_desc.autoReactiveMax = 1.0f; + + Projection reprojection = p_params.previous_projection.flipped_y() * p_params.previous_transform.affine_inverse() * p_params.current_transform * p_params.current_projection.flipped_y().inverse(); + RendererRD::MaterialStorage::store_camera(reprojection, dispatch_desc.reprojectionMatrix); + + FfxErrorCode result = ffxFsr2ContextDispatch(&p_params.context->fsr_context, &dispatch_desc); + ERR_FAIL_COND(result != FFX_OK); +} diff --git a/servers/rendering/renderer_rd/effects/fsr2.h b/servers/rendering/renderer_rd/effects/fsr2.h new file mode 100644 index 000000000000..b4ea9b051971 --- /dev/null +++ b/servers/rendering/renderer_rd/effects/fsr2.h @@ -0,0 +1,202 @@ +/**************************************************************************/ +/* fsr2.h */ +/**************************************************************************/ +/* This file is part of: */ +/* GODOT ENGINE */ +/* https://godotengine.org */ +/**************************************************************************/ +/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */ +/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */ +/* */ +/* Permission is hereby granted, free of charge, to any person obtaining */ +/* a copy of this software and associated documentation files (the */ +/* "Software"), to deal in the Software without restriction, including */ +/* without limitation the rights to use, copy, modify, merge, publish, */ +/* distribute, sublicense, and/or sell copies of the Software, and to */ +/* permit persons to whom the Software is furnished to do so, subject to */ +/* the following conditions: */ +/* */ +/* The above copyright notice and this permission notice shall be */ +/* included in all copies or substantial portions of the Software. */ +/* */ +/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */ +/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */ +/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */ +/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */ +/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */ +/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */ +/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ +/**************************************************************************/ + +#ifndef FSR2_RD_H +#define FSR2_RD_H + +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_accumulate_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_autogen_reactive_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_compute_luminance_pyramid_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_depth_clip_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_lock_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_rcas_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_reconstruct_previous_depth_pass.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/effects/fsr2_tcr_autogen_pass.glsl.gen.h" + +// This flag doesn't actually control anything GCC specific in FSR2. It determines +// if symbols should be exported, which is not required for Godot. +#ifndef FFX_GCC +#define FFX_GCC +#endif + +#include "thirdparty/amd-fsr2/ffx_fsr2.h" + +#define FSR2_MAX_QUEUED_FRAMES (4) +#define FSR2_MAX_UNIFORM_BUFFERS (4) +#define FSR2_MAX_BUFFERED_DESCRIPTORS (FFX_FSR2_PASS_COUNT * FSR2_MAX_QUEUED_FRAMES) +#define FSR2_UBO_RING_BUFFER_SIZE (FSR2_MAX_BUFFERED_DESCRIPTORS * FSR2_MAX_UNIFORM_BUFFERS) + +namespace RendererRD { +class FSR2Context { +public: + enum ResourceID : uint32_t { + RESOURCE_ID_DYNAMIC = 0xFFFFFFFF + }; + + struct Resources { + LocalVector rids; + LocalVector> mip_slice_rids; + LocalVector ids; + LocalVector descriptions; + LocalVector dynamic_list; + LocalVector free_list; + + uint32_t add(RID p_rid, bool p_dynamic, uint32_t p_id, FfxResourceDescription p_description) { + uint32_t ret_index; + if (free_list.is_empty()) { + ret_index = rids.size(); + uint32_t new_size = ret_index + 1; + rids.resize(new_size); + mip_slice_rids.resize(new_size); + ids.resize(new_size); + descriptions.resize(new_size); + } else { + uint32_t end_index = free_list.size() - 1; + ret_index = free_list[end_index]; + free_list.resize(end_index); + } + + rids[ret_index] = p_rid; + mip_slice_rids[ret_index].clear(); + ids[ret_index] = p_id; + descriptions[ret_index] = p_description; + + if (p_dynamic) { + dynamic_list.push_back(ret_index); + } + + return ret_index; + } + + void remove(uint32_t p_index) { + DEV_ASSERT(p_index < rids.size()); + free_list.push_back(p_index); + rids[p_index] = RID(); + mip_slice_rids[p_index].clear(); + ids[p_index] = 0; + descriptions[p_index] = {}; + dynamic_list.erase(p_index); + } + + uint32_t size() const { + return rids.size(); + } + }; + + struct Scratch { + Resources resources; + LocalVector gpu_jobs; + RID ubo_ring_buffer[FSR2_UBO_RING_BUFFER_SIZE]; + uint32_t ubo_ring_buffer_index = 0; + FfxDevice device = nullptr; + }; + + Scratch scratch; + FfxFsr2Context fsr_context; + FfxFsr2ContextDescription fsr_desc; + + ~FSR2Context(); +}; + +class FSR2Effect { +public: + struct RootSignature { + // Proxy structure to store the shader required by RD that uses the terminology used by the FSR2 API. + RID shader_rid; + }; + + struct Pipeline { + RID pipeline_rid; + }; + + struct Pass { + ShaderRD *shader; + RID shader_version; + RootSignature root_signature; + uint32_t shader_variant = 0; + Pipeline pipeline; + Vector sampled_bindings; + Vector storage_bindings; + Vector uniform_bindings; + }; + + struct Device { + Pass passes[FFX_FSR2_PASS_COUNT]; + FfxDeviceCapabilities capabilities; + RID point_clamp_sampler; + RID linear_clamp_sampler; + }; + + struct Parameters { + FSR2Context *context; + Size2i internal_size; + RID color; + RID depth; + RID velocity; + RID reactive; + RID exposure; + RID output; + float z_near = 0.0f; + float z_far = 0.0f; + float fovy = 0.0f; + Vector2 jitter; + float delta_time = 0.0f; + float sharpness = 0.0f; + bool reset_accumulation = false; + Projection current_projection; + Transform3D current_transform; + Projection previous_projection; + Transform3D previous_transform; + }; + + FSR2Effect(); + ~FSR2Effect(); + FSR2Context *create_context(Size2i p_internal_size, Size2i p_target_size); + void upscale(const Parameters &p_params); + +private: + struct { + Fsr2DepthClipPassShaderRD depth_clip; + Fsr2ReconstructPreviousDepthPassShaderRD reconstruct_previous_depth; + Fsr2LockPassShaderRD lock; + Fsr2AccumulatePassShaderRD accumulate; + Fsr2AccumulatePassShaderRD accumulate_sharpen; + Fsr2RcasPassShaderRD rcas; + Fsr2ComputeLuminancePyramidPassShaderRD compute_luminance_pyramid; + Fsr2AutogenReactivePassShaderRD autogen_reactive; + Fsr2TcrAutogenPassShaderRD tcr_autogen; + } shaders; + + Device device; +}; + +} // namespace RendererRD + +#endif // FSR2_RD_H diff --git a/servers/rendering/renderer_rd/effects/taa.cpp b/servers/rendering/renderer_rd/effects/taa.cpp index 61e0d3866ca5..c1037ec11a2f 100644 --- a/servers/rendering/renderer_rd/effects/taa.cpp +++ b/servers/rendering/renderer_rd/effects/taa.cpp @@ -47,20 +47,6 @@ TAA::~TAA() { taa_shader.version_free(shader_version); } -void TAA::msaa_resolve(Ref p_render_buffers) { - if (!p_render_buffers->has_velocity_buffer(true)) { - // nothing to resolve - return; - } - - for (uint32_t v = 0; v < p_render_buffers->get_view_count(); v++) { - RID velocity_buffer_msaa = p_render_buffers->get_velocity_buffer(true, v); - RID velocity_buffer = p_render_buffers->get_velocity_buffer(false, v); - - RD::get_singleton()->texture_resolve_multisample(velocity_buffer_msaa, velocity_buffer); - } -} - void TAA::resolve(RID p_frame, RID p_temp, RID p_depth, RID p_velocity, RID p_prev_velocity, RID p_history, Size2 p_resolution, float p_z_near, float p_z_far) { UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton(); ERR_FAIL_NULL(uniform_set_cache); diff --git a/servers/rendering/renderer_rd/effects/taa.h b/servers/rendering/renderer_rd/effects/taa.h index 9e7ad76fb8f1..f50e297fe56c 100644 --- a/servers/rendering/renderer_rd/effects/taa.h +++ b/servers/rendering/renderer_rd/effects/taa.h @@ -45,7 +45,6 @@ class TAA { TAA(); ~TAA(); - void msaa_resolve(Ref p_render_buffers); void process(Ref p_render_buffers, RD::DataFormat p_format, float p_z_near, float p_z_far); private: diff --git a/servers/rendering/renderer_rd/environment/sky.cpp b/servers/rendering/renderer_rd/environment/sky.cpp index f94c701adaf6..79859790429a 100644 --- a/servers/rendering/renderer_rd/environment/sky.cpp +++ b/servers/rendering/renderer_rd/environment/sky.cpp @@ -973,7 +973,7 @@ SkyRD::~SkyRD() { } } -void SkyRD::setup_sky(RID p_env, Ref p_render_buffers, const PagedArray &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, RendererSceneRenderRD *p_scene_render) { +void SkyRD::setup_sky(RID p_env, Ref p_render_buffers, const PagedArray &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, Vector2 p_jitter, RendererSceneRenderRD *p_scene_render) { RendererRD::LightStorage *light_storage = RendererRD::LightStorage::get_singleton(); RendererRD::MaterialStorage *material_storage = RendererRD::MaterialStorage::get_singleton(); ERR_FAIL_COND(p_env.is_null()); @@ -1173,18 +1173,21 @@ void SkyRD::setup_sky(RID p_env, Ref p_render_buffers, con } } + Projection correction; + correction.add_jitter_offset(p_jitter); + sky_scene_state.view_count = p_view_count; sky_scene_state.cam_transform = p_cam_transform; - sky_scene_state.cam_projection = p_cam_projection; // We only use this when rendering a single view. + sky_scene_state.cam_projection = correction * p_cam_projection; // We only use this when rendering a single view. // Our info in our UBO is only used if we're rendering stereo. for (uint32_t i = 0; i < p_view_count; i++) { - Projection view_inv_projection = p_view_projections[i].inverse(); + Projection view_inv_projection = (correction * p_view_projections[i]).inverse(); if (p_view_count > 1) { RendererRD::MaterialStorage::store_camera(p_cam_projection * view_inv_projection, sky_scene_state.ubo.combined_reprojection[i]); } else { Projection ident; - RendererRD::MaterialStorage::store_camera(ident, sky_scene_state.ubo.combined_reprojection[i]); + RendererRD::MaterialStorage::store_camera(correction, sky_scene_state.ubo.combined_reprojection[i]); } RendererRD::MaterialStorage::store_camera(view_inv_projection, sky_scene_state.ubo.view_inv_projections[i]); diff --git a/servers/rendering/renderer_rd/environment/sky.h b/servers/rendering/renderer_rd/environment/sky.h index ee2d81757ce6..b146a416f9b8 100644 --- a/servers/rendering/renderer_rd/environment/sky.h +++ b/servers/rendering/renderer_rd/environment/sky.h @@ -294,7 +294,7 @@ class SkyRD { void set_texture_format(RD::DataFormat p_texture_format); ~SkyRD(); - void setup_sky(RID p_env, Ref p_render_buffers, const PagedArray &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, RendererSceneRenderRD *p_scene_render); + void setup_sky(RID p_env, Ref p_render_buffers, const PagedArray &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, Vector2 p_jitter, RendererSceneRenderRD *p_scene_render); void update_radiance_buffers(Ref p_render_buffers, RID p_env, const Vector3 &p_global_pos, double p_time, float p_luminance_multiplier = 1.0); void update_res_buffers(Ref p_render_buffers, RID p_env, double p_time, float p_luminance_multiplier = 1.0); void draw_sky(RD::DrawListID p_draw_list, Ref p_render_buffers, RID p_env, RID p_fb, double p_time, float p_luminance_multiplier = 1.0); diff --git a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp index 2e64946fd265..c3c662ca5ef6 100644 --- a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp +++ b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp @@ -105,6 +105,12 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::ensure_voxelgi() } } +void RenderForwardClustered::RenderBufferDataForwardClustered::ensure_fsr2(RendererRD::FSR2Effect *effect) { + if (fsr2_context == nullptr) { + fsr2_context = effect->create_context(render_buffers->get_internal_size(), render_buffers->get_target_size()); + } +} + void RenderForwardClustered::RenderBufferDataForwardClustered::free_data() { // JIC, should already have been cleared if (render_buffers) { @@ -120,6 +126,11 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::free_data() { cluster_builder = nullptr; } + if (fsr2_context) { + memdelete(fsr2_context); + fsr2_context = nullptr; + } + if (!render_sdfgi_uniform_set.is_null() && RD::get_singleton()->uniform_set_is_valid(render_sdfgi_uniform_set)) { RD::get_singleton()->free(render_sdfgi_uniform_set); } @@ -230,6 +241,14 @@ RID RenderForwardClustered::RenderBufferDataForwardClustered::get_specular_only_ return FramebufferCacheRD::get_singleton()->get_cache_multiview(render_buffers->get_view_count(), specular); } +RID RenderForwardClustered::RenderBufferDataForwardClustered::get_velocity_only_fb() { + bool use_msaa = render_buffers->get_msaa_3d() != RS::VIEWPORT_MSAA_DISABLED; + + RID velocity = render_buffers->get_texture(RB_SCOPE_BUFFERS, use_msaa ? RB_TEX_VELOCITY_MSAA : RB_TEX_VELOCITY); + + return FramebufferCacheRD::get_singleton()->get_cache_multiview(render_buffers->get_view_count(), velocity); +} + void RenderForwardClustered::setup_render_buffer_data(Ref p_render_buffers) { Ref data; data.instantiate(); @@ -582,7 +601,7 @@ void RenderForwardClustered::_render_list_with_threads(RenderListParameters *p_p } } -void RenderForwardClustered::_setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers, bool p_pancake_shadows, int p_index) { +void RenderForwardClustered::_setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers, bool p_apply_alpha_multiplier, bool p_pancake_shadows, int p_index) { RendererRD::LightStorage *light_storage = RendererRD::LightStorage::get_singleton(); Ref rd = p_render_data->render_buffers; @@ -598,7 +617,7 @@ void RenderForwardClustered::_setup_environment(const RenderDataRD *p_render_dat } } - p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers); + p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers, p_apply_alpha_multiplier); // now do implementation UBO @@ -1581,15 +1600,19 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co RENDER_TIMESTAMP("Setup 3D Scene"); // check if we need motion vectors + bool motion_vectors_required; if (get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS) { - p_render_data->scene_data->calculate_motion_vectors = true; + motion_vectors_required = true; } else if (!is_reflection_probe && rb->get_use_taa()) { - p_render_data->scene_data->calculate_motion_vectors = true; + motion_vectors_required = true; + } else if (rb->get_scaling_3d_mode() == RenderingServer::ViewportScaling3DMode::VIEWPORT_SCALING_3D_MODE_FSR2) { + motion_vectors_required = true; } else { - p_render_data->scene_data->calculate_motion_vectors = false; + motion_vectors_required = false; } //p_render_data->scene_data->subsurface_scatter_width = subsurface_scatter_size; + p_render_data->scene_data->calculate_motion_vectors = motion_vectors_required; p_render_data->scene_data->directional_light_count = 0; p_render_data->scene_data->opaque_prepass_threshold = 0.99f; @@ -1607,6 +1630,10 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co bool using_voxelgi = false; bool reverse_cull = p_render_data->scene_data->cam_transform.basis.determinant() < 0; bool using_ssil = !is_reflection_probe && p_render_data->environment.is_valid() && environment_get_ssil_enabled(p_render_data->environment); + bool using_taa = taa && rb->get_use_taa(); + bool using_fsr2 = fsr2_effect && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR2; + bool using_reactive = using_fsr2; + bool using_derived_mvs = using_fsr2; if (is_reflection_probe) { uint32_t resolution = light_storage->reflection_probe_instance_get_resolution(p_render_data->reflection_probe); @@ -1625,7 +1652,7 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co } else { screen_size = rb->get_internal_size(); - if (rb->get_use_taa() || get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS) { + if (p_render_data->scene_data->calculate_motion_vectors) { color_pass_flags |= COLOR_PASS_FLAG_MOTION_VECTORS; scene_shader.enable_advanced_shader_group(); } @@ -1792,9 +1819,9 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co correction.set_depth_correction(true); Projection projection = correction * p_render_data->scene_data->cam_projection; - sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, this); + sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, Vector2(0.0f, 0.0f), this); } else { - sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, this); + sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, p_render_data->scene_data->taa_jitter, this); } sky_energy_multiplier *= bg_energy_multiplier; @@ -1889,7 +1916,7 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co p_render_data->scene_data->directional_light_count = p_render_data->directional_light_count; p_render_data->scene_data->opaque_prepass_threshold = 0.0f; - _setup_environment(p_render_data, is_reflection_probe, screen_size, !is_reflection_probe, p_default_bg_color, true); + _setup_environment(p_render_data, is_reflection_probe, screen_size, !is_reflection_probe, p_default_bg_color, true, using_reactive); RENDER_TIMESTAMP("Render Opaque Pass"); @@ -1912,7 +1939,9 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co if (rb_data.is_valid()) { c.push_back(Color(0, 0, 0, 0)); // Separate specular - c.push_back(Color(0, 0, 0, 0)); // Motion vectors + + // Motion vectors need to be cleared to an invalid value when the effects support deriving the information from the camera movement and the depth buffer. + c.push_back(using_derived_mvs ? Color(-1, -1, 0, 0) : Color(0, 0, 0, 0)); } } @@ -1923,6 +1952,12 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co RD::get_singleton()->draw_list_begin(rb_data->get_specular_only_fb(), RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_CONTINUE); RD::get_singleton()->draw_list_end(); } + + if (will_continue_color && using_reactive) { + // Close the motion vectors framebuffer if the transparent pass prefers to use the reactive mask instead. + RD::get_singleton()->draw_list_begin(rb_data->get_velocity_only_fb(), RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_CONTINUE); + RD::get_singleton()->draw_list_end(); + } } RD::get_singleton()->draw_command_end_label(); @@ -2049,6 +2084,11 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co { uint32_t transparent_color_pass_flags = (color_pass_flags | COLOR_PASS_FLAG_TRANSPARENT) & ~(COLOR_PASS_FLAG_SEPARATE_SPECULAR); + if (using_reactive) { + // Motion vectors on transparent draw calls are not required when using the reactive mask. + transparent_color_pass_flags &= ~(COLOR_PASS_FLAG_MOTION_VECTORS); + } + RID alpha_framebuffer = rb_data.is_valid() ? rb_data->get_color_pass_fb(transparent_color_pass_flags) : color_only_framebuffer; RenderListParameters render_list_params(render_list[RENDER_LIST_ALPHA].elements.ptr(), render_list[RENDER_LIST_ALPHA].element_info.ptr(), render_list[RENDER_LIST_ALPHA].elements.size(), false, PASS_MODE_COLOR, transparent_color_pass_flags, rb_data.is_null(), p_render_data->directional_light_soft_shadows, rp_uniform_set, get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_WIREFRAME, Vector2(), p_render_data->scene_data->lod_distance_multiplier, p_render_data->scene_data->screen_mesh_lod_threshold, p_render_data->scene_data->view_count); _render_list_with_threads(&render_list_params, alpha_framebuffer, can_continue_color ? RD::INITIAL_ACTION_CONTINUE : RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, can_continue_depth ? RD::INITIAL_ACTION_CONTINUE : RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ); @@ -2061,12 +2101,14 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co RD::get_singleton()->draw_command_begin_label("Resolve"); if (rb_data.is_valid() && rb->get_msaa_3d() != RS::VIEWPORT_MSAA_DISABLED) { + bool resolve_velocity_buffer = (using_taa || using_fsr2) && rb->has_velocity_buffer(true); for (uint32_t v = 0; v < rb->get_view_count(); v++) { RD::get_singleton()->texture_resolve_multisample(rb->get_color_msaa(v), rb->get_internal_texture(v)); resolve_effects->resolve_depth(rb->get_depth_msaa(v), rb->get_depth_texture(v), rb->get_internal_size(), texture_multisamples[rb->get_msaa_3d()]); - } - if (taa && rb->get_use_taa()) { - taa->msaa_resolve(rb); + + if (resolve_velocity_buffer) { + RD::get_singleton()->texture_resolve_multisample(rb->get_velocity_buffer(true, v), rb->get_velocity_buffer(false, v)); + } } } @@ -2079,9 +2121,49 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co } RD::get_singleton()->draw_command_end_label(); - if (rb_data.is_valid() && taa && rb->get_use_taa()) { - RENDER_TIMESTAMP("TAA") - taa->process(rb, _render_buffers_get_color_format(), p_render_data->scene_data->z_near, p_render_data->scene_data->z_far); + if (rb_data.is_valid()) { + if (using_fsr2 && fsr2_effect) { + rb->ensure_upscaled(); + rb_data->ensure_fsr2(fsr2_effect); + + RID exposure; + if (RSG::camera_attributes->camera_attributes_uses_auto_exposure(p_render_data->camera_attributes)) { + exposure = luminance->get_current_luminance_buffer(rb); + } + + RENDER_TIMESTAMP("FSR2"); + for (uint32_t v = 0; v < rb->get_view_count(); v++) { + // FIXME: The engine does not provide a way to reset the accumulation. + real_t fov = p_render_data->scene_data->cam_projection.get_fov(); + real_t aspect = p_render_data->scene_data->cam_projection.get_aspect(); + real_t fovy = p_render_data->scene_data->cam_projection.get_fovy(fov, aspect); + Vector2 jitter = p_render_data->scene_data->taa_jitter * Vector2(rb->get_internal_size()) * 0.5f; + RendererRD::FSR2Effect::Parameters params; + params.context = rb_data->get_fsr2_context(); + params.internal_size = rb->get_internal_size(); + params.sharpness = CLAMP(1.0f - (rb->get_fsr_sharpness() / 2.0f), 0.0f, 1.0f); + params.color = rb->get_internal_texture(v); + params.depth = rb->get_depth_texture(v); + params.velocity = rb->get_velocity_buffer(false, v); + params.reactive = rb->get_internal_texture_reactive(v); + params.exposure = exposure; + params.output = rb->get_upscaled_texture(v); + params.z_near = p_render_data->scene_data->z_near; + params.z_far = p_render_data->scene_data->z_far; + params.fovy = fovy; + params.jitter = jitter; + params.delta_time = float(time_step); + params.reset_accumulation = false; + params.current_projection = p_render_data->scene_data->cam_projection; + params.current_transform = p_render_data->scene_data->cam_transform; + params.previous_projection = p_render_data->scene_data->prev_cam_projection; + params.previous_transform = p_render_data->scene_data->prev_cam_transform; + fsr2_effect->upscale(params); + } + } else if (using_taa) { + RENDER_TIMESTAMP("TAA"); + taa->process(rb, _render_buffers_get_color_format(), p_render_data->scene_data->z_near, p_render_data->scene_data->z_far); + } } if (rb_data.is_valid()) { @@ -2354,7 +2436,7 @@ void RenderForwardClustered::_render_shadow_append(RID p_framebuffer, const Page render_data.instances = &p_instances; render_data.render_info = p_render_info; - _setup_environment(&render_data, true, Vector2(1, 1), !p_flip_y, Color(), false, p_use_pancake, shadow_pass_index); + _setup_environment(&render_data, true, Vector2(1, 1), !p_flip_y, Color(), false, false, p_use_pancake, shadow_pass_index); if (get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_DISABLE_LOD) { scene_data.screen_mesh_lod_threshold = 0.0; @@ -2450,7 +2532,7 @@ void RenderForwardClustered::_render_particle_collider_heightfield(RID p_fb, con _update_render_base_uniform_set(); - _setup_environment(&render_data, true, Vector2(1, 1), true, Color(), false, false); + _setup_environment(&render_data, true, Vector2(1, 1), true, Color(), false, false, false); PassMode pass_mode = PASS_MODE_SHADOW; @@ -3982,6 +4064,7 @@ RenderForwardClustered::RenderForwardClustered() { resolve_effects = memnew(RendererRD::Resolve()); taa = memnew(RendererRD::TAA); + fsr2_effect = memnew(RendererRD::FSR2Effect); ss_effects = memnew(RendererRD::SSEffects); } @@ -3996,6 +4079,11 @@ RenderForwardClustered::~RenderForwardClustered() { taa = nullptr; } + if (fsr2_effect) { + memdelete(fsr2_effect); + fsr2_effect = nullptr; + } + if (resolve_effects != nullptr) { memdelete(resolve_effects); resolve_effects = nullptr; diff --git a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h index a89c77c652ee..76edee48648b 100644 --- a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h +++ b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h @@ -33,6 +33,7 @@ #include "core/templates/paged_allocator.h" #include "servers/rendering/renderer_rd/cluster_builder_rd.h" +#include "servers/rendering/renderer_rd/effects/fsr2.h" #include "servers/rendering/renderer_rd/effects/resolve.h" #include "servers/rendering/renderer_rd/effects/ss_effects.h" #include "servers/rendering/renderer_rd/effects/taa.h" @@ -100,6 +101,7 @@ class RenderForwardClustered : public RendererSceneRenderRD { private: RenderSceneBuffersRD *render_buffers = nullptr; + RendererRD::FSR2Context *fsr2_context = nullptr; public: ClusterBuilderRD *cluster_builder = nullptr; @@ -140,10 +142,14 @@ class RenderForwardClustered : public RendererSceneRenderRD { RID get_voxelgi(uint32_t p_layer) { return render_buffers->get_texture_slice(RB_SCOPE_FORWARD_CLUSTERED, RB_TEX_VOXEL_GI, p_layer, 0); } RID get_voxelgi_msaa(uint32_t p_layer) { return render_buffers->get_texture_slice(RB_SCOPE_FORWARD_CLUSTERED, RB_TEX_VOXEL_GI_MSAA, p_layer, 0); } + void ensure_fsr2(RendererRD::FSR2Effect *effect); + RendererRD::FSR2Context *get_fsr2_context() const { return fsr2_context; } + RID get_color_only_fb(); RID get_color_pass_fb(uint32_t p_color_pass_flags); RID get_depth_fb(DepthFrameBufferType p_type = DEPTH_FB); RID get_specular_only_fb(); + RID get_velocity_only_fb(); virtual void configure(RenderSceneBuffersRD *p_render_buffers) override; virtual void free_data() override; @@ -346,7 +352,7 @@ class RenderForwardClustered : public RendererSceneRenderRD { static RenderForwardClustered *singleton; - void _setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers = false, bool p_pancake_shadows = false, int p_index = 0); + void _setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers = false, bool p_apply_alpha_multiplier = false, bool p_pancake_shadows = false, int p_index = 0); void _setup_voxelgis(const PagedArray &p_voxelgis); void _setup_lightmaps(const RenderDataRD *p_render_data, const PagedArray &p_lightmaps, const Transform3D &p_cam_transform); @@ -564,6 +570,7 @@ class RenderForwardClustered : public RendererSceneRenderRD { RendererRD::Resolve *resolve_effects = nullptr; RendererRD::TAA *taa = nullptr; + RendererRD::FSR2Effect *fsr2_effect = nullptr; RendererRD::SSEffects *ss_effects = nullptr; /* Cluster builder */ diff --git a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp index 3f01ffa3c96c..9a4aed5fe005 100644 --- a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp +++ b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp @@ -823,9 +823,9 @@ void RenderForwardMobile::_render_scene(RenderDataRD *p_render_data, const Color correction.set_depth_correction(true); Projection projection = correction * p_render_data->scene_data->cam_projection; - sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, this); + sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, Vector2(0.0f, 0.0f), this); } else { - sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, this); + sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, p_render_data->scene_data->taa_jitter, this); } sky_energy_multiplier *= bg_energy_multiplier; @@ -1906,7 +1906,7 @@ void RenderForwardMobile::_setup_environment(const RenderDataRD *p_render_data, } } - p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers); + p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers, false); } void RenderForwardMobile::_fill_element_info(RenderListType p_render_list, uint32_t p_offset, int32_t p_max_elements) { diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp index 20e24dba0eb3..dea7e8c819f9 100644 --- a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp +++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp @@ -340,14 +340,16 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende // Glow, auto exposure and DoF (if enabled). - Size2i internal_size = rb->get_internal_size(); Size2i target_size = rb->get_target_size(); - bool can_use_effects = target_size.x >= 8 && target_size.y >= 8; // FIXME I think this should check internal size, we do all our post processing at this size... bool can_use_storage = _render_buffers_can_be_storage(); + bool use_fsr = fsr && can_use_effects && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR; + bool use_upscaled_texture = rb->has_upscaled_texture() && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR2; + RID render_target = rb->get_render_target(); - RID internal_texture = rb->get_internal_texture(); + RID color_texture = use_upscaled_texture ? rb->get_upscaled_texture() : rb->get_internal_texture(); + Size2i color_size = use_upscaled_texture ? target_size : rb->get_internal_size(); if (can_use_effects && RSG::camera_attributes->camera_attributes_uses_dof(p_render_data->camera_attributes)) { RENDER_TIMESTAMP("Depth of Field"); @@ -416,7 +418,7 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende double step = RSG::camera_attributes->camera_attributes_get_auto_exposure_adjust_speed(p_render_data->camera_attributes) * time_step; float auto_exposure_min_sensitivity = RSG::camera_attributes->camera_attributes_get_auto_exposure_min_sensitivity(p_render_data->camera_attributes); float auto_exposure_max_sensitivity = RSG::camera_attributes->camera_attributes_get_auto_exposure_max_sensitivity(p_render_data->camera_attributes); - luminance->luminance_reduction(internal_texture, internal_size, luminance_buffers, auto_exposure_min_sensitivity, auto_exposure_max_sensitivity, step, set_immediate); + luminance->luminance_reduction(color_texture, color_size, luminance_buffers, auto_exposure_min_sensitivity, auto_exposure_max_sensitivity, step, set_immediate); // Swap final reduce with prev luminance. @@ -525,7 +527,7 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende } tonemap.use_debanding = rb->get_use_debanding(); - tonemap.texture_size = Vector2i(rb->get_internal_size().x, rb->get_internal_size().y); + tonemap.texture_size = Vector2i(color_size.x, color_size.y); if (p_render_data->environment.is_valid()) { tonemap.tonemap_mode = environment_get_tone_mapper(p_render_data->environment); @@ -555,7 +557,8 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende tonemap.convert_to_srgb = !texture_storage->render_target_is_using_hdr(render_target); RID dest_fb; - if (fsr && can_use_effects && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR) { + bool use_intermediate_fb = use_fsr; + if (use_intermediate_fb) { // If we use FSR to upscale we need to write our result into an intermediate buffer. // Note that this is cached so we only create the texture the first time. RID dest_texture = rb->create_texture(SNAME("Tonemapper"), SNAME("destination"), _render_buffers_get_color_format(), RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT); @@ -567,12 +570,12 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende dest_fb = texture_storage->render_target_get_rd_framebuffer(render_target); } - tone_mapper->tonemapper(internal_texture, dest_fb, tonemap); + tone_mapper->tonemapper(color_texture, dest_fb, tonemap); RD::get_singleton()->draw_command_end_label(); } - if (fsr && can_use_effects && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR) { + if (use_fsr) { RD::get_singleton()->draw_command_begin_label("FSR 1.0 Upscale"); for (uint32_t v = 0; v < rb->get_view_count(); v++) { @@ -732,6 +735,11 @@ void RendererSceneRenderRD::_render_buffers_debug_draw(const RenderDataRD *p_ren } } + if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER) { + Size2 rtsize = texture_storage->render_target_get_size(render_target); + copy_effects->copy_to_fb_rect(rb->get_internal_texture(), texture_storage->render_target_get_rd_framebuffer(render_target), Rect2(Vector2(), rtsize), false, false); + } + if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_NORMAL_BUFFER && _render_buffers_get_normal_texture(rb).is_valid()) { Size2 rtsize = texture_storage->render_target_get_size(render_target); copy_effects->copy_to_fb_rect(_render_buffers_get_normal_texture(rb), texture_storage->render_target_get_rd_framebuffer(render_target), Rect2(Vector2(), rtsize), false, false); @@ -745,7 +753,12 @@ void RendererSceneRenderRD::_render_buffers_debug_draw(const RenderDataRD *p_ren } if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS && _render_buffers_get_velocity_texture(rb).is_valid()) { - debug_effects->draw_motion_vectors(_render_buffers_get_velocity_texture(rb), texture_storage->render_target_get_rd_framebuffer(render_target), rb->get_internal_size()); + RID velocity = _render_buffers_get_velocity_texture(rb); + RID depth = rb->get_depth_texture(); + RID dest_fb = texture_storage->render_target_get_rd_framebuffer(render_target); + Size2i resolution = rb->get_internal_size(); + + debug_effects->draw_motion_vectors(velocity, depth, dest_fb, p_render_data->scene_data->cam_projection, p_render_data->scene_data->cam_transform, p_render_data->scene_data->prev_cam_projection, p_render_data->scene_data->prev_cam_transform, resolution); } } diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_accumulate_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_accumulate_pass.glsl new file mode 100644 index 000000000000..f76c921cb406 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_accumulate_pass.glsl @@ -0,0 +1,8 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "motion_vector_inc.glsl" +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_autogen_reactive_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_autogen_reactive_pass.glsl new file mode 100644 index 000000000000..647121fd6f9c --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_autogen_reactive_pass.glsl @@ -0,0 +1,8 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "motion_vector_inc.glsl" +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_compute_luminance_pyramid_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_compute_luminance_pyramid_pass.glsl new file mode 100644 index 000000000000..37504c2e530b --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_compute_luminance_pyramid_pass.glsl @@ -0,0 +1,7 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_depth_clip_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_depth_clip_pass.glsl new file mode 100644 index 000000000000..0145a9f120d5 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_depth_clip_pass.glsl @@ -0,0 +1,8 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "motion_vector_inc.glsl" +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_lock_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_lock_pass.glsl new file mode 100644 index 000000000000..8c8430d4b1c3 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_lock_pass.glsl @@ -0,0 +1,7 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_rcas_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_rcas_pass.glsl new file mode 100644 index 000000000000..4120cfe64495 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_rcas_pass.glsl @@ -0,0 +1,7 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_reconstruct_previous_depth_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_reconstruct_previous_depth_pass.glsl new file mode 100644 index 000000000000..81a148782a93 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_reconstruct_previous_depth_pass.glsl @@ -0,0 +1,8 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "motion_vector_inc.glsl" +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr2_tcr_autogen_pass.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr2_tcr_autogen_pass.glsl new file mode 100644 index 000000000000..1dede3adcae5 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr2_tcr_autogen_pass.glsl @@ -0,0 +1,8 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#include "motion_vector_inc.glsl" +#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl" diff --git a/servers/rendering/renderer_rd/shaders/effects/motion_vector_inc.glsl b/servers/rendering/renderer_rd/shaders/effects/motion_vector_inc.glsl new file mode 100644 index 000000000000..cbf202653e78 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/motion_vector_inc.glsl @@ -0,0 +1,6 @@ +vec2 derive_motion_vector(vec2 uv, float depth, mat4 reprojection_matrix) { + vec4 previous_pos_ndc = reprojection_matrix * vec4(uv * 2.0f - 1.0f, depth * 2.0f - 1.0f, 1.0f); + return 0.5f + (previous_pos_ndc.xy / previous_pos_ndc.w) * 0.5f - uv; +} + +#define FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS_FUNCTION(i, j, k) derive_motion_vector(i, j, k) diff --git a/servers/rendering/renderer_rd/shaders/effects/motion_vectors.glsl b/servers/rendering/renderer_rd/shaders/effects/motion_vectors.glsl index 80e4f51565c0..d02ffe0b4f23 100644 --- a/servers/rendering/renderer_rd/shaders/effects/motion_vectors.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/motion_vectors.glsl @@ -18,14 +18,19 @@ void main() { #VERSION_DEFINES +#include "motion_vector_inc.glsl" + layout(location = 0) in vec2 uv_interp; layout(set = 0, binding = 0) uniform sampler2D source_velocity; +layout(set = 0, binding = 1) uniform sampler2D source_depth; layout(location = 0) out vec4 frag_color; layout(push_constant, std430) uniform Params { + highp mat4 reprojection_matrix; vec2 resolution; + bool force_derive_from_depth; } params; @@ -49,7 +54,14 @@ void main() { vec2 pos_pixel = uv_interp * params.resolution; vec2 cell_pos_pixel = floor(pos_pixel / cell_size) * cell_size + (cell_size * 0.5f); vec2 cell_pos_uv = cell_pos_pixel / params.resolution; - vec2 cell_pos_previous_uv = cell_pos_uv + textureLod(source_velocity, cell_pos_uv, 0.0f).xy; + vec2 cell_pos_velocity = textureLod(source_velocity, cell_pos_uv, 0.0f).xy; + bool derive_velocity = params.force_derive_from_depth || all(lessThanEqual(cell_pos_velocity, vec2(-1.0f, -1.0f))); + if (derive_velocity) { + float depth = textureLod(source_depth, cell_pos_uv, 0.0f).x; + cell_pos_velocity = derive_motion_vector(cell_pos_uv, depth, params.reprojection_matrix); + } + + vec2 cell_pos_previous_uv = cell_pos_uv + cell_pos_velocity; // Draw the shapes. float epsilon = 1e-6f; @@ -76,5 +88,10 @@ void main() { alpha = 0.0f; } + if (derive_velocity) { + color = vec3(1.0f, 1.0f, 1.0f) - color; + alpha *= 0.5f; + } + frag_color = vec4(color, alpha); } diff --git a/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl b/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl index 8f3c704afeb7..b4f05ea75123 100644 --- a/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl +++ b/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl @@ -2273,6 +2273,8 @@ void fragment_shader(in SceneData scene_data) { #else //MODE_SEPARATE_SPECULAR + alpha *= scene_data.pass_alpha_multiplier; + #ifdef MODE_UNSHADED frag_color = vec4(albedo, alpha); #else diff --git a/servers/rendering/renderer_rd/shaders/scene_data_inc.glsl b/servers/rendering/renderer_rd/shaders/scene_data_inc.glsl index b57ee1852175..f42fafc68a24 100644 --- a/servers/rendering/renderer_rd/shaders/scene_data_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_data_inc.glsl @@ -64,6 +64,6 @@ struct SceneData { bool pancake_shadows; uint camera_visible_layers; - uint pad2; + float pass_alpha_multiplier; uint pad3; }; diff --git a/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.cpp b/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.cpp index 71bc21d5d489..80e5b57fb571 100644 --- a/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.cpp +++ b/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.cpp @@ -130,6 +130,11 @@ void RenderSceneBuffersRD::configure(const RenderSceneBuffersConfiguration *p_co texture_mipmap_bias -= 0.5; } + if (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2) { + // Same reasoning as TAA, except AMD recommends this bias instead. + texture_mipmap_bias -= 1.0; + } + if (screen_space_aa == RS::VIEWPORT_SCREEN_SPACE_AA_FXAA) { // Use negative mipmap LOD bias when FXAA is enabled to compensate for loss of sharpness. // If both TAA and FXAA are enabled, combine their negative LOD biases together. @@ -387,6 +392,10 @@ const RD::TextureFormat RenderSceneBuffersRD::get_texture_format(const StringNam } RID RenderSceneBuffersRD::get_texture_slice(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers, const uint32_t p_mipmaps) { + return get_texture_slice_view(p_context, p_texture_name, p_layer, p_mipmap, p_layers, p_mipmaps, RD::TextureView()); +} + +RID RenderSceneBuffersRD::get_texture_slice_view(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers, const uint32_t p_mipmaps, RD::TextureView p_view) { NTKey key(p_context, p_texture_name); // check if this is a known texture @@ -403,19 +412,20 @@ RID RenderSceneBuffersRD::get_texture_slice(const StringName &p_context, const S ERR_FAIL_COND_V(p_mipmap + p_mipmaps > named_texture.format.mipmaps, RID()); // asking the whole thing? just return the original - if (p_layer == 0 && p_mipmap == 0 && named_texture.format.array_layers == p_layers && named_texture.format.mipmaps == p_mipmaps) { + RD::TextureView default_view = RD::TextureView(); + if (p_layer == 0 && p_mipmap == 0 && named_texture.format.array_layers == p_layers && named_texture.format.mipmaps == p_mipmaps && p_view == default_view) { return named_texture.texture; } // see if we have this - NTSliceKey slice_key(p_layer, p_layers, p_mipmap, p_mipmaps); + NTSliceKey slice_key(p_layer, p_layers, p_mipmap, p_mipmaps, p_view); if (named_texture.slices.has(slice_key)) { return named_texture.slices[slice_key]; } // create our slice RID &slice = named_texture.slices[slice_key]; - slice = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), named_texture.texture, p_layer, p_mipmap, p_mipmaps, p_layers > 1 ? RD::TEXTURE_SLICE_2D_ARRAY : RD::TEXTURE_SLICE_2D, p_layers); + slice = RD::get_singleton()->texture_create_shared_from_slice(p_view, named_texture.texture, p_layer, p_mipmap, p_mipmaps, p_layers > 1 ? RD::TEXTURE_SLICE_2D_ARRAY : RD::TEXTURE_SLICE_2D, p_layers); Array arr; arr.push_back(p_context); @@ -424,7 +434,12 @@ RID RenderSceneBuffersRD::get_texture_slice(const StringName &p_context, const S arr.push_back(itos(p_layers)); arr.push_back(itos(p_mipmap)); arr.push_back(itos(p_mipmaps)); - RD::get_singleton()->set_resource_name(slice, String("RenderBuffer {0}/{1}, layer {2}/{3}, mipmap {4}/{5}").format(arr)); + arr.push_back(itos(p_view.format_override)); + arr.push_back(itos(p_view.swizzle_r)); + arr.push_back(itos(p_view.swizzle_g)); + arr.push_back(itos(p_view.swizzle_b)); + arr.push_back(itos(p_view.swizzle_a)); + RD::get_singleton()->set_resource_name(slice, String("RenderBuffer {0}/{1}, layer {2}/{3}, mipmap {4}/{5}, view {6}/{7}/{8}/{9}/{10}").format(arr)); // and return our slice return slice; @@ -593,6 +608,16 @@ RID RenderSceneBuffersRD::get_depth_texture(const uint32_t p_layer) { } } +// Upscaled texture. + +void RenderSceneBuffersRD::ensure_upscaled() { + if (!has_upscaled_texture()) { + uint32_t usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | (can_be_storage ? RD::TEXTURE_USAGE_STORAGE_BIT : 0) | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT; + usage_bits |= RD::TEXTURE_USAGE_INPUT_ATTACHMENT_BIT; + create_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, target_size); + } +} + // Velocity texture. void RenderSceneBuffersRD::ensure_velocity() { diff --git a/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.h b/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.h index 85140c674c00..d324ff7b5b98 100644 --- a/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.h +++ b/servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.h @@ -31,6 +31,7 @@ #ifndef RENDER_SCENE_BUFFERS_RD_H #define RENDER_SCENE_BUFFERS_RD_H +#include "../effects/fsr2.h" #include "../effects/vrs.h" #include "../framebuffer_cache_rd.h" #include "core/templates/hash_map.h" @@ -46,6 +47,7 @@ #define RB_TEXTURE SNAME("texture") #define RB_TEX_COLOR SNAME("color") #define RB_TEX_COLOR_MSAA SNAME("color_msaa") +#define RB_TEX_COLOR_UPSCALED SNAME("color_upscaled") #define RB_TEX_DEPTH SNAME("depth") #define RB_TEX_DEPTH_MSAA SNAME("depth_msaa") #define RB_TEX_VELOCITY SNAME("velocity") @@ -113,9 +115,10 @@ class RenderSceneBuffersRD : public RenderSceneBuffers { uint32_t layers; uint32_t mipmap; uint32_t mipmaps; + RD::TextureView texture_view; bool operator==(const NTSliceKey &p_val) const { - return (layer == p_val.layer) && (layers == p_val.layers) && (mipmap == p_val.mipmap) && (mipmaps == p_val.mipmaps); + return (layer == p_val.layer) && (layers == p_val.layers) && (mipmap == p_val.mipmap) && (mipmaps == p_val.mipmaps) && (texture_view == p_val.texture_view); } static uint32_t hash(const NTSliceKey &p_val) { @@ -123,15 +126,21 @@ class RenderSceneBuffersRD : public RenderSceneBuffers { h = hash_murmur3_one_32(p_val.layers, h); h = hash_murmur3_one_32(p_val.mipmap, h); h = hash_murmur3_one_32(p_val.mipmaps, h); + h = hash_murmur3_one_32(p_val.texture_view.format_override); + h = hash_murmur3_one_32(p_val.texture_view.swizzle_r, h); + h = hash_murmur3_one_32(p_val.texture_view.swizzle_g, h); + h = hash_murmur3_one_32(p_val.texture_view.swizzle_b, h); + h = hash_murmur3_one_32(p_val.texture_view.swizzle_a, h); return hash_fmix32(h); } NTSliceKey() {} - NTSliceKey(uint32_t p_layer, uint32_t p_layers, uint32_t p_mipmap, uint32_t p_mipmaps) { + NTSliceKey(uint32_t p_layer, uint32_t p_layers, uint32_t p_mipmap, uint32_t p_mipmaps, RD::TextureView p_texture_view) { layer = p_layer; layers = p_layers; mipmap = p_mipmap; mipmaps = p_mipmaps; + texture_view = p_texture_view; } }; @@ -184,6 +193,7 @@ class RenderSceneBuffersRD : public RenderSceneBuffers { RID get_texture(const StringName &p_context, const StringName &p_texture_name) const; const RD::TextureFormat get_texture_format(const StringName &p_context, const StringName &p_texture_name) const; RID get_texture_slice(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers = 1, const uint32_t p_mipmaps = 1); + RID get_texture_slice_view(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers = 1, const uint32_t p_mipmaps = 1, RD::TextureView p_view = RD::TextureView()); Size2i get_texture_slice_size(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_mipmap); void clear_context(const StringName &p_context); @@ -224,6 +234,14 @@ class RenderSceneBuffersRD : public RenderSceneBuffers { _FORCE_INLINE_ RID get_internal_texture(const uint32_t p_layer) { return get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_COLOR, p_layer, 0); } + _FORCE_INLINE_ RID get_internal_texture_reactive(const uint32_t p_layer) { + RD::TextureView alpha_only_view; + alpha_only_view.swizzle_r = RD::TEXTURE_SWIZZLE_A; + alpha_only_view.swizzle_g = RD::TEXTURE_SWIZZLE_A; + alpha_only_view.swizzle_b = RD::TEXTURE_SWIZZLE_A; + alpha_only_view.swizzle_a = RD::TEXTURE_SWIZZLE_A; + return get_texture_slice_view(RB_SCOPE_BUFFERS, RB_TEX_COLOR, p_layer, 0, 1, 1, alpha_only_view); + } _FORCE_INLINE_ RID get_color_msaa() const { return get_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_MSAA); } @@ -245,6 +263,19 @@ class RenderSceneBuffersRD : public RenderSceneBuffers { // back buffer (color) RID get_back_buffer_texture() const { return has_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0) ? get_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0) : RID(); } // We (re)use our blur texture here. + // Upscaled. + void ensure_upscaled(); + + _FORCE_INLINE_ bool has_upscaled_texture() const { + return has_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED); + } + _FORCE_INLINE_ RID get_upscaled_texture() const { + return get_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED); + } + _FORCE_INLINE_ RID get_upscaled_texture(const uint32_t p_layer) { + return get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED, p_layer, 0); + } + // Velocity, currently only used by TAA (Clustered) but we'll be using this in other places soon too. void ensure_velocity(); diff --git a/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.cpp b/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.cpp index 27c435eeba77..40891f9a6312 100644 --- a/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.cpp +++ b/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.cpp @@ -38,7 +38,7 @@ RID RenderSceneDataRD::create_uniform_buffer() { return RD::get_singleton()->uniform_buffer_create(sizeof(UBODATA)); } -void RenderSceneDataRD::update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers) { +void RenderSceneDataRD::update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers, bool p_apply_alpha_multiplier) { RendererSceneRenderRD *render_scene_render = RendererSceneRenderRD::get_singleton(); UBODATA ubo_data; @@ -89,6 +89,7 @@ void RenderSceneDataRD::update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p RendererRD::MaterialStorage::store_soft_shadow_kernel(render_scene_render->penumbra_shadow_kernel_get(), ubo.penumbra_shadow_kernel); RendererRD::MaterialStorage::store_soft_shadow_kernel(render_scene_render->soft_shadow_kernel_get(), ubo.soft_shadow_kernel); ubo.camera_visible_layers = camera_visible_layers; + ubo.pass_alpha_multiplier = p_opaque_render_buffers && p_apply_alpha_multiplier ? 0.0f : 1.0f; ubo.viewport_size[0] = p_screen_size.x; ubo.viewport_size[1] = p_screen_size.y; diff --git a/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.h b/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.h index 7546998a9b5b..f183207b5714 100644 --- a/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.h +++ b/servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.h @@ -77,7 +77,7 @@ class RenderSceneDataRD { float time_step; RID create_uniform_buffer(); - void update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers); + void update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers, bool p_apply_alpha_multiplier); RID get_uniform_buffer(); private: @@ -144,7 +144,7 @@ class RenderSceneDataRD { uint32_t pancake_shadows; uint32_t camera_visible_layers; - uint32_t pad2; + float pass_alpha_multiplier; uint32_t pad3; }; diff --git a/servers/rendering/renderer_scene_cull.cpp b/servers/rendering/renderer_scene_cull.cpp index 45bbcf51c4d1..3629463c01d3 100644 --- a/servers/rendering/renderer_scene_cull.cpp +++ b/servers/rendering/renderer_scene_cull.cpp @@ -37,6 +37,19 @@ #include +/* HALTON SEQUENCE */ + +float get_halton_value(int index, int base) { + float f = 1; + float r = 0; + while (index > 0) { + f = f / static_cast(base); + r = r + f * (index % base); + index = index / base; + } + return r * 2.0f - 1.0f; +}; + /* CAMERA API */ RID RendererSceneCull::camera_allocate() { @@ -2498,15 +2511,26 @@ bool RendererSceneCull::_light_instance_update_shadow(Instance *p_instance, cons return animated_material_found; } -void RendererSceneCull::render_camera(const Ref &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, bool p_use_taa, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref &p_xr_interface, RenderInfo *r_render_info) { +void RendererSceneCull::render_camera(const Ref &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, uint32_t p_jitter_phase_count, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref &p_xr_interface, RenderInfo *r_render_info) { #ifndef _3D_DISABLED Camera *camera = camera_owner.get_or_null(p_camera); ERR_FAIL_COND(!camera); Vector2 jitter; - if (p_use_taa) { - jitter = taa_jitter_array[RSG::rasterizer->get_frame_number() % TAA_JITTER_COUNT] / p_viewport_size; + if (p_jitter_phase_count > 0) { + uint32_t current_jitter_count = camera_jitter_array.size(); + if (p_jitter_phase_count != current_jitter_count) { + // Resize the jitter array and fill it with the pre-computed Halton sequence. + camera_jitter_array.resize(p_jitter_phase_count); + + for (uint32_t i = current_jitter_count; i < p_jitter_phase_count; i++) { + camera_jitter_array[i].x = get_halton_value(i, 2); + camera_jitter_array[i].y = get_halton_value(i, 3); + } + } + + jitter = camera_jitter_array[RSG::rasterizer->get_frame_number() % p_jitter_phase_count] / p_viewport_size; } RendererSceneRender::CameraData camera_data; @@ -4113,17 +4137,6 @@ void RendererSceneCull::set_scene_render(RendererSceneRender *p_scene_render) { geometry_instance_pair_mask = scene_render->geometry_instance_get_pair_mask(); } -float get_halton_value(int index, int base) { - float f = 1; - float r = 0; - while (index > 0) { - f = f / static_cast(base); - r = r + f * (index % base); - index = index / base; - } - return r * 2.0f - 1.0f; -}; - RendererSceneCull::RendererSceneCull() { render_pass = 1; singleton = this; @@ -4148,12 +4161,6 @@ RendererSceneCull::RendererSceneCull() { thread_cull_threshold = GLOBAL_GET("rendering/limits/spatial_indexer/threaded_cull_minimum_instances"); thread_cull_threshold = MAX(thread_cull_threshold, (uint32_t)WorkerThreadPool::get_singleton()->get_thread_count()); //make sure there is at least one thread per CPU - taa_jitter_array.resize(TAA_JITTER_COUNT); - for (int i = 0; i < TAA_JITTER_COUNT; i++) { - taa_jitter_array[i].x = get_halton_value(i, 2); - taa_jitter_array[i].y = get_halton_value(i, 3); - } - dummy_occlusion_culling = memnew(RendererSceneOcclusionCull); } diff --git a/servers/rendering/renderer_scene_cull.h b/servers/rendering/renderer_scene_cull.h index 29e5ea29fd63..e3e20b85029c 100644 --- a/servers/rendering/renderer_scene_cull.h +++ b/servers/rendering/renderer_scene_cull.h @@ -954,8 +954,7 @@ class RendererSceneCull : public RenderingMethod { uint32_t geometry_instance_pair_mask = 0; // used in traditional forward, unnecessary on clustered - const int TAA_JITTER_COUNT = 16; - LocalVector taa_jitter_array; + LocalVector camera_jitter_array; virtual RID instance_allocate(); virtual void instance_initialize(RID p_rid); @@ -1089,7 +1088,7 @@ class RendererSceneCull : public RenderingMethod { void _render_scene(const RendererSceneRender::CameraData *p_camera_data, const Ref &p_render_buffers, RID p_environment, RID p_force_camera_attributes, uint32_t p_visible_layers, RID p_scenario, RID p_viewport, RID p_shadow_atlas, RID p_reflection_probe, int p_reflection_probe_pass, float p_screen_mesh_lod_threshold, bool p_using_shadows = true, RenderInfo *r_render_info = nullptr); void render_empty_scene(const Ref &p_render_buffers, RID p_scenario, RID p_shadow_atlas); - void render_camera(const Ref &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, bool p_use_taa, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref &p_xr_interface, RenderingMethod::RenderInfo *r_render_info = nullptr); + void render_camera(const Ref &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, uint32_t p_jitter_phase_count, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref &p_xr_interface, RenderingMethod::RenderInfo *r_render_info = nullptr); void update_dirty_instances(); void render_particle_colliders(); diff --git a/servers/rendering/renderer_viewport.cpp b/servers/rendering/renderer_viewport.cpp index 5880bf395123..a8772320108c 100644 --- a/servers/rendering/renderer_viewport.cpp +++ b/servers/rendering/renderer_viewport.cpp @@ -118,28 +118,33 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) { } else { float scaling_3d_scale = p_viewport->scaling_3d_scale; RS::ViewportScaling3DMode scaling_3d_mode = p_viewport->scaling_3d_mode; + bool scaling_3d_is_fsr = (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR) || (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2); - if ((scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR) && (scaling_3d_scale > 1.0)) { + if (scaling_3d_is_fsr && (scaling_3d_scale > 1.0)) { // FSR is not designed for downsampling. // Fall back to bilinear scaling. + WARN_PRINT_ONCE("FSR 3D resolution scaling is not designed for downsampling. Falling back to bilinear 3D resolution scaling."); scaling_3d_mode = RS::VIEWPORT_SCALING_3D_MODE_BILINEAR; } - if ((scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR) && !p_viewport->fsr_enabled) { + bool upscaler_available = p_viewport->fsr_enabled; + if (scaling_3d_is_fsr && !upscaler_available) { // FSR is not actually available. // Fall back to bilinear scaling. - WARN_PRINT_ONCE("FSR 1.0 3D resolution scaling is not available. Falling back to bilinear 3D resolution scaling."); + WARN_PRINT_ONCE("FSR 3D resolution scaling is not available. Falling back to bilinear 3D resolution scaling."); scaling_3d_mode = RS::VIEWPORT_SCALING_3D_MODE_BILINEAR; } - if (scaling_3d_scale == 1.0) { - scaling_3d_mode = RS::VIEWPORT_SCALING_3D_MODE_OFF; - } - int width; int height; int render_width; int render_height; + uint32_t jitter_phase_count = 0; + + if (p_viewport->use_taa) { + // Default jitter count for TAA. + jitter_phase_count = 16; + } switch (scaling_3d_mode) { case RS::VIEWPORT_SCALING_3D_MODE_BILINEAR: @@ -151,10 +156,14 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) { render_height = height; break; case RS::VIEWPORT_SCALING_3D_MODE_FSR: + case RS::VIEWPORT_SCALING_3D_MODE_FSR2: width = p_viewport->size.width; height = p_viewport->size.height; render_width = MAX(width * scaling_3d_scale, 1.0); // width / (width * scaling) render_height = MAX(height * scaling_3d_scale, 1.0); + + // Implementation has been copied from ffxFsr2GetJitterPhaseCount. + jitter_phase_count = uint32_t(8.0f * pow(float(width) / render_width, 2.0f)); break; case RS::VIEWPORT_SCALING_3D_MODE_OFF: width = p_viewport->size.width; @@ -175,6 +184,7 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) { } p_viewport->internal_size = Size2(render_width, render_height); + p_viewport->jitter_phase_count = jitter_phase_count; // At resolution scales lower than 1.0, use negative texture mipmap bias // to compensate for the loss of sharpness. @@ -221,7 +231,7 @@ void RendererViewport::_draw_3d(Viewport *p_viewport) { } float screen_mesh_lod_threshold = p_viewport->mesh_lod_threshold / float(p_viewport->size.width); - RSG::scene->render_camera(p_viewport->render_buffers, p_viewport->camera, p_viewport->scenario, p_viewport->self, p_viewport->internal_size, p_viewport->use_taa, screen_mesh_lod_threshold, p_viewport->shadow_atlas, xr_interface, &p_viewport->render_info); + RSG::scene->render_camera(p_viewport->render_buffers, p_viewport->camera, p_viewport->scenario, p_viewport->self, p_viewport->internal_size, p_viewport->jitter_phase_count, screen_mesh_lod_threshold, p_viewport->shadow_atlas, xr_interface, &p_viewport->render_info); RENDER_TIMESTAMP("< Render 3D Scene"); } @@ -825,8 +835,20 @@ void RendererViewport::viewport_set_use_xr(RID p_viewport, bool p_use_xr) { void RendererViewport::viewport_set_scaling_3d_mode(RID p_viewport, RS::ViewportScaling3DMode p_mode) { Viewport *viewport = viewport_owner.get_or_null(p_viewport); ERR_FAIL_COND(!viewport); + ERR_FAIL_COND_EDMSG(p_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2 && OS::get_singleton()->get_current_rendering_method() != "forward_plus", "FSR2 is only available when using the Forward+ renderer."); + if (viewport->scaling_3d_mode == p_mode) { + return; + } + + bool motion_vectors_before = _viewport_requires_motion_vectors(viewport); viewport->scaling_3d_mode = p_mode; + + bool motion_vectors_after = _viewport_requires_motion_vectors(viewport); + if (motion_vectors_before != motion_vectors_after) { + num_viewports_with_motion_vectors += motion_vectors_after ? 1 : -1; + } + _configure_3d_render_buffers(viewport); } @@ -888,6 +910,10 @@ void RendererViewport::_viewport_set_size(Viewport *p_viewport, int p_width, int } } +bool RendererViewport::_viewport_requires_motion_vectors(Viewport *p_viewport) { + return p_viewport->use_taa || p_viewport->scaling_3d_mode == RenderingServer::VIEWPORT_SCALING_3D_MODE_FSR2; +} + void RendererViewport::viewport_set_active(RID p_viewport, bool p_active) { Viewport *viewport = viewport_owner.get_or_null(p_viewport); ERR_FAIL_COND(!viewport); @@ -1193,8 +1219,15 @@ void RendererViewport::viewport_set_use_taa(RID p_viewport, bool p_use_taa) { if (viewport->use_taa == p_use_taa) { return; } + + bool motion_vectors_before = _viewport_requires_motion_vectors(viewport); viewport->use_taa = p_use_taa; - num_viewports_with_motion_vectors += p_use_taa ? 1 : -1; + + bool motion_vectors_after = _viewport_requires_motion_vectors(viewport); + if (motion_vectors_before != motion_vectors_after) { + num_viewports_with_motion_vectors += motion_vectors_after ? 1 : -1; + } + _configure_3d_render_buffers(viewport); } @@ -1379,7 +1412,7 @@ bool RendererViewport::free(RID p_rid) { RendererSceneOcclusionCull::get_singleton()->remove_buffer(p_rid); } - if (viewport->use_taa) { + if (_viewport_requires_motion_vectors(viewport)) { num_viewports_with_motion_vectors--; } diff --git a/servers/rendering/renderer_viewport.h b/servers/rendering/renderer_viewport.h index 3bfb1afd51c0..44de6d880420 100644 --- a/servers/rendering/renderer_viewport.h +++ b/servers/rendering/renderer_viewport.h @@ -63,6 +63,7 @@ class RendererViewport { float fsr_sharpness = 0.2f; float texture_mipmap_bias = 0.0f; bool fsr_enabled = false; + uint32_t jitter_phase_count = 0; RS::ViewportUpdateMode update_mode = RenderingServer::VIEWPORT_UPDATE_WHEN_VISIBLE; RID render_target; RID render_target_texture; @@ -203,6 +204,7 @@ class RendererViewport { private: Vector _sort_active_viewports(); void _viewport_set_size(Viewport *p_viewport, int p_width, int p_height, uint32_t p_view_count); + bool _viewport_requires_motion_vectors(Viewport *p_viewport); void _configure_3d_render_buffers(Viewport *p_viewport); void _draw_3d(Viewport *p_viewport); void _draw_viewport(Viewport *p_viewport); diff --git a/servers/rendering/rendering_device.h b/servers/rendering/rendering_device.h index 1ade1b25c4d5..90c9b0c88c59 100644 --- a/servers/rendering/rendering_device.h +++ b/servers/rendering/rendering_device.h @@ -518,6 +518,22 @@ class RenderingDevice : public Object { TextureSwizzle swizzle_b; TextureSwizzle swizzle_a; + bool operator==(const TextureView &p_view) const { + if (format_override != p_view.format_override) { + return false; + } else if (swizzle_r != p_view.swizzle_r) { + return false; + } else if (swizzle_g != p_view.swizzle_g) { + return false; + } else if (swizzle_b != p_view.swizzle_b) { + return false; + } else if (swizzle_a != p_view.swizzle_a) { + return false; + } else { + return true; + } + } + TextureView() { format_override = DATA_FORMAT_MAX; //means, use same as format swizzle_r = TEXTURE_SWIZZLE_R; @@ -1270,6 +1286,8 @@ class RenderingDevice : public Object { LIMIT_MAX_VIEWPORT_DIMENSIONS_X, LIMIT_MAX_VIEWPORT_DIMENSIONS_Y, LIMIT_SUBGROUP_SIZE, + LIMIT_SUBGROUP_MIN_SIZE, + LIMIT_SUBGROUP_MAX_SIZE, LIMIT_SUBGROUP_IN_SHADERS, // Set flags using SHADER_STAGE_VERTEX_BIT, SHADER_STAGE_FRAGMENT_BIT, etc. LIMIT_SUBGROUP_OPERATIONS, LIMIT_VRS_TEXEL_WIDTH, diff --git a/servers/rendering/rendering_method.h b/servers/rendering/rendering_method.h index f705603a1cbc..d1c6c1cbf902 100644 --- a/servers/rendering/rendering_method.h +++ b/servers/rendering/rendering_method.h @@ -301,7 +301,7 @@ class RenderingMethod { int info[RS::VIEWPORT_RENDER_INFO_TYPE_MAX][RS::VIEWPORT_RENDER_INFO_MAX] = {}; }; - virtual void render_camera(const Ref &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, bool p_use_taa, float p_mesh_lod_threshold, RID p_shadow_atlas, Ref &p_xr_interface, RenderInfo *r_render_info = nullptr) = 0; + virtual void render_camera(const Ref &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, uint32_t p_jitter_phase_count, float p_mesh_lod_threshold, RID p_shadow_atlas, Ref &p_xr_interface, RenderInfo *r_render_info = nullptr) = 0; virtual void update() = 0; virtual void render_probes() = 0; diff --git a/servers/rendering/storage/render_scene_buffers.cpp b/servers/rendering/storage/render_scene_buffers.cpp index 19496c0116d2..96e649285457 100644 --- a/servers/rendering/storage/render_scene_buffers.cpp +++ b/servers/rendering/storage/render_scene_buffers.cpp @@ -49,7 +49,7 @@ void RenderSceneBuffersConfiguration::_bind_methods() { ClassDB::bind_method(D_METHOD("get_scaling_3d_mode"), &RenderSceneBuffersConfiguration::get_scaling_3d_mode); ClassDB::bind_method(D_METHOD("set_scaling_3d_mode", "scaling_3d_mode"), &RenderSceneBuffersConfiguration::set_scaling_3d_mode); - ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast)"), "set_scaling_3d_mode", "get_scaling_3d_mode"); // TODO VIEWPORT_SCALING_3D_MODE_OFF is possible here too, but we can't specify an enum string for it. + ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast),FSR 2.2 (Slow)"), "set_scaling_3d_mode", "get_scaling_3d_mode"); // TODO VIEWPORT_SCALING_3D_MODE_OFF is possible here too, but we can't specify an enum string for it. ClassDB::bind_method(D_METHOD("get_msaa_3d"), &RenderSceneBuffersConfiguration::get_msaa_3d); ClassDB::bind_method(D_METHOD("set_msaa_3d", "msaa_3d"), &RenderSceneBuffersConfiguration::set_msaa_3d); diff --git a/servers/rendering_server.cpp b/servers/rendering_server.cpp index 45ba0b3c089b..b7c40600cb3a 100644 --- a/servers/rendering_server.cpp +++ b/servers/rendering_server.cpp @@ -2252,6 +2252,7 @@ void RenderingServer::_bind_methods() { BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_BILINEAR); BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_FSR); + BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_FSR2); BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_MAX); BIND_ENUM_CONSTANT(VIEWPORT_UPDATE_DISABLED); @@ -2329,6 +2330,7 @@ void RenderingServer::_bind_methods() { BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_CLUSTER_REFLECTION_PROBES); BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_OCCLUDERS); BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_MOTION_VECTORS); + BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER); BIND_ENUM_CONSTANT(VIEWPORT_VRS_DISABLED); BIND_ENUM_CONSTANT(VIEWPORT_VRS_TEXTURE); @@ -2959,7 +2961,7 @@ void RenderingServer::init() { GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/anti_aliasing/screen_space_roughness_limiter/amount", PROPERTY_HINT_RANGE, "0.01,4.0,0.01"), 0.25); GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/anti_aliasing/screen_space_roughness_limiter/limit", PROPERTY_HINT_RANGE, "0.01,1.0,0.01"), 0.18); - GLOBAL_DEF(PropertyInfo(Variant::INT, "rendering/scaling_3d/mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast)"), 0); + GLOBAL_DEF(PropertyInfo(Variant::INT, "rendering/scaling_3d/mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast),FSR 2.2 (Slow)"), 0); GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/scaling_3d/scale", PROPERTY_HINT_RANGE, "0.25,2.0,0.01"), 1.0); GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/scaling_3d/fsr_sharpness", PROPERTY_HINT_RANGE, "0,2,0.1"), 0.2f); GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/textures/default_filters/texture_mipmap_bias", PROPERTY_HINT_RANGE, "-2,2,0.001"), 0.0f); diff --git a/servers/rendering_server.h b/servers/rendering_server.h index 1528a957ce45..6b2ba562ce9b 100644 --- a/servers/rendering_server.h +++ b/servers/rendering_server.h @@ -807,6 +807,7 @@ class RenderingServer : public Object { enum ViewportScaling3DMode { VIEWPORT_SCALING_3D_MODE_BILINEAR, VIEWPORT_SCALING_3D_MODE_FSR, + VIEWPORT_SCALING_3D_MODE_FSR2, VIEWPORT_SCALING_3D_MODE_MAX, VIEWPORT_SCALING_3D_MODE_OFF = 255, // for internal use only }; @@ -971,6 +972,7 @@ class RenderingServer : public Object { VIEWPORT_DEBUG_DRAW_CLUSTER_REFLECTION_PROBES, VIEWPORT_DEBUG_DRAW_OCCLUDERS, VIEWPORT_DEBUG_DRAW_MOTION_VECTORS, + VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER, }; virtual void viewport_set_debug_draw(RID p_viewport, ViewportDebugDraw p_draw) = 0; diff --git a/thirdparty/README.md b/thirdparty/README.md index ff8f1d559123..659a899bb66b 100644 --- a/thirdparty/README.md +++ b/thirdparty/README.md @@ -16,6 +16,19 @@ Files extracted from upstream source: - `ffx_a.h` and `ffx_fsr1.h` from `ffx-fsr` - `license.txt` +## amd-fsr2 + +- Upstream: https://github.com/GPUOpen-Effects/FidelityFX-FSR2 +- Version: 2.2.1 (1680d1edd5c034f88ebbbb793d8b88f8842cf804, 2023) +- License: MIT + +Files extracted from upstream source: + +- `ffx_*.cpp` and `ffx_*.h` from `src/ffx-fsr2-api` +- `shaders` folder from `src/ffx-fsr2-api` with `ffx_*.hlsl` files excluded +- `LICENSE.txt` + +Apply `patches` to add the new options required by Godot and general compilation fixes. ## astcenc diff --git a/thirdparty/amd-fsr2/LICENSE.txt b/thirdparty/amd-fsr2/LICENSE.txt new file mode 100644 index 000000000000..c066ae106362 --- /dev/null +++ b/thirdparty/amd-fsr2/LICENSE.txt @@ -0,0 +1,21 @@ +FidelityFX Super Resolution 2.2 +================================= +Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in +all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +THE SOFTWARE. diff --git a/thirdparty/amd-fsr2/ffx_assert.cpp b/thirdparty/amd-fsr2/ffx_assert.cpp new file mode 100644 index 000000000000..8a70ad501a12 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_assert.cpp @@ -0,0 +1,81 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#include "ffx_assert.h" +#include // for malloc() + +#ifdef _WIN32 +#ifndef WIN32_LEAN_AND_MEAN +#define WIN32_LEAN_AND_MEAN +#endif +#include // required for OutputDebugString() +#include // required for sprintf_s +#endif // #ifndef _WIN32 + +static FfxAssertCallback s_assertCallback; + +// set the printing callback function +void ffxAssertSetPrintingCallback(FfxAssertCallback callback) +{ + s_assertCallback = callback; + return; +} + +// implementation of assert reporting +bool ffxAssertReport(const char* file, int32_t line, const char* condition, const char* message) +{ + if (!file) { + + return true; + } + +#ifdef _WIN32 + // form the final assertion string and output to the TTY. + const size_t bufferSize = static_cast(snprintf(nullptr, 0, "%s(%d): ASSERTION FAILED. %s\n", file, line, message ? message : condition)) + 1; + char* tempBuf = static_cast(malloc(bufferSize)); + if (!tempBuf) { + + return true; + } + + if (!message) { + sprintf_s(tempBuf, bufferSize, "%s(%d): ASSERTION FAILED. %s\n", file, line, condition); + } else { + sprintf_s(tempBuf, bufferSize, "%s(%d): ASSERTION FAILED. %s\n", file, line, message); + } + + if (!s_assertCallback) { + OutputDebugStringA(tempBuf); + } else { + s_assertCallback(tempBuf); + } + + // free the buffer. + free(tempBuf); + +#else + FFX_UNUSED(line); + FFX_UNUSED(condition); + FFX_UNUSED(message); +#endif + + return true; +} diff --git a/thirdparty/amd-fsr2/ffx_assert.h b/thirdparty/amd-fsr2/ffx_assert.h new file mode 100644 index 000000000000..ae32d2a73345 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_assert.h @@ -0,0 +1,132 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#pragma once + +#include "ffx_types.h" +#include "ffx_util.h" + +#ifdef __cplusplus +extern "C" { +#endif // #ifdef __cplusplus + +#ifdef _DEBUG +#ifdef _WIN32 + +#ifdef DISABLE_FFX_DEBUG_BREAK +#define FFX_DEBUG_BREAK \ + { \ + } +#else +/// Macro to force the debugger to break at this point in the code. +#define FFX_DEBUG_BREAK __debugbreak(); +#endif +#else +#define FFX_DEBUG_BREAK \ + { \ + } +#endif +#else +// don't allow debug break in release builds. +#define FFX_DEBUG_BREAK +#endif + +/// A typedef for the callback function for assert printing. +/// +/// This can be used to re-route printing of assert messages from the FFX backend +/// to another destination. For example instead of the default behaviour of printing +/// the assert messages to the debugger's TTY the message can be re-routed to a +/// MessageBox in a GUI application. +/// +/// @param [in] message The message generated by the assert. +/// +typedef void (*FfxAssertCallback)(const char* message); + +/// Function to report an assert. +/// +/// @param [in] file The name of the file as a string. +/// @param [in] line The index of the line in the file. +/// @param [in] condition The boolean condition that was tested. +/// @param [in] msg The optional message to print. +/// +/// @returns +/// Always returns true. +/// +FFX_API bool ffxAssertReport(const char* file, int32_t line, const char* condition, const char* msg); + +/// Provides the ability to set a callback for assert messages. +/// +/// @param [in] callback The callback function that will receive assert messages. +/// +FFX_API void ffxAssertSetPrintingCallback(FfxAssertCallback callback); + +#ifdef _DEBUG +/// Standard assert macro. +#define FFX_ASSERT(condition) \ + do \ + { \ + if (!(condition) && ffxAssertReport(__FILE__, __LINE__, #condition, NULL)) \ + FFX_DEBUG_BREAK \ + } while (0) + +/// Assert macro with message. +#define FFX_ASSERT_MESSAGE(condition, msg) \ + do \ + { \ + if (!(condition) && ffxAssertReport(__FILE__, __LINE__, #condition, msg)) \ + FFX_DEBUG_BREAK \ + } while (0) + +/// Assert macro that always fails. +#define FFX_ASSERT_FAIL(message) \ + do \ + { \ + ffxAssertReport(__FILE__, __LINE__, NULL, message); \ + FFX_DEBUG_BREAK \ + } while (0) +#else +// asserts disabled +#define FFX_ASSERT(condition) \ + do \ + { \ + FFX_UNUSED(condition); \ + } while (0) + +#define FFX_ASSERT_MESSAGE(condition, message) \ + do \ + { \ + FFX_UNUSED(condition); \ + FFX_UNUSED(message); \ + } while (0) + +#define FFX_ASSERT_FAIL(message) \ + do \ + { \ + FFX_UNUSED(message); \ + } while (0) +#endif // #if _DEBUG + +/// Simple static assert. +#define FFX_STATIC_ASSERT(condition) static_assert(condition, #condition) + +#ifdef __cplusplus +} +#endif // #ifdef __cplusplus diff --git a/thirdparty/amd-fsr2/ffx_error.h b/thirdparty/amd-fsr2/ffx_error.h new file mode 100644 index 000000000000..7ba7d9c4ea47 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_error.h @@ -0,0 +1,59 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#pragma once + +#include "ffx_types.h" + +/// Typedef for error codes returned from functions in the FidelityFX SDK. +typedef int32_t FfxErrorCode; + +static const FfxErrorCode FFX_OK = 0; ///< The operation completed successfully. +static const FfxErrorCode FFX_ERROR_INVALID_POINTER = 0x80000000; ///< The operation failed due to an invalid pointer. +static const FfxErrorCode FFX_ERROR_INVALID_ALIGNMENT = 0x80000001; ///< The operation failed due to an invalid alignment. +static const FfxErrorCode FFX_ERROR_INVALID_SIZE = 0x80000002; ///< The operation failed due to an invalid size. +static const FfxErrorCode FFX_EOF = 0x80000003; ///< The end of the file was encountered. +static const FfxErrorCode FFX_ERROR_INVALID_PATH = 0x80000004; ///< The operation failed because the specified path was invalid. +static const FfxErrorCode FFX_ERROR_EOF = 0x80000005; ///< The operation failed because end of file was reached. +static const FfxErrorCode FFX_ERROR_MALFORMED_DATA = 0x80000006; ///< The operation failed because of some malformed data. +static const FfxErrorCode FFX_ERROR_OUT_OF_MEMORY = 0x80000007; ///< The operation failed because it ran out memory. +static const FfxErrorCode FFX_ERROR_INCOMPLETE_INTERFACE = 0x80000008; ///< The operation failed because the interface was not fully configured. +static const FfxErrorCode FFX_ERROR_INVALID_ENUM = 0x80000009; ///< The operation failed because of an invalid enumeration value. +static const FfxErrorCode FFX_ERROR_INVALID_ARGUMENT = 0x8000000a; ///< The operation failed because an argument was invalid. +static const FfxErrorCode FFX_ERROR_OUT_OF_RANGE = 0x8000000b; ///< The operation failed because a value was out of range. +static const FfxErrorCode FFX_ERROR_NULL_DEVICE = 0x8000000c; ///< The operation failed because a device was null. +static const FfxErrorCode FFX_ERROR_BACKEND_API_ERROR = 0x8000000d; ///< The operation failed because the backend API returned an error code. +static const FfxErrorCode FFX_ERROR_INSUFFICIENT_MEMORY = 0x8000000e; ///< The operation failed because there was not enough memory. + +/// Helper macro to return error code y from a function when a specific condition, x, is not met. +#define FFX_RETURN_ON_ERROR(x, y) \ + if (!(x)) \ + { \ + return (y); \ + } + +/// Helper macro to return error code x from a function when it is not FFX_OK. +#define FFX_VALIDATE(x) \ + { \ + FfxErrorCode ret = x; \ + FFX_RETURN_ON_ERROR(ret == FFX_OK, ret); \ + } + diff --git a/thirdparty/amd-fsr2/ffx_fsr2.cpp b/thirdparty/amd-fsr2/ffx_fsr2.cpp new file mode 100644 index 000000000000..864f7f129429 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_fsr2.cpp @@ -0,0 +1,1373 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#include // for max used inside SPD CPU code. +#include // for fabs, abs, sinf, sqrt, etc. +#include // for memset +#include // for FLT_EPSILON +#include "ffx_fsr2.h" +#define FFX_CPU +#include "shaders/ffx_core.h" +#include "shaders/ffx_fsr1.h" +#include "shaders/ffx_spd.h" +#include "shaders/ffx_fsr2_callbacks_hlsl.h" + +#include "ffx_fsr2_maximum_bias.h" + +#ifdef __clang__ +#pragma clang diagnostic ignored "-Wunused-variable" +#endif + +// -- GODOT start -- +#ifndef _countof +#define _countof(array) (sizeof(array) / sizeof(array[0])) +#endif + +#ifndef _MSC_VER +#include +#define wcscpy_s wcscpy +#endif +// -- GODOT end -- + +// max queued frames for descriptor management +static const uint32_t FSR2_MAX_QUEUED_FRAMES = 16; + +#include "ffx_fsr2_private.h" + +// lists to map shader resource bindpoint name to resource identifier +typedef struct ResourceBinding +{ + uint32_t index; + wchar_t name[64]; +}ResourceBinding; + +static const ResourceBinding srvResourceBindingTable[] = +{ + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR, L"r_input_color_jittered"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY, L"r_input_opaque_only"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS, L"r_input_motion_vectors"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH, L"r_input_depth" }, + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE, L"r_input_exposure"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE, L"r_auto_exposure"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK, L"r_reactive_mask"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK, L"r_transparency_and_composition_mask"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH, L"r_reconstructed_previous_nearest_depth"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS, L"r_dilated_motion_vectors"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREVIOUS_DILATED_MOTION_VECTORS, L"r_previous_dilated_motion_vectors"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH, L"r_dilatedDepth"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR, L"r_internal_upscaled_color"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS, L"r_lock_status"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR, L"r_prepared_input_color"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY, L"r_luma_history" }, + {FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT, L"r_rcas_input"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LANCZOS_LUT, L"r_lanczos_lut"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE, L"r_imgMips"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE, L"r_img_mip_shading_change"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_5, L"r_img_mip_5"}, + {FFX_FSR2_RESOURCE_IDENTITIER_UPSAMPLE_MAXIMUM_BIAS_LUT, L"r_upsample_maximum_bias_lut"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS, L"r_dilated_reactive_masks"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS, L"r_new_locks"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA, L"r_lock_input_luma"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR, L"r_input_prev_color_pre_alpha"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR, L"r_input_prev_color_post_alpha"}, +}; + +static const ResourceBinding uavResourceBindingTable[] = +{ + {FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH, L"rw_reconstructed_previous_nearest_depth"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS, L"rw_dilated_motion_vectors"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH, L"rw_dilatedDepth"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR, L"rw_internal_upscaled_color"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS, L"rw_lock_status"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR, L"rw_prepared_input_color"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY, L"rw_luma_history"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT, L"rw_upscaled_output"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE, L"rw_img_mip_shading_change"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_5, L"rw_img_mip_5"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS, L"rw_dilated_reactive_masks"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE, L"rw_auto_exposure"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_SPD_ATOMIC_COUNT, L"rw_spd_global_atomic"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS, L"rw_new_locks"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA, L"rw_lock_input_luma"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE, L"rw_output_autoreactive"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION, L"rw_output_autocomposition"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR, L"rw_output_prev_color_pre_alpha"}, + {FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR, L"rw_output_prev_color_post_alpha"}, +}; + +static const ResourceBinding cbResourceBindingTable[] = +{ + {FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_FSR2, L"cbFSR2"}, + {FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_SPD, L"cbSPD"}, + {FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_RCAS, L"cbRCAS"}, + {FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_GENREACTIVE, L"cbGenerateReactive"}, +}; + +// Broad structure of the root signature. +typedef enum Fsr2RootSignatureLayout { + + FSR2_ROOT_SIGNATURE_LAYOUT_UAVS, + FSR2_ROOT_SIGNATURE_LAYOUT_SRVS, + FSR2_ROOT_SIGNATURE_LAYOUT_CONSTANTS, + FSR2_ROOT_SIGNATURE_LAYOUT_CONSTANTS_REGISTER_1, + FSR2_ROOT_SIGNATURE_LAYOUT_PARAMETER_COUNT +} Fsr2RootSignatureLayout; + +typedef struct Fsr2RcasConstants { + + uint32_t rcasConfig[4]; +} FfxRcasConstants; + +typedef struct Fsr2SpdConstants { + + uint32_t mips; + uint32_t numworkGroups; + uint32_t workGroupOffset[2]; + uint32_t renderSize[2]; +} Fsr2SpdConstants; + +typedef struct Fsr2GenerateReactiveConstants +{ + float scale; + float threshold; + float binaryValue; + uint32_t flags; + +} Fsr2GenerateReactiveConstants; + +typedef struct Fsr2GenerateReactiveConstants2 +{ + float autoTcThreshold; + float autoTcScale; + float autoReactiveScale; + float autoReactiveMax; + +} Fsr2GenerateReactiveConstants2; + +typedef union Fsr2SecondaryUnion { + + Fsr2RcasConstants rcas; + Fsr2SpdConstants spd; + Fsr2GenerateReactiveConstants2 autogenReactive; +} Fsr2SecondaryUnion; + +typedef struct Fsr2ResourceDescription { + + uint32_t id; + const wchar_t* name; + FfxResourceUsage usage; + FfxSurfaceFormat format; + uint32_t width; + uint32_t height; + uint32_t mipCount; + FfxResourceFlags flags; + uint32_t initDataSize; + void* initData; +} Fsr2ResourceDescription; + +FfxConstantBuffer globalFsr2ConstantBuffers[4] = { + { sizeof(Fsr2Constants) / sizeof(uint32_t) }, + { sizeof(Fsr2SpdConstants) / sizeof(uint32_t) }, + { sizeof(Fsr2RcasConstants) / sizeof(uint32_t) }, + { sizeof(Fsr2GenerateReactiveConstants) / sizeof(uint32_t) } +}; + +// Lanczos +static float lanczos2(float value) +{ + return abs(value) < FFX_EPSILON ? 1.f : (sinf(FFX_PI * value) / (FFX_PI * value)) * (sinf(0.5f * FFX_PI * value) / (0.5f * FFX_PI * value)); +} + +// Calculate halton number for index and base. +static float halton(int32_t index, int32_t base) +{ + float f = 1.0f, result = 0.0f; + + for (int32_t currentIndex = index; currentIndex > 0;) { + + f /= (float)base; + result = result + f * (float)(currentIndex % base); + currentIndex = (uint32_t)(floorf((float)(currentIndex) / (float)(base))); + } + + return result; +} + +static void fsr2DebugCheckDispatch(FfxFsr2Context_Private* context, const FfxFsr2DispatchDescription* params) +{ + if (params->commandList == nullptr) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"commandList is null"); + } + + if (params->color.resource == nullptr) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"color resource is null"); + } + + if (params->depth.resource == nullptr) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"depth resource is null"); + } + + if (params->motionVectors.resource == nullptr) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"motionVectors resource is null"); + } + + if (params->exposure.resource != nullptr) + { + if ((context->contextDescription.flags & FFX_FSR2_ENABLE_AUTO_EXPOSURE) == FFX_FSR2_ENABLE_AUTO_EXPOSURE) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"exposure resource provided, however auto exposure flag is present"); + } + } + + if (params->output.resource == nullptr) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"output resource is null"); + } + + if (fabs(params->jitterOffset.x) > 1.0f || fabs(params->jitterOffset.y) > 1.0f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"jitterOffset contains value outside of expected range [-1.0, 1.0]"); + } + + if ((params->motionVectorScale.x > (float)context->contextDescription.maxRenderSize.width) || + (params->motionVectorScale.y > (float)context->contextDescription.maxRenderSize.height)) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"motionVectorScale contains scale value greater than maxRenderSize"); + } + if ((params->motionVectorScale.x == 0.0f) || + (params->motionVectorScale.y == 0.0f)) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"motionVectorScale contains zero scale value"); + } + + if ((params->renderSize.width > context->contextDescription.maxRenderSize.width) || + (params->renderSize.height > context->contextDescription.maxRenderSize.height)) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"renderSize is greater than context maxRenderSize"); + } + if ((params->renderSize.width == 0) || + (params->renderSize.height == 0)) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"renderSize contains zero dimension"); + } + + if (params->sharpness < 0.0f || params->sharpness > 1.0f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"sharpness contains value outside of expected range [0.0, 1.0]"); + } + + if (params->frameTimeDelta < 1.0f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, L"frameTimeDelta is less than 1.0f - this value should be milliseconds (~16.6f for 60fps)"); + } + + if (params->preExposure == 0.0f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"preExposure provided as 0.0f which is invalid"); + } + + bool infiniteDepth = (context->contextDescription.flags & FFX_FSR2_ENABLE_DEPTH_INFINITE) == FFX_FSR2_ENABLE_DEPTH_INFINITE; + bool inverseDepth = (context->contextDescription.flags & FFX_FSR2_ENABLE_DEPTH_INVERTED) == FFX_FSR2_ENABLE_DEPTH_INVERTED; + + if (inverseDepth) + { + if (params->cameraNear < params->cameraFar) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, + L"FFX_FSR2_ENABLE_DEPTH_INVERTED flag is present yet cameraNear is less than cameraFar"); + } + if (infiniteDepth) + { + if (params->cameraNear != FLT_MAX) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, + L"FFX_FSR2_ENABLE_DEPTH_INFINITE and FFX_FSR2_ENABLE_DEPTH_INVERTED present, yet cameraNear != FLT_MAX"); + } + } + if (params->cameraFar < 0.075f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, + L"FFX_FSR2_ENABLE_DEPTH_INFINITE and FFX_FSR2_ENABLE_DEPTH_INVERTED present, cameraFar value is very low which may result in depth separation artefacting"); + } + } + else + { + if (params->cameraNear > params->cameraFar) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, + L"cameraNear is greater than cameraFar in non-inverted-depth context"); + } + if (infiniteDepth) + { + if (params->cameraFar != FLT_MAX) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, + L"FFX_FSR2_ENABLE_DEPTH_INFINITE and FFX_FSR2_ENABLE_DEPTH_INVERTED present, yet cameraFar != FLT_MAX"); + } + } + if (params->cameraNear < 0.075f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_WARNING, + L"FFX_FSR2_ENABLE_DEPTH_INFINITE and FFX_FSR2_ENABLE_DEPTH_INVERTED present, cameraNear value is very low which may result in depth separation artefacting"); + } + } + + if (params->cameraFovAngleVertical <= 0.0f) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"cameraFovAngleVertical is 0.0f - this value should be > 0.0f"); + } + if (params->cameraFovAngleVertical > FFX_PI) + { + context->contextDescription.fpMessage(FFX_FSR2_MESSAGE_TYPE_ERROR, L"cameraFovAngleVertical is greater than 180 degrees/PI"); + } +} + +static FfxErrorCode patchResourceBindings(FfxPipelineState* inoutPipeline) +{ + for (uint32_t srvIndex = 0; srvIndex < inoutPipeline->srvCount; ++srvIndex) + { + int32_t mapIndex = 0; + for (mapIndex = 0; mapIndex < _countof(srvResourceBindingTable); ++mapIndex) + { + if (0 == wcscmp(srvResourceBindingTable[mapIndex].name, inoutPipeline->srvResourceBindings[srvIndex].name)) + break; + } + if (mapIndex == _countof(srvResourceBindingTable)) + return FFX_ERROR_INVALID_ARGUMENT; + + inoutPipeline->srvResourceBindings[srvIndex].resourceIdentifier = srvResourceBindingTable[mapIndex].index; + } + + for (uint32_t uavIndex = 0; uavIndex < inoutPipeline->uavCount; ++uavIndex) + { + int32_t mapIndex = 0; + for (mapIndex = 0; mapIndex < _countof(uavResourceBindingTable); ++mapIndex) + { + if (0 == wcscmp(uavResourceBindingTable[mapIndex].name, inoutPipeline->uavResourceBindings[uavIndex].name)) + break; + } + if (mapIndex == _countof(uavResourceBindingTable)) + return FFX_ERROR_INVALID_ARGUMENT; + + inoutPipeline->uavResourceBindings[uavIndex].resourceIdentifier = uavResourceBindingTable[mapIndex].index; + } + + for (uint32_t cbIndex = 0; cbIndex < inoutPipeline->constCount; ++cbIndex) + { + int32_t mapIndex = 0; + for (mapIndex = 0; mapIndex < _countof(cbResourceBindingTable); ++mapIndex) + { + if (0 == wcscmp(cbResourceBindingTable[mapIndex].name, inoutPipeline->cbResourceBindings[cbIndex].name)) + break; + } + if (mapIndex == _countof(cbResourceBindingTable)) + return FFX_ERROR_INVALID_ARGUMENT; + + inoutPipeline->cbResourceBindings[cbIndex].resourceIdentifier = cbResourceBindingTable[mapIndex].index; + } + + return FFX_OK; +} + + +static FfxErrorCode createPipelineStates(FfxFsr2Context_Private* context) +{ + FFX_ASSERT(context); + + const size_t samplerCount = 2; + FfxFilterType samplers[samplerCount]; + samplers[0] = FFX_FILTER_TYPE_POINT; + samplers[1] = FFX_FILTER_TYPE_LINEAR; + + const size_t rootConstantCount = 2; + uint32_t rootConstants[rootConstantCount]; + rootConstants[0] = sizeof(Fsr2Constants) / sizeof(uint32_t); + rootConstants[1] = sizeof(Fsr2SecondaryUnion) / sizeof(uint32_t); + + FfxPipelineDescription pipelineDescription; + pipelineDescription.contextFlags = context->contextDescription.flags; + pipelineDescription.samplerCount = samplerCount; + pipelineDescription.samplers = samplers; + pipelineDescription.rootConstantBufferCount = rootConstantCount; + pipelineDescription.rootConstantBufferSizes = rootConstants; + + // New interface: will handle RootSignature in backend + // set up pipeline descriptor (basically RootSignature and binding) + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_COMPUTE_LUMINANCE_PYRAMID, &pipelineDescription, &context->pipelineComputeLuminancePyramid)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_RCAS, &pipelineDescription, &context->pipelineRCAS)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_GENERATE_REACTIVE, &pipelineDescription, &context->pipelineGenerateReactive)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_TCR_AUTOGENERATE, &pipelineDescription, &context->pipelineTcrAutogenerate)); + + pipelineDescription.rootConstantBufferCount = 1; + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_DEPTH_CLIP, &pipelineDescription, &context->pipelineDepthClip)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_RECONSTRUCT_PREVIOUS_DEPTH, &pipelineDescription, &context->pipelineReconstructPreviousDepth)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_LOCK, &pipelineDescription, &context->pipelineLock)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_ACCUMULATE, &pipelineDescription, &context->pipelineAccumulate)); + FFX_VALIDATE(context->contextDescription.callbacks.fpCreatePipeline(&context->contextDescription.callbacks, FFX_FSR2_PASS_ACCUMULATE_SHARPEN, &pipelineDescription, &context->pipelineAccumulateSharpen)); + + // for each pipeline: re-route/fix-up IDs based on names + patchResourceBindings(&context->pipelineDepthClip); + patchResourceBindings(&context->pipelineReconstructPreviousDepth); + patchResourceBindings(&context->pipelineLock); + patchResourceBindings(&context->pipelineAccumulate); + patchResourceBindings(&context->pipelineComputeLuminancePyramid); + patchResourceBindings(&context->pipelineAccumulateSharpen); + patchResourceBindings(&context->pipelineRCAS); + patchResourceBindings(&context->pipelineGenerateReactive); + patchResourceBindings(&context->pipelineTcrAutogenerate); + + return FFX_OK; +} + +static FfxErrorCode generateReactiveMaskInternal(FfxFsr2Context_Private* contextPrivate, const FfxFsr2DispatchDescription* params); + +static FfxErrorCode fsr2Create(FfxFsr2Context_Private* context, const FfxFsr2ContextDescription* contextDescription) +{ + FFX_ASSERT(context); + FFX_ASSERT(contextDescription); + + // Setup the data for implementation. + memset(context, 0, sizeof(FfxFsr2Context_Private)); + context->device = contextDescription->device; + + memcpy(&context->contextDescription, contextDescription, sizeof(FfxFsr2ContextDescription)); + + if ((context->contextDescription.flags & FFX_FSR2_ENABLE_DEBUG_CHECKING) == FFX_FSR2_ENABLE_DEBUG_CHECKING) + { + if (context->contextDescription.fpMessage == nullptr) + { + FFX_ASSERT(context->contextDescription.fpMessage != nullptr); + // remove the debug checking flag - we have no message function + context->contextDescription.flags &= ~FFX_FSR2_ENABLE_DEBUG_CHECKING; + } + } + + // Create the device. + FfxErrorCode errorCode = context->contextDescription.callbacks.fpCreateBackendContext(&context->contextDescription.callbacks, context->device); + FFX_RETURN_ON_ERROR(errorCode == FFX_OK, errorCode); + + // call out for device caps. + errorCode = context->contextDescription.callbacks.fpGetDeviceCapabilities(&context->contextDescription.callbacks, &context->deviceCapabilities, context->device); + FFX_RETURN_ON_ERROR(errorCode == FFX_OK, errorCode); + + // set defaults + context->firstExecution = true; + context->resourceFrameIndex = 0; + + context->constants.displaySize[0] = contextDescription->displaySize.width; + context->constants.displaySize[1] = contextDescription->displaySize.height; + + // generate the data for the LUT. + const uint32_t lanczos2LutWidth = 128; + int16_t lanczos2Weights[lanczos2LutWidth] = { }; + + for (uint32_t currentLanczosWidthIndex = 0; currentLanczosWidthIndex < lanczos2LutWidth; currentLanczosWidthIndex++) { + + const float x = 2.0f * currentLanczosWidthIndex / float(lanczos2LutWidth - 1); + const float y = lanczos2(x); + lanczos2Weights[currentLanczosWidthIndex] = int16_t(roundf(y * 32767.0f)); + } + + // upload path only supports R16_SNORM, let's go and convert + int16_t maximumBias[FFX_FSR2_MAXIMUM_BIAS_TEXTURE_WIDTH * FFX_FSR2_MAXIMUM_BIAS_TEXTURE_HEIGHT]; + for (uint32_t i = 0; i < FFX_FSR2_MAXIMUM_BIAS_TEXTURE_WIDTH * FFX_FSR2_MAXIMUM_BIAS_TEXTURE_HEIGHT; ++i) { + + maximumBias[i] = int16_t(roundf(ffxFsr2MaximumBias[i] / 2.0f * 32767.0f)); + } + + uint8_t defaultReactiveMaskData = 0U; + uint32_t atomicInitData = 0U; + float defaultExposure[] = { 0.0f, 0.0f }; + const FfxResourceType texture1dResourceType = (context->contextDescription.flags & FFX_FSR2_ENABLE_TEXTURE1D_USAGE) ? FFX_RESOURCE_TYPE_TEXTURE1D : FFX_RESOURCE_TYPE_TEXTURE2D; + + // declare internal resources needed + const Fsr2ResourceDescription internalSurfaceDesc[] = { + + { FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR, L"FSR2_PreparedInputColor", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH, L"FSR2_ReconstructedPrevNearestDepth", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R32_UINT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_1, L"FSR2_InternalDilatedVelocity1", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16G16_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_2, L"FSR2_InternalDilatedVelocity2", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16G16_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH, L"FSR2_DilatedDepth", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R32_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_1, L"FSR2_LockStatus1", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16G16_FLOAT, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_2, L"FSR2_LockStatus2", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16G16_FLOAT, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA, L"FSR2_LockInputLuma", (FfxResourceUsage)(FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS, L"FSR2_NewLocks", (FfxResourceUsage)(FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R8_UNORM, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_1, L"FSR2_InternalUpscaled1", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_2, L"FSR2_InternalUpscaled2", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE, L"FSR2_ExposureMips", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R16_FLOAT, contextDescription->maxRenderSize.width / 2, contextDescription->maxRenderSize.height / 2, 0, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_1, L"FSR2_LumaHistory1", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R8G8B8A8_UNORM, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_2, L"FSR2_LumaHistory2", (FfxResourceUsage)(FFX_RESOURCE_USAGE_RENDERTARGET | FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R8G8B8A8_UNORM, contextDescription->displaySize.width, contextDescription->displaySize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_SPD_ATOMIC_COUNT, L"FSR2_SpdAtomicCounter", (FfxResourceUsage)(FFX_RESOURCE_USAGE_UAV), + FFX_SURFACE_FORMAT_R32_UINT, 1, 1, 1, FFX_RESOURCE_FLAGS_ALIASABLE, sizeof(atomicInitData), &atomicInitData }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS, L"FSR2_DilatedReactiveMasks", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R8G8_UNORM, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_ALIASABLE }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_LANCZOS_LUT, L"FSR2_LanczosLutData", FFX_RESOURCE_USAGE_READ_ONLY, + FFX_SURFACE_FORMAT_R16_SNORM, lanczos2LutWidth, 1, 1, FFX_RESOURCE_FLAGS_NONE, sizeof(lanczos2Weights), lanczos2Weights }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_REACTIVITY, L"FSR2_DefaultReactiviyMask", FFX_RESOURCE_USAGE_READ_ONLY, + FFX_SURFACE_FORMAT_R8_UNORM, 1, 1, 1, FFX_RESOURCE_FLAGS_NONE, sizeof(defaultReactiveMaskData), &defaultReactiveMaskData }, + + { FFX_FSR2_RESOURCE_IDENTITIER_UPSAMPLE_MAXIMUM_BIAS_LUT, L"FSR2_MaximumUpsampleBias", FFX_RESOURCE_USAGE_READ_ONLY, + FFX_SURFACE_FORMAT_R16_SNORM, FFX_FSR2_MAXIMUM_BIAS_TEXTURE_WIDTH, FFX_FSR2_MAXIMUM_BIAS_TEXTURE_HEIGHT, 1, FFX_RESOURCE_FLAGS_NONE, sizeof(maximumBias), maximumBias }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_EXPOSURE, L"FSR2_DefaultExposure", FFX_RESOURCE_USAGE_READ_ONLY, + FFX_SURFACE_FORMAT_R32G32_FLOAT, 1, 1, 1, FFX_RESOURCE_FLAGS_NONE, sizeof(defaultExposure), defaultExposure }, + + { FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE, L"FSR2_AutoExposure", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R32G32_FLOAT, 1, 1, 1, FFX_RESOURCE_FLAGS_NONE }, + + + // only one for now, will need pingpont to respect the motion vectors + { FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE, L"FSR2_AutoReactive", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R8_UNORM, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + { FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION, L"FSR2_AutoComposition", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R8_UNORM, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + { FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_1, L"FSR2_PrevPreAlpha0", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R11G11B10_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + { FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_1, L"FSR2_PrevPostAlpha0", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R11G11B10_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + { FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_2, L"FSR2_PrevPreAlpha1", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R11G11B10_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + { FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_2, L"FSR2_PrevPostAlpha1", FFX_RESOURCE_USAGE_UAV, + FFX_SURFACE_FORMAT_R11G11B10_FLOAT, contextDescription->maxRenderSize.width, contextDescription->maxRenderSize.height, 1, FFX_RESOURCE_FLAGS_NONE }, + + }; + + // clear the SRV resources to NULL. + memset(context->srvResources, 0, sizeof(context->srvResources)); + + for (int32_t currentSurfaceIndex = 0; currentSurfaceIndex < FFX_ARRAY_ELEMENTS(internalSurfaceDesc); ++currentSurfaceIndex) { + + const Fsr2ResourceDescription* currentSurfaceDescription = &internalSurfaceDesc[currentSurfaceIndex]; + const FfxResourceType resourceType = currentSurfaceDescription->height > 1 ? FFX_RESOURCE_TYPE_TEXTURE2D : texture1dResourceType; + const FfxResourceDescription resourceDescription = { resourceType, currentSurfaceDescription->format, currentSurfaceDescription->width, currentSurfaceDescription->height, 1, currentSurfaceDescription->mipCount }; + const FfxResourceStates initialState = (currentSurfaceDescription->usage == FFX_RESOURCE_USAGE_READ_ONLY) ? FFX_RESOURCE_STATE_COMPUTE_READ : FFX_RESOURCE_STATE_UNORDERED_ACCESS; + const FfxCreateResourceDescription createResourceDescription = { FFX_HEAP_TYPE_DEFAULT, resourceDescription, initialState, currentSurfaceDescription->initDataSize, currentSurfaceDescription->initData, currentSurfaceDescription->name, currentSurfaceDescription->usage, currentSurfaceDescription->id }; + + FFX_VALIDATE(context->contextDescription.callbacks.fpCreateResource(&context->contextDescription.callbacks, &createResourceDescription, &context->srvResources[currentSurfaceDescription->id])); + } + + // copy resources to uavResrouces list + memcpy(context->uavResources, context->srvResources, sizeof(context->srvResources)); + + // avoid compiling pipelines on first render + { + context->refreshPipelineStates = false; + errorCode = createPipelineStates(context); + FFX_RETURN_ON_ERROR(errorCode == FFX_OK, errorCode); + } + return FFX_OK; +} + +static void fsr2SafeReleasePipeline(FfxFsr2Context_Private* context, FfxPipelineState* pipeline) +{ + FFX_ASSERT(pipeline); + + context->contextDescription.callbacks.fpDestroyPipeline(&context->contextDescription.callbacks, pipeline); +} + +static void fsr2SafeReleaseResource(FfxFsr2Context_Private* context, FfxResourceInternal resource) +{ + context->contextDescription.callbacks.fpDestroyResource(&context->contextDescription.callbacks, resource); +} + +static void fsr2SafeReleaseDevice(FfxFsr2Context_Private* context, FfxDevice* device) +{ + if (*device == nullptr) { + return; + } + + context->contextDescription.callbacks.fpDestroyBackendContext(&context->contextDescription.callbacks); + *device = nullptr; +} + +static FfxErrorCode fsr2Release(FfxFsr2Context_Private* context) +{ + FFX_ASSERT(context); + + fsr2SafeReleasePipeline(context, &context->pipelineDepthClip); + fsr2SafeReleasePipeline(context, &context->pipelineReconstructPreviousDepth); + fsr2SafeReleasePipeline(context, &context->pipelineLock); + fsr2SafeReleasePipeline(context, &context->pipelineAccumulate); + fsr2SafeReleasePipeline(context, &context->pipelineAccumulateSharpen); + fsr2SafeReleasePipeline(context, &context->pipelineRCAS); + fsr2SafeReleasePipeline(context, &context->pipelineComputeLuminancePyramid); + fsr2SafeReleasePipeline(context, &context->pipelineGenerateReactive); + fsr2SafeReleasePipeline(context, &context->pipelineTcrAutogenerate); + + // unregister resources not created internally + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT] = { FFX_FSR2_RESOURCE_IDENTIFIER_NULL }; + + // release internal resources + for (int32_t currentResourceIndex = 0; currentResourceIndex < FFX_FSR2_RESOURCE_IDENTIFIER_COUNT; ++currentResourceIndex) { + + fsr2SafeReleaseResource(context, context->srvResources[currentResourceIndex]); + } + + fsr2SafeReleaseDevice(context, &context->device); + + return FFX_OK; +} + +static void setupDeviceDepthToViewSpaceDepthParams(FfxFsr2Context_Private* context, const FfxFsr2DispatchDescription* params) +{ + const bool bInverted = (context->contextDescription.flags & FFX_FSR2_ENABLE_DEPTH_INVERTED) == FFX_FSR2_ENABLE_DEPTH_INVERTED; + const bool bInfinite = (context->contextDescription.flags & FFX_FSR2_ENABLE_DEPTH_INFINITE) == FFX_FSR2_ENABLE_DEPTH_INFINITE; + + // make sure it has no impact if near and far plane values are swapped in dispatch params + // the flags "inverted" and "infinite" will decide what transform to use + float fMin = FFX_MINIMUM(params->cameraNear, params->cameraFar); + float fMax = FFX_MAXIMUM(params->cameraNear, params->cameraFar); + + if (bInverted) { + float tmp = fMin; + fMin = fMax; + fMax = tmp; + } + + // a 0 0 0 x + // 0 b 0 0 y + // 0 0 c d z + // 0 0 e 0 1 + + const float fQ = fMax / (fMin - fMax); + const float d = -1.0f; // for clarity + + const float matrix_elem_c[2][2] = { + fQ, // non reversed, non infinite + -1.0f - FLT_EPSILON, // non reversed, infinite + fQ, // reversed, non infinite + 0.0f + FLT_EPSILON // reversed, infinite + }; + + const float matrix_elem_e[2][2] = { + fQ * fMin, // non reversed, non infinite + -fMin - FLT_EPSILON, // non reversed, infinite + fQ * fMin, // reversed, non infinite + fMax, // reversed, infinite + }; + + context->constants.deviceToViewDepth[0] = d * matrix_elem_c[bInverted][bInfinite]; + context->constants.deviceToViewDepth[1] = matrix_elem_e[bInverted][bInfinite]; + + // revert x and y coords + const float aspect = params->renderSize.width / float(params->renderSize.height); + const float cotHalfFovY = cosf(0.5f * params->cameraFovAngleVertical) / sinf(0.5f * params->cameraFovAngleVertical); + const float a = cotHalfFovY / aspect; + const float b = cotHalfFovY; + + context->constants.deviceToViewDepth[2] = (1.0f / a); + context->constants.deviceToViewDepth[3] = (1.0f / b); +} + +static void scheduleDispatch(FfxFsr2Context_Private* context, const FfxFsr2DispatchDescription* params, const FfxPipelineState* pipeline, uint32_t dispatchX, uint32_t dispatchY) +{ + FfxComputeJobDescription jobDescriptor = {}; + + for (uint32_t currentShaderResourceViewIndex = 0; currentShaderResourceViewIndex < pipeline->srvCount; ++currentShaderResourceViewIndex) { + + const uint32_t currentResourceId = pipeline->srvResourceBindings[currentShaderResourceViewIndex].resourceIdentifier; + const FfxResourceInternal currentResource = context->srvResources[currentResourceId]; + jobDescriptor.srvs[currentShaderResourceViewIndex] = currentResource; + wcscpy_s(jobDescriptor.srvNames[currentShaderResourceViewIndex], pipeline->srvResourceBindings[currentShaderResourceViewIndex].name); + } + + for (uint32_t currentUnorderedAccessViewIndex = 0; currentUnorderedAccessViewIndex < pipeline->uavCount; ++currentUnorderedAccessViewIndex) { + + const uint32_t currentResourceId = pipeline->uavResourceBindings[currentUnorderedAccessViewIndex].resourceIdentifier; + wcscpy_s(jobDescriptor.uavNames[currentUnorderedAccessViewIndex], pipeline->uavResourceBindings[currentUnorderedAccessViewIndex].name); + + if (currentResourceId >= FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0 && currentResourceId <= FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_12) + { + const FfxResourceInternal currentResource = context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE]; + jobDescriptor.uavs[currentUnorderedAccessViewIndex] = currentResource; + jobDescriptor.uavMip[currentUnorderedAccessViewIndex] = currentResourceId - FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0; + } + else + { + const FfxResourceInternal currentResource = context->uavResources[currentResourceId]; + jobDescriptor.uavs[currentUnorderedAccessViewIndex] = currentResource; + jobDescriptor.uavMip[currentUnorderedAccessViewIndex] = 0; + } + } + + jobDescriptor.dimensions[0] = dispatchX; + jobDescriptor.dimensions[1] = dispatchY; + jobDescriptor.dimensions[2] = 1; + jobDescriptor.pipeline = *pipeline; + + for (uint32_t currentRootConstantIndex = 0; currentRootConstantIndex < pipeline->constCount; ++currentRootConstantIndex) { + wcscpy_s( jobDescriptor.cbNames[currentRootConstantIndex], pipeline->cbResourceBindings[currentRootConstantIndex].name); + jobDescriptor.cbs[currentRootConstantIndex] = globalFsr2ConstantBuffers[pipeline->cbResourceBindings[currentRootConstantIndex].resourceIdentifier]; + jobDescriptor.cbSlotIndex[currentRootConstantIndex] = pipeline->cbResourceBindings[currentRootConstantIndex].slotIndex; + } + + FfxGpuJobDescription dispatchJob = { FFX_GPU_JOB_COMPUTE }; + dispatchJob.computeJobDescriptor = jobDescriptor; + + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &dispatchJob); +} + +static FfxErrorCode fsr2Dispatch(FfxFsr2Context_Private* context, const FfxFsr2DispatchDescription* params) +{ + if ((context->contextDescription.flags & FFX_FSR2_ENABLE_DEBUG_CHECKING) == FFX_FSR2_ENABLE_DEBUG_CHECKING) + { + fsr2DebugCheckDispatch(context, params); + } + // take a short cut to the command list + FfxCommandList commandList = params->commandList; + + // try and refresh shaders first. Early exit in case of error. + if (context->refreshPipelineStates) { + + context->refreshPipelineStates = false; + + const FfxErrorCode errorCode = createPipelineStates(context); + FFX_RETURN_ON_ERROR(errorCode == FFX_OK, errorCode); + } + + if (context->firstExecution) + { + FfxGpuJobDescription clearJob = { FFX_GPU_JOB_CLEAR_FLOAT }; + + const float clearValuesToZeroFloat[]{ 0.f, 0.f, 0.f, 0.f }; + memcpy(clearJob.clearJobDescriptor.color, clearValuesToZeroFloat, 4 * sizeof(float)); + + clearJob.clearJobDescriptor.target = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_1]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + clearJob.clearJobDescriptor.target = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_2]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + clearJob.clearJobDescriptor.target = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + } + + // Prepare per frame descriptor tables + const bool isOddFrame = !!(context->resourceFrameIndex & 1); + const uint32_t currentCpuOnlyTableBase = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_COUNT : 0; + const uint32_t currentGpuTableBase = 2 * FFX_FSR2_RESOURCE_IDENTIFIER_COUNT * context->resourceFrameIndex; + const uint32_t lockStatusSrvResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_2 : FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_1; + const uint32_t lockStatusUavResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_1 : FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_2; + const uint32_t upscaledColorSrvResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_2 : FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_1; + const uint32_t upscaledColorUavResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_1 : FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_2; + const uint32_t dilatedMotionVectorsResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_2 : FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_1; + const uint32_t previousDilatedMotionVectorsResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_1 : FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_2; + const uint32_t lumaHistorySrvResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_2 : FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_1; + const uint32_t lumaHistoryUavResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_1 : FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_2; + + const uint32_t prevPreAlphaColorSrvResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_2 : FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_1; + const uint32_t prevPreAlphaColorUavResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_1 : FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_2; + const uint32_t prevPostAlphaColorSrvResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_2 : FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_1; + const uint32_t prevPostAlphaColorUavResourceIndex = isOddFrame ? FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_1 : FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_2; + + const bool resetAccumulation = params->reset || context->firstExecution; + context->firstExecution = false; + + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->color, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR]); + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->depth, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH]); + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->motionVectors, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS]); + + // if auto exposure is enabled use the auto exposure SRV, otherwise what the app sends. + if (context->contextDescription.flags & FFX_FSR2_ENABLE_AUTO_EXPOSURE) { + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE] = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE]; + } else { + if (ffxFsr2ResourceIsNull(params->exposure)) { + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE] = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_EXPOSURE]; + } else { + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->exposure, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE]); + } + } + + if (params->enableAutoReactive) + { + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->colorOpaqueOnly, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR]); + } + + if (ffxFsr2ResourceIsNull(params->reactive)) { + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK] = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_REACTIVITY]; + } + else { + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->reactive, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK]); + } + + if (ffxFsr2ResourceIsNull(params->transparencyAndComposition)) { + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK] = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_REACTIVITY]; + } else { + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->transparencyAndComposition, &context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK]); + } + + context->contextDescription.callbacks.fpRegisterResource(&context->contextDescription.callbacks, ¶ms->output, &context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT]); + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS] = context->srvResources[lockStatusSrvResourceIndex]; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR] = context->srvResources[upscaledColorSrvResourceIndex]; + context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS] = context->uavResources[lockStatusUavResourceIndex]; + context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR] = context->uavResources[upscaledColorUavResourceIndex]; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT] = context->uavResources[upscaledColorUavResourceIndex]; + + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS] = context->srvResources[dilatedMotionVectorsResourceIndex]; + context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS] = context->uavResources[dilatedMotionVectorsResourceIndex]; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREVIOUS_DILATED_MOTION_VECTORS] = context->srvResources[previousDilatedMotionVectorsResourceIndex]; + + context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY] = context->uavResources[lumaHistoryUavResourceIndex]; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY] = context->srvResources[lumaHistorySrvResourceIndex]; + + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR] = context->srvResources[prevPreAlphaColorSrvResourceIndex]; + context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR] = context->uavResources[prevPreAlphaColorUavResourceIndex]; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR] = context->srvResources[prevPostAlphaColorSrvResourceIndex]; + context->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR] = context->uavResources[prevPostAlphaColorUavResourceIndex]; + + // actual resource size may differ from render/display resolution (e.g. due to Hw/API restrictions), so query the descriptor for UVs adjustment + const FfxResourceDescription resourceDescInputColor = context->contextDescription.callbacks.fpGetResourceDescription(&context->contextDescription.callbacks, context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR]); + const FfxResourceDescription resourceDescLockStatus = context->contextDescription.callbacks.fpGetResourceDescription(&context->contextDescription.callbacks, context->srvResources[lockStatusSrvResourceIndex]); + const FfxResourceDescription resourceDescReactiveMask = context->contextDescription.callbacks.fpGetResourceDescription(&context->contextDescription.callbacks, context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK]); + FFX_ASSERT(resourceDescInputColor.type == FFX_RESOURCE_TYPE_TEXTURE2D); + FFX_ASSERT(resourceDescLockStatus.type == FFX_RESOURCE_TYPE_TEXTURE2D); + + context->constants.jitterOffset[0] = params->jitterOffset.x; + context->constants.jitterOffset[1] = params->jitterOffset.y; + context->constants.renderSize[0] = int32_t(params->renderSize.width ? params->renderSize.width : resourceDescInputColor.width); + context->constants.renderSize[1] = int32_t(params->renderSize.height ? params->renderSize.height : resourceDescInputColor.height); + context->constants.maxRenderSize[0] = int32_t(context->contextDescription.maxRenderSize.width); + context->constants.maxRenderSize[1] = int32_t(context->contextDescription.maxRenderSize.height); + context->constants.inputColorResourceDimensions[0] = resourceDescInputColor.width; + context->constants.inputColorResourceDimensions[1] = resourceDescInputColor.height; + + // compute the horizontal FOV for the shader from the vertical one. + const float aspectRatio = (float)params->renderSize.width / (float)params->renderSize.height; + const float cameraAngleHorizontal = atan(tan(params->cameraFovAngleVertical / 2) * aspectRatio) * 2; + context->constants.tanHalfFOV = tanf(cameraAngleHorizontal * 0.5f); + context->constants.viewSpaceToMetersFactor = (params->viewSpaceToMetersFactor > 0.0f) ? params->viewSpaceToMetersFactor : 1.0f; + + // compute params to enable device depth to view space depth computation in shader + setupDeviceDepthToViewSpaceDepthParams(context, params); + + // To be updated if resource is larger than the actual image size + context->constants.downscaleFactor[0] = float(context->constants.renderSize[0]) / context->contextDescription.displaySize.width; + context->constants.downscaleFactor[1] = float(context->constants.renderSize[1]) / context->contextDescription.displaySize.height; + context->constants.previousFramePreExposure = context->constants.preExposure; + context->constants.preExposure = (params->preExposure != 0) ? params->preExposure : 1.0f; + + // motion vector data + const int32_t* motionVectorsTargetSize = (context->contextDescription.flags & FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS) ? context->constants.displaySize : context->constants.renderSize; + + context->constants.motionVectorScale[0] = (params->motionVectorScale.x / motionVectorsTargetSize[0]); + context->constants.motionVectorScale[1] = (params->motionVectorScale.y / motionVectorsTargetSize[1]); + + // compute jitter cancellation + if (context->contextDescription.flags & FFX_FSR2_ENABLE_MOTION_VECTORS_JITTER_CANCELLATION) { + + context->constants.motionVectorJitterCancellation[0] = (context->previousJitterOffset[0] - context->constants.jitterOffset[0]) / motionVectorsTargetSize[0]; + context->constants.motionVectorJitterCancellation[1] = (context->previousJitterOffset[1] - context->constants.jitterOffset[1]) / motionVectorsTargetSize[1]; + + context->previousJitterOffset[0] = context->constants.jitterOffset[0]; + context->previousJitterOffset[1] = context->constants.jitterOffset[1]; + } + + // lock data, assuming jitter sequence length computation for now + const int32_t jitterPhaseCount = ffxFsr2GetJitterPhaseCount(params->renderSize.width, context->contextDescription.displaySize.width); + + // init on first frame + if (resetAccumulation || context->constants.jitterPhaseCount == 0) { + context->constants.jitterPhaseCount = (float)jitterPhaseCount; + } else { + const int32_t jitterPhaseCountDelta = (int32_t)(jitterPhaseCount - context->constants.jitterPhaseCount); + if (jitterPhaseCountDelta > 0) { + context->constants.jitterPhaseCount++; + } else if (jitterPhaseCountDelta < 0) { + context->constants.jitterPhaseCount--; + } + } + + // convert delta time to seconds and clamp to [0, 1]. + context->constants.deltaTime = FFX_MAXIMUM(0.0f, FFX_MINIMUM(1.0f, params->frameTimeDelta / 1000.0f)); + + if (resetAccumulation) { + context->constants.frameIndex = 0; + } else { + context->constants.frameIndex++; + } + + // shading change usage of the SPD mip levels. + context->constants.lumaMipLevelToUse = uint32_t(FFX_FSR2_SHADING_CHANGE_MIP_LEVEL); + + const float mipDiv = float(2 << context->constants.lumaMipLevelToUse); + context->constants.lumaMipDimensions[0] = uint32_t(context->constants.maxRenderSize[0] / mipDiv); + context->constants.lumaMipDimensions[1] = uint32_t(context->constants.maxRenderSize[1] / mipDiv); + + // -- GODOT start -- + memcpy(context->constants.reprojectionMatrix, params->reprojectionMatrix, sizeof(context->constants.reprojectionMatrix)); + // -- GODOT end -- + + // reactive mask bias + const int32_t threadGroupWorkRegionDim = 8; + const int32_t dispatchSrcX = (context->constants.renderSize[0] + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + const int32_t dispatchSrcY = (context->constants.renderSize[1] + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + const int32_t dispatchDstX = (context->contextDescription.displaySize.width + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + const int32_t dispatchDstY = (context->contextDescription.displaySize.height + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + + // Clear reconstructed depth for max depth store. + if (resetAccumulation) { + + FfxGpuJobDescription clearJob = { FFX_GPU_JOB_CLEAR_FLOAT }; + + // LockStatus resource has no sign bit, callback functions are compensating for this. + // Clearing the resource must follow the same logic. + float clearValuesLockStatus[4]{}; + clearValuesLockStatus[LOCK_LIFETIME_REMAINING] = 0.0f; + clearValuesLockStatus[LOCK_TEMPORAL_LUMA] = 0.0f; + + memcpy(clearJob.clearJobDescriptor.color, clearValuesLockStatus, 4 * sizeof(float)); + clearJob.clearJobDescriptor.target = context->srvResources[lockStatusSrvResourceIndex]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + + const float clearValuesToZeroFloat[]{ 0.f, 0.f, 0.f, 0.f }; + memcpy(clearJob.clearJobDescriptor.color, clearValuesToZeroFloat, 4 * sizeof(float)); + clearJob.clearJobDescriptor.target = context->srvResources[upscaledColorSrvResourceIndex]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + + clearJob.clearJobDescriptor.target = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + + //if (context->contextDescription.flags & FFX_FSR2_ENABLE_AUTO_EXPOSURE) + // Auto exposure always used to track luma changes in locking logic + { + const float clearValuesExposure[]{ -1.f, 1e8f, 0.f, 0.f }; + memcpy(clearJob.clearJobDescriptor.color, clearValuesExposure, 4 * sizeof(float)); + clearJob.clearJobDescriptor.target = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE]; + context->contextDescription.callbacks.fpScheduleGpuJob(&context->contextDescription.callbacks, &clearJob); + } + } + + // Auto exposure + uint32_t dispatchThreadGroupCountXY[2]; + uint32_t workGroupOffset[2]; + uint32_t numWorkGroupsAndMips[2]; + uint32_t rectInfo[4] = { 0, 0, params->renderSize.width, params->renderSize.height }; + SpdSetup(dispatchThreadGroupCountXY, workGroupOffset, numWorkGroupsAndMips, rectInfo); + + // downsample + Fsr2SpdConstants luminancePyramidConstants; + luminancePyramidConstants.numworkGroups = numWorkGroupsAndMips[0]; + luminancePyramidConstants.mips = numWorkGroupsAndMips[1]; + luminancePyramidConstants.workGroupOffset[0] = workGroupOffset[0]; + luminancePyramidConstants.workGroupOffset[1] = workGroupOffset[1]; + luminancePyramidConstants.renderSize[0] = params->renderSize.width; + luminancePyramidConstants.renderSize[1] = params->renderSize.height; + + // compute the constants. + Fsr2RcasConstants rcasConsts = {}; + const float sharpenessRemapped = (-2.0f * params->sharpness) + 2.0f; + FsrRcasCon(rcasConsts.rcasConfig, sharpenessRemapped); + + Fsr2GenerateReactiveConstants2 genReactiveConsts = {}; + genReactiveConsts.autoTcThreshold = params->autoTcThreshold; + genReactiveConsts.autoTcScale = params->autoTcScale; + genReactiveConsts.autoReactiveScale = params->autoReactiveScale; + genReactiveConsts.autoReactiveMax = params->autoReactiveMax; + + // initialize constantBuffers data + memcpy(&globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_FSR2].data, &context->constants, globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_FSR2].uint32Size * sizeof(uint32_t)); + memcpy(&globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_SPD].data, &luminancePyramidConstants, globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_SPD].uint32Size * sizeof(uint32_t)); + memcpy(&globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_RCAS].data, &rcasConsts, globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_RCAS].uint32Size * sizeof(uint32_t)); + memcpy(&globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_GENREACTIVE].data, &genReactiveConsts, globalFsr2ConstantBuffers[FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_GENREACTIVE].uint32Size * sizeof(uint32_t)); + + // Auto reactive + if (params->enableAutoReactive) + { + generateReactiveMaskInternal(context, params); + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK] = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE]; + context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK] = context->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION]; + } + scheduleDispatch(context, params, &context->pipelineComputeLuminancePyramid, dispatchThreadGroupCountXY[0], dispatchThreadGroupCountXY[1]); + scheduleDispatch(context, params, &context->pipelineReconstructPreviousDepth, dispatchSrcX, dispatchSrcY); + scheduleDispatch(context, params, &context->pipelineDepthClip, dispatchSrcX, dispatchSrcY); + + const bool sharpenEnabled = params->enableSharpening; + + scheduleDispatch(context, params, &context->pipelineLock, dispatchSrcX, dispatchSrcY); + scheduleDispatch(context, params, sharpenEnabled ? &context->pipelineAccumulateSharpen : &context->pipelineAccumulate, dispatchDstX, dispatchDstY); + + // RCAS + if (sharpenEnabled) { + + // dispatch RCAS + const int32_t threadGroupWorkRegionDimRCAS = 16; + const int32_t dispatchX = (context->contextDescription.displaySize.width + (threadGroupWorkRegionDimRCAS - 1)) / threadGroupWorkRegionDimRCAS; + const int32_t dispatchY = (context->contextDescription.displaySize.height + (threadGroupWorkRegionDimRCAS - 1)) / threadGroupWorkRegionDimRCAS; + scheduleDispatch(context, params, &context->pipelineRCAS, dispatchX, dispatchY); + } + + context->resourceFrameIndex = (context->resourceFrameIndex + 1) % FSR2_MAX_QUEUED_FRAMES; + + // Fsr2MaxQueuedFrames must be an even number. + FFX_STATIC_ASSERT((FSR2_MAX_QUEUED_FRAMES & 1) == 0); + + context->contextDescription.callbacks.fpExecuteGpuJobs(&context->contextDescription.callbacks, commandList); + + // release dynamic resources + context->contextDescription.callbacks.fpUnregisterResources(&context->contextDescription.callbacks); + + return FFX_OK; +} + +FfxErrorCode ffxFsr2ContextCreate(FfxFsr2Context* context, const FfxFsr2ContextDescription* contextDescription) +{ + // zero context memory + memset(context, 0, sizeof(FfxFsr2Context)); + + // check pointers are valid. + FFX_RETURN_ON_ERROR( + context, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + contextDescription, + FFX_ERROR_INVALID_POINTER); + + // validate that all callbacks are set for the interface + FFX_RETURN_ON_ERROR(contextDescription->callbacks.fpGetDeviceCapabilities, FFX_ERROR_INCOMPLETE_INTERFACE); + FFX_RETURN_ON_ERROR(contextDescription->callbacks.fpCreateBackendContext, FFX_ERROR_INCOMPLETE_INTERFACE); + FFX_RETURN_ON_ERROR(contextDescription->callbacks.fpDestroyBackendContext, FFX_ERROR_INCOMPLETE_INTERFACE); + + // if a scratch buffer is declared, then we must have a size + if (contextDescription->callbacks.scratchBuffer) { + + FFX_RETURN_ON_ERROR(contextDescription->callbacks.scratchBufferSize, FFX_ERROR_INCOMPLETE_INTERFACE); + } + + // ensure the context is large enough for the internal context. + FFX_STATIC_ASSERT(sizeof(FfxFsr2Context) >= sizeof(FfxFsr2Context_Private)); + + // create the context. + FfxFsr2Context_Private* contextPrivate = (FfxFsr2Context_Private*)(context); + const FfxErrorCode errorCode = fsr2Create(contextPrivate, contextDescription); + + return errorCode; +} + +FfxErrorCode ffxFsr2ContextDestroy(FfxFsr2Context* context) +{ + FFX_RETURN_ON_ERROR( + context, + FFX_ERROR_INVALID_POINTER); + + // destroy the context. + FfxFsr2Context_Private* contextPrivate = (FfxFsr2Context_Private*)(context); + const FfxErrorCode errorCode = fsr2Release(contextPrivate); + return errorCode; +} + +FfxErrorCode ffxFsr2ContextDispatch(FfxFsr2Context* context, const FfxFsr2DispatchDescription* dispatchParams) +{ + FFX_RETURN_ON_ERROR( + context, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + dispatchParams, + FFX_ERROR_INVALID_POINTER); + + FfxFsr2Context_Private* contextPrivate = (FfxFsr2Context_Private*)(context); + + // validate that renderSize is within the maximum. + FFX_RETURN_ON_ERROR( + dispatchParams->renderSize.width <= contextPrivate->contextDescription.maxRenderSize.width, + FFX_ERROR_OUT_OF_RANGE); + FFX_RETURN_ON_ERROR( + dispatchParams->renderSize.height <= contextPrivate->contextDescription.maxRenderSize.height, + FFX_ERROR_OUT_OF_RANGE); + FFX_RETURN_ON_ERROR( + contextPrivate->device, + FFX_ERROR_NULL_DEVICE); + + // dispatch the FSR2 passes. + const FfxErrorCode errorCode = fsr2Dispatch(contextPrivate, dispatchParams); + return errorCode; +} + +float ffxFsr2GetUpscaleRatioFromQualityMode(FfxFsr2QualityMode qualityMode) +{ + switch (qualityMode) { + + case FFX_FSR2_QUALITY_MODE_QUALITY: + return 1.5f; + case FFX_FSR2_QUALITY_MODE_BALANCED: + return 1.7f; + case FFX_FSR2_QUALITY_MODE_PERFORMANCE: + return 2.0f; + case FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE: + return 3.0f; + default: + return 0.0f; + } +} + +FfxErrorCode ffxFsr2GetRenderResolutionFromQualityMode( + uint32_t* renderWidth, + uint32_t* renderHeight, + uint32_t displayWidth, + uint32_t displayHeight, + FfxFsr2QualityMode qualityMode) +{ + FFX_RETURN_ON_ERROR( + renderWidth, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + renderHeight, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + FFX_FSR2_QUALITY_MODE_QUALITY <= qualityMode && qualityMode <= FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE, + FFX_ERROR_INVALID_ENUM); + + // scale by the predefined ratios in each dimension. + const float ratio = ffxFsr2GetUpscaleRatioFromQualityMode(qualityMode); + const uint32_t scaledDisplayWidth = (uint32_t)((float)displayWidth / ratio); + const uint32_t scaledDisplayHeight = (uint32_t)((float)displayHeight / ratio); + *renderWidth = scaledDisplayWidth; + *renderHeight = scaledDisplayHeight; + + return FFX_OK; +} + +FfxErrorCode ffxFsr2ContextEnqueueRefreshPipelineRequest(FfxFsr2Context* context) +{ + FFX_RETURN_ON_ERROR( + context, + FFX_ERROR_INVALID_POINTER); + + FfxFsr2Context_Private* contextPrivate = (FfxFsr2Context_Private*)context; + contextPrivate->refreshPipelineStates = true; + + return FFX_OK; +} + +int32_t ffxFsr2GetJitterPhaseCount(int32_t renderWidth, int32_t displayWidth) +{ + const float basePhaseCount = 8.0f; + const int32_t jitterPhaseCount = int32_t(basePhaseCount * pow((float(displayWidth) / renderWidth), 2.0f)); + return jitterPhaseCount; +} + +FfxErrorCode ffxFsr2GetJitterOffset(float* outX, float* outY, int32_t index, int32_t phaseCount) +{ + FFX_RETURN_ON_ERROR( + outX, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + outY, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + phaseCount > 0, + FFX_ERROR_INVALID_ARGUMENT); + + const float x = halton((index % phaseCount) + 1, 2) - 0.5f; + const float y = halton((index % phaseCount) + 1, 3) - 0.5f; + + *outX = x; + *outY = y; + return FFX_OK; +} + +FFX_API bool ffxFsr2ResourceIsNull(FfxResource resource) +{ + return resource.resource == NULL; +} + +FfxErrorCode ffxFsr2ContextGenerateReactiveMask(FfxFsr2Context* context, const FfxFsr2GenerateReactiveDescription* params) +{ + FFX_RETURN_ON_ERROR( + context, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + params, + FFX_ERROR_INVALID_POINTER); + FFX_RETURN_ON_ERROR( + params->commandList, + FFX_ERROR_INVALID_POINTER); + + FfxFsr2Context_Private* contextPrivate = (FfxFsr2Context_Private*)(context); + + FFX_RETURN_ON_ERROR( + contextPrivate->device, + FFX_ERROR_NULL_DEVICE); + + if (contextPrivate->refreshPipelineStates) { + + createPipelineStates(contextPrivate); + contextPrivate->refreshPipelineStates = false; + } + + // take a short cut to the command list + FfxCommandList commandList = params->commandList; + + FfxPipelineState* pipeline = &contextPrivate->pipelineGenerateReactive; + + const int32_t threadGroupWorkRegionDim = 8; + const int32_t dispatchSrcX = (params->renderSize.width + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + const int32_t dispatchSrcY = (params->renderSize.height + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + + // save internal reactive resource + FfxResourceInternal internalReactive = contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE]; + + FfxComputeJobDescription jobDescriptor = {}; + contextPrivate->contextDescription.callbacks.fpRegisterResource(&contextPrivate->contextDescription.callbacks, ¶ms->colorOpaqueOnly, &contextPrivate->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY]); + contextPrivate->contextDescription.callbacks.fpRegisterResource(&contextPrivate->contextDescription.callbacks, ¶ms->colorPreUpscale, &contextPrivate->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR]); + contextPrivate->contextDescription.callbacks.fpRegisterResource(&contextPrivate->contextDescription.callbacks, ¶ms->outReactive, &contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE]); + + jobDescriptor.uavs[0] = contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE]; + + wcscpy_s(jobDescriptor.srvNames[0], pipeline->srvResourceBindings[0].name); + wcscpy_s(jobDescriptor.srvNames[1], pipeline->srvResourceBindings[1].name); + wcscpy_s(jobDescriptor.uavNames[0], pipeline->uavResourceBindings[0].name); + + jobDescriptor.dimensions[0] = dispatchSrcX; + jobDescriptor.dimensions[1] = dispatchSrcY; + jobDescriptor.dimensions[2] = 1; + jobDescriptor.pipeline = *pipeline; + + for (uint32_t currentShaderResourceViewIndex = 0; currentShaderResourceViewIndex < pipeline->srvCount; ++currentShaderResourceViewIndex) { + + const uint32_t currentResourceId = pipeline->srvResourceBindings[currentShaderResourceViewIndex].resourceIdentifier; + const FfxResourceInternal currentResource = contextPrivate->srvResources[currentResourceId]; + jobDescriptor.srvs[currentShaderResourceViewIndex] = currentResource; + wcscpy_s(jobDescriptor.srvNames[currentShaderResourceViewIndex], pipeline->srvResourceBindings[currentShaderResourceViewIndex].name); + } + + Fsr2GenerateReactiveConstants constants = {}; + constants.scale = params->scale; + constants.threshold = params->cutoffThreshold; + constants.binaryValue = params->binaryValue; + constants.flags = params->flags; + + jobDescriptor.cbs[0].uint32Size = sizeof(constants); + memcpy(&jobDescriptor.cbs[0].data, &constants, sizeof(constants)); + wcscpy_s(jobDescriptor.cbNames[0], pipeline->cbResourceBindings[0].name); + + FfxGpuJobDescription dispatchJob = { FFX_GPU_JOB_COMPUTE }; + dispatchJob.computeJobDescriptor = jobDescriptor; + + contextPrivate->contextDescription.callbacks.fpScheduleGpuJob(&contextPrivate->contextDescription.callbacks, &dispatchJob); + + contextPrivate->contextDescription.callbacks.fpExecuteGpuJobs(&contextPrivate->contextDescription.callbacks, commandList); + + // restore internal reactive + contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE] = internalReactive; + + return FFX_OK; +} + +static FfxErrorCode generateReactiveMaskInternal(FfxFsr2Context_Private* contextPrivate, const FfxFsr2DispatchDescription* params) +{ + if (contextPrivate->refreshPipelineStates) { + + createPipelineStates(contextPrivate); + contextPrivate->refreshPipelineStates = false; + } + + // take a short cut to the command list + FfxCommandList commandList = params->commandList; + + FfxPipelineState* pipeline = &contextPrivate->pipelineTcrAutogenerate; + + const int32_t threadGroupWorkRegionDim = 8; + const int32_t dispatchSrcX = (params->renderSize.width + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + const int32_t dispatchSrcY = (params->renderSize.height + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim; + + FfxComputeJobDescription jobDescriptor = {}; + contextPrivate->contextDescription.callbacks.fpRegisterResource(&contextPrivate->contextDescription.callbacks, ¶ms->colorOpaqueOnly, &contextPrivate->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY]); + contextPrivate->contextDescription.callbacks.fpRegisterResource(&contextPrivate->contextDescription.callbacks, ¶ms->color, &contextPrivate->srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR]); + + jobDescriptor.uavs[0] = contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE]; + jobDescriptor.uavs[1] = contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION]; + jobDescriptor.uavs[2] = contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR]; + jobDescriptor.uavs[3] = contextPrivate->uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR]; + + wcscpy_s(jobDescriptor.uavNames[0], pipeline->uavResourceBindings[0].name); + wcscpy_s(jobDescriptor.uavNames[1], pipeline->uavResourceBindings[1].name); + wcscpy_s(jobDescriptor.uavNames[2], pipeline->uavResourceBindings[2].name); + wcscpy_s(jobDescriptor.uavNames[3], pipeline->uavResourceBindings[3].name); + + jobDescriptor.dimensions[0] = dispatchSrcX; + jobDescriptor.dimensions[1] = dispatchSrcY; + jobDescriptor.dimensions[2] = 1; + jobDescriptor.pipeline = *pipeline; + + for (uint32_t currentShaderResourceViewIndex = 0; currentShaderResourceViewIndex < pipeline->srvCount; ++currentShaderResourceViewIndex) { + + const uint32_t currentResourceId = pipeline->srvResourceBindings[currentShaderResourceViewIndex].resourceIdentifier; + const FfxResourceInternal currentResource = contextPrivate->srvResources[currentResourceId]; + jobDescriptor.srvs[currentShaderResourceViewIndex] = currentResource; + wcscpy_s(jobDescriptor.srvNames[currentShaderResourceViewIndex], pipeline->srvResourceBindings[currentShaderResourceViewIndex].name); + } + + for (uint32_t currentRootConstantIndex = 0; currentRootConstantIndex < pipeline->constCount; ++currentRootConstantIndex) { + wcscpy_s(jobDescriptor.cbNames[currentRootConstantIndex], pipeline->cbResourceBindings[currentRootConstantIndex].name); + jobDescriptor.cbs[currentRootConstantIndex] = globalFsr2ConstantBuffers[pipeline->cbResourceBindings[currentRootConstantIndex].resourceIdentifier]; + jobDescriptor.cbSlotIndex[currentRootConstantIndex] = pipeline->cbResourceBindings[currentRootConstantIndex].slotIndex; + } + + FfxGpuJobDescription dispatchJob = { FFX_GPU_JOB_COMPUTE }; + dispatchJob.computeJobDescriptor = jobDescriptor; + + contextPrivate->contextDescription.callbacks.fpScheduleGpuJob(&contextPrivate->contextDescription.callbacks, &dispatchJob); + + return FFX_OK; +} diff --git a/thirdparty/amd-fsr2/ffx_fsr2.h b/thirdparty/amd-fsr2/ffx_fsr2.h new file mode 100644 index 000000000000..7df3773cccb0 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_fsr2.h @@ -0,0 +1,458 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + +// @defgroup FSR2 + +#pragma once + +// Include the interface for the backend of the FSR2 API. +#include "ffx_fsr2_interface.h" + +/// FidelityFX Super Resolution 2 major version. +/// +/// @ingroup FSR2 +#define FFX_FSR2_VERSION_MAJOR (2) + +/// FidelityFX Super Resolution 2 minor version. +/// +/// @ingroup FSR2 +#define FFX_FSR2_VERSION_MINOR (2) + +/// FidelityFX Super Resolution 2 patch version. +/// +/// @ingroup FSR2 +#define FFX_FSR2_VERSION_PATCH (1) + +/// The size of the context specified in 32bit values. +/// +/// @ingroup FSR2 +#define FFX_FSR2_CONTEXT_SIZE (16536) + +#if defined(__cplusplus) +extern "C" { +#endif // #if defined(__cplusplus) + +/// An enumeration of all the quality modes supported by FidelityFX Super +/// Resolution 2 upscaling. +/// +/// In order to provide a consistent user experience across multiple +/// applications which implement FSR2. It is strongly recommended that the +/// following preset scaling factors are made available through your +/// application's user interface. +/// +/// If your application does not expose the notion of preset scaling factors +/// for upscaling algorithms (perhaps instead implementing a fixed ratio which +/// is immutable) or implementing a more dynamic scaling scheme (such as +/// dynamic resolution scaling), then there is no need to use these presets. +/// +/// Please note that FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE is +/// an optional mode which may introduce significant quality degradation in the +/// final image. As such it is recommended that you evaluate the final results +/// of using this scaling mode before deciding if you should include it in your +/// application. +/// +/// @ingroup FSR2 +typedef enum FfxFsr2QualityMode { + + FFX_FSR2_QUALITY_MODE_QUALITY = 1, ///< Perform upscaling with a per-dimension upscaling ratio of 1.5x. + FFX_FSR2_QUALITY_MODE_BALANCED = 2, ///< Perform upscaling with a per-dimension upscaling ratio of 1.7x. + FFX_FSR2_QUALITY_MODE_PERFORMANCE = 3, ///< Perform upscaling with a per-dimension upscaling ratio of 2.0x. + FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE = 4 ///< Perform upscaling with a per-dimension upscaling ratio of 3.0x. +} FfxFsr2QualityMode; + +/// An enumeration of bit flags used when creating a +/// FfxFsr2Context. See FfxFsr2ContextDescription. +/// +/// @ingroup FSR2 +typedef enum FfxFsr2InitializationFlagBits { + + FFX_FSR2_ENABLE_HIGH_DYNAMIC_RANGE = (1<<0), ///< A bit indicating if the input color data provided is using a high-dynamic range. + FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS = (1<<1), ///< A bit indicating if the motion vectors are rendered at display resolution. + FFX_FSR2_ENABLE_MOTION_VECTORS_JITTER_CANCELLATION = (1<<2), ///< A bit indicating that the motion vectors have the jittering pattern applied to them. + FFX_FSR2_ENABLE_DEPTH_INVERTED = (1<<3), ///< A bit indicating that the input depth buffer data provided is inverted [1..0]. + FFX_FSR2_ENABLE_DEPTH_INFINITE = (1<<4), ///< A bit indicating that the input depth buffer data provided is using an infinite far plane. + FFX_FSR2_ENABLE_AUTO_EXPOSURE = (1<<5), ///< A bit indicating if automatic exposure should be applied to input color data. + FFX_FSR2_ENABLE_DYNAMIC_RESOLUTION = (1<<6), ///< A bit indicating that the application uses dynamic resolution scaling. + FFX_FSR2_ENABLE_TEXTURE1D_USAGE = (1<<7), ///< A bit indicating that the backend should use 1D textures. + FFX_FSR2_ENABLE_DEBUG_CHECKING = (1<<8), ///< A bit indicating that the runtime should check some API values and report issues. +} FfxFsr2InitializationFlagBits; + +/// A structure encapsulating the parameters required to initialize FidelityFX +/// Super Resolution 2 upscaling. +/// +/// @ingroup FSR2 +typedef struct FfxFsr2ContextDescription { + + uint32_t flags; ///< A collection of FfxFsr2InitializationFlagBits. + FfxDimensions2D maxRenderSize; ///< The maximum size that rendering will be performed at. + FfxDimensions2D displaySize; ///< The size of the presentation resolution targeted by the upscaling process. + FfxFsr2Interface callbacks; ///< A set of pointers to the backend implementation for FSR 2.0. + FfxDevice device; ///< The abstracted device which is passed to some callback functions. + + FfxFsr2Message fpMessage; ///< A pointer to a function that can recieve messages from the runtime. +} FfxFsr2ContextDescription; + +/// A structure encapsulating the parameters for dispatching the various passes +/// of FidelityFX Super Resolution 2. +/// +/// @ingroup FSR2 +typedef struct FfxFsr2DispatchDescription { + + FfxCommandList commandList; ///< The FfxCommandList to record FSR2 rendering commands into. + FfxResource color; ///< A FfxResource containing the color buffer for the current frame (at render resolution). + FfxResource depth; ///< A FfxResource containing 32bit depth values for the current frame (at render resolution). + FfxResource motionVectors; ///< A FfxResource containing 2-dimensional motion vectors (at render resolution if FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS is not set). + FfxResource exposure; ///< A optional FfxResource containing a 1x1 exposure value. + FfxResource reactive; ///< A optional FfxResource containing alpha value of reactive objects in the scene. + FfxResource transparencyAndComposition; ///< A optional FfxResource containing alpha value of special objects in the scene. + FfxResource output; ///< A FfxResource containing the output color buffer for the current frame (at presentation resolution). + FfxFloatCoords2D jitterOffset; ///< The subpixel jitter offset applied to the camera. + FfxFloatCoords2D motionVectorScale; ///< The scale factor to apply to motion vectors. + FfxDimensions2D renderSize; ///< The resolution that was used for rendering the input resources. + bool enableSharpening; ///< Enable an additional sharpening pass. + float sharpness; ///< The sharpness value between 0 and 1, where 0 is no additional sharpness and 1 is maximum additional sharpness. + float frameTimeDelta; ///< The time elapsed since the last frame (expressed in milliseconds). + float preExposure; ///< The pre exposure value (must be > 0.0f) + bool reset; ///< A boolean value which when set to true, indicates the camera has moved discontinuously. + float cameraNear; ///< The distance to the near plane of the camera. + float cameraFar; ///< The distance to the far plane of the camera. + float cameraFovAngleVertical; ///< The camera angle field of view in the vertical direction (expressed in radians). + float viewSpaceToMetersFactor; ///< The scale factor to convert view space units to meters + + // EXPERIMENTAL reactive mask generation parameters + bool enableAutoReactive; ///< A boolean value to indicate internal reactive autogeneration should be used + FfxResource colorOpaqueOnly; ///< A FfxResource containing the opaque only color buffer for the current frame (at render resolution). + float autoTcThreshold; ///< Cutoff value for TC + float autoTcScale; ///< A value to scale the transparency and composition mask + float autoReactiveScale; ///< A value to scale the reactive mask + float autoReactiveMax; ///< A value to clamp the reactive mask + + // -- GODOT start -- + float reprojectionMatrix[16]; ///< The matrix used for reprojecting pixels with invalid motion vectors by using the depth. + // -- GODOT end -- + +} FfxFsr2DispatchDescription; + +/// A structure encapsulating the parameters for automatic generation of a reactive mask +/// +/// @ingroup FSR2 +typedef struct FfxFsr2GenerateReactiveDescription { + + FfxCommandList commandList; ///< The FfxCommandList to record FSR2 rendering commands into. + FfxResource colorOpaqueOnly; ///< A FfxResource containing the opaque only color buffer for the current frame (at render resolution). + FfxResource colorPreUpscale; ///< A FfxResource containing the opaque+translucent color buffer for the current frame (at render resolution). + FfxResource outReactive; ///< A FfxResource containing the surface to generate the reactive mask into. + FfxDimensions2D renderSize; ///< The resolution that was used for rendering the input resources. + float scale; ///< A value to scale the output + float cutoffThreshold; ///< A threshold value to generate a binary reactive mask + float binaryValue; ///< A value to set for the binary reactive mask + uint32_t flags; ///< Flags to determine how to generate the reactive mask +} FfxFsr2GenerateReactiveDescription; + +/// A structure encapsulating the FidelityFX Super Resolution 2 context. +/// +/// This sets up an object which contains all persistent internal data and +/// resources that are required by FSR2. +/// +/// The FfxFsr2Context object should have a lifetime matching +/// your use of FSR2. Before destroying the FSR2 context care should be taken +/// to ensure the GPU is not accessing the resources created or used by FSR2. +/// It is therefore recommended that the GPU is idle before destroying the +/// FSR2 context. +/// +/// @ingroup FSR2 +typedef struct FfxFsr2Context { + + uint32_t data[FFX_FSR2_CONTEXT_SIZE]; ///< An opaque set of uint32_t which contain the data for the context. +} FfxFsr2Context; + +/// Create a FidelityFX Super Resolution 2 context from the parameters +/// programmed to the FfxFsr2CreateParams structure. +/// +/// The context structure is the main object used to interact with the FSR2 +/// API, and is responsible for the management of the internal resources used +/// by the FSR2 algorithm. When this API is called, multiple calls will be +/// made via the pointers contained in the callbacks structure. +/// These callbacks will attempt to retreive the device capabilities, and +/// create the internal resources, and pipelines required by FSR2's +/// frame-to-frame function. Depending on the precise configuration used when +/// creating the FfxFsr2Context a different set of resources and +/// pipelines might be requested via the callback functions. +/// +/// The flags included in the flags field of +/// FfxFsr2Context how match the configuration of your +/// application as well as the intended use of FSR2. It is important that these +/// flags are set correctly (as well as a correct programmed +/// FfxFsr2DispatchDescription) to ensure correct operation. It is +/// recommended to consult the overview documentation for further details on +/// how FSR2 should be integerated into an application. +/// +/// When the FfxFsr2Context is created, you should use the +/// ffxFsr2ContextDispatch function each frame where FSR2 +/// upscaling should be applied. See the documentation of +/// ffxFsr2ContextDispatch for more details. +/// +/// The FfxFsr2Context should be destroyed when use of it is +/// completed, typically when an application is unloaded or FSR2 upscaling is +/// disabled by a user. To destroy the FSR2 context you should call +/// ffxFsr2ContextDestroy. +/// +/// @param [out] context A pointer to a FfxFsr2Context structure to populate. +/// @param [in] contextDescription A pointer to a FfxFsr2ContextDescription structure. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// FFX_ERROR_CODE_NULL_POINTER The operation failed because either context or contextDescription was NULL. +/// @retval +/// FFX_ERROR_INCOMPLETE_INTERFACE The operation failed because the FfxFsr2ContextDescription.callbacks was not fully specified. +/// @retval +/// FFX_ERROR_BACKEND_API_ERROR The operation failed because of an error returned from the backend. +/// +/// @ingroup FSR2 +FFX_API FfxErrorCode ffxFsr2ContextCreate(FfxFsr2Context* context, const FfxFsr2ContextDescription* contextDescription); + +/// Dispatch the various passes that constitute FidelityFX Super Resolution 2. +/// +/// FSR2 is a composite effect, meaning that it is compromised of multiple +/// constituent passes (implemented as one or more clears, copies and compute +/// dispatches). The ffxFsr2ContextDispatch function is the +/// function which (via the use of the functions contained in the +/// callbacks field of the FfxFsr2Context +/// structure) utlimately generates the sequence of graphics API calls required +/// each frame. +/// +/// As with the creation of the FfxFsr2Context correctly +/// programming the FfxFsr2DispatchDescription is key to ensuring +/// the correct operation of FSR2. It is particularly important to ensure that +/// camera jitter is correctly applied to your application's projection matrix +/// (or camera origin for raytraced applications). FSR2 provides the +/// ffxFsr2GetJitterPhaseCount and +/// ffxFsr2GetJitterOffset entry points to help applications +/// correctly compute the camera jitter. Whatever jitter pattern is used by the +/// application it should be correctly programmed to the +/// jitterOffset field of the dispatchDescription +/// structure. For more guidance on camera jitter please consult the +/// documentation for ffxFsr2GetJitterOffset as well as the +/// accompanying overview documentation for FSR2. +/// +/// @param [in] context A pointer to a FfxFsr2Context structure. +/// @param [in] dispatchDescription A pointer to a FfxFsr2DispatchDescription structure. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// FFX_ERROR_CODE_NULL_POINTER The operation failed because either context or dispatchDescription was NULL. +/// @retval +/// FFX_ERROR_OUT_OF_RANGE The operation failed because dispatchDescription.renderSize was larger than the maximum render resolution. +/// @retval +/// FFX_ERROR_NULL_DEVICE The operation failed because the device inside the context was NULL. +/// @retval +/// FFX_ERROR_BACKEND_API_ERROR The operation failed because of an error returned from the backend. +/// +/// @ingroup FSR2 +FFX_API FfxErrorCode ffxFsr2ContextDispatch(FfxFsr2Context* context, const FfxFsr2DispatchDescription* dispatchDescription); + +/// A helper function generate a Reactive mask from an opaque only texure and one containing translucent objects. +/// +/// @param [in] context A pointer to a FfxFsr2Context structure. +/// @param [in] params A pointer to a FfxFsr2GenerateReactiveDescription structure +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// +/// @ingroup FSR2 +FFX_API FfxErrorCode ffxFsr2ContextGenerateReactiveMask(FfxFsr2Context* context, const FfxFsr2GenerateReactiveDescription* params); + +/// Destroy the FidelityFX Super Resolution context. +/// +/// @param [out] context A pointer to a FfxFsr2Context structure to destroy. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// FFX_ERROR_CODE_NULL_POINTER The operation failed because either context was NULL. +/// +/// @ingroup FSR2 +FFX_API FfxErrorCode ffxFsr2ContextDestroy(FfxFsr2Context* context); + +/// Get the upscale ratio from the quality mode. +/// +/// The following table enumerates the mapping of the quality modes to +/// per-dimension scaling ratios. +/// +/// Quality preset | Scale factor +/// ----------------------------------------------------- | ------------- +/// FFX_FSR2_QUALITY_MODE_QUALITY | 1.5x +/// FFX_FSR2_QUALITY_MODE_BALANCED | 1.7x +/// FFX_FSR2_QUALITY_MODE_PERFORMANCE | 2.0x +/// FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE | 3.0x +/// +/// Passing an invalid qualityMode will return 0.0f. +/// +/// @param [in] qualityMode The quality mode preset. +/// +/// @returns +/// The upscaling the per-dimension upscaling ratio for +/// qualityMode according to the table above. +/// +/// @ingroup FSR2 +FFX_API float ffxFsr2GetUpscaleRatioFromQualityMode(FfxFsr2QualityMode qualityMode); + +/// A helper function to calculate the rendering resolution from a target +/// resolution and desired quality level. +/// +/// This function applies the scaling factor returned by +/// ffxFsr2GetUpscaleRatioFromQualityMode to each dimension. +/// +/// @param [out] renderWidth A pointer to a uint32_t which will hold the calculated render resolution width. +/// @param [out] renderHeight A pointer to a uint32_t which will hold the calculated render resolution height. +/// @param [in] displayWidth The target display resolution width. +/// @param [in] displayHeight The target display resolution height. +/// @param [in] qualityMode The desired quality mode for FSR 2 upscaling. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// FFX_ERROR_INVALID_POINTER Either renderWidth or renderHeight was NULL. +/// @retval +/// FFX_ERROR_INVALID_ENUM An invalid quality mode was specified. +/// +/// @ingroup FSR2 +FFX_API FfxErrorCode ffxFsr2GetRenderResolutionFromQualityMode( + uint32_t* renderWidth, + uint32_t* renderHeight, + uint32_t displayWidth, + uint32_t displayHeight, + FfxFsr2QualityMode qualityMode); + +/// A helper function to calculate the jitter phase count from display +/// resolution. +/// +/// For more detailed information about the application of camera jitter to +/// your application's rendering please refer to the +/// ffxFsr2GetJitterOffset function. +/// +/// The table below shows the jitter phase count which this function +/// would return for each of the quality presets. +/// +/// Quality preset | Scale factor | Phase count +/// ----------------------------------------------------- | ------------- | --------------- +/// FFX_FSR2_QUALITY_MODE_QUALITY | 1.5x | 18 +/// FFX_FSR2_QUALITY_MODE_BALANCED | 1.7x | 23 +/// FFX_FSR2_QUALITY_MODE_PERFORMANCE | 2.0x | 32 +/// FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE | 3.0x | 72 +/// Custom | [1..n]x | ceil(8*n^2) +/// +/// @param [in] renderWidth The render resolution width. +/// @param [in] displayWidth The display resolution width. +/// +/// @returns +/// The jitter phase count for the scaling factor between renderWidth and displayWidth. +/// +/// @ingroup FSR2 +FFX_API int32_t ffxFsr2GetJitterPhaseCount(int32_t renderWidth, int32_t displayWidth); + +/// A helper function to calculate the subpixel jitter offset. +/// +/// FSR2 relies on the application to apply sub-pixel jittering while rendering. +/// This is typically included in the projection matrix of the camera. To make +/// the application of camera jitter simple, the FSR2 API provides a small set +/// of utility function which computes the sub-pixel jitter offset for a +/// particular frame within a sequence of separate jitter offsets. To begin, the +/// index within the jitter phase must be computed. To calculate the +/// sequence's length, you can call the ffxFsr2GetJitterPhaseCount +/// function. The index should be a value which is incremented each frame modulo +/// the length of the sequence computed by ffxFsr2GetJitterPhaseCount. +/// The index within the jitter phase is passed to +/// ffxFsr2GetJitterOffset via the index parameter. +/// +/// This function uses a Halton(2,3) sequence to compute the jitter offset. +/// The ultimate index used for the sequence is index % +/// phaseCount. +/// +/// It is important to understand that the values returned from the +/// ffxFsr2GetJitterOffset function are in unit pixel space, and +/// in order to composite this correctly into a projection matrix we must +/// convert them into projection offsets. This is done as per the pseudo code +/// listing which is shown below. +/// +/// const int32_t jitterPhaseCount = ffxFsr2GetJitterPhaseCount(renderWidth, displayWidth); +/// +/// float jitterX = 0; +/// float jitterY = 0; +/// ffxFsr2GetJitterOffset(&jitterX, &jitterY, index, jitterPhaseCount); +/// +/// const float jitterX = 2.0f * jitterX / (float)renderWidth; +/// const float jitterY = -2.0f * jitterY / (float)renderHeight; +/// const Matrix4 jitterTranslationMatrix = translateMatrix(Matrix3::identity, Vector3(jitterX, jitterY, 0)); +/// const Matrix4 jitteredProjectionMatrix = jitterTranslationMatrix * projectionMatrix; +/// +/// Jitter should be applied to all rendering. This includes opaque, alpha +/// transparent, and raytraced objects. For rasterized objects, the sub-pixel +/// jittering values calculated by the iffxFsr2GetJitterOffset +/// function can be applied to the camera projection matrix which is ultimately +/// used to perform transformations during vertex shading. For raytraced +/// rendering, the sub-pixel jitter should be applied to the ray's origin, +/// often the camera's position. +/// +/// Whether you elect to use the ffxFsr2GetJitterOffset function +/// or your own sequence generator, you must program the +/// jitterOffset field of the +/// FfxFsr2DispatchParameters structure in order to inform FSR2 +/// of the jitter offset that has been applied in order to render each frame. +/// +/// If not using the recommended ffxFsr2GetJitterOffset function, +/// care should be taken that your jitter sequence never generates a null vector; +/// that is value of 0 in both the X and Y dimensions. +/// +/// @param [out] outX A pointer to a float which will contain the subpixel jitter offset for the x dimension. +/// @param [out] outY A pointer to a float which will contain the subpixel jitter offset for the y dimension. +/// @param [in] index The index within the jitter sequence. +/// @param [in] phaseCount The length of jitter phase. See ffxFsr2GetJitterPhaseCount. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// FFX_ERROR_INVALID_POINTER Either outX or outY was NULL. +/// @retval +/// FFX_ERROR_INVALID_ARGUMENT Argument phaseCount must be greater than 0. +/// +/// @ingroup FSR2 +FFX_API FfxErrorCode ffxFsr2GetJitterOffset(float* outX, float* outY, int32_t index, int32_t phaseCount); + +/// A helper function to check if a resource is +/// FFX_FSR2_RESOURCE_IDENTIFIER_NULL. +/// +/// @param [in] resource A FfxResource. +/// +/// @returns +/// true The resource was not FFX_FSR2_RESOURCE_IDENTIFIER_NULL. +/// @returns +/// false The resource was FFX_FSR2_RESOURCE_IDENTIFIER_NULL. +/// +/// @ingroup FSR2 +FFX_API bool ffxFsr2ResourceIsNull(FfxResource resource); + +#if defined(__cplusplus) +} +#endif // #if defined(__cplusplus) diff --git a/thirdparty/amd-fsr2/ffx_fsr2_interface.h b/thirdparty/amd-fsr2/ffx_fsr2_interface.h new file mode 100644 index 000000000000..b6be9760a719 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_fsr2_interface.h @@ -0,0 +1,395 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#pragma once + +#include "ffx_assert.h" +#include "ffx_types.h" +#include "ffx_error.h" + +// Include the FSR2 resources defined in the HLSL code. This shared here to avoid getting out of sync. +#define FFX_CPU +#include "shaders/ffx_fsr2_resources.h" +#include "shaders/ffx_fsr2_common.h" + +#if defined(__cplusplus) +extern "C" { +#endif // #if defined(__cplusplus) + +FFX_FORWARD_DECLARE(FfxFsr2Interface); + +/// An enumeration of all the passes which constitute the FSR2 algorithm. +/// +/// FSR2 is implemented as a composite of several compute passes each +/// computing a key part of the final result. Each call to the +/// FfxFsr2ScheduleGpuJobFunc callback function will +/// correspond to a single pass included in FfxFsr2Pass. For a +/// more comprehensive description of each pass, please refer to the FSR2 +/// reference documentation. +/// +/// Please note in some cases e.g.: FFX_FSR2_PASS_ACCUMULATE +/// and FFX_FSR2_PASS_ACCUMULATE_SHARPEN either one pass or the +/// other will be used (they are mutually exclusive). The choice of which will +/// depend on the way the FfxFsr2Context is created and the +/// precise contents of FfxFsr2DispatchParamters each time a call +/// is made to ffxFsr2ContextDispatch. +/// +/// @ingroup FSR2 +typedef enum FfxFsr2Pass { + + FFX_FSR2_PASS_DEPTH_CLIP = 0, ///< A pass which performs depth clipping. + FFX_FSR2_PASS_RECONSTRUCT_PREVIOUS_DEPTH = 1, ///< A pass which performs reconstruction of previous frame's depth. + FFX_FSR2_PASS_LOCK = 2, ///< A pass which calculates pixel locks. + FFX_FSR2_PASS_ACCUMULATE = 3, ///< A pass which performs upscaling. + FFX_FSR2_PASS_ACCUMULATE_SHARPEN = 4, ///< A pass which performs upscaling when sharpening is used. + FFX_FSR2_PASS_RCAS = 5, ///< A pass which performs sharpening. + FFX_FSR2_PASS_COMPUTE_LUMINANCE_PYRAMID = 6, ///< A pass which generates the luminance mipmap chain for the current frame. + FFX_FSR2_PASS_GENERATE_REACTIVE = 7, ///< An optional pass to generate a reactive mask + FFX_FSR2_PASS_TCR_AUTOGENERATE = 8, ///< An optional pass to generate a texture-and-composition and reactive masks + + FFX_FSR2_PASS_COUNT ///< The number of passes performed by FSR2. +} FfxFsr2Pass; + +typedef enum FfxFsr2MsgType { + FFX_FSR2_MESSAGE_TYPE_ERROR = 0, + FFX_FSR2_MESSAGE_TYPE_WARNING = 1, + FFX_FSR2_MESSAGE_TYPE_COUNT +} FfxFsr2MsgType; + +/// Create and initialize the backend context. +/// +/// The callback function sets up the backend context for rendering. +/// It will create or reference the device and create required internal data structures. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] device The FfxDevice obtained by ffxGetDevice(DX12/VK/...). +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2CreateBackendContextFunc)( + FfxFsr2Interface* backendInterface, + FfxDevice device); + +/// Get a list of capabilities of the device. +/// +/// When creating an FfxFsr2Context it is desirable for the FSR2 +/// core implementation to be aware of certain characteristics of the platform +/// that is being targetted. This is because some optimizations which FSR2 +/// attempts to perform are more effective on certain classes of hardware than +/// others, or are not supported by older hardware. In order to avoid cases +/// where optimizations actually have the effect of decreasing performance, or +/// reduce the breadth of support provided by FSR2, FSR2 queries the +/// capabilities of the device to make such decisions. +/// +/// For target platforms with fixed hardware support you need not implement +/// this callback function by querying the device, but instead may hardcore +/// what features are available on the platform. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [out] outDeviceCapabilities The device capabilities structure to fill out. +/// @param [in] device The device to query for capabilities. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode(*FfxFsr2GetDeviceCapabilitiesFunc)( + FfxFsr2Interface* backendInterface, + FfxDeviceCapabilities* outDeviceCapabilities, + FfxDevice device); + +/// Destroy the backend context and dereference the device. +/// +/// This function is called when the FfxFsr2Context is destroyed. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode(*FfxFsr2DestroyBackendContextFunc)( + FfxFsr2Interface* backendInterface); + +/// Create a resource. +/// +/// This callback is intended for the backend to create internal resources. +/// +/// Please note: It is also possible that the creation of resources might +/// itself cause additional resources to be created by simply calling the +/// FfxFsr2CreateResourceFunc function pointer again. This is +/// useful when handling the initial creation of resources which must be +/// initialized. The flow in such a case would be an initial call to create the +/// CPU-side resource, another to create the GPU-side resource, and then a call +/// to schedule a copy render job to move the data between the two. Typically +/// this type of function call flow is only seen during the creation of an +/// FfxFsr2Context. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] createResourceDescription A pointer to a FfxCreateResourceDescription. +/// @param [out] outResource A pointer to a FfxResource object. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2CreateResourceFunc)( + FfxFsr2Interface* backendInterface, + const FfxCreateResourceDescription* createResourceDescription, + FfxResourceInternal* outResource); + +/// Register a resource in the backend for the current frame. +/// +/// Since FSR2 and the backend are not aware how many different +/// resources will get passed to FSR2 over time, it's not safe +/// to register all resources simultaneously in the backend. +/// Also passed resources may not be valid after the dispatch call. +/// As a result it's safest to register them as FfxResourceInternal +/// and clear them at the end of the dispatch call. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] inResource A pointer to a FfxResource. +/// @param [out] outResource A pointer to a FfxResourceInternal object. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode(*FfxFsr2RegisterResourceFunc)( + FfxFsr2Interface* backendInterface, + const FfxResource* inResource, + FfxResourceInternal* outResource); + +/// Unregister all temporary FfxResourceInternal from the backend. +/// +/// Unregister FfxResourceInternal referencing resources passed to +/// a function as a parameter. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode(*FfxFsr2UnregisterResourcesFunc)( + FfxFsr2Interface* backendInterface); + +/// Retrieve a FfxResourceDescription matching a +/// FfxResource structure. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] resource A pointer to a FfxResource object. +/// +/// @returns +/// A description of the resource. +/// +/// @ingroup FSR2 +typedef FfxResourceDescription (*FfxFsr2GetResourceDescriptionFunc)( + FfxFsr2Interface* backendInterface, + FfxResourceInternal resource); + +/// Destroy a resource +/// +/// This callback is intended for the backend to release an internal resource. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] resource A pointer to a FfxResource object. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2DestroyResourceFunc)( + FfxFsr2Interface* backendInterface, + FfxResourceInternal resource); + +/// Create a render pipeline. +/// +/// A rendering pipeline contains the shader as well as resource bindpoints +/// and samplers. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] pass The identifier for the pass. +/// @param [in] pipelineDescription A pointer to a FfxPipelineDescription describing the pipeline to be created. +/// @param [out] outPipeline A pointer to a FfxPipelineState structure which should be populated. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2CreatePipelineFunc)( + FfxFsr2Interface* backendInterface, + FfxFsr2Pass pass, + const FfxPipelineDescription* pipelineDescription, + FfxPipelineState* outPipeline); + +/// Destroy a render pipeline. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [out] pipeline A pointer to a FfxPipelineState structure which should be released. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2DestroyPipelineFunc)( + FfxFsr2Interface* backendInterface, + FfxPipelineState* pipeline); + +/// Schedule a render job to be executed on the next call of +/// FfxFsr2ExecuteGpuJobsFunc. +/// +/// Render jobs can perform one of three different tasks: clear, copy or +/// compute dispatches. +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] job A pointer to a FfxGpuJobDescription structure. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2ScheduleGpuJobFunc)( + FfxFsr2Interface* backendInterface, + const FfxGpuJobDescription* job); + +/// Execute scheduled render jobs on the comandList provided. +/// +/// The recording of the graphics API commands should take place in this +/// callback function, the render jobs which were previously enqueued (via +/// callbacks made to FfxFsr2ScheduleGpuJobFunc) should be +/// processed in the order they were received. Advanced users might choose to +/// reorder the rendering jobs, but should do so with care to respect the +/// resource dependencies. +/// +/// Depending on the precise contents of FfxFsr2DispatchDescription a +/// different number of render jobs might have previously been enqueued (for +/// example if sharpening is toggled on and off). +/// +/// @param [in] backendInterface A pointer to the backend interface. +/// @param [in] commandList A pointer to a FfxCommandList structure. +/// +/// @retval +/// FFX_OK The operation completed successfully. +/// @retval +/// Anything else The operation failed. +/// +/// @ingroup FSR2 +typedef FfxErrorCode (*FfxFsr2ExecuteGpuJobsFunc)( + FfxFsr2Interface* backendInterface, + FfxCommandList commandList); + +/// Pass a string message +/// +/// Used for debug messages. +/// +/// @param [in] type The type of message. +/// @param [in] message A string message to pass. +/// +/// +/// @ingroup FSR2 +typedef void(*FfxFsr2Message)( + FfxFsr2MsgType type, + const wchar_t* message); + +/// A structure encapsulating the interface between the core implentation of +/// the FSR2 algorithm and any graphics API that it should ultimately call. +/// +/// This set of functions serves as an abstraction layer between FSR2 and the +/// API used to implement it. While FSR2 ships with backends for DirectX12 and +/// Vulkan, it is possible to implement your own backend for other platforms or +/// which sits ontop of your engine's own abstraction layer. For details on the +/// expectations of what each function should do you should refer the +/// description of the following function pointer types: +/// +/// FfxFsr2CreateDeviceFunc +/// FfxFsr2GetDeviceCapabilitiesFunc +/// FfxFsr2DestroyDeviceFunc +/// FfxFsr2CreateResourceFunc +/// FfxFsr2GetResourceDescriptionFunc +/// FfxFsr2DestroyResourceFunc +/// FfxFsr2CreatePipelineFunc +/// FfxFsr2DestroyPipelineFunc +/// FfxFsr2ScheduleGpuJobFunc +/// FfxFsr2ExecuteGpuJobsFunc +/// +/// Depending on the graphics API that is abstracted by the backend, it may be +/// required that the backend is to some extent stateful. To ensure that +/// applications retain full control to manage the memory used by FSR2, the +/// scratchBuffer and scratchBufferSize fields are +/// provided. A backend should provide a means of specifying how much scratch +/// memory is required for its internal implementation (e.g: via a function +/// or constant value). The application is that responsible for allocating that +/// memory and providing it when setting up the FSR2 backend. Backends provided +/// with FSR2 do not perform dynamic memory allocations, and instead +/// suballocate all memory from the scratch buffers provided. +/// +/// The scratchBuffer and scratchBufferSize fields +/// should be populated according to the requirements of each backend. For +/// example, if using the DirectX 12 backend you should call the +/// ffxFsr2GetScratchMemorySizeDX12 function. It is not required +/// that custom backend implementations use a scratch buffer. +/// +/// @ingroup FSR2 +typedef struct FfxFsr2Interface { + + FfxFsr2CreateBackendContextFunc fpCreateBackendContext; ///< A callback function to create and initialize the backend context. + FfxFsr2GetDeviceCapabilitiesFunc fpGetDeviceCapabilities; ///< A callback function to query device capabilites. + FfxFsr2DestroyBackendContextFunc fpDestroyBackendContext; ///< A callback function to destroy the backendcontext. This also dereferences the device. + FfxFsr2CreateResourceFunc fpCreateResource; ///< A callback function to create a resource. + FfxFsr2RegisterResourceFunc fpRegisterResource; ///< A callback function to register an external resource. + FfxFsr2UnregisterResourcesFunc fpUnregisterResources; ///< A callback function to unregister external resource. + FfxFsr2GetResourceDescriptionFunc fpGetResourceDescription; ///< A callback function to retrieve a resource description. + FfxFsr2DestroyResourceFunc fpDestroyResource; ///< A callback function to destroy a resource. + FfxFsr2CreatePipelineFunc fpCreatePipeline; ///< A callback function to create a render or compute pipeline. + FfxFsr2DestroyPipelineFunc fpDestroyPipeline; ///< A callback function to destroy a render or compute pipeline. + FfxFsr2ScheduleGpuJobFunc fpScheduleGpuJob; ///< A callback function to schedule a render job. + FfxFsr2ExecuteGpuJobsFunc fpExecuteGpuJobs; ///< A callback function to execute all queued render jobs. + + void* scratchBuffer; ///< A preallocated buffer for memory utilized internally by the backend. + size_t scratchBufferSize; ///< Size of the buffer pointed to by scratchBuffer. +} FfxFsr2Interface; + +#if defined(__cplusplus) +} +#endif // #if defined(__cplusplus) diff --git a/thirdparty/amd-fsr2/ffx_fsr2_maximum_bias.h b/thirdparty/amd-fsr2/ffx_fsr2_maximum_bias.h new file mode 100644 index 000000000000..5fdbd0cdcd00 --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_fsr2_maximum_bias.h @@ -0,0 +1,46 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// @internal + +#pragma once + +static const int FFX_FSR2_MAXIMUM_BIAS_TEXTURE_WIDTH = 16; +static const int FFX_FSR2_MAXIMUM_BIAS_TEXTURE_HEIGHT = 16; +static const float ffxFsr2MaximumBias[] = { + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.876f, 1.809f, 1.772f, 1.753f, 1.748f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.869f, 1.801f, 1.764f, 1.745f, 1.739f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.976f, 1.841f, 1.774f, 1.737f, 1.716f, 1.71f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.914f, 1.784f, 1.716f, 1.673f, 1.649f, 1.641f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.793f, 1.676f, 1.604f, 1.562f, 1.54f, 1.533f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.802f, 1.619f, 1.536f, 1.492f, 1.467f, 1.454f, 1.449f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.812f, 1.575f, 1.496f, 1.456f, 1.432f, 1.416f, 1.408f, 1.405f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.555f, 1.479f, 1.438f, 1.413f, 1.398f, 1.387f, 1.381f, 1.379f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.812f, 1.555f, 1.474f, 1.43f, 1.404f, 1.387f, 1.376f, 1.368f, 1.363f, 1.362f, + 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.802f, 1.575f, 1.479f, 1.43f, 1.401f, 1.382f, 1.369f, 1.36f, 1.354f, 1.351f, 1.35f, + 2.0f, 2.0f, 1.976f, 1.914f, 1.793f, 1.619f, 1.496f, 1.438f, 1.404f, 1.382f, 1.367f, 1.357f, 1.349f, 1.344f, 1.341f, 1.34f, + 1.876f, 1.869f, 1.841f, 1.784f, 1.676f, 1.536f, 1.456f, 1.413f, 1.387f, 1.369f, 1.357f, 1.347f, 1.341f, 1.336f, 1.333f, 1.332f, + 1.809f, 1.801f, 1.774f, 1.716f, 1.604f, 1.492f, 1.432f, 1.398f, 1.376f, 1.36f, 1.349f, 1.341f, 1.335f, 1.33f, 1.328f, 1.327f, + 1.772f, 1.764f, 1.737f, 1.673f, 1.562f, 1.467f, 1.416f, 1.387f, 1.368f, 1.354f, 1.344f, 1.336f, 1.33f, 1.326f, 1.323f, 1.323f, + 1.753f, 1.745f, 1.716f, 1.649f, 1.54f, 1.454f, 1.408f, 1.381f, 1.363f, 1.351f, 1.341f, 1.333f, 1.328f, 1.323f, 1.321f, 1.32f, + 1.748f, 1.739f, 1.71f, 1.641f, 1.533f, 1.449f, 1.405f, 1.379f, 1.362f, 1.35f, 1.34f, 1.332f, 1.327f, 1.323f, 1.32f, 1.319f, + +}; diff --git a/thirdparty/amd-fsr2/ffx_fsr2_private.h b/thirdparty/amd-fsr2/ffx_fsr2_private.h new file mode 100644 index 000000000000..0face069b60a --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_fsr2_private.h @@ -0,0 +1,86 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#pragma once + +// Constants for FSR2 DX12 dispatches. Must be kept in sync with cbFSR2 in ffx_fsr2_callbacks_hlsl.h +typedef struct Fsr2Constants { + + int32_t renderSize[2]; + int32_t maxRenderSize[2]; + int32_t displaySize[2]; + int32_t inputColorResourceDimensions[2]; + int32_t lumaMipDimensions[2]; + int32_t lumaMipLevelToUse; + int32_t frameIndex; + + float deviceToViewDepth[4]; + float jitterOffset[2]; + float motionVectorScale[2]; + float downscaleFactor[2]; + float motionVectorJitterCancellation[2]; + float preExposure; + float previousFramePreExposure; + float tanHalfFOV; + float jitterPhaseCount; + float deltaTime; + float dynamicResChangeFactor; + float viewSpaceToMetersFactor; + + // -- GODOT start -- + float pad; + float reprojectionMatrix[16]; + // -- GODOT end -- +} Fsr2Constants; + +struct FfxFsr2ContextDescription; +struct FfxDeviceCapabilities; +struct FfxPipelineState; +struct FfxResource; + +// FfxFsr2Context_Private +// The private implementation of the FSR2 context. +typedef struct FfxFsr2Context_Private { + + FfxFsr2ContextDescription contextDescription; + Fsr2Constants constants; + FfxDevice device; + FfxDeviceCapabilities deviceCapabilities; + FfxPipelineState pipelineDepthClip; + FfxPipelineState pipelineReconstructPreviousDepth; + FfxPipelineState pipelineLock; + FfxPipelineState pipelineAccumulate; + FfxPipelineState pipelineAccumulateSharpen; + FfxPipelineState pipelineRCAS; + FfxPipelineState pipelineComputeLuminancePyramid; + FfxPipelineState pipelineGenerateReactive; + FfxPipelineState pipelineTcrAutogenerate; + + // 2 arrays of resources, as e.g. FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS will use different resources when bound as SRV vs when bound as UAV + FfxResourceInternal srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_COUNT]; + FfxResourceInternal uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_COUNT]; + + bool firstExecution; + bool refreshPipelineStates; + uint32_t resourceFrameIndex; + float previousJitterOffset[2]; + int32_t jitterPhaseCountRemaining; +} FfxFsr2Context_Private; diff --git a/thirdparty/amd-fsr2/ffx_types.h b/thirdparty/amd-fsr2/ffx_types.h new file mode 100644 index 000000000000..8b65219b50ea --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_types.h @@ -0,0 +1,367 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#pragma once + +#include +// -- GODOT start -- +#include +// -- GODOT end -- + +#if defined (FFX_GCC) +/// FidelityFX exported functions +#define FFX_API +#else +/// FidelityFX exported functions +#define FFX_API __declspec(dllexport) +#endif // #if defined (FFX_GCC) + +/// Maximum supported number of simultaneously bound SRVs. +#define FFX_MAX_NUM_SRVS 16 + +/// Maximum supported number of simultaneously bound UAVs. +#define FFX_MAX_NUM_UAVS 8 + +/// Maximum number of constant buffers bound. +#define FFX_MAX_NUM_CONST_BUFFERS 2 + +/// Maximum size of bound constant buffers. +#define FFX_MAX_CONST_SIZE 64 + +/// Off by default warnings +#if defined(_MSC_VER) +#pragma warning(disable : 4365 4710 4820 5039) +#elif defined(__clang__) +#pragma clang diagnostic ignored "-Wunused-parameter" +#pragma clang diagnostic ignored "-Wmissing-field-initializers" +#pragma clang diagnostic ignored "-Wsign-compare" +#pragma clang diagnostic ignored "-Wunused-function" +#pragma clang diagnostic ignored "-Wignored-qualifiers" +#elif defined(__GNUC__) +#pragma GCC diagnostic ignored "-Wunused-function" +#endif + +#ifdef __cplusplus +extern "C" { +#endif // #ifdef __cplusplus + +/// An enumeration of surface formats. +typedef enum FfxSurfaceFormat { + + FFX_SURFACE_FORMAT_UNKNOWN, ///< Unknown format + FFX_SURFACE_FORMAT_R32G32B32A32_TYPELESS, ///< 32 bit per channel, 4 channel typeless format + FFX_SURFACE_FORMAT_R32G32B32A32_FLOAT, ///< 32 bit per channel, 4 channel float format + FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT, ///< 16 bit per channel, 4 channel float format + FFX_SURFACE_FORMAT_R16G16B16A16_UNORM, ///< 16 bit per channel, 4 channel unsigned normalized format + FFX_SURFACE_FORMAT_R32G32_FLOAT, ///< 32 bit per channel, 2 channel float format + FFX_SURFACE_FORMAT_R32_UINT, ///< 32 bit per channel, 1 channel float format + FFX_SURFACE_FORMAT_R8G8B8A8_TYPELESS, ///< 8 bit per channel, 4 channel float format + FFX_SURFACE_FORMAT_R8G8B8A8_UNORM, ///< 8 bit per channel, 4 channel unsigned normalized format + FFX_SURFACE_FORMAT_R11G11B10_FLOAT, ///< 32 bit 3 channel float format + FFX_SURFACE_FORMAT_R16G16_FLOAT, ///< 16 bit per channel, 2 channel float format + FFX_SURFACE_FORMAT_R16G16_UINT, ///< 16 bit per channel, 2 channel unsigned int format + FFX_SURFACE_FORMAT_R16_FLOAT, ///< 16 bit per channel, 1 channel float format + FFX_SURFACE_FORMAT_R16_UINT, ///< 16 bit per channel, 1 channel unsigned int format + FFX_SURFACE_FORMAT_R16_UNORM, ///< 16 bit per channel, 1 channel unsigned normalized format + FFX_SURFACE_FORMAT_R16_SNORM, ///< 16 bit per channel, 1 channel signed normalized format + FFX_SURFACE_FORMAT_R8_UNORM, ///< 8 bit per channel, 1 channel unsigned normalized format + FFX_SURFACE_FORMAT_R8_UINT, ///< 8 bit per channel, 1 channel unsigned int format + FFX_SURFACE_FORMAT_R8G8_UNORM, ///< 8 bit per channel, 2 channel unsigned normalized format + FFX_SURFACE_FORMAT_R32_FLOAT ///< 32 bit per channel, 1 channel float format +} FfxSurfaceFormat; + +/// An enumeration of resource usage. +typedef enum FfxResourceUsage { + + FFX_RESOURCE_USAGE_READ_ONLY = 0, ///< No usage flags indicate a resource is read only. + FFX_RESOURCE_USAGE_RENDERTARGET = (1<<0), ///< Indicates a resource will be used as render target. + FFX_RESOURCE_USAGE_UAV = (1<<1), ///< Indicates a resource will be used as UAV. +} FfxResourceUsage; + +/// An enumeration of resource states. +typedef enum FfxResourceStates { + + FFX_RESOURCE_STATE_UNORDERED_ACCESS = (1<<0), ///< Indicates a resource is in the state to be used as UAV. + FFX_RESOURCE_STATE_COMPUTE_READ = (1 << 1), ///< Indicates a resource is in the state to be read by compute shaders. + FFX_RESOURCE_STATE_COPY_SRC = (1 << 2), ///< Indicates a resource is in the state to be used as source in a copy command. + FFX_RESOURCE_STATE_COPY_DEST = (1 << 3), ///< Indicates a resource is in the state to be used as destination in a copy command. + FFX_RESOURCE_STATE_GENERIC_READ = (FFX_RESOURCE_STATE_COPY_SRC | FFX_RESOURCE_STATE_COMPUTE_READ), ///< Indicates a resource is in generic (slow) read state. +} FfxResourceStates; + +/// An enumeration of surface dimensions. +typedef enum FfxResourceDimension { + + FFX_RESOURCE_DIMENSION_TEXTURE_1D, ///< A resource with a single dimension. + FFX_RESOURCE_DIMENSION_TEXTURE_2D, ///< A resource with two dimensions. +} FfxResourceDimension; + +/// An enumeration of surface dimensions. +typedef enum FfxResourceFlags { + + FFX_RESOURCE_FLAGS_NONE = 0, ///< No flags. + FFX_RESOURCE_FLAGS_ALIASABLE = (1<<0), ///< A bit indicating a resource does not need to persist across frames. +} FfxResourceFlags; + +/// An enumeration of all resource view types. +typedef enum FfxResourceViewType { + + FFX_RESOURCE_VIEW_UNORDERED_ACCESS, ///< The resource view is an unordered access view (UAV). + FFX_RESOURCE_VIEW_SHADER_READ, ///< The resource view is a shader resource view (SRV). +} FfxResourceViewType; + +/// The type of filtering to perform when reading a texture. +typedef enum FfxFilterType { + + FFX_FILTER_TYPE_POINT, ///< Point sampling. + FFX_FILTER_TYPE_LINEAR ///< Sampling with interpolation. +} FfxFilterType; + +/// An enumeration of all supported shader models. +typedef enum FfxShaderModel { + + FFX_SHADER_MODEL_5_1, ///< Shader model 5.1. + FFX_SHADER_MODEL_6_0, ///< Shader model 6.0. + FFX_SHADER_MODEL_6_1, ///< Shader model 6.1. + FFX_SHADER_MODEL_6_2, ///< Shader model 6.2. + FFX_SHADER_MODEL_6_3, ///< Shader model 6.3. + FFX_SHADER_MODEL_6_4, ///< Shader model 6.4. + FFX_SHADER_MODEL_6_5, ///< Shader model 6.5. + FFX_SHADER_MODEL_6_6, ///< Shader model 6.6. + FFX_SHADER_MODEL_6_7, ///< Shader model 6.7. +} FfxShaderModel; + +// An enumeration for different resource types +typedef enum FfxResourceType { + + FFX_RESOURCE_TYPE_BUFFER, ///< The resource is a buffer. + FFX_RESOURCE_TYPE_TEXTURE1D, ///< The resource is a 1-dimensional texture. + FFX_RESOURCE_TYPE_TEXTURE2D, ///< The resource is a 2-dimensional texture. + FFX_RESOURCE_TYPE_TEXTURE3D, ///< The resource is a 3-dimensional texture. +} FfxResourceType; + +/// An enumeration for different heap types +typedef enum FfxHeapType { + + FFX_HEAP_TYPE_DEFAULT = 0, ///< Local memory. + FFX_HEAP_TYPE_UPLOAD ///< Heap used for uploading resources. +} FfxHeapType; + +/// An enumberation for different render job types +typedef enum FfxGpuJobType { + + FFX_GPU_JOB_CLEAR_FLOAT = 0, ///< The GPU job is performing a floating-point clear. + FFX_GPU_JOB_COPY = 1, ///< The GPU job is performing a copy. + FFX_GPU_JOB_COMPUTE = 2, ///< The GPU job is performing a compute dispatch. +} FfxGpuJobType; + +/// A typedef representing the graphics device. +typedef void* FfxDevice; + +/// A typedef representing a command list or command buffer. +typedef void* FfxCommandList; + +/// A typedef for a root signature. +typedef void* FfxRootSignature; + +/// A typedef for a pipeline state object. +typedef void* FfxPipeline; + +/// A structure encapasulating a collection of device capabilities. +typedef struct FfxDeviceCapabilities { + + FfxShaderModel minimumSupportedShaderModel; ///< The minimum shader model supported by the device. + uint32_t waveLaneCountMin; ///< The minimum supported wavefront width. + uint32_t waveLaneCountMax; ///< The maximum supported wavefront width. + bool fp16Supported; ///< The device supports FP16 in hardware. + bool raytracingSupported; ///< The device supports raytracing. +} FfxDeviceCapabilities; + +/// A structure encapsulating a 2-dimensional point, using 32bit unsigned integers. +typedef struct FfxDimensions2D { + + uint32_t width; ///< The width of a 2-dimensional range. + uint32_t height; ///< The height of a 2-dimensional range. +} FfxDimensions2D; + +/// A structure encapsulating a 2-dimensional point, +typedef struct FfxIntCoords2D { + + int32_t x; ///< The x coordinate of a 2-dimensional point. + int32_t y; ///< The y coordinate of a 2-dimensional point. +} FfxIntCoords2D; + +/// A structure encapsulating a 2-dimensional set of floating point coordinates. +typedef struct FfxFloatCoords2D { + + float x; ///< The x coordinate of a 2-dimensional point. + float y; ///< The y coordinate of a 2-dimensional point. +} FfxFloatCoords2D; + +/// A structure describing a resource. +typedef struct FfxResourceDescription { + + FfxResourceType type; ///< The type of the resource. + FfxSurfaceFormat format; ///< The surface format. + uint32_t width; ///< The width of the resource. + uint32_t height; ///< The height of the resource. + uint32_t depth; ///< The depth of the resource. + uint32_t mipCount; ///< Number of mips (or 0 for full mipchain). + FfxResourceFlags flags; ///< A set of FfxResourceFlags flags. +} FfxResourceDescription; + +/// An outward facing structure containing a resource +typedef struct FfxResource { + void* resource; ///< pointer to the resource. + wchar_t name[64]; + FfxResourceDescription description; + FfxResourceStates state; + bool isDepth; + uint64_t descriptorData; +} FfxResource; + +/// An internal structure containing a handle to a resource and resource views +typedef struct FfxResourceInternal { + int32_t internalIndex; ///< The index of the resource. +} FfxResourceInternal; + + +/// A structure defining a resource bind point +typedef struct FfxResourceBinding +{ + uint32_t slotIndex; + uint32_t resourceIdentifier; + wchar_t name[64]; +}FfxResourceBinding; + +/// A structure encapsulating a single pass of an algorithm. +typedef struct FfxPipelineState { + + FfxRootSignature rootSignature; ///< The pipelines rootSignature + FfxPipeline pipeline; ///< The pipeline object + uint32_t uavCount; ///< Count of UAVs used in this pipeline + uint32_t srvCount; ///< Count of SRVs used in this pipeline + uint32_t constCount; ///< Count of constant buffers used in this pipeline + + FfxResourceBinding uavResourceBindings[FFX_MAX_NUM_UAVS]; ///< Array of ResourceIdentifiers bound as UAVs + FfxResourceBinding srvResourceBindings[FFX_MAX_NUM_SRVS]; ///< Array of ResourceIdentifiers bound as SRVs + FfxResourceBinding cbResourceBindings[FFX_MAX_NUM_CONST_BUFFERS]; ///< Array of ResourceIdentifiers bound as CBs +} FfxPipelineState; + +/// A structure containing the data required to create a resource. +typedef struct FfxCreateResourceDescription { + + FfxHeapType heapType; ///< The heap type to hold the resource, typically FFX_HEAP_TYPE_DEFAULT. + FfxResourceDescription resourceDescription; ///< A resource description. + FfxResourceStates initalState; ///< The initial resource state. + uint32_t initDataSize; ///< Size of initial data buffer. + void* initData; ///< Buffer containing data to fill the resource. + const wchar_t* name; ///< Name of the resource. + FfxResourceUsage usage; ///< Resource usage flags. + uint32_t id; ///< Internal resource ID. +} FfxCreateResourceDescription; + +/// A structure containing the description used to create a +/// FfxPipeline structure. +/// +/// A pipeline is the name given to a shader and the collection of state that +/// is required to dispatch it. In the context of FSR2 and its architecture +/// this means that a FfxPipelineDescription will map to either a +/// monolithic object in an explicit API (such as a +/// PipelineStateObject in DirectX 12). Or a shader and some +/// ancillary API objects (in something like DirectX 11). +/// +/// The contextFlags field contains a copy of the flags passed +/// to ffxFsr2ContextCreate via the flags field of +/// the FfxFsr2InitializationParams structure. These flags are +/// used to determine which permutation of a pipeline for a specific +/// FfxFsr2Pass should be used to implement the features required +/// by each application, as well as to acheive the best performance on specific +/// target hardware configurations. +/// +/// When using one of the provided backends for FSR2 (such as DirectX 12 or +/// Vulkan) the data required to create a pipeline is compiled offline and +/// included into the backend library that you are using. For cases where the +/// backend interface is overriden by providing custom callback function +/// implementations care should be taken to respect the contents of the +/// contextFlags field in order to correctly support the options +/// provided by FSR2, and acheive best performance. +/// +/// @ingroup FSR2 +typedef struct FfxPipelineDescription { + + uint32_t contextFlags; ///< A collection of FfxFsr2InitializationFlagBits which were passed to the context. + FfxFilterType* samplers; ///< Array of static samplers. + size_t samplerCount; ///< The number of samples contained inside samplers. + const uint32_t* rootConstantBufferSizes; ///< Array containing the sizes of the root constant buffers (count of 32 bit elements). + uint32_t rootConstantBufferCount; ///< The number of root constants contained within rootConstantBufferSizes. +} FfxPipelineDescription; + +/// A structure containing a constant buffer. +typedef struct FfxConstantBuffer { + + uint32_t uint32Size; ///< Size of 32 bit chunks used in the constant buffer + uint32_t data[FFX_MAX_CONST_SIZE]; ///< Constant buffer data +}FfxConstantBuffer; + +/// A structure describing a clear render job. +typedef struct FfxClearFloatJobDescription { + + float color[4]; ///< The clear color of the resource. + FfxResourceInternal target; ///< The resource to be cleared. +} FfxClearFloatJobDescription; + +/// A structure describing a compute render job. +typedef struct FfxComputeJobDescription { + + FfxPipelineState pipeline; ///< Compute pipeline for the render job. + uint32_t dimensions[3]; ///< Dispatch dimensions. + FfxResourceInternal srvs[FFX_MAX_NUM_SRVS]; ///< SRV resources to be bound in the compute job. + wchar_t srvNames[FFX_MAX_NUM_SRVS][64]; + FfxResourceInternal uavs[FFX_MAX_NUM_UAVS]; ///< UAV resources to be bound in the compute job. + uint32_t uavMip[FFX_MAX_NUM_UAVS]; ///< Mip level of UAV resources to be bound in the compute job. + wchar_t uavNames[FFX_MAX_NUM_UAVS][64]; + FfxConstantBuffer cbs[FFX_MAX_NUM_CONST_BUFFERS]; ///< Constant buffers to be bound in the compute job. + wchar_t cbNames[FFX_MAX_NUM_CONST_BUFFERS][64]; + uint32_t cbSlotIndex[FFX_MAX_NUM_CONST_BUFFERS]; ///< Slot index in the descriptor table +} FfxComputeJobDescription; + +/// A structure describing a copy render job. +typedef struct FfxCopyJobDescription +{ + FfxResourceInternal src; ///< Source resource for the copy. + FfxResourceInternal dst; ///< Destination resource for the copy. +} FfxCopyJobDescription; + +/// A structure describing a single render job. +typedef struct FfxGpuJobDescription{ + + FfxGpuJobType jobType; ///< Type of the job. + + union { + FfxClearFloatJobDescription clearJobDescriptor; ///< Clear job descriptor. Valid when jobType is FFX_RENDER_JOB_CLEAR_FLOAT. + FfxCopyJobDescription copyJobDescriptor; ///< Copy job descriptor. Valid when jobType is FFX_RENDER_JOB_COPY. + FfxComputeJobDescription computeJobDescriptor; ///< Compute job descriptor. Valid when jobType is FFX_RENDER_JOB_COMPUTE. + }; +} FfxGpuJobDescription; + +#ifdef __cplusplus +} +#endif // #ifdef __cplusplus diff --git a/thirdparty/amd-fsr2/ffx_util.h b/thirdparty/amd-fsr2/ffx_util.h new file mode 100644 index 000000000000..ca4324ea832e --- /dev/null +++ b/thirdparty/amd-fsr2/ffx_util.h @@ -0,0 +1,78 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#pragma once + +#include "ffx_types.h" + +/// The value of Pi. +const float FFX_PI = 3.141592653589793f; + +/// An epsilon value for floating point numbers. +const float FFX_EPSILON = 1e-06f; + +/// Helper macro to create the version number. +#define FFX_MAKE_VERSION(major, minor, patch) ((major << 22) | (minor << 12) | patch) + +///< Use this to specify no version. +#define FFX_UNSPECIFIED_VERSION 0xFFFFAD00 + +/// Helper macro to avoid warnings about unused variables. +#define FFX_UNUSED(x) ((void)(x)) + +/// Helper macro to align an integer to the specified power of 2 boundary +#define FFX_ALIGN_UP(x, y) (((x) + ((y)-1)) & ~((y)-1)) + +/// Helper macro to check if a value is aligned. +#define FFX_IS_ALIGNED(x) (((x) != 0) && ((x) & ((x)-1))) + +/// Helper macro to stringify a value. +#define FFX_STR(s) FFX_XSTR(s) +#define FFX_XSTR(s) #s + +/// Helper macro to forward declare a structure. +#define FFX_FORWARD_DECLARE(x) typedef struct x x + +/// Helper macro to return the maximum of two values. +#define FFX_MAXIMUM(x, y) (((x) > (y)) ? (x) : (y)) + +/// Helper macro to return the minimum of two values. +#define FFX_MINIMUM(x, y) (((x) < (y)) ? (x) : (y)) + +/// Helper macro to do safe free on a pointer. +#define FFX_SAFE_FREE(x) \ + if (x) \ + free(x) + +/// Helper macro to return the abs of an integer value. +#define FFX_ABSOLUTE(x) (((x) < 0) ? (-(x)) : (x)) + +/// Helper macro to return sign of a value. +#define FFX_SIGN(x) (((x) < 0) ? -1 : 1) + +/// Helper macro to work out the number of elements in an array. +#define FFX_ARRAY_ELEMENTS(x) (int32_t)((sizeof(x) / sizeof(0 [x])) / ((size_t)(!(sizeof(x) % sizeof(0 [x]))))) + +/// The maximum length of a path that can be specified to the FidelityFX API. +#define FFX_MAXIMUM_PATH (260) + +/// Helper macro to check if the specified key is set in a bitfield. +#define FFX_CONTAINS_FLAG(options, key) ((options & key) == key) diff --git a/thirdparty/amd-fsr2/patches/godot-changes.patch b/thirdparty/amd-fsr2/patches/godot-changes.patch new file mode 100644 index 0000000000000000000000000000000000000000..513d8a5a29530ba70954a7e4ad2f7b16da1e6845 GIT binary patch literal 21350 zcmeI4ZExGi5y$tf3-mh>cW-oc?D+Yi*B0m6ahf(xTwBStJrshhhuA&aab-JodbqFN z_W#?_iWDW9q)Z7toIpq-m&@Im*_r>&%r5oc|FqqO>*|Qyl8z6$y3={nJ=gEKu6nNR zZgkz#^=Fkf-IYp~^u(>o)>OjN53VsNS#~WwaidnwDrn`G?iZrsT77oZOXODErd!qb zs@qY?hFf>rZdu2U+tO86S6$s(bGy2Ve9aYqXT#AFJT6tlAMGvrqdHE%ao>o#C2_Uo zp6L^*eLG(jXOZrHGlzmp#QU2S~jd7|fGE*A&oX}XWxsoNL*@7)V`lBjj<$Kl>{5V_yDf9d+% z)4QzS-wkRGb4aaQeQv~2LuZkHE|dh>rD@BOz12&~-A2snqb=CcKPC5$|DX$9uS-`d zq4%!W67-1cme+_IKSn*>rG;yKKIkq_FnZR5je5{G#`apzgD3ZXH1DL#YvPcSLrL(& zUmdwU_ZOA^*}aoA`5fr#P zZ47iBioXZ7+tYzH?CTp&;oC6nK-Au9G@7E6@vDg%o;mE_=Lxt5k?ut6v6m3GL)ov? zlJOpDJR6jlG=^I{=<^h*7RT_jOVS!EG3nCpt?=$rGG!!=MC*+tcF`XTblv1`+HJB+ z%fD8z7t@v#Kg!6cr_nH53$0^cR|6V8$(F(xRnO0=taW<)SnJvRm>n(%@ycLckOG!+ zC8=ZMMWqi?gYqBUKXlg8$lKT+c})YSAEfD#*HrNEN}7Sb6n$sG>pm5AP49VrU8p@U;<@~+NWMbky;tT%P=eOs zIkZQ~xxbT*D8o2XHE*EygOH@Cobq@xkN-+zT%0$X!vY0{uoe;Jd+*_?N#%mv^JcT} zWTRV}#rq&hU?~ zq)lP=MbB(DX4c@jGg+GT%X36vLD#T7Jm+)aPM-geH6Q+{_9n5`;D?!3o7Busn70BF z-0KW`p?7TMMjYLXZzP)L#_|C886Dll8%#zE!KQ6AEaoIeAd15~iRDL}Mvfu!eJAZd zp539k`kAKTHSofZEgTS)J#FS0*CMnhXQC}>$fkHclZ44$5X&1 z_!q2<%)z9HH{3=VrLMe3-LBEvG$@ zXuuf}pXCEXUKX#2yx5`7-Mtj9?f3Jj#Fj?;p?WrS#jpD8>*8rV2%Eq>Y{0UQQ!o{sW);L@#*WuY{l;iY&j6P;GqZ`#VX`gYTm z+*YKI*c?{Y*7>@(sIE$|y_RUM!pf@UWfIG*mV#i-pwTn=VXUyk=dm@L2ESE4x~@8J zrMrBe==+s>qwjrPz0vVPXLXfv6)BVSN>_i-8Q(ATd+ML9>&~8@ro>ta_ur{TT*~SR z&(Y6;e)rsK-``8svYrh`jJnW9x(|yr%zBr!szJPh1%xPm!=o@Q*~}l?xn;D$fw=Zd z(a`X#J8$*VEv*__1q9e~xu`-(Rt^5xkrtzv~11El$B}>DRvdlc%Y!UhAR&-I?qNtEOdTaO~}- z;dS`f>vTg@q07f5DcPJr2FKpgj>SC|$M`TVfhH^$DPdzpj;#qOcrE@}*Rq~ZC4oJM zwjhDJ`)i?oW7@VSqF_%v@4LT=(^Js_MIhN^$Bw5U*bv@a@PvqGS#4WW%dr&&OV8Ka zeaVH@K3ff(jJnaM*d1Ur7|I!57R_|oIyIJV-c6vcAz7l$hW3Vuw&HD?exf_92-~Q? z*O)Wj=--i^jGvggO^^^W@-k1|1UWmZ(tcu?MMizl zFNB-JyF*y>C$DxXkBa#8dynhE&%*zW_Z*}>+VQP@)J#Es?%Mmf<(^K?GjZRu@jNX9 zZW*$MpD8UHKz3?YUTAq5uqVa^%O=~ptx;qj4l6BPd7v$w;gNO}^{r`)$@pygh;T!- zXXTMj#@DpG)5)j_Xd8Q{b5Hf>;Hl06eey84>UcM=zci|V6Kx^Os-jx*<~w~<0ZZ0| zIb@RBy0_^gxY?je5gw+4AVqo(Xff=Y;itvfcJqi9v?n91Cxz8pc1Q$op(~FiD{8ry<&MBYOrd{*@ILq>8#^<*4X*!mE#@TD?+sYU7+~{#7UoU_C)AGLTn#ra& z?R`b6OK!R)n+RTG$VQJwv&bXp%DQx8I}r_ARuh|x`gO+Z$!mFA@cv4Ez>OmCJFhXe z&KdT~anE+%Jb2x@keB+P4?9wAza^3NxV$1QguQ{IzAq8op7)-EuT7kFs)(1ppr+xE zQ_VOeJuTMd*-uo4@*nzoRZgb@zh&ZzY3YK+DOu+-XF5lD_TByY@SU@YS&C zLx11)H1-siP(LY)hSy=F!M-w+q~3~TzpS?Gd|Bw+RY}lgbdW4+$b>KW zUSldPr!#%(lJwxeqI-r8S)({=X+QQgEYGfN4{E%Bo17Cd>0@?fb~0*PN3xl)cK%*c zBu`HE{BeCf5P!CcPmVpSjbVEZhQ0O+{k>#1 z*NxLR_U{oFeT;L8-np+5>Q>N-Fm%X#PJ9Bl@C%0Axw zl{xlevyOw0NKVV%T3`-f4@}FKu_NAgud`Y_i`Ca^sKR%ElaL8p7eBi`GTrs@>aKVs z6Skr&dsl5u7H$#jtTJKMaXPKYwhH}7#x(P6nfbJQ#~6~m;>y0?pYpoyEKm#!H~)jx z3-;68%mk;3_M1i`-g)Tw{UE#gCPQyIdG=xPw!@CRwJl!<^q)=ktPC&H!J0A@gbc|X zP3CA)?rr#dp9sI7bd}7(k*>vg${Z2PX`_HdYz-g933Y6C!HebKz=s73l-ks11ng`;oTqJXa_x# z-tP4Gv(tMObd5tD{0`z7_TcWQ7f^>x3OkLs<~ zzw59H9o`%cdm*QfSH*iyBhs00i1&D#@_&ed$oPUt?0{ysBcB!3!CRUYB9-H2I+#PIJW(Z#J2n#DAh7fF8w5q(I2nF-fIU;g2bl_sO^dgd-fuIyldDV#Tp0y?{}lEh h6p>W&ZKtZ TypeName; +#define FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix TypeName; + +#define FFX_16BIT_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType##16_t TypeName; +#define FFX_16BIT_VECTOR( TypeName, BaseComponentType, COL ) typedef vector TypeName; +#define FFX_16BIT_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix TypeName; + +#else //FFX_HLSL_6_2 + +#define FFX_MIN16_SCALAR( TypeName, BaseComponentType ) typedef min16##BaseComponentType TypeName; +#define FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL ) typedef vector TypeName; +#define FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix TypeName; + +#define FFX_16BIT_SCALAR( TypeName, BaseComponentType ) FFX_MIN16_SCALAR( TypeName, BaseComponentType ); +#define FFX_16BIT_VECTOR( TypeName, BaseComponentType, COL ) FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL ); +#define FFX_16BIT_MATRIX( TypeName, BaseComponentType, ROW, COL ) FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ); + +#endif //FFX_HLSL_6_2 + +#else //FFX_HALF + +#define FFX_MIN16_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType TypeName; +#define FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL ) typedef vector TypeName; +#define FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix TypeName; + +#define FFX_16BIT_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType TypeName; +#define FFX_16BIT_VECTOR( TypeName, BaseComponentType, COL ) typedef vector TypeName; +#define FFX_16BIT_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix TypeName; + +#endif //FFX_HALF + +#if defined(FFX_GPU) +// Common typedefs: +#if defined(FFX_HLSL) +FFX_MIN16_SCALAR( FFX_MIN16_F , float ); +FFX_MIN16_VECTOR( FFX_MIN16_F2, float, 2 ); +FFX_MIN16_VECTOR( FFX_MIN16_F3, float, 3 ); +FFX_MIN16_VECTOR( FFX_MIN16_F4, float, 4 ); + +FFX_MIN16_SCALAR( FFX_MIN16_I, int ); +FFX_MIN16_VECTOR( FFX_MIN16_I2, int, 2 ); +FFX_MIN16_VECTOR( FFX_MIN16_I3, int, 3 ); +FFX_MIN16_VECTOR( FFX_MIN16_I4, int, 4 ); + +FFX_MIN16_SCALAR( FFX_MIN16_U, uint ); +FFX_MIN16_VECTOR( FFX_MIN16_U2, uint, 2 ); +FFX_MIN16_VECTOR( FFX_MIN16_U3, uint, 3 ); +FFX_MIN16_VECTOR( FFX_MIN16_U4, uint, 4 ); + +FFX_16BIT_SCALAR( FFX_F16_t , float ); +FFX_16BIT_VECTOR( FFX_F16_t2, float, 2 ); +FFX_16BIT_VECTOR( FFX_F16_t3, float, 3 ); +FFX_16BIT_VECTOR( FFX_F16_t4, float, 4 ); + +FFX_16BIT_SCALAR( FFX_I16_t, int ); +FFX_16BIT_VECTOR( FFX_I16_t2, int, 2 ); +FFX_16BIT_VECTOR( FFX_I16_t3, int, 3 ); +FFX_16BIT_VECTOR( FFX_I16_t4, int, 4 ); + +FFX_16BIT_SCALAR( FFX_U16_t, uint ); +FFX_16BIT_VECTOR( FFX_U16_t2, uint, 2 ); +FFX_16BIT_VECTOR( FFX_U16_t3, uint, 3 ); +FFX_16BIT_VECTOR( FFX_U16_t4, uint, 4 ); + +#define TYPEDEF_MIN16_TYPES(Prefix) \ +typedef FFX_MIN16_F Prefix##_F; \ +typedef FFX_MIN16_F2 Prefix##_F2; \ +typedef FFX_MIN16_F3 Prefix##_F3; \ +typedef FFX_MIN16_F4 Prefix##_F4; \ +typedef FFX_MIN16_I Prefix##_I; \ +typedef FFX_MIN16_I2 Prefix##_I2; \ +typedef FFX_MIN16_I3 Prefix##_I3; \ +typedef FFX_MIN16_I4 Prefix##_I4; \ +typedef FFX_MIN16_U Prefix##_U; \ +typedef FFX_MIN16_U2 Prefix##_U2; \ +typedef FFX_MIN16_U3 Prefix##_U3; \ +typedef FFX_MIN16_U4 Prefix##_U4; + +#define TYPEDEF_16BIT_TYPES(Prefix) \ +typedef FFX_16BIT_F Prefix##_F; \ +typedef FFX_16BIT_F2 Prefix##_F2; \ +typedef FFX_16BIT_F3 Prefix##_F3; \ +typedef FFX_16BIT_F4 Prefix##_F4; \ +typedef FFX_16BIT_I Prefix##_I; \ +typedef FFX_16BIT_I2 Prefix##_I2; \ +typedef FFX_16BIT_I3 Prefix##_I3; \ +typedef FFX_16BIT_I4 Prefix##_I4; \ +typedef FFX_16BIT_U Prefix##_U; \ +typedef FFX_16BIT_U2 Prefix##_U2; \ +typedef FFX_16BIT_U3 Prefix##_U3; \ +typedef FFX_16BIT_U4 Prefix##_U4; + +#define TYPEDEF_FULL_PRECISION_TYPES(Prefix) \ +typedef FfxFloat32 Prefix##_F; \ +typedef FfxFloat32x2 Prefix##_F2; \ +typedef FfxFloat32x3 Prefix##_F3; \ +typedef FfxFloat32x4 Prefix##_F4; \ +typedef FfxInt32 Prefix##_I; \ +typedef FfxInt32x2 Prefix##_I2; \ +typedef FfxInt32x3 Prefix##_I3; \ +typedef FfxInt32x4 Prefix##_I4; \ +typedef FfxUInt32 Prefix##_U; \ +typedef FfxUInt32x2 Prefix##_U2; \ +typedef FfxUInt32x3 Prefix##_U3; \ +typedef FfxUInt32x4 Prefix##_U4; +#endif // #if defined(FFX_HLSL) + +#if defined(FFX_GLSL) + +#if FFX_HALF + +#define FFX_MIN16_F float16_t +#define FFX_MIN16_F2 f16vec2 +#define FFX_MIN16_F3 f16vec3 +#define FFX_MIN16_F4 f16vec4 + +#define FFX_MIN16_I int16_t +#define FFX_MIN16_I2 i16vec2 +#define FFX_MIN16_I3 i16vec3 +#define FFX_MIN16_I4 i16vec4 + +#define FFX_MIN16_U uint16_t +#define FFX_MIN16_U2 u16vec2 +#define FFX_MIN16_U3 u16vec3 +#define FFX_MIN16_U4 u16vec4 + +#define FFX_16BIT_F float16_t +#define FFX_16BIT_F2 f16vec2 +#define FFX_16BIT_F3 f16vec3 +#define FFX_16BIT_F4 f16vec4 + +#define FFX_16BIT_I int16_t +#define FFX_16BIT_I2 i16vec2 +#define FFX_16BIT_I3 i16vec3 +#define FFX_16BIT_I4 i16vec4 + +#define FFX_16BIT_U uint16_t +#define FFX_16BIT_U2 u16vec2 +#define FFX_16BIT_U3 u16vec3 +#define FFX_16BIT_U4 u16vec4 + +#else // FFX_HALF + +#define FFX_MIN16_F float +#define FFX_MIN16_F2 vec2 +#define FFX_MIN16_F3 vec3 +#define FFX_MIN16_F4 vec4 + +#define FFX_MIN16_I int +#define FFX_MIN16_I2 ivec2 +#define FFX_MIN16_I3 ivec3 +#define FFX_MIN16_I4 ivec4 + +#define FFX_MIN16_U uint +#define FFX_MIN16_U2 uvec2 +#define FFX_MIN16_U3 uvec3 +#define FFX_MIN16_U4 uvec4 + +#define FFX_16BIT_F float +#define FFX_16BIT_F2 vec2 +#define FFX_16BIT_F3 vec3 +#define FFX_16BIT_F4 vec4 + +#define FFX_16BIT_I int +#define FFX_16BIT_I2 ivec2 +#define FFX_16BIT_I3 ivec3 +#define FFX_16BIT_I4 ivec4 + +#define FFX_16BIT_U uint +#define FFX_16BIT_U2 uvec2 +#define FFX_16BIT_U3 uvec3 +#define FFX_16BIT_U4 uvec4 + +#endif // FFX_HALF + +#endif // #if defined(FFX_GLSL) + +#endif // #if defined(FFX_GPU) +#endif // #ifndef FFX_COMMON_TYPES_H diff --git a/thirdparty/amd-fsr2/shaders/ffx_core.h b/thirdparty/amd-fsr2/shaders/ffx_core.h new file mode 100644 index 000000000000..4e687d6e3d6b --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core.h @@ -0,0 +1,52 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// @defgroup Core +/// @defgroup HLSL +/// @defgroup GLSL +/// @defgroup GPU +/// @defgroup CPU +/// @defgroup CAS +/// @defgroup FSR1 + +#if !defined(FFX_CORE_H) +#define FFX_CORE_H + +#include "ffx_common_types.h" + +#if defined(FFX_CPU) + #include "ffx_core_cpu.h" +#endif // #if defined(FFX_CPU) + +#if defined(FFX_GLSL) && defined(FFX_GPU) + #include "ffx_core_glsl.h" +#endif // #if defined(FFX_GLSL) && defined(FFX_GPU) + +#if defined(FFX_HLSL) && defined(FFX_GPU) + #include "ffx_core_hlsl.h" +#endif // #if defined(FFX_HLSL) && defined(FFX_GPU) + +#if defined(FFX_GPU) + #include "ffx_core_gpu_common.h" + #include "ffx_core_gpu_common_half.h" + #include "ffx_core_portability.h" +#endif // #if defined(FFX_GPU) +#endif // #if !defined(FFX_CORE_H) \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_core_cpu.h b/thirdparty/amd-fsr2/shaders/ffx_core_cpu.h new file mode 100644 index 000000000000..3bf0295bfc62 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core_cpu.h @@ -0,0 +1,332 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// A define for a true value in a boolean expression. +/// +/// @ingroup CPU +#define FFX_TRUE (1) + +/// A define for a false value in a boolean expression. +/// +/// @ingroup CPU +#define FFX_FALSE (0) + +#if !defined(FFX_STATIC) +/// A define to abstract declaration of static variables and functions. +/// +/// @ingroup CPU +#define FFX_STATIC static +#endif // #if !defined(FFX_STATIC) + +#ifdef __clang__ +#pragma clang diagnostic ignored "-Wunused-variable" +#endif + +/// Interpret the bit layout of an IEEE-754 floating point value as an unsigned integer. +/// +/// @param [in] x A 32bit floating value. +/// +/// @returns +/// An unsigned 32bit integer value containing the bit pattern of x. +/// +/// @ingroup CPU +FFX_STATIC FfxUInt32 ffxAsUInt32(FfxFloat32 x) +{ + union + { + FfxFloat32 f; + FfxUInt32 u; + } bits; + + bits.f = x; + return bits.u; +} + +FFX_STATIC FfxFloat32 ffxDot2(FfxFloat32x2 a, FfxFloat32x2 b) +{ + return a[0] * b[0] + a[1] * b[1]; +} + +FFX_STATIC FfxFloat32 ffxDot3(FfxFloat32x3 a, FfxFloat32x3 b) +{ + return a[0] * b[0] + a[1] * b[1] + a[2] * b[2]; +} + +FFX_STATIC FfxFloat32 ffxDot4(FfxFloat32x4 a, FfxFloat32x4 b) +{ + return a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3]; +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxLerp(FfxFloat32 x, FfxFloat32 y, FfxFloat32 t) +{ + return y * t + (-x * t + x); +} + +/// Compute the reciprocal of a value. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxReciprocal(FfxFloat32 a) +{ + return 1.0f / a; +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxSqrt(FfxFloat32 x) +{ + return sqrt(x); +} + +FFX_STATIC FfxUInt32 AShrSU1(FfxUInt32 a, FfxUInt32 b) +{ + return FfxUInt32(FfxInt32(a) >> FfxInt32(b)); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxFract(FfxFloat32 a) +{ + return a - floor(a); +} + +/// Compute the reciprocal square root of a value. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 rsqrt(FfxFloat32 a) +{ + return ffxReciprocal(ffxSqrt(a)); +} + +FFX_STATIC FfxFloat32 ffxMin(FfxFloat32 x, FfxFloat32 y) +{ + return x < y ? x : y; +} + +FFX_STATIC FfxUInt32 ffxMin(FfxUInt32 x, FfxUInt32 y) +{ + return x < y ? x : y; +} + +FFX_STATIC FfxFloat32 ffxMax(FfxFloat32 x, FfxFloat32 y) +{ + return x > y ? x : y; +} + +FFX_STATIC FfxUInt32 ffxMax(FfxUInt32 x, FfxUInt32 y) +{ + return x > y ? x : y; +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxSaturate(FfxFloat32 a) +{ + return ffxMin(1.0f, ffxMax(0.0f, a)); +} + +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +FFX_STATIC void opAAddOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b) +{ + d[0] = a[0] + b; + d[1] = a[1] + b; + d[2] = a[2] + b; + return; +} + +FFX_STATIC void opACpyF3(FfxFloat32x3 d, FfxFloat32x3 a) +{ + d[0] = a[0]; + d[1] = a[1]; + d[2] = a[2]; + return; +} + +FFX_STATIC void opAMulF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32x3 b) +{ + d[0] = a[0] * b[0]; + d[1] = a[1] * b[1]; + d[2] = a[2] * b[2]; + return; +} + +FFX_STATIC void opAMulOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b) +{ + d[0] = a[0] * b; + d[1] = a[1] * b; + d[2] = a[2] * b; + return; +} + +FFX_STATIC void opARcpF3(FfxFloat32x3 d, FfxFloat32x3 a) +{ + d[0] = ffxReciprocal(a[0]); + d[1] = ffxReciprocal(a[1]); + d[2] = ffxReciprocal(a[2]); + return; +} + +/// Convert FfxFloat32 to half (in lower 16-bits of output). +/// +/// This function implements the same fast technique that is documented here: ftp://ftp.fox-toolkit.org/pub/fasthalffloatconversion.pdf +/// +/// The function supports denormals. +/// +/// Some conversion rules are to make computations possibly "safer" on the GPU, +/// -INF & -NaN -> -65504 +/// +INF & +NaN -> +65504 +/// +/// @param [in] f The 32bit floating point value to convert. +/// +/// @returns +/// The closest 16bit floating point value to f. +/// +/// @ingroup CPU +FFX_STATIC FfxUInt32 f32tof16(FfxFloat32 f) +{ + static FfxUInt16 base[512] = { + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100, 0x0200, 0x0400, + 0x0800, 0x0c00, 0x1000, 0x1400, 0x1800, 0x1c00, 0x2000, 0x2400, 0x2800, 0x2c00, 0x3000, 0x3400, 0x3800, 0x3c00, 0x4000, 0x4400, 0x4800, 0x4c00, 0x5000, + 0x5400, 0x5800, 0x5c00, 0x6000, 0x6400, 0x6800, 0x6c00, 0x7000, 0x7400, 0x7800, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8001, 0x8002, + 0x8004, 0x8008, 0x8010, 0x8020, 0x8040, 0x8080, 0x8100, 0x8200, 0x8400, 0x8800, 0x8c00, 0x9000, 0x9400, 0x9800, 0x9c00, 0xa000, 0xa400, 0xa800, 0xac00, + 0xb000, 0xb400, 0xb800, 0xbc00, 0xc000, 0xc400, 0xc800, 0xcc00, 0xd000, 0xd400, 0xd800, 0xdc00, 0xe000, 0xe400, 0xe800, 0xec00, 0xf000, 0xf400, 0xf800, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff + }; + + static FfxUInt8 shift[512] = { + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18 + }; + + union + { + FfxFloat32 f; + FfxUInt32 u; + } bits; + + bits.f = f; + FfxUInt32 u = bits.u; + FfxUInt32 i = u >> 23; + return (FfxUInt32)(base[i]) + ((u & 0x7fffff) >> shift[i]); +} + +/// Pack 2x32-bit floating point values in a single 32bit value. +/// +/// This function first converts each component of value into their nearest 16-bit floating +/// point representation, and then stores the X and Y components in the lower and upper 16 bits of the +/// 32bit unsigned integer respectively. +/// +/// @param [in] value A 2-dimensional floating point value to convert and pack. +/// +/// @returns +/// A packed 32bit value containing 2 16bit floating point values. +/// +/// @ingroup CPU +FFX_STATIC FfxUInt32 packHalf2x16(FfxFloat32x2 a) +{ + return f32tof16(a[0]) + (f32tof16(a[1]) << 16); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_core_glsl.h b/thirdparty/amd-fsr2/shaders/ffx_core_glsl.h new file mode 100644 index 000000000000..6ec58f3c625c --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core_glsl.h @@ -0,0 +1,1669 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// A define for abstracting shared memory between shading languages. +/// +/// @ingroup GPU +#define FFX_GROUPSHARED shared + +/// A define for abstracting compute memory barriers between shading languages. +/// +/// @ingroup GPU +#define FFX_GROUP_MEMORY_BARRIER() barrier() + +/// A define added to accept static markup on functions to aid CPU/GPU portability of code. +/// +/// @ingroup GPU +#define FFX_STATIC + +/// A define for abstracting loop unrolling between shading languages. +/// +/// @ingroup GPU +#define FFX_UNROLL + +/// A define for abstracting a 'greater than' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_GREATER_THAN(x, y) greaterThan(x, y) + +/// A define for abstracting a 'greater than or equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_GREATER_THAN_EQUAL(x, y) greaterThanEqual(x, y) + +/// A define for abstracting a 'less than' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_LESS_THAN(x, y) lessThan(x, y) + +/// A define for abstracting a 'less than or equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_LESS_THAN_EQUAL(x, y) lessThanEqual(x, y) + +/// A define for abstracting an 'equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_EQUAL(x, y) equal(x, y) + +/// A define for abstracting a 'not equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_NOT_EQUAL(x, y) notEqual(x, y) + +/// Broadcast a scalar value to a 1-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32(x) FfxFloat32(x) + +/// Broadcast a scalar value to a 2-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X2(x) FfxFloat32x2(FfxFloat32(x)) + +/// Broadcast a scalar value to a 3-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X3(x) FfxFloat32x3(FfxFloat32(x)) + +/// Broadcast a scalar value to a 4-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X4(x) FfxFloat32x4(FfxFloat32(x)) + +/// Broadcast a scalar value to a 1-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32(x) FfxUInt32(x) + +/// Broadcast a scalar value to a 2-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X2(x) FfxUInt32x2(FfxUInt32(x)) + +/// Broadcast a scalar value to a 3-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X3(x) FfxUInt32x3(FfxUInt32(x)) + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X4(x) FfxUInt32x4(FfxUInt32(x)) + +/// Broadcast a scalar value to a 1-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32(x) FfxInt32(x) + +/// Broadcast a scalar value to a 2-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X2(x) FfxInt32x2(FfxInt32(x)) + +/// Broadcast a scalar value to a 3-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X3(x) FfxInt32x3(FfxInt32(x)) + +/// Broadcast a scalar value to a 4-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X4(x) FfxInt32x4(FfxInt32(x)) + +/// Broadcast a scalar value to a 1-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16(x) FFX_MIN16_F(x) + +/// Broadcast a scalar value to a 2-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X2(x) FFX_MIN16_F2(FFX_MIN16_F(x)) + +/// Broadcast a scalar value to a 3-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X3(x) FFX_MIN16_F3(FFX_MIN16_F(x)) + +/// Broadcast a scalar value to a 4-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X4(x) FFX_MIN16_F4(FFX_MIN16_F(x)) + +/// Broadcast a scalar value to a 1-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16(x) FFX_MIN16_U(x) + +/// Broadcast a scalar value to a 2-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X2(x) FFX_MIN16_U2(FFX_MIN16_U(x)) + +/// Broadcast a scalar value to a 3-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X3(x) FFX_MIN16_U3(FFX_MIN16_U(x)) + +/// Broadcast a scalar value to a 4-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X4(x) FFX_MIN16_U4(FFX_MIN16_U(x)) + +/// Broadcast a scalar value to a 1-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16(x) FFX_MIN16_I(x) + +/// Broadcast a scalar value to a 2-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X2(x) FFX_MIN16_I2(FFX_MIN16_I(x)) + +/// Broadcast a scalar value to a 3-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X3(x) FFX_MIN16_I3(FFX_MIN16_I(x)) + +/// Broadcast a scalar value to a 4-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X4(x) FFX_MIN16_I4(FFX_MIN16_I(x)) + +#if !defined(FFX_SKIP_EXT) +#if FFX_HALF + #extension GL_EXT_shader_16bit_storage : require + #extension GL_EXT_shader_explicit_arithmetic_types : require +#endif // FFX_HALF + +#if defined(FFX_LONG) + #extension GL_ARB_gpu_shader_int64 : require + #extension GL_NV_shader_atomic_int64 : require +#endif // #if defined(FFX_LONG) + +#if defined(FFX_WAVE) + #extension GL_KHR_shader_subgroup_arithmetic : require + #extension GL_KHR_shader_subgroup_ballot : require + #extension GL_KHR_shader_subgroup_quad : require + #extension GL_KHR_shader_subgroup_shuffle : require +#endif // #if defined(FFX_WAVE) +#endif // #if !defined(FFX_SKIP_EXT) + +// Forward declarations +FfxFloat32 ffxSqrt(FfxFloat32 x); +FfxFloat32x2 ffxSqrt(FfxFloat32x2 x); +FfxFloat32x3 ffxSqrt(FfxFloat32x3 x); +FfxFloat32x4 ffxSqrt(FfxFloat32x4 x); + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32 ffxAsFloat(FfxUInt32 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxAsFloat(FfxUInt32x2 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxAsFloat(FfxUInt32x3 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxAsFloat(FfxUInt32x4 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32 ffxAsUInt32(FfxFloat32 x) +{ + return floatBitsToUint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxAsUInt32(FfxFloat32x2 x) +{ + return floatBitsToUint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxAsUInt32(FfxFloat32x3 x) +{ + return floatBitsToUint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxAsUInt32(FfxFloat32x4 x) +{ + return floatBitsToUint(x); +} + +/// Convert a 32bit IEEE 754 floating point value to its nearest 16bit equivalent. +/// +/// @param [in] value The value to convert. +/// +/// @returns +/// The nearest 16bit equivalent of value. +/// +/// @ingroup GLSL +FfxUInt32 f32tof16(FfxFloat32 value) +{ + return packHalf2x16(FfxFloat32x2(value, 0.0)); +} + +/// Broadcast a scalar value to a 2-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional floating point vector with value in each component. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxBroadcast2(FfxFloat32 value) +{ + return FfxFloat32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional floating point vector with value in each component. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxBroadcast3(FfxFloat32 value) +{ + return FfxFloat32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional floating point vector with value in each component. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxBroadcast4(FfxFloat32 value) +{ + return FfxFloat32x4(value, value, value, value); +} + +/// Broadcast a scalar value to a 2-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional signed integer vector with value in each component. +/// +/// @ingroup GLSL +FfxInt32x2 ffxBroadcast2(FfxInt32 value) +{ + return FfxInt32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional signed integer vector with value in each component. +/// +/// @ingroup GLSL +FfxInt32x3 ffxBroadcast3(FfxInt32 value) +{ + return FfxInt32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional signed integer vector with value in each component. +/// +/// @ingroup GLSL +FfxInt32x4 ffxBroadcast4(FfxInt32 value) +{ + return FfxInt32x4(value, value, value, value); +} + +/// Broadcast a scalar value to a 2-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxBroadcast2(FfxUInt32 value) +{ + return FfxUInt32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxBroadcast3(FfxUInt32 value) +{ + return FfxUInt32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxBroadcast4(FfxUInt32 value) +{ + return FfxUInt32x4(value, value, value, value); +} + +/// +/// +/// @ingroup GLSL +FfxUInt32 bitfieldExtract(FfxUInt32 src, FfxUInt32 off, FfxUInt32 bits) +{ + return bitfieldExtract(src, FfxInt32(off), FfxInt32(bits)); +} + +/// +/// +/// @ingroup GLSL +FfxUInt32 bitfieldInsert(FfxUInt32 src, FfxUInt32 ins, FfxUInt32 mask) +{ + return (ins & mask) | (src & (~mask)); +} + +// Proxy for V_BFI_B32 where the 'mask' is set as 'bits', 'mask=(1<mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32 ffxLerp(FfxFloat32 x, FfxFloat32 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxLerp(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxLerp(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxLerp(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxLerp(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxLerp(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxLerp(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 t) +{ + return mix(x, y, t); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32 ffxMax3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxMax3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxMax3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxMax3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32 ffxMax3(FfxUInt32 x, FfxUInt32 y, FfxUInt32 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxMax3(FfxUInt32x2 x, FfxUInt32x2 y, FfxUInt32x2 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxMax3(FfxUInt32x3 x, FfxUInt32x3 y, FfxUInt32x3 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxMax3(FfxUInt32x4 x, FfxUInt32x4 y, FfxUInt32x4 z) +{ + return max(x, max(y, z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32 ffxMed3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxMed3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxMed3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxMed3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_I32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxInt32 ffxMed3(FfxInt32 x, FfxInt32 y, FfxInt32 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_I32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxInt32x2 ffxMed3(FfxInt32x2 x, FfxInt32x2 y, FfxInt32x2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_I32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxInt32x3 ffxMed3(FfxInt32x3 x, FfxInt32x3 y, FfxInt32x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_I32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxInt32x4 ffxMed3(FfxInt32x4 x, FfxInt32x4 y, FfxInt32x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN and RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32 ffxMin3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxMin3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxMin3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxMin3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32 ffxMin3(FfxUInt32 x, FfxUInt32 y, FfxUInt32 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxMin3(FfxUInt32x2 x, FfxUInt32x2 y, FfxUInt32x2 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxMin3(FfxUInt32x3 x, FfxUInt32x3 y, FfxUInt32x3 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxMin3(FfxUInt32x4 x, FfxUInt32x4 y, FfxUInt32x4 z) +{ + return min(x, min(y, z)); +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32 rcp(FfxFloat32 x) +{ + return FfxFloat32(1.0) / x; +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32x2 rcp(FfxFloat32x2 x) +{ + return ffxBroadcast2(1.0) / x; +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32x3 rcp(FfxFloat32x3 x) +{ + return ffxBroadcast3(1.0) / x; +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32x4 rcp(FfxFloat32x4 x) +{ + return ffxBroadcast4(1.0) / x; +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32 rsqrt(FfxFloat32 x) +{ + return FfxFloat32(1.0) / ffxSqrt(x); +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32x2 rsqrt(FfxFloat32x2 x) +{ + return ffxBroadcast2(1.0) / ffxSqrt(x); +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32x3 rsqrt(FfxFloat32x3 x) +{ + return ffxBroadcast3(1.0) / ffxSqrt(x); +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32x4 rsqrt(FfxFloat32x4 x) +{ + return ffxBroadcast4(1.0) / ffxSqrt(x); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32 ffxSaturate(FfxFloat32 x) +{ + return clamp(x, FfxFloat32(0.0), FfxFloat32(1.0)); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxSaturate(FfxFloat32x2 x) +{ + return clamp(x, ffxBroadcast2(0.0), ffxBroadcast2(1.0)); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxSaturate(FfxFloat32x3 x) +{ + return clamp(x, ffxBroadcast3(0.0), ffxBroadcast3(1.0)); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxSaturate(FfxFloat32x4 x) +{ + return clamp(x, ffxBroadcast4(0.0), ffxBroadcast4(1.0)); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32 ffxFract(FfxFloat32 x) +{ + return fract(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxFract(FfxFloat32x2 x) +{ + return fract(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxFract(FfxFloat32x3 x) +{ + return fract(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxFract(FfxFloat32x4 x) +{ + return fract(x); +} + +FfxUInt32 AShrSU1(FfxUInt32 a, FfxUInt32 b) +{ + return FfxUInt32(FfxInt32(a) >> FfxInt32(b)); +} + +#if FFX_HALF + +#define FFX_UINT32_TO_FLOAT16X2(x) unpackFloat2x16(FfxUInt32(x)) + +FfxFloat16x4 ffxUint32x2ToFloat16x4(FfxUInt32x2 x) +{ + return FfxFloat16x4(unpackFloat2x16(x.x), unpackFloat2x16(x.y)); +} +#define FFX_UINT32X2_TO_FLOAT16X4(x) ffxUint32x2ToFloat16x4(FfxUInt32x2(x)) +#define FFX_UINT32_TO_UINT16X2(x) unpackUint2x16(FfxUInt32(x)) +#define FFX_UINT32X2_TO_UINT16X4(x) unpackUint4x16(pack64(FfxUInt32x2(x))) +//------------------------------------------------------------------------------------------------------------------------------ +#define FFX_FLOAT16X2_TO_UINT32(x) packFloat2x16(FfxFloat16x2(x)) +FfxUInt32x2 ffxFloat16x4ToUint32x2(FfxFloat16x4 x) +{ + return FfxUInt32x2(packFloat2x16(x.xy), packFloat2x16(x.zw)); +} +#define FFX_FLOAT16X4_TO_UINT32X2(x) ffxFloat16x4ToUint32x2(FfxFloat16x4(x)) +#define FFX_UINT16X2_TO_UINT32(x) packUint2x16(FfxUInt16x2(x)) +#define FFX_UINT16X4_TO_UINT32X2(x) unpack32(packUint4x16(FfxUInt16x4(x))) +//============================================================================================================================== +#define FFX_TO_UINT16(x) halfBitsToUint16(FfxFloat16(x)) +#define FFX_TO_UINT16X2(x) halfBitsToUint16(FfxFloat16x2(x)) +#define FFX_TO_UINT16X3(x) halfBitsToUint16(FfxFloat16x3(x)) +#define FFX_TO_UINT16X4(x) halfBitsToUint16(FfxFloat16x4(x)) +//------------------------------------------------------------------------------------------------------------------------------ +#define FFX_TO_FLOAT16(x) uint16BitsToHalf(FfxUInt16(x)) +#define FFX_TO_FLOAT16X2(x) uint16BitsToHalf(FfxUInt16x2(x)) +#define FFX_TO_FLOAT16X3(x) uint16BitsToHalf(FfxUInt16x3(x)) +#define FFX_TO_FLOAT16X4(x) uint16BitsToHalf(FfxUInt16x4(x)) +//============================================================================================================================== +FfxFloat16 ffxBroadcastFloat16(FfxFloat16 a) +{ + return FfxFloat16(a); +} +FfxFloat16x2 ffxBroadcastFloat16x2(FfxFloat16 a) +{ + return FfxFloat16x2(a, a); +} +FfxFloat16x3 ffxBroadcastFloat16x3(FfxFloat16 a) +{ + return FfxFloat16x3(a, a, a); +} +FfxFloat16x4 ffxBroadcastFloat16x4(FfxFloat16 a) +{ + return FfxFloat16x4(a, a, a, a); +} +#define FFX_BROADCAST_FLOAT16(a) FfxFloat16(a) +#define FFX_BROADCAST_FLOAT16X2(a) FfxFloat16x2(FfxFloat16(a)) +#define FFX_BROADCAST_FLOAT16X3(a) FfxFloat16x3(FfxFloat16(a)) +#define FFX_BROADCAST_FLOAT16X4(a) FfxFloat16x4(FfxFloat16(a)) +//------------------------------------------------------------------------------------------------------------------------------ +FfxInt16 ffxBroadcastInt16(FfxInt16 a) +{ + return FfxInt16(a); +} +FfxInt16x2 ffxBroadcastInt16x2(FfxInt16 a) +{ + return FfxInt16x2(a, a); +} +FfxInt16x3 ffxBroadcastInt16x3(FfxInt16 a) +{ + return FfxInt16x3(a, a, a); +} +FfxInt16x4 ffxBroadcastInt16x4(FfxInt16 a) +{ + return FfxInt16x4(a, a, a, a); +} +#define FFX_BROADCAST_INT16(a) FfxInt16(a) +#define FFX_BROADCAST_INT16X2(a) FfxInt16x2(FfxInt16(a)) +#define FFX_BROADCAST_INT16X3(a) FfxInt16x3(FfxInt16(a)) +#define FFX_BROADCAST_INT16X4(a) FfxInt16x4(FfxInt16(a)) +//------------------------------------------------------------------------------------------------------------------------------ +FfxUInt16 ffxBroadcastUInt16(FfxUInt16 a) +{ + return FfxUInt16(a); +} +FfxUInt16x2 ffxBroadcastUInt16x2(FfxUInt16 a) +{ + return FfxUInt16x2(a, a); +} +FfxUInt16x3 ffxBroadcastUInt16x3(FfxUInt16 a) +{ + return FfxUInt16x3(a, a, a); +} +FfxUInt16x4 ffxBroadcastUInt16x4(FfxUInt16 a) +{ + return FfxUInt16x4(a, a, a, a); +} +#define FFX_BROADCAST_UINT16(a) FfxUInt16(a) +#define FFX_BROADCAST_UINT16X2(a) FfxUInt16x2(FfxUInt16(a)) +#define FFX_BROADCAST_UINT16X3(a) FfxUInt16x3(FfxUInt16(a)) +#define FFX_BROADCAST_UINT16X4(a) FfxUInt16x4(FfxUInt16(a)) +//============================================================================================================================== +FfxUInt16 ffxAbsHalf(FfxUInt16 a) +{ + return FfxUInt16(abs(FfxInt16(a))); +} +FfxUInt16x2 ffxAbsHalf(FfxUInt16x2 a) +{ + return FfxUInt16x2(abs(FfxInt16x2(a))); +} +FfxUInt16x3 ffxAbsHalf(FfxUInt16x3 a) +{ + return FfxUInt16x3(abs(FfxInt16x3(a))); +} +FfxUInt16x4 ffxAbsHalf(FfxUInt16x4 a) +{ + return FfxUInt16x4(abs(FfxInt16x4(a))); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxClampHalf(FfxFloat16 x, FfxFloat16 n, FfxFloat16 m) +{ + return clamp(x, n, m); +} +FfxFloat16x2 ffxClampHalf(FfxFloat16x2 x, FfxFloat16x2 n, FfxFloat16x2 m) +{ + return clamp(x, n, m); +} +FfxFloat16x3 ffxClampHalf(FfxFloat16x3 x, FfxFloat16x3 n, FfxFloat16x3 m) +{ + return clamp(x, n, m); +} +FfxFloat16x4 ffxClampHalf(FfxFloat16x4 x, FfxFloat16x4 n, FfxFloat16x4 m) +{ + return clamp(x, n, m); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxFract(FfxFloat16 x) +{ + return fract(x); +} +FfxFloat16x2 ffxFract(FfxFloat16x2 x) +{ + return fract(x); +} +FfxFloat16x3 ffxFract(FfxFloat16x3 x) +{ + return fract(x); +} +FfxFloat16x4 ffxFract(FfxFloat16x4 x) +{ + return fract(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxLerp(FfxFloat16 x, FfxFloat16 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x2 ffxLerp(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x2 ffxLerp(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 a) +{ + return mix(x, y, a); +} +FfxFloat16x3 ffxLerp(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 a) +{ + return mix(x, y, a); +} +FfxFloat16x3 ffxLerp(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x4 ffxLerp(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x4 ffxLerp(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 a) +{ + return mix(x, y, a); +} +//------------------------------------------------------------------------------------------------------------------------------ +// No packed version of ffxMid3. +FfxFloat16 ffxMed3Half(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxFloat16x2 ffxMed3Half(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxFloat16x3 ffxMed3Half(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxFloat16x4 ffxMed3Half(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxInt16 ffxMed3Half(FfxInt16 x, FfxInt16 y, FfxInt16 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxInt16x2 ffxMed3Half(FfxInt16x2 x, FfxInt16x2 y, FfxInt16x2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxInt16x3 ffxMed3Half(FfxInt16x3 x, FfxInt16x3 y, FfxInt16x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FfxInt16x4 ffxMed3Half(FfxInt16x4 x, FfxInt16x4 y, FfxInt16x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +// No packed version of ffxMax3. +FfxFloat16 ffxMax3Half(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + return max(x, max(y, z)); +} +FfxFloat16x2 ffxMax3Half(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + return max(x, max(y, z)); +} +FfxFloat16x3 ffxMax3Half(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + return max(x, max(y, z)); +} +FfxFloat16x4 ffxMax3Half(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + return max(x, max(y, z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +// No packed version of ffxMin3. +FfxFloat16 ffxMin3Half(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + return min(x, min(y, z)); +} +FfxFloat16x2 ffxMin3Half(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + return min(x, min(y, z)); +} +FfxFloat16x3 ffxMin3Half(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + return min(x, min(y, z)); +} +FfxFloat16x4 ffxMin3Half(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + return min(x, min(y, z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxReciprocalHalf(FfxFloat16 x) +{ + return FFX_BROADCAST_FLOAT16(1.0) / x; +} +FfxFloat16x2 ffxReciprocalHalf(FfxFloat16x2 x) +{ + return FFX_BROADCAST_FLOAT16X2(1.0) / x; +} +FfxFloat16x3 ffxReciprocalHalf(FfxFloat16x3 x) +{ + return FFX_BROADCAST_FLOAT16X3(1.0) / x; +} +FfxFloat16x4 ffxReciprocalHalf(FfxFloat16x4 x) +{ + return FFX_BROADCAST_FLOAT16X4(1.0) / x; +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxReciprocalSquareRootHalf(FfxFloat16 x) +{ + return FFX_BROADCAST_FLOAT16(1.0) / sqrt(x); +} +FfxFloat16x2 ffxReciprocalSquareRootHalf(FfxFloat16x2 x) +{ + return FFX_BROADCAST_FLOAT16X2(1.0) / sqrt(x); +} +FfxFloat16x3 ffxReciprocalSquareRootHalf(FfxFloat16x3 x) +{ + return FFX_BROADCAST_FLOAT16X3(1.0) / sqrt(x); +} +FfxFloat16x4 ffxReciprocalSquareRootHalf(FfxFloat16x4 x) +{ + return FFX_BROADCAST_FLOAT16X4(1.0) / sqrt(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxSaturate(FfxFloat16 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16(0.0), FFX_BROADCAST_FLOAT16(1.0)); +} +FfxFloat16x2 ffxSaturate(FfxFloat16x2 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16X2(0.0), FFX_BROADCAST_FLOAT16X2(1.0)); +} +FfxFloat16x3 ffxSaturate(FfxFloat16x3 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16X3(0.0), FFX_BROADCAST_FLOAT16X3(1.0)); +} +FfxFloat16x4 ffxSaturate(FfxFloat16x4 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16X4(0.0), FFX_BROADCAST_FLOAT16X4(1.0)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxUInt16 ffxBitShiftRightHalf(FfxUInt16 a, FfxUInt16 b) +{ + return FfxUInt16(FfxInt16(a) >> FfxInt16(b)); +} +FfxUInt16x2 ffxBitShiftRightHalf(FfxUInt16x2 a, FfxUInt16x2 b) +{ + return FfxUInt16x2(FfxInt16x2(a) >> FfxInt16x2(b)); +} +FfxUInt16x3 ffxBitShiftRightHalf(FfxUInt16x3 a, FfxUInt16x3 b) +{ + return FfxUInt16x3(FfxInt16x3(a) >> FfxInt16x3(b)); +} +FfxUInt16x4 ffxBitShiftRightHalf(FfxUInt16x4 a, FfxUInt16x4 b) +{ + return FfxUInt16x4(FfxInt16x4(a) >> FfxInt16x4(b)); +} +#endif // FFX_HALF + +#if defined(FFX_WAVE) +// Where 'x' must be a compile time literal. +FfxFloat32 AWaveXorF1(FfxFloat32 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxFloat32x2 AWaveXorF2(FfxFloat32x2 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxFloat32x3 AWaveXorF3(FfxFloat32x3 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxFloat32x4 AWaveXorF4(FfxFloat32x4 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32 AWaveXorU1(FfxUInt32 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32x2 AWaveXorU2(FfxUInt32x2 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32x3 AWaveXorU3(FfxUInt32x3 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32x4 AWaveXorU4(FfxUInt32x4 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} + +//------------------------------------------------------------------------------------------------------------------------------ +#if FFX_HALF +FfxFloat16x2 ffxWaveXorFloat16x2(FfxFloat16x2 v, FfxUInt32 x) +{ + return FFX_UINT32_TO_FLOAT16X2(subgroupShuffleXor(FFX_FLOAT16X2_TO_UINT32(v), x)); +} +FfxFloat16x4 ffxWaveXorFloat16x4(FfxFloat16x4 v, FfxUInt32 x) +{ + return FFX_UINT32X2_TO_FLOAT16X4(subgroupShuffleXor(FFX_FLOAT16X4_TO_UINT32X2(v), x)); +} +FfxUInt16x2 ffxWaveXorUint16x2(FfxUInt16x2 v, FfxUInt32 x) +{ + return FFX_UINT32_TO_UINT16X2(subgroupShuffleXor(FFX_UINT16X2_TO_UINT32(v), x)); +} +FfxUInt16x4 ffxWaveXorUint16x4(FfxUInt16x4 v, FfxUInt32 x) +{ + return FFX_UINT32X2_TO_UINT16X4(subgroupShuffleXor(FFX_UINT16X4_TO_UINT32X2(v), x)); +} +#endif // FFX_HALF +#endif // #if defined(FFX_WAVE) diff --git a/thirdparty/amd-fsr2/shaders/ffx_core_gpu_common.h b/thirdparty/amd-fsr2/shaders/ffx_core_gpu_common.h new file mode 100644 index 000000000000..ae07642f0df3 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core_gpu_common.h @@ -0,0 +1,2784 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// A define for a true value in a boolean expression. +/// +/// @ingroup GPU +#define FFX_TRUE (true) + +/// A define for a false value in a boolean expression. +/// +/// @ingroup GPU +#define FFX_FALSE (false) + +/// A define value for positive infinity. +/// +/// @ingroup GPU +#define FFX_POSITIVE_INFINITY_FLOAT ffxAsFloat(0x7f800000u) + +/// A define value for negative infinity. +/// +/// @ingroup GPU +#define FFX_NEGATIVE_INFINITY_FLOAT ffxAsFloat(0xff800000u) + +/// A define value for PI. +/// +/// @ingroup GPU +#define FFX_PI (3.14159) + + +/// Compute the reciprocal of value. +/// +/// @param [in] value The value to compute the reciprocal of. +/// +/// @returns +/// The 1 / value. +/// +/// @ingroup GPU +FfxFloat32 ffxReciprocal(FfxFloat32 value) +{ + return rcp(value); +} + +/// Compute the reciprocal of value. +/// +/// @param [in] value The value to compute the reciprocal of. +/// +/// @returns +/// The 1 / value. +/// +/// @ingroup GPU +FfxFloat32x2 ffxReciprocal(FfxFloat32x2 value) +{ + return rcp(value); +} + +/// Compute the reciprocal of value. +/// +/// @param [in] value The value to compute the reciprocal of. +/// +/// @returns +/// The 1 / value. +/// +/// @ingroup GPU +FfxFloat32x3 ffxReciprocal(FfxFloat32x3 value) +{ + return rcp(value); +} + +/// Compute the reciprocal of value. +/// +/// @param [in] value The value to compute the reciprocal of. +/// +/// @returns +/// The 1 / value. +/// +/// @ingroup GPU +FfxFloat32x4 ffxReciprocal(FfxFloat32x4 value) +{ + return rcp(value); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32 ffxMin(FfxFloat32 x, FfxFloat32 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32x2 ffxMin(FfxFloat32x2 x, FfxFloat32x2 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32x3 ffxMin(FfxFloat32x3 x, FfxFloat32x3 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32x4 ffxMin(FfxFloat32x4 x, FfxFloat32x4 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32 ffxMin(FfxInt32 x, FfxInt32 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32x2 ffxMin(FfxInt32x2 x, FfxInt32x2 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32x3 ffxMin(FfxInt32x3 x, FfxInt32x3 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32x4 ffxMin(FfxInt32x4 x, FfxInt32x4 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32 ffxMin(FfxUInt32 x, FfxUInt32 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32x2 ffxMin(FfxUInt32x2 x, FfxUInt32x2 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32x3 ffxMin(FfxUInt32x3 x, FfxUInt32x3 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32x4 ffxMin(FfxUInt32x4 x, FfxUInt32x4 y) +{ + return min(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32 ffxMax(FfxFloat32 x, FfxFloat32 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32x2 ffxMax(FfxFloat32x2 x, FfxFloat32x2 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32x3 ffxMax(FfxFloat32x3 x, FfxFloat32x3 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat32x4 ffxMax(FfxFloat32x4 x, FfxFloat32x4 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32 ffxMax(FfxInt32 x, FfxInt32 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32x2 ffxMax(FfxInt32x2 x, FfxInt32x2 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32x3 ffxMax(FfxInt32x3 x, FfxInt32x3 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt32x4 ffxMax(FfxInt32x4 x, FfxInt32x4 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32 ffxMax(FfxUInt32 x, FfxUInt32 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32x2 ffxMax(FfxUInt32x2 x, FfxUInt32x2 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32x3 ffxMax(FfxUInt32x3 x, FfxUInt32x3 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt32x4 ffxMax(FfxUInt32x4 x, FfxUInt32x4 y) +{ + return max(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat32 ffxPow(FfxFloat32 x, FfxFloat32 y) +{ + return pow(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat32x2 ffxPow(FfxFloat32x2 x, FfxFloat32x2 y) +{ + return pow(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat32x3 ffxPow(FfxFloat32x3 x, FfxFloat32x3 y) +{ + return pow(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat32x4 ffxPow(FfxFloat32x4 x, FfxFloat32x4 y) +{ + return pow(x, y); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat32 ffxSqrt(FfxFloat32 x) +{ + return sqrt(x); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat32x2 ffxSqrt(FfxFloat32x2 x) +{ + return sqrt(x); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat32x3 ffxSqrt(FfxFloat32x3 x) +{ + return sqrt(x); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat32x4 ffxSqrt(FfxFloat32x4 x) +{ + return sqrt(x); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat32 ffxCopySignBit(FfxFloat32 d, FfxFloat32 s) +{ + return ffxAsFloat(ffxAsUInt32(d) | (ffxAsUInt32(s) & FfxUInt32(0x80000000u))); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat32x2 ffxCopySignBit(FfxFloat32x2 d, FfxFloat32x2 s) +{ + return ffxAsFloat(ffxAsUInt32(d) | (ffxAsUInt32(s) & ffxBroadcast2(0x80000000u))); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat32x3 ffxCopySignBit(FfxFloat32x3 d, FfxFloat32x3 s) +{ + return ffxAsFloat(ffxAsUInt32(d) | (ffxAsUInt32(s) & ffxBroadcast3(0x80000000u))); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat32x4 ffxCopySignBit(FfxFloat32x4 d, FfxFloat32x4 s) +{ + return ffxAsFloat(ffxAsUInt32(d) | (ffxAsUInt32(s) & ffxBroadcast4(0x80000000u))); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat32 ffxIsSigned(FfxFloat32 m) +{ + return ffxSaturate(m * FfxFloat32(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat32x2 ffxIsSigned(FfxFloat32x2 m) +{ + return ffxSaturate(m * ffxBroadcast2(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat32x3 ffxIsSigned(FfxFloat32x3 m) +{ + return ffxSaturate(m * ffxBroadcast3(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against for have the sign set. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or positive. +/// +/// @ingroup GPU +FfxFloat32x4 ffxIsSigned(FfxFloat32x4 m) +{ + return ffxSaturate(m * ffxBroadcast4(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat32 ffxIsGreaterThanZero(FfxFloat32 m) +{ + return ffxSaturate(m * FfxFloat32(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat32x2 ffxIsGreaterThanZero(FfxFloat32x2 m) +{ + return ffxSaturate(m * ffxBroadcast2(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat32x3 ffxIsGreaterThanZero(FfxFloat32x3 m) +{ + return ffxSaturate(m * ffxBroadcast3(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat32x4 ffxIsGreaterThanZero(FfxFloat32x4 m) +{ + return ffxSaturate(m * ffxBroadcast4(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// Convert a 32bit floating point value to sortable integer. +/// +/// - If sign bit=0, flip the sign bit (positives). +/// - If sign bit=1, flip all bits (negatives). +/// +/// The function has the side effects that: +/// - Larger integers are more positive values. +/// - Float zero is mapped to center of integers (so clear to integer zero is a nice default for atomic max usage). +/// +/// @param [in] value The floating point value to make sortable. +/// +/// @returns +/// The sortable integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxFloatToSortableInteger(FfxUInt32 value) +{ + return value ^ ((AShrSU1(value, FfxUInt32(31))) | FfxUInt32(0x80000000)); +} + +/// Convert a sortable integer to a 32bit floating point value. +/// +/// The function has the side effects that: +/// - If sign bit=1, flip the sign bit (positives). +/// - If sign bit=0, flip all bits (negatives). +/// +/// @param [in] value The floating point value to make sortable. +/// +/// @returns +/// The sortable integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxSortableIntegerToFloat(FfxUInt32 value) +{ + return value ^ ((~AShrSU1(value, FfxUInt32(31))) | FfxUInt32(0x80000000)); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateSqrt(FfxFloat32 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> FfxUInt32(1)) + FfxUInt32(0x1fbc4639)); +} + +/// Calculate a low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateReciprocal(FfxFloat32 a) +{ + return ffxAsFloat(FfxUInt32(0x7ef07ebb) - ffxAsUInt32(a)); +} + +/// Calculate a medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateReciprocalMedium(FfxFloat32 value) +{ + FfxFloat32 b = ffxAsFloat(FfxUInt32(0x7ef19fff) - ffxAsUInt32(value)); + return b * (-b * value + FfxFloat32(2.0)); +} + +/// Calculate a low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal square root for. +/// +/// @returns +/// An approximation of the reciprocal square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateReciprocalSquareRoot(FfxFloat32 a) +{ + return ffxAsFloat(FfxUInt32(0x5f347d74) - (ffxAsUInt32(a) >> FfxUInt32(1))); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateSqrt(FfxFloat32x2 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast2(1u)) + ffxBroadcast2(0x1fbc4639u)); +} + +/// Calculate a low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateReciprocal(FfxFloat32x2 a) +{ + return ffxAsFloat(ffxBroadcast2(0x7ef07ebbu) - ffxAsUInt32(a)); +} + +/// Calculate a medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateReciprocalMedium(FfxFloat32x2 a) +{ + FfxFloat32x2 b = ffxAsFloat(ffxBroadcast2(0x7ef19fffu) - ffxAsUInt32(a)); + return b * (-b * a + ffxBroadcast2(2.0f)); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateReciprocalSquareRoot(FfxFloat32x2 a) +{ + return ffxAsFloat(ffxBroadcast2(0x5f347d74u) - (ffxAsUInt32(a) >> ffxBroadcast2(1u))); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateSqrt(FfxFloat32x3 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast3(1u)) + ffxBroadcast3(0x1fbc4639u)); +} + +/// Calculate a low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateReciprocal(FfxFloat32x3 a) +{ + return ffxAsFloat(ffxBroadcast3(0x7ef07ebbu) - ffxAsUInt32(a)); +} + +/// Calculate a medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateReciprocalMedium(FfxFloat32x3 a) +{ + FfxFloat32x3 b = ffxAsFloat(ffxBroadcast3(0x7ef19fffu) - ffxAsUInt32(a)); + return b * (-b * a + ffxBroadcast3(2.0f)); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateReciprocalSquareRoot(FfxFloat32x3 a) +{ + return ffxAsFloat(ffxBroadcast3(0x5f347d74u) - (ffxAsUInt32(a) >> ffxBroadcast3(1u))); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateSqrt(FfxFloat32x4 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast4(1u)) + ffxBroadcast4(0x1fbc4639u)); +} + +/// Calculate a low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateReciprocal(FfxFloat32x4 a) +{ + return ffxAsFloat(ffxBroadcast4(0x7ef07ebbu) - ffxAsUInt32(a)); +} + +/// Calculate a medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateReciprocalMedium(FfxFloat32x4 a) +{ + FfxFloat32x4 b = ffxAsFloat(ffxBroadcast4(0x7ef19fffu) - ffxAsUInt32(a)); + return b * (-b * a + ffxBroadcast4(2.0f)); +} + +/// Calculate a low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] value The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateReciprocalSquareRoot(FfxFloat32x4 a) +{ + return ffxAsFloat(ffxBroadcast4(0x5f347d74u) - (ffxAsUInt32(a) >> ffxBroadcast4(1u))); +} + +/// Calculate dot product of 'a' and 'b'. +/// +/// @param [in] a First vector input. +/// @param [in] b Second vector input. +/// +/// @returns +/// The value of a dot b. +/// +/// @ingroup GPU +FfxFloat32 ffxDot2(FfxFloat32x2 a, FfxFloat32x2 b) +{ + return dot(a, b); +} + +/// Calculate dot product of 'a' and 'b'. +/// +/// @param [in] a First vector input. +/// @param [in] b Second vector input. +/// +/// @returns +/// The value of a dot b. +/// +/// @ingroup GPU +FfxFloat32 ffxDot3(FfxFloat32x3 a, FfxFloat32x3 b) +{ + return dot(a, b); +} + +/// Calculate dot product of 'a' and 'b'. +/// +/// @param [in] a First vector input. +/// @param [in] b Second vector input. +/// +/// @returns +/// The value of a dot b. +/// +/// @ingroup GPU +FfxFloat32 ffxDot4(FfxFloat32x4 a, FfxFloat32x4 b) +{ + return dot(a, b); +} + + +/// Compute an approximate conversion from PQ to Gamma2 space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and Gamma2. +/// +/// @returns +/// The value a converted into Gamma2. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximatePQToGamma2Medium(FfxFloat32 a) +{ + return a * a * a * a; +} + +/// Compute an approximate conversion from PQ to linear space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and linear. +/// +/// @returns +/// The value a converted into linear. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximatePQToLinear(FfxFloat32 a) +{ + return a * a * a * a * a * a * a * a; +} + +/// Compute an approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateGamma2ToPQ(FfxFloat32 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> FfxUInt32(2)) + FfxUInt32(0x2F9A4E46)); +} + +/// Compute a more accurate approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateGamma2ToPQMedium(FfxFloat32 a) +{ + FfxFloat32 b = ffxAsFloat((ffxAsUInt32(a) >> FfxUInt32(2)) + FfxUInt32(0x2F9A4E46)); + FfxFloat32 b4 = b * b * b * b; + return b - b * (b4 - a) / (FfxFloat32(4.0) * b4); +} + +/// Compute a high accuracy approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateGamma2ToPQHigh(FfxFloat32 a) +{ + return ffxSqrt(ffxSqrt(a)); +} + +/// Compute an approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateLinearToPQ(FfxFloat32 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> FfxUInt32(3)) + FfxUInt32(0x378D8723)); +} + +/// Compute a more accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateLinearToPQMedium(FfxFloat32 a) +{ + FfxFloat32 b = ffxAsFloat((ffxAsUInt32(a) >> FfxUInt32(3)) + FfxUInt32(0x378D8723)); + FfxFloat32 b8 = b * b * b * b * b * b * b * b; + return b - b * (b8 - a) / (FfxFloat32(8.0) * b8); +} + +/// Compute a very accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32 ffxApproximateLinearToPQHigh(FfxFloat32 a) +{ + return ffxSqrt(ffxSqrt(ffxSqrt(a))); +} + +/// Compute an approximate conversion from PQ to Gamma2 space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and Gamma2. +/// +/// @returns +/// The value a converted into Gamma2. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximatePQToGamma2Medium(FfxFloat32x2 a) +{ + return a * a * a * a; +} + +/// Compute an approximate conversion from PQ to linear space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and linear. +/// +/// @returns +/// The value a converted into linear. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximatePQToLinear(FfxFloat32x2 a) +{ + return a * a * a * a * a * a * a * a; +} + +/// Compute an approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateGamma2ToPQ(FfxFloat32x2 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast2(2u)) + ffxBroadcast2(0x2F9A4E46u)); +} + +/// Compute a more accurate approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateGamma2ToPQMedium(FfxFloat32x2 a) +{ + FfxFloat32x2 b = ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast2(2u)) + ffxBroadcast2(0x2F9A4E46u)); + FfxFloat32x2 b4 = b * b * b * b; + return b - b * (b4 - a) / (FfxFloat32(4.0) * b4); +} + +/// Compute a high accuracy approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateGamma2ToPQHigh(FfxFloat32x2 a) +{ + return ffxSqrt(ffxSqrt(a)); +} + +/// Compute an approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateLinearToPQ(FfxFloat32x2 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast2(3u)) + ffxBroadcast2(0x378D8723u)); +} + +/// Compute a more accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateLinearToPQMedium(FfxFloat32x2 a) +{ + FfxFloat32x2 b = ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast2(3u)) + ffxBroadcast2(0x378D8723u)); + FfxFloat32x2 b8 = b * b * b * b * b * b * b * b; + return b - b * (b8 - a) / (FfxFloat32(8.0) * b8); +} + +/// Compute a very accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x2 ffxApproximateLinearToPQHigh(FfxFloat32x2 a) +{ + return ffxSqrt(ffxSqrt(ffxSqrt(a))); +} + +/// Compute an approximate conversion from PQ to Gamma2 space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and Gamma2. +/// +/// @returns +/// The value a converted into Gamma2. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximatePQToGamma2Medium(FfxFloat32x3 a) +{ + return a * a * a * a; +} + +/// Compute an approximate conversion from PQ to linear space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and linear. +/// +/// @returns +/// The value a converted into linear. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximatePQToLinear(FfxFloat32x3 a) +{ + return a * a * a * a * a * a * a * a; +} + +/// Compute an approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateGamma2ToPQ(FfxFloat32x3 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast3(2u)) + ffxBroadcast3(0x2F9A4E46u)); +} + +/// Compute a more accurate approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateGamma2ToPQMedium(FfxFloat32x3 a) +{ + FfxFloat32x3 b = ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast3(2u)) + ffxBroadcast3(0x2F9A4E46u)); + FfxFloat32x3 b4 = b * b * b * b; + return b - b * (b4 - a) / (FfxFloat32(4.0) * b4); +} + +/// Compute a high accuracy approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateGamma2ToPQHigh(FfxFloat32x3 a) +{ + return ffxSqrt(ffxSqrt(a)); +} + +/// Compute an approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateLinearToPQ(FfxFloat32x3 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast3(3u)) + ffxBroadcast3(0x378D8723u)); +} + +/// Compute a more accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateLinearToPQMedium(FfxFloat32x3 a) +{ + FfxFloat32x3 b = ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast3(3u)) + ffxBroadcast3(0x378D8723u)); + FfxFloat32x3 b8 = b * b * b * b * b * b * b * b; + return b - b * (b8 - a) / (FfxFloat32(8.0) * b8); +} + +/// Compute a very accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x3 ffxApproximateLinearToPQHigh(FfxFloat32x3 a) +{ + return ffxSqrt(ffxSqrt(ffxSqrt(a))); +} + +/// Compute an approximate conversion from PQ to Gamma2 space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and Gamma2. +/// +/// @returns +/// The value a converted into Gamma2. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximatePQToGamma2Medium(FfxFloat32x4 a) +{ + return a * a * a * a; +} + +/// Compute an approximate conversion from PQ to linear space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between PQ and linear. +/// +/// @returns +/// The value a converted into linear. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximatePQToLinear(FfxFloat32x4 a) +{ + return a * a * a * a * a * a * a * a; +} + +/// Compute an approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateGamma2ToPQ(FfxFloat32x4 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast4(2u)) + ffxBroadcast4(0x2F9A4E46u)); +} + +/// Compute a more accurate approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateGamma2ToPQMedium(FfxFloat32x4 a) +{ + FfxFloat32x4 b = ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast4(2u)) + ffxBroadcast4(0x2F9A4E46u)); + FfxFloat32x4 b4 = b * b * b * b * b * b * b * b; + return b - b * (b4 - a) / (FfxFloat32(4.0) * b4); +} + +/// Compute a high accuracy approximate conversion from gamma2 to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between gamma2 and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateGamma2ToPQHigh(FfxFloat32x4 a) +{ + return ffxSqrt(ffxSqrt(a)); +} + +/// Compute an approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateLinearToPQ(FfxFloat32x4 a) +{ + return ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast4(3u)) + ffxBroadcast4(0x378D8723u)); +} + +/// Compute a more accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateLinearToPQMedium(FfxFloat32x4 a) +{ + FfxFloat32x4 b = ffxAsFloat((ffxAsUInt32(a) >> ffxBroadcast4(3u)) + ffxBroadcast4(0x378D8723u)); + FfxFloat32x4 b8 = b * b * b * b * b * b * b * b; + return b - b * (b8 - a) / (FfxFloat32(8.0) * b8); +} + +/// Compute a very accurate approximate conversion from linear to PQ space. +/// +/// PQ is very close to x^(1/8). The functions below Use the fast FfxFloat32 approximation method to do +/// PQ conversions to and from Gamma2 (4th power and fast 4th root), and PQ to and from Linear +/// (8th power and fast 8th root). The maximum error is approximately 0.2%. +/// +/// @param a The value to convert between linear and PQ. +/// +/// @returns +/// The value a converted into PQ. +/// +/// @ingroup GPU +FfxFloat32x4 ffxApproximateLinearToPQHigh(FfxFloat32x4 a) +{ + return ffxSqrt(ffxSqrt(ffxSqrt(a))); +} + +// An approximation of sine. +// +// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +// is {-1/4 to 1/4} representing {-1 to 1}. +// +// @param [in] value The value to calculate approximate sine for. +// +// @returns +// The approximate sine of value. +FfxFloat32 ffxParabolicSin(FfxFloat32 value) +{ + return value * abs(value) - value; +} + +// An approximation of sine. +// +// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +// is {-1/4 to 1/4} representing {-1 to 1}. +// +// @param [in] value The value to calculate approximate sine for. +// +// @returns +// The approximate sine of value. +FfxFloat32x2 ffxParabolicSin(FfxFloat32x2 x) +{ + return x * abs(x) - x; +} + +// An approximation of cosine. +// +// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +// is {-1/4 to 1/4} representing {-1 to 1}. +// +// @param [in] value The value to calculate approximate cosine for. +// +// @returns +// The approximate cosine of value. +FfxFloat32 ffxParabolicCos(FfxFloat32 x) +{ + x = ffxFract(x * FfxFloat32(0.5) + FfxFloat32(0.75)); + x = x * FfxFloat32(2.0) - FfxFloat32(1.0); + return ffxParabolicSin(x); +} + +// An approximation of cosine. +// +// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +// is {-1/4 to 1/4} representing {-1 to 1}. +// +// @param [in] value The value to calculate approximate cosine for. +// +// @returns +// The approximate cosine of value. +FfxFloat32x2 ffxParabolicCos(FfxFloat32x2 x) +{ + x = ffxFract(x * ffxBroadcast2(0.5f) + ffxBroadcast2(0.75f)); + x = x * ffxBroadcast2(2.0f) - ffxBroadcast2(1.0f); + return ffxParabolicSin(x); +} + +// An approximation of both sine and cosine. +// +// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +// is {-1/4 to 1/4} representing {-1 to 1}. +// +// @param [in] value The value to calculate approximate cosine for. +// +// @returns +// A FfxFloat32x2 containing approximations of both sine and cosine of value. +FfxFloat32x2 ffxParabolicSinCos(FfxFloat32 x) +{ + FfxFloat32 y = ffxFract(x * FfxFloat32(0.5) + FfxFloat32(0.75)); + y = y * FfxFloat32(2.0) - FfxFloat32(1.0); + return ffxParabolicSin(FfxFloat32x2(x, y)); +} + +/// Conditional free logic AND operation using values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt32 ffxZeroOneAnd(FfxUInt32 x, FfxUInt32 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt32x2 ffxZeroOneAnd(FfxUInt32x2 x, FfxUInt32x2 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt32x3 ffxZeroOneAnd(FfxUInt32x3 x, FfxUInt32x3 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt32x4 ffxZeroOneAnd(FfxUInt32x4 x, FfxUInt32x4 y) +{ + return min(x, y); +} + +/// Conditional free logic NOT operation using two values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt32 ffxZeroOneAnd(FfxUInt32 x) +{ + return x ^ FfxUInt32(1); +} + +/// Conditional free logic NOT operation using two values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt32x2 ffxZeroOneAnd(FfxUInt32x2 x) +{ + return x ^ ffxBroadcast2(1u); +} + +/// Conditional free logic NOT operation using two values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt32x3 ffxZeroOneAnd(FfxUInt32x3 x) +{ + return x ^ ffxBroadcast3(1u); +} + +/// Conditional free logic NOT operation using two values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt32x4 ffxZeroOneAnd(FfxUInt32x4 x) +{ + return x ^ ffxBroadcast4(1u); +} + +/// Conditional free logic OR operation using two values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt32 ffxZeroOneOr(FfxUInt32 x, FfxUInt32 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt32x2 ffxZeroOneOr(FfxUInt32x2 x, FfxUInt32x2 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt32x3 ffxZeroOneOr(FfxUInt32x3 x, FfxUInt32x3 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt32x4 ffxZeroOneOr(FfxUInt32x4 x, FfxUInt32x4 y) +{ + return max(x, y); +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxUInt32 ffxZeroOneAndToU1(FfxFloat32 x) +{ + return FfxUInt32(FfxFloat32(1.0) - x); +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxUInt32x2 ffxZeroOneAndToU2(FfxFloat32x2 x) +{ + return FfxUInt32x2(ffxBroadcast2(1.0) - x); +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxUInt32x3 ffxZeroOneAndToU3(FfxFloat32x3 x) +{ + return FfxUInt32x3(ffxBroadcast3(1.0) - x); +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxUInt32x4 ffxZeroOneAndToU4(FfxFloat32x4 x) +{ + return FfxUInt32x4(ffxBroadcast4(1.0) - x); +} + +/// Conditional free logic AND operation using two values followed by a NOT operation +/// using the resulting value and a third value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32 ffxZeroOneAndOr(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return ffxSaturate(x * y + z); +} + +/// Conditional free logic AND operation using two values followed by a NOT operation +/// using the resulting value and a third value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32x2 ffxZeroOneAndOr(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return ffxSaturate(x * y + z); +} + +/// Conditional free logic AND operation using two values followed by a NOT operation +/// using the resulting value and a third value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32x3 ffxZeroOneAndOr(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return ffxSaturate(x * y + z); +} + +/// Conditional free logic AND operation using two values followed by a NOT operation +/// using the resulting value and a third value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32x4 ffxZeroOneAndOr(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return ffxSaturate(x * y + z); +} + +/// Given a value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat32 ffxZeroOneIsGreaterThanZero(FfxFloat32 x) +{ + return ffxSaturate(x * FfxFloat32(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// Given a value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat32x2 ffxZeroOneIsGreaterThanZero(FfxFloat32x2 x) +{ + return ffxSaturate(x * ffxBroadcast2(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// Given a value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat32x3 ffxZeroOneIsGreaterThanZero(FfxFloat32x3 x) +{ + return ffxSaturate(x * ffxBroadcast3(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// Given a value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat32x4 ffxZeroOneIsGreaterThanZero(FfxFloat32x4 x) +{ + return ffxSaturate(x * ffxBroadcast4(FFX_POSITIVE_INFINITY_FLOAT)); +} + +/// Conditional free logic signed NOT operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32 ffxZeroOneAnd(FfxFloat32 x) +{ + return FfxFloat32(1.0) - x; +} + +/// Conditional free logic signed NOT operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32x2 ffxZeroOneAnd(FfxFloat32x2 x) +{ + return ffxBroadcast2(1.0) - x; +} + +/// Conditional free logic signed NOT operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32x3 ffxZeroOneAnd(FfxFloat32x3 x) +{ + return ffxBroadcast3(1.0) - x; +} + +/// Conditional free logic signed NOT operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat32x4 ffxZeroOneAnd(FfxFloat32x4 x) +{ + return ffxBroadcast4(1.0) - x; +} + +/// Conditional free logic OR operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat32 ffxZeroOneOr(FfxFloat32 x, FfxFloat32 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat32x2 ffxZeroOneOr(FfxFloat32x2 x, FfxFloat32x2 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat32x3 ffxZeroOneOr(FfxFloat32x3 x, FfxFloat32x3 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat32x4 ffxZeroOneOr(FfxFloat32x4 x, FfxFloat32x4 y) +{ + return max(x, y); +} + +/// Choose between two FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat32 ffxZeroOneSelect(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + FfxFloat32 r = (-x) * z + z; + return x * y + r; +} + +/// Choose between two FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat32x2 ffxZeroOneSelect(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + FfxFloat32x2 r = (-x) * z + z; + return x * y + r; +} + +/// Choose between two FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat32x3 ffxZeroOneSelect(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + FfxFloat32x3 r = (-x) * z + z; + return x * y + r; +} + +/// Choose between two FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat32x4 ffxZeroOneSelect(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + FfxFloat32x4 r = (-x) * z + z; + return x * y + r; +} + +/// Given a value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat32 ffxZeroOneIsSigned(FfxFloat32 x) +{ + return ffxSaturate(x * FfxFloat32(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// Given a value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat32x2 ffxZeroOneIsSigned(FfxFloat32x2 x) +{ + return ffxSaturate(x * ffxBroadcast2(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// Given a value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat32x3 ffxZeroOneIsSigned(FfxFloat32x3 x) +{ + return ffxSaturate(x * ffxBroadcast3(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// Given a value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat32x4 ffxZeroOneIsSigned(FfxFloat32x4 x) +{ + return ffxSaturate(x * ffxBroadcast4(FFX_NEGATIVE_INFINITY_FLOAT)); +} + +/// Compute a Rec.709 color space. +/// +/// Rec.709 is used for some HDTVs. +/// +/// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +/// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +/// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +/// +/// @param [in] color The color to convert to Rec. 709. +/// +/// @returns +/// The color in linear space. +/// +/// @ingroup GPU +FfxFloat32 ffxRec709FromLinear(FfxFloat32 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.018 * 4.5, 4.5, 0.45); + FfxFloat32x2 k = FfxFloat32x2(1.099, -0.099); + return clamp(j.x, color * j.y, pow(color, j.z) * k.x + k.y); +} + +/// Compute a Rec.709 color space. +/// +/// Rec.709 is used for some HDTVs. +/// +/// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +/// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +/// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +/// +/// @param [in] color The color to convert to Rec. 709. +/// +/// @returns +/// The color in linear space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxRec709FromLinear(FfxFloat32x2 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.018 * 4.5, 4.5, 0.45); + FfxFloat32x2 k = FfxFloat32x2(1.099, -0.099); + return clamp(j.xx, color * j.yy, pow(color, j.zz) * k.xx + k.yy); +} + +/// Compute a Rec.709 color space. +/// +/// Rec.709 is used for some HDTVs. +/// +/// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +/// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +/// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +/// +/// @param [in] color The color to convert to Rec. 709. +/// +/// @returns +/// The color in linear space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxRec709FromLinear(FfxFloat32x3 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.018 * 4.5, 4.5, 0.45); + FfxFloat32x2 k = FfxFloat32x2(1.099, -0.099); + return clamp(j.xxx, color * j.yyy, pow(color, j.zzz) * k.xxx + k.yyy); +} + +/// Compute a gamma value from a linear value. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// Note: 'rcpX' is '1/x', where the 'x' is what would be used in ffxLinearFromGamma. +/// +/// @param [in] value The value to convert to gamma space from linear. +/// @param [in] power The reciprocal of power value used for the gamma curve. +/// +/// @returns +/// A value in gamma space. +/// +/// @ingroup GPU +FfxFloat32 ffxGammaFromLinear(FfxFloat32 color, FfxFloat32 rcpX) +{ + return pow(color, FfxFloat32(rcpX)); +} + +/// Compute a gamma value from a linear value. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// Note: 'rcpX' is '1/x', where the 'x' is what would be used in ffxLinearFromGamma. +/// +/// @param [in] value The value to convert to gamma space from linear. +/// @param [in] power The reciprocal of power value used for the gamma curve. +/// +/// @returns +/// A value in gamma space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxGammaFromLinear(FfxFloat32x2 color, FfxFloat32 rcpX) +{ + return pow(color, ffxBroadcast2(rcpX)); +} + +/// Compute a gamma value from a linear value. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// Note: 'rcpX' is '1/x', where the 'x' is what would be used in ffxLinearFromGamma. +/// +/// @param [in] value The value to convert to gamma space from linear. +/// @param [in] power The reciprocal of power value used for the gamma curve. +/// +/// @returns +/// A value in gamma space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxGammaFromLinear(FfxFloat32x3 color, FfxFloat32 rcpX) +{ + return pow(color, ffxBroadcast3(rcpX)); +} + +/// Compute a PQ value from a linear value. +/// +/// @param [in] value The value to convert to PQ from linear. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32 ffxPQToLinear(FfxFloat32 x) +{ + FfxFloat32 p = pow(x, FfxFloat32(0.159302)); + return pow((FfxFloat32(0.835938) + FfxFloat32(18.8516) * p) / (FfxFloat32(1.0) + FfxFloat32(18.6875) * p), FfxFloat32(78.8438)); +} + +/// Compute a PQ value from a linear value. +/// +/// @param [in] value The value to convert to PQ from linear. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxPQToLinear(FfxFloat32x2 x) +{ + FfxFloat32x2 p = pow(x, ffxBroadcast2(0.159302)); + return pow((ffxBroadcast2(0.835938) + ffxBroadcast2(18.8516) * p) / (ffxBroadcast2(1.0) + ffxBroadcast2(18.6875) * p), ffxBroadcast2(78.8438)); +} + +/// Compute a PQ value from a linear value. +/// +/// @param [in] value The value to convert to PQ from linear. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxPQToLinear(FfxFloat32x3 x) +{ + FfxFloat32x3 p = pow(x, ffxBroadcast3(0.159302)); + return pow((ffxBroadcast3(0.835938) + ffxBroadcast3(18.8516) * p) / (ffxBroadcast3(1.0) + ffxBroadcast3(18.6875) * p), ffxBroadcast3(78.8438)); +} + +/// Compute a linear value from a SRGB value. +/// +/// @param [in] value The value to convert to linear from SRGB. +/// +/// @returns +/// A value in SRGB space. +/// +/// @ingroup GPU +FfxFloat32 ffxSrgbToLinear(FfxFloat32 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.0031308 * 12.92, 12.92, 1.0 / 2.4); + FfxFloat32x2 k = FfxFloat32x2(1.055, -0.055); + return clamp(j.x, color * j.y, pow(color, j.z) * k.x + k.y); +} + +/// Compute a linear value from a SRGB value. +/// +/// @param [in] value The value to convert to linear from SRGB. +/// +/// @returns +/// A value in SRGB space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxSrgbToLinear(FfxFloat32x2 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.0031308 * 12.92, 12.92, 1.0 / 2.4); + FfxFloat32x2 k = FfxFloat32x2(1.055, -0.055); + return clamp(j.xx, color * j.yy, pow(color, j.zz) * k.xx + k.yy); +} + +/// Compute a linear value from a SRGB value. +/// +/// @param [in] value The value to convert to linear from SRGB. +/// +/// @returns +/// A value in SRGB space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxSrgbToLinear(FfxFloat32x3 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.0031308 * 12.92, 12.92, 1.0 / 2.4); + FfxFloat32x2 k = FfxFloat32x2(1.055, -0.055); + return clamp(j.xxx, color * j.yyy, pow(color, j.zzz) * k.xxx + k.yyy); +} + +/// Compute a linear value from a REC.709 value. +/// +/// @param [in] color The value to convert to linear from REC.709. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32 ffxLinearFromRec709(FfxFloat32 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.081 / 4.5, 1.0 / 4.5, 1.0 / 0.45); + FfxFloat32x2 k = FfxFloat32x2(1.0 / 1.099, 0.099 / 1.099); + return ffxZeroOneSelect(ffxZeroOneIsSigned(color - j.x), color * j.y, pow(color * k.x + k.y, j.z)); +} + +/// Compute a linear value from a REC.709 value. +/// +/// @param [in] color The value to convert to linear from REC.709. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxLinearFromRec709(FfxFloat32x2 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.081 / 4.5, 1.0 / 4.5, 1.0 / 0.45); + FfxFloat32x2 k = FfxFloat32x2(1.0 / 1.099, 0.099 / 1.099); + return ffxZeroOneSelect(ffxZeroOneIsSigned(color - j.xx), color * j.yy, pow(color * k.xx + k.yy, j.zz)); +} + +/// Compute a linear value from a REC.709 value. +/// +/// @param [in] color The value to convert to linear from REC.709. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxLinearFromRec709(FfxFloat32x3 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.081 / 4.5, 1.0 / 4.5, 1.0 / 0.45); + FfxFloat32x2 k = FfxFloat32x2(1.0 / 1.099, 0.099 / 1.099); + return ffxZeroOneSelect(ffxZeroOneIsSigned(color - j.xxx), color * j.yyy, pow(color * k.xxx + k.yyy, j.zzz)); +} + +/// Compute a linear value from a value in a gamma space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] color The value to convert to linear in gamma space. +/// @param [in] power The power value used for the gamma curve. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32 ffxLinearFromGamma(FfxFloat32 color, FfxFloat32 power) +{ + return pow(color, FfxFloat32(power)); +} + +/// Compute a linear value from a value in a gamma space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] color The value to convert to linear in gamma space. +/// @param [in] power The power value used for the gamma curve. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxLinearFromGamma(FfxFloat32x2 color, FfxFloat32 power) +{ + return pow(color, ffxBroadcast2(power)); +} + +/// Compute a linear value from a value in a gamma space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] color The value to convert to linear in gamma space. +/// @param [in] power The power value used for the gamma curve. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxLinearFromGamma(FfxFloat32x3 color, FfxFloat32 power) +{ + return pow(color, ffxBroadcast3(power)); +} + +/// Compute a linear value from a value in a PQ space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] value The value to convert to linear in PQ space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32 ffxLinearFromPQ(FfxFloat32 x) +{ + FfxFloat32 p = pow(x, FfxFloat32(0.0126833)); + return pow(ffxSaturate(p - FfxFloat32(0.835938)) / (FfxFloat32(18.8516) - FfxFloat32(18.6875) * p), FfxFloat32(6.27739)); +} + +/// Compute a linear value from a value in a PQ space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] value The value to convert to linear in PQ space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxLinearFromPQ(FfxFloat32x2 x) +{ + FfxFloat32x2 p = pow(x, ffxBroadcast2(0.0126833)); + return pow(ffxSaturate(p - ffxBroadcast2(0.835938)) / (ffxBroadcast2(18.8516) - ffxBroadcast2(18.6875) * p), ffxBroadcast2(6.27739)); +} + +/// Compute a linear value from a value in a PQ space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] value The value to convert to linear in PQ space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxLinearFromPQ(FfxFloat32x3 x) +{ + FfxFloat32x3 p = pow(x, ffxBroadcast3(0.0126833)); + return pow(ffxSaturate(p - ffxBroadcast3(0.835938)) / (ffxBroadcast3(18.8516) - ffxBroadcast3(18.6875) * p), ffxBroadcast3(6.27739)); +} + +/// Compute a linear value from a value in a SRGB space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] value The value to convert to linear in SRGB space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32 ffxLinearFromSrgb(FfxFloat32 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.04045 / 12.92, 1.0 / 12.92, 2.4); + FfxFloat32x2 k = FfxFloat32x2(1.0 / 1.055, 0.055 / 1.055); + return ffxZeroOneSelect(ffxZeroOneIsSigned(color - j.x), color * j.y, pow(color * k.x + k.y, j.z)); +} + +/// Compute a linear value from a value in a SRGB space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] value The value to convert to linear in SRGB space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x2 ffxLinearFromSrgb(FfxFloat32x2 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.04045 / 12.92, 1.0 / 12.92, 2.4); + FfxFloat32x2 k = FfxFloat32x2(1.0 / 1.055, 0.055 / 1.055); + return ffxZeroOneSelect(ffxZeroOneIsSigned(color - j.xx), color * j.yy, pow(color * k.xx + k.yy, j.zz)); +} + +/// Compute a linear value from a value in a SRGB space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] value The value to convert to linear in SRGB space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat32x3 ffxLinearFromSrgb(FfxFloat32x3 color) +{ + FfxFloat32x3 j = FfxFloat32x3(0.04045 / 12.92, 1.0 / 12.92, 2.4); + FfxFloat32x2 k = FfxFloat32x2(1.0 / 1.055, 0.055 / 1.055); + return ffxZeroOneSelect(ffxZeroOneIsSigned(color - j.xxx), color * j.yyy, pow(color * k.xxx + k.yyy, j.zzz)); +} + +/// A remapping of 64x1 to 8x8 imposing rotated 2x2 pixel quads in quad linear. +/// +/// 543210 +/// ====== +/// ..xxx. +/// yy...y +/// +/// @param [in] a The input 1D coordinates to remap. +/// +/// @returns +/// The remapped 2D coordinates. +/// +/// @ingroup GPU +FfxUInt32x2 ffxRemapForQuad(FfxUInt32 a) +{ + return FfxUInt32x2(bitfieldExtract(a, 1u, 3u), bitfieldInsertMask(bitfieldExtract(a, 3u, 3u), a, 1u)); +} + +/// A helper function performing a remap 64x1 to 8x8 remapping which is necessary for 2D wave reductions. +/// +/// The 64-wide lane indices to 8x8 remapping is performed as follows: +/// +/// 00 01 08 09 10 11 18 19 +/// 02 03 0a 0b 12 13 1a 1b +/// 04 05 0c 0d 14 15 1c 1d +/// 06 07 0e 0f 16 17 1e 1f +/// 20 21 28 29 30 31 38 39 +/// 22 23 2a 2b 32 33 3a 3b +/// 24 25 2c 2d 34 35 3c 3d +/// 26 27 2e 2f 36 37 3e 3f +/// +/// @param [in] a The input 1D coordinate to remap. +/// +/// @returns +/// The remapped 2D coordinates. +/// +/// @ingroup GPU +FfxUInt32x2 ffxRemapForWaveReduction(FfxUInt32 a) +{ + return FfxUInt32x2(bitfieldInsertMask(bitfieldExtract(a, 2u, 3u), a, 1u), bitfieldInsertMask(bitfieldExtract(a, 3u, 3u), bitfieldExtract(a, 1u, 2u), 2u)); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_core_gpu_common_half.h b/thirdparty/amd-fsr2/shaders/ffx_core_gpu_common_half.h new file mode 100644 index 000000000000..c46ccb36575d --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core_gpu_common_half.h @@ -0,0 +1,2978 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#if FFX_HALF +#if FFX_HLSL_6_2 +/// A define value for 16bit positive infinity. +/// +/// @ingroup GPU +#define FFX_POSITIVE_INFINITY_HALF FFX_TO_FLOAT16((uint16_t)0x7c00u) + +/// A define value for 16bit negative infinity. +/// +/// @ingroup GPU +#define FFX_NEGATIVE_INFINITY_HALF FFX_TO_FLOAT16((uint16_t)0xfc00u) +#else +/// A define value for 16bit positive infinity. +/// +/// @ingroup GPU +#define FFX_POSITIVE_INFINITY_HALF FFX_TO_FLOAT16(0x7c00u) + +/// A define value for 16bit negative infinity. +/// +/// @ingroup GPU +#define FFX_NEGATIVE_INFINITY_HALF FFX_TO_FLOAT16(0xfc00u) +#endif // FFX_HLSL_6_2 + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16 ffxMin(FfxFloat16 x, FfxFloat16 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16x2 ffxMin(FfxFloat16x2 x, FfxFloat16x2 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16x3 ffxMin(FfxFloat16x3 x, FfxFloat16x3 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16x4 ffxMin(FfxFloat16x4 x, FfxFloat16x4 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16 ffxMin(FfxInt16 x, FfxInt16 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16x2 ffxMin(FfxInt16x2 x, FfxInt16x2 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16x3 ffxMin(FfxInt16x3 x, FfxInt16x3 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16x4 ffxMin(FfxInt16x4 x, FfxInt16x4 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16 ffxMin(FfxUInt16 x, FfxUInt16 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16x2 ffxMin(FfxUInt16x2 x, FfxUInt16x2 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16x3 ffxMin(FfxUInt16x3 x, FfxUInt16x3 y) +{ + return min(x, y); +} + +/// Compute the min of two values. +/// +/// @param [in] x The first value to compute the min of. +/// @param [in] y The second value to compute the min of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16x4 ffxMin(FfxUInt16x4 x, FfxUInt16x4 y) +{ + return min(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16 ffxMax(FfxFloat16 x, FfxFloat16 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16x2 ffxMax(FfxFloat16x2 x, FfxFloat16x2 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16x3 ffxMax(FfxFloat16x3 x, FfxFloat16x3 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxFloat16x4 ffxMax(FfxFloat16x4 x, FfxFloat16x4 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16 ffxMax(FfxInt16 x, FfxInt16 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16x2 ffxMax(FfxInt16x2 x, FfxInt16x2 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16x3 ffxMax(FfxInt16x3 x, FfxInt16x3 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxInt16x4 ffxMax(FfxInt16x4 x, FfxInt16x4 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16 ffxMax(FfxUInt16 x, FfxUInt16 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16x2 ffxMax(FfxUInt16x2 x, FfxUInt16x2 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16x3 ffxMax(FfxUInt16x3 x, FfxUInt16x3 y) +{ + return max(x, y); +} + +/// Compute the max of two values. +/// +/// @param [in] x The first value to compute the max of. +/// @param [in] y The second value to compute the max of. +/// +/// @returns +/// The the lowest of two values. +/// +/// @ingroup GPU +FfxUInt16x4 ffxMax(FfxUInt16x4 x, FfxUInt16x4 y) +{ + return max(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat16 ffxPow(FfxFloat16 x, FfxFloat16 y) +{ + return pow(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPow(FfxFloat16x2 x, FfxFloat16x2 y) +{ + return pow(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat16x3 ffxPow(FfxFloat16x3 x, FfxFloat16x3 y) +{ + return pow(x, y); +} + +/// Compute the value of the first parameter raised to the power of the second. +/// +/// @param [in] x The value to raise to the power y. +/// @param [in] y The power to which to raise x. +/// +/// @returns +/// The value of the first parameter raised to the power of the second. +/// +/// @ingroup GPU +FfxFloat16x4 ffxPow(FfxFloat16x4 x, FfxFloat16x4 y) +{ + return pow(x, y); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat16 ffxSqrt(FfxFloat16 x) +{ + return sqrt(x); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat16x2 ffxSqrt(FfxFloat16x2 x) +{ + return sqrt(x); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat16x3 ffxSqrt(FfxFloat16x3 x) +{ + return sqrt(x); +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup GPU +FfxFloat16x4 ffxSqrt(FfxFloat16x4 x) +{ + return sqrt(x); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat16 ffxCopySignBitHalf(FfxFloat16 d, FfxFloat16 s) +{ + return FFX_TO_FLOAT16(FFX_TO_UINT16(d) | (FFX_TO_UINT16(s) & FFX_BROADCAST_UINT16(0x8000u))); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat16x2 ffxCopySignBitHalf(FfxFloat16x2 d, FfxFloat16x2 s) +{ + return FFX_TO_FLOAT16X2(FFX_TO_UINT16X2(d) | (FFX_TO_UINT16X2(s) & FFX_BROADCAST_UINT16X2(0x8000u))); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat16x3 ffxCopySignBitHalf(FfxFloat16x3 d, FfxFloat16x3 s) +{ + return FFX_TO_FLOAT16X3(FFX_TO_UINT16X3(d) | (FFX_TO_UINT16X3(s) & FFX_BROADCAST_UINT16X3(0x8000u))); +} + +/// Copy the sign bit from 's' to positive 'd'. +/// +/// @param [in] d The value to copy the sign bit into. +/// @param [in] s The value to copy the sign bit from. +/// +/// @returns +/// The value of d with the sign bit from s. +/// +/// @ingroup GPU +FfxFloat16x4 ffxCopySignBitHalf(FfxFloat16x4 d, FfxFloat16x4 s) +{ + return FFX_TO_FLOAT16X4(FFX_TO_UINT16X4(d) | (FFX_TO_UINT16X4(s) & FFX_BROADCAST_UINT16X4(0x8000u))); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat16 ffxIsSignedHalf(FfxFloat16 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat16x2 ffxIsSignedHalf(FfxFloat16x2 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16X2(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat16x3 ffxIsSignedHalf(FfxFloat16x3 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16X3(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 0 +/// m >= 0 := 0 +/// m < 0 := 1 +/// +/// Uses the following useful floating point logic, +/// saturate(+a*(-INF)==-INF) := 0 +/// saturate( 0*(-INF)== NaN) := 0 +/// saturate(-a*(-INF)==+INF) := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against 0. +/// +/// @returns +/// 1.0 when the value is negative, or 0.0 when the value is 0 or position. +/// +/// @ingroup GPU +FfxFloat16x4 ffxIsSignedHalf(FfxFloat16x4 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16X4(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat16 ffxIsGreaterThanZeroHalf(FfxFloat16 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16(FFX_POSITIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat16x2 ffxIsGreaterThanZeroHalf(FfxFloat16x2 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16X2(FFX_POSITIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat16x3 ffxIsGreaterThanZeroHalf(FfxFloat16x3 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16X3(FFX_POSITIVE_INFINITY_HALF)); +} + +/// A single operation to return the following: +/// m = NaN := 1 +/// m > 0 := 0 +/// m <= 0 := 1 +/// +/// This function is useful when creating masks for branch-free logic. +/// +/// @param [in] m The value to test against zero. +/// +/// @returns +/// 1.0 when the value is position, or 0.0 when the value is 0 or negative. +/// +/// @ingroup GPU +FfxFloat16x4 ffxIsGreaterThanZeroHalf(FfxFloat16x4 m) +{ + return ffxSaturate(m * FFX_BROADCAST_FLOAT16X4(FFX_POSITIVE_INFINITY_HALF)); +} + +/// Convert a 16bit floating point value to sortable integer. +/// +/// - If sign bit=0, flip the sign bit (positives). +/// - If sign bit=1, flip all bits (negatives). +/// +/// The function has the side effects that: +/// - Larger integers are more positive values. +/// - Float zero is mapped to center of integers (so clear to integer zero is a nice default for atomic max usage). +/// +/// @param [in] x The floating point value to make sortable. +/// +/// @returns +/// The sortable integer value. +/// +/// @ingroup GPU +FfxUInt16 ffxFloatToSortableIntegerHalf(FfxUInt16 x) +{ + return x ^ ((ffxBitShiftRightHalf(x, FFX_BROADCAST_UINT16(15))) | FFX_BROADCAST_UINT16(0x8000)); +} + +/// Convert a sortable integer to a 16bit floating point value. +/// +/// The function has the side effects that: +/// - If sign bit=1, flip the sign bit (positives). +/// - If sign bit=0, flip all bits (negatives). +/// +/// @param [in] x The sortable integer value to make floating point. +/// +/// @returns +/// The floating point value. +/// +/// @ingroup GPU +FfxUInt16 ffxSortableIntegerToFloatHalf(FfxUInt16 x) +{ + return x ^ ((~ffxBitShiftRightHalf(x, FFX_BROADCAST_UINT16(15))) | FFX_BROADCAST_UINT16(0x8000)); +} + +/// Convert a pair of 16bit floating point values to a pair of sortable integers. +/// +/// - If sign bit=0, flip the sign bit (positives). +/// - If sign bit=1, flip all bits (negatives). +/// +/// The function has the side effects that: +/// - Larger integers are more positive values. +/// - Float zero is mapped to center of integers (so clear to integer zero is a nice default for atomic max usage). +/// +/// @param [in] x The floating point values to make sortable. +/// +/// @returns +/// The sortable integer values. +/// +/// @ingroup GPU +FfxUInt16x2 ffxFloatToSortableIntegerHalf(FfxUInt16x2 x) +{ + return x ^ ((ffxBitShiftRightHalf(x, FFX_BROADCAST_UINT16X2(15))) | FFX_BROADCAST_UINT16X2(0x8000)); +} + +/// Convert a pair of sortable integers to a pair of 16bit floating point values. +/// +/// The function has the side effects that: +/// - If sign bit=1, flip the sign bit (positives). +/// - If sign bit=0, flip all bits (negatives). +/// +/// @param [in] x The sortable integer values to make floating point. +/// +/// @returns +/// The floating point values. +/// +/// @ingroup GPU +FfxUInt16x2 ffxSortableIntegerToFloatHalf(FfxUInt16x2 x) +{ + return x ^ ((~ffxBitShiftRightHalf(x, FFX_BROADCAST_UINT16X2(15))) | FFX_BROADCAST_UINT16X2(0x8000)); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// [Zero] Y0 [Zero] X0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesZeroY0ZeroX0(FfxUInt32x2 i) +{ + return ((i.x) & 0xffu) | ((i.y << 16) & 0xff0000u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// [Zero] Y1 [Zero] X1 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesZeroY1ZeroX1(FfxUInt32x2 i) +{ + return ((i.x >> 8) & 0xffu) | ((i.y << 8) & 0xff0000u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// [Zero] Y2 [Zero] X2 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesZeroY2ZeroX2(FfxUInt32x2 i) +{ + return ((i.x >> 16) & 0xffu) | ((i.y) & 0xff0000u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// [Zero] Y3 [Zero] X3 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesZeroY3ZeroX3(FfxUInt32x2 i) +{ + return ((i.x >> 24) & 0xffu) | ((i.y >> 8) & 0xff0000u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y3 Y2 Y1 X0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY3Y2Y1X0(FfxUInt32x2 i) +{ + return ((i.x) & 0x000000ffu) | (i.y & 0xffffff00u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y3 Y2 Y1 X2 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY3Y2Y1X2(FfxUInt32x2 i) +{ + return ((i.x >> 16) & 0x000000ffu) | (i.y & 0xffffff00u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y3 Y2 X0 Y0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY3Y2X0Y0(FfxUInt32x2 i) +{ + return ((i.x << 8) & 0x0000ff00u) | (i.y & 0xffff00ffu); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y3 Y2 X2 Y0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY3Y2X2Y0(FfxUInt32x2 i) +{ + return ((i.x >> 8) & 0x0000ff00u) | (i.y & 0xffff00ffu); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y3 X0 Y1 Y0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY3X0Y1Y0(FfxUInt32x2 i) +{ + return ((i.x << 16) & 0x00ff0000u) | (i.y & 0xff00ffffu); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y3 X2 Y1 Y0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY3X2Y1Y0(FfxUInt32x2 i) +{ + return ((i.x) & 0x00ff0000u) | (i.y & 0xff00ffffu); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// X0 Y2 Y1 Y0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesX0Y2Y1Y0(FfxUInt32x2 i) +{ + return ((i.x << 24) & 0xff000000u) | (i.y & 0x00ffffffu); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// X2 Y2 Y1 Y0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesX2Y2Y1Y0(FfxUInt32x2 i) +{ + return ((i.x << 8) & 0xff000000u) | (i.y & 0x00ffffffu); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y2 X2 Y0 X0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY2X2Y0X0(FfxUInt32x2 i) +{ + return ((i.x) & 0x00ff00ffu) | ((i.y << 8) & 0xff00ff00u); +} + +/// Packs the bytes from the X and Y components of a FfxUInt32x2 into a single 32-bit integer. +/// +/// The resulting integer will contain bytes in the following order, from most to least significant: +/// Y2 Y0 X2 X0 +/// +/// @param [in] i The integer pair to pack. +/// +/// @returns +/// The packed integer value. +/// +/// @ingroup GPU +FfxUInt32 ffxPackBytesY2Y0X2X0(FfxUInt32x2 i) +{ + return (((i.x) & 0xffu) | ((i.x >> 8) & 0xff00u) | ((i.y << 16) & 0xff0000u) | ((i.y << 8) & 0xff000000u)); +} + +/// Takes two Float16x2 values x and y, normalizes them and builds a single Uint16x2 value in the format {{x0,y0},{x1,y1}}. +/// +/// @param [in] x The first float16x2 value to pack. +/// @param [in] y The second float16x2 value to pack. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt16x2 ffxPackX0Y0X1Y1UnsignedToUint16x2(FfxFloat16x2 x, FfxFloat16x2 y) +{ + x *= FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0); + y *= FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0); + return FFX_UINT32_TO_UINT16X2(ffxPackBytesY2X2Y0X0(FfxUInt32x2(FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(x)), FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(y))))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[0:7], +/// d.y[0:7] into r.y[0:7], i.x[8:15] into r.x[8:15], r.y[8:15] and i.y[0:15] into r.x[16:31], r.y[16:31] using 3 ops. +/// +/// r=ffxPermuteUByte0Float16x2ToUint2(d,i) +/// Where 'k0' is an SGPR with {1.0/32768.0} packed into the lower 16-bits +/// Where 'k1' is an SGPR with 0x???? +/// Where 'k2' is an SGPR with 0x???? +/// V_PK_FMA_F16 i,i,k0.x,0 +/// V_PERM_B32 r.x,i,i,k1 +/// V_PERM_B32 r.y,i,i,k2 +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteUByte0Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0))); + return FfxUInt32x2(ffxPackBytesY3Y2Y1X0(FfxUInt32x2(d.x, b)), ffxPackBytesY3Y2Y1X2(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[8:15], +/// d.y[0:7] into r.y[8:15], i.x[0:7] into r.x[0:7], r.y[0:7] and i.y[0:15] into r.x[16:31], r.y[16:31] using 3 ops. +/// +/// r=ffxPermuteUByte1Float16x2ToUint2(d,i) +/// Where 'k0' is an SGPR with {1.0/32768.0} packed into the lower 16-bits +/// Where 'k1' is an SGPR with 0x???? +/// Where 'k2' is an SGPR with 0x???? +/// V_PK_FMA_F16 i,i,k0.x,0 +/// V_PERM_B32 r.x,i,i,k1 +/// V_PERM_B32 r.y,i,i,k2 +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteUByte1Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0))); + return FfxUInt32x2(ffxPackBytesY3Y2X0Y0(FfxUInt32x2(d.x, b)), ffxPackBytesY3Y2X2Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[16:23], +/// d.y[0:7] into r.y[16:23], i.x[0:15] into r.x[0:15], r.y[0:15] and i.y[8:15] into r.x[24:31], r.y[24:31] using 3 ops. +/// +/// r=ffxPermuteUByte2Float16x2ToUint2(d,i) +/// Where 'k0' is an SGPR with {1.0/32768.0} packed into the lower 16-bits +/// Where 'k1' is an SGPR with 0x???? +/// Where 'k2' is an SGPR with 0x???? +/// V_PK_FMA_F16 i,i,k0.x,0 +/// V_PERM_B32 r.x,i,i,k1 +/// V_PERM_B32 r.y,i,i,k2 +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteUByte2Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0))); + return FfxUInt32x2(ffxPackBytesY3X0Y1Y0(FfxUInt32x2(d.x, b)), ffxPackBytesY3X2Y1Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[24:31], +/// d.y[0:7] into r.y[24:31], i.x[0:15] into r.x[0:15], r.y[0:15] and i.y[0:7] into r.x[16:23], r.y[16:23] using 3 ops. +/// +/// r=ffxPermuteUByte3Float16x2ToUint2(d,i) +/// Where 'k0' is an SGPR with {1.0/32768.0} packed into the lower 16-bits +/// Where 'k1' is an SGPR with 0x???? +/// Where 'k2' is an SGPR with 0x???? +/// V_PK_FMA_F16 i,i,k0.x,0 +/// V_PERM_B32 r.x,i,i,k1 +/// V_PERM_B32 r.y,i,i,k2 +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteUByte3Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0))); + return FfxUInt32x2(ffxPackBytesX0Y2Y1Y0(FfxUInt32x2(d.x, b)), ffxPackBytesX2Y2Y1Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[0:7] into r.x[0:7] and i.y[0:7] into r.y[0:7] using 2 ops. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteUByte0Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY0ZeroX0(i))) * FFX_BROADCAST_FLOAT16X2(32768.0); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[8:15] into r.x[0:7] and i.y[8:15] into r.y[0:7] using 2 ops. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteUByte1Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY1ZeroX1(i))) * FFX_BROADCAST_FLOAT16X2(32768.0); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[16:23] into r.x[0:7] and i.y[16:23] into r.y[0:7] using 2 ops. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteUByte2Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY2ZeroX2(i))) * FFX_BROADCAST_FLOAT16X2(32768.0); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[24:31] into r.x[0:7] and i.y[24:31] into r.y[0:7] using 2 ops. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteUByte3Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY3ZeroX3(i))) * FFX_BROADCAST_FLOAT16X2(32768.0); +} + +/// Takes two Float16x2 values x and y, normalizes them and builds a single Uint16x2 value in the format {{x0,y0},{x1,y1}}. +/// +/// @param [in] x The first float16x2 value to pack. +/// @param [in] y The second float16x2 value to pack. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt16x2 ffxPackX0Y0X1Y1SignedToUint16x2(FfxFloat16x2 x, FfxFloat16x2 y) +{ + x = x * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0); + y = y * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0); + return FFX_UINT32_TO_UINT16X2(ffxPackBytesY2X2Y0X0(FfxUInt32x2(FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(x)), FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(y))))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[0:7], +/// d.y[0:7] into r.y[0:7], i.x[8:15] into r.x[8:15], r.y[8:15] and i.y[0:15] into r.x[16:31], r.y[16:31] using 3 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteSByte0Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))); + return FfxUInt32x2(ffxPackBytesY3Y2Y1X0(FfxUInt32x2(d.x, b)), ffxPackBytesY3Y2Y1X2(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[8:15], +/// d.y[0:7] into r.y[8:15], i.x[0:7] into r.x[0:7], r.y[0:7] and i.y[0:15] into r.x[16:31], r.y[16:31] using 3 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteSByte1Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))); + return FfxUInt32x2(ffxPackBytesY3Y2X0Y0(FfxUInt32x2(d.x, b)), ffxPackBytesY3Y2X2Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[16:23], +/// d.y[0:7] into r.y[16:23], i.x[0:15] into r.x[0:15], r.y[0:15] and i.y[8:15] into r.x[24:31], r.y[24:31] using 3 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteSByte2Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))); + return FfxUInt32x2(ffxPackBytesY3X0Y1Y0(FfxUInt32x2(d.x, b)), ffxPackBytesY3X2Y1Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[24:31], +/// d.y[0:7] into r.y[24:31], i.x[0:15] into r.x[0:15], r.y[0:15] and i.y[0:7] into r.x[16:23], r.y[16:23] using 3 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteSByte3Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))); + return FfxUInt32x2(ffxPackBytesX0Y2Y1Y0(FfxUInt32x2(d.x, b)), ffxPackBytesX2Y2Y1Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[0:7], +/// d.y[0:7] into r.y[0:7], i.x[8:15] into r.x[8:15], r.y[8:15] and i.y[0:15] into r.x[16:31], r.y[16:31] using 3 ops. +/// +/// Zero-based flips the MSB bit of the byte (making 128 "exact zero" actually zero). +/// This is useful if there is a desire for cleared values to decode as zero. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteZeroBasedSByte0Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))) ^ 0x00800080u; + return FfxUInt32x2(ffxPackBytesY3Y2Y1X0(FfxUInt32x2(d.x, b)), ffxPackBytesY3Y2Y1X2(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[8:15], +/// d.y[0:7] into r.y[8:15], i.x[0:7] into r.x[0:7], r.y[0:7] and i.y[0:15] into r.x[16:31], r.y[16:31] using 3 ops. +/// +/// Zero-based flips the MSB bit of the byte (making 128 "exact zero" actually zero). +/// This is useful if there is a desire for cleared values to decode as zero. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteZeroBasedSByte1Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))) ^ 0x00800080u; + return FfxUInt32x2(ffxPackBytesY3Y2X0Y0(FfxUInt32x2(d.x, b)), ffxPackBytesY3Y2X2Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[16:23], +/// d.y[0:7] into r.y[16:23], i.x[0:15] into r.x[0:15], r.y[0:15] and i.y[8:15] into r.x[24:31], r.y[24:31] using 3 ops. +/// +/// Zero-based flips the MSB bit of the byte (making 128 "exact zero" actually zero). +/// This is useful if there is a desire for cleared values to decode as zero. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteZeroBasedSByte2Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))) ^ 0x00800080u; + return FfxUInt32x2(ffxPackBytesY3X0Y1Y0(FfxUInt32x2(d.x, b)), ffxPackBytesY3X2Y1Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value d, Float16x2 value i and a resulting FfxUInt32x2 value r, this function packs d.x[0:7] into r.x[24:31], +/// d.y[0:7] into r.y[24:31], i.x[0:15] into r.x[0:15], r.y[0:15] and i.y[0:7] into r.x[16:23], r.y[16:23] using 3 ops. +/// +/// Zero-based flips the MSB bit of the byte (making 128 "exact zero" actually zero). +/// This is useful if there is a desire for cleared values to decode as zero. +/// +/// Handles signed byte values. +/// +/// @param [in] d The FfxUInt32x2 value to be packed. +/// @param [in] i The FfxFloat16x2 value to be packed. +/// +/// @returns +/// The packed FfxUInt32x2 value. +/// +/// @ingroup GPU +FfxUInt32x2 ffxPermuteZeroBasedSByte3Float16x2ToUint2(FfxUInt32x2 d, FfxFloat16x2 i) +{ + FfxUInt32 b = FFX_UINT16X2_TO_UINT32(FFX_TO_UINT16X2(i * FFX_BROADCAST_FLOAT16X2(1.0 / 32768.0) + FFX_BROADCAST_FLOAT16X2(0.25 / 32768.0))) ^ 0x00800080u; + return FfxUInt32x2(ffxPackBytesX0Y2Y1Y0(FfxUInt32x2(d.x, b)), ffxPackBytesX2Y2Y1Y0(FfxUInt32x2(d.y, b))); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[0:7] into r.x[0:7] and i.y[0:7] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteSByte0Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY0ZeroX0(i))) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[8:15] into r.x[0:7] and i.y[8:15] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteSByte1Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY1ZeroX1(i))) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[16:23] into r.x[0:7] and i.y[16:23] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteSByte2Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY2ZeroX2(i))) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[24:31] into r.x[0:7] and i.y[24:31] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteSByte3Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY3ZeroX3(i))) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[0:7] into r.x[0:7] and i.y[0:7] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteZeroBasedSByte0Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY0ZeroX0(i) ^ 0x00800080u)) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[8:15] into r.x[0:7] and i.y[8:15] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteZeroBasedSByte1Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY1ZeroX1(i) ^ 0x00800080u)) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[16:23] into r.x[0:7] and i.y[16:23] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteZeroBasedSByte2Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY2ZeroX2(i) ^ 0x00800080u)) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Given a FfxUInt32x2 value i and a resulting Float16x2 value r, this function packs i.x[24:31] into r.x[0:7] and i.y[24:31] into r.y[0:7] using 2 ops. +/// +/// Handles signed byte values. +/// +/// @param [in] i The FfxUInt32x2 value to be unpacked. +/// +/// @returns +/// The unpacked FfxFloat16x2. +/// +/// @ingroup GPU +FfxFloat16x2 ffxPermuteZeroBasedSByte3Uint2ToFloat16x2(FfxUInt32x2 i) +{ + return FFX_TO_FLOAT16X2(FFX_UINT32_TO_UINT16X2(ffxPackBytesZeroY3ZeroX3(i) ^ 0x00800080u)) * FFX_BROADCAST_FLOAT16X2(32768.0) - FFX_BROADCAST_FLOAT16X2(0.25); +} + +/// Calculate a half-precision low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16 ffxApproximateSqrtHalf(FfxFloat16 a) +{ + return FFX_TO_FLOAT16((FFX_TO_UINT16(a) >> FFX_BROADCAST_UINT16(1)) + FFX_BROADCAST_UINT16(0x1de2)); +} + +/// Calculate a half-precision low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x2 ffxApproximateSqrtHalf(FfxFloat16x2 a) +{ + return FFX_TO_FLOAT16X2((FFX_TO_UINT16X2(a) >> FFX_BROADCAST_UINT16X2(1)) + FFX_BROADCAST_UINT16X2(0x1de2)); +} + +/// Calculate a half-precision low-quality approximation for the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the square root for. +/// +/// @returns +/// An approximation of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x3 ffxApproximateSqrtHalf(FfxFloat16x3 a) +{ + return FFX_TO_FLOAT16X3((FFX_TO_UINT16X3(a) >> FFX_BROADCAST_UINT16X3(1)) + FFX_BROADCAST_UINT16X3(0x1de2)); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16 ffxApproximateReciprocalHalf(FfxFloat16 a) +{ + return FFX_TO_FLOAT16(FFX_BROADCAST_UINT16(0x7784) - FFX_TO_UINT16(a)); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x2 ffxApproximateReciprocalHalf(FfxFloat16x2 a) +{ + return FFX_TO_FLOAT16X2(FFX_BROADCAST_UINT16X2(0x7784) - FFX_TO_UINT16X2(a)); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x3 ffxApproximateReciprocalHalf(FfxFloat16x3 a) +{ + return FFX_TO_FLOAT16X3(FFX_BROADCAST_UINT16X3(0x7784) - FFX_TO_UINT16X3(a)); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x4 ffxApproximateReciprocalHalf(FfxFloat16x4 a) +{ + return FFX_TO_FLOAT16X4(FFX_BROADCAST_UINT16X4(0x7784) - FFX_TO_UINT16X4(a)); +} + +/// Calculate a half-precision medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat16 ffxApproximateReciprocalMediumHalf(FfxFloat16 a) +{ + FfxFloat16 b = FFX_TO_FLOAT16(FFX_BROADCAST_UINT16(0x778d) - FFX_TO_UINT16(a)); + return b * (-b * a + FFX_BROADCAST_FLOAT16(2.0)); +} + +/// Calculate a half-precision medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat16x2 ffxApproximateReciprocalMediumHalf(FfxFloat16x2 a) +{ + FfxFloat16x2 b = FFX_TO_FLOAT16X2(FFX_BROADCAST_UINT16X2(0x778d) - FFX_TO_UINT16X2(a)); + return b * (-b * a + FFX_BROADCAST_FLOAT16X2(2.0)); +} + +/// Calculate a half-precision medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat16x3 ffxApproximateReciprocalMediumHalf(FfxFloat16x3 a) +{ + FfxFloat16x3 b = FFX_TO_FLOAT16X3(FFX_BROADCAST_UINT16X3(0x778d) - FFX_TO_UINT16X3(a)); + return b * (-b * a + FFX_BROADCAST_FLOAT16X3(2.0)); +} + +/// Calculate a half-precision medium-quality approximation for the reciprocal of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal for. +/// +/// @returns +/// An approximation of the reciprocal, estimated to medium quality. +/// +/// @ingroup GPU +FfxFloat16x4 ffxApproximateReciprocalMediumHalf(FfxFloat16x4 a) +{ + FfxFloat16x4 b = FFX_TO_FLOAT16X4(FFX_BROADCAST_UINT16X4(0x778d) - FFX_TO_UINT16X4(a)); + return b * (-b * a + FFX_BROADCAST_FLOAT16X4(2.0)); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal of the square root for. +/// +/// @returns +/// An approximation of the reciprocal of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16 ffxApproximateReciprocalSquareRootHalf(FfxFloat16 a) +{ + return FFX_TO_FLOAT16(FFX_BROADCAST_UINT16(0x59a3) - (FFX_TO_UINT16(a) >> FFX_BROADCAST_UINT16(1))); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal of the square root for. +/// +/// @returns +/// An approximation of the reciprocal of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x2 ffxApproximateReciprocalSquareRootHalf(FfxFloat16x2 a) +{ + return FFX_TO_FLOAT16X2(FFX_BROADCAST_UINT16X2(0x59a3) - (FFX_TO_UINT16X2(a) >> FFX_BROADCAST_UINT16X2(1))); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal of the square root for. +/// +/// @returns +/// An approximation of the reciprocal of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x3 ffxApproximateReciprocalSquareRootHalf(FfxFloat16x3 a) +{ + return FFX_TO_FLOAT16X3(FFX_BROADCAST_UINT16X3(0x59a3) - (FFX_TO_UINT16X3(a) >> FFX_BROADCAST_UINT16X3(1))); +} + +/// Calculate a half-precision low-quality approximation for the reciprocal of the square root of a value. +/// +/// For additional information on the approximation family of functions, you can refer to Michal Drobot's excellent +/// presentation materials: +/// +/// - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf +/// - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h +/// +/// @param [in] a The value to calculate an approximate to the reciprocal of the square root for. +/// +/// @returns +/// An approximation of the reciprocal of the square root, estimated to low quality. +/// +/// @ingroup GPU +FfxFloat16x4 ffxApproximateReciprocalSquareRootHalf(FfxFloat16x4 a) +{ + return FFX_TO_FLOAT16X4(FFX_BROADCAST_UINT16X4(0x59a3) - (FFX_TO_UINT16X4(a) >> FFX_BROADCAST_UINT16X4(1))); +} + +/// An approximation of sine. +/// +/// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +/// is {-1/4 to 1/4} representing {-1 to 1}. +/// +/// @param [in] x The value to calculate approximate sine for. +/// +/// @returns +/// The approximate sine of value. +FfxFloat16 ffxParabolicSinHalf(FfxFloat16 x) +{ + return x * abs(x) - x; +} + +/// An approximation of sine. +/// +/// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +/// is {-1/4 to 1/4} representing {-1 to 1}. +/// +/// @param [in] x The value to calculate approximate sine for. +/// +/// @returns +/// The approximate sine of value. +FfxFloat16x2 ffxParabolicSinHalf(FfxFloat16x2 x) +{ + return x * abs(x) - x; +} + +/// An approximation of cosine. +/// +/// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +/// is {-1/4 to 1/4} representing {-1 to 1}. +/// +/// @param [in] x The value to calculate approximate cosine for. +/// +/// @returns +/// The approximate cosine of value. +FfxFloat16 ffxParabolicCosHalf(FfxFloat16 x) +{ + x = ffxFract(x * FFX_BROADCAST_FLOAT16(0.5) + FFX_BROADCAST_FLOAT16(0.75)); + x = x * FFX_BROADCAST_FLOAT16(2.0) - FFX_BROADCAST_FLOAT16(1.0); + return ffxParabolicSinHalf(x); +} + +/// An approximation of cosine. +/// +/// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +/// is {-1/4 to 1/4} representing {-1 to 1}. +/// +/// @param [in] x The value to calculate approximate cosine for. +/// +/// @returns +/// The approximate cosine of value. +FfxFloat16x2 ffxParabolicCosHalf(FfxFloat16x2 x) +{ + x = ffxFract(x * FFX_BROADCAST_FLOAT16X2(0.5) + FFX_BROADCAST_FLOAT16X2(0.75)); + x = x * FFX_BROADCAST_FLOAT16X2(2.0) - FFX_BROADCAST_FLOAT16X2(1.0); + return ffxParabolicSinHalf(x); +} + +/// An approximation of both sine and cosine. +/// +/// Valid input range is {-1 to 1} representing {0 to 2 pi}, and the output range +/// is {-1/4 to 1/4} representing {-1 to 1}. +/// +/// @param [in] x The value to calculate approximate cosine for. +/// +/// @returns +/// A FfxFloat32x2 containing approximations of both sine and cosine of value. +FfxFloat16x2 ffxParabolicSinCosHalf(FfxFloat16 x) +{ + FfxFloat16 y = ffxFract(x * FFX_BROADCAST_FLOAT16(0.5) + FFX_BROADCAST_FLOAT16(0.75)); + y = y * FFX_BROADCAST_FLOAT16(2.0) - FFX_BROADCAST_FLOAT16(1.0); + return ffxParabolicSinHalf(FfxFloat16x2(x, y)); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt16 ffxZeroOneAndHalf(FfxUInt16 x, FfxUInt16 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt16x2 ffxZeroOneAndHalf(FfxUInt16x2 x, FfxUInt16x2 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt16x3 ffxZeroOneAndHalf(FfxUInt16x3 x, FfxUInt16x3 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxUInt16x4 ffxZeroOneAndHalf(FfxUInt16x4 x, FfxUInt16x4 y) +{ + return min(x, y); +} + +/// Conditional free logic NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt16 ffxZeroOneNotHalf(FfxUInt16 x) +{ + return x ^ FFX_BROADCAST_UINT16(1); +} + +/// Conditional free logic NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt16x2 ffxZeroOneNotHalf(FfxUInt16x2 x) +{ + return x ^ FFX_BROADCAST_UINT16X2(1); +} + +/// Conditional free logic NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt16x3 ffxZeroOneNotHalf(FfxUInt16x3 x) +{ + return x ^ FFX_BROADCAST_UINT16X3(1); +} + +/// Conditional free logic NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the NOT operator. +/// @param [in] y The second value to be fed into the NOT operator. +/// +/// @returns +/// Result of the NOT operation. +/// +/// @ingroup GPU +FfxUInt16x4 ffxZeroOneNotHalf(FfxUInt16x4 x) +{ + return x ^ FFX_BROADCAST_UINT16X4(1); +} + +/// Conditional free logic OR operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt16 ffxZeroOneOrHalf(FfxUInt16 x, FfxUInt16 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt16x2 ffxZeroOneOrHalf(FfxUInt16x2 x, FfxUInt16x2 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt16x3 ffxZeroOneOrHalf(FfxUInt16x3 x, FfxUInt16x3 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxUInt16x4 ffxZeroOneOrHalf(FfxUInt16x4 x, FfxUInt16x4 y) +{ + return max(x, y); +} + +/// Convert a half-precision FfxFloat32 value between 0.0f and 1.0f to a half-precision Uint. +/// +/// @param [in] x The value to converted to a Uint. +/// +/// @returns +/// The converted Uint value. +/// +/// @ingroup GPU +FfxUInt16 ffxZeroOneFloat16ToUint16(FfxFloat16 x) +{ + return FFX_TO_UINT16(x * FFX_TO_FLOAT16(FFX_TO_UINT16(1))); +} + +/// Convert a half-precision FfxFloat32 value between 0.0f and 1.0f to a half-precision Uint. +/// +/// @param [in] x The value to converted to a Uint. +/// +/// @returns +/// The converted Uint value. +/// +/// @ingroup GPU +FfxUInt16x2 ffxZeroOneFloat16x2ToUint16x2(FfxFloat16x2 x) +{ + return FFX_TO_UINT16X2(x * FFX_TO_FLOAT16X2(FfxUInt16x2(1, 1))); +} + +/// Convert a half-precision FfxFloat32 value between 0.0f and 1.0f to a half-precision Uint. +/// +/// @param [in] x The value to converted to a Uint. +/// +/// @returns +/// The converted Uint value. +/// +/// @ingroup GPU +FfxUInt16x3 ffxZeroOneFloat16x3ToUint16x3(FfxFloat16x3 x) +{ + return FFX_TO_UINT16X3(x * FFX_TO_FLOAT16X3(FfxUInt16x3(1, 1, 1))); +} + +/// Convert a half-precision FfxFloat32 value between 0.0f and 1.0f to a half-precision Uint. +/// +/// @param [in] x The value to converted to a Uint. +/// +/// @returns +/// The converted Uint value. +/// +/// @ingroup GPU +FfxUInt16x4 ffxZeroOneFloat16x4ToUint16x4(FfxFloat16x4 x) +{ + return FFX_TO_UINT16X4(x * FFX_TO_FLOAT16X4(FfxUInt16x4(1, 1, 1, 1))); +} + +/// Convert a half-precision FfxUInt32 value between 0 and 1 to a half-precision FfxFloat32. +/// +/// @param [in] x The value to converted to a half-precision FfxFloat32. +/// +/// @returns +/// The converted half-precision FfxFloat32 value. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneUint16ToFloat16(FfxUInt16 x) +{ + return FFX_TO_FLOAT16(x * FFX_TO_UINT16(FFX_TO_FLOAT16(1.0))); +} + +/// Convert a half-precision FfxUInt32 value between 0 and 1 to a half-precision FfxFloat32. +/// +/// @param [in] x The value to converted to a half-precision FfxFloat32. +/// +/// @returns +/// The converted half-precision FfxFloat32 value. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneUint16x2ToFloat16x2(FfxUInt16x2 x) +{ + return FFX_TO_FLOAT16X2(x * FFX_TO_UINT16X2(FfxUInt16x2(FFX_TO_FLOAT16(1.0), FFX_TO_FLOAT16(1.0)))); +} + +/// Convert a half-precision FfxUInt32 value between 0 and 1 to a half-precision FfxFloat32. +/// +/// @param [in] x The value to converted to a half-precision FfxFloat32. +/// +/// @returns +/// The converted half-precision FfxFloat32 value. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneUint16x3ToFloat16x3(FfxUInt16x3 x) +{ + return FFX_TO_FLOAT16X3(x * FFX_TO_UINT16X3(FfxUInt16x3(FFX_TO_FLOAT16(1.0), FFX_TO_FLOAT16(1.0), FFX_TO_FLOAT16(1.0)))); +} + +/// Convert a half-precision FfxUInt32 value between 0 and 1 to a half-precision FfxFloat32. +/// +/// @param [in] x The value to converted to a half-precision FfxFloat32. +/// +/// @returns +/// The converted half-precision FfxFloat32 value. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneUint16x4ToFloat16x4(FfxUInt16x4 x) +{ + return FFX_TO_FLOAT16X4(x * FFX_TO_UINT16X4(FfxUInt16x4(FFX_TO_FLOAT16(1.0), FFX_TO_FLOAT16(1.0), FFX_TO_FLOAT16(1.0), FFX_TO_FLOAT16(1.0)))); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneAndHalf(FfxFloat16 x, FfxFloat16 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneAndHalf(FfxFloat16x2 x, FfxFloat16x2 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneAndHalf(FfxFloat16x3 x, FfxFloat16x3 y) +{ + return min(x, y); +} + +/// Conditional free logic AND operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// +/// @returns +/// Result of the AND operation. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneAndHalf(FfxFloat16x4 x, FfxFloat16x4 y) +{ + return min(x, y); +} + +/// Conditional free logic AND NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND NOT operator. +/// @param [in] y The second value to be fed into the AND NOT operator. +/// +/// @returns +/// Result of the AND NOT operation. +/// +/// @ingroup GPU +FfxFloat16 ffxSignedZeroOneAndOrHalf(FfxFloat16 x, FfxFloat16 y) +{ + return (-x) * y + FFX_BROADCAST_FLOAT16(1.0); +} + +/// Conditional free logic AND NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND NOT operator. +/// @param [in] y The second value to be fed into the AND NOT operator. +/// +/// @returns +/// Result of the AND NOT operation. +/// +/// @ingroup GPU +FfxFloat16x2 ffxSignedZeroOneAndOrHalf(FfxFloat16x2 x, FfxFloat16x2 y) +{ + return (-x) * y + FFX_BROADCAST_FLOAT16X2(1.0); +} + +/// Conditional free logic AND NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND NOT operator. +/// @param [in] y The second value to be fed into the AND NOT operator. +/// +/// @returns +/// Result of the AND NOT operation. +/// +/// @ingroup GPU +FfxFloat16x3 ffxSignedZeroOneAndOrHalf(FfxFloat16x3 x, FfxFloat16x3 y) +{ + return (-x) * y + FFX_BROADCAST_FLOAT16X3(1.0); +} + +/// Conditional free logic AND NOT operation using two half-precision values. +/// +/// @param [in] x The first value to be fed into the AND NOT operator. +/// @param [in] y The second value to be fed into the AND NOT operator. +/// +/// @returns +/// Result of the AND NOT operation. +/// +/// @ingroup GPU +FfxFloat16x4 ffxSignedZeroOneAndOrHalf(FfxFloat16x4 x, FfxFloat16x4 y) +{ + return (-x) * y + FFX_BROADCAST_FLOAT16X4(1.0); +} + +/// Conditional free logic AND operation using two half-precision values followed by +/// a NOT operation using the resulting value and a third half-precision value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneAndOrHalf(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + return ffxSaturate(x * y + z); +} + +/// Conditional free logic AND operation using two half-precision values followed by +/// a NOT operation using the resulting value and a third half-precision value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneAndOrHalf(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + return ffxSaturate(x * y + z); +} + +/// Conditional free logic AND operation using two half-precision values followed by +/// a NOT operation using the resulting value and a third half-precision value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneAndOrHalf(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + return ffxSaturate(x * y + z); +} + +/// Conditional free logic AND operation using two half-precision values followed by +/// a NOT operation using the resulting value and a third half-precision value. +/// +/// @param [in] x The first value to be fed into the AND operator. +/// @param [in] y The second value to be fed into the AND operator. +/// @param [in] z The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneAndOrHalf(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + return ffxSaturate(x * y + z); +} + +/// Given a half-precision value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneIsGreaterThanZeroHalf(FfxFloat16 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16(FFX_POSITIVE_INFINITY_HALF)); +} + +/// Given a half-precision value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneIsGreaterThanZeroHalf(FfxFloat16x2 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16X2(FFX_POSITIVE_INFINITY_HALF)); +} + +/// Given a half-precision value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneIsGreaterThanZeroHalf(FfxFloat16x3 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16X3(FFX_POSITIVE_INFINITY_HALF)); +} + +/// Given a half-precision value, returns 1.0 if greater than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the greater than zero comparison. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneIsGreaterThanZeroHalf(FfxFloat16x4 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16X4(FFX_POSITIVE_INFINITY_HALF)); +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneNotHalf(FfxFloat16 x) +{ + return FFX_BROADCAST_FLOAT16(1.0) - x; +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneNotHalf(FfxFloat16x2 x) +{ + return FFX_BROADCAST_FLOAT16X2(1.0) - x; +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneNotHalf(FfxFloat16x3 x) +{ + return FFX_BROADCAST_FLOAT16X3(1.0) - x; +} + +/// Conditional free logic signed NOT operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the AND OR operator. +/// +/// @returns +/// Result of the AND OR operation. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneNotHalf(FfxFloat16x4 x) +{ + return FFX_BROADCAST_FLOAT16X4(1.0) - x; +} + +/// Conditional free logic OR operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneOrHalf(FfxFloat16 x, FfxFloat16 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneOrHalf(FfxFloat16x2 x, FfxFloat16x2 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneOrHalf(FfxFloat16x3 x, FfxFloat16x3 y) +{ + return max(x, y); +} + +/// Conditional free logic OR operation using two half-precision FfxFloat32 values. +/// +/// @param [in] x The first value to be fed into the OR operator. +/// @param [in] y The second value to be fed into the OR operator. +/// +/// @returns +/// Result of the OR operation. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneOrHalf(FfxFloat16x4 x, FfxFloat16x4 y) +{ + return max(x, y); +} + +/// Choose between two half-precision FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneSelectHalf(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + FfxFloat16 r = (-x) * z + z; + return x * y + r; +} + +/// Choose between two half-precision FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneSelectHalf(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + FfxFloat16x2 r = (-x) * z + z; + return x * y + r; +} + +/// Choose between two half-precision FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneSelectHalf(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + FfxFloat16x3 r = (-x) * z + z; + return x * y + r; +} + +/// Choose between two half-precision FfxFloat32 values if the first paramter is greater than zero. +/// +/// @param [in] x The value to compare against zero. +/// @param [in] y The value to return if the comparision is greater than zero. +/// @param [in] z The value to return if the comparision is less than or equal to zero. +/// +/// @returns +/// The selected value. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneSelectHalf(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + FfxFloat16x4 r = (-x) * z + z; + return x * y + r; +} + +/// Given a half-precision value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat16 ffxZeroOneIsSignedHalf(FfxFloat16 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// Given a half-precision value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat16x2 ffxZeroOneIsSignedHalf(FfxFloat16x2 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16X2(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// Given a half-precision value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat16x3 ffxZeroOneIsSignedHalf(FfxFloat16x3 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16X3(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// Given a half-precision value, returns 1.0 if less than zero and 0.0 if not. +/// +/// @param [in] x The value to be compared. +/// +/// @returns +/// Result of the sign value. +/// +/// @ingroup GPU +FfxFloat16x4 ffxZeroOneIsSignedHalf(FfxFloat16x4 x) +{ + return ffxSaturate(x * FFX_BROADCAST_FLOAT16X4(FFX_NEGATIVE_INFINITY_HALF)); +} + +/// Compute a Rec.709 color space. +/// +/// Rec.709 is used for some HDTVs. +/// +/// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +/// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +/// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +/// +/// @param [in] c The color to convert to Rec. 709. +/// +/// @returns +/// The color in Rec.709 space. +/// +/// @ingroup GPU +FfxFloat16 ffxRec709FromLinearHalf(FfxFloat16 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.018 * 4.5, 4.5, 0.45); + FfxFloat16x2 k = FfxFloat16x2(1.099, -0.099); + return clamp(j.x, c * j.y, pow(c, j.z) * k.x + k.y); +} + +/// Compute a Rec.709 color space. +/// +/// Rec.709 is used for some HDTVs. +/// +/// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +/// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +/// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +/// +/// @param [in] c The color to convert to Rec. 709. +/// +/// @returns +/// The color in Rec.709 space. +/// +/// @ingroup GPU +FfxFloat16x2 ffxRec709FromLinearHalf(FfxFloat16x2 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.018 * 4.5, 4.5, 0.45); + FfxFloat16x2 k = FfxFloat16x2(1.099, -0.099); + return clamp(j.xx, c * j.yy, pow(c, j.zz) * k.xx + k.yy); +} + +/// Compute a Rec.709 color space. +/// +/// Rec.709 is used for some HDTVs. +/// +/// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times. +/// (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range). +/// (b.) For 8-bit 709, steps {0 to 20.7} are in the linear region (8% of the encoding range). +/// +/// @param [in] c The color to convert to Rec. 709. +/// +/// @returns +/// The color in Rec.709 space. +/// +/// @ingroup GPU +FfxFloat16x3 ffxRec709FromLinearHalf(FfxFloat16x3 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.018 * 4.5, 4.5, 0.45); + FfxFloat16x2 k = FfxFloat16x2(1.099, -0.099); + return clamp(j.xxx, c * j.yyy, pow(c, j.zzz) * k.xxx + k.yyy); +} + +/// Compute a gamma value from a linear value. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// Note: 'rcpX' is '1/x', where the 'x' is what would be used in ffxLinearFromGammaHalf. +/// +/// @param [in] c The value to convert to gamma space from linear. +/// @param [in] rcpX The reciprocal of power value used for the gamma curve. +/// +/// @returns +/// A value in gamma space. +/// +/// @ingroup GPU +FfxFloat16 ffxGammaFromLinearHalf(FfxFloat16 c, FfxFloat16 rcpX) +{ + return pow(c, FFX_BROADCAST_FLOAT16(rcpX)); +} + +/// Compute a gamma value from a linear value. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// Note: 'rcpX' is '1/x', where the 'x' is what would be used in ffxLinearFromGammaHalf. +/// +/// @param [in] c The value to convert to gamma space from linear. +/// @param [in] rcpX The reciprocal of power value used for the gamma curve. +/// +/// @returns +/// A value in gamma space. +/// +/// @ingroup GPU +FfxFloat16x2 ffxGammaFromLinearHalf(FfxFloat16x2 c, FfxFloat16 rcpX) +{ + return pow(c, FFX_BROADCAST_FLOAT16X2(rcpX)); +} + +/// Compute a gamma value from a linear value. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// Note: 'rcpX' is '1/x', where the 'x' is what would be used in ffxLinearFromGammaHalf. +/// +/// @param [in] c The value to convert to gamma space from linear. +/// @param [in] rcpX The reciprocal of power value used for the gamma curve. +/// +/// @returns +/// A value in gamma space. +/// +/// @ingroup GPU +FfxFloat16x3 ffxGammaFromLinearHalf(FfxFloat16x3 c, FfxFloat16 rcpX) +{ + return pow(c, FFX_BROADCAST_FLOAT16X3(rcpX)); +} + +/// Compute an SRGB value from a linear value. +/// +/// @param [in] c The value to convert to SRGB from linear. +/// +/// @returns +/// A value in SRGB space. +/// +/// @ingroup GPU +FfxFloat16 ffxSrgbFromLinearHalf(FfxFloat16 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.0031308 * 12.92, 12.92, 1.0 / 2.4); + FfxFloat16x2 k = FfxFloat16x2(1.055, -0.055); + return clamp(j.x, c * j.y, pow(c, j.z) * k.x + k.y); +} + +/// Compute an SRGB value from a linear value. +/// +/// @param [in] c The value to convert to SRGB from linear. +/// +/// @returns +/// A value in SRGB space. +/// +/// @ingroup GPU +FfxFloat16x2 ffxSrgbFromLinearHalf(FfxFloat16x2 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.0031308 * 12.92, 12.92, 1.0 / 2.4); + FfxFloat16x2 k = FfxFloat16x2(1.055, -0.055); + return clamp(j.xx, c * j.yy, pow(c, j.zz) * k.xx + k.yy); +} + +/// Compute an SRGB value from a linear value. +/// +/// @param [in] c The value to convert to SRGB from linear. +/// +/// @returns +/// A value in SRGB space. +/// +/// @ingroup GPU +FfxFloat16x3 ffxSrgbFromLinearHalf(FfxFloat16x3 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.0031308 * 12.92, 12.92, 1.0 / 2.4); + FfxFloat16x2 k = FfxFloat16x2(1.055, -0.055); + return clamp(j.xxx, c * j.yyy, pow(c, j.zzz) * k.xxx + k.yyy); +} + +/// Compute the square root of a value. +/// +/// @param [in] c The value to compute the square root for. +/// +/// @returns +/// A square root of the input value. +/// +/// @ingroup GPU +FfxFloat16 ffxSquareRootHalf(FfxFloat16 c) +{ + return sqrt(c); +} + +/// Compute the square root of a value. +/// +/// @param [in] c The value to compute the square root for. +/// +/// @returns +/// A square root of the input value. +/// +/// @ingroup GPU +FfxFloat16x2 ffxSquareRootHalf(FfxFloat16x2 c) +{ + return sqrt(c); +} + +/// Compute the square root of a value. +/// +/// @param [in] c The value to compute the square root for. +/// +/// @returns +/// A square root of the input value. +/// +/// @ingroup GPU +FfxFloat16x3 ffxSquareRootHalf(FfxFloat16x3 c) +{ + return sqrt(c); +} + +/// Compute the cube root of a value. +/// +/// @param [in] c The value to compute the cube root for. +/// +/// @returns +/// A cube root of the input value. +/// +/// @ingroup GPU +FfxFloat16 ffxCubeRootHalf(FfxFloat16 c) +{ + return pow(c, FFX_BROADCAST_FLOAT16(1.0 / 3.0)); +} + +/// Compute the cube root of a value. +/// +/// @param [in] c The value to compute the cube root for. +/// +/// @returns +/// A cube root of the input value. +/// +/// @ingroup GPU +FfxFloat16x2 ffxCubeRootHalf(FfxFloat16x2 c) +{ + return pow(c, FFX_BROADCAST_FLOAT16X2(1.0 / 3.0)); +} + +/// Compute the cube root of a value. +/// +/// @param [in] c The value to compute the cube root for. +/// +/// @returns +/// A cube root of the input value. +/// +/// @ingroup GPU +FfxFloat16x3 ffxCubeRootHalf(FfxFloat16x3 c) +{ + return pow(c, FFX_BROADCAST_FLOAT16X3(1.0 / 3.0)); +} + +/// Compute a linear value from a REC.709 value. +/// +/// @param [in] c The value to convert to linear from REC.709. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16 ffxLinearFromRec709Half(FfxFloat16 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.081 / 4.5, 1.0 / 4.5, 1.0 / 0.45); + FfxFloat16x2 k = FfxFloat16x2(1.0 / 1.099, 0.099 / 1.099); + return ffxZeroOneSelectHalf(ffxZeroOneIsSignedHalf(c - j.x), c * j.y, pow(c * k.x + k.y, j.z)); +} + +/// Compute a linear value from a REC.709 value. +/// +/// @param [in] c The value to convert to linear from REC.709. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16x2 ffxLinearFromRec709Half(FfxFloat16x2 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.081 / 4.5, 1.0 / 4.5, 1.0 / 0.45); + FfxFloat16x2 k = FfxFloat16x2(1.0 / 1.099, 0.099 / 1.099); + return ffxZeroOneSelectHalf(ffxZeroOneIsSignedHalf(c - j.xx), c * j.yy, pow(c * k.xx + k.yy, j.zz)); +} + +/// Compute a linear value from a REC.709 value. +/// +/// @param [in] c The value to convert to linear from REC.709. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16x3 ffxLinearFromRec709Half(FfxFloat16x3 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.081 / 4.5, 1.0 / 4.5, 1.0 / 0.45); + FfxFloat16x2 k = FfxFloat16x2(1.0 / 1.099, 0.099 / 1.099); + return ffxZeroOneSelectHalf(ffxZeroOneIsSignedHalf(c - j.xxx), c * j.yyy, pow(c * k.xxx + k.yyy, j.zzz)); +} + +/// Compute a linear value from a value in a gamma space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] c The value to convert to linear in gamma space. +/// @param [in] x The power value used for the gamma curve. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16 ffxLinearFromGammaHalf(FfxFloat16 c, FfxFloat16 x) +{ + return pow(c, FFX_BROADCAST_FLOAT16(x)); +} + +/// Compute a linear value from a value in a gamma space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] c The value to convert to linear in gamma space. +/// @param [in] x The power value used for the gamma curve. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16x2 ffxLinearFromGammaHalf(FfxFloat16x2 c, FfxFloat16 x) +{ + return pow(c, FFX_BROADCAST_FLOAT16X2(x)); +} + +/// Compute a linear value from a value in a gamma space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] c The value to convert to linear in gamma space. +/// @param [in] x The power value used for the gamma curve. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16x3 ffxLinearFromGammaHalf(FfxFloat16x3 c, FfxFloat16 x) +{ + return pow(c, FFX_BROADCAST_FLOAT16X3(x)); +} + +/// Compute a linear value from a value in a SRGB space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] c The value to convert to linear in SRGB space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16 ffxLinearFromSrgbHalf(FfxFloat16 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.04045 / 12.92, 1.0 / 12.92, 2.4); + FfxFloat16x2 k = FfxFloat16x2(1.0 / 1.055, 0.055 / 1.055); + return ffxZeroOneSelectHalf(ffxZeroOneIsSignedHalf(c - j.x), c * j.y, pow(c * k.x + k.y, j.z)); +} + +/// Compute a linear value from a value in a SRGB space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] c The value to convert to linear in SRGB space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16x2 ffxLinearFromSrgbHalf(FfxFloat16x2 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.04045 / 12.92, 1.0 / 12.92, 2.4); + FfxFloat16x2 k = FfxFloat16x2(1.0 / 1.055, 0.055 / 1.055); + return ffxZeroOneSelectHalf(ffxZeroOneIsSignedHalf(c - j.xx), c * j.yy, pow(c * k.xx + k.yy, j.zz)); +} + +/// Compute a linear value from a value in a SRGB space. +/// +/// Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native. +/// +/// @param [in] c The value to convert to linear in SRGB space. +/// +/// @returns +/// A value in linear space. +/// +/// @ingroup GPU +FfxFloat16x3 ffxLinearFromSrgbHalf(FfxFloat16x3 c) +{ + FfxFloat16x3 j = FfxFloat16x3(0.04045 / 12.92, 1.0 / 12.92, 2.4); + FfxFloat16x2 k = FfxFloat16x2(1.0 / 1.055, 0.055 / 1.055); + return ffxZeroOneSelectHalf(ffxZeroOneIsSignedHalf(c - j.xxx), c * j.yyy, pow(c * k.xxx + k.yyy, j.zzz)); +} + +/// A remapping of 64x1 to 8x8 imposing rotated 2x2 pixel quads in quad linear. +/// +/// 543210 +/// ====== +/// ..xxx. +/// yy...y +/// +/// @param [in] a The input 1D coordinates to remap. +/// +/// @returns +/// The remapped 2D coordinates. +/// +/// @ingroup GPU +FfxUInt16x2 ffxRemapForQuadHalf(FfxUInt32 a) +{ + return FfxUInt16x2(bitfieldExtract(a, 1u, 3u), bitfieldInsertMask(bitfieldExtract(a, 3u, 3u), a, 1u)); +} + +/// A helper function performing a remap 64x1 to 8x8 remapping which is necessary for 2D wave reductions. +/// +/// The 64-wide lane indices to 8x8 remapping is performed as follows: +/// +/// 00 01 08 09 10 11 18 19 +/// 02 03 0a 0b 12 13 1a 1b +/// 04 05 0c 0d 14 15 1c 1d +/// 06 07 0e 0f 16 17 1e 1f +/// 20 21 28 29 30 31 38 39 +/// 22 23 2a 2b 32 33 3a 3b +/// 24 25 2c 2d 34 35 3c 3d +/// 26 27 2e 2f 36 37 3e 3f +/// +/// @param [in] a The input 1D coordinate to remap. +/// +/// @returns +/// The remapped 2D coordinates. +/// +/// @ingroup GPU +FfxUInt16x2 ffxRemapForWaveReductionHalf(FfxUInt32 a) +{ + return FfxUInt16x2(bitfieldInsertMask(bitfieldExtract(a, 2u, 3u), a, 1u), bitfieldInsertMask(bitfieldExtract(a, 3u, 3u), bitfieldExtract(a, 1u, 2u), 2u)); +} + +#endif // FFX_HALF diff --git a/thirdparty/amd-fsr2/shaders/ffx_core_hlsl.h b/thirdparty/amd-fsr2/shaders/ffx_core_hlsl.h new file mode 100644 index 000000000000..ad4ff6552d16 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core_hlsl.h @@ -0,0 +1,1502 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// A define for abstracting shared memory between shading languages. +/// +/// @ingroup GPU +#define FFX_GROUPSHARED groupshared + +/// A define for abstracting compute memory barriers between shading languages. +/// +/// @ingroup GPU +#define FFX_GROUP_MEMORY_BARRIER GroupMemoryBarrierWithGroupSync + +/// A define added to accept static markup on functions to aid CPU/GPU portability of code. +/// +/// @ingroup GPU +#define FFX_STATIC static + +/// A define for abstracting loop unrolling between shading languages. +/// +/// @ingroup GPU +#define FFX_UNROLL [unroll] + +/// A define for abstracting a 'greater than' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_GREATER_THAN(x, y) x > y + +/// A define for abstracting a 'greater than or equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_GREATER_THAN_EQUAL(x, y) x >= y + +/// A define for abstracting a 'less than' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_LESS_THAN(x, y) x < y + +/// A define for abstracting a 'less than or equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_LESS_THAN_EQUAL(x, y) x <= y + +/// A define for abstracting an 'equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_EQUAL(x, y) x == y + +/// A define for abstracting a 'not equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_NOT_EQUAL(x, y) x != y + +/// Broadcast a scalar value to a 1-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32(x) FfxFloat32(x) + +/// Broadcast a scalar value to a 2-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X2(x) FfxFloat32(x) + +/// Broadcast a scalar value to a 3-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X3(x) FfxFloat32(x) + +/// Broadcast a scalar value to a 4-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X4(x) FfxFloat32(x) + +/// Broadcast a scalar value to a 1-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32(x) FfxUInt32(x) + +/// Broadcast a scalar value to a 2-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X2(x) FfxUInt32(x) + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X3(x) FfxUInt32(x) + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X4(x) FfxUInt32(x) + +/// Broadcast a scalar value to a 1-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32(x) FfxInt32(x) + +/// Broadcast a scalar value to a 2-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X2(x) FfxInt32(x) + +/// Broadcast a scalar value to a 3-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X3(x) FfxInt32(x) + +/// Broadcast a scalar value to a 4-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X4(x) FfxInt32(x) + +/// Broadcast a scalar value to a 1-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16(a) FFX_MIN16_F(a) + +/// Broadcast a scalar value to a 2-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X2(a) FFX_MIN16_F(a) + +/// Broadcast a scalar value to a 3-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X3(a) FFX_MIN16_F(a) + +/// Broadcast a scalar value to a 4-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X4(a) FFX_MIN16_F(a) + +/// Broadcast a scalar value to a 1-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16(a) FFX_MIN16_U(a) + +/// Broadcast a scalar value to a 2-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X2(a) FFX_MIN16_U(a) + +/// Broadcast a scalar value to a 3-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X3(a) FFX_MIN16_U(a) + +/// Broadcast a scalar value to a 4-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X4(a) FFX_MIN16_U(a) + +/// Broadcast a scalar value to a 1-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16(a) FFX_MIN16_I(a) + +/// Broadcast a scalar value to a 2-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X2(a) FFX_MIN16_I(a) + +/// Broadcast a scalar value to a 3-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X3(a) FFX_MIN16_I(a) + +/// Broadcast a scalar value to a 4-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X4(a) FFX_MIN16_I(a) + +/// Pack 2x32-bit floating point values in a single 32bit value. +/// +/// This function first converts each component of value into their nearest 16-bit floating +/// point representation, and then stores the X and Y components in the lower and upper 16 bits of the +/// 32bit unsigned integer respectively. +/// +/// @param [in] value A 2-dimensional floating point value to convert and pack. +/// +/// @returns +/// A packed 32bit value containing 2 16bit floating point values. +/// +/// @ingroup HLSL +FfxUInt32 packHalf2x16(FfxFloat32x2 value) +{ + return f32tof16(value.x) | (f32tof16(value.y) << 16); +} + +/// Broadcast a scalar value to a 2-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional floating point vector with value in each component. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxBroadcast2(FfxFloat32 value) +{ + return FfxFloat32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional floating point vector with value in each component. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxBroadcast3(FfxFloat32 value) +{ + return FfxFloat32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional floating point vector with value in each component. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxBroadcast4(FfxFloat32 value) +{ + return FfxFloat32x4(value, value, value, value); +} + +/// Broadcast a scalar value to a 2-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional signed integer vector with value in each component. +/// +/// @ingroup HLSL +FfxInt32x2 ffxBroadcast2(FfxInt32 value) +{ + return FfxInt32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional signed integer vector with value in each component. +/// +/// @ingroup HLSL +FfxUInt32x3 ffxBroadcast3(FfxInt32 value) +{ + return FfxUInt32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional signed integer vector with value in each component. +/// +/// @ingroup HLSL +FfxInt32x4 ffxBroadcast4(FfxInt32 value) +{ + return FfxInt32x4(value, value, value, value); +} + +/// Broadcast a scalar value to a 2-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup HLSL +FfxUInt32x2 ffxBroadcast2(FfxUInt32 value) +{ + return FfxUInt32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup HLSL +FfxUInt32x3 ffxBroadcast3(FfxUInt32 value) +{ + return FfxUInt32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup HLSL +FfxUInt32x4 ffxBroadcast4(FfxUInt32 value) +{ + return FfxUInt32x4(value, value, value, value); +} + +FfxUInt32 bitfieldExtract(FfxUInt32 src, FfxUInt32 off, FfxUInt32 bits) +{ + FfxUInt32 mask = (1u << bits) - 1; + return (src >> off) & mask; +} + +FfxUInt32 bitfieldInsert(FfxUInt32 src, FfxUInt32 ins, FfxUInt32 mask) +{ + return (ins & mask) | (src & (~mask)); +} + +FfxUInt32 bitfieldInsertMask(FfxUInt32 src, FfxUInt32 ins, FfxUInt32 bits) +{ + FfxUInt32 mask = (1u << bits) - 1; + return (ins & mask) | (src & (~mask)); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup HLSL +FfxUInt32 ffxAsUInt32(FfxFloat32 x) +{ + return asuint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup HLSL +FfxUInt32x2 ffxAsUInt32(FfxFloat32x2 x) +{ + return asuint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup HLSL +FfxUInt32x3 ffxAsUInt32(FfxFloat32x3 x) +{ + return asuint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup HLSL +FfxUInt32x4 ffxAsUInt32(FfxFloat32x4 x) +{ + return asuint(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup HLSL +FfxFloat32 ffxAsFloat(FfxUInt32 x) +{ + return asfloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxAsFloat(FfxUInt32x2 x) +{ + return asfloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxAsFloat(FfxUInt32x3 x) +{ + return asfloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxAsFloat(FfxUInt32x4 x) +{ + return asfloat(x); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32 ffxLerp(FfxFloat32 x, FfxFloat32 y, FfxFloat32 t) +{ + return lerp(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxLerp(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32 t) +{ + return lerp(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxLerp(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 t) +{ + return lerp(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxLerp(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32 t) +{ + return lerp(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxLerp(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 t) +{ + return lerp(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxLerp(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32 t) +{ + return lerp(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the HLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxLerp(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 t) +{ + return lerp(x, y, t); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup HLSL +FfxFloat32 ffxSaturate(FfxFloat32 x) +{ + return saturate(x); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxSaturate(FfxFloat32x2 x) +{ + return saturate(x); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxSaturate(FfxFloat32x3 x) +{ + return saturate(x); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxSaturate(FfxFloat32x4 x) +{ + return saturate(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32 ffxFract(FfxFloat32 x) +{ + return x - floor(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxFract(FfxFloat32x2 x) +{ + return x - floor(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxFract(FfxFloat32x3 x) +{ + return x - floor(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxFract(FfxFloat32x4 x) +{ + return x - floor(x); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32 ffxMax3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxMax3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxMax3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxMax3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32 ffxMax3(FfxUInt32 x, FfxUInt32 y, FfxUInt32 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32x2 ffxMax3(FfxUInt32x2 x, FfxUInt32x2 y, FfxUInt32x2 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32x3 ffxMax3(FfxUInt32x3 x, FfxUInt32x3 y, FfxUInt32x3 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32x4 ffxMax3(FfxUInt32x4 x, FfxUInt32x4 y, FfxUInt32x4 z) +{ + return max(x, max(y, z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32 ffxMed3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxMed3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxMed3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxMed3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxInt32 ffxMed3(FfxInt32 x, FfxInt32 y, FfxInt32 z) +{ + return max(min(x, y), min(max(x, y), z)); + // return min(max(min(y, z), x), max(y, z)); + // return max(max(x, y), z) == x ? max(y, z) : (max(max(x, y), z) == y ? max(x, z) : max(x, y)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxInt32x2 ffxMed3(FfxInt32x2 x, FfxInt32x2 y, FfxInt32x2 z) +{ + return max(min(x, y), min(max(x, y), z)); + // return min(max(min(y, z), x), max(y, z)); + // return max(max(x, y), z) == x ? max(y, z) : (max(max(x, y), z) == y ? max(x, z) : max(x, y)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxInt32x3 ffxMed3(FfxInt32x3 x, FfxInt32x3 y, FfxInt32x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_I32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup HLSL +FfxInt32x4 ffxMed3(FfxInt32x4 x, FfxInt32x4 y, FfxInt32x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_I32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32 ffxMin3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_I32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxMin3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_I32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxMin3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxMin3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32 ffxMin3(FfxUInt32 x, FfxUInt32 y, FfxUInt32 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32x2 ffxMin3(FfxUInt32x2 x, FfxUInt32x2 y, FfxUInt32x2 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32x3 ffxMin3(FfxUInt32x3 x, FfxUInt32x3 y, FfxUInt32x3 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup HLSL +FfxUInt32x4 ffxMin3(FfxUInt32x4 x, FfxUInt32x4 y, FfxUInt32x4 z) +{ + return min(x, min(y, z)); +} + + +FfxUInt32 AShrSU1(FfxUInt32 a, FfxUInt32 b) +{ + return FfxUInt32(FfxInt32(a) >> FfxInt32(b)); +} + +//============================================================================================================================== +// HLSL HALF +//============================================================================================================================== +#if FFX_HALF + +//============================================================================================================================== +// Need to use manual unpack to get optimal execution (don't use packed types in buffers directly). +// Unpack requires this pattern: https://gpuopen.com/first-steps-implementing-fp16/ +FFX_MIN16_F2 ffxUint32ToFloat16x2(FfxUInt32 x) +{ + FfxFloat32x2 t = f16tof32(FfxUInt32x2(x & 0xFFFF, x >> 16)); + return FFX_MIN16_F2(t); +} +FFX_MIN16_F4 ffxUint32x2ToFloat16x4(FfxUInt32x2 x) +{ + return FFX_MIN16_F4(ffxUint32ToFloat16x2(x.x), ffxUint32ToFloat16x2(x.y)); +} +FFX_MIN16_U2 ffxUint32ToUint16x2(FfxUInt32 x) +{ + FfxUInt32x2 t = FfxUInt32x2(x & 0xFFFF, x >> 16); + return FFX_MIN16_U2(t); +} +FFX_MIN16_U4 ffxUint32x2ToUint16x4(FfxUInt32x2 x) +{ + return FFX_MIN16_U4(ffxUint32ToUint16x2(x.x), ffxUint32ToUint16x2(x.y)); +} +#define FFX_UINT32_TO_FLOAT16X2(x) ffxUint32ToFloat16x2(FfxUInt32(x)) +#define FFX_UINT32X2_TO_FLOAT16X4(x) ffxUint32x2ToFloat16x4(FfxUInt32x2(x)) +#define FFX_UINT32_TO_UINT16X2(x) ffxUint32ToUint16x2(FfxUInt32(x)) +#define FFX_UINT32X2_TO_UINT16X4(x) ffxUint32x2ToUint16x4(FfxUInt32x2(x)) +//------------------------------------------------------------------------------------------------------------------------------ +FfxUInt32 FFX_MIN16_F2ToUint32(FFX_MIN16_F2 x) +{ + return f32tof16(x.x) + (f32tof16(x.y) << 16); +} +FfxUInt32x2 FFX_MIN16_F4ToUint32x2(FFX_MIN16_F4 x) +{ + return FfxUInt32x2(FFX_MIN16_F2ToUint32(x.xy), FFX_MIN16_F2ToUint32(x.zw)); +} +FfxUInt32 FFX_MIN16_U2ToUint32(FFX_MIN16_U2 x) +{ + return FfxUInt32(x.x) + (FfxUInt32(x.y) << 16); +} +FfxUInt32x2 FFX_MIN16_U4ToUint32x2(FFX_MIN16_U4 x) +{ + return FfxUInt32x2(FFX_MIN16_U2ToUint32(x.xy), FFX_MIN16_U2ToUint32(x.zw)); +} +#define FFX_FLOAT16X2_TO_UINT32(x) FFX_MIN16_F2ToUint32(FFX_MIN16_F2(x)) +#define FFX_FLOAT16X4_TO_UINT32X2(x) FFX_MIN16_F4ToUint32x2(FFX_MIN16_F4(x)) +#define FFX_UINT16X2_TO_UINT32(x) FFX_MIN16_U2ToUint32(FFX_MIN16_U2(x)) +#define FFX_UINT16X4_TO_UINT32X2(x) FFX_MIN16_U4ToUint32x2(FFX_MIN16_U4(x)) + +#if defined(FFX_HLSL_6_2) && !defined(FFX_NO_16_BIT_CAST) +#define FFX_TO_UINT16(x) asuint16(x) +#define FFX_TO_UINT16X2(x) asuint16(x) +#define FFX_TO_UINT16X3(x) asuint16(x) +#define FFX_TO_UINT16X4(x) asuint16(x) +#else +#define FFX_TO_UINT16(a) FFX_MIN16_U(f32tof16(FfxFloat32(a))) +#define FFX_TO_UINT16X2(a) FFX_MIN16_U2(FFX_TO_UINT16((a).x), FFX_TO_UINT16((a).y)) +#define FFX_TO_UINT16X3(a) FFX_MIN16_U3(FFX_TO_UINT16((a).x), FFX_TO_UINT16((a).y), FFX_TO_UINT16((a).z)) +#define FFX_TO_UINT16X4(a) FFX_MIN16_U4(FFX_TO_UINT16((a).x), FFX_TO_UINT16((a).y), FFX_TO_UINT16((a).z), FFX_TO_UINT16((a).w)) +#endif // #if defined(FFX_HLSL_6_2) && !defined(FFX_NO_16_BIT_CAST) + +#if defined(FFX_HLSL_6_2) && !defined(FFX_NO_16_BIT_CAST) +#define FFX_TO_FLOAT16(x) asfloat16(x) +#define FFX_TO_FLOAT16X2(x) asfloat16(x) +#define FFX_TO_FLOAT16X3(x) asfloat16(x) +#define FFX_TO_FLOAT16X4(x) asfloat16(x) +#else +#define FFX_TO_FLOAT16(a) FFX_MIN16_F(f16tof32(FfxUInt32(a))) +#define FFX_TO_FLOAT16X2(a) FFX_MIN16_F2(FFX_TO_FLOAT16((a).x), FFX_TO_FLOAT16((a).y)) +#define FFX_TO_FLOAT16X3(a) FFX_MIN16_F3(FFX_TO_FLOAT16((a).x), FFX_TO_FLOAT16((a).y), FFX_TO_FLOAT16((a).z)) +#define FFX_TO_FLOAT16X4(a) FFX_MIN16_F4(FFX_TO_FLOAT16((a).x), FFX_TO_FLOAT16((a).y), FFX_TO_FLOAT16((a).z), FFX_TO_FLOAT16((a).w)) +#endif // #if defined(FFX_HLSL_6_2) && !defined(FFX_NO_16_BIT_CAST) + +//============================================================================================================================== +#define FFX_BROADCAST_FLOAT16(a) FFX_MIN16_F(a) +#define FFX_BROADCAST_FLOAT16X2(a) FFX_MIN16_F(a) +#define FFX_BROADCAST_FLOAT16X3(a) FFX_MIN16_F(a) +#define FFX_BROADCAST_FLOAT16X4(a) FFX_MIN16_F(a) + +//------------------------------------------------------------------------------------------------------------------------------ +#define FFX_BROADCAST_INT16(a) FFX_MIN16_I(a) +#define FFX_BROADCAST_INT16X2(a) FFX_MIN16_I(a) +#define FFX_BROADCAST_INT16X3(a) FFX_MIN16_I(a) +#define FFX_BROADCAST_INT16X4(a) FFX_MIN16_I(a) + +//------------------------------------------------------------------------------------------------------------------------------ +#define FFX_BROADCAST_UINT16(a) FFX_MIN16_U(a) +#define FFX_BROADCAST_UINT16X2(a) FFX_MIN16_U(a) +#define FFX_BROADCAST_UINT16X3(a) FFX_MIN16_U(a) +#define FFX_BROADCAST_UINT16X4(a) FFX_MIN16_U(a) + +//============================================================================================================================== +FFX_MIN16_U ffxAbsHalf(FFX_MIN16_U a) +{ + return FFX_MIN16_U(abs(FFX_MIN16_I(a))); +} +FFX_MIN16_U2 ffxAbsHalf(FFX_MIN16_U2 a) +{ + return FFX_MIN16_U2(abs(FFX_MIN16_I2(a))); +} +FFX_MIN16_U3 ffxAbsHalf(FFX_MIN16_U3 a) +{ + return FFX_MIN16_U3(abs(FFX_MIN16_I3(a))); +} +FFX_MIN16_U4 ffxAbsHalf(FFX_MIN16_U4 a) +{ + return FFX_MIN16_U4(abs(FFX_MIN16_I4(a))); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxClampHalf(FFX_MIN16_F x, FFX_MIN16_F n, FFX_MIN16_F m) +{ + return max(n, min(x, m)); +} +FFX_MIN16_F2 ffxClampHalf(FFX_MIN16_F2 x, FFX_MIN16_F2 n, FFX_MIN16_F2 m) +{ + return max(n, min(x, m)); +} +FFX_MIN16_F3 ffxClampHalf(FFX_MIN16_F3 x, FFX_MIN16_F3 n, FFX_MIN16_F3 m) +{ + return max(n, min(x, m)); +} +FFX_MIN16_F4 ffxClampHalf(FFX_MIN16_F4 x, FFX_MIN16_F4 n, FFX_MIN16_F4 m) +{ + return max(n, min(x, m)); +} +//------------------------------------------------------------------------------------------------------------------------------ +// V_FRACT_F16 (note DX frac() is different). +FFX_MIN16_F ffxFract(FFX_MIN16_F x) +{ + return x - floor(x); +} +FFX_MIN16_F2 ffxFract(FFX_MIN16_F2 x) +{ + return x - floor(x); +} +FFX_MIN16_F3 ffxFract(FFX_MIN16_F3 x) +{ + return x - floor(x); +} +FFX_MIN16_F4 ffxFract(FFX_MIN16_F4 x) +{ + return x - floor(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxLerp(FFX_MIN16_F x, FFX_MIN16_F y, FFX_MIN16_F a) +{ + return lerp(x, y, a); +} +FFX_MIN16_F2 ffxLerp(FFX_MIN16_F2 x, FFX_MIN16_F2 y, FFX_MIN16_F a) +{ + return lerp(x, y, a); +} +FFX_MIN16_F2 ffxLerp(FFX_MIN16_F2 x, FFX_MIN16_F2 y, FFX_MIN16_F2 a) +{ + return lerp(x, y, a); +} +FFX_MIN16_F3 ffxLerp(FFX_MIN16_F3 x, FFX_MIN16_F3 y, FFX_MIN16_F a) +{ + return lerp(x, y, a); +} +FFX_MIN16_F3 ffxLerp(FFX_MIN16_F3 x, FFX_MIN16_F3 y, FFX_MIN16_F3 a) +{ + return lerp(x, y, a); +} +FFX_MIN16_F4 ffxLerp(FFX_MIN16_F4 x, FFX_MIN16_F4 y, FFX_MIN16_F a) +{ + return lerp(x, y, a); +} +FFX_MIN16_F4 ffxLerp(FFX_MIN16_F4 x, FFX_MIN16_F4 y, FFX_MIN16_F4 a) +{ + return lerp(x, y, a); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxMax3Half(FFX_MIN16_F x, FFX_MIN16_F y, FFX_MIN16_F z) +{ + return max(x, max(y, z)); +} +FFX_MIN16_F2 ffxMax3Half(FFX_MIN16_F2 x, FFX_MIN16_F2 y, FFX_MIN16_F2 z) +{ + return max(x, max(y, z)); +} +FFX_MIN16_F3 ffxMax3Half(FFX_MIN16_F3 x, FFX_MIN16_F3 y, FFX_MIN16_F3 z) +{ + return max(x, max(y, z)); +} +FFX_MIN16_F4 ffxMax3Half(FFX_MIN16_F4 x, FFX_MIN16_F4 y, FFX_MIN16_F4 z) +{ + return max(x, max(y, z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxMin3Half(FFX_MIN16_F x, FFX_MIN16_F y, FFX_MIN16_F z) +{ + return min(x, min(y, z)); +} +FFX_MIN16_F2 ffxMin3Half(FFX_MIN16_F2 x, FFX_MIN16_F2 y, FFX_MIN16_F2 z) +{ + return min(x, min(y, z)); +} +FFX_MIN16_F3 ffxMin3Half(FFX_MIN16_F3 x, FFX_MIN16_F3 y, FFX_MIN16_F3 z) +{ + return min(x, min(y, z)); +} +FFX_MIN16_F4 ffxMin3Half(FFX_MIN16_F4 x, FFX_MIN16_F4 y, FFX_MIN16_F4 z) +{ + return min(x, min(y, z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxMed3Half(FFX_MIN16_F x, FFX_MIN16_F y, FFX_MIN16_F z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FFX_MIN16_F2 ffxMed3Half(FFX_MIN16_F2 x, FFX_MIN16_F2 y, FFX_MIN16_F2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FFX_MIN16_F3 ffxMed3Half(FFX_MIN16_F3 x, FFX_MIN16_F3 y, FFX_MIN16_F3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FFX_MIN16_F4 ffxMed3Half(FFX_MIN16_F4 x, FFX_MIN16_F4 y, FFX_MIN16_F4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_I ffxMed3Half(FFX_MIN16_I x, FFX_MIN16_I y, FFX_MIN16_I z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FFX_MIN16_I2 ffxMed3Half(FFX_MIN16_I2 x, FFX_MIN16_I2 y, FFX_MIN16_I2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FFX_MIN16_I3 ffxMed3Half(FFX_MIN16_I3 x, FFX_MIN16_I3 y, FFX_MIN16_I3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +FFX_MIN16_I4 ffxMed3Half(FFX_MIN16_I4 x, FFX_MIN16_I4 y, FFX_MIN16_I4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxReciprocalHalf(FFX_MIN16_F x) +{ + return rcp(x); +} +FFX_MIN16_F2 ffxReciprocalHalf(FFX_MIN16_F2 x) +{ + return rcp(x); +} +FFX_MIN16_F3 ffxReciprocalHalf(FFX_MIN16_F3 x) +{ + return rcp(x); +} +FFX_MIN16_F4 ffxReciprocalHalf(FFX_MIN16_F4 x) +{ + return rcp(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxReciprocalSquareRootHalf(FFX_MIN16_F x) +{ + return rsqrt(x); +} +FFX_MIN16_F2 ffxReciprocalSquareRootHalf(FFX_MIN16_F2 x) +{ + return rsqrt(x); +} +FFX_MIN16_F3 ffxReciprocalSquareRootHalf(FFX_MIN16_F3 x) +{ + return rsqrt(x); +} +FFX_MIN16_F4 ffxReciprocalSquareRootHalf(FFX_MIN16_F4 x) +{ + return rsqrt(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_F ffxSaturate(FFX_MIN16_F x) +{ + return saturate(x); +} +FFX_MIN16_F2 ffxSaturate(FFX_MIN16_F2 x) +{ + return saturate(x); +} +FFX_MIN16_F3 ffxSaturate(FFX_MIN16_F3 x) +{ + return saturate(x); +} +FFX_MIN16_F4 ffxSaturate(FFX_MIN16_F4 x) +{ + return saturate(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FFX_MIN16_U ffxBitShiftRightHalf(FFX_MIN16_U a, FFX_MIN16_U b) +{ + return FFX_MIN16_U(FFX_MIN16_I(a) >> FFX_MIN16_I(b)); +} +FFX_MIN16_U2 ffxBitShiftRightHalf(FFX_MIN16_U2 a, FFX_MIN16_U2 b) +{ + return FFX_MIN16_U2(FFX_MIN16_I2(a) >> FFX_MIN16_I2(b)); +} +FFX_MIN16_U3 ffxBitShiftRightHalf(FFX_MIN16_U3 a, FFX_MIN16_U3 b) +{ + return FFX_MIN16_U3(FFX_MIN16_I3(a) >> FFX_MIN16_I3(b)); +} +FFX_MIN16_U4 ffxBitShiftRightHalf(FFX_MIN16_U4 a, FFX_MIN16_U4 b) +{ + return FFX_MIN16_U4(FFX_MIN16_I4(a) >> FFX_MIN16_I4(b)); +} +#endif // FFX_HALF + +//============================================================================================================================== +// HLSL WAVE +//============================================================================================================================== +#if defined(FFX_WAVE) +// Where 'x' must be a compile time literal. +FfxFloat32 AWaveXorF1(FfxFloat32 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxFloat32x2 AWaveXorF2(FfxFloat32x2 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxFloat32x3 AWaveXorF3(FfxFloat32x3 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxFloat32x4 AWaveXorF4(FfxFloat32x4 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxUInt32 AWaveXorU1(FfxUInt32 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxUInt32x2 AWaveXorU1(FfxUInt32x2 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxUInt32x3 AWaveXorU1(FfxUInt32x3 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} +FfxUInt32x4 AWaveXorU1(FfxUInt32x4 v, FfxUInt32 x) +{ + return WaveReadLaneAt(v, WaveGetLaneIndex() ^ x); +} + +#if FFX_HALF +FfxFloat16x2 ffxWaveXorFloat16x2(FfxFloat16x2 v, FfxUInt32 x) +{ + return FFX_UINT32_TO_FLOAT16X2(WaveReadLaneAt(FFX_FLOAT16X2_TO_UINT32(v), WaveGetLaneIndex() ^ x)); +} +FfxFloat16x4 ffxWaveXorFloat16x4(FfxFloat16x4 v, FfxUInt32 x) +{ + return FFX_UINT32X2_TO_FLOAT16X4(WaveReadLaneAt(FFX_FLOAT16X4_TO_UINT32X2(v), WaveGetLaneIndex() ^ x)); +} +FfxUInt16x2 ffxWaveXorUint16x2(FfxUInt16x2 v, FfxUInt32 x) +{ + return FFX_UINT32_TO_UINT16X2(WaveReadLaneAt(FFX_UINT16X2_TO_UINT32(v), WaveGetLaneIndex() ^ x)); +} +FfxUInt16x4 ffxWaveXorUint16x4(FfxUInt16x4 v, FfxUInt32 x) +{ + return AW4_FFX_UINT32(WaveReadLaneAt(FFX_UINT32_AW4(v), WaveGetLaneIndex() ^ x)); +} +#endif // FFX_HALF +#endif // #if defined(FFX_WAVE) diff --git a/thirdparty/amd-fsr2/shaders/ffx_core_portability.h b/thirdparty/amd-fsr2/shaders/ffx_core_portability.h new file mode 100644 index 000000000000..45be05973a84 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_core_portability.h @@ -0,0 +1,50 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +FfxFloat32x3 opAAddOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b) +{ + d = a + ffxBroadcast3(b); + return d; +} + +FfxFloat32x3 opACpyF3(FfxFloat32x3 d, FfxFloat32x3 a) +{ + d = a; + return d; +} + +FfxFloat32x3 opAMulF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32x3 b) +{ + d = a * b; + return d; +} + +FfxFloat32x3 opAMulOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b) +{ + d = a * ffxBroadcast3(b); + return d; +} + +FfxFloat32x3 opARcpF3(FfxFloat32x3 d, FfxFloat32x3 a) +{ + d = rcp(a); + return d; +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr1.h b/thirdparty/amd-fsr2/shaders/ffx_fsr1.h new file mode 100644 index 000000000000..1ac23cf3de3d --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr1.h @@ -0,0 +1,1250 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifdef __clang__ +#pragma clang diagnostic ignored "-Wunused-variable" +#endif + +/// Setup required constant values for EASU (works on CPU or GPU). +/// +/// @param [out] con0 +/// @param [out] con1 +/// @param [out] con2 +/// @param [out] con3 +/// @param [in] inputViewportInPixelsX The rendered image resolution being upscaled in X dimension. +/// @param [in] inputViewportInPixelsY The rendered image resolution being upscaled in Y dimension. +/// @param [in] inputSizeInPixelsX The resolution of the resource containing the input image (useful for dynamic resolution) in X dimension. +/// @param [in] inputSizeInPixelsY The resolution of the resource containing the input image (useful for dynamic resolution) in Y dimension. +/// @param [in] outputSizeInPixelsX The display resolution which the input image gets upscaled to in X dimension. +/// @param [in] outputSizeInPixelsY The display resolution which the input image gets upscaled to in Y dimension. +/// +/// @ingroup FSR1 +FFX_STATIC void ffxFsrPopulateEasuConstants( + FFX_PARAMETER_INOUT FfxUInt32x4 con0, + FFX_PARAMETER_INOUT FfxUInt32x4 con1, + FFX_PARAMETER_INOUT FfxUInt32x4 con2, + FFX_PARAMETER_INOUT FfxUInt32x4 con3, + FFX_PARAMETER_IN FfxFloat32 inputViewportInPixelsX, + FFX_PARAMETER_IN FfxFloat32 inputViewportInPixelsY, + FFX_PARAMETER_IN FfxFloat32 inputSizeInPixelsX, + FFX_PARAMETER_IN FfxFloat32 inputSizeInPixelsY, + FFX_PARAMETER_IN FfxFloat32 outputSizeInPixelsX, + FFX_PARAMETER_IN FfxFloat32 outputSizeInPixelsY) +{ + // Output integer position to a pixel position in viewport. + con0[0] = ffxAsUInt32(inputViewportInPixelsX * ffxReciprocal(outputSizeInPixelsX)); + con0[1] = ffxAsUInt32(inputViewportInPixelsY * ffxReciprocal(outputSizeInPixelsY)); + con0[2] = ffxAsUInt32(FfxFloat32(0.5) * inputViewportInPixelsX * ffxReciprocal(outputSizeInPixelsX) - FfxFloat32(0.5)); + con0[3] = ffxAsUInt32(FfxFloat32(0.5) * inputViewportInPixelsY * ffxReciprocal(outputSizeInPixelsY) - FfxFloat32(0.5)); + + // Viewport pixel position to normalized image space. + // This is used to get upper-left of 'F' tap. + con1[0] = ffxAsUInt32(ffxReciprocal(inputSizeInPixelsX)); + con1[1] = ffxAsUInt32(ffxReciprocal(inputSizeInPixelsY)); + + // Centers of gather4, first offset from upper-left of 'F'. + // +---+---+ + // | | | + // +--(0)--+ + // | b | c | + // +---F---+---+---+ + // | e | f | g | h | + // +--(1)--+--(2)--+ + // | i | j | k | l | + // +---+---+---+---+ + // | n | o | + // +--(3)--+ + // | | | + // +---+---+ + con1[2] = ffxAsUInt32(FfxFloat32(1.0) * ffxReciprocal(inputSizeInPixelsX)); + con1[3] = ffxAsUInt32(FfxFloat32(-1.0) * ffxReciprocal(inputSizeInPixelsY)); + + // These are from (0) instead of 'F'. + con2[0] = ffxAsUInt32(FfxFloat32(-1.0) * ffxReciprocal(inputSizeInPixelsX)); + con2[1] = ffxAsUInt32(FfxFloat32(2.0) * ffxReciprocal(inputSizeInPixelsY)); + con2[2] = ffxAsUInt32(FfxFloat32(1.0) * ffxReciprocal(inputSizeInPixelsX)); + con2[3] = ffxAsUInt32(FfxFloat32(2.0) * ffxReciprocal(inputSizeInPixelsY)); + con3[0] = ffxAsUInt32(FfxFloat32(0.0) * ffxReciprocal(inputSizeInPixelsX)); + con3[1] = ffxAsUInt32(FfxFloat32(4.0) * ffxReciprocal(inputSizeInPixelsY)); + con3[2] = con3[3] = 0; +} + +/// Setup required constant values for EASU (works on CPU or GPU). +/// +/// @param [out] con0 +/// @param [out] con1 +/// @param [out] con2 +/// @param [out] con3 +/// @param [in] inputViewportInPixelsX The resolution of the input in the X dimension. +/// @param [in] inputViewportInPixelsY The resolution of the input in the Y dimension. +/// @param [in] inputSizeInPixelsX The input size in pixels in the X dimension. +/// @param [in] inputSizeInPixelsY The input size in pixels in the Y dimension. +/// @param [in] outputSizeInPixelsX The output size in pixels in the X dimension. +/// @param [in] outputSizeInPixelsY The output size in pixels in the Y dimension. +/// @param [in] inputOffsetInPixelsX The input image offset in the X dimension into the resource containing it (useful for dynamic resolution). +/// @param [in] inputOffsetInPixelsY The input image offset in the Y dimension into the resource containing it (useful for dynamic resolution). +/// +/// @ingroup FSR1 +FFX_STATIC void ffxFsrPopulateEasuConstantsOffset( + FFX_PARAMETER_INOUT FfxUInt32x4 con0, + FFX_PARAMETER_INOUT FfxUInt32x4 con1, + FFX_PARAMETER_INOUT FfxUInt32x4 con2, + FFX_PARAMETER_INOUT FfxUInt32x4 con3, + FFX_PARAMETER_IN FfxFloat32 inputViewportInPixelsX, + FFX_PARAMETER_IN FfxFloat32 inputViewportInPixelsY, + FFX_PARAMETER_IN FfxFloat32 inputSizeInPixelsX, + FFX_PARAMETER_IN FfxFloat32 inputSizeInPixelsY, + FFX_PARAMETER_IN FfxFloat32 outputSizeInPixelsX, + FFX_PARAMETER_IN FfxFloat32 outputSizeInPixelsY, + FFX_PARAMETER_IN FfxFloat32 inputOffsetInPixelsX, + FFX_PARAMETER_IN FfxFloat32 inputOffsetInPixelsY) +{ + ffxFsrPopulateEasuConstants( + con0, + con1, + con2, + con3, + inputViewportInPixelsX, + inputViewportInPixelsY, + inputSizeInPixelsX, + inputSizeInPixelsY, + outputSizeInPixelsX, + outputSizeInPixelsY); + + // override + con0[2] = ffxAsUInt32(FfxFloat32(0.5) * inputViewportInPixelsX * ffxReciprocal(outputSizeInPixelsX) - FfxFloat32(0.5) + inputOffsetInPixelsX); + con0[3] = ffxAsUInt32(FfxFloat32(0.5) * inputViewportInPixelsY * ffxReciprocal(outputSizeInPixelsY) - FfxFloat32(0.5) + inputOffsetInPixelsY); +} + +#if defined(FFX_GPU) && defined(FFX_FSR_EASU_FLOAT) +// Input callback prototypes, need to be implemented by calling shader +FfxFloat32x4 FsrEasuRF(FfxFloat32x2 p); +FfxFloat32x4 FsrEasuGF(FfxFloat32x2 p); +FfxFloat32x4 FsrEasuBF(FfxFloat32x2 p); + +// Filtering for a given tap for the scalar. +void fsrEasuTapFloat( + FFX_PARAMETER_INOUT FfxFloat32x3 accumulatedColor, // Accumulated color, with negative lobe. + FFX_PARAMETER_INOUT FfxFloat32 accumulatedWeight, // Accumulated weight. + FFX_PARAMETER_IN FfxFloat32x2 pixelOffset, // Pixel offset from resolve position to tap. + FFX_PARAMETER_IN FfxFloat32x2 gradientDirection, // Gradient direction. + FFX_PARAMETER_IN FfxFloat32x2 length, // Length. + FFX_PARAMETER_IN FfxFloat32 negativeLobeStrength, // Negative lobe strength. + FFX_PARAMETER_IN FfxFloat32 clippingPoint, // Clipping point. + FFX_PARAMETER_IN FfxFloat32x3 color) // Tap color. +{ + // Rotate offset by direction. + FfxFloat32x2 rotatedOffset; + rotatedOffset.x = (pixelOffset.x * (gradientDirection.x)) + (pixelOffset.y * gradientDirection.y); + rotatedOffset.y = (pixelOffset.x * (-gradientDirection.y)) + (pixelOffset.y * gradientDirection.x); + + // Anisotropy. + rotatedOffset *= length; + + // Compute distance^2. + FfxFloat32 distanceSquared = rotatedOffset.x * rotatedOffset.x + rotatedOffset.y * rotatedOffset.y; + + // Limit to the window as at corner, 2 taps can easily be outside. + distanceSquared = ffxMin(distanceSquared, clippingPoint); + + // Approximation of lancos2 without sin() or rcp(), or sqrt() to get x. + // (25/16 * (2/5 * x^2 - 1)^2 - (25/16 - 1)) * (1/4 * x^2 - 1)^2 + // |_______________________________________| |_______________| + // base window + // The general form of the 'base' is, + // (a*(b*x^2-1)^2-(a-1)) + // Where 'a=1/(2*b-b^2)' and 'b' moves around the negative lobe. + FfxFloat32 weightB = FfxFloat32(2.0 / 5.0) * distanceSquared + FfxFloat32(-1.0); + FfxFloat32 weightA = negativeLobeStrength * distanceSquared + FfxFloat32(-1.0); + weightB *= weightB; + weightA *= weightA; + weightB = FfxFloat32(25.0 / 16.0) * weightB + FfxFloat32(-(25.0 / 16.0 - 1.0)); + FfxFloat32 weight = weightB * weightA; + + // Do weighted average. + accumulatedColor += color * weight; + accumulatedWeight += weight; +} + +// Accumulate direction and length. +void fsrEasuSetFloat( + FFX_PARAMETER_INOUT FfxFloat32x2 direction, + FFX_PARAMETER_INOUT FfxFloat32 length, + FFX_PARAMETER_IN FfxFloat32x2 pp, + FFX_PARAMETER_IN FfxBoolean biS, + FFX_PARAMETER_IN FfxBoolean biT, + FFX_PARAMETER_IN FfxBoolean biU, + FFX_PARAMETER_IN FfxBoolean biV, + FFX_PARAMETER_IN FfxFloat32 lA, + FFX_PARAMETER_IN FfxFloat32 lB, + FFX_PARAMETER_IN FfxFloat32 lC, + FFX_PARAMETER_IN FfxFloat32 lD, + FFX_PARAMETER_IN FfxFloat32 lE) +{ + // Compute bilinear weight, branches factor out as predicates are compiler time immediates. + // s t + // u v + FfxFloat32 weight = FfxFloat32(0.0); + if (biS) + weight = (FfxFloat32(1.0) - pp.x) * (FfxFloat32(1.0) - pp.y); + if (biT) + weight = pp.x * (FfxFloat32(1.0) - pp.y); + if (biU) + weight = (FfxFloat32(1.0) - pp.x) * pp.y; + if (biV) + weight = pp.x * pp.y; + + // Direction is the '+' diff. + // a + // b c d + // e + // Then takes magnitude from abs average of both sides of 'c'. + // Length converts gradient reversal to 0, smoothly to non-reversal at 1, shaped, then adding horz and vert terms. + FfxFloat32 dc = lD - lC; + FfxFloat32 cb = lC - lB; + FfxFloat32 lengthX = max(abs(dc), abs(cb)); + lengthX = ffxApproximateReciprocal(lengthX); + FfxFloat32 directionX = lD - lB; + direction.x += directionX * weight; + lengthX = ffxSaturate(abs(directionX) * lengthX); + lengthX *= lengthX; + length += lengthX * weight; + + // Repeat for the y axis. + FfxFloat32 ec = lE - lC; + FfxFloat32 ca = lC - lA; + FfxFloat32 lengthY = max(abs(ec), abs(ca)); + lengthY = ffxApproximateReciprocal(lengthY); + FfxFloat32 directionY = lE - lA; + direction.y += directionY * weight; + lengthY = ffxSaturate(abs(directionY) * lengthY); + lengthY *= lengthY; + length += lengthY * weight; +} + +/// Apply edge-aware spatial upsampling using 32bit floating point precision calculations. +/// +/// @param [out] outPixel The computed color of a pixel. +/// @param [in] integerPosition Integer pixel position within the output. +/// @param [in] con0 The first constant value generated by ffxFsrPopulateEasuConstants. +/// @param [in] con1 The second constant value generated by ffxFsrPopulateEasuConstants. +/// @param [in] con2 The third constant value generated by ffxFsrPopulateEasuConstants. +/// @param [in] con3 The fourth constant value generated by ffxFsrPopulateEasuConstants. +/// +/// @ingroup FSR +void ffxFsrEasuFloat( + FFX_PARAMETER_OUT FfxFloat32x3 pix, + FFX_PARAMETER_IN FfxUInt32x2 ip, + FFX_PARAMETER_IN FfxUInt32x4 con0, + FFX_PARAMETER_IN FfxUInt32x4 con1, + FFX_PARAMETER_IN FfxUInt32x4 con2, + FFX_PARAMETER_IN FfxUInt32x4 con3) +{ + // Get position of 'f'. + FfxFloat32x2 pp = FfxFloat32x2(ip) * ffxAsFloat(con0.xy) + ffxAsFloat(con0.zw); + FfxFloat32x2 fp = floor(pp); + pp -= fp; + + // 12-tap kernel. + // b c + // e f g h + // i j k l + // n o + // Gather 4 ordering. + // a b + // r g + // For packed FP16, need either {rg} or {ab} so using the following setup for gather in all versions, + // a b <- unused (z) + // r g + // a b a b + // r g r g + // a b + // r g <- unused (z) + // Allowing dead-code removal to remove the 'z's. + FfxFloat32x2 p0 = fp * ffxAsFloat(con1.xy) + ffxAsFloat(con1.zw); + + // These are from p0 to avoid pulling two constants on pre-Navi hardware. + FfxFloat32x2 p1 = p0 + ffxAsFloat(con2.xy); + FfxFloat32x2 p2 = p0 + ffxAsFloat(con2.zw); + FfxFloat32x2 p3 = p0 + ffxAsFloat(con3.xy); + FfxFloat32x4 bczzR = FsrEasuRF(p0); + FfxFloat32x4 bczzG = FsrEasuGF(p0); + FfxFloat32x4 bczzB = FsrEasuBF(p0); + FfxFloat32x4 ijfeR = FsrEasuRF(p1); + FfxFloat32x4 ijfeG = FsrEasuGF(p1); + FfxFloat32x4 ijfeB = FsrEasuBF(p1); + FfxFloat32x4 klhgR = FsrEasuRF(p2); + FfxFloat32x4 klhgG = FsrEasuGF(p2); + FfxFloat32x4 klhgB = FsrEasuBF(p2); + FfxFloat32x4 zzonR = FsrEasuRF(p3); + FfxFloat32x4 zzonG = FsrEasuGF(p3); + FfxFloat32x4 zzonB = FsrEasuBF(p3); + + // Simplest multi-channel approximate luma possible (luma times 2, in 2 FMA/MAD). + FfxFloat32x4 bczzL = bczzB * ffxBroadcast4(0.5) + (bczzR * ffxBroadcast4(0.5) + bczzG); + FfxFloat32x4 ijfeL = ijfeB * ffxBroadcast4(0.5) + (ijfeR * ffxBroadcast4(0.5) + ijfeG); + FfxFloat32x4 klhgL = klhgB * ffxBroadcast4(0.5) + (klhgR * ffxBroadcast4(0.5) + klhgG); + FfxFloat32x4 zzonL = zzonB * ffxBroadcast4(0.5) + (zzonR * ffxBroadcast4(0.5) + zzonG); + + // Rename. + FfxFloat32 bL = bczzL.x; + FfxFloat32 cL = bczzL.y; + FfxFloat32 iL = ijfeL.x; + FfxFloat32 jL = ijfeL.y; + FfxFloat32 fL = ijfeL.z; + FfxFloat32 eL = ijfeL.w; + FfxFloat32 kL = klhgL.x; + FfxFloat32 lL = klhgL.y; + FfxFloat32 hL = klhgL.z; + FfxFloat32 gL = klhgL.w; + FfxFloat32 oL = zzonL.z; + FfxFloat32 nL = zzonL.w; + + // Accumulate for bilinear interpolation. + FfxFloat32x2 dir = ffxBroadcast2(0.0); + FfxFloat32 len = FfxFloat32(0.0); + fsrEasuSetFloat(dir, len, pp, FFX_TRUE, FFX_FALSE, FFX_FALSE, FFX_FALSE, bL, eL, fL, gL, jL); + fsrEasuSetFloat(dir, len, pp, FFX_FALSE, FFX_TRUE, FFX_FALSE, FFX_FALSE, cL, fL, gL, hL, kL); + fsrEasuSetFloat(dir, len, pp, FFX_FALSE, FFX_FALSE, FFX_TRUE, FFX_FALSE, fL, iL, jL, kL, nL); + fsrEasuSetFloat(dir, len, pp, FFX_FALSE, FFX_FALSE, FFX_FALSE, FFX_TRUE, gL, jL, kL, lL, oL); + + // Normalize with approximation, and cleanup close to zero. + FfxFloat32x2 dir2 = dir * dir; + FfxFloat32 dirR = dir2.x + dir2.y; + FfxUInt32 zro = dirR < FfxFloat32(1.0 / 32768.0); + dirR = ffxApproximateReciprocalSquareRoot(dirR); + dirR = zro ? FfxFloat32(1.0) : dirR; + dir.x = zro ? FfxFloat32(1.0) : dir.x; + dir *= ffxBroadcast2(dirR); + + // Transform from {0 to 2} to {0 to 1} range, and shape with square. + len = len * FfxFloat32(0.5); + len *= len; + + // Stretch kernel {1.0 vert|horz, to sqrt(2.0) on diagonal}. + FfxFloat32 stretch = (dir.x * dir.x + dir.y * dir.y) * ffxApproximateReciprocal(max(abs(dir.x), abs(dir.y))); + + // Anisotropic length after rotation, + // x := 1.0 lerp to 'stretch' on edges + // y := 1.0 lerp to 2x on edges + FfxFloat32x2 len2 = FfxFloat32x2(FfxFloat32(1.0) + (stretch - FfxFloat32(1.0)) * len, FfxFloat32(1.0) + FfxFloat32(-0.5) * len); + + // Based on the amount of 'edge', + // the window shifts from +/-{sqrt(2.0) to slightly beyond 2.0}. + FfxFloat32 lob = FfxFloat32(0.5) + FfxFloat32((1.0 / 4.0 - 0.04) - 0.5) * len; + + // Set distance^2 clipping point to the end of the adjustable window. + FfxFloat32 clp = ffxApproximateReciprocal(lob); + + // Accumulation mixed with min/max of 4 nearest. + // b c + // e f g h + // i j k l + // n o + FfxFloat32x3 min4 = + ffxMin(ffxMin3(FfxFloat32x3(ijfeR.z, ijfeG.z, ijfeB.z), FfxFloat32x3(klhgR.w, klhgG.w, klhgB.w), FfxFloat32x3(ijfeR.y, ijfeG.y, ijfeB.y)), + FfxFloat32x3(klhgR.x, klhgG.x, klhgB.x)); + FfxFloat32x3 max4 = + max(ffxMax3(FfxFloat32x3(ijfeR.z, ijfeG.z, ijfeB.z), FfxFloat32x3(klhgR.w, klhgG.w, klhgB.w), FfxFloat32x3(ijfeR.y, ijfeG.y, ijfeB.y)), FfxFloat32x3(klhgR.x, klhgG.x, klhgB.x)); + + // Accumulation. + FfxFloat32x3 aC = ffxBroadcast3(0.0); + FfxFloat32 aW = FfxFloat32(0.0); + fsrEasuTapFloat(aC, aW, FfxFloat32x2(0.0, -1.0) - pp, dir, len2, lob, clp, FfxFloat32x3(bczzR.x, bczzG.x, bczzB.x)); // b + fsrEasuTapFloat(aC, aW, FfxFloat32x2(1.0, -1.0) - pp, dir, len2, lob, clp, FfxFloat32x3(bczzR.y, bczzG.y, bczzB.y)); // c + fsrEasuTapFloat(aC, aW, FfxFloat32x2(-1.0, 1.0) - pp, dir, len2, lob, clp, FfxFloat32x3(ijfeR.x, ijfeG.x, ijfeB.x)); // i + fsrEasuTapFloat(aC, aW, FfxFloat32x2(0.0, 1.0) - pp, dir, len2, lob, clp, FfxFloat32x3(ijfeR.y, ijfeG.y, ijfeB.y)); // j + fsrEasuTapFloat(aC, aW, FfxFloat32x2(0.0, 0.0) - pp, dir, len2, lob, clp, FfxFloat32x3(ijfeR.z, ijfeG.z, ijfeB.z)); // f + fsrEasuTapFloat(aC, aW, FfxFloat32x2(-1.0, 0.0) - pp, dir, len2, lob, clp, FfxFloat32x3(ijfeR.w, ijfeG.w, ijfeB.w)); // e + fsrEasuTapFloat(aC, aW, FfxFloat32x2(1.0, 1.0) - pp, dir, len2, lob, clp, FfxFloat32x3(klhgR.x, klhgG.x, klhgB.x)); // k + fsrEasuTapFloat(aC, aW, FfxFloat32x2(2.0, 1.0) - pp, dir, len2, lob, clp, FfxFloat32x3(klhgR.y, klhgG.y, klhgB.y)); // l + fsrEasuTapFloat(aC, aW, FfxFloat32x2(2.0, 0.0) - pp, dir, len2, lob, clp, FfxFloat32x3(klhgR.z, klhgG.z, klhgB.z)); // h + fsrEasuTapFloat(aC, aW, FfxFloat32x2(1.0, 0.0) - pp, dir, len2, lob, clp, FfxFloat32x3(klhgR.w, klhgG.w, klhgB.w)); // g + fsrEasuTapFloat(aC, aW, FfxFloat32x2(1.0, 2.0) - pp, dir, len2, lob, clp, FfxFloat32x3(zzonR.z, zzonG.z, zzonB.z)); // o + fsrEasuTapFloat(aC, aW, FfxFloat32x2(0.0, 2.0) - pp, dir, len2, lob, clp, FfxFloat32x3(zzonR.w, zzonG.w, zzonB.w)); // n + + // Normalize and dering. + pix = ffxMin(max4, max(min4, aC * ffxBroadcast3(rcp(aW)))); +} +#endif // #if defined(FFX_GPU) && defined(FFX_FSR_EASU_FLOAT) + +#if defined(FFX_GPU) && FFX_HALF == 1 && defined(FFX_FSR_EASU_HALF) +// Input callback prototypes, need to be implemented by calling shader +FfxFloat16x4 FsrEasuRH(FfxFloat32x2 p); +FfxFloat16x4 FsrEasuGH(FfxFloat32x2 p); +FfxFloat16x4 FsrEasuBH(FfxFloat32x2 p); + +// This runs 2 taps in parallel. +void FsrEasuTapH( + FFX_PARAMETER_INOUT FfxFloat16x2 aCR, + FFX_PARAMETER_INOUT FfxFloat16x2 aCG, + FFX_PARAMETER_INOUT FfxFloat16x2 aCB, + FFX_PARAMETER_INOUT FfxFloat16x2 aW, + FFX_PARAMETER_IN FfxFloat16x2 offX, + FFX_PARAMETER_IN FfxFloat16x2 offY, + FFX_PARAMETER_IN FfxFloat16x2 dir, + FFX_PARAMETER_IN FfxFloat16x2 len, + FFX_PARAMETER_IN FfxFloat16 lob, + FFX_PARAMETER_IN FfxFloat16 clp, + FFX_PARAMETER_IN FfxFloat16x2 cR, + FFX_PARAMETER_IN FfxFloat16x2 cG, + FFX_PARAMETER_IN FfxFloat16x2 cB) +{ + FfxFloat16x2 vX, vY; + vX = offX * dir.xx + offY * dir.yy; + vY = offX * (-dir.yy) + offY * dir.xx; + vX *= len.x; + vY *= len.y; + FfxFloat16x2 d2 = vX * vX + vY * vY; + d2 = min(d2, FFX_BROADCAST_FLOAT16X2(clp)); + FfxFloat16x2 wB = FFX_BROADCAST_FLOAT16X2(2.0 / 5.0) * d2 + FFX_BROADCAST_FLOAT16X2(-1.0); + FfxFloat16x2 wA = FFX_BROADCAST_FLOAT16X2(lob) * d2 + FFX_BROADCAST_FLOAT16X2(-1.0); + wB *= wB; + wA *= wA; + wB = FFX_BROADCAST_FLOAT16X2(25.0 / 16.0) * wB + FFX_BROADCAST_FLOAT16X2(-(25.0 / 16.0 - 1.0)); + FfxFloat16x2 w = wB * wA; + aCR += cR * w; + aCG += cG * w; + aCB += cB * w; + aW += w; +} + +// This runs 2 taps in parallel. +void FsrEasuSetH( + FFX_PARAMETER_INOUT FfxFloat16x2 dirPX, + FFX_PARAMETER_INOUT FfxFloat16x2 dirPY, + FFX_PARAMETER_INOUT FfxFloat16x2 lenP, + FFX_PARAMETER_IN FfxFloat16x2 pp, + FFX_PARAMETER_IN FfxBoolean biST, + FFX_PARAMETER_IN FfxBoolean biUV, + FFX_PARAMETER_IN FfxFloat16x2 lA, + FFX_PARAMETER_IN FfxFloat16x2 lB, + FFX_PARAMETER_IN FfxFloat16x2 lC, + FFX_PARAMETER_IN FfxFloat16x2 lD, + FFX_PARAMETER_IN FfxFloat16x2 lE) +{ + FfxFloat16x2 w = FFX_BROADCAST_FLOAT16X2(0.0); + + if (biST) + w = (FfxFloat16x2(1.0, 0.0) + FfxFloat16x2(-pp.x, pp.x)) * FFX_BROADCAST_FLOAT16X2(FFX_BROADCAST_FLOAT16(1.0) - pp.y); + + if (biUV) + w = (FfxFloat16x2(1.0, 0.0) + FfxFloat16x2(-pp.x, pp.x)) * FFX_BROADCAST_FLOAT16X2(pp.y); + + // ABS is not free in the packed FP16 path. + FfxFloat16x2 dc = lD - lC; + FfxFloat16x2 cb = lC - lB; + FfxFloat16x2 lenX = max(abs(dc), abs(cb)); + lenX = ffxReciprocalHalf(lenX); + + FfxFloat16x2 dirX = lD - lB; + dirPX += dirX * w; + lenX = ffxSaturate(abs(dirX) * lenX); + lenX *= lenX; + lenP += lenX * w; + FfxFloat16x2 ec = lE - lC; + FfxFloat16x2 ca = lC - lA; + FfxFloat16x2 lenY = max(abs(ec), abs(ca)); + lenY = ffxReciprocalHalf(lenY); + FfxFloat16x2 dirY = lE - lA; + dirPY += dirY * w; + lenY = ffxSaturate(abs(dirY) * lenY); + lenY *= lenY; + lenP += lenY * w; +} + +void FsrEasuH( + FFX_PARAMETER_OUT FfxFloat16x3 pix, + FFX_PARAMETER_IN FfxUInt32x2 ip, + FFX_PARAMETER_IN FfxUInt32x4 con0, + FFX_PARAMETER_IN FfxUInt32x4 con1, + FFX_PARAMETER_IN FfxUInt32x4 con2, + FFX_PARAMETER_IN FfxUInt32x4 con3) +{ + FfxFloat32x2 pp = FfxFloat32x2(ip) * ffxAsFloat(con0.xy) + ffxAsFloat(con0.zw); + FfxFloat32x2 fp = floor(pp); + pp -= fp; + FfxFloat16x2 ppp = FfxFloat16x2(pp); + + FfxFloat32x2 p0 = fp * ffxAsFloat(con1.xy) + ffxAsFloat(con1.zw); + FfxFloat32x2 p1 = p0 + ffxAsFloat(con2.xy); + FfxFloat32x2 p2 = p0 + ffxAsFloat(con2.zw); + FfxFloat32x2 p3 = p0 + ffxAsFloat(con3.xy); + FfxFloat16x4 bczzR = FsrEasuRH(p0); + FfxFloat16x4 bczzG = FsrEasuGH(p0); + FfxFloat16x4 bczzB = FsrEasuBH(p0); + FfxFloat16x4 ijfeR = FsrEasuRH(p1); + FfxFloat16x4 ijfeG = FsrEasuGH(p1); + FfxFloat16x4 ijfeB = FsrEasuBH(p1); + FfxFloat16x4 klhgR = FsrEasuRH(p2); + FfxFloat16x4 klhgG = FsrEasuGH(p2); + FfxFloat16x4 klhgB = FsrEasuBH(p2); + FfxFloat16x4 zzonR = FsrEasuRH(p3); + FfxFloat16x4 zzonG = FsrEasuGH(p3); + FfxFloat16x4 zzonB = FsrEasuBH(p3); + + FfxFloat16x4 bczzL = bczzB * FFX_BROADCAST_FLOAT16X4(0.5) + (bczzR * FFX_BROADCAST_FLOAT16X4(0.5) + bczzG); + FfxFloat16x4 ijfeL = ijfeB * FFX_BROADCAST_FLOAT16X4(0.5) + (ijfeR * FFX_BROADCAST_FLOAT16X4(0.5) + ijfeG); + FfxFloat16x4 klhgL = klhgB * FFX_BROADCAST_FLOAT16X4(0.5) + (klhgR * FFX_BROADCAST_FLOAT16X4(0.5) + klhgG); + FfxFloat16x4 zzonL = zzonB * FFX_BROADCAST_FLOAT16X4(0.5) + (zzonR * FFX_BROADCAST_FLOAT16X4(0.5) + zzonG); + FfxFloat16 bL = bczzL.x; + FfxFloat16 cL = bczzL.y; + FfxFloat16 iL = ijfeL.x; + FfxFloat16 jL = ijfeL.y; + FfxFloat16 fL = ijfeL.z; + FfxFloat16 eL = ijfeL.w; + FfxFloat16 kL = klhgL.x; + FfxFloat16 lL = klhgL.y; + FfxFloat16 hL = klhgL.z; + FfxFloat16 gL = klhgL.w; + FfxFloat16 oL = zzonL.z; + FfxFloat16 nL = zzonL.w; + + // This part is different, accumulating 2 taps in parallel. + FfxFloat16x2 dirPX = FFX_BROADCAST_FLOAT16X2(0.0); + FfxFloat16x2 dirPY = FFX_BROADCAST_FLOAT16X2(0.0); + FfxFloat16x2 lenP = FFX_BROADCAST_FLOAT16X2(0.0); + FsrEasuSetH(dirPX, + dirPY, + lenP, + ppp, + FfxUInt32(true), + FfxUInt32(false), + FfxFloat16x2(bL, cL), + FfxFloat16x2(eL, fL), + FfxFloat16x2(fL, gL), + FfxFloat16x2(gL, hL), + FfxFloat16x2(jL, kL)); + FsrEasuSetH(dirPX, + dirPY, + lenP, + ppp, + FfxUInt32(false), + FfxUInt32(true), + FfxFloat16x2(fL, gL), + FfxFloat16x2(iL, jL), + FfxFloat16x2(jL, kL), + FfxFloat16x2(kL, lL), + FfxFloat16x2(nL, oL)); + FfxFloat16x2 dir = FfxFloat16x2(dirPX.r + dirPX.g, dirPY.r + dirPY.g); + FfxFloat16 len = lenP.r + lenP.g; + + FfxFloat16x2 dir2 = dir * dir; + FfxFloat16 dirR = dir2.x + dir2.y; + FfxBoolean zro = FfxBoolean(dirR < FFX_BROADCAST_FLOAT16(1.0 / 32768.0)); + dirR = ffxApproximateReciprocalSquareRootHalf(dirR); + dirR = (zro > 0) ? FFX_BROADCAST_FLOAT16(1.0) : dirR; + dir.x = (zro > 0) ? FFX_BROADCAST_FLOAT16(1.0) : dir.x; + dir *= FFX_BROADCAST_FLOAT16X2(dirR); + len = len * FFX_BROADCAST_FLOAT16(0.5); + len *= len; + FfxFloat16 stretch = (dir.x * dir.x + dir.y * dir.y) * ffxApproximateReciprocalHalf(max(abs(dir.x), abs(dir.y))); + FfxFloat16x2 len2 = + FfxFloat16x2(FFX_BROADCAST_FLOAT16(1.0) + (stretch - FFX_BROADCAST_FLOAT16(1.0)) * len, FFX_BROADCAST_FLOAT16(1.0) + FFX_BROADCAST_FLOAT16(-0.5) * len); + FfxFloat16 lob = FFX_BROADCAST_FLOAT16(0.5) + FFX_BROADCAST_FLOAT16((1.0 / 4.0 - 0.04) - 0.5) * len; + FfxFloat16 clp = ffxApproximateReciprocalHalf(lob); + + // FP16 is different, using packed trick to do min and max in same operation. + FfxFloat16x2 bothR = + max(max(FfxFloat16x2(-ijfeR.z, ijfeR.z), FfxFloat16x2(-klhgR.w, klhgR.w)), max(FfxFloat16x2(-ijfeR.y, ijfeR.y), FfxFloat16x2(-klhgR.x, klhgR.x))); + FfxFloat16x2 bothG = + max(max(FfxFloat16x2(-ijfeG.z, ijfeG.z), FfxFloat16x2(-klhgG.w, klhgG.w)), max(FfxFloat16x2(-ijfeG.y, ijfeG.y), FfxFloat16x2(-klhgG.x, klhgG.x))); + FfxFloat16x2 bothB = + max(max(FfxFloat16x2(-ijfeB.z, ijfeB.z), FfxFloat16x2(-klhgB.w, klhgB.w)), max(FfxFloat16x2(-ijfeB.y, ijfeB.y), FfxFloat16x2(-klhgB.x, klhgB.x))); + + // This part is different for FP16, working pairs of taps at a time. + FfxFloat16x2 pR = FFX_BROADCAST_FLOAT16X2(0.0); + FfxFloat16x2 pG = FFX_BROADCAST_FLOAT16X2(0.0); + FfxFloat16x2 pB = FFX_BROADCAST_FLOAT16X2(0.0); + FfxFloat16x2 pW = FFX_BROADCAST_FLOAT16X2(0.0); + FsrEasuTapH(pR, pG, pB, pW, FfxFloat16x2(0.0, 1.0) - ppp.xx, FfxFloat16x2(-1.0, -1.0) - ppp.yy, dir, len2, lob, clp, bczzR.xy, bczzG.xy, bczzB.xy); + FsrEasuTapH(pR, pG, pB, pW, FfxFloat16x2(-1.0, 0.0) - ppp.xx, FfxFloat16x2(1.0, 1.0) - ppp.yy, dir, len2, lob, clp, ijfeR.xy, ijfeG.xy, ijfeB.xy); + FsrEasuTapH(pR, pG, pB, pW, FfxFloat16x2(0.0, -1.0) - ppp.xx, FfxFloat16x2(0.0, 0.0) - ppp.yy, dir, len2, lob, clp, ijfeR.zw, ijfeG.zw, ijfeB.zw); + FsrEasuTapH(pR, pG, pB, pW, FfxFloat16x2(1.0, 2.0) - ppp.xx, FfxFloat16x2(1.0, 1.0) - ppp.yy, dir, len2, lob, clp, klhgR.xy, klhgG.xy, klhgB.xy); + FsrEasuTapH(pR, pG, pB, pW, FfxFloat16x2(2.0, 1.0) - ppp.xx, FfxFloat16x2(0.0, 0.0) - ppp.yy, dir, len2, lob, clp, klhgR.zw, klhgG.zw, klhgB.zw); + FsrEasuTapH(pR, pG, pB, pW, FfxFloat16x2(1.0, 0.0) - ppp.xx, FfxFloat16x2(2.0, 2.0) - ppp.yy, dir, len2, lob, clp, zzonR.zw, zzonG.zw, zzonB.zw); + FfxFloat16x3 aC = FfxFloat16x3(pR.x + pR.y, pG.x + pG.y, pB.x + pB.y); + FfxFloat16 aW = pW.x + pW.y; + + // Slightly different for FP16 version due to combined min and max. + pix = min(FfxFloat16x3(bothR.y, bothG.y, bothB.y), max(-FfxFloat16x3(bothR.x, bothG.x, bothB.x), aC * FFX_BROADCAST_FLOAT16X3(ffxReciprocalHalf(aW)))); +} +#endif // #if defined(FFX_GPU) && defined(FFX_HALF) && defined(FFX_FSR_EASU_HALF) + +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [RCAS] ROBUST CONTRAST ADAPTIVE SHARPENING +// +//------------------------------------------------------------------------------------------------------------------------------ +// CAS uses a simplified mechanism to convert local contrast into a variable amount of sharpness. +// RCAS uses a more exact mechanism, solving for the maximum local sharpness possible before clipping. +// RCAS also has a built in process to limit sharpening of what it detects as possible noise. +// RCAS sharper does not support scaling, as it should be applied after EASU scaling. +// Pass EASU output straight into RCAS, no color conversions necessary. +//------------------------------------------------------------------------------------------------------------------------------ +// RCAS is based on the following logic. +// RCAS uses a 5 tap filter in a cross pattern (same as CAS), +// w n +// w 1 w for taps w m e +// w s +// Where 'w' is the negative lobe weight. +// output = (w*(n+e+w+s)+m)/(4*w+1) +// RCAS solves for 'w' by seeing where the signal might clip out of the {0 to 1} input range, +// 0 == (w*(n+e+w+s)+m)/(4*w+1) -> w = -m/(n+e+w+s) +// 1 == (w*(n+e+w+s)+m)/(4*w+1) -> w = (1-m)/(n+e+w+s-4*1) +// Then chooses the 'w' which results in no clipping, limits 'w', and multiplies by the 'sharp' amount. +// This solution above has issues with MSAA input as the steps along the gradient cause edge detection issues. +// So RCAS uses 4x the maximum and 4x the minimum (depending on equation)in place of the individual taps. +// As well as switching from 'm' to either the minimum or maximum (depending on side), to help in energy conservation. +// This stabilizes RCAS. +// RCAS does a simple highpass which is normalized against the local contrast then shaped, +// 0.25 +// 0.25 -1 0.25 +// 0.25 +// This is used as a noise detection filter, to reduce the effect of RCAS on grain, and focus on real edges. +// +// GLSL example for the required callbacks : +// +// FfxFloat16x4 FsrRcasLoadH(FfxInt16x2 p){return FfxFloat16x4(imageLoad(imgSrc,FfxInt32x2(p)));} +// void FsrRcasInputH(inout FfxFloat16 r,inout FfxFloat16 g,inout FfxFloat16 b) +// { +// //do any simple input color conversions here or leave empty if none needed +// } +// +// FsrRcasCon need to be called from the CPU or GPU to set up constants. +// Including a GPU example here, the 'con' value would be stored out to a constant buffer. +// +// FfxUInt32x4 con; +// FsrRcasCon(con, +// 0.0); // The scale is {0.0 := maximum sharpness, to N>0, where N is the number of stops (halving) of the reduction of sharpness}. +// --------------- +// RCAS sharpening supports a CAS-like pass-through alpha via, +// #define FSR_RCAS_PASSTHROUGH_ALPHA 1 +// RCAS also supports a define to enable a more expensive path to avoid some sharpening of noise. +// Would suggest it is better to apply film grain after RCAS sharpening (and after scaling) instead of using this define, +// #define FSR_RCAS_DENOISE 1 +//============================================================================================================================== +// This is set at the limit of providing unnatural results for sharpening. +#define FSR_RCAS_LIMIT (0.25-(1.0/16.0)) +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// CONSTANT SETUP +//============================================================================================================================== +// Call to setup required constant values (works on CPU or GPU). + FFX_STATIC void FsrRcasCon(FfxUInt32x4 con, + // The scale is {0.0 := maximum, to N>0, where N is the number of stops (halving) of the reduction of sharpness}. + FfxFloat32 sharpness) + { + // Transform from stops to linear value. + sharpness = exp2(-sharpness); + FfxFloat32x2 hSharp = {sharpness, sharpness}; + con[0] = ffxAsUInt32(sharpness); + con[1] = packHalf2x16(hSharp); + con[2] = 0; + con[3] = 0; + } + //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// NON-PACKED 32-BIT VERSION +//============================================================================================================================== +#if defined(FFX_GPU)&&defined(FSR_RCAS_F) + // Input callback prototypes that need to be implemented by calling shader + FfxFloat32x4 FsrRcasLoadF(FfxInt32x2 p); + void FsrRcasInputF(inout FfxFloat32 r,inout FfxFloat32 g,inout FfxFloat32 b); +//------------------------------------------------------------------------------------------------------------------------------ + void FsrRcasF(out FfxFloat32 pixR, // Output values, non-vector so port between RcasFilter() and RcasFilterH() is easy. + out FfxFloat32 pixG, + out FfxFloat32 pixB, +#ifdef FSR_RCAS_PASSTHROUGH_ALPHA + out FfxFloat32 pixA, +#endif + FfxUInt32x2 ip, // Integer pixel position in output. + FfxUInt32x4 con) + { // Constant generated by RcasSetup(). + // Algorithm uses minimal 3x3 pixel neighborhood. + // b + // d e f + // h + FfxInt32x2 sp = FfxInt32x2(ip); + FfxFloat32x3 b = FsrRcasLoadF(sp + FfxInt32x2(0, -1)).rgb; + FfxFloat32x3 d = FsrRcasLoadF(sp + FfxInt32x2(-1, 0)).rgb; +#ifdef FSR_RCAS_PASSTHROUGH_ALPHA + FfxFloat32x4 ee = FsrRcasLoadF(sp); + FfxFloat32x3 e = ee.rgb; + pixA = ee.a; +#else + FfxFloat32x3 e = FsrRcasLoadF(sp).rgb; +#endif + FfxFloat32x3 f = FsrRcasLoadF(sp + FfxInt32x2(1, 0)).rgb; + FfxFloat32x3 h = FsrRcasLoadF(sp + FfxInt32x2(0, 1)).rgb; + // Rename (32-bit) or regroup (16-bit). + FfxFloat32 bR = b.r; + FfxFloat32 bG = b.g; + FfxFloat32 bB = b.b; + FfxFloat32 dR = d.r; + FfxFloat32 dG = d.g; + FfxFloat32 dB = d.b; + FfxFloat32 eR = e.r; + FfxFloat32 eG = e.g; + FfxFloat32 eB = e.b; + FfxFloat32 fR = f.r; + FfxFloat32 fG = f.g; + FfxFloat32 fB = f.b; + FfxFloat32 hR = h.r; + FfxFloat32 hG = h.g; + FfxFloat32 hB = h.b; + // Run optional input transform. + FsrRcasInputF(bR, bG, bB); + FsrRcasInputF(dR, dG, dB); + FsrRcasInputF(eR, eG, eB); + FsrRcasInputF(fR, fG, fB); + FsrRcasInputF(hR, hG, hB); + // Luma times 2. + FfxFloat32 bL = bB * FfxFloat32(0.5) + (bR * FfxFloat32(0.5) + bG); + FfxFloat32 dL = dB * FfxFloat32(0.5) + (dR * FfxFloat32(0.5) + dG); + FfxFloat32 eL = eB * FfxFloat32(0.5) + (eR * FfxFloat32(0.5) + eG); + FfxFloat32 fL = fB * FfxFloat32(0.5) + (fR * FfxFloat32(0.5) + fG); + FfxFloat32 hL = hB * FfxFloat32(0.5) + (hR * FfxFloat32(0.5) + hG); + // Noise detection. + FfxFloat32 nz = FfxFloat32(0.25) * bL + FfxFloat32(0.25) * dL + FfxFloat32(0.25) * fL + FfxFloat32(0.25) * hL - eL; + nz = ffxSaturate(abs(nz) * ffxApproximateReciprocalMedium(ffxMax3(ffxMax3(bL, dL, eL), fL, hL) - ffxMin3(ffxMin3(bL, dL, eL), fL, hL))); + nz = FfxFloat32(-0.5) * nz + FfxFloat32(1.0); + // Min and max of ring. + FfxFloat32 mn4R = ffxMin(ffxMin3(bR, dR, fR), hR); + FfxFloat32 mn4G = ffxMin(ffxMin3(bG, dG, fG), hG); + FfxFloat32 mn4B = ffxMin(ffxMin3(bB, dB, fB), hB); + FfxFloat32 mx4R = max(ffxMax3(bR, dR, fR), hR); + FfxFloat32 mx4G = max(ffxMax3(bG, dG, fG), hG); + FfxFloat32 mx4B = max(ffxMax3(bB, dB, fB), hB); + // Immediate constants for peak range. + FfxFloat32x2 peakC = FfxFloat32x2(1.0, -1.0 * 4.0); + // Limiters, these need to be high precision RCPs. + FfxFloat32 hitMinR = mn4R * rcp(FfxFloat32(4.0) * mx4R); + FfxFloat32 hitMinG = mn4G * rcp(FfxFloat32(4.0) * mx4G); + FfxFloat32 hitMinB = mn4B * rcp(FfxFloat32(4.0) * mx4B); + FfxFloat32 hitMaxR = (peakC.x - mx4R) * rcp(FfxFloat32(4.0) * mn4R + peakC.y); + FfxFloat32 hitMaxG = (peakC.x - mx4G) * rcp(FfxFloat32(4.0) * mn4G + peakC.y); + FfxFloat32 hitMaxB = (peakC.x - mx4B) * rcp(FfxFloat32(4.0) * mn4B + peakC.y); + FfxFloat32 lobeR = max(-hitMinR, hitMaxR); + FfxFloat32 lobeG = max(-hitMinG, hitMaxG); + FfxFloat32 lobeB = max(-hitMinB, hitMaxB); + FfxFloat32 lobe = max(FfxFloat32(-FSR_RCAS_LIMIT), ffxMin(ffxMax3(lobeR, lobeG, lobeB), FfxFloat32(0.0))) * ffxAsFloat + (con.x); + // Apply noise removal. +#ifdef FSR_RCAS_DENOISE + lobe *= nz; +#endif + // Resolve, which needs the medium precision rcp approximation to avoid visible tonality changes. + FfxFloat32 rcpL = ffxApproximateReciprocalMedium(FfxFloat32(4.0) * lobe + FfxFloat32(1.0)); + pixR = (lobe * bR + lobe * dR + lobe * hR + lobe * fR + eR) * rcpL; + pixG = (lobe * bG + lobe * dG + lobe * hG + lobe * fG + eG) * rcpL; + pixB = (lobe * bB + lobe * dB + lobe * hB + lobe * fB + eB) * rcpL; + return; + } +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// NON-PACKED 16-BIT VERSION +//============================================================================================================================== +#if defined(FFX_GPU) && FFX_HALF == 1 && defined(FSR_RCAS_H) + // Input callback prototypes that need to be implemented by calling shader + FfxFloat16x4 FsrRcasLoadH(FfxInt16x2 p); + void FsrRcasInputH(inout FfxFloat16 r,inout FfxFloat16 g,inout FfxFloat16 b); +//------------------------------------------------------------------------------------------------------------------------------ + void FsrRcasH( + out FfxFloat16 pixR, // Output values, non-vector so port between RcasFilter() and RcasFilterH() is easy. + out FfxFloat16 pixG, + out FfxFloat16 pixB, + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + out FfxFloat16 pixA, + #endif + FfxUInt32x2 ip, // Integer pixel position in output. + FfxUInt32x4 con){ // Constant generated by RcasSetup(). + // Sharpening algorithm uses minimal 3x3 pixel neighborhood. + // b + // d e f + // h + FfxInt16x2 sp=FfxInt16x2(ip); + FfxFloat16x3 b=FsrRcasLoadH(sp+FfxInt16x2( 0,-1)).rgb; + FfxFloat16x3 d=FsrRcasLoadH(sp+FfxInt16x2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + FfxFloat16x4 ee=FsrRcasLoadH(sp); + FfxFloat16x3 e=ee.rgb;pixA=ee.a; + #else + FfxFloat16x3 e=FsrRcasLoadH(sp).rgb; + #endif + FfxFloat16x3 f=FsrRcasLoadH(sp+FfxInt16x2( 1, 0)).rgb; + FfxFloat16x3 h=FsrRcasLoadH(sp+FfxInt16x2( 0, 1)).rgb; + // Rename (32-bit) or regroup (16-bit). + FfxFloat16 bR=b.r; + FfxFloat16 bG=b.g; + FfxFloat16 bB=b.b; + FfxFloat16 dR=d.r; + FfxFloat16 dG=d.g; + FfxFloat16 dB=d.b; + FfxFloat16 eR=e.r; + FfxFloat16 eG=e.g; + FfxFloat16 eB=e.b; + FfxFloat16 fR=f.r; + FfxFloat16 fG=f.g; + FfxFloat16 fB=f.b; + FfxFloat16 hR=h.r; + FfxFloat16 hG=h.g; + FfxFloat16 hB=h.b; + // Run optional input transform. + FsrRcasInputH(bR,bG,bB); + FsrRcasInputH(dR,dG,dB); + FsrRcasInputH(eR,eG,eB); + FsrRcasInputH(fR,fG,fB); + FsrRcasInputH(hR,hG,hB); + // Luma times 2. + FfxFloat16 bL=bB*FFX_BROADCAST_FLOAT16(0.5)+(bR*FFX_BROADCAST_FLOAT16(0.5)+bG); + FfxFloat16 dL=dB*FFX_BROADCAST_FLOAT16(0.5)+(dR*FFX_BROADCAST_FLOAT16(0.5)+dG); + FfxFloat16 eL=eB*FFX_BROADCAST_FLOAT16(0.5)+(eR*FFX_BROADCAST_FLOAT16(0.5)+eG); + FfxFloat16 fL=fB*FFX_BROADCAST_FLOAT16(0.5)+(fR*FFX_BROADCAST_FLOAT16(0.5)+fG); + FfxFloat16 hL=hB*FFX_BROADCAST_FLOAT16(0.5)+(hR*FFX_BROADCAST_FLOAT16(0.5)+hG); + // Noise detection. + FfxFloat16 nz=FFX_BROADCAST_FLOAT16(0.25)*bL+FFX_BROADCAST_FLOAT16(0.25)*dL+FFX_BROADCAST_FLOAT16(0.25)*fL+FFX_BROADCAST_FLOAT16(0.25)*hL-eL; + nz=ffxSaturate(abs(nz)*ffxApproximateReciprocalMediumHalf(ffxMax3Half(ffxMax3Half(bL,dL,eL),fL,hL)-ffxMin3Half(ffxMin3Half(bL,dL,eL),fL,hL))); + nz=FFX_BROADCAST_FLOAT16(-0.5)*nz+FFX_BROADCAST_FLOAT16(1.0); + // Min and max of ring. + FfxFloat16 mn4R=min(ffxMin3Half(bR,dR,fR),hR); + FfxFloat16 mn4G=min(ffxMin3Half(bG,dG,fG),hG); + FfxFloat16 mn4B=min(ffxMin3Half(bB,dB,fB),hB); + FfxFloat16 mx4R=max(ffxMax3Half(bR,dR,fR),hR); + FfxFloat16 mx4G=max(ffxMax3Half(bG,dG,fG),hG); + FfxFloat16 mx4B=max(ffxMax3Half(bB,dB,fB),hB); + // Immediate constants for peak range. + FfxFloat16x2 peakC=FfxFloat16x2(1.0,-1.0*4.0); + // Limiters, these need to be high precision RCPs. + FfxFloat16 hitMinR=mn4R*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16(4.0)*mx4R); + FfxFloat16 hitMinG=mn4G*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16(4.0)*mx4G); + FfxFloat16 hitMinB=mn4B*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16(4.0)*mx4B); + FfxFloat16 hitMaxR=(peakC.x-mx4R)*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16(4.0)*mn4R+peakC.y); + FfxFloat16 hitMaxG=(peakC.x-mx4G)*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16(4.0)*mn4G+peakC.y); + FfxFloat16 hitMaxB=(peakC.x-mx4B)*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16(4.0)*mn4B+peakC.y); + FfxFloat16 lobeR=max(-hitMinR,hitMaxR); + FfxFloat16 lobeG=max(-hitMinG,hitMaxG); + FfxFloat16 lobeB=max(-hitMinB,hitMaxB); + FfxFloat16 lobe=max(FFX_BROADCAST_FLOAT16(-FSR_RCAS_LIMIT),min(ffxMax3Half(lobeR,lobeG,lobeB),FFX_BROADCAST_FLOAT16(0.0)))*FFX_UINT32_TO_FLOAT16X2(con.y).x; + // Apply noise removal. + #ifdef FSR_RCAS_DENOISE + lobe*=nz; + #endif + // Resolve, which needs the medium precision rcp approximation to avoid visible tonality changes. + FfxFloat16 rcpL=ffxApproximateReciprocalMediumHalf(FFX_BROADCAST_FLOAT16(4.0)*lobe+FFX_BROADCAST_FLOAT16(1.0)); + pixR=(lobe*bR+lobe*dR+lobe*hR+lobe*fR+eR)*rcpL; + pixG=(lobe*bG+lobe*dG+lobe*hG+lobe*fG+eG)*rcpL; + pixB=(lobe*bB+lobe*dB+lobe*hB+lobe*fB+eB)*rcpL; +} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// PACKED 16-BIT VERSION +//============================================================================================================================== +#if defined(FFX_GPU)&& FFX_HALF == 1 && defined(FSR_RCAS_HX2) + // Input callback prototypes that need to be implemented by the calling shader + FfxFloat16x4 FsrRcasLoadHx2(FfxInt16x2 p); + void FsrRcasInputHx2(inout FfxFloat16x2 r,inout FfxFloat16x2 g,inout FfxFloat16x2 b); +//------------------------------------------------------------------------------------------------------------------------------ + // Can be used to convert from packed Structures of Arrays to Arrays of Structures for store. + void FsrRcasDepackHx2(out FfxFloat16x4 pix0,out FfxFloat16x4 pix1,FfxFloat16x2 pixR,FfxFloat16x2 pixG,FfxFloat16x2 pixB){ + #ifdef FFX_HLSL + // Invoke a slower path for DX only, since it won't allow uninitialized values. + pix0.a=pix1.a=0.0; + #endif + pix0.rgb=FfxFloat16x3(pixR.x,pixG.x,pixB.x); + pix1.rgb=FfxFloat16x3(pixR.y,pixG.y,pixB.y);} +//------------------------------------------------------------------------------------------------------------------------------ + void FsrRcasHx2( + // Output values are for 2 8x8 tiles in a 16x8 region. + // pix.x = left 8x8 tile + // pix.y = right 8x8 tile + // This enables later processing to easily be packed as well. + out FfxFloat16x2 pixR, + out FfxFloat16x2 pixG, + out FfxFloat16x2 pixB, + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + out FfxFloat16x2 pixA, + #endif + FfxUInt32x2 ip, // Integer pixel position in output. + FfxUInt32x4 con){ // Constant generated by RcasSetup(). + // No scaling algorithm uses minimal 3x3 pixel neighborhood. + FfxInt16x2 sp0=FfxInt16x2(ip); + FfxFloat16x3 b0=FsrRcasLoadHx2(sp0+FfxInt16x2( 0,-1)).rgb; + FfxFloat16x3 d0=FsrRcasLoadHx2(sp0+FfxInt16x2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + FfxFloat16x4 ee0=FsrRcasLoadHx2(sp0); + FfxFloat16x3 e0=ee0.rgb;pixA.r=ee0.a; + #else + FfxFloat16x3 e0=FsrRcasLoadHx2(sp0).rgb; + #endif + FfxFloat16x3 f0=FsrRcasLoadHx2(sp0+FfxInt16x2( 1, 0)).rgb; + FfxFloat16x3 h0=FsrRcasLoadHx2(sp0+FfxInt16x2( 0, 1)).rgb; + FfxInt16x2 sp1=sp0+FfxInt16x2(8,0); + FfxFloat16x3 b1=FsrRcasLoadHx2(sp1+FfxInt16x2( 0,-1)).rgb; + FfxFloat16x3 d1=FsrRcasLoadHx2(sp1+FfxInt16x2(-1, 0)).rgb; + #ifdef FSR_RCAS_PASSTHROUGH_ALPHA + FfxFloat16x4 ee1=FsrRcasLoadHx2(sp1); + FfxFloat16x3 e1=ee1.rgb;pixA.g=ee1.a; + #else + FfxFloat16x3 e1=FsrRcasLoadHx2(sp1).rgb; + #endif + FfxFloat16x3 f1=FsrRcasLoadHx2(sp1+FfxInt16x2( 1, 0)).rgb; + FfxFloat16x3 h1=FsrRcasLoadHx2(sp1+FfxInt16x2( 0, 1)).rgb; + // Arrays of Structures to Structures of Arrays conversion. + FfxFloat16x2 bR=FfxFloat16x2(b0.r,b1.r); + FfxFloat16x2 bG=FfxFloat16x2(b0.g,b1.g); + FfxFloat16x2 bB=FfxFloat16x2(b0.b,b1.b); + FfxFloat16x2 dR=FfxFloat16x2(d0.r,d1.r); + FfxFloat16x2 dG=FfxFloat16x2(d0.g,d1.g); + FfxFloat16x2 dB=FfxFloat16x2(d0.b,d1.b); + FfxFloat16x2 eR=FfxFloat16x2(e0.r,e1.r); + FfxFloat16x2 eG=FfxFloat16x2(e0.g,e1.g); + FfxFloat16x2 eB=FfxFloat16x2(e0.b,e1.b); + FfxFloat16x2 fR=FfxFloat16x2(f0.r,f1.r); + FfxFloat16x2 fG=FfxFloat16x2(f0.g,f1.g); + FfxFloat16x2 fB=FfxFloat16x2(f0.b,f1.b); + FfxFloat16x2 hR=FfxFloat16x2(h0.r,h1.r); + FfxFloat16x2 hG=FfxFloat16x2(h0.g,h1.g); + FfxFloat16x2 hB=FfxFloat16x2(h0.b,h1.b); + // Run optional input transform. + FsrRcasInputHx2(bR,bG,bB); + FsrRcasInputHx2(dR,dG,dB); + FsrRcasInputHx2(eR,eG,eB); + FsrRcasInputHx2(fR,fG,fB); + FsrRcasInputHx2(hR,hG,hB); + // Luma times 2. + FfxFloat16x2 bL=bB*FFX_BROADCAST_FLOAT16X2(0.5)+(bR*FFX_BROADCAST_FLOAT16X2(0.5)+bG); + FfxFloat16x2 dL=dB*FFX_BROADCAST_FLOAT16X2(0.5)+(dR*FFX_BROADCAST_FLOAT16X2(0.5)+dG); + FfxFloat16x2 eL=eB*FFX_BROADCAST_FLOAT16X2(0.5)+(eR*FFX_BROADCAST_FLOAT16X2(0.5)+eG); + FfxFloat16x2 fL=fB*FFX_BROADCAST_FLOAT16X2(0.5)+(fR*FFX_BROADCAST_FLOAT16X2(0.5)+fG); + FfxFloat16x2 hL=hB*FFX_BROADCAST_FLOAT16X2(0.5)+(hR*FFX_BROADCAST_FLOAT16X2(0.5)+hG); + // Noise detection. + FfxFloat16x2 nz=FFX_BROADCAST_FLOAT16X2(0.25)*bL+FFX_BROADCAST_FLOAT16X2(0.25)*dL+FFX_BROADCAST_FLOAT16X2(0.25)*fL+FFX_BROADCAST_FLOAT16X2(0.25)*hL-eL; + nz=ffxSaturate(abs(nz)*ffxApproximateReciprocalMediumHalf(ffxMax3Half(ffxMax3Half(bL,dL,eL),fL,hL)-ffxMin3Half(ffxMin3Half(bL,dL,eL),fL,hL))); + nz=FFX_BROADCAST_FLOAT16X2(-0.5)*nz+FFX_BROADCAST_FLOAT16X2(1.0); + // Min and max of ring. + FfxFloat16x2 mn4R=min(ffxMin3Half(bR,dR,fR),hR); + FfxFloat16x2 mn4G=min(ffxMin3Half(bG,dG,fG),hG); + FfxFloat16x2 mn4B=min(ffxMin3Half(bB,dB,fB),hB); + FfxFloat16x2 mx4R=max(ffxMax3Half(bR,dR,fR),hR); + FfxFloat16x2 mx4G=max(ffxMax3Half(bG,dG,fG),hG); + FfxFloat16x2 mx4B=max(ffxMax3Half(bB,dB,fB),hB); + // Immediate constants for peak range. + FfxFloat16x2 peakC=FfxFloat16x2(1.0,-1.0*4.0); + // Limiters, these need to be high precision RCPs. + FfxFloat16x2 hitMinR=mn4R*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16X2(4.0)*mx4R); + FfxFloat16x2 hitMinG=mn4G*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16X2(4.0)*mx4G); + FfxFloat16x2 hitMinB=mn4B*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16X2(4.0)*mx4B); + FfxFloat16x2 hitMaxR=(peakC.x-mx4R)*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16X2(4.0)*mn4R+peakC.y); + FfxFloat16x2 hitMaxG=(peakC.x-mx4G)*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16X2(4.0)*mn4G+peakC.y); + FfxFloat16x2 hitMaxB=(peakC.x-mx4B)*ffxReciprocalHalf(FFX_BROADCAST_FLOAT16X2(4.0)*mn4B+peakC.y); + FfxFloat16x2 lobeR=max(-hitMinR,hitMaxR); + FfxFloat16x2 lobeG=max(-hitMinG,hitMaxG); + FfxFloat16x2 lobeB=max(-hitMinB,hitMaxB); + FfxFloat16x2 lobe=max(FFX_BROADCAST_FLOAT16X2(-FSR_RCAS_LIMIT),min(ffxMax3Half(lobeR,lobeG,lobeB),FFX_BROADCAST_FLOAT16X2(0.0)))*FFX_BROADCAST_FLOAT16X2(FFX_UINT32_TO_FLOAT16X2(con.y).x); + // Apply noise removal. + #ifdef FSR_RCAS_DENOISE + lobe*=nz; + #endif + // Resolve, which needs the medium precision rcp approximation to avoid visible tonality changes. + FfxFloat16x2 rcpL=ffxApproximateReciprocalMediumHalf(FFX_BROADCAST_FLOAT16X2(4.0)*lobe+FFX_BROADCAST_FLOAT16X2(1.0)); + pixR=(lobe*bR+lobe*dR+lobe*hR+lobe*fR+eR)*rcpL; + pixG=(lobe*bG+lobe*dG+lobe*hG+lobe*fG+eG)*rcpL; + pixB=(lobe*bB+lobe*dB+lobe*hB+lobe*fB+eB)*rcpL;} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [LFGA] LINEAR FILM GRAIN APPLICATOR +// +//------------------------------------------------------------------------------------------------------------------------------ +// Adding output-resolution film grain after scaling is a good way to mask both rendering and scaling artifacts. +// Suggest using tiled blue noise as film grain input, with peak noise frequency set for a specific look and feel. +// The 'Lfga*()' functions provide a convenient way to introduce grain. +// These functions limit grain based on distance to signal limits. +// This is done so that the grain is temporally energy preserving, and thus won't modify image tonality. +// Grain application should be done in a linear colorspace. +// The grain should be temporally changing, but have a temporal sum per pixel that adds to zero (non-biased). +//------------------------------------------------------------------------------------------------------------------------------ +// Usage, +// FsrLfga*( +// color, // In/out linear colorspace color {0 to 1} ranged. +// grain, // Per pixel grain texture value {-0.5 to 0.5} ranged, input is 3-channel to support colored grain. +// amount); // Amount of grain (0 to 1} ranged. +//------------------------------------------------------------------------------------------------------------------------------ +// Example if grain texture is monochrome: 'FsrLfgaF(color,ffxBroadcast3(grain),amount)' +//============================================================================================================================== +#if defined(FFX_GPU) + // Maximum grain is the minimum distance to the signal limit. + void FsrLfgaF(inout FfxFloat32x3 c, FfxFloat32x3 t, FfxFloat32 a) + { + c += (t * ffxBroadcast3(a)) * ffxMin(ffxBroadcast3(1.0) - c, c); + } +#endif +//============================================================================================================================== +#if defined(FFX_GPU)&& FFX_HALF == 1 + // Half precision version (slower). + void FsrLfgaH(inout FfxFloat16x3 c, FfxFloat16x3 t, FfxFloat16 a) + { + c += (t * FFX_BROADCAST_FLOAT16X3(a)) * min(FFX_BROADCAST_FLOAT16X3(1.0) - c, c); + } + //------------------------------------------------------------------------------------------------------------------------------ + // Packed half precision version (faster). + void FsrLfgaHx2(inout FfxFloat16x2 cR,inout FfxFloat16x2 cG,inout FfxFloat16x2 cB,FfxFloat16x2 tR,FfxFloat16x2 tG,FfxFloat16x2 tB,FfxFloat16 a){ + cR+=(tR*FFX_BROADCAST_FLOAT16X2(a))*min(FFX_BROADCAST_FLOAT16X2(1.0)-cR,cR);cG+=(tG*FFX_BROADCAST_FLOAT16X2(a))*min(FFX_BROADCAST_FLOAT16X2(1.0)-cG,cG);cB+=(tB*FFX_BROADCAST_FLOAT16X2(a))*min(FFX_BROADCAST_FLOAT16X2(1.0)-cB,cB);} +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [SRTM] SIMPLE REVERSIBLE TONE-MAPPER +// +//------------------------------------------------------------------------------------------------------------------------------ +// This provides a way to take linear HDR color {0 to FP16_MAX} and convert it into a temporary {0 to 1} ranged post-tonemapped linear. +// The tonemapper preserves RGB ratio, which helps maintain HDR color bleed during filtering. +//------------------------------------------------------------------------------------------------------------------------------ +// Reversible tonemapper usage, +// FsrSrtm*(color); // {0 to FP16_MAX} converted to {0 to 1}. +// FsrSrtmInv*(color); // {0 to 1} converted into {0 to 32768, output peak safe for FP16}. +//============================================================================================================================== +#if defined(FFX_GPU) + void FsrSrtmF(inout FfxFloat32x3 c) + { + c *= ffxBroadcast3(rcp(ffxMax3(c.r, c.g, c.b) + FfxFloat32(1.0))); + } + // The extra max solves the c=1.0 case (which is a /0). + void FsrSrtmInvF(inout FfxFloat32x3 c){c*=ffxBroadcast3(rcp(max(FfxFloat32(1.0/32768.0),FfxFloat32(1.0)-ffxMax3(c.r,c.g,c.b))));} +#endif +//============================================================================================================================== +#if defined(FFX_GPU )&& FFX_HALF == 1 + void FsrSrtmH(inout FfxFloat16x3 c) + { + c *= FFX_BROADCAST_FLOAT16X3(ffxReciprocalHalf(ffxMax3Half(c.r, c.g, c.b) + FFX_BROADCAST_FLOAT16(1.0))); + } + void FsrSrtmInvH(inout FfxFloat16x3 c) + { + c *= FFX_BROADCAST_FLOAT16X3(ffxReciprocalHalf(max(FFX_BROADCAST_FLOAT16(1.0 / 32768.0), FFX_BROADCAST_FLOAT16(1.0) - ffxMax3Half(c.r, c.g, c.b)))); + } + //------------------------------------------------------------------------------------------------------------------------------ + void FsrSrtmHx2(inout FfxFloat16x2 cR, inout FfxFloat16x2 cG, inout FfxFloat16x2 cB) + { + FfxFloat16x2 rcp = ffxReciprocalHalf(ffxMax3Half(cR, cG, cB) + FFX_BROADCAST_FLOAT16X2(1.0)); + cR *= rcp; + cG *= rcp; + cB *= rcp; + } + void FsrSrtmInvHx2(inout FfxFloat16x2 cR,inout FfxFloat16x2 cG,inout FfxFloat16x2 cB) + { + FfxFloat16x2 rcp=ffxReciprocalHalf(max(FFX_BROADCAST_FLOAT16X2(1.0/32768.0),FFX_BROADCAST_FLOAT16X2(1.0)-ffxMax3Half(cR,cG,cB))); + cR*=rcp; + cG*=rcp; + cB*=rcp; + } +#endif +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//_____________________________________________________________/\_______________________________________________________________ +//============================================================================================================================== +// +// FSR - [TEPD] TEMPORAL ENERGY PRESERVING DITHER +// +//------------------------------------------------------------------------------------------------------------------------------ +// Temporally energy preserving dithered {0 to 1} linear to gamma 2.0 conversion. +// Gamma 2.0 is used so that the conversion back to linear is just to square the color. +// The conversion comes in 8-bit and 10-bit modes, designed for output to 8-bit UNORM or 10:10:10:2 respectively. +// Given good non-biased temporal blue noise as dither input, +// the output dither will temporally conserve energy. +// This is done by choosing the linear nearest step point instead of perceptual nearest. +// See code below for details. +//------------------------------------------------------------------------------------------------------------------------------ +// DX SPEC RULES FOR FLOAT->UNORM 8-BIT CONVERSION +// =============================================== +// - Output is 'FfxUInt32(floor(saturate(n)*255.0+0.5))'. +// - Thus rounding is to nearest. +// - NaN gets converted to zero. +// - INF is clamped to {0.0 to 1.0}. +//============================================================================================================================== +#if defined(FFX_GPU) + // Hand tuned integer position to dither value, with more values than simple checkerboard. + // Only 32-bit has enough precision for this compddation. + // Output is {0 to <1}. + FfxFloat32 FsrTepdDitF(FfxUInt32x2 p, FfxUInt32 f) + { + FfxFloat32 x = FfxFloat32(p.x + f); + FfxFloat32 y = FfxFloat32(p.y); + // The 1.61803 golden ratio. + FfxFloat32 a = FfxFloat32((1.0 + ffxSqrt(5.0f)) / 2.0); + // Number designed to provide a good visual pattern. + FfxFloat32 b = FfxFloat32(1.0 / 3.69); + x = x * a + (y * b); + return ffxFract(x); + } + //------------------------------------------------------------------------------------------------------------------------------ + // This version is 8-bit gamma 2.0. + // The 'c' input is {0 to 1}. + // Output is {0 to 1} ready for image store. + void FsrTepdC8F(inout FfxFloat32x3 c, FfxFloat32 dit) + { + FfxFloat32x3 n = ffxSqrt(c); + n = floor(n * ffxBroadcast3(255.0)) * ffxBroadcast3(1.0 / 255.0); + FfxFloat32x3 a = n * n; + FfxFloat32x3 b = n + ffxBroadcast3(1.0 / 255.0); + b = b * b; + // Ratio of 'a' to 'b' required to produce 'c'. + // ffxApproximateReciprocal() won't work here (at least for very high dynamic ranges). + // ffxApproximateReciprocalMedium() is an IADD,FMA,MUL. + FfxFloat32x3 r = (c - b) * ffxApproximateReciprocalMedium(a - b); + // Use the ratio as a cutoff to choose 'a' or 'b'. + // ffxIsGreaterThanZero() is a MUL. + c = ffxSaturate(n + ffxIsGreaterThanZero(ffxBroadcast3(dit) - r) * ffxBroadcast3(1.0 / 255.0)); + } + //------------------------------------------------------------------------------------------------------------------------------ + // This version is 10-bit gamma 2.0. + // The 'c' input is {0 to 1}. + // Output is {0 to 1} ready for image store. + void FsrTepdC10F(inout FfxFloat32x3 c, FfxFloat32 dit) + { + FfxFloat32x3 n = ffxSqrt(c); + n = floor(n * ffxBroadcast3(1023.0)) * ffxBroadcast3(1.0 / 1023.0); + FfxFloat32x3 a = n * n; + FfxFloat32x3 b = n + ffxBroadcast3(1.0 / 1023.0); + b = b * b; + FfxFloat32x3 r = (c - b) * ffxApproximateReciprocalMedium(a - b); + c = ffxSaturate(n + ffxIsGreaterThanZero(ffxBroadcast3(dit) - r) * ffxBroadcast3(1.0 / 1023.0)); + } +#endif +//============================================================================================================================== +#if defined(FFX_GPU)&& FFX_HALF == 1 + FfxFloat16 FsrTepdDitH(FfxUInt32x2 p, FfxUInt32 f) + { + FfxFloat32 x = FfxFloat32(p.x + f); + FfxFloat32 y = FfxFloat32(p.y); + FfxFloat32 a = FfxFloat32((1.0 + ffxSqrt(5.0f)) / 2.0); + FfxFloat32 b = FfxFloat32(1.0 / 3.69); + x = x * a + (y * b); + return FfxFloat16(ffxFract(x)); + } + //------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC8H(inout FfxFloat16x3 c, FfxFloat16 dit) + { + FfxFloat16x3 n = sqrt(c); + n = floor(n * FFX_BROADCAST_FLOAT16X3(255.0)) * FFX_BROADCAST_FLOAT16X3(1.0 / 255.0); + FfxFloat16x3 a = n * n; + FfxFloat16x3 b = n + FFX_BROADCAST_FLOAT16X3(1.0 / 255.0); + b = b * b; + FfxFloat16x3 r = (c - b) * ffxApproximateReciprocalMediumHalf(a - b); + c = ffxSaturate(n + ffxIsGreaterThanZeroHalf(FFX_BROADCAST_FLOAT16X3(dit) - r) * FFX_BROADCAST_FLOAT16X3(1.0 / 255.0)); + } + //------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC10H(inout FfxFloat16x3 c, FfxFloat16 dit) + { + FfxFloat16x3 n = sqrt(c); + n = floor(n * FFX_BROADCAST_FLOAT16X3(1023.0)) * FFX_BROADCAST_FLOAT16X3(1.0 / 1023.0); + FfxFloat16x3 a = n * n; + FfxFloat16x3 b = n + FFX_BROADCAST_FLOAT16X3(1.0 / 1023.0); + b = b * b; + FfxFloat16x3 r = (c - b) * ffxApproximateReciprocalMediumHalf(a - b); + c = ffxSaturate(n + ffxIsGreaterThanZeroHalf(FFX_BROADCAST_FLOAT16X3(dit) - r) * FFX_BROADCAST_FLOAT16X3(1.0 / 1023.0)); + } + //============================================================================================================================== + // This computes dither for positions 'p' and 'p+{8,0}'. + FfxFloat16x2 FsrTepdDitHx2(FfxUInt32x2 p, FfxUInt32 f) + { + FfxFloat32x2 x; + x.x = FfxFloat32(p.x + f); + x.y = x.x + FfxFloat32(8.0); + FfxFloat32 y = FfxFloat32(p.y); + FfxFloat32 a = FfxFloat32((1.0 + ffxSqrt(5.0f)) / 2.0); + FfxFloat32 b = FfxFloat32(1.0 / 3.69); + x = x * ffxBroadcast2(a) + ffxBroadcast2(y * b); + return FfxFloat16x2(ffxFract(x)); + } + //------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC8Hx2(inout FfxFloat16x2 cR, inout FfxFloat16x2 cG, inout FfxFloat16x2 cB, FfxFloat16x2 dit) + { + FfxFloat16x2 nR = sqrt(cR); + FfxFloat16x2 nG = sqrt(cG); + FfxFloat16x2 nB = sqrt(cB); + nR = floor(nR * FFX_BROADCAST_FLOAT16X2(255.0)) * FFX_BROADCAST_FLOAT16X2(1.0 / 255.0); + nG = floor(nG * FFX_BROADCAST_FLOAT16X2(255.0)) * FFX_BROADCAST_FLOAT16X2(1.0 / 255.0); + nB = floor(nB * FFX_BROADCAST_FLOAT16X2(255.0)) * FFX_BROADCAST_FLOAT16X2(1.0 / 255.0); + FfxFloat16x2 aR = nR * nR; + FfxFloat16x2 aG = nG * nG; + FfxFloat16x2 aB = nB * nB; + FfxFloat16x2 bR = nR + FFX_BROADCAST_FLOAT16X2(1.0 / 255.0); + bR = bR * bR; + FfxFloat16x2 bG = nG + FFX_BROADCAST_FLOAT16X2(1.0 / 255.0); + bG = bG * bG; + FfxFloat16x2 bB = nB + FFX_BROADCAST_FLOAT16X2(1.0 / 255.0); + bB = bB * bB; + FfxFloat16x2 rR = (cR - bR) * ffxApproximateReciprocalMediumHalf(aR - bR); + FfxFloat16x2 rG = (cG - bG) * ffxApproximateReciprocalMediumHalf(aG - bG); + FfxFloat16x2 rB = (cB - bB) * ffxApproximateReciprocalMediumHalf(aB - bB); + cR = ffxSaturate(nR + ffxIsGreaterThanZeroHalf(dit - rR) * FFX_BROADCAST_FLOAT16X2(1.0 / 255.0)); + cG = ffxSaturate(nG + ffxIsGreaterThanZeroHalf(dit - rG) * FFX_BROADCAST_FLOAT16X2(1.0 / 255.0)); + cB = ffxSaturate(nB + ffxIsGreaterThanZeroHalf(dit - rB) * FFX_BROADCAST_FLOAT16X2(1.0 / 255.0)); + } + //------------------------------------------------------------------------------------------------------------------------------ + void FsrTepdC10Hx2(inout FfxFloat16x2 cR,inout FfxFloat16x2 cG,inout FfxFloat16x2 cB,FfxFloat16x2 dit){ + FfxFloat16x2 nR=sqrt(cR); + FfxFloat16x2 nG=sqrt(cG); + FfxFloat16x2 nB=sqrt(cB); + nR=floor(nR*FFX_BROADCAST_FLOAT16X2(1023.0))*FFX_BROADCAST_FLOAT16X2(1.0/1023.0); + nG=floor(nG*FFX_BROADCAST_FLOAT16X2(1023.0))*FFX_BROADCAST_FLOAT16X2(1.0/1023.0); + nB=floor(nB*FFX_BROADCAST_FLOAT16X2(1023.0))*FFX_BROADCAST_FLOAT16X2(1.0/1023.0); + FfxFloat16x2 aR=nR*nR; + FfxFloat16x2 aG=nG*nG; + FfxFloat16x2 aB=nB*nB; + FfxFloat16x2 bR=nR+FFX_BROADCAST_FLOAT16X2(1.0/1023.0);bR=bR*bR; + FfxFloat16x2 bG=nG+FFX_BROADCAST_FLOAT16X2(1.0/1023.0);bG=bG*bG; + FfxFloat16x2 bB=nB+FFX_BROADCAST_FLOAT16X2(1.0/1023.0);bB=bB*bB; + FfxFloat16x2 rR=(cR-bR)*ffxApproximateReciprocalMediumHalf(aR-bR); + FfxFloat16x2 rG=(cG-bG)*ffxApproximateReciprocalMediumHalf(aG-bG); + FfxFloat16x2 rB=(cB-bB)*ffxApproximateReciprocalMediumHalf(aB-bB); + cR=ffxSaturate(nR+ffxIsGreaterThanZeroHalf(dit-rR)*FFX_BROADCAST_FLOAT16X2(1.0/1023.0)); + cG=ffxSaturate(nG+ffxIsGreaterThanZeroHalf(dit-rG)*FFX_BROADCAST_FLOAT16X2(1.0/1023.0)); + cB = ffxSaturate(nB + ffxIsGreaterThanZeroHalf(dit - rB) * FFX_BROADCAST_FLOAT16X2(1.0 / 1023.0)); +} +#endif diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate.h new file mode 100644 index 000000000000..7bd5892cb90f --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate.h @@ -0,0 +1,295 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_ACCUMULATE_H +#define FFX_FSR2_ACCUMULATE_H + +FfxFloat32 GetPxHrVelocity(FfxFloat32x2 fMotionVector) +{ + return length(fMotionVector * DisplaySize()); +} +#if FFX_HALF +FFX_MIN16_F GetPxHrVelocity(FFX_MIN16_F2 fMotionVector) +{ + return length(fMotionVector * FFX_MIN16_F2(DisplaySize())); +} +#endif + +void Accumulate(const AccumulationPassCommonParams params, FFX_PARAMETER_INOUT FfxFloat32x3 fHistoryColor, FfxFloat32x3 fAccumulation, FFX_PARAMETER_IN FfxFloat32x4 fUpsampledColorAndWeight) +{ + // Aviod invalid values when accumulation and upsampled weight is 0 + fAccumulation = ffxMax(FSR2_EPSILON.xxx, fAccumulation + fUpsampledColorAndWeight.www); + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + //YCoCg -> RGB -> Tonemap -> YCoCg (Use RGB tonemapper to avoid color desaturation) + fUpsampledColorAndWeight.xyz = RGBToYCoCg(Tonemap(YCoCgToRGB(fUpsampledColorAndWeight.xyz))); + fHistoryColor = RGBToYCoCg(Tonemap(YCoCgToRGB(fHistoryColor))); +#endif + + const FfxFloat32x3 fAlpha = fUpsampledColorAndWeight.www / fAccumulation; + fHistoryColor = ffxLerp(fHistoryColor, fUpsampledColorAndWeight.xyz, fAlpha); + + fHistoryColor = YCoCgToRGB(fHistoryColor); + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + fHistoryColor = InverseTonemap(fHistoryColor); +#endif +} + +void RectifyHistory( + const AccumulationPassCommonParams params, + RectificationBox clippingBox, + FFX_PARAMETER_INOUT FfxFloat32x3 fHistoryColor, + FFX_PARAMETER_INOUT FfxFloat32x3 fAccumulation, + FfxFloat32 fLockContributionThisFrame, + FfxFloat32 fTemporalReactiveFactor, + FfxFloat32 fLumaInstabilityFactor) +{ + FfxFloat32 fScaleFactorInfluence = ffxMin(20.0f, ffxPow(FfxFloat32(1.0f / length(DownscaleFactor().x * DownscaleFactor().y)), 3.0f)); + + const FfxFloat32 fVecolityFactor = ffxSaturate(params.fHrVelocity / 20.0f); + const FfxFloat32 fBoxScaleT = ffxMax(params.fDepthClipFactor, ffxMax(params.fAccumulationMask, fVecolityFactor)); + FfxFloat32 fBoxScale = ffxLerp(fScaleFactorInfluence, 1.0f, fBoxScaleT); + + FfxFloat32x3 fScaledBoxVec = clippingBox.boxVec * fBoxScale; + FfxFloat32x3 boxMin = clippingBox.boxCenter - fScaledBoxVec; + FfxFloat32x3 boxMax = clippingBox.boxCenter + fScaledBoxVec; + FfxFloat32x3 boxCenter = clippingBox.boxCenter; + FfxFloat32 boxVecSize = length(clippingBox.boxVec); + + boxMin = ffxMax(clippingBox.aabbMin, boxMin); + boxMax = ffxMin(clippingBox.aabbMax, boxMax); + + if (any(FFX_GREATER_THAN(boxMin, fHistoryColor)) || any(FFX_GREATER_THAN(fHistoryColor, boxMax))) { + + const FfxFloat32x3 fClampedHistoryColor = clamp(fHistoryColor, boxMin, boxMax); + + FfxFloat32x3 fHistoryContribution = ffxMax(fLumaInstabilityFactor, fLockContributionThisFrame).xxx; + + const FfxFloat32 fReactiveFactor = params.fDilatedReactiveFactor; + const FfxFloat32 fReactiveContribution = 1.0f - ffxPow(fReactiveFactor, 1.0f / 2.0f); + fHistoryContribution *= fReactiveContribution; + + // Scale history color using rectification info, also using accumulation mask to avoid potential invalid color protection + fHistoryColor = ffxLerp(fClampedHistoryColor, fHistoryColor, ffxSaturate(fHistoryContribution)); + + // Scale accumulation using rectification info + const FfxFloat32x3 fAccumulationMin = ffxMin(fAccumulation, FFX_BROADCAST_FLOAT32X3(0.1f)); + fAccumulation = ffxLerp(fAccumulationMin, fAccumulation, ffxSaturate(fHistoryContribution)); + } +} + +void WriteUpscaledOutput(FfxInt32x2 iPxHrPos, FfxFloat32x3 fUpscaledColor) +{ + StoreUpscaledOutput(iPxHrPos, fUpscaledColor); +} + +void FinalizeLockStatus(const AccumulationPassCommonParams params, FfxFloat32x2 fLockStatus, FfxFloat32 fUpsampledWeight) +{ + // we expect similar motion for next frame + // kill lock if that location is outside screen, avoid locks to be clamped to screen borders + FfxFloat32x2 fEstimatedUvNextFrame = params.fHrUv - params.fMotionVector; + if (IsUvInside(fEstimatedUvNextFrame) == false) { + KillLock(fLockStatus); + } + else { + // Decrease lock lifetime + const FfxFloat32 fLifetimeDecreaseLanczosMax = FfxFloat32(JitterSequenceLength()) * FfxFloat32(fAverageLanczosWeightPerFrame); + const FfxFloat32 fLifetimeDecrease = FfxFloat32(fUpsampledWeight / fLifetimeDecreaseLanczosMax); + fLockStatus[LOCK_LIFETIME_REMAINING] = ffxMax(FfxFloat32(0), fLockStatus[LOCK_LIFETIME_REMAINING] - fLifetimeDecrease); + } + + StoreLockStatus(params.iPxHrPos, fLockStatus); +} + + +FfxFloat32x3 ComputeBaseAccumulationWeight(const AccumulationPassCommonParams params, FfxFloat32 fThisFrameReactiveFactor, FfxBoolean bInMotionLastFrame, FfxFloat32 fUpsampledWeight, LockState lockState) +{ + // Always assume max accumulation was reached + FfxFloat32 fBaseAccumulation = fMaxAccumulationLanczosWeight * FfxFloat32(params.bIsExistingSample) * (1.0f - fThisFrameReactiveFactor) * (1.0f - params.fDepthClipFactor); + + fBaseAccumulation = ffxMin(fBaseAccumulation, ffxLerp(fBaseAccumulation, fUpsampledWeight * 10.0f, ffxMax(FfxFloat32(bInMotionLastFrame), ffxSaturate(params.fHrVelocity * FfxFloat32(10))))); + + fBaseAccumulation = ffxMin(fBaseAccumulation, ffxLerp(fBaseAccumulation, fUpsampledWeight, ffxSaturate(params.fHrVelocity / FfxFloat32(20)))); + + return fBaseAccumulation.xxx; +} + +FfxFloat32 ComputeLumaInstabilityFactor(const AccumulationPassCommonParams params, RectificationBox clippingBox, FfxFloat32 fThisFrameReactiveFactor, FfxFloat32 fLuminanceDiff) +{ + const FfxFloat32 fUnormThreshold = 1.0f / 255.0f; + const FfxInt32 N_MINUS_1 = 0; + const FfxInt32 N_MINUS_2 = 1; + const FfxInt32 N_MINUS_3 = 2; + const FfxInt32 N_MINUS_4 = 3; + + FfxFloat32 fCurrentFrameLuma = clippingBox.boxCenter.x; + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + fCurrentFrameLuma = fCurrentFrameLuma / (1.0f + ffxMax(0.0f, fCurrentFrameLuma)); +#endif + + fCurrentFrameLuma = round(fCurrentFrameLuma * 255.0f) / 255.0f; + + const FfxBoolean bSampleLumaHistory = (ffxMax(ffxMax(params.fDepthClipFactor, params.fAccumulationMask), fLuminanceDiff) < 0.1f) && (params.bIsNewSample == false); + FfxFloat32x4 fCurrentFrameLumaHistory = bSampleLumaHistory ? SampleLumaHistory(params.fReprojectedHrUv) : FFX_BROADCAST_FLOAT32X4(0.0f); + + FfxFloat32 fLumaInstability = 0.0f; + FfxFloat32 fDiffs0 = (fCurrentFrameLuma - fCurrentFrameLumaHistory[N_MINUS_1]); + + FfxFloat32 fMin = abs(fDiffs0); + + if (fMin >= fUnormThreshold) + { + for (int i = N_MINUS_2; i <= N_MINUS_4; i++) { + FfxFloat32 fDiffs1 = (fCurrentFrameLuma - fCurrentFrameLumaHistory[i]); + + if (sign(fDiffs0) == sign(fDiffs1)) { + + // Scale difference to protect historically similar values + const FfxFloat32 fMinBias = 1.0f; + fMin = ffxMin(fMin, abs(fDiffs1) * fMinBias); + } + } + + const FfxFloat32 fBoxSize = clippingBox.boxVec.x; + const FfxFloat32 fBoxSizeFactor = ffxPow(ffxSaturate(fBoxSize / 0.1f), 6.0f); + + fLumaInstability = FfxFloat32(fMin != abs(fDiffs0)) * fBoxSizeFactor; + fLumaInstability = FfxFloat32(fLumaInstability > fUnormThreshold); + + fLumaInstability *= 1.0f - ffxMax(params.fAccumulationMask, ffxPow(fThisFrameReactiveFactor, 1.0f / 6.0f)); + } + + //shift history + fCurrentFrameLumaHistory[N_MINUS_4] = fCurrentFrameLumaHistory[N_MINUS_3]; + fCurrentFrameLumaHistory[N_MINUS_3] = fCurrentFrameLumaHistory[N_MINUS_2]; + fCurrentFrameLumaHistory[N_MINUS_2] = fCurrentFrameLumaHistory[N_MINUS_1]; + fCurrentFrameLumaHistory[N_MINUS_1] = fCurrentFrameLuma; + + StoreLumaHistory(params.iPxHrPos, fCurrentFrameLumaHistory); + + return fLumaInstability * FfxFloat32(fCurrentFrameLumaHistory[N_MINUS_4] != 0); +} + +FfxFloat32 ComputeTemporalReactiveFactor(const AccumulationPassCommonParams params, FfxFloat32 fTemporalReactiveFactor) +{ + FfxFloat32 fNewFactor = ffxMin(0.99f, fTemporalReactiveFactor); + + fNewFactor = ffxMax(fNewFactor, ffxLerp(fNewFactor, 0.4f, ffxSaturate(params.fHrVelocity))); + + fNewFactor = ffxMax(fNewFactor * fNewFactor, ffxMax(params.fDepthClipFactor * 0.1f, params.fDilatedReactiveFactor)); + + // Force reactive factor for new samples + fNewFactor = params.bIsNewSample ? 1.0f : fNewFactor; + + if (ffxSaturate(params.fHrVelocity * 10.0f) >= 1.0f) { + fNewFactor = ffxMax(FSR2_EPSILON, fNewFactor) * -1.0f; + } + + return fNewFactor; +} + +AccumulationPassCommonParams InitParams(FfxInt32x2 iPxHrPos) +{ + AccumulationPassCommonParams params; + + params.iPxHrPos = iPxHrPos; + const FfxFloat32x2 fHrUv = (iPxHrPos + 0.5f) / DisplaySize(); + params.fHrUv = fHrUv; + + const FfxFloat32x2 fLrUvJittered = fHrUv + Jitter() / RenderSize(); + params.fLrUv_HwSampler = ClampUv(fLrUvJittered, RenderSize(), MaxRenderSize()); + + params.fMotionVector = GetMotionVector(iPxHrPos, fHrUv); + params.fHrVelocity = GetPxHrVelocity(params.fMotionVector); + + ComputeReprojectedUVs(params, params.fReprojectedHrUv, params.bIsExistingSample); + + params.fDepthClipFactor = ffxSaturate(SampleDepthClip(params.fLrUv_HwSampler)); + + const FfxFloat32x2 fDilatedReactiveMasks = SampleDilatedReactiveMasks(params.fLrUv_HwSampler); + params.fDilatedReactiveFactor = fDilatedReactiveMasks.x; + params.fAccumulationMask = fDilatedReactiveMasks.y; + params.bIsResetFrame = (0 == FrameIndex()); + + params.bIsNewSample = (params.bIsExistingSample == false || params.bIsResetFrame); + + return params; +} + +void Accumulate(FfxInt32x2 iPxHrPos) +{ + const AccumulationPassCommonParams params = InitParams(iPxHrPos); + + FfxFloat32x3 fHistoryColor = FfxFloat32x3(0, 0, 0); + FfxFloat32x2 fLockStatus; + InitializeNewLockSample(fLockStatus); + + FfxFloat32 fTemporalReactiveFactor = 0.0f; + FfxBoolean bInMotionLastFrame = FFX_FALSE; + LockState lockState = { FFX_FALSE , FFX_FALSE }; + if (params.bIsExistingSample && !params.bIsResetFrame) { + ReprojectHistoryColor(params, fHistoryColor, fTemporalReactiveFactor, bInMotionLastFrame); + lockState = ReprojectHistoryLockStatus(params, fLockStatus); + } + + FfxFloat32 fThisFrameReactiveFactor = ffxMax(params.fDilatedReactiveFactor, fTemporalReactiveFactor); + + FfxFloat32 fLuminanceDiff = 0.0f; + FfxFloat32 fLockContributionThisFrame = 0.0f; + UpdateLockStatus(params, fThisFrameReactiveFactor, lockState, fLockStatus, fLockContributionThisFrame, fLuminanceDiff); + + // Load upsampled input color + RectificationBox clippingBox; + FfxFloat32x4 fUpsampledColorAndWeight = ComputeUpsampledColorAndWeight(params, clippingBox, fThisFrameReactiveFactor); + + const FfxFloat32 fLumaInstabilityFactor = ComputeLumaInstabilityFactor(params, clippingBox, fThisFrameReactiveFactor, fLuminanceDiff); + + + FfxFloat32x3 fAccumulation = ComputeBaseAccumulationWeight(params, fThisFrameReactiveFactor, bInMotionLastFrame, fUpsampledColorAndWeight.w, lockState); + + if (params.bIsNewSample) { + fHistoryColor = YCoCgToRGB(fUpsampledColorAndWeight.xyz); + } + else { + RectifyHistory(params, clippingBox, fHistoryColor, fAccumulation, fLockContributionThisFrame, fThisFrameReactiveFactor, fLumaInstabilityFactor); + + Accumulate(params, fHistoryColor, fAccumulation, fUpsampledColorAndWeight); + } + + fHistoryColor = UnprepareRgb(fHistoryColor, Exposure()); + + FinalizeLockStatus(params, fLockStatus, fUpsampledColorAndWeight.w); + + // Get new temporal reactive factor + fTemporalReactiveFactor = ComputeTemporalReactiveFactor(params, fThisFrameReactiveFactor); + + StoreInternalColorAndWeight(iPxHrPos, FfxFloat32x4(fHistoryColor, fTemporalReactiveFactor)); + + // Output final color when RCAS is disabled +#if FFX_FSR2_OPTION_APPLY_SHARPENING == 0 + WriteUpscaledOutput(iPxHrPos, fHistoryColor); +#endif + StoreNewLocks(iPxHrPos, 0); +} + +#endif // FFX_FSR2_ACCUMULATE_H diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl new file mode 100644 index 000000000000..d2306fec4c14 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl @@ -0,0 +1,92 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require +// Needed for rw_upscaled_output declaration +#extension GL_EXT_shader_image_load_formatted : require + +#define FSR2_BIND_SRV_INPUT_EXPOSURE 0 +#define FSR2_BIND_SRV_DILATED_REACTIVE_MASKS 1 +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS +#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 2 +#else +#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 2 +#endif +#define FSR2_BIND_SRV_INTERNAL_UPSCALED 3 +#define FSR2_BIND_SRV_LOCK_STATUS 4 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 6 +#define FSR2_BIND_SRV_LUMA_INSTABILITY 7 +#define FSR2_BIND_SRV_LANCZOS_LUT 8 +#define FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT 9 +#define FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS 10 +#define FSR2_BIND_SRV_AUTO_EXPOSURE 11 +#define FSR2_BIND_SRV_LUMA_HISTORY 12 + +#define FSR2_BIND_UAV_INTERNAL_UPSCALED 13 +#define FSR2_BIND_UAV_LOCK_STATUS 14 +#define FSR2_BIND_UAV_UPSCALED_OUTPUT 15 +#define FSR2_BIND_UAV_NEW_LOCKS 16 +#define FSR2_BIND_UAV_LUMA_HISTORY 17 + +#define FSR2_BIND_CB_FSR2 18 + +// -- GODOT start -- +#if FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS +#define FSR2_BIND_SRV_INPUT_DEPTH 5 +#endif +// -- GODOT end -- + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_upsample.h" +#include "ffx_fsr2_postprocess_lock_status.h" +#include "ffx_fsr2_reproject.h" +#include "ffx_fsr2_accumulate.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; + +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + uvec2 uGroupId = gl_WorkGroupID.xy; + const uint GroupRows = (uint(DisplaySize().y) + FFX_FSR2_THREAD_GROUP_HEIGHT - 1) / FFX_FSR2_THREAD_GROUP_HEIGHT; + uGroupId.y = GroupRows - uGroupId.y - 1; + + uvec2 uDispatchThreadId = uGroupId * uvec2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + gl_LocalInvocationID.xy; + + Accumulate(ivec2(uDispatchThreadId)); +} \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl new file mode 100644 index 000000000000..e62b4459246e --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl @@ -0,0 +1,93 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_INPUT_OPAQUE_ONLY 0 +#define FSR2_BIND_SRV_INPUT_COLOR 1 +#define FSR2_BIND_UAV_AUTOREACTIVE 2 +#define FSR2_BIND_CB_REACTIVE 3 +#define FSR2_BIND_CB_FSR2 4 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +// layout (set = 1, binding = FSR2_BIND_SRV_PRE_ALPHA_COLOR) uniform texture2D r_input_color_pre_alpha; +// layout (set = 1, binding = FSR2_BIND_SRV_POST_ALPHA_COLOR) uniform texture2D r_input_color_post_alpha; +// layout (set = 1, binding = FSR2_BIND_UAV_REACTIVE, r8) uniform image2D rw_output_reactive_mask; + + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +#if defined(FSR2_BIND_CB_REACTIVE) +layout (set = 1, binding = FSR2_BIND_CB_REACTIVE, std140) uniform cbGenerateReactive_t +{ + float scale; + float threshold; + float binaryValue; + uint flags; +} cbGenerateReactive; +#endif + +FFX_FSR2_NUM_THREADS +void main() +{ + FfxUInt32x2 uDispatchThreadId = gl_GlobalInvocationID.xy; + + FfxFloat32x3 ColorPreAlpha = LoadOpaqueOnly(FFX_MIN16_I2(uDispatchThreadId)).rgb; + FfxFloat32x3 ColorPostAlpha = LoadInputColor(FFX_MIN16_I2(uDispatchThreadId)).rgb; + + if ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_TONEMAP) != 0) + { + ColorPreAlpha = Tonemap(ColorPreAlpha); + ColorPostAlpha = Tonemap(ColorPostAlpha); + } + + if ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_INVERSETONEMAP) != 0) + { + ColorPreAlpha = InverseTonemap(ColorPreAlpha); + ColorPostAlpha = InverseTonemap(ColorPostAlpha); + } + + FfxFloat32 out_reactive_value = 0.f; + FfxFloat32x3 delta = abs(ColorPostAlpha - ColorPreAlpha); + + out_reactive_value = ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_USE_COMPONENTS_MAX)!=0) ? max(delta.x, max(delta.y, delta.z)) : length(delta); + out_reactive_value *= cbGenerateReactive.scale; + + out_reactive_value = ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_THRESHOLD)!=0) ? ((out_reactive_value < cbGenerateReactive.threshold) ? 0 : cbGenerateReactive.binaryValue) : out_reactive_value; + + imageStore(rw_output_autoreactive, FfxInt32x2(uDispatchThreadId), vec4(out_reactive_value)); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_glsl.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_glsl.h new file mode 100644 index 000000000000..45279bd35744 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_glsl.h @@ -0,0 +1,704 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +#include "ffx_fsr2_resources.h" + +#if defined(FFX_GPU) +#include "ffx_core.h" +#endif // #if defined(FFX_GPU) + +#if defined(FFX_GPU) +#ifndef FFX_FSR2_PREFER_WAVE64 +#define FFX_FSR2_PREFER_WAVE64 +#endif // #if defined(FFX_GPU) + +#if defined(FSR2_BIND_CB_FSR2) + layout (set = 1, binding = FSR2_BIND_CB_FSR2, std140) uniform cbFSR2_t + { + FfxInt32x2 iRenderSize; + FfxInt32x2 iMaxRenderSize; + FfxInt32x2 iDisplaySize; + FfxInt32x2 iInputColorResourceDimensions; + FfxInt32x2 iLumaMipDimensions; + FfxInt32 iLumaMipLevelToUse; + FfxInt32 iFrameIndex; + + FfxFloat32x4 fDeviceToViewDepth; + FfxFloat32x2 fJitter; + FfxFloat32x2 fMotionVectorScale; + FfxFloat32x2 fDownscaleFactor; + FfxFloat32x2 fMotionVectorJitterCancellation; + FfxFloat32 fPreExposure; + FfxFloat32 fPreviousFramePreExposure; + FfxFloat32 fTanHalfFOV; + FfxFloat32 fJitterSequenceLength; + FfxFloat32 fDeltaTime; + FfxFloat32 fDynamicResChangeFactor; + FfxFloat32 fViewSpaceToMetersFactor; + + // -- GODOT start -- + FfxFloat32 fPad; + mat4 mReprojectionMatrix; + // -- GODOT end -- + } cbFSR2; +#endif + +FfxInt32x2 RenderSize() +{ + return cbFSR2.iRenderSize; +} + +FfxInt32x2 MaxRenderSize() +{ + return cbFSR2.iMaxRenderSize; +} + +FfxInt32x2 DisplaySize() +{ + return cbFSR2.iDisplaySize; +} + +FfxInt32x2 InputColorResourceDimensions() +{ + return cbFSR2.iInputColorResourceDimensions; +} + +FfxInt32x2 LumaMipDimensions() +{ + return cbFSR2.iLumaMipDimensions; +} + +FfxInt32 LumaMipLevelToUse() +{ + return cbFSR2.iLumaMipLevelToUse; +} + +FfxInt32 FrameIndex() +{ + return cbFSR2.iFrameIndex; +} + +FfxFloat32x4 DeviceToViewSpaceTransformFactors() +{ + return cbFSR2.fDeviceToViewDepth; +} + +FfxFloat32x2 Jitter() +{ + return cbFSR2.fJitter; +} + +FfxFloat32x2 MotionVectorScale() +{ + return cbFSR2.fMotionVectorScale; +} + +FfxFloat32x2 DownscaleFactor() +{ + return cbFSR2.fDownscaleFactor; +} + +FfxFloat32x2 MotionVectorJitterCancellation() +{ + return cbFSR2.fMotionVectorJitterCancellation; +} + +FfxFloat32 PreExposure() +{ + return cbFSR2.fPreExposure; +} + +FfxFloat32 PreviousFramePreExposure() +{ + return cbFSR2.fPreviousFramePreExposure; +} + +FfxFloat32 TanHalfFoV() +{ + return cbFSR2.fTanHalfFOV; +} + +FfxFloat32 JitterSequenceLength() +{ + return cbFSR2.fJitterSequenceLength; +} + +FfxFloat32 DeltaTime() +{ + return cbFSR2.fDeltaTime; +} + +FfxFloat32 DynamicResChangeFactor() +{ + return cbFSR2.fDynamicResChangeFactor; +} + +FfxFloat32 ViewSpaceToMetersFactor() +{ + return cbFSR2.fViewSpaceToMetersFactor; +} + +layout (set = 0, binding = 0) uniform sampler s_PointClamp; +layout (set = 0, binding = 1) uniform sampler s_LinearClamp; + +// SRVs +#if defined(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY) + layout (set = 1, binding = FSR2_BIND_SRV_INPUT_OPAQUE_ONLY) uniform texture2D r_input_opaque_only; +#endif +#if defined(FSR2_BIND_SRV_INPUT_COLOR) + layout (set = 1, binding = FSR2_BIND_SRV_INPUT_COLOR) uniform texture2D r_input_color_jittered; +#endif +#if defined(FSR2_BIND_SRV_INPUT_MOTION_VECTORS) + layout (set = 1, binding = FSR2_BIND_SRV_INPUT_MOTION_VECTORS) uniform texture2D r_input_motion_vectors; +#endif +#if defined(FSR2_BIND_SRV_INPUT_DEPTH) + layout (set = 1, binding = FSR2_BIND_SRV_INPUT_DEPTH) uniform texture2D r_input_depth; +#endif +#if defined(FSR2_BIND_SRV_INPUT_EXPOSURE) + layout (set = 1, binding = FSR2_BIND_SRV_INPUT_EXPOSURE) uniform texture2D r_input_exposure; +#endif +#if defined(FSR2_BIND_SRV_AUTO_EXPOSURE) + layout(set = 1, binding = FSR2_BIND_SRV_AUTO_EXPOSURE) uniform texture2D r_auto_exposure; +#endif +#if defined(FSR2_BIND_SRV_REACTIVE_MASK) + layout (set = 1, binding = FSR2_BIND_SRV_REACTIVE_MASK) uniform texture2D r_reactive_mask; +#endif +#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) + layout (set = 1, binding = FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) uniform texture2D r_transparency_and_composition_mask; +#endif +#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) + layout (set = 1, binding = FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) uniform utexture2D r_reconstructed_previous_nearest_depth; +#endif +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) + layout (set = 1, binding = FSR2_BIND_SRV_DILATED_MOTION_VECTORS) uniform texture2D r_dilated_motion_vectors; +#endif +#if defined (FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS) + layout(set = 1, binding = FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS) uniform texture2D r_previous_dilated_motion_vectors; +#endif +#if defined(FSR2_BIND_SRV_DILATED_DEPTH) + layout (set = 1, binding = FSR2_BIND_SRV_DILATED_DEPTH) uniform texture2D r_dilatedDepth; +#endif +#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED) + layout (set = 1, binding = FSR2_BIND_SRV_INTERNAL_UPSCALED) uniform texture2D r_internal_upscaled_color; +#endif +#if defined(FSR2_BIND_SRV_LOCK_STATUS) + layout (set = 1, binding = FSR2_BIND_SRV_LOCK_STATUS) uniform texture2D r_lock_status; +#endif +#if defined(FSR2_BIND_SRV_LOCK_INPUT_LUMA) + layout (set = 1, binding = FSR2_BIND_SRV_LOCK_INPUT_LUMA) uniform texture2D r_lock_input_luma; +#endif +#if defined(FSR2_BIND_SRV_NEW_LOCKS) + layout(set = 1, binding = FSR2_BIND_SRV_NEW_LOCKS) uniform texture2D r_new_locks; +#endif +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) + layout (set = 1, binding = FSR2_BIND_SRV_PREPARED_INPUT_COLOR) uniform texture2D r_prepared_input_color; +#endif +#if defined(FSR2_BIND_SRV_LUMA_HISTORY) + layout (set = 1, binding = FSR2_BIND_SRV_LUMA_HISTORY) uniform texture2D r_luma_history; +#endif +#if defined(FSR2_BIND_SRV_RCAS_INPUT) + layout (set = 1, binding = FSR2_BIND_SRV_RCAS_INPUT) uniform texture2D r_rcas_input; +#endif +#if defined(FSR2_BIND_SRV_LANCZOS_LUT) + layout (set = 1, binding = FSR2_BIND_SRV_LANCZOS_LUT) uniform texture2D r_lanczos_lut; +#endif +#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) + layout (set = 1, binding = FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) uniform texture2D r_imgMips; +#endif +#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) + layout (set = 1, binding = FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) uniform texture2D r_upsample_maximum_bias_lut; +#endif +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) + layout (set = 1, binding = FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) uniform texture2D r_dilated_reactive_masks; +#endif +#if defined(FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR) + layout(set = 1, binding = FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR) uniform texture2D r_input_prev_color_pre_alpha; +#endif +#if defined(FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR) + layout(set = 1, binding = FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR) uniform texture2D r_input_prev_color_post_alpha; +#endif + +// UAV +#if defined FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH + layout (set = 1, binding = FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH, r32ui) uniform uimage2D rw_reconstructed_previous_nearest_depth; +#endif +#if defined FSR2_BIND_UAV_DILATED_MOTION_VECTORS + layout (set = 1, binding = FSR2_BIND_UAV_DILATED_MOTION_VECTORS, rg16f) writeonly uniform image2D rw_dilated_motion_vectors; +#endif +#if defined FSR2_BIND_UAV_DILATED_DEPTH + layout (set = 1, binding = FSR2_BIND_UAV_DILATED_DEPTH, r16f) writeonly uniform image2D rw_dilatedDepth; +#endif +#if defined FSR2_BIND_UAV_INTERNAL_UPSCALED + layout (set = 1, binding = FSR2_BIND_UAV_INTERNAL_UPSCALED, rgba16f) writeonly uniform image2D rw_internal_upscaled_color; +#endif +#if defined FSR2_BIND_UAV_LOCK_STATUS + layout (set = 1, binding = FSR2_BIND_UAV_LOCK_STATUS, rg16f) uniform image2D rw_lock_status; +#endif +#if defined(FSR2_BIND_UAV_LOCK_INPUT_LUMA) + layout(set = 1, binding = FSR2_BIND_UAV_LOCK_INPUT_LUMA, r16f) writeonly uniform image2D rw_lock_input_luma; +#endif +#if defined FSR2_BIND_UAV_NEW_LOCKS + layout(set = 1, binding = FSR2_BIND_UAV_NEW_LOCKS, r8) uniform image2D rw_new_locks; +#endif +#if defined FSR2_BIND_UAV_PREPARED_INPUT_COLOR + layout (set = 1, binding = FSR2_BIND_UAV_PREPARED_INPUT_COLOR, rgba16) writeonly uniform image2D rw_prepared_input_color; +#endif +#if defined FSR2_BIND_UAV_LUMA_HISTORY + layout (set = 1, binding = FSR2_BIND_UAV_LUMA_HISTORY, rgba8) uniform image2D rw_luma_history; +#endif +#if defined FSR2_BIND_UAV_UPSCALED_OUTPUT + layout (set = 1, binding = FSR2_BIND_UAV_UPSCALED_OUTPUT /* app controlled format */) writeonly uniform image2D rw_upscaled_output; +#endif +#if defined FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE + layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE, r16f) coherent uniform image2D rw_img_mip_shading_change; +#endif +#if defined FSR2_BIND_UAV_EXPOSURE_MIP_5 + layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE_MIP_5, r16f) coherent uniform image2D rw_img_mip_5; +#endif +#if defined FSR2_BIND_UAV_DILATED_REACTIVE_MASKS + layout (set = 1, binding = FSR2_BIND_UAV_DILATED_REACTIVE_MASKS, rg8) writeonly uniform image2D rw_dilated_reactive_masks; +#endif +#if defined FSR2_BIND_UAV_EXPOSURE + layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE, rg32f) uniform image2D rw_exposure; +#endif +#if defined FSR2_BIND_UAV_AUTO_EXPOSURE + layout(set = 1, binding = FSR2_BIND_UAV_AUTO_EXPOSURE, rg32f) uniform image2D rw_auto_exposure; +#endif +#if defined FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC + layout (set = 1, binding = FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC, r32ui) coherent uniform uimage2D rw_spd_global_atomic; +#endif + +#if defined FSR2_BIND_UAV_AUTOREACTIVE + layout(set = 1, binding = FSR2_BIND_UAV_AUTOREACTIVE, r32f) uniform image2D rw_output_autoreactive; +#endif +#if defined FSR2_BIND_UAV_AUTOCOMPOSITION + layout(set = 1, binding = FSR2_BIND_UAV_AUTOCOMPOSITION, r32f) uniform image2D rw_output_autocomposition; +#endif +#if defined FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR + layout(set = 1, binding = FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR, r11f_g11f_b10f) uniform image2D rw_output_prev_color_pre_alpha; +#endif +#if defined FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR + layout(set = 1, binding = FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR, r11f_g11f_b10f) uniform image2D rw_output_prev_color_post_alpha; +#endif + +#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) +FfxFloat32 LoadMipLuma(FfxInt32x2 iPxPos, FfxInt32 mipLevel) +{ + return texelFetch(r_imgMips, iPxPos, FfxInt32(mipLevel)).r; +} +#endif + +#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) +FfxFloat32 SampleMipLuma(FfxFloat32x2 fUV, FfxInt32 mipLevel) +{ + return textureLod(sampler2D(r_imgMips, s_LinearClamp), fUV, FfxFloat32(mipLevel)).r; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_DEPTH) +FfxFloat32 LoadInputDepth(FfxInt32x2 iPxPos) +{ + return texelFetch(r_input_depth, iPxPos, 0).r; +} +#endif + +#if defined(FSR2_BIND_SRV_REACTIVE_MASK) +FfxFloat32 LoadReactiveMask(FfxInt32x2 iPxPos) +{ +// -- GODOT start -- +#if FFX_FSR2_OPTION_GODOT_REACTIVE_MASK_CLAMP + return min(texelFetch(r_reactive_mask, FfxInt32x2(iPxPos), 0).r, 0.9f); +#else + return texelFetch(r_reactive_mask, FfxInt32x2(iPxPos), 0).r; +#endif +// -- GODOT end -- +} +#endif + +#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) +FfxFloat32 LoadTransparencyAndCompositionMask(FfxUInt32x2 iPxPos) +{ + return texelFetch(r_transparency_and_composition_mask, FfxInt32x2(iPxPos), 0).r; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_COLOR) +FfxFloat32x3 LoadInputColor(FfxInt32x2 iPxPos) +{ + return texelFetch(r_input_color_jittered, iPxPos, 0).rgb; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_COLOR) +FfxFloat32x3 SampleInputColor(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_input_color_jittered, s_LinearClamp), fUV, 0.0f).rgb; +} +#endif + +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) +FfxFloat32x3 LoadPreparedInputColor(FfxInt32x2 iPxPos) +{ + return texelFetch(r_prepared_input_color, iPxPos, 0).xyz; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_MOTION_VECTORS) +FfxFloat32x2 LoadInputMotionVector(FfxInt32x2 iPxDilatedMotionVectorPos) +{ + FfxFloat32x2 fSrcMotionVector = texelFetch(r_input_motion_vectors, iPxDilatedMotionVectorPos, 0).xy; + +// -- GODOT start -- +#if FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS + bool bInvalidMotionVector = all(lessThanEqual(fSrcMotionVector, vec2(-1.0f, -1.0f))); + if (bInvalidMotionVector) + { + FfxFloat32 fSrcDepth = LoadInputDepth(iPxDilatedMotionVectorPos); + FfxFloat32x2 fUv = (iPxDilatedMotionVectorPos + FfxFloat32(0.5)) / RenderSize(); + fSrcMotionVector = FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS_FUNCTION(fUv, fSrcDepth, cbFSR2.mReprojectionMatrix); + } +#endif +// -- GODOT end -- + + FfxFloat32x2 fUvMotionVector = fSrcMotionVector * MotionVectorScale(); + +#if FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS + fUvMotionVector -= MotionVectorJitterCancellation(); +#endif + + return fUvMotionVector; +} +#endif + +#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED) +FfxFloat32x4 LoadHistory(FfxInt32x2 iPxHistory) +{ + return texelFetch(r_internal_upscaled_color, iPxHistory, 0); +} +#endif + +#if defined(FSR2_BIND_UAV_LUMA_HISTORY) +void StoreLumaHistory(FfxInt32x2 iPxPos, FfxFloat32x4 fLumaHistory) +{ + imageStore(rw_luma_history, FfxInt32x2(iPxPos), fLumaHistory); +} +#endif + +#if defined(FSR2_BIND_SRV_LUMA_HISTORY) +FfxFloat32x4 SampleLumaHistory(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_luma_history, s_LinearClamp), fUV, 0.0f); +} +#endif + +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) +void StoreReprojectedHistory(FfxInt32x2 iPxHistory, FfxFloat32x4 fHistory) +{ + imageStore(rw_internal_upscaled_color, iPxHistory, fHistory); +} +#endif + +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) +void StoreInternalColorAndWeight(FfxInt32x2 iPxPos, FfxFloat32x4 fColorAndWeight) +{ + imageStore(rw_internal_upscaled_color, FfxInt32x2(iPxPos), fColorAndWeight); +} +#endif + +#if defined(FSR2_BIND_UAV_UPSCALED_OUTPUT) +void StoreUpscaledOutput(FfxInt32x2 iPxPos, FfxFloat32x3 fColor) +{ + imageStore(rw_upscaled_output, FfxInt32x2(iPxPos), FfxFloat32x4(fColor, 1.f)); +} +#endif + +#if defined(FSR2_BIND_SRV_LOCK_STATUS) +FfxFloat32x2 LoadLockStatus(FfxInt32x2 iPxPos) +{ + FfxFloat32x2 fLockStatus = texelFetch(r_lock_status, iPxPos, 0).rg; + + return fLockStatus; +} +#endif + +#if defined(FSR2_BIND_UAV_LOCK_STATUS) +void StoreLockStatus(FfxInt32x2 iPxPos, FfxFloat32x2 fLockstatus) +{ + imageStore(rw_lock_status, iPxPos, vec4(fLockstatus, 0.0f, 0.0f)); +} +#endif + +#if defined(FSR2_BIND_SRV_LOCK_INPUT_LUMA) +FfxFloat32 LoadLockInputLuma(FfxInt32x2 iPxPos) +{ + return texelFetch(r_lock_input_luma, iPxPos, 0).r; +} +#endif + +#if defined(FSR2_BIND_UAV_LOCK_INPUT_LUMA) +void StoreLockInputLuma(FfxInt32x2 iPxPos, FfxFloat32 fLuma) +{ + imageStore(rw_lock_input_luma, iPxPos, vec4(fLuma, 0, 0, 0)); +} +#endif + +#if defined(FSR2_BIND_SRV_NEW_LOCKS) +FfxFloat32 LoadNewLocks(FfxInt32x2 iPxPos) +{ + return texelFetch(r_new_locks, iPxPos, 0).r; +} +#endif + +#if defined(FSR2_BIND_UAV_NEW_LOCKS) +FfxFloat32 LoadRwNewLocks(FfxInt32x2 iPxPos) +{ + return imageLoad(rw_new_locks, iPxPos).r; +} +#endif + +#if defined(FSR2_BIND_UAV_NEW_LOCKS) +void StoreNewLocks(FfxInt32x2 iPxPos, FfxFloat32 newLock) +{ + imageStore(rw_new_locks, iPxPos, vec4(newLock, 0, 0, 0)); +} +#endif + +#if defined(FSR2_BIND_UAV_PREPARED_INPUT_COLOR) +void StorePreparedInputColor(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x4 fTonemapped) +{ + imageStore(rw_prepared_input_color, iPxPos, fTonemapped); +} +#endif + +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) +FfxFloat32 SampleDepthClip(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_prepared_input_color, s_LinearClamp), fUV, 0.0f).w; +} +#endif + +#if defined(FSR2_BIND_SRV_LOCK_STATUS) +FfxFloat32x2 SampleLockStatus(FfxFloat32x2 fUV) +{ + FfxFloat32x2 fLockStatus = textureLod(sampler2D(r_lock_status, s_LinearClamp), fUV, 0.0f).rg; + return fLockStatus; +} +#endif + +#if defined(FSR2_BIND_SRV_DEPTH) +FfxFloat32 LoadSceneDepth(FfxInt32x2 iPxInput) +{ + return texelFetch(r_input_depth, iPxInput, 0).r; +} +#endif + +#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) +FfxFloat32 LoadReconstructedPrevDepth(FfxInt32x2 iPxPos) +{ + return uintBitsToFloat(texelFetch(r_reconstructed_previous_nearest_depth, iPxPos, 0).r); +} +#endif + +#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) +void StoreReconstructedDepth(FfxInt32x2 iPxSample, FfxFloat32 fDepth) +{ + FfxUInt32 uDepth = floatBitsToUint(fDepth); + + #if FFX_FSR2_OPTION_INVERTED_DEPTH + imageAtomicMax(rw_reconstructed_previous_nearest_depth, iPxSample, uDepth); + #else + imageAtomicMin(rw_reconstructed_previous_nearest_depth, iPxSample, uDepth); // min for standard, max for inverted depth + #endif +} +#endif + +#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) +void SetReconstructedDepth(FfxInt32x2 iPxSample, FfxUInt32 uValue) +{ + imageStore(rw_reconstructed_previous_nearest_depth, iPxSample, uvec4(uValue, 0, 0, 0)); +} +#endif + +#if defined(FSR2_BIND_UAV_DILATED_DEPTH) +void StoreDilatedDepth(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32 fDepth) +{ + //FfxUInt32 uDepth = f32tof16(fDepth); + imageStore(rw_dilatedDepth, iPxPos, vec4(fDepth, 0.0f, 0.0f, 0.0f)); +} +#endif + +#if defined(FSR2_BIND_UAV_DILATED_MOTION_VECTORS) +void StoreDilatedMotionVector(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fMotionVector) +{ + imageStore(rw_dilated_motion_vectors, iPxPos, vec4(fMotionVector, 0.0f, 0.0f)); +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) +FfxFloat32x2 LoadDilatedMotionVector(FfxInt32x2 iPxInput) +{ + return texelFetch(r_dilated_motion_vectors, iPxInput, 0).rg; +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) +FfxFloat32x2 SampleDilatedMotionVector(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_dilated_motion_vectors, s_LinearClamp), fUV, 0.0f).rg; +} +#endif + +#if defined(FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS) +FfxFloat32x2 LoadPreviousDilatedMotionVector(FfxInt32x2 iPxInput) +{ + return texelFetch(r_previous_dilated_motion_vectors, iPxInput, 0).rg; +} + +FfxFloat32x2 SamplePreviousDilatedMotionVector(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_previous_dilated_motion_vectors, s_LinearClamp), fUV, 0.0f).xy; +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_DEPTH) +FfxFloat32 LoadDilatedDepth(FfxInt32x2 iPxInput) +{ + return texelFetch(r_dilatedDepth, iPxInput, 0).r; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_EXPOSURE) +FfxFloat32 Exposure() +{ + FfxFloat32 exposure = texelFetch(r_input_exposure, FfxInt32x2(0, 0), 0).x; + + if (exposure == 0.0f) { + exposure = 1.0f; + } + + return exposure; +} +#endif + +#if defined(FSR2_BIND_SRV_AUTO_EXPOSURE) +FfxFloat32 AutoExposure() +{ + FfxFloat32 exposure = texelFetch(r_auto_exposure, FfxInt32x2(0, 0), 0).x; + + if (exposure == 0.0f) { + exposure = 1.0f; + } + + return exposure; +} +#endif + +FfxFloat32 SampleLanczos2Weight(FfxFloat32 x) +{ +#if defined(FSR2_BIND_SRV_LANCZOS_LUT) + return textureLod(sampler2D(r_lanczos_lut, s_LinearClamp), FfxFloat32x2(x / 2.0f, 0.5f), 0.0f).x; +#else + return 0.f; +#endif +} + +#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) +FfxFloat32 SampleUpsampleMaximumBias(FfxFloat32x2 uv) +{ + // Stored as a SNORM, so make sure to multiply by 2 to retrieve the actual expected range. + return FfxFloat32(2.0f) * FfxFloat32(textureLod(sampler2D(r_upsample_maximum_bias_lut, s_LinearClamp), abs(uv) * 2.0f, 0.0f).r); +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) +FfxFloat32x2 SampleDilatedReactiveMasks(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_dilated_reactive_masks, s_LinearClamp), fUV, 0.0f).rg; +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) +FfxFloat32x2 LoadDilatedReactiveMasks(FFX_PARAMETER_IN FfxInt32x2 iPxPos) +{ + return texelFetch(r_dilated_reactive_masks, iPxPos, 0).rg; +} +#endif + +#if defined(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS) +void StoreDilatedReactiveMasks(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fDilatedReactiveMasks) +{ + imageStore(rw_dilated_reactive_masks, iPxPos, vec4(fDilatedReactiveMasks, 0.0f, 0.0f)); +} +#endif + +#if defined(FFX_INTERNAL) +FfxFloat32x4 SampleDebug(FfxFloat32x2 fUV) +{ + return textureLod(sampler2D(r_debug_out, s_LinearClamp), fUV, 0.0f).rgba; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY) +FfxFloat32x3 LoadOpaqueOnly(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos) +{ + return texelFetch(r_input_opaque_only, iPxPos, 0).xyz; +} +#endif + +#if defined(FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR) +FfxFloat32x3 LoadPrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos) +{ + return texelFetch(r_input_prev_color_pre_alpha, iPxPos, 0).xyz; +} +#endif + +#if defined(FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR) +FfxFloat32x3 LoadPrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos) +{ + return texelFetch(r_input_prev_color_post_alpha, iPxPos, 0).xyz; +} +#endif + +#if defined(FSR2_BIND_UAV_AUTOREACTIVE) +#if defined(FSR2_BIND_UAV_AUTOCOMPOSITION) +void StoreAutoReactive(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F2 fReactive) +{ + imageStore(rw_output_autoreactive, iPxPos, vec4(FfxFloat32(fReactive.x), 0.0f, 0.0f, 0.0f)); + + imageStore(rw_output_autocomposition, iPxPos, vec4(FfxFloat32(fReactive.y), 0.0f, 0.0f, 0.0f)); +} +#endif +#endif + +#if defined(FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR) +void StorePrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color) +{ + imageStore(rw_output_prev_color_pre_alpha, iPxPos, vec4(color, 0.0f)); +} +#endif + +#if defined(FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR) +void StorePrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color) +{ + imageStore(rw_output_prev_color_post_alpha, iPxPos, vec4(color, 0.0f)); +} +#endif + +#endif // #if defined(FFX_GPU) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_hlsl.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_hlsl.h new file mode 100644 index 000000000000..fd722b307e2f --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_hlsl.h @@ -0,0 +1,799 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#include "ffx_fsr2_resources.h" + +#if defined(FFX_GPU) +#ifdef __hlsl_dx_compiler +#pragma dxc diagnostic push +#pragma dxc diagnostic ignored "-Wambig-lit-shift" +#endif //__hlsl_dx_compiler +#include "ffx_core.h" +#ifdef __hlsl_dx_compiler +#pragma dxc diagnostic pop +#endif //__hlsl_dx_compiler +#endif // #if defined(FFX_GPU) + +#if defined(FFX_GPU) +#ifndef FFX_FSR2_PREFER_WAVE64 +#define FFX_FSR2_PREFER_WAVE64 +#endif // #if defined(FFX_GPU) + +#if defined(FFX_GPU) +#pragma warning(disable: 3205) // conversion from larger type to smaller +#endif // #if defined(FFX_GPU) + +#define DECLARE_SRV_REGISTER(regIndex) t##regIndex +#define DECLARE_UAV_REGISTER(regIndex) u##regIndex +#define DECLARE_CB_REGISTER(regIndex) b##regIndex +#define FFX_FSR2_DECLARE_SRV(regIndex) register(DECLARE_SRV_REGISTER(regIndex)) +#define FFX_FSR2_DECLARE_UAV(regIndex) register(DECLARE_UAV_REGISTER(regIndex)) +#define FFX_FSR2_DECLARE_CB(regIndex) register(DECLARE_CB_REGISTER(regIndex)) + +#if defined(FSR2_BIND_CB_FSR2) || defined(FFX_INTERNAL) + cbuffer cbFSR2 : FFX_FSR2_DECLARE_CB(FSR2_BIND_CB_FSR2) + { + FfxInt32x2 iRenderSize; + FfxInt32x2 iMaxRenderSize; + FfxInt32x2 iDisplaySize; + FfxInt32x2 iInputColorResourceDimensions; + FfxInt32x2 iLumaMipDimensions; + FfxInt32 iLumaMipLevelToUse; + FfxInt32 iFrameIndex; + + FfxFloat32x4 fDeviceToViewDepth; + FfxFloat32x2 fJitter; + FfxFloat32x2 fMotionVectorScale; + FfxFloat32x2 fDownscaleFactor; + FfxFloat32x2 fMotionVectorJitterCancellation; + FfxFloat32 fPreExposure; + FfxFloat32 fPreviousFramePreExposure; + FfxFloat32 fTanHalfFOV; + FfxFloat32 fJitterSequenceLength; + FfxFloat32 fDeltaTime; + FfxFloat32 fDynamicResChangeFactor; + FfxFloat32 fViewSpaceToMetersFactor; + }; + +#define FFX_FSR2_CONSTANT_BUFFER_1_SIZE (sizeof(cbFSR2) / 4) // Number of 32-bit values. This must be kept in sync with the cbFSR2 size. +#endif + +#if defined(FFX_GPU) +#define FFX_FSR2_ROOTSIG_STRINGIFY(p) FFX_FSR2_ROOTSIG_STR(p) +#define FFX_FSR2_ROOTSIG_STR(p) #p +#define FFX_FSR2_ROOTSIG [RootSignature( "DescriptorTable(UAV(u0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \ + "DescriptorTable(SRV(t0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \ + "RootConstants(num32BitConstants=" FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_CONSTANT_BUFFER_1_SIZE) ", b0), " \ + "StaticSampler(s0, filter = FILTER_MIN_MAG_MIP_POINT, " \ + "addressU = TEXTURE_ADDRESS_CLAMP, " \ + "addressV = TEXTURE_ADDRESS_CLAMP, " \ + "addressW = TEXTURE_ADDRESS_CLAMP, " \ + "comparisonFunc = COMPARISON_NEVER, " \ + "borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK), " \ + "StaticSampler(s1, filter = FILTER_MIN_MAG_MIP_LINEAR, " \ + "addressU = TEXTURE_ADDRESS_CLAMP, " \ + "addressV = TEXTURE_ADDRESS_CLAMP, " \ + "addressW = TEXTURE_ADDRESS_CLAMP, " \ + "comparisonFunc = COMPARISON_NEVER, " \ + "borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK)" )] + +#define FFX_FSR2_CONSTANT_BUFFER_2_SIZE 6 // Number of 32-bit values. This must be kept in sync with max( cbRCAS , cbSPD) size. + +#define FFX_FSR2_CB2_ROOTSIG [RootSignature( "DescriptorTable(UAV(u0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \ + "DescriptorTable(SRV(t0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \ + "RootConstants(num32BitConstants=" FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_CONSTANT_BUFFER_1_SIZE) ", b0), " \ + "RootConstants(num32BitConstants=" FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_CONSTANT_BUFFER_2_SIZE) ", b1), " \ + "StaticSampler(s0, filter = FILTER_MIN_MAG_MIP_POINT, " \ + "addressU = TEXTURE_ADDRESS_CLAMP, " \ + "addressV = TEXTURE_ADDRESS_CLAMP, " \ + "addressW = TEXTURE_ADDRESS_CLAMP, " \ + "comparisonFunc = COMPARISON_NEVER, " \ + "borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK), " \ + "StaticSampler(s1, filter = FILTER_MIN_MAG_MIP_LINEAR, " \ + "addressU = TEXTURE_ADDRESS_CLAMP, " \ + "addressV = TEXTURE_ADDRESS_CLAMP, " \ + "addressW = TEXTURE_ADDRESS_CLAMP, " \ + "comparisonFunc = COMPARISON_NEVER, " \ + "borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK)" )] +#if defined(FFX_FSR2_EMBED_ROOTSIG) +#define FFX_FSR2_EMBED_ROOTSIG_CONTENT FFX_FSR2_ROOTSIG +#define FFX_FSR2_EMBED_CB2_ROOTSIG_CONTENT FFX_FSR2_CB2_ROOTSIG +#else +#define FFX_FSR2_EMBED_ROOTSIG_CONTENT +#define FFX_FSR2_EMBED_CB2_ROOTSIG_CONTENT +#endif // #if FFX_FSR2_EMBED_ROOTSIG +#endif // #if defined(FFX_GPU) + +/* Define getter functions in the order they are defined in the CB! */ +FfxInt32x2 RenderSize() +{ + return iRenderSize; +} + +FfxInt32x2 MaxRenderSize() +{ + return iMaxRenderSize; +} + +FfxInt32x2 DisplaySize() +{ + return iDisplaySize; +} + +FfxInt32x2 InputColorResourceDimensions() +{ + return iInputColorResourceDimensions; +} + +FfxInt32x2 LumaMipDimensions() +{ + return iLumaMipDimensions; +} + +FfxInt32 LumaMipLevelToUse() +{ + return iLumaMipLevelToUse; +} + +FfxInt32 FrameIndex() +{ + return iFrameIndex; +} + +FfxFloat32x2 Jitter() +{ + return fJitter; +} + +FfxFloat32x4 DeviceToViewSpaceTransformFactors() +{ + return fDeviceToViewDepth; +} + +FfxFloat32x2 MotionVectorScale() +{ + return fMotionVectorScale; +} + +FfxFloat32x2 DownscaleFactor() +{ + return fDownscaleFactor; +} + +FfxFloat32x2 MotionVectorJitterCancellation() +{ + return fMotionVectorJitterCancellation; +} + +FfxFloat32 PreExposure() +{ + return fPreExposure; +} + +FfxFloat32 PreviousFramePreExposure() +{ + return fPreviousFramePreExposure; +} + +FfxFloat32 TanHalfFoV() +{ + return fTanHalfFOV; +} + +FfxFloat32 JitterSequenceLength() +{ + return fJitterSequenceLength; +} + +FfxFloat32 DeltaTime() +{ + return fDeltaTime; +} + +FfxFloat32 DynamicResChangeFactor() +{ + return fDynamicResChangeFactor; +} + +FfxFloat32 ViewSpaceToMetersFactor() +{ + return fViewSpaceToMetersFactor; +} + + +SamplerState s_PointClamp : register(s0); +SamplerState s_LinearClamp : register(s1); + +// SRVs +#if defined(FFX_INTERNAL) + Texture2D r_input_opaque_only : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY); + Texture2D r_input_color_jittered : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR); + Texture2D r_input_motion_vectors : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS); + Texture2D r_input_depth : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH); + Texture2D r_input_exposure : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE); + Texture2D r_auto_exposure : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE); + Texture2D r_reactive_mask : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK); + Texture2D r_transparency_and_composition_mask : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK); + Texture2D r_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH); + Texture2D r_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS); + Texture2D r_previous_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREVIOUS_DILATED_MOTION_VECTORS); + Texture2D r_dilatedDepth : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH); + Texture2D r_internal_upscaled_color : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR); + Texture2D r_lock_status : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS); + Texture2D r_lock_input_luma : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA); + Texture2D r_new_locks : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS); + Texture2D r_prepared_input_color : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR); + Texture2D r_luma_history : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY); + Texture2D r_rcas_input : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT); + Texture2D r_lanczos_lut : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LANCZOS_LUT); + Texture2D r_imgMips : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE); + Texture2D r_upsample_maximum_bias_lut : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTITIER_UPSAMPLE_MAXIMUM_BIAS_LUT); + Texture2D r_dilated_reactive_masks : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS); + Texture2D r_input_prev_color_pre_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR); + Texture2D r_input_prev_color_post_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR); + + Texture2D r_debug_out : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DEBUG_OUTPUT); + + // UAV declarations + RWTexture2D rw_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH); + RWTexture2D rw_dilated_motion_vectors : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS); + RWTexture2D rw_dilatedDepth : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH); + RWTexture2D rw_internal_upscaled_color : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR); + RWTexture2D rw_lock_status : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS); + RWTexture2D rw_lock_input_luma : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA); + RWTexture2D rw_new_locks : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS); + RWTexture2D rw_prepared_input_color : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR); + RWTexture2D rw_luma_history : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY); + RWTexture2D rw_upscaled_output : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT); + + globallycoherent RWTexture2D rw_img_mip_shading_change : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE); + globallycoherent RWTexture2D rw_img_mip_5 : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_5); + RWTexture2D rw_dilated_reactive_masks : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS); + RWTexture2D rw_auto_exposure : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE); + globallycoherent RWTexture2D rw_spd_global_atomic : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_SPD_ATOMIC_COUNT); + RWTexture2D rw_debug_out : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DEBUG_OUTPUT); + + RWTexture2D rw_output_autoreactive : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE); + RWTexture2D rw_output_autocomposition : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION); + RWTexture2D rw_output_prev_color_pre_alpha : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR); + RWTexture2D rw_output_prev_color_post_alpha : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR); + +#else // #if defined(FFX_INTERNAL) + #if defined FSR2_BIND_SRV_INPUT_COLOR + Texture2D r_input_color_jittered : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_COLOR); + #endif + #if defined FSR2_BIND_SRV_INPUT_OPAQUE_ONLY + Texture2D r_input_opaque_only : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY); + #endif + #if defined FSR2_BIND_SRV_INPUT_MOTION_VECTORS + Texture2D r_input_motion_vectors : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_MOTION_VECTORS); + #endif + #if defined FSR2_BIND_SRV_INPUT_DEPTH + Texture2D r_input_depth : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_DEPTH); + #endif + #if defined FSR2_BIND_SRV_INPUT_EXPOSURE + Texture2D r_input_exposure : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_EXPOSURE); + #endif + #if defined FSR2_BIND_SRV_AUTO_EXPOSURE + Texture2D r_auto_exposure : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_AUTO_EXPOSURE); + #endif + #if defined FSR2_BIND_SRV_REACTIVE_MASK + Texture2D r_reactive_mask : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_REACTIVE_MASK); + #endif + #if defined FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK + Texture2D r_transparency_and_composition_mask : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK); + #endif + #if defined FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH + Texture2D r_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH); + #endif + #if defined FSR2_BIND_SRV_DILATED_MOTION_VECTORS + Texture2D r_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_DILATED_MOTION_VECTORS); + #endif + #if defined FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS + Texture2D r_previous_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS); + #endif + #if defined FSR2_BIND_SRV_DILATED_DEPTH + Texture2D r_dilatedDepth : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_DILATED_DEPTH); + #endif + #if defined FSR2_BIND_SRV_INTERNAL_UPSCALED + Texture2D r_internal_upscaled_color : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INTERNAL_UPSCALED); + #endif + #if defined FSR2_BIND_SRV_LOCK_STATUS + Texture2D r_lock_status : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LOCK_STATUS); + #endif + #if defined FSR2_BIND_SRV_LOCK_INPUT_LUMA + Texture2D r_lock_input_luma : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LOCK_INPUT_LUMA); + #endif + #if defined FSR2_BIND_SRV_NEW_LOCKS + Texture2D r_new_locks : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_NEW_LOCKS); + #endif + #if defined FSR2_BIND_SRV_PREPARED_INPUT_COLOR + Texture2D r_prepared_input_color : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_PREPARED_INPUT_COLOR); + #endif + #if defined FSR2_BIND_SRV_LUMA_HISTORY + Texture2D r_luma_history : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LUMA_HISTORY); + #endif + #if defined FSR2_BIND_SRV_RCAS_INPUT + Texture2D r_rcas_input : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_RCAS_INPUT); + #endif + #if defined FSR2_BIND_SRV_LANCZOS_LUT + Texture2D r_lanczos_lut : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LANCZOS_LUT); + #endif + #if defined FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS + Texture2D r_imgMips : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS); + #endif + #if defined FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT + Texture2D r_upsample_maximum_bias_lut : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT); + #endif + #if defined FSR2_BIND_SRV_DILATED_REACTIVE_MASKS + Texture2D r_dilated_reactive_masks : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS); + #endif + + #if defined FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR + Texture2D r_input_prev_color_pre_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR); + #endif + #if defined FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR + Texture2D r_input_prev_color_post_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR); + #endif + + // UAV declarations + #if defined FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH + RWTexture2D rw_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH); + #endif + #if defined FSR2_BIND_UAV_DILATED_MOTION_VECTORS + RWTexture2D rw_dilated_motion_vectors : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_DILATED_MOTION_VECTORS); + #endif + #if defined FSR2_BIND_UAV_DILATED_DEPTH + RWTexture2D rw_dilatedDepth : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_DILATED_DEPTH); + #endif + #if defined FSR2_BIND_UAV_INTERNAL_UPSCALED + RWTexture2D rw_internal_upscaled_color : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_INTERNAL_UPSCALED); + #endif + #if defined FSR2_BIND_UAV_LOCK_STATUS + RWTexture2D rw_lock_status : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_LOCK_STATUS); + #endif + #if defined FSR2_BIND_UAV_LOCK_INPUT_LUMA + RWTexture2D rw_lock_input_luma : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_LOCK_INPUT_LUMA); + #endif + #if defined FSR2_BIND_UAV_NEW_LOCKS + RWTexture2D rw_new_locks : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_NEW_LOCKS); + #endif + #if defined FSR2_BIND_UAV_PREPARED_INPUT_COLOR + RWTexture2D rw_prepared_input_color : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_PREPARED_INPUT_COLOR); + #endif + #if defined FSR2_BIND_UAV_LUMA_HISTORY + RWTexture2D rw_luma_history : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_LUMA_HISTORY); + #endif + #if defined FSR2_BIND_UAV_UPSCALED_OUTPUT + RWTexture2D rw_upscaled_output : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_UPSCALED_OUTPUT); + #endif + #if defined FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE + globallycoherent RWTexture2D rw_img_mip_shading_change : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE); + #endif + #if defined FSR2_BIND_UAV_EXPOSURE_MIP_5 + globallycoherent RWTexture2D rw_img_mip_5 : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_EXPOSURE_MIP_5); + #endif + #if defined FSR2_BIND_UAV_DILATED_REACTIVE_MASKS + RWTexture2D rw_dilated_reactive_masks : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS); + #endif + #if defined FSR2_BIND_UAV_EXPOSURE + RWTexture2D rw_exposure : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_EXPOSURE); + #endif + #if defined FSR2_BIND_UAV_AUTO_EXPOSURE + RWTexture2D rw_auto_exposure : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_AUTO_EXPOSURE); + #endif + #if defined FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC + globallycoherent RWTexture2D rw_spd_global_atomic : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC); + #endif + + #if defined FSR2_BIND_UAV_AUTOREACTIVE + RWTexture2D rw_output_autoreactive : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_AUTOREACTIVE); + #endif + #if defined FSR2_BIND_UAV_AUTOCOMPOSITION + RWTexture2D rw_output_autocomposition : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_AUTOCOMPOSITION); + #endif + #if defined FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR + RWTexture2D rw_output_prev_color_pre_alpha : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR); + #endif + #if defined FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR + RWTexture2D rw_output_prev_color_post_alpha : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR); + #endif +#endif // #if defined(FFX_INTERNAL) + +#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) || defined(FFX_INTERNAL) +FfxFloat32 LoadMipLuma(FfxUInt32x2 iPxPos, FfxUInt32 mipLevel) +{ + return r_imgMips.mips[mipLevel][iPxPos]; +} +#endif + +#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) || defined(FFX_INTERNAL) +FfxFloat32 SampleMipLuma(FfxFloat32x2 fUV, FfxUInt32 mipLevel) +{ + return r_imgMips.SampleLevel(s_LinearClamp, fUV, mipLevel); +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_DEPTH) || defined(FFX_INTERNAL) +FfxFloat32 LoadInputDepth(FfxUInt32x2 iPxPos) +{ + return r_input_depth[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_DEPTH) || defined(FFX_INTERNAL) +FfxFloat32 SampleInputDepth(FfxFloat32x2 fUV) +{ + return r_input_depth.SampleLevel(s_LinearClamp, fUV, 0).x; +} +#endif + +#if defined(FSR2_BIND_SRV_REACTIVE_MASK) || defined(FFX_INTERNAL) +FfxFloat32 LoadReactiveMask(FfxUInt32x2 iPxPos) +{ + return r_reactive_mask[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) || defined(FFX_INTERNAL) +FfxFloat32 LoadTransparencyAndCompositionMask(FfxUInt32x2 iPxPos) +{ + return r_transparency_and_composition_mask[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_COLOR) || defined(FFX_INTERNAL) +FfxFloat32x3 LoadInputColor(FfxUInt32x2 iPxPos) +{ + return r_input_color_jittered[iPxPos].rgb; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_COLOR) || defined(FFX_INTERNAL) +FfxFloat32x3 SampleInputColor(FfxFloat32x2 fUV) +{ + return r_input_color_jittered.SampleLevel(s_LinearClamp, fUV, 0).rgb; +} +#endif + +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) || defined(FFX_INTERNAL) +FfxFloat32x3 LoadPreparedInputColor(FfxUInt32x2 iPxPos) +{ + return r_prepared_input_color[iPxPos].xyz; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_MOTION_VECTORS) || defined(FFX_INTERNAL) +FfxFloat32x2 LoadInputMotionVector(FfxUInt32x2 iPxDilatedMotionVectorPos) +{ + FfxFloat32x2 fSrcMotionVector = r_input_motion_vectors[iPxDilatedMotionVectorPos].xy; + + FfxFloat32x2 fUvMotionVector = fSrcMotionVector * MotionVectorScale(); + +#if FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS + fUvMotionVector -= MotionVectorJitterCancellation(); +#endif + + return fUvMotionVector; +} +#endif + +#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED) || defined(FFX_INTERNAL) +FfxFloat32x4 LoadHistory(FfxUInt32x2 iPxHistory) +{ + return r_internal_upscaled_color[iPxHistory]; +} +#endif + +#if defined(FSR2_BIND_UAV_LUMA_HISTORY) || defined(FFX_INTERNAL) +void StoreLumaHistory(FfxUInt32x2 iPxPos, FfxFloat32x4 fLumaHistory) +{ + rw_luma_history[iPxPos] = fLumaHistory; +} +#endif + +#if defined(FSR2_BIND_SRV_LUMA_HISTORY) || defined(FFX_INTERNAL) +FfxFloat32x4 SampleLumaHistory(FfxFloat32x2 fUV) +{ + return r_luma_history.SampleLevel(s_LinearClamp, fUV, 0); +} +#endif + +#if defined(FFX_INTERNAL) +FfxFloat32x4 SampleDebug(FfxFloat32x2 fUV) +{ + return r_debug_out.SampleLevel(s_LinearClamp, fUV, 0).w; +} +#endif + +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) || defined(FFX_INTERNAL) +void StoreReprojectedHistory(FfxUInt32x2 iPxHistory, FfxFloat32x4 fHistory) +{ + rw_internal_upscaled_color[iPxHistory] = fHistory; +} +#endif + +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) || defined(FFX_INTERNAL) +void StoreInternalColorAndWeight(FfxUInt32x2 iPxPos, FfxFloat32x4 fColorAndWeight) +{ + rw_internal_upscaled_color[iPxPos] = fColorAndWeight; +} +#endif + +#if defined(FSR2_BIND_UAV_UPSCALED_OUTPUT) || defined(FFX_INTERNAL) +void StoreUpscaledOutput(FfxUInt32x2 iPxPos, FfxFloat32x3 fColor) +{ + rw_upscaled_output[iPxPos] = FfxFloat32x4(fColor, 1.f); +} +#endif + +//LOCK_LIFETIME_REMAINING == 0 +//Should make LockInitialLifetime() return a const 1.0f later +#if defined(FSR2_BIND_SRV_LOCK_STATUS) || defined(FFX_INTERNAL) +FfxFloat32x2 LoadLockStatus(FfxUInt32x2 iPxPos) +{ + return r_lock_status[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_UAV_LOCK_STATUS) || defined(FFX_INTERNAL) +void StoreLockStatus(FfxUInt32x2 iPxPos, FfxFloat32x2 fLockStatus) +{ + rw_lock_status[iPxPos] = fLockStatus; +} +#endif + +#if defined(FSR2_BIND_SRV_LOCK_INPUT_LUMA) || defined(FFX_INTERNAL) +FfxFloat32 LoadLockInputLuma(FfxUInt32x2 iPxPos) +{ + return r_lock_input_luma[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_UAV_LOCK_INPUT_LUMA) || defined(FFX_INTERNAL) +void StoreLockInputLuma(FfxUInt32x2 iPxPos, FfxFloat32 fLuma) +{ + rw_lock_input_luma[iPxPos] = fLuma; +} +#endif + +#if defined(FSR2_BIND_SRV_NEW_LOCKS) || defined(FFX_INTERNAL) +FfxFloat32 LoadNewLocks(FfxUInt32x2 iPxPos) +{ + return r_new_locks[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_UAV_NEW_LOCKS) || defined(FFX_INTERNAL) +FfxFloat32 LoadRwNewLocks(FfxUInt32x2 iPxPos) +{ + return rw_new_locks[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_UAV_NEW_LOCKS) || defined(FFX_INTERNAL) +void StoreNewLocks(FfxUInt32x2 iPxPos, FfxFloat32 newLock) +{ + rw_new_locks[iPxPos] = newLock; +} +#endif + +#if defined(FSR2_BIND_UAV_PREPARED_INPUT_COLOR) || defined(FFX_INTERNAL) +void StorePreparedInputColor(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x4 fTonemapped) +{ + rw_prepared_input_color[iPxPos] = fTonemapped; +} +#endif + +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) || defined(FFX_INTERNAL) +FfxFloat32 SampleDepthClip(FfxFloat32x2 fUV) +{ + return r_prepared_input_color.SampleLevel(s_LinearClamp, fUV, 0).w; +} +#endif + +#if defined(FSR2_BIND_SRV_LOCK_STATUS) || defined(FFX_INTERNAL) +FfxFloat32x2 SampleLockStatus(FfxFloat32x2 fUV) +{ + FfxFloat32x2 fLockStatus = r_lock_status.SampleLevel(s_LinearClamp, fUV, 0); + return fLockStatus; +} +#endif + +#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) || defined(FFX_INTERNAL) +FfxFloat32 LoadReconstructedPrevDepth(FfxUInt32x2 iPxPos) +{ + return asfloat(r_reconstructed_previous_nearest_depth[iPxPos]); +} +#endif + +#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) || defined(FFX_INTERNAL) +void StoreReconstructedDepth(FfxUInt32x2 iPxSample, FfxFloat32 fDepth) +{ + FfxUInt32 uDepth = asuint(fDepth); + + #if FFX_FSR2_OPTION_INVERTED_DEPTH + InterlockedMax(rw_reconstructed_previous_nearest_depth[iPxSample], uDepth); + #else + InterlockedMin(rw_reconstructed_previous_nearest_depth[iPxSample], uDepth); // min for standard, max for inverted depth + #endif +} +#endif + +#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) || defined(FFX_INTERNAL) +void SetReconstructedDepth(FfxUInt32x2 iPxSample, const FfxUInt32 uValue) +{ + rw_reconstructed_previous_nearest_depth[iPxSample] = uValue; +} +#endif + +#if defined(FSR2_BIND_UAV_DILATED_DEPTH) || defined(FFX_INTERNAL) +void StoreDilatedDepth(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32 fDepth) +{ + rw_dilatedDepth[iPxPos] = fDepth; +} +#endif + +#if defined(FSR2_BIND_UAV_DILATED_MOTION_VECTORS) || defined(FFX_INTERNAL) +void StoreDilatedMotionVector(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fMotionVector) +{ + rw_dilated_motion_vectors[iPxPos] = fMotionVector; +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) || defined(FFX_INTERNAL) +FfxFloat32x2 LoadDilatedMotionVector(FfxUInt32x2 iPxInput) +{ + return r_dilated_motion_vectors[iPxInput].xy; +} +#endif + +#if defined(FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS) || defined(FFX_INTERNAL) +FfxFloat32x2 LoadPreviousDilatedMotionVector(FfxUInt32x2 iPxInput) +{ + return r_previous_dilated_motion_vectors[iPxInput].xy; +} + +FfxFloat32x2 SamplePreviousDilatedMotionVector(FfxFloat32x2 uv) +{ + return r_previous_dilated_motion_vectors.SampleLevel(s_LinearClamp, uv, 0).xy; +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_DEPTH) || defined(FFX_INTERNAL) +FfxFloat32 LoadDilatedDepth(FfxUInt32x2 iPxInput) +{ + return r_dilatedDepth[iPxInput]; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_EXPOSURE) || defined(FFX_INTERNAL) +FfxFloat32 Exposure() +{ + FfxFloat32 exposure = r_input_exposure[FfxUInt32x2(0, 0)].x; + + if (exposure == 0.0f) { + exposure = 1.0f; + } + + return exposure; +} +#endif + +#if defined(FSR2_BIND_SRV_AUTO_EXPOSURE) || defined(FFX_INTERNAL) +FfxFloat32 AutoExposure() +{ + FfxFloat32 exposure = r_auto_exposure[FfxUInt32x2(0, 0)].x; + + if (exposure == 0.0f) { + exposure = 1.0f; + } + + return exposure; +} +#endif + +FfxFloat32 SampleLanczos2Weight(FfxFloat32 x) +{ +#if defined(FSR2_BIND_SRV_LANCZOS_LUT) || defined(FFX_INTERNAL) + return r_lanczos_lut.SampleLevel(s_LinearClamp, FfxFloat32x2(x / 2, 0.5f), 0); +#else + return 0.f; +#endif +} + +#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) || defined(FFX_INTERNAL) +FfxFloat32 SampleUpsampleMaximumBias(FfxFloat32x2 uv) +{ + // Stored as a SNORM, so make sure to multiply by 2 to retrieve the actual expected range. + return FfxFloat32(2.0) * r_upsample_maximum_bias_lut.SampleLevel(s_LinearClamp, abs(uv) * 2.0, 0); +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) || defined(FFX_INTERNAL) +FfxFloat32x2 SampleDilatedReactiveMasks(FfxFloat32x2 fUV) +{ + return r_dilated_reactive_masks.SampleLevel(s_LinearClamp, fUV, 0); +} +#endif + +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) || defined(FFX_INTERNAL) +FfxFloat32x2 LoadDilatedReactiveMasks(FFX_PARAMETER_IN FfxUInt32x2 iPxPos) +{ + return r_dilated_reactive_masks[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS) || defined(FFX_INTERNAL) +void StoreDilatedReactiveMasks(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fDilatedReactiveMasks) +{ + rw_dilated_reactive_masks[iPxPos] = fDilatedReactiveMasks; +} +#endif + +#if defined(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY) || defined(FFX_INTERNAL) +FfxFloat32x3 LoadOpaqueOnly(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos) +{ + return r_input_opaque_only[iPxPos].xyz; +} +#endif + +#if defined(FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR) || defined(FFX_INTERNAL) +FfxFloat32x3 LoadPrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos) +{ + return r_input_prev_color_pre_alpha[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR) || defined(FFX_INTERNAL) +FfxFloat32x3 LoadPrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos) +{ + return r_input_prev_color_post_alpha[iPxPos]; +} +#endif + +#if defined(FSR2_BIND_UAV_AUTOREACTIVE) || defined(FFX_INTERNAL) +#if defined(FSR2_BIND_UAV_AUTOCOMPOSITION) || defined(FFX_INTERNAL) +void StoreAutoReactive(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F2 fReactive) +{ + rw_output_autoreactive[iPxPos] = fReactive.x; + + rw_output_autocomposition[iPxPos] = fReactive.y; +} +#endif +#endif + +#if defined(FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR) || defined(FFX_INTERNAL) +void StorePrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color) +{ + rw_output_prev_color_pre_alpha[iPxPos] = color; + +} +#endif + +#if defined(FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR) || defined(FFX_INTERNAL) +void StorePrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color) +{ + rw_output_prev_color_post_alpha[iPxPos] = color; +} +#endif + +#endif // #if defined(FFX_GPU) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_common.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_common.h new file mode 100644 index 000000000000..0c72aa84943e --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_common.h @@ -0,0 +1,565 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#if !defined(FFX_FSR2_COMMON_H) +#define FFX_FSR2_COMMON_H + +#if defined(FFX_CPU) || defined(FFX_GPU) +//Locks +#define LOCK_LIFETIME_REMAINING 0 +#define LOCK_TEMPORAL_LUMA 1 +#endif // #if defined(FFX_CPU) || defined(FFX_GPU) + +#if defined(FFX_GPU) +FFX_STATIC const FfxFloat32 FSR2_FP16_MIN = 6.10e-05f; +FFX_STATIC const FfxFloat32 FSR2_FP16_MAX = 65504.0f; +FFX_STATIC const FfxFloat32 FSR2_EPSILON = 1e-03f; +FFX_STATIC const FfxFloat32 FSR2_TONEMAP_EPSILON = 1.0f / FSR2_FP16_MAX; +FFX_STATIC const FfxFloat32 FSR2_FLT_MAX = 3.402823466e+38f; +FFX_STATIC const FfxFloat32 FSR2_FLT_MIN = 1.175494351e-38f; + +// treat vector truncation warnings as errors +#pragma warning(error: 3206) + +// suppress warnings +#pragma warning(disable: 3205) // conversion from larger type to smaller +#pragma warning(disable: 3571) // in ffxPow(f, e), f could be negative + +// Reconstructed depth usage +FFX_STATIC const FfxFloat32 fReconstructedDepthBilinearWeightThreshold = 0.01f; + +// Accumulation +FFX_STATIC const FfxFloat32 fUpsampleLanczosWeightScale = 1.0f / 12.0f; +FFX_STATIC const FfxFloat32 fMaxAccumulationLanczosWeight = 1.0f; +FFX_STATIC const FfxFloat32 fAverageLanczosWeightPerFrame = 0.74f * fUpsampleLanczosWeightScale; // Average lanczos weight for jitter accumulated samples +FFX_STATIC const FfxFloat32 fAccumulationMaxOnMotion = 3.0f * fUpsampleLanczosWeightScale; + +// Auto exposure +FFX_STATIC const FfxFloat32 resetAutoExposureAverageSmoothing = 1e8f; + +struct AccumulationPassCommonParams +{ + FfxInt32x2 iPxHrPos; + FfxFloat32x2 fHrUv; + FfxFloat32x2 fLrUv_HwSampler; + FfxFloat32x2 fMotionVector; + FfxFloat32x2 fReprojectedHrUv; + FfxFloat32 fHrVelocity; + FfxFloat32 fDepthClipFactor; + FfxFloat32 fDilatedReactiveFactor; + FfxFloat32 fAccumulationMask; + + FfxBoolean bIsResetFrame; + FfxBoolean bIsExistingSample; + FfxBoolean bIsNewSample; +}; + +struct LockState +{ + FfxBoolean NewLock; //Set for both unique new and re-locked new + FfxBoolean WasLockedPrevFrame; //Set to identify if the pixel was already locked (relock) +}; + +void InitializeNewLockSample(FFX_PARAMETER_OUT FfxFloat32x2 fLockStatus) +{ + fLockStatus = FfxFloat32x2(0, 0); +} + +#if FFX_HALF +void InitializeNewLockSample(FFX_PARAMETER_OUT FFX_MIN16_F2 fLockStatus) +{ + fLockStatus = FFX_MIN16_F2(0, 0); +} +#endif + + +void KillLock(FFX_PARAMETER_INOUT FfxFloat32x2 fLockStatus) +{ + fLockStatus[LOCK_LIFETIME_REMAINING] = 0; +} + +#if FFX_HALF +void KillLock(FFX_PARAMETER_INOUT FFX_MIN16_F2 fLockStatus) +{ + fLockStatus[LOCK_LIFETIME_REMAINING] = FFX_MIN16_F(0); +} +#endif + +struct RectificationBox +{ + FfxFloat32x3 boxCenter; + FfxFloat32x3 boxVec; + FfxFloat32x3 aabbMin; + FfxFloat32x3 aabbMax; + FfxFloat32 fBoxCenterWeight; +}; +#if FFX_HALF +struct RectificationBoxMin16 +{ + FFX_MIN16_F3 boxCenter; + FFX_MIN16_F3 boxVec; + FFX_MIN16_F3 aabbMin; + FFX_MIN16_F3 aabbMax; + FFX_MIN16_F fBoxCenterWeight; +}; +#endif + +void RectificationBoxReset(FFX_PARAMETER_INOUT RectificationBox rectificationBox) +{ + rectificationBox.fBoxCenterWeight = FfxFloat32(0); + + rectificationBox.boxCenter = FfxFloat32x3(0, 0, 0); + rectificationBox.boxVec = FfxFloat32x3(0, 0, 0); + rectificationBox.aabbMin = FfxFloat32x3(FSR2_FLT_MAX, FSR2_FLT_MAX, FSR2_FLT_MAX); + rectificationBox.aabbMax = -FfxFloat32x3(FSR2_FLT_MAX, FSR2_FLT_MAX, FSR2_FLT_MAX); +} +#if FFX_HALF +void RectificationBoxReset(FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox) +{ + rectificationBox.fBoxCenterWeight = FFX_MIN16_F(0); + + rectificationBox.boxCenter = FFX_MIN16_F3(0, 0, 0); + rectificationBox.boxVec = FFX_MIN16_F3(0, 0, 0); + rectificationBox.aabbMin = FFX_MIN16_F3(FSR2_FP16_MAX, FSR2_FP16_MAX, FSR2_FP16_MAX); + rectificationBox.aabbMax = -FFX_MIN16_F3(FSR2_FP16_MAX, FSR2_FP16_MAX, FSR2_FP16_MAX); +} +#endif + +void RectificationBoxAddInitialSample(FFX_PARAMETER_INOUT RectificationBox rectificationBox, const FfxFloat32x3 colorSample, const FfxFloat32 fSampleWeight) +{ + rectificationBox.aabbMin = colorSample; + rectificationBox.aabbMax = colorSample; + + FfxFloat32x3 weightedSample = colorSample * fSampleWeight; + rectificationBox.boxCenter = weightedSample; + rectificationBox.boxVec = colorSample * weightedSample; + rectificationBox.fBoxCenterWeight = fSampleWeight; +} + +void RectificationBoxAddSample(FfxBoolean bInitialSample, FFX_PARAMETER_INOUT RectificationBox rectificationBox, const FfxFloat32x3 colorSample, const FfxFloat32 fSampleWeight) +{ + if (bInitialSample) { + RectificationBoxAddInitialSample(rectificationBox, colorSample, fSampleWeight); + } else { + rectificationBox.aabbMin = ffxMin(rectificationBox.aabbMin, colorSample); + rectificationBox.aabbMax = ffxMax(rectificationBox.aabbMax, colorSample); + + FfxFloat32x3 weightedSample = colorSample * fSampleWeight; + rectificationBox.boxCenter += weightedSample; + rectificationBox.boxVec += colorSample * weightedSample; + rectificationBox.fBoxCenterWeight += fSampleWeight; + } +} +#if FFX_HALF +void RectificationBoxAddInitialSample(FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox, const FFX_MIN16_F3 colorSample, const FFX_MIN16_F fSampleWeight) +{ + rectificationBox.aabbMin = colorSample; + rectificationBox.aabbMax = colorSample; + + FFX_MIN16_F3 weightedSample = colorSample * fSampleWeight; + rectificationBox.boxCenter = weightedSample; + rectificationBox.boxVec = colorSample * weightedSample; + rectificationBox.fBoxCenterWeight = fSampleWeight; +} + +void RectificationBoxAddSample(FfxBoolean bInitialSample, FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox, const FFX_MIN16_F3 colorSample, const FFX_MIN16_F fSampleWeight) +{ + if (bInitialSample) { + RectificationBoxAddInitialSample(rectificationBox, colorSample, fSampleWeight); + } else { + rectificationBox.aabbMin = ffxMin(rectificationBox.aabbMin, colorSample); + rectificationBox.aabbMax = ffxMax(rectificationBox.aabbMax, colorSample); + + FFX_MIN16_F3 weightedSample = colorSample * fSampleWeight; + rectificationBox.boxCenter += weightedSample; + rectificationBox.boxVec += colorSample * weightedSample; + rectificationBox.fBoxCenterWeight += fSampleWeight; + } +} +#endif + +void RectificationBoxComputeVarianceBoxData(FFX_PARAMETER_INOUT RectificationBox rectificationBox) +{ + rectificationBox.fBoxCenterWeight = (abs(rectificationBox.fBoxCenterWeight) > FfxFloat32(FSR2_EPSILON) ? rectificationBox.fBoxCenterWeight : FfxFloat32(1.f)); + rectificationBox.boxCenter /= rectificationBox.fBoxCenterWeight; + rectificationBox.boxVec /= rectificationBox.fBoxCenterWeight; + FfxFloat32x3 stdDev = sqrt(abs(rectificationBox.boxVec - rectificationBox.boxCenter * rectificationBox.boxCenter)); + rectificationBox.boxVec = stdDev; +} +#if FFX_HALF +void RectificationBoxComputeVarianceBoxData(FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox) +{ + rectificationBox.fBoxCenterWeight = (abs(rectificationBox.fBoxCenterWeight) > FFX_MIN16_F(FSR2_EPSILON) ? rectificationBox.fBoxCenterWeight : FFX_MIN16_F(1.f)); + rectificationBox.boxCenter /= rectificationBox.fBoxCenterWeight; + rectificationBox.boxVec /= rectificationBox.fBoxCenterWeight; + FFX_MIN16_F3 stdDev = sqrt(abs(rectificationBox.boxVec - rectificationBox.boxCenter * rectificationBox.boxCenter)); + rectificationBox.boxVec = stdDev; +} +#endif + +FfxFloat32x3 SafeRcp3(FfxFloat32x3 v) +{ + return (all(FFX_NOT_EQUAL(v, FfxFloat32x3(0, 0, 0)))) ? (FfxFloat32x3(1, 1, 1) / v) : FfxFloat32x3(0, 0, 0); +} +#if FFX_HALF +FFX_MIN16_F3 SafeRcp3(FFX_MIN16_F3 v) +{ + return (all(FFX_NOT_EQUAL(v, FFX_MIN16_F3(0, 0, 0)))) ? (FFX_MIN16_F3(1, 1, 1) / v) : FFX_MIN16_F3(0, 0, 0); +} +#endif + +FfxFloat32 MinDividedByMax(const FfxFloat32 v0, const FfxFloat32 v1) +{ + const FfxFloat32 m = ffxMax(v0, v1); + return m != 0 ? ffxMin(v0, v1) / m : 0; +} + +#if FFX_HALF +FFX_MIN16_F MinDividedByMax(const FFX_MIN16_F v0, const FFX_MIN16_F v1) +{ + const FFX_MIN16_F m = ffxMax(v0, v1); + return m != FFX_MIN16_F(0) ? ffxMin(v0, v1) / m : FFX_MIN16_F(0); +} +#endif + +FfxFloat32x3 YCoCgToRGB(FfxFloat32x3 fYCoCg) +{ + FfxFloat32x3 fRgb; + + fRgb = FfxFloat32x3( + fYCoCg.x + fYCoCg.y - fYCoCg.z, + fYCoCg.x + fYCoCg.z, + fYCoCg.x - fYCoCg.y - fYCoCg.z); + + return fRgb; +} +#if FFX_HALF +FFX_MIN16_F3 YCoCgToRGB(FFX_MIN16_F3 fYCoCg) +{ + FFX_MIN16_F3 fRgb; + + fRgb = FFX_MIN16_F3( + fYCoCg.x + fYCoCg.y - fYCoCg.z, + fYCoCg.x + fYCoCg.z, + fYCoCg.x - fYCoCg.y - fYCoCg.z); + + return fRgb; +} +#endif + +FfxFloat32x3 RGBToYCoCg(FfxFloat32x3 fRgb) +{ + FfxFloat32x3 fYCoCg; + + fYCoCg = FfxFloat32x3( + 0.25f * fRgb.r + 0.5f * fRgb.g + 0.25f * fRgb.b, + 0.5f * fRgb.r - 0.5f * fRgb.b, + -0.25f * fRgb.r + 0.5f * fRgb.g - 0.25f * fRgb.b); + + return fYCoCg; +} +#if FFX_HALF +FFX_MIN16_F3 RGBToYCoCg(FFX_MIN16_F3 fRgb) +{ + FFX_MIN16_F3 fYCoCg; + + fYCoCg = FFX_MIN16_F3( + 0.25 * fRgb.r + 0.5 * fRgb.g + 0.25 * fRgb.b, + 0.5 * fRgb.r - 0.5 * fRgb.b, + -0.25 * fRgb.r + 0.5 * fRgb.g - 0.25 * fRgb.b); + + return fYCoCg; +} +#endif + +FfxFloat32 RGBToLuma(FfxFloat32x3 fLinearRgb) +{ + return dot(fLinearRgb, FfxFloat32x3(0.2126f, 0.7152f, 0.0722f)); +} +#if FFX_HALF +FFX_MIN16_F RGBToLuma(FFX_MIN16_F3 fLinearRgb) +{ + return dot(fLinearRgb, FFX_MIN16_F3(0.2126f, 0.7152f, 0.0722f)); +} +#endif + +FfxFloat32 RGBToPerceivedLuma(FfxFloat32x3 fLinearRgb) +{ + FfxFloat32 fLuminance = RGBToLuma(fLinearRgb); + + FfxFloat32 fPercievedLuminance = 0; + if (fLuminance <= 216.0f / 24389.0f) { + fPercievedLuminance = fLuminance * (24389.0f / 27.0f); + } + else { + fPercievedLuminance = ffxPow(fLuminance, 1.0f / 3.0f) * 116.0f - 16.0f; + } + + return fPercievedLuminance * 0.01f; +} +#if FFX_HALF +FFX_MIN16_F RGBToPerceivedLuma(FFX_MIN16_F3 fLinearRgb) +{ + FFX_MIN16_F fLuminance = RGBToLuma(fLinearRgb); + + FFX_MIN16_F fPercievedLuminance = FFX_MIN16_F(0); + if (fLuminance <= FFX_MIN16_F(216.0f / 24389.0f)) { + fPercievedLuminance = fLuminance * FFX_MIN16_F(24389.0f / 27.0f); + } + else { + fPercievedLuminance = ffxPow(fLuminance, FFX_MIN16_F(1.0f / 3.0f)) * FFX_MIN16_F(116.0f) - FFX_MIN16_F(16.0f); + } + + return fPercievedLuminance * FFX_MIN16_F(0.01f); +} +#endif + +FfxFloat32x3 Tonemap(FfxFloat32x3 fRgb) +{ + return fRgb / (ffxMax(ffxMax(0.f, fRgb.r), ffxMax(fRgb.g, fRgb.b)) + 1.f).xxx; +} + +FfxFloat32x3 InverseTonemap(FfxFloat32x3 fRgb) +{ + return fRgb / ffxMax(FSR2_TONEMAP_EPSILON, 1.f - ffxMax(fRgb.r, ffxMax(fRgb.g, fRgb.b))).xxx; +} + +#if FFX_HALF +FFX_MIN16_F3 Tonemap(FFX_MIN16_F3 fRgb) +{ + return fRgb / (ffxMax(ffxMax(FFX_MIN16_F(0.f), fRgb.r), ffxMax(fRgb.g, fRgb.b)) + FFX_MIN16_F(1.f)).xxx; +} + +FFX_MIN16_F3 InverseTonemap(FFX_MIN16_F3 fRgb) +{ + return fRgb / ffxMax(FFX_MIN16_F(FSR2_TONEMAP_EPSILON), FFX_MIN16_F(1.f) - ffxMax(fRgb.r, ffxMax(fRgb.g, fRgb.b))).xxx; +} +#endif + +FfxInt32x2 ClampLoad(FfxInt32x2 iPxSample, FfxInt32x2 iPxOffset, FfxInt32x2 iTextureSize) +{ + FfxInt32x2 result = iPxSample + iPxOffset; + result.x = (iPxOffset.x < 0) ? ffxMax(result.x, 0) : result.x; + result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - 1) : result.x; + result.y = (iPxOffset.y < 0) ? ffxMax(result.y, 0) : result.y; + result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - 1) : result.y; + return result; + + // return ffxMed3(iPxSample + iPxOffset, FfxInt32x2(0, 0), iTextureSize - FfxInt32x2(1, 1)); +} +#if FFX_HALF +FFX_MIN16_I2 ClampLoad(FFX_MIN16_I2 iPxSample, FFX_MIN16_I2 iPxOffset, FFX_MIN16_I2 iTextureSize) +{ + FFX_MIN16_I2 result = iPxSample + iPxOffset; + result.x = (iPxOffset.x < 0) ? ffxMax(result.x, FFX_MIN16_I(0)) : result.x; + result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - FFX_MIN16_I(1)) : result.x; + result.y = (iPxOffset.y < 0) ? ffxMax(result.y, FFX_MIN16_I(0)) : result.y; + result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - FFX_MIN16_I(1)) : result.y; + return result; + + // return ffxMed3Half(iPxSample + iPxOffset, FFX_MIN16_I2(0, 0), iTextureSize - FFX_MIN16_I2(1, 1)); +} +#endif + +FfxFloat32x2 ClampUv(FfxFloat32x2 fUv, FfxInt32x2 iTextureSize, FfxInt32x2 iResourceSize) +{ + const FfxFloat32x2 fSampleLocation = fUv * iTextureSize; + const FfxFloat32x2 fClampedLocation = ffxMax(FfxFloat32x2(0.5f, 0.5f), ffxMin(fSampleLocation, FfxFloat32x2(iTextureSize) - FfxFloat32x2(0.5f, 0.5f))); + const FfxFloat32x2 fClampedUv = fClampedLocation / FfxFloat32x2(iResourceSize); + + return fClampedUv; +} + +FfxBoolean IsOnScreen(FfxInt32x2 pos, FfxInt32x2 size) +{ + return all(FFX_LESS_THAN(FfxUInt32x2(pos), FfxUInt32x2(size))); +} +#if FFX_HALF +FfxBoolean IsOnScreen(FFX_MIN16_I2 pos, FFX_MIN16_I2 size) +{ + return all(FFX_LESS_THAN(FFX_MIN16_U2(pos), FFX_MIN16_U2(size))); +} +#endif + +FfxFloat32 ComputeAutoExposureFromLavg(FfxFloat32 Lavg) +{ + Lavg = exp(Lavg); + + const FfxFloat32 S = 100.0f; //ISO arithmetic speed + const FfxFloat32 K = 12.5f; + FfxFloat32 ExposureISO100 = log2((Lavg * S) / K); + + const FfxFloat32 q = 0.65f; + FfxFloat32 Lmax = (78.0f / (q * S)) * ffxPow(2.0f, ExposureISO100); + + return 1 / Lmax; +} +#if FFX_HALF +FFX_MIN16_F ComputeAutoExposureFromLavg(FFX_MIN16_F Lavg) +{ + Lavg = exp(Lavg); + + const FFX_MIN16_F S = FFX_MIN16_F(100.0f); //ISO arithmetic speed + const FFX_MIN16_F K = FFX_MIN16_F(12.5f); + const FFX_MIN16_F ExposureISO100 = log2((Lavg * S) / K); + + const FFX_MIN16_F q = FFX_MIN16_F(0.65f); + const FFX_MIN16_F Lmax = (FFX_MIN16_F(78.0f) / (q * S)) * ffxPow(FFX_MIN16_F(2.0f), ExposureISO100); + + return FFX_MIN16_F(1) / Lmax; +} +#endif + +FfxInt32x2 ComputeHrPosFromLrPos(FfxInt32x2 iPxLrPos) +{ + FfxFloat32x2 fSrcJitteredPos = FfxFloat32x2(iPxLrPos) + 0.5f - Jitter(); + FfxFloat32x2 fLrPosInHr = (fSrcJitteredPos / RenderSize()) * DisplaySize(); + FfxInt32x2 iPxHrPos = FfxInt32x2(floor(fLrPosInHr)); + return iPxHrPos; +} +#if FFX_HALF +FFX_MIN16_I2 ComputeHrPosFromLrPos(FFX_MIN16_I2 iPxLrPos) +{ + FFX_MIN16_F2 fSrcJitteredPos = FFX_MIN16_F2(iPxLrPos) + FFX_MIN16_F(0.5f) - FFX_MIN16_F2(Jitter()); + FFX_MIN16_F2 fLrPosInHr = (fSrcJitteredPos / FFX_MIN16_F2(RenderSize())) * FFX_MIN16_F2(DisplaySize()); + FFX_MIN16_I2 iPxHrPos = FFX_MIN16_I2(floor(fLrPosInHr)); + return iPxHrPos; +} +#endif + +FfxFloat32x2 ComputeNdc(FfxFloat32x2 fPxPos, FfxInt32x2 iSize) +{ + return fPxPos / FfxFloat32x2(iSize) * FfxFloat32x2(2.0f, -2.0f) + FfxFloat32x2(-1.0f, 1.0f); +} + +FfxFloat32 GetViewSpaceDepth(FfxFloat32 fDeviceDepth) +{ + const FfxFloat32x4 fDeviceToViewDepth = DeviceToViewSpaceTransformFactors(); + + // fDeviceToViewDepth details found in ffx_fsr2.cpp + return (fDeviceToViewDepth[1] / (fDeviceDepth - fDeviceToViewDepth[0])); +} + +FfxFloat32 GetViewSpaceDepthInMeters(FfxFloat32 fDeviceDepth) +{ + return GetViewSpaceDepth(fDeviceDepth) * ViewSpaceToMetersFactor(); +} + +FfxFloat32x3 GetViewSpacePosition(FfxInt32x2 iViewportPos, FfxInt32x2 iViewportSize, FfxFloat32 fDeviceDepth) +{ + const FfxFloat32x4 fDeviceToViewDepth = DeviceToViewSpaceTransformFactors(); + + const FfxFloat32 Z = GetViewSpaceDepth(fDeviceDepth); + + const FfxFloat32x2 fNdcPos = ComputeNdc(iViewportPos, iViewportSize); + const FfxFloat32 X = fDeviceToViewDepth[2] * fNdcPos.x * Z; + const FfxFloat32 Y = fDeviceToViewDepth[3] * fNdcPos.y * Z; + + return FfxFloat32x3(X, Y, Z); +} + +FfxFloat32x3 GetViewSpacePositionInMeters(FfxInt32x2 iViewportPos, FfxInt32x2 iViewportSize, FfxFloat32 fDeviceDepth) +{ + return GetViewSpacePosition(iViewportPos, iViewportSize, fDeviceDepth) * ViewSpaceToMetersFactor(); +} + +FfxFloat32 GetMaxDistanceInMeters() +{ +#if FFX_FSR2_OPTION_INVERTED_DEPTH + return GetViewSpaceDepth(0.0f) * ViewSpaceToMetersFactor(); +#else + return GetViewSpaceDepth(1.0f) * ViewSpaceToMetersFactor(); +#endif +} + +FfxFloat32x3 PrepareRgb(FfxFloat32x3 fRgb, FfxFloat32 fExposure, FfxFloat32 fPreExposure) +{ + fRgb /= fPreExposure; + fRgb *= fExposure; + + fRgb = clamp(fRgb, 0.0f, FSR2_FP16_MAX); + + return fRgb; +} + +FfxFloat32x3 UnprepareRgb(FfxFloat32x3 fRgb, FfxFloat32 fExposure) +{ + fRgb /= fExposure; + fRgb *= PreExposure(); + + return fRgb; +} + + +struct BilinearSamplingData +{ + FfxInt32x2 iOffsets[4]; + FfxFloat32 fWeights[4]; + FfxInt32x2 iBasePos; +}; + +BilinearSamplingData GetBilinearSamplingData(FfxFloat32x2 fUv, FfxInt32x2 iSize) +{ + BilinearSamplingData data; + + FfxFloat32x2 fPxSample = (fUv * iSize) - FfxFloat32x2(0.5f, 0.5f); + data.iBasePos = FfxInt32x2(floor(fPxSample)); + FfxFloat32x2 fPxFrac = ffxFract(fPxSample); + + data.iOffsets[0] = FfxInt32x2(0, 0); + data.iOffsets[1] = FfxInt32x2(1, 0); + data.iOffsets[2] = FfxInt32x2(0, 1); + data.iOffsets[3] = FfxInt32x2(1, 1); + + data.fWeights[0] = (1 - fPxFrac.x) * (1 - fPxFrac.y); + data.fWeights[1] = (fPxFrac.x) * (1 - fPxFrac.y); + data.fWeights[2] = (1 - fPxFrac.x) * (fPxFrac.y); + data.fWeights[3] = (fPxFrac.x) * (fPxFrac.y); + + return data; +} + +struct PlaneData +{ + FfxFloat32x3 fNormal; + FfxFloat32 fDistanceFromOrigin; +}; + +PlaneData GetPlaneFromPoints(FfxFloat32x3 fP0, FfxFloat32x3 fP1, FfxFloat32x3 fP2) +{ + PlaneData plane; + + FfxFloat32x3 v0 = fP0 - fP1; + FfxFloat32x3 v1 = fP0 - fP2; + plane.fNormal = normalize(cross(v0, v1)); + plane.fDistanceFromOrigin = -dot(fP0, plane.fNormal); + + return plane; +} + +FfxFloat32 PointToPlaneDistance(PlaneData plane, FfxFloat32x3 fPoint) +{ + return abs(dot(plane.fNormal, fPoint) + plane.fDistanceFromOrigin); +} + +#endif // #if defined(FFX_GPU) + +#endif //!defined(FFX_FSR2_COMMON_H) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid.h new file mode 100644 index 000000000000..c63f1820e08d --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid.h @@ -0,0 +1,189 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +FFX_GROUPSHARED FfxUInt32 spdCounter; + +#ifndef SPD_PACKED_ONLY +FFX_GROUPSHARED FfxFloat32 spdIntermediateR[16][16]; +FFX_GROUPSHARED FfxFloat32 spdIntermediateG[16][16]; +FFX_GROUPSHARED FfxFloat32 spdIntermediateB[16][16]; +FFX_GROUPSHARED FfxFloat32 spdIntermediateA[16][16]; + +FfxFloat32x4 SpdLoadSourceImage(FfxFloat32x2 tex, FfxUInt32 slice) +{ + FfxFloat32x2 fUv = (tex + 0.5f + Jitter()) / RenderSize(); + fUv = ClampUv(fUv, RenderSize(), InputColorResourceDimensions()); + FfxFloat32x3 fRgb = SampleInputColor(fUv); + + fRgb /= PreExposure(); + + //compute log luma + const FfxFloat32 fLogLuma = log(ffxMax(FSR2_EPSILON, RGBToLuma(fRgb))); + + // Make sure out of screen pixels contribute no value to the end result + const FfxFloat32 result = all(FFX_LESS_THAN(tex, RenderSize())) ? fLogLuma : 0.0f; + + return FfxFloat32x4(result, 0, 0, 0); +} + +FfxFloat32x4 SpdLoad(FfxInt32x2 tex, FfxUInt32 slice) +{ + return SPD_LoadMipmap5(tex); +} + +void SpdStore(FfxInt32x2 pix, FfxFloat32x4 outValue, FfxUInt32 index, FfxUInt32 slice) +{ + if (index == LumaMipLevelToUse() || index == 5) + { + SPD_SetMipmap(pix, index, outValue.r); + } + + if (index == MipCount() - 1) { //accumulate on 1x1 level + + if (all(FFX_EQUAL(pix, FfxInt32x2(0, 0)))) + { + FfxFloat32 prev = SPD_LoadExposureBuffer().y; + FfxFloat32 result = outValue.r; + + if (prev < resetAutoExposureAverageSmoothing) // Compare Lavg, so small or negative values + { + FfxFloat32 rate = 1.0f; + result = prev + (result - prev) * (1 - exp(-DeltaTime() * rate)); + } + FfxFloat32x2 spdOutput = FfxFloat32x2(ComputeAutoExposureFromLavg(result), result); + SPD_SetExposureBuffer(spdOutput); + } + } +} + +void SpdIncreaseAtomicCounter(FfxUInt32 slice) +{ + SPD_IncreaseAtomicCounter(spdCounter); +} + +FfxUInt32 SpdGetAtomicCounter() +{ + return spdCounter; +} + +void SpdResetAtomicCounter(FfxUInt32 slice) +{ + SPD_ResetAtomicCounter(); +} + +FfxFloat32x4 SpdLoadIntermediate(FfxUInt32 x, FfxUInt32 y) +{ + return FfxFloat32x4( + spdIntermediateR[x][y], + spdIntermediateG[x][y], + spdIntermediateB[x][y], + spdIntermediateA[x][y]); +} +void SpdStoreIntermediate(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value) +{ + spdIntermediateR[x][y] = value.x; + spdIntermediateG[x][y] = value.y; + spdIntermediateB[x][y] = value.z; + spdIntermediateA[x][y] = value.w; +} +FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3) +{ + return (v0 + v1 + v2 + v3) * 0.25f; +} +#endif + +// define fetch and store functions Packed +#if FFX_HALF +#error Callback must be implemented + +FFX_GROUPSHARED FfxFloat16x2 spdIntermediateRG[16][16]; +FFX_GROUPSHARED FfxFloat16x2 spdIntermediateBA[16][16]; + +FfxFloat16x4 SpdLoadSourceImageH(FfxFloat32x2 tex, FfxUInt32 slice) +{ + return FfxFloat16x4(imgDst[0][FfxFloat32x3(tex, slice)]); +} +FfxFloat16x4 SpdLoadH(FfxInt32x2 p, FfxUInt32 slice) +{ + return FfxFloat16x4(imgDst6[FfxUInt32x3(p, slice)]); +} +void SpdStoreH(FfxInt32x2 p, FfxFloat16x4 value, FfxUInt32 mip, FfxUInt32 slice) +{ + if (index == LumaMipLevelToUse() || index == 5) + { + imgDst6[FfxUInt32x3(p, slice)] = FfxFloat32x4(value); + return; + } + imgDst[mip + 1][FfxUInt32x3(p, slice)] = FfxFloat32x4(value); +} +void SpdIncreaseAtomicCounter(FfxUInt32 slice) +{ + InterlockedAdd(rw_spd_global_atomic[FfxInt16x2(0, 0)].counter[slice], 1, spdCounter); +} +FfxUInt32 SpdGetAtomicCounter() +{ + return spdCounter; +} +void SpdResetAtomicCounter(FfxUInt32 slice) +{ + rw_spd_global_atomic[FfxInt16x2(0, 0)].counter[slice] = 0; +} +FfxFloat16x4 SpdLoadIntermediateH(FfxUInt32 x, FfxUInt32 y) +{ + return FfxFloat16x4( + spdIntermediateRG[x][y].x, + spdIntermediateRG[x][y].y, + spdIntermediateBA[x][y].x, + spdIntermediateBA[x][y].y); +} +void SpdStoreIntermediateH(FfxUInt32 x, FfxUInt32 y, FfxFloat16x4 value) +{ + spdIntermediateRG[x][y] = value.xy; + spdIntermediateBA[x][y] = value.zw; +} +FfxFloat16x4 SpdReduce4H(FfxFloat16x4 v0, FfxFloat16x4 v1, FfxFloat16x4 v2, FfxFloat16x4 v3) +{ + return (v0 + v1 + v2 + v3) * FfxFloat16(0.25); +} +#endif + +#include "ffx_spd.h" + +void ComputeAutoExposure(FfxUInt32x3 WorkGroupId, FfxUInt32 LocalThreadIndex) +{ +#if FFX_HALF + SpdDownsampleH( + FfxUInt32x2(WorkGroupId.xy), + FfxUInt32(LocalThreadIndex), + FfxUInt32(MipCount()), + FfxUInt32(NumWorkGroups()), + FfxUInt32(WorkGroupId.z), + FfxUInt32x2(WorkGroupOffset())); +#else + SpdDownsample( + FfxUInt32x2(WorkGroupId.xy), + FfxUInt32(LocalThreadIndex), + FfxUInt32(MipCount()), + FfxUInt32(NumWorkGroups()), + FfxUInt32(WorkGroupId.z), + FfxUInt32x2(WorkGroupOffset())); +#endif +} \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl new file mode 100644 index 000000000000..088e425452cc --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl @@ -0,0 +1,134 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_INPUT_COLOR 0 +#define FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC 1 +#define FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE 2 +#define FSR2_BIND_UAV_EXPOSURE_MIP_5 3 +#define FSR2_BIND_UAV_AUTO_EXPOSURE 4 +#define FSR2_BIND_CB_FSR2 5 +#define FSR2_BIND_CB_SPD 6 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +#if defined(FSR2_BIND_CB_SPD) + layout (set = 1, binding = FSR2_BIND_CB_SPD, std140) uniform cbSPD_t + { + uint mips; + uint numWorkGroups; + uvec2 workGroupOffset; + uvec2 renderSize; + } cbSPD; + + uint MipCount() + { + return cbSPD.mips; + } + + uint NumWorkGroups() + { + return cbSPD.numWorkGroups; + } + + uvec2 WorkGroupOffset() + { + return cbSPD.workGroupOffset; + } + + uvec2 SPD_RenderSize() + { + return cbSPD.renderSize; + } +#endif + +vec2 SPD_LoadExposureBuffer() +{ + return imageLoad(rw_auto_exposure, ivec2(0,0)).xy; +} + +void SPD_SetExposureBuffer(vec2 value) +{ + imageStore(rw_auto_exposure, ivec2(0,0), vec4(value, 0.0f, 0.0f)); +} + +vec4 SPD_LoadMipmap5(ivec2 iPxPos) +{ + return vec4(imageLoad(rw_img_mip_5, iPxPos).x, 0.0f, 0.0f, 0.0f); +} + +void SPD_SetMipmap(ivec2 iPxPos, uint slice, float value) +{ + switch (slice) + { + case FFX_FSR2_SHADING_CHANGE_MIP_LEVEL: + imageStore(rw_img_mip_shading_change, iPxPos, vec4(value, 0.0f, 0.0f, 0.0f)); + break; + case 5: + imageStore(rw_img_mip_5, iPxPos, vec4(value, 0.0f, 0.0f, 0.0f)); + break; + default: + + // avoid flattened side effect +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE) + imageStore(rw_img_mip_shading_change, iPxPos, vec4(imageLoad(rw_img_mip_shading_change, iPxPos).x, 0.0f, 0.0f, 0.0f)); +#elif defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) + imageStore(rw_img_mip_5, iPxPos, vec4(imageLoad(rw_img_mip_5, iPxPos).x, 0.0f, 0.0f, 0.0f)); +#endif + break; + } +} + +void SPD_IncreaseAtomicCounter(inout uint spdCounter) +{ + spdCounter = imageAtomicAdd(rw_spd_global_atomic, ivec2(0,0), 1); +} + +void SPD_ResetAtomicCounter() +{ + imageStore(rw_spd_global_atomic, ivec2(0,0), uvec4(0)); +} + +#include "ffx_fsr2_compute_luminance_pyramid.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 256 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + ComputeAutoExposure(gl_WorkGroupID.xyz, gl_LocalInvocationIndex); +} \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip.h new file mode 100644 index 000000000000..fa4c975a23fd --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip.h @@ -0,0 +1,258 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_DEPTH_CLIP_H +#define FFX_FSR2_DEPTH_CLIP_H + +FFX_STATIC const FfxFloat32 DepthClipBaseScale = 4.0f; + +FfxFloat32 ComputeDepthClip(FfxFloat32x2 fUvSample, FfxFloat32 fCurrentDepthSample) +{ + FfxFloat32 fCurrentDepthViewSpace = GetViewSpaceDepth(fCurrentDepthSample); + BilinearSamplingData bilinearInfo = GetBilinearSamplingData(fUvSample, RenderSize()); + + FfxFloat32 fDilatedSum = 0.0f; + FfxFloat32 fDepth = 0.0f; + FfxFloat32 fWeightSum = 0.0f; + for (FfxInt32 iSampleIndex = 0; iSampleIndex < 4; iSampleIndex++) { + + const FfxInt32x2 iOffset = bilinearInfo.iOffsets[iSampleIndex]; + const FfxInt32x2 iSamplePos = bilinearInfo.iBasePos + iOffset; + + if (IsOnScreen(iSamplePos, RenderSize())) { + const FfxFloat32 fWeight = bilinearInfo.fWeights[iSampleIndex]; + if (fWeight > fReconstructedDepthBilinearWeightThreshold) { + + const FfxFloat32 fPrevDepthSample = LoadReconstructedPrevDepth(iSamplePos); + const FfxFloat32 fPrevNearestDepthViewSpace = GetViewSpaceDepth(fPrevDepthSample); + + const FfxFloat32 fDepthDiff = fCurrentDepthViewSpace - fPrevNearestDepthViewSpace; + + if (fDepthDiff > 0.0f) { + +#if FFX_FSR2_OPTION_INVERTED_DEPTH + const FfxFloat32 fPlaneDepth = ffxMin(fPrevDepthSample, fCurrentDepthSample); +#else + const FfxFloat32 fPlaneDepth = ffxMax(fPrevDepthSample, fCurrentDepthSample); +#endif + + const FfxFloat32x3 fCenter = GetViewSpacePosition(FfxInt32x2(RenderSize() * 0.5f), RenderSize(), fPlaneDepth); + const FfxFloat32x3 fCorner = GetViewSpacePosition(FfxInt32x2(0, 0), RenderSize(), fPlaneDepth); + + const FfxFloat32 fHalfViewportWidth = length(FfxFloat32x2(RenderSize())); + const FfxFloat32 fDepthThreshold = ffxMax(fCurrentDepthViewSpace, fPrevNearestDepthViewSpace); + + const FfxFloat32 Ksep = 1.37e-05f; + const FfxFloat32 Kfov = length(fCorner) / length(fCenter); + const FfxFloat32 fRequiredDepthSeparation = Ksep * Kfov * fHalfViewportWidth * fDepthThreshold; + + const FfxFloat32 fResolutionFactor = ffxSaturate(length(FfxFloat32x2(RenderSize())) / length(FfxFloat32x2(1920.0f, 1080.0f))); + const FfxFloat32 fPower = ffxLerp(1.0f, 3.0f, fResolutionFactor); + fDepth += ffxPow(ffxSaturate(FfxFloat32(fRequiredDepthSeparation / fDepthDiff)), fPower) * fWeight; + fWeightSum += fWeight; + } + } + } + } + + return (fWeightSum > 0) ? ffxSaturate(1.0f - fDepth / fWeightSum) : 0.0f; +} + +FfxFloat32 ComputeMotionDivergence(FfxInt32x2 iPxPos, FfxInt32x2 iPxInputMotionVectorSize) +{ + FfxFloat32 minconvergence = 1.0f; + + FfxFloat32x2 fMotionVectorNucleus = LoadInputMotionVector(iPxPos); + FfxFloat32 fNucleusVelocityLr = length(fMotionVectorNucleus * RenderSize()); + FfxFloat32 fMaxVelocityUv = length(fMotionVectorNucleus); + + const FfxFloat32 MotionVectorVelocityEpsilon = 1e-02f; + + if (fNucleusVelocityLr > MotionVectorVelocityEpsilon) { + for (FfxInt32 y = -1; y <= 1; ++y) { + for (FfxInt32 x = -1; x <= 1; ++x) { + + FfxInt32x2 sp = ClampLoad(iPxPos, FfxInt32x2(x, y), iPxInputMotionVectorSize); + + FfxFloat32x2 fMotionVector = LoadInputMotionVector(sp); + FfxFloat32 fVelocityUv = length(fMotionVector); + + fMaxVelocityUv = ffxMax(fVelocityUv, fMaxVelocityUv); + fVelocityUv = ffxMax(fVelocityUv, fMaxVelocityUv); + minconvergence = ffxMin(minconvergence, dot(fMotionVector / fVelocityUv, fMotionVectorNucleus / fVelocityUv)); + } + } + } + + return ffxSaturate(1.0f - minconvergence) * ffxSaturate(fMaxVelocityUv / 0.01f); +} + +FfxFloat32 ComputeDepthDivergence(FfxInt32x2 iPxPos) +{ + const FfxFloat32 fMaxDistInMeters = GetMaxDistanceInMeters(); + FfxFloat32 fDepthMax = 0.0f; + FfxFloat32 fDepthMin = fMaxDistInMeters; + + FfxInt32 iMaxDistFound = 0; + + for (FfxInt32 y = -1; y < 2; y++) { + for (FfxInt32 x = -1; x < 2; x++) { + + const FfxInt32x2 iOffset = FfxInt32x2(x, y); + const FfxInt32x2 iSamplePos = iPxPos + iOffset; + + const FfxFloat32 fOnScreenFactor = IsOnScreen(iSamplePos, RenderSize()) ? 1.0f : 0.0f; + FfxFloat32 fDepth = GetViewSpaceDepthInMeters(LoadDilatedDepth(iSamplePos)) * fOnScreenFactor; + + iMaxDistFound |= FfxInt32(fMaxDistInMeters == fDepth); + + fDepthMin = ffxMin(fDepthMin, fDepth); + fDepthMax = ffxMax(fDepthMax, fDepth); + } + } + + return (1.0f - fDepthMin / fDepthMax) * (FfxBoolean(iMaxDistFound) ? 0.0f : 1.0f); +} + +FfxFloat32 ComputeTemporalMotionDivergence(FfxInt32x2 iPxPos) +{ + const FfxFloat32x2 fUv = FfxFloat32x2(iPxPos + 0.5f) / RenderSize(); + + FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos); + FfxFloat32x2 fReprojectedUv = fUv + fMotionVector; + fReprojectedUv = ClampUv(fReprojectedUv, RenderSize(), MaxRenderSize()); + FfxFloat32x2 fPrevMotionVector = SamplePreviousDilatedMotionVector(fReprojectedUv); + + float fPxDistance = length(fMotionVector * DisplaySize()); + return fPxDistance > 1.0f ? ffxLerp(0.0f, 1.0f - ffxSaturate(length(fPrevMotionVector) / length(fMotionVector)), ffxSaturate(ffxPow(fPxDistance / 20.0f, 3.0f))) : 0; +} + +void PreProcessReactiveMasks(FfxInt32x2 iPxLrPos, FfxFloat32 fMotionDivergence) +{ + // Compensate for bilinear sampling in accumulation pass + + FfxFloat32x3 fReferenceColor = LoadInputColor(iPxLrPos).xyz; + FfxFloat32x2 fReactiveFactor = FfxFloat32x2(0.0f, fMotionDivergence); + + float fMasksSum = 0.0f; + + FfxFloat32x3 fColorSamples[9]; + FfxFloat32 fReactiveSamples[9]; + FfxFloat32 fTransparencyAndCompositionSamples[9]; + + FFX_UNROLL + for (FfxInt32 y = -1; y < 2; y++) { + FFX_UNROLL + for (FfxInt32 x = -1; x < 2; x++) { + + const FfxInt32x2 sampleCoord = ClampLoad(iPxLrPos, FfxInt32x2(x, y), FfxInt32x2(RenderSize())); + + FfxInt32 sampleIdx = (y + 1) * 3 + x + 1; + + FfxFloat32x3 fColorSample = LoadInputColor(sampleCoord).xyz; + FfxFloat32 fReactiveSample = LoadReactiveMask(sampleCoord); + FfxFloat32 fTransparencyAndCompositionSample = LoadTransparencyAndCompositionMask(sampleCoord); + + fColorSamples[sampleIdx] = fColorSample; + fReactiveSamples[sampleIdx] = fReactiveSample; + fTransparencyAndCompositionSamples[sampleIdx] = fTransparencyAndCompositionSample; + + fMasksSum += (fReactiveSample + fTransparencyAndCompositionSample); + } + } + + if (fMasksSum > 0) + { + for (FfxInt32 sampleIdx = 0; sampleIdx < 9; sampleIdx++) + { + FfxFloat32x3 fColorSample = fColorSamples[sampleIdx]; + FfxFloat32 fReactiveSample = fReactiveSamples[sampleIdx]; + FfxFloat32 fTransparencyAndCompositionSample = fTransparencyAndCompositionSamples[sampleIdx]; + + const FfxFloat32 fMaxLenSq = ffxMax(dot(fReferenceColor, fReferenceColor), dot(fColorSample, fColorSample)); + const FfxFloat32 fSimilarity = dot(fReferenceColor, fColorSample) / fMaxLenSq; + + // Increase power for non-similar samples + const FfxFloat32 fPowerBiasMax = 6.0f; + const FfxFloat32 fSimilarityPower = 1.0f + (fPowerBiasMax - fSimilarity * fPowerBiasMax); + const FfxFloat32 fWeightedReactiveSample = ffxPow(fReactiveSample, fSimilarityPower); + const FfxFloat32 fWeightedTransparencyAndCompositionSample = ffxPow(fTransparencyAndCompositionSample, fSimilarityPower); + + fReactiveFactor = ffxMax(fReactiveFactor, FfxFloat32x2(fWeightedReactiveSample, fWeightedTransparencyAndCompositionSample)); + } + } + + StoreDilatedReactiveMasks(iPxLrPos, fReactiveFactor); +} + +FfxFloat32x3 ComputePreparedInputColor(FfxInt32x2 iPxLrPos) +{ + //We assume linear data. if non-linear input (sRGB, ...), + //then we should convert to linear first and back to sRGB on output. + FfxFloat32x3 fRgb = ffxMax(FfxFloat32x3(0, 0, 0), LoadInputColor(iPxLrPos)); + + fRgb = PrepareRgb(fRgb, Exposure(), PreExposure()); + + const FfxFloat32x3 fPreparedYCoCg = RGBToYCoCg(fRgb); + + return fPreparedYCoCg; +} + +FfxFloat32 EvaluateSurface(FfxInt32x2 iPxPos, FfxFloat32x2 fMotionVector) +{ + FfxFloat32 d0 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, -1))); + FfxFloat32 d1 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, 0))); + FfxFloat32 d2 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, 1))); + + return 1.0f - FfxFloat32(((d0 - d1) > (d1 * 0.01f)) && ((d1 - d2) > (d2 * 0.01f))); +} + +void DepthClip(FfxInt32x2 iPxPos) +{ + FfxFloat32x2 fDepthUv = (iPxPos + 0.5f) / RenderSize(); + FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos); + + // Discard tiny mvs + fMotionVector *= FfxFloat32(length(fMotionVector * DisplaySize()) > 0.01f); + + const FfxFloat32x2 fDilatedUv = fDepthUv + fMotionVector; + const FfxFloat32 fDilatedDepth = LoadDilatedDepth(iPxPos); + const FfxFloat32 fCurrentDepthViewSpace = GetViewSpaceDepth(LoadInputDepth(iPxPos)); + + // Compute prepared input color and depth clip + FfxFloat32 fDepthClip = ComputeDepthClip(fDilatedUv, fDilatedDepth) * EvaluateSurface(iPxPos, fMotionVector); + FfxFloat32x3 fPreparedYCoCg = ComputePreparedInputColor(iPxPos); + StorePreparedInputColor(iPxPos, FfxFloat32x4(fPreparedYCoCg, fDepthClip)); + + // Compute dilated reactive mask +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS + FfxInt32x2 iSamplePos = iPxPos; +#else + FfxInt32x2 iSamplePos = ComputeHrPosFromLrPos(iPxPos); +#endif + + FfxFloat32 fMotionDivergence = ComputeMotionDivergence(iSamplePos, RenderSize()); + FfxFloat32 fTemporalMotionDifference = ffxSaturate(ComputeTemporalMotionDivergence(iPxPos) - ComputeDepthDivergence(iPxPos)); + + PreProcessReactiveMasks(iPxPos, ffxMax(fTemporalMotionDifference, fMotionDivergence)); +} + +#endif //!defined( FFX_FSR2_DEPTH_CLIPH ) \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl new file mode 100644 index 000000000000..65cc8b67eff1 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl @@ -0,0 +1,67 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH 0 +#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 1 +#define FSR2_BIND_SRV_DILATED_DEPTH 2 +#define FSR2_BIND_SRV_REACTIVE_MASK 3 +#define FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK 4 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 5 +#define FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS 6 +#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 7 +#define FSR2_BIND_SRV_INPUT_COLOR 8 +#define FSR2_BIND_SRV_INPUT_DEPTH 9 +#define FSR2_BIND_SRV_INPUT_EXPOSURE 10 + +#define FSR2_BIND_UAV_DEPTH_CLIP 11 +#define FSR2_BIND_UAV_DILATED_REACTIVE_MASKS 12 +#define FSR2_BIND_UAV_PREPARED_INPUT_COLOR 13 + +#define FSR2_BIND_CB_FSR2 14 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_depth_clip.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + DepthClip(ivec2(gl_GlobalInvocationID.xy)); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_begin.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_begin.h new file mode 100644 index 000000000000..3bd4d5d91290 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_begin.h @@ -0,0 +1 @@ +// This file doesn't exist in this version of FSR. \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_end.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_end.h new file mode 100644 index 000000000000..3bd4d5d91290 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_end.h @@ -0,0 +1 @@ +// This file doesn't exist in this version of FSR. \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_lock.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_lock.h new file mode 100644 index 000000000000..8347fa86bcdf --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_lock.h @@ -0,0 +1,115 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_LOCK_H +#define FFX_FSR2_LOCK_H + +void ClearResourcesForNextFrame(in FfxInt32x2 iPxHrPos) +{ + if (all(FFX_LESS_THAN(iPxHrPos, FfxInt32x2(RenderSize())))) + { +#if FFX_FSR2_OPTION_INVERTED_DEPTH + const FfxUInt32 farZ = 0x0; +#else + const FfxUInt32 farZ = 0x3f800000; +#endif + SetReconstructedDepth(iPxHrPos, farZ); + } +} + +FfxBoolean ComputeThinFeatureConfidence(FfxInt32x2 pos) +{ + const FfxInt32 RADIUS = 1; + + FfxFloat32 fNucleus = LoadLockInputLuma(pos); + + FfxFloat32 similar_threshold = 1.05f; + FfxFloat32 dissimilarLumaMin = FSR2_FLT_MAX; + FfxFloat32 dissimilarLumaMax = 0; + + /* + 0 1 2 + 3 4 5 + 6 7 8 + */ + + #define SETBIT(x) (1U << x) + + FfxUInt32 mask = SETBIT(4); //flag fNucleus as similar + + const FfxUInt32 uNumRejectionMasks = 4; + const FfxUInt32 uRejectionMasks[uNumRejectionMasks] = { + SETBIT(0) | SETBIT(1) | SETBIT(3) | SETBIT(4), //Upper left + SETBIT(1) | SETBIT(2) | SETBIT(4) | SETBIT(5), //Upper right + SETBIT(3) | SETBIT(4) | SETBIT(6) | SETBIT(7), //Lower left + SETBIT(4) | SETBIT(5) | SETBIT(7) | SETBIT(8), //Lower right + }; + + FfxInt32 idx = 0; + FFX_UNROLL + for (FfxInt32 y = -RADIUS; y <= RADIUS; y++) { + FFX_UNROLL + for (FfxInt32 x = -RADIUS; x <= RADIUS; x++, idx++) { + if (x == 0 && y == 0) continue; + + FfxInt32x2 samplePos = ClampLoad(pos, FfxInt32x2(x, y), FfxInt32x2(RenderSize())); + + FfxFloat32 sampleLuma = LoadLockInputLuma(samplePos); + FfxFloat32 difference = ffxMax(sampleLuma, fNucleus) / ffxMin(sampleLuma, fNucleus); + + if (difference > 0 && (difference < similar_threshold)) { + mask |= SETBIT(idx); + } else { + dissimilarLumaMin = ffxMin(dissimilarLumaMin, sampleLuma); + dissimilarLumaMax = ffxMax(dissimilarLumaMax, sampleLuma); + } + } + } + + FfxBoolean isRidge = fNucleus > dissimilarLumaMax || fNucleus < dissimilarLumaMin; + + if (FFX_FALSE == isRidge) { + + return false; + } + + FFX_UNROLL + for (FfxInt32 i = 0; i < 4; i++) { + + if ((mask & uRejectionMasks[i]) == uRejectionMasks[i]) { + return false; + } + } + + return true; +} + +void ComputeLock(FfxInt32x2 iPxLrPos) +{ + if (ComputeThinFeatureConfidence(iPxLrPos)) + { + StoreNewLocks(ComputeHrPosFromLrPos(iPxLrPos), 1.f); + } + + ClearResourcesForNextFrame(iPxLrPos); +} + +#endif // FFX_FSR2_LOCK_H diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl new file mode 100644 index 000000000000..0adce1bb1127 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl @@ -0,0 +1,56 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_LOCK_INPUT_LUMA 0 +#define FSR2_BIND_UAV_NEW_LOCKS 1 +#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 2 +#define FSR2_BIND_CB_FSR2 3 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_lock.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + uvec2 uDispatchThreadId = gl_WorkGroupID.xy * uvec2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + gl_LocalInvocationID.xy; + + ComputeLock(ivec2(uDispatchThreadId)); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_postprocess_lock_status.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_postprocess_lock_status.h new file mode 100644 index 000000000000..cee9e148ba16 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_postprocess_lock_status.h @@ -0,0 +1,106 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_POSTPROCESS_LOCK_STATUS_H +#define FFX_FSR2_POSTPROCESS_LOCK_STATUS_H + +FfxFloat32x4 WrapShadingChangeLuma(FfxInt32x2 iPxSample) +{ + return FfxFloat32x4(LoadMipLuma(iPxSample, LumaMipLevelToUse()), 0, 0, 0); +} + +#if FFX_HALF +FFX_MIN16_F4 WrapShadingChangeLuma(FFX_MIN16_I2 iPxSample) +{ + return FFX_MIN16_F4(LoadMipLuma(iPxSample, LumaMipLevelToUse()), 0, 0, 0); +} +#endif + +#if FFX_FSR2_OPTION_POSTPROCESSLOCKSTATUS_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBilinearSamplesMin16(FetchShadingChangeLumaSamples, WrapShadingChangeLuma) +#else +DeclareCustomFetchBicubicSamples(FetchShadingChangeLumaSamples, WrapShadingChangeLuma) +#endif +DeclareCustomTextureSample(ShadingChangeLumaSample, Lanczos2, FetchShadingChangeLumaSamples) + +FfxFloat32 GetShadingChangeLuma(FfxInt32x2 iPxHrPos, FfxFloat32x2 fUvCoord) +{ + FfxFloat32 fShadingChangeLuma = 0; + +#if 0 + fShadingChangeLuma = Exposure() * exp(ShadingChangeLumaSample(fUvCoord, LumaMipDimensions()).x); +#else + + const FfxFloat32 fDiv = FfxFloat32(2 << LumaMipLevelToUse()); + FfxInt32x2 iMipRenderSize = FfxInt32x2(RenderSize() / fDiv); + + fUvCoord = ClampUv(fUvCoord, iMipRenderSize, LumaMipDimensions()); + fShadingChangeLuma = Exposure() * exp(FfxFloat32(SampleMipLuma(fUvCoord, LumaMipLevelToUse()))); +#endif + + fShadingChangeLuma = ffxPow(fShadingChangeLuma, 1.0f / 6.0f); + + return fShadingChangeLuma; +} + +void UpdateLockStatus(AccumulationPassCommonParams params, + FFX_PARAMETER_INOUT FfxFloat32 fReactiveFactor, LockState state, + FFX_PARAMETER_INOUT FfxFloat32x2 fLockStatus, + FFX_PARAMETER_OUT FfxFloat32 fLockContributionThisFrame, + FFX_PARAMETER_OUT FfxFloat32 fLuminanceDiff) { + + const FfxFloat32 fShadingChangeLuma = GetShadingChangeLuma(params.iPxHrPos, params.fHrUv); + + //init temporal shading change factor, init to -1 or so in reproject to know if "true new"? + fLockStatus[LOCK_TEMPORAL_LUMA] = (fLockStatus[LOCK_TEMPORAL_LUMA] == FfxFloat32(0.0f)) ? fShadingChangeLuma : fLockStatus[LOCK_TEMPORAL_LUMA]; + + FfxFloat32 fPreviousShadingChangeLuma = fLockStatus[LOCK_TEMPORAL_LUMA]; + + fLuminanceDiff = 1.0f - MinDividedByMax(fPreviousShadingChangeLuma, fShadingChangeLuma); + + if (state.NewLock) { + fLockStatus[LOCK_TEMPORAL_LUMA] = fShadingChangeLuma; + + fLockStatus[LOCK_LIFETIME_REMAINING] = (fLockStatus[LOCK_LIFETIME_REMAINING] != 0.0f) ? 2.0f : 1.0f; + } + else if(fLockStatus[LOCK_LIFETIME_REMAINING] <= 1.0f) { + fLockStatus[LOCK_TEMPORAL_LUMA] = ffxLerp(fLockStatus[LOCK_TEMPORAL_LUMA], FfxFloat32(fShadingChangeLuma), 0.5f); + } + else { + if (fLuminanceDiff > 0.1f) { + KillLock(fLockStatus); + } + } + + fReactiveFactor = ffxMax(fReactiveFactor, ffxSaturate((fLuminanceDiff - 0.1f) * 10.0f)); + fLockStatus[LOCK_LIFETIME_REMAINING] *= (1.0f - fReactiveFactor); + + fLockStatus[LOCK_LIFETIME_REMAINING] *= ffxSaturate(1.0f - params.fAccumulationMask); + fLockStatus[LOCK_LIFETIME_REMAINING] *= FfxFloat32(params.fDepthClipFactor < 0.1f); + + // Compute this frame lock contribution + const FfxFloat32 fLifetimeContribution = ffxSaturate(fLockStatus[LOCK_LIFETIME_REMAINING] - 1.0f); + const FfxFloat32 fShadingChangeContribution = ffxSaturate(MinDividedByMax(fLockStatus[LOCK_TEMPORAL_LUMA], fShadingChangeLuma)); + + fLockContributionThisFrame = ffxSaturate(ffxSaturate(fLifetimeContribution * 4.0f) * fShadingChangeContribution); +} + +#endif //!defined( FFX_FSR2_POSTPROCESS_LOCK_STATUS_H ) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas.h new file mode 100644 index 000000000000..d9006cd8ee00 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas.h @@ -0,0 +1,67 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#define GROUP_SIZE 8 + +#define FSR_RCAS_DENOISE 1 + +void WriteUpscaledOutput(FFX_MIN16_U2 iPxHrPos, FfxFloat32x3 fUpscaledColor) +{ + StoreUpscaledOutput(FFX_MIN16_I2(iPxHrPos), fUpscaledColor); +} + +#define FSR_RCAS_F +FfxFloat32x4 FsrRcasLoadF(FfxInt32x2 p) +{ + FfxFloat32x4 fColor = LoadRCAS_Input(p); + + fColor.rgb = PrepareRgb(fColor.rgb, Exposure(), PreExposure()); + + return fColor; +} + +void FsrRcasInputF(inout FfxFloat32 r, inout FfxFloat32 g, inout FfxFloat32 b) {} + +#include "ffx_fsr1.h" + + +void CurrFilter(FFX_MIN16_U2 pos) +{ + FfxFloat32x3 c; + FsrRcasF(c.r, c.g, c.b, pos, RCASConfig()); + + c = UnprepareRgb(c, Exposure()); + + WriteUpscaledOutput(pos, c); +} + +void RCAS(FfxUInt32x3 LocalThreadId, FfxUInt32x3 WorkGroupId, FfxUInt32x3 Dtid) +{ + // Do remapping of local xy in workgroup for a more PS-like swizzle pattern. + FfxUInt32x2 gxy = ffxRemapForQuad(LocalThreadId.x) + FfxUInt32x2(WorkGroupId.x << 4u, WorkGroupId.y << 4u); + CurrFilter(FFX_MIN16_U2(gxy)); + gxy.x += 8u; + CurrFilter(FFX_MIN16_U2(gxy)); + gxy.y += 8u; + CurrFilter(FFX_MIN16_U2(gxy)); + gxy.x -= 8u; + CurrFilter(FFX_MIN16_U2(gxy)); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl new file mode 100644 index 000000000000..f78fa53e6ebd --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl @@ -0,0 +1,80 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require +// Needed for rw_upscaled_output declaration +#extension GL_EXT_shader_image_load_formatted : require + +#define FSR2_BIND_SRV_INPUT_EXPOSURE 0 +#define FSR2_BIND_SRV_RCAS_INPUT 1 +#define FSR2_BIND_UAV_UPSCALED_OUTPUT 2 +#define FSR2_BIND_CB_FSR2 3 +#define FSR2_BIND_CB_RCAS 4 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +//Move to prototype shader! +#if defined(FSR2_BIND_CB_RCAS) + layout (set = 1, binding = FSR2_BIND_CB_RCAS, std140) uniform cbRCAS_t + { + uvec4 rcasConfig; + } cbRCAS; + + uvec4 RCASConfig() + { + return cbRCAS.rcasConfig; + } +#else + uvec4 RCASConfig() + { + return uvec4(0); + } +#endif + +vec4 LoadRCAS_Input(FfxInt32x2 iPxPos) +{ + return texelFetch(r_rcas_input, iPxPos, 0); +} + +#include "ffx_fsr2_rcas.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 64 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + RCAS(gl_LocalInvocationID.xyz, gl_WorkGroupID.xyz, gl_GlobalInvocationID.xyz); +} \ No newline at end of file diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h new file mode 100644 index 000000000000..e9ccc4bc8c28 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h @@ -0,0 +1,145 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H +#define FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H + +void ReconstructPrevDepth(FfxInt32x2 iPxPos, FfxFloat32 fDepth, FfxFloat32x2 fMotionVector, FfxInt32x2 iPxDepthSize) +{ + fMotionVector *= FfxFloat32(length(fMotionVector * DisplaySize()) > 0.1f); + + FfxFloat32x2 fUv = (iPxPos + FfxFloat32(0.5)) / iPxDepthSize; + FfxFloat32x2 fReprojectedUv = fUv + fMotionVector; + + BilinearSamplingData bilinearInfo = GetBilinearSamplingData(fReprojectedUv, RenderSize()); + + // Project current depth into previous frame locations. + // Push to all pixels having some contribution if reprojection is using bilinear logic. + for (FfxInt32 iSampleIndex = 0; iSampleIndex < 4; iSampleIndex++) { + + const FfxInt32x2 iOffset = bilinearInfo.iOffsets[iSampleIndex]; + FfxFloat32 fWeight = bilinearInfo.fWeights[iSampleIndex]; + + if (fWeight > fReconstructedDepthBilinearWeightThreshold) { + + FfxInt32x2 iStorePos = bilinearInfo.iBasePos + iOffset; + if (IsOnScreen(iStorePos, iPxDepthSize)) { + StoreReconstructedDepth(iStorePos, fDepth); + } + } + } +} + +void FindNearestDepth(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxInt32x2 iPxSize, FFX_PARAMETER_OUT FfxFloat32 fNearestDepth, FFX_PARAMETER_OUT FfxInt32x2 fNearestDepthCoord) +{ + const FfxInt32 iSampleCount = 9; + const FfxInt32x2 iSampleOffsets[iSampleCount] = { + FfxInt32x2(+0, +0), + FfxInt32x2(+1, +0), + FfxInt32x2(+0, +1), + FfxInt32x2(+0, -1), + FfxInt32x2(-1, +0), + FfxInt32x2(-1, +1), + FfxInt32x2(+1, +1), + FfxInt32x2(-1, -1), + FfxInt32x2(+1, -1), + }; + + // pull out the depth loads to allow SC to batch them + FfxFloat32 depth[9]; + FfxInt32 iSampleIndex = 0; + FFX_UNROLL + for (iSampleIndex = 0; iSampleIndex < iSampleCount; ++iSampleIndex) { + + FfxInt32x2 iPos = iPxPos + iSampleOffsets[iSampleIndex]; + depth[iSampleIndex] = LoadInputDepth(iPos); + } + + // find closest depth + fNearestDepthCoord = iPxPos; + fNearestDepth = depth[0]; + FFX_UNROLL + for (iSampleIndex = 1; iSampleIndex < iSampleCount; ++iSampleIndex) { + + FfxInt32x2 iPos = iPxPos + iSampleOffsets[iSampleIndex]; + if (IsOnScreen(iPos, iPxSize)) { + + FfxFloat32 fNdDepth = depth[iSampleIndex]; +#if FFX_FSR2_OPTION_INVERTED_DEPTH + if (fNdDepth > fNearestDepth) { +#else + if (fNdDepth < fNearestDepth) { +#endif + fNearestDepthCoord = iPos; + fNearestDepth = fNdDepth; + } + } + } +} + +FfxFloat32 ComputeLockInputLuma(FfxInt32x2 iPxLrPos) +{ + //We assume linear data. if non-linear input (sRGB, ...), + //then we should convert to linear first and back to sRGB on output. + FfxFloat32x3 fRgb = ffxMax(FfxFloat32x3(0, 0, 0), LoadInputColor(iPxLrPos)); + + // Use internal auto exposure for locking logic + fRgb /= PreExposure(); + fRgb *= Exposure(); + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + fRgb = Tonemap(fRgb); +#endif + + //compute luma used to lock pixels, if used elsewhere the ffxPow must be moved! + const FfxFloat32 fLockInputLuma = ffxPow(RGBToPerceivedLuma(fRgb), FfxFloat32(1.0 / 6.0)); + + return fLockInputLuma; +} + +void ReconstructAndDilate(FfxInt32x2 iPxLrPos) +{ + FfxFloat32 fDilatedDepth; + FfxInt32x2 iNearestDepthCoord; + + FindNearestDepth(iPxLrPos, RenderSize(), fDilatedDepth, iNearestDepthCoord); + +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS + FfxInt32x2 iSamplePos = iPxLrPos; + FfxInt32x2 iMotionVectorPos = iNearestDepthCoord; +#else + FfxInt32x2 iSamplePos = ComputeHrPosFromLrPos(iPxLrPos); + FfxInt32x2 iMotionVectorPos = ComputeHrPosFromLrPos(iNearestDepthCoord); +#endif + + FfxFloat32x2 fDilatedMotionVector = LoadInputMotionVector(iMotionVectorPos); + + StoreDilatedDepth(iPxLrPos, fDilatedDepth); + StoreDilatedMotionVector(iPxLrPos, fDilatedMotionVector); + + ReconstructPrevDepth(iPxLrPos, fDilatedDepth, fDilatedMotionVector, RenderSize()); + + FfxFloat32 fLockInputLuma = ComputeLockInputLuma(iPxLrPos); + StoreLockInputLuma(iPxLrPos, fLockInputLuma); +} + + +#endif //!defined( FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H ) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl new file mode 100644 index 000000000000..25c18c0622da --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl @@ -0,0 +1,65 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 0 +#define FSR2_BIND_SRV_INPUT_DEPTH 1 +#define FSR2_BIND_SRV_INPUT_COLOR 2 +#define FSR2_BIND_SRV_INPUT_EXPOSURE 3 +#define FSR2_BIND_SRV_LUMA_HISTORY 4 + +#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 5 +#define FSR2_BIND_UAV_DILATED_MOTION_VECTORS 6 +#define FSR2_BIND_UAV_DILATED_DEPTH 7 +#define FSR2_BIND_UAV_PREPARED_INPUT_COLOR 8 +#define FSR2_BIND_UAV_LUMA_HISTORY 9 +#define FSR2_BIND_UAV_LUMA_INSTABILITY 10 +#define FSR2_BIND_UAV_LOCK_INPUT_LUMA 11 + +#define FSR2_BIND_CB_FSR2 12 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + ReconstructAndDilate(FFX_MIN16_I2(gl_GlobalInvocationID.xy)); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_reproject.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_reproject.h new file mode 100644 index 000000000000..f7f396129e14 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_reproject.h @@ -0,0 +1,136 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_REPROJECT_H +#define FFX_FSR2_REPROJECT_H + +#ifndef FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE +#define FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE 0 // Reference +#endif + +FfxFloat32x4 WrapHistory(FfxInt32x2 iPxSample) +{ + return LoadHistory(iPxSample); +} + +#if FFX_HALF +FFX_MIN16_F4 WrapHistory(FFX_MIN16_I2 iPxSample) +{ + return FFX_MIN16_F4(LoadHistory(iPxSample)); +} +#endif + + +#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBicubicSamplesMin16(FetchHistorySamples, WrapHistory) +DeclareCustomTextureSampleMin16(HistorySample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchHistorySamples) +#else +DeclareCustomFetchBicubicSamples(FetchHistorySamples, WrapHistory) +DeclareCustomTextureSample(HistorySample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchHistorySamples) +#endif + +FfxFloat32x4 WrapLockStatus(FfxInt32x2 iPxSample) +{ + FfxFloat32x4 fSample = FfxFloat32x4(LoadLockStatus(iPxSample), 0.0f, 0.0f); + return fSample; +} + +#if FFX_HALF +FFX_MIN16_F4 WrapLockStatus(FFX_MIN16_I2 iPxSample) +{ + FFX_MIN16_F4 fSample = FFX_MIN16_F4(LoadLockStatus(iPxSample), 0.0, 0.0); + + return fSample; +} +#endif + +#if 1 +#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBilinearSamplesMin16(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSampleMin16(LockStatusSample, Bilinear, FetchLockStatusSamples) +#else +DeclareCustomFetchBilinearSamples(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSample(LockStatusSample, Bilinear, FetchLockStatusSamples) +#endif +#else +#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBicubicSamplesMin16(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSampleMin16(LockStatusSample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchLockStatusSamples) +#else +DeclareCustomFetchBicubicSamples(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSample(LockStatusSample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchLockStatusSamples) +#endif +#endif + +FfxFloat32x2 GetMotionVector(FfxInt32x2 iPxHrPos, FfxFloat32x2 fHrUv) +{ +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS + FfxFloat32x2 fDilatedMotionVector = LoadDilatedMotionVector(FFX_MIN16_I2(fHrUv * RenderSize())); +#else + FfxFloat32x2 fDilatedMotionVector = LoadInputMotionVector(iPxHrPos); +#endif + + return fDilatedMotionVector; +} + +FfxBoolean IsUvInside(FfxFloat32x2 fUv) +{ + return (fUv.x >= 0.0f && fUv.x <= 1.0f) && (fUv.y >= 0.0f && fUv.y <= 1.0f); +} + +void ComputeReprojectedUVs(const AccumulationPassCommonParams params, FFX_PARAMETER_OUT FfxFloat32x2 fReprojectedHrUv, FFX_PARAMETER_OUT FfxBoolean bIsExistingSample) +{ + fReprojectedHrUv = params.fHrUv + params.fMotionVector; + + bIsExistingSample = IsUvInside(fReprojectedHrUv); +} + +void ReprojectHistoryColor(const AccumulationPassCommonParams params, FFX_PARAMETER_OUT FfxFloat32x3 fHistoryColor, FFX_PARAMETER_OUT FfxFloat32 fTemporalReactiveFactor, FFX_PARAMETER_OUT FfxBoolean bInMotionLastFrame) +{ + FfxFloat32x4 fHistory = HistorySample(params.fReprojectedHrUv, DisplaySize()); + + fHistoryColor = PrepareRgb(fHistory.rgb, Exposure(), PreviousFramePreExposure()); + + fHistoryColor = RGBToYCoCg(fHistoryColor); + + //Compute temporal reactivity info + fTemporalReactiveFactor = ffxSaturate(abs(fHistory.w)); + bInMotionLastFrame = (fHistory.w < 0.0f); +} + +LockState ReprojectHistoryLockStatus(const AccumulationPassCommonParams params, FFX_PARAMETER_OUT FfxFloat32x2 fReprojectedLockStatus) +{ + LockState state = { FFX_FALSE, FFX_FALSE }; + const FfxFloat32 fNewLockIntensity = LoadRwNewLocks(params.iPxHrPos); + state.NewLock = fNewLockIntensity > (127.0f / 255.0f); + + FfxFloat32 fInPlaceLockLifetime = state.NewLock ? fNewLockIntensity : 0; + + fReprojectedLockStatus = SampleLockStatus(params.fReprojectedHrUv); + + if (fReprojectedLockStatus[LOCK_LIFETIME_REMAINING] != FfxFloat32(0.0f)) { + state.WasLockedPrevFrame = true; + } + + return state; +} + +#endif //!defined( FFX_FSR2_REPROJECT_H ) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_resources.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_resources.h new file mode 100644 index 000000000000..535dbc383c7b --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_resources.h @@ -0,0 +1,105 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_RESOURCES_H +#define FFX_FSR2_RESOURCES_H + +#if defined(FFX_CPU) || defined(FFX_GPU) +#define FFX_FSR2_RESOURCE_IDENTIFIER_NULL 0 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY 1 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR 2 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS 3 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH 4 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE 5 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK 6 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK 7 +#define FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH 8 +#define FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS 9 +#define FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH 10 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR 11 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS 12 +#define FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS 13 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR 14 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY 15 +#define FFX_FSR2_RESOURCE_IDENTIFIER_DEBUG_OUTPUT 16 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LANCZOS_LUT 17 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SPD_ATOMIC_COUNT 18 +#define FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT 19 +#define FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT 20 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_1 21 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_2 22 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_1 23 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_2 24 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_REACTIVITY 25 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_TRANSPARENCY_AND_COMPOSITION 26 +#define FFX_FSR2_RESOURCE_IDENTITIER_UPSAMPLE_MAXIMUM_BIAS_LUT 27 +#define FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS 28 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE 29 // same as FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0 29 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_1 30 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_2 31 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_3 32 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_4 33 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_5 34 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_6 35 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_7 36 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_8 37 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_9 38 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_10 39 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_11 40 +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_12 41 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_EXPOSURE 42 +#define FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE 43 +#define FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE 44 +#define FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION 45 + +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR 46 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR 47 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_1 48 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_1 49 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_2 50 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_2 51 +#define FFX_FSR2_RESOURCE_IDENTIFIER_PREVIOUS_DILATED_MOTION_VECTORS 52 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_1 53 +#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_2 54 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_1 55 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_2 56 +#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA 57 + +// Shading change detection mip level setting, value must be in the range [FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0, FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_12] +#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_4 +#define FFX_FSR2_SHADING_CHANGE_MIP_LEVEL (FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE - FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE) + +#define FFX_FSR2_RESOURCE_IDENTIFIER_COUNT 58 + +#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_FSR2 0 +#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_SPD 1 +#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_RCAS 2 +#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_GENREACTIVE 3 + +#define FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_TONEMAP 1 +#define FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_INVERSETONEMAP 2 +#define FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_THRESHOLD 4 +#define FFX_FSR2_AUTOREACTIVEFLAGS_USE_COMPONENTS_MAX 8 + +#endif // #if defined(FFX_CPU) || defined(FFX_GPU) + +#endif //!defined( FFX_FSR2_RESOURCES_H ) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_sample.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_sample.h new file mode 100644 index 000000000000..f94f40aa793c --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_sample.h @@ -0,0 +1,605 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_SAMPLE_H +#define FFX_FSR2_SAMPLE_H + +// suppress warnings +#ifdef FFX_HLSL +#pragma warning(disable: 4008) // potentially divide by zero +#endif //FFX_HLSL + +struct FetchedBilinearSamples { + + FfxFloat32x4 fColor00; + FfxFloat32x4 fColor10; + + FfxFloat32x4 fColor01; + FfxFloat32x4 fColor11; +}; + +struct FetchedBicubicSamples { + + FfxFloat32x4 fColor00; + FfxFloat32x4 fColor10; + FfxFloat32x4 fColor20; + FfxFloat32x4 fColor30; + + FfxFloat32x4 fColor01; + FfxFloat32x4 fColor11; + FfxFloat32x4 fColor21; + FfxFloat32x4 fColor31; + + FfxFloat32x4 fColor02; + FfxFloat32x4 fColor12; + FfxFloat32x4 fColor22; + FfxFloat32x4 fColor32; + + FfxFloat32x4 fColor03; + FfxFloat32x4 fColor13; + FfxFloat32x4 fColor23; + FfxFloat32x4 fColor33; +}; + +#if FFX_HALF +struct FetchedBilinearSamplesMin16 { + + FFX_MIN16_F4 fColor00; + FFX_MIN16_F4 fColor10; + + FFX_MIN16_F4 fColor01; + FFX_MIN16_F4 fColor11; +}; + +struct FetchedBicubicSamplesMin16 { + + FFX_MIN16_F4 fColor00; + FFX_MIN16_F4 fColor10; + FFX_MIN16_F4 fColor20; + FFX_MIN16_F4 fColor30; + + FFX_MIN16_F4 fColor01; + FFX_MIN16_F4 fColor11; + FFX_MIN16_F4 fColor21; + FFX_MIN16_F4 fColor31; + + FFX_MIN16_F4 fColor02; + FFX_MIN16_F4 fColor12; + FFX_MIN16_F4 fColor22; + FFX_MIN16_F4 fColor32; + + FFX_MIN16_F4 fColor03; + FFX_MIN16_F4 fColor13; + FFX_MIN16_F4 fColor23; + FFX_MIN16_F4 fColor33; +}; +#else //FFX_HALF +#define FetchedBicubicSamplesMin16 FetchedBicubicSamples +#define FetchedBilinearSamplesMin16 FetchedBilinearSamples +#endif //FFX_HALF + +FfxFloat32x4 Linear(FfxFloat32x4 A, FfxFloat32x4 B, FfxFloat32 t) +{ + return A + (B - A) * t; +} + +FfxFloat32x4 Bilinear(FetchedBilinearSamples BilinearSamples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Linear(BilinearSamples.fColor00, BilinearSamples.fColor10, fPxFrac.x); + FfxFloat32x4 fColorX1 = Linear(BilinearSamples.fColor01, BilinearSamples.fColor11, fPxFrac.x); + FfxFloat32x4 fColorXY = Linear(fColorX0, fColorX1, fPxFrac.y); + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Linear(FFX_MIN16_F4 A, FFX_MIN16_F4 B, FFX_MIN16_F t) +{ + return A + (B - A) * t; +} + +FFX_MIN16_F4 Bilinear(FetchedBilinearSamplesMin16 BilinearSamples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Linear(BilinearSamples.fColor00, BilinearSamples.fColor10, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Linear(BilinearSamples.fColor01, BilinearSamples.fColor11, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Linear(fColorX0, fColorX1, fPxFrac.y); + return fColorXY; +} +#endif + +FfxFloat32 Lanczos2NoClamp(FfxFloat32 x) +{ + const FfxFloat32 PI = 3.141592653589793f; // TODO: share SDK constants + return abs(x) < FSR2_EPSILON ? 1.f : (sin(PI * x) / (PI * x)) * (sin(0.5f * PI * x) / (0.5f * PI * x)); +} + +FfxFloat32 Lanczos2(FfxFloat32 x) +{ + x = ffxMin(abs(x), 2.0f); + return Lanczos2NoClamp(x); +} + +#if FFX_HALF + +#if 0 +FFX_MIN16_F Lanczos2NoClamp(FFX_MIN16_F x) +{ + const FFX_MIN16_F PI = FFX_MIN16_F(3.141592653589793f); // TODO: share SDK constants + return abs(x) < FFX_MIN16_F(FSR2_EPSILON) ? FFX_MIN16_F(1.f) : (sin(PI * x) / (PI * x)) * (sin(FFX_MIN16_F(0.5f) * PI * x) / (FFX_MIN16_F(0.5f) * PI * x)); +} +#endif + +FFX_MIN16_F Lanczos2(FFX_MIN16_F x) +{ + x = ffxMin(abs(x), FFX_MIN16_F(2.0f)); + return FFX_MIN16_F(Lanczos2NoClamp(x)); +} +#endif //FFX_HALF + +// FSR1 lanczos approximation. Input is x*x and must be <= 4. +FfxFloat32 Lanczos2ApproxSqNoClamp(FfxFloat32 x2) +{ + FfxFloat32 a = (2.0f / 5.0f) * x2 - 1; + FfxFloat32 b = (1.0f / 4.0f) * x2 - 1; + return ((25.0f / 16.0f) * a * a - (25.0f / 16.0f - 1)) * (b * b); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2ApproxSqNoClamp(FFX_MIN16_F x2) +{ + FFX_MIN16_F a = FFX_MIN16_F(2.0f / 5.0f) * x2 - FFX_MIN16_F(1); + FFX_MIN16_F b = FFX_MIN16_F(1.0f / 4.0f) * x2 - FFX_MIN16_F(1); + return (FFX_MIN16_F(25.0f / 16.0f) * a * a - FFX_MIN16_F(25.0f / 16.0f - 1)) * (b * b); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2ApproxSq(FfxFloat32 x2) +{ + x2 = ffxMin(x2, 4.0f); + return Lanczos2ApproxSqNoClamp(x2); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2ApproxSq(FFX_MIN16_F x2) +{ + x2 = ffxMin(x2, FFX_MIN16_F(4.0f)); + return Lanczos2ApproxSqNoClamp(x2); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2ApproxNoClamp(FfxFloat32 x) +{ + return Lanczos2ApproxSqNoClamp(x * x); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2ApproxNoClamp(FFX_MIN16_F x) +{ + return Lanczos2ApproxSqNoClamp(x * x); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2Approx(FfxFloat32 x) +{ + return Lanczos2ApproxSq(x * x); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2Approx(FFX_MIN16_F x) +{ + return Lanczos2ApproxSq(x * x); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2_UseLUT(FfxFloat32 x) +{ + return SampleLanczos2Weight(abs(x)); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2_UseLUT(FFX_MIN16_F x) +{ + return FFX_MIN16_F(SampleLanczos2Weight(abs(x))); +} +#endif //FFX_HALF + +FfxFloat32x4 Lanczos2_UseLUT(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t) +{ + FfxFloat32 fWeight0 = Lanczos2_UseLUT(-1.f - t); + FfxFloat32 fWeight1 = Lanczos2_UseLUT(-0.f - t); + FfxFloat32 fWeight2 = Lanczos2_UseLUT(+1.f - t); + FfxFloat32 fWeight3 = Lanczos2_UseLUT(+2.f - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} +#if FFX_HALF +FFX_MIN16_F4 Lanczos2_UseLUT(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t) +{ + FFX_MIN16_F fWeight0 = Lanczos2_UseLUT(FFX_MIN16_F(-1.f) - t); + FFX_MIN16_F fWeight1 = Lanczos2_UseLUT(FFX_MIN16_F(-0.f) - t); + FFX_MIN16_F fWeight2 = Lanczos2_UseLUT(FFX_MIN16_F(+1.f) - t); + FFX_MIN16_F fWeight3 = Lanczos2_UseLUT(FFX_MIN16_F(+2.f) - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} +#endif + +FfxFloat32x4 Lanczos2(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t) +{ + FfxFloat32 fWeight0 = Lanczos2(-1.f - t); + FfxFloat32 fWeight1 = Lanczos2(-0.f - t); + FfxFloat32 fWeight2 = Lanczos2(+1.f - t); + FfxFloat32 fWeight3 = Lanczos2(+2.f - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} + +FfxFloat32x4 Lanczos2(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Lanczos2(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FfxFloat32x4 fColorX1 = Lanczos2(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FfxFloat32x4 fColorX2 = Lanczos2(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FfxFloat32x4 fColorX3 = Lanczos2(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FfxFloat32x4 fColorXY = Lanczos2(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FfxFloat32x4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FfxFloat32x4 fDeringingMin = fDeringingSamples[0]; + FfxFloat32x4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) { + + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t) +{ + FFX_MIN16_F fWeight0 = Lanczos2(FFX_MIN16_F(-1.f) - t); + FFX_MIN16_F fWeight1 = Lanczos2(FFX_MIN16_F(-0.f) - t); + FFX_MIN16_F fWeight2 = Lanczos2(FFX_MIN16_F(+1.f) - t); + FFX_MIN16_F fWeight3 = Lanczos2(FFX_MIN16_F(+2.f) - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} + +FFX_MIN16_F4 Lanczos2(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Lanczos2(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Lanczos2(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FFX_MIN16_F4 fColorX2 = Lanczos2(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FFX_MIN16_F4 fColorX3 = Lanczos2(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Lanczos2(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FFX_MIN16_F4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0]; + FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} +#endif //FFX_HALF + + +FfxFloat32x4 Lanczos2LUT(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Lanczos2_UseLUT(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FfxFloat32x4 fColorX1 = Lanczos2_UseLUT(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FfxFloat32x4 fColorX2 = Lanczos2_UseLUT(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FfxFloat32x4 fColorX3 = Lanczos2_UseLUT(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FfxFloat32x4 fColorXY = Lanczos2_UseLUT(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FfxFloat32x4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FfxFloat32x4 fDeringingMin = fDeringingSamples[0]; + FfxFloat32x4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) { + + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2LUT(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Lanczos2_UseLUT(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Lanczos2_UseLUT(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FFX_MIN16_F4 fColorX2 = Lanczos2_UseLUT(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FFX_MIN16_F4 fColorX3 = Lanczos2_UseLUT(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Lanczos2_UseLUT(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FFX_MIN16_F4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0]; + FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} +#endif //FFX_HALF + + + +FfxFloat32x4 Lanczos2Approx(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t) +{ + FfxFloat32 fWeight0 = Lanczos2ApproxNoClamp(-1.f - t); + FfxFloat32 fWeight1 = Lanczos2ApproxNoClamp(-0.f - t); + FfxFloat32 fWeight2 = Lanczos2ApproxNoClamp(+1.f - t); + FfxFloat32 fWeight3 = Lanczos2ApproxNoClamp(+2.f - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2Approx(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t) +{ + FFX_MIN16_F fWeight0 = Lanczos2ApproxNoClamp(FFX_MIN16_F(-1.f) - t); + FFX_MIN16_F fWeight1 = Lanczos2ApproxNoClamp(FFX_MIN16_F(-0.f) - t); + FFX_MIN16_F fWeight2 = Lanczos2ApproxNoClamp(FFX_MIN16_F(+1.f) - t); + FFX_MIN16_F fWeight3 = Lanczos2ApproxNoClamp(FFX_MIN16_F(+2.f) - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} +#endif //FFX_HALF + +FfxFloat32x4 Lanczos2Approx(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Lanczos2Approx(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FfxFloat32x4 fColorX1 = Lanczos2Approx(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FfxFloat32x4 fColorX2 = Lanczos2Approx(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FfxFloat32x4 fColorX3 = Lanczos2Approx(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FfxFloat32x4 fColorXY = Lanczos2Approx(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FfxFloat32x4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FfxFloat32x4 fDeringingMin = fDeringingSamples[0]; + FfxFloat32x4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2Approx(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Lanczos2Approx(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Lanczos2Approx(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FFX_MIN16_F4 fColorX2 = Lanczos2Approx(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FFX_MIN16_F4 fColorX3 = Lanczos2Approx(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Lanczos2Approx(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FFX_MIN16_F4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0]; + FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} +#endif + +// Clamp by offset direction. Assuming iPxSample is already in range and iPxOffset is compile time constant. +FfxInt32x2 ClampCoord(FfxInt32x2 iPxSample, FfxInt32x2 iPxOffset, FfxInt32x2 iTextureSize) +{ + FfxInt32x2 result = iPxSample + iPxOffset; + result.x = (iPxOffset.x < 0) ? ffxMax(result.x, 0) : result.x; + result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - 1) : result.x; + result.y = (iPxOffset.y < 0) ? ffxMax(result.y, 0) : result.y; + result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - 1) : result.y; + return result; +} +#if FFX_HALF +FFX_MIN16_I2 ClampCoord(FFX_MIN16_I2 iPxSample, FFX_MIN16_I2 iPxOffset, FFX_MIN16_I2 iTextureSize) +{ + FFX_MIN16_I2 result = iPxSample + iPxOffset; + result.x = (iPxOffset.x < FFX_MIN16_I(0)) ? ffxMax(result.x, FFX_MIN16_I(0)) : result.x; + result.x = (iPxOffset.x > FFX_MIN16_I(0)) ? ffxMin(result.x, iTextureSize.x - FFX_MIN16_I(1)) : result.x; + result.y = (iPxOffset.y < FFX_MIN16_I(0)) ? ffxMax(result.y, FFX_MIN16_I(0)) : result.y; + result.y = (iPxOffset.y > FFX_MIN16_I(0)) ? ffxMin(result.y, iTextureSize.y - FFX_MIN16_I(1)) : result.y; + return result; +} +#endif //FFX_HALF + + +#define DeclareCustomFetchBicubicSamplesWithType(SampleType, TextureType, AddrType, Name, LoadTexture) \ + SampleType Name(AddrType iPxSample, AddrType iTextureSize) \ + { \ + SampleType Samples; \ + \ + Samples.fColor00 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, -1), iTextureSize))); \ + Samples.fColor10 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, -1), iTextureSize))); \ + Samples.fColor20 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, -1), iTextureSize))); \ + Samples.fColor30 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, -1), iTextureSize))); \ + \ + Samples.fColor01 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +0), iTextureSize))); \ + Samples.fColor11 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +0), iTextureSize))); \ + Samples.fColor21 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +0), iTextureSize))); \ + Samples.fColor31 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +0), iTextureSize))); \ + \ + Samples.fColor02 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +1), iTextureSize))); \ + Samples.fColor12 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +1), iTextureSize))); \ + Samples.fColor22 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +1), iTextureSize))); \ + Samples.fColor32 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +1), iTextureSize))); \ + \ + Samples.fColor03 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +2), iTextureSize))); \ + Samples.fColor13 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +2), iTextureSize))); \ + Samples.fColor23 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +2), iTextureSize))); \ + Samples.fColor33 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +2), iTextureSize))); \ + \ + return Samples; \ + } + +#define DeclareCustomFetchBicubicSamples(Name, LoadTexture) \ + DeclareCustomFetchBicubicSamplesWithType(FetchedBicubicSamples, FfxFloat32x4, FfxInt32x2, Name, LoadTexture) + +#define DeclareCustomFetchBicubicSamplesMin16(Name, LoadTexture) \ + DeclareCustomFetchBicubicSamplesWithType(FetchedBicubicSamplesMin16, FFX_MIN16_F4, FfxInt32x2, Name, LoadTexture) + +#define DeclareCustomFetchBilinearSamplesWithType(SampleType, TextureType,AddrType, Name, LoadTexture) \ + SampleType Name(AddrType iPxSample, AddrType iTextureSize) \ + { \ + SampleType Samples; \ + Samples.fColor00 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +0), iTextureSize))); \ + Samples.fColor10 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +0), iTextureSize))); \ + Samples.fColor01 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +1), iTextureSize))); \ + Samples.fColor11 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +1), iTextureSize))); \ + return Samples; \ + } + +#define DeclareCustomFetchBilinearSamples(Name, LoadTexture) \ + DeclareCustomFetchBilinearSamplesWithType(FetchedBilinearSamples, FfxFloat32x4, FfxInt32x2, Name, LoadTexture) + +#define DeclareCustomFetchBilinearSamplesMin16(Name, LoadTexture) \ + DeclareCustomFetchBilinearSamplesWithType(FetchedBilinearSamplesMin16, FFX_MIN16_F4, FfxInt32x2, Name, LoadTexture) + +// BE CAREFUL: there is some precision issues and (3253, 125) leading to (3252.9989778, 125.001102) +// is common, so iPxSample can "jitter" +#define DeclareCustomTextureSample(Name, InterpolateSamples, FetchSamples) \ + FfxFloat32x4 Name(FfxFloat32x2 fUvSample, FfxInt32x2 iTextureSize) \ + { \ + FfxFloat32x2 fPxSample = (fUvSample * FfxFloat32x2(iTextureSize)) - FfxFloat32x2(0.5f, 0.5f); \ + /* Clamp base coords */ \ + fPxSample.x = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.x), fPxSample.x)); \ + fPxSample.y = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.y), fPxSample.y)); \ + /* */ \ + FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); \ + FfxFloat32x2 fPxFrac = ffxFract(fPxSample); \ + FfxFloat32x4 fColorXY = FfxFloat32x4(InterpolateSamples(FetchSamples(iPxSample, iTextureSize), fPxFrac)); \ + return fColorXY; \ + } + +#define DeclareCustomTextureSampleMin16(Name, InterpolateSamples, FetchSamples) \ + FFX_MIN16_F4 Name(FfxFloat32x2 fUvSample, FfxInt32x2 iTextureSize) \ + { \ + FfxFloat32x2 fPxSample = (fUvSample * FfxFloat32x2(iTextureSize)) - FfxFloat32x2(0.5f, 0.5f); \ + /* Clamp base coords */ \ + fPxSample.x = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.x), fPxSample.x)); \ + fPxSample.y = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.y), fPxSample.y)); \ + /* */ \ + FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); \ + FFX_MIN16_F2 fPxFrac = FFX_MIN16_F2(ffxFract(fPxSample)); \ + FFX_MIN16_F4 fColorXY = FFX_MIN16_F4(InterpolateSamples(FetchSamples(iPxSample, iTextureSize), fPxFrac)); \ + return fColorXY; \ + } + +#define FFX_FSR2_CONCAT_ID(x, y) x ## y +#define FFX_FSR2_CONCAT(x, y) FFX_FSR2_CONCAT_ID(x, y) +#define FFX_FSR2_SAMPLER_1D_0 Lanczos2 +#define FFX_FSR2_SAMPLER_1D_1 Lanczos2LUT +#define FFX_FSR2_SAMPLER_1D_2 Lanczos2Approx + +#define FFX_FSR2_GET_LANCZOS_SAMPLER1D(x) FFX_FSR2_CONCAT(FFX_FSR2_SAMPLER_1D_, x) + +#endif //!defined( FFX_FSR2_SAMPLE_H ) diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen.h new file mode 100644 index 000000000000..101b75d25e4d --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen.h @@ -0,0 +1,250 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#define USE_YCOCG 1 + +#define fAutogenEpsilon 0.01f + +// EXPERIMENTAL + +FFX_MIN16_F ComputeAutoTC_01(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx) +{ + FfxFloat32x3 colorPreAlpha = LoadOpaqueOnly(uDispatchThreadId); + FfxFloat32x3 colorPostAlpha = LoadInputColor(uDispatchThreadId); + FfxFloat32x3 colorPrevPreAlpha = LoadPrevPreAlpha(iPrevIdx); + FfxFloat32x3 colorPrevPostAlpha = LoadPrevPostAlpha(iPrevIdx); + +#if USE_YCOCG + colorPreAlpha = RGBToYCoCg(colorPreAlpha); + colorPostAlpha = RGBToYCoCg(colorPostAlpha); + colorPrevPreAlpha = RGBToYCoCg(colorPrevPreAlpha); + colorPrevPostAlpha = RGBToYCoCg(colorPrevPostAlpha); +#endif + + FfxFloat32x3 colorDeltaCurr = colorPostAlpha - colorPreAlpha; + FfxFloat32x3 colorDeltaPrev = colorPrevPostAlpha - colorPrevPreAlpha; + bool hasAlpha = any(FFX_GREATER_THAN(abs(colorDeltaCurr), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon))); + bool hadAlpha = any(FFX_GREATER_THAN(abs(colorDeltaPrev), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon))); + + FfxFloat32x3 X = colorPreAlpha; + FfxFloat32x3 Y = colorPostAlpha; + FfxFloat32x3 Z = colorPrevPreAlpha; + FfxFloat32x3 W = colorPrevPostAlpha; + + FFX_MIN16_F retVal = FFX_MIN16_F(ffxSaturate(dot(abs(abs(Y - X) - abs(W - Z)), FfxFloat32x3(1, 1, 1)))); + + // cleanup very small values + retVal = (retVal < getTcThreshold()) ? FFX_MIN16_F(0.0f) : FFX_MIN16_F(1.f); + + return retVal; +} + +// works ok: thin edges +FFX_MIN16_F ComputeAutoTC_02(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx) +{ + FfxFloat32x3 colorPreAlpha = LoadOpaqueOnly(uDispatchThreadId); + FfxFloat32x3 colorPostAlpha = LoadInputColor(uDispatchThreadId); + FfxFloat32x3 colorPrevPreAlpha = LoadPrevPreAlpha(iPrevIdx); + FfxFloat32x3 colorPrevPostAlpha = LoadPrevPostAlpha(iPrevIdx); + +#if USE_YCOCG + colorPreAlpha = RGBToYCoCg(colorPreAlpha); + colorPostAlpha = RGBToYCoCg(colorPostAlpha); + colorPrevPreAlpha = RGBToYCoCg(colorPrevPreAlpha); + colorPrevPostAlpha = RGBToYCoCg(colorPrevPostAlpha); +#endif + + FfxFloat32x3 colorDelta = colorPostAlpha - colorPreAlpha; + FfxFloat32x3 colorPrevDelta = colorPrevPostAlpha - colorPrevPreAlpha; + bool hasAlpha = any(FFX_GREATER_THAN(abs(colorDelta), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon))); + bool hadAlpha = any(FFX_GREATER_THAN(abs(colorPrevDelta), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon))); + + FfxFloat32x3 delta = colorPostAlpha - colorPreAlpha; //prev+1*d = post => d = color, alpha = + FfxFloat32x3 deltaPrev = colorPrevPostAlpha - colorPrevPreAlpha; + + FfxFloat32x3 X = colorPrevPreAlpha; + FfxFloat32x3 N = colorPreAlpha - colorPrevPreAlpha; + FfxFloat32x3 YAminusXA = colorPrevPostAlpha - colorPrevPreAlpha; + FfxFloat32x3 NminusNA = colorPostAlpha - colorPrevPostAlpha; + + FfxFloat32x3 A = (hasAlpha || hadAlpha) ? NminusNA / max(FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon), N) : FfxFloat32x3(0, 0, 0); + + FFX_MIN16_F retVal = FFX_MIN16_F( max(max(A.x, A.y), A.z) ); + + // only pixels that have significantly changed in color shuold be considered + retVal = ffxSaturate(retVal * FFX_MIN16_F(length(colorPostAlpha - colorPrevPostAlpha)) ); + + return retVal; +} + +// This function computes the TransparencyAndComposition mask: +// This mask indicates pixels that should discard locks and apply color clamping. +// +// Typically this is the case for translucent pixels (that don't write depth values) or pixels where the correctness of +// the MVs can not be guaranteed (e.g. procedutal movement or vegetation that does not have MVs to reduce the cost during rasterization) +// Also, large changes in color due to changed lighting should be marked to remove locks on pixels with "old" lighting. +// +// This function takes a opaque only and a final texture and uses internal copies of those textures from the last frame. +// The function tries to determine where the color changes between opaque only and final image to determine the pixels that use transparency. +// Also it uses the previous frames and detects where the use of transparency changed to mark those pixels. +// Additionally it marks pixels where the color changed significantly in the opaque only image, e.g. due to lighting or texture animation. +// +// In the final step it stores the current textures in internal textures for the next frame + +FFX_MIN16_F ComputeTransparencyAndComposition(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx) +{ + FFX_MIN16_F retVal = ComputeAutoTC_02(uDispatchThreadId, iPrevIdx); + + // [branch] + if (retVal > FFX_MIN16_F(0.01f)) + { + retVal = ComputeAutoTC_01(uDispatchThreadId, iPrevIdx); + } + return retVal; +} + +float computeSolidEdge(FFX_MIN16_I2 curPos, FFX_MIN16_I2 prevPos) +{ + float lum[9]; + int i = 0; + for (int y = -1; y < 2; ++y) + { + for (int x = -1; x < 2; ++x) + { + FfxFloat32x3 curCol = LoadOpaqueOnly(curPos + FFX_MIN16_I2(x, y)).rgb; + FfxFloat32x3 prevCol = LoadPrevPreAlpha(prevPos + FFX_MIN16_I2(x, y)).rgb; + lum[i++] = length(curCol - prevCol); + } + } + + //float gradX = abs(lum[3] - lum[4]) + abs(lum[5] - lum[4]); + //float gradY = abs(lum[1] - lum[4]) + abs(lum[7] - lum[4]); + + //return sqrt(gradX * gradX + gradY * gradY); + + float gradX = abs(lum[3] - lum[4]) * abs(lum[5] - lum[4]); + float gradY = abs(lum[1] - lum[4]) * abs(lum[7] - lum[4]); + + return sqrt(sqrt(gradX * gradY)); +} + +float computeAlphaEdge(FFX_MIN16_I2 curPos, FFX_MIN16_I2 prevPos) +{ + float lum[9]; + int i = 0; + for (int y = -1; y < 2; ++y) + { + for (int x = -1; x < 2; ++x) + { + FfxFloat32x3 curCol = abs(LoadInputColor(curPos + FFX_MIN16_I2(x, y)).rgb - LoadOpaqueOnly(curPos + FFX_MIN16_I2(x, y)).rgb); + FfxFloat32x3 prevCol = abs(LoadPrevPostAlpha(prevPos + FFX_MIN16_I2(x, y)).rgb - LoadPrevPreAlpha(prevPos + FFX_MIN16_I2(x, y)).rgb); + lum[i++] = length(curCol - prevCol); + } + } + + //float gradX = abs(lum[3] - lum[4]) + abs(lum[5] - lum[4]); + //float gradY = abs(lum[1] - lum[4]) + abs(lum[7] - lum[4]); + + //return sqrt(gradX * gradX + gradY * gradY); + + float gradX = abs(lum[3] - lum[4]) * abs(lum[5] - lum[4]); + float gradY = abs(lum[1] - lum[4]) * abs(lum[7] - lum[4]); + + return sqrt(sqrt(gradX * gradY)); +} + +FFX_MIN16_F ComputeAabbOverlap(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx) +{ + FFX_MIN16_F retVal = FFX_MIN16_F(0.f); + + FfxFloat32x2 fMotionVector = LoadInputMotionVector(uDispatchThreadId); + FfxFloat32x3 colorPreAlpha = LoadOpaqueOnly(uDispatchThreadId); + FfxFloat32x3 colorPostAlpha = LoadInputColor(uDispatchThreadId); + FfxFloat32x3 colorPrevPreAlpha = LoadPrevPreAlpha(iPrevIdx); + FfxFloat32x3 colorPrevPostAlpha = LoadPrevPostAlpha(iPrevIdx); + +#if USE_YCOCG + colorPreAlpha = RGBToYCoCg(colorPreAlpha); + colorPostAlpha = RGBToYCoCg(colorPostAlpha); + colorPrevPreAlpha = RGBToYCoCg(colorPrevPreAlpha); + colorPrevPostAlpha = RGBToYCoCg(colorPrevPostAlpha); +#endif + FfxFloat32x3 minPrev = FFX_MIN16_F3(+1000.f, +1000.f, +1000.f); + FfxFloat32x3 maxPrev = FFX_MIN16_F3(-1000.f, -1000.f, -1000.f); + for (int y = -1; y < 2; ++y) + { + for (int x = -1; x < 2; ++x) + { + FfxFloat32x3 W = LoadPrevPostAlpha(iPrevIdx + FFX_MIN16_I2(x, y)); + +#if USE_YCOCG + W = RGBToYCoCg(W); +#endif + minPrev = min(minPrev, W); + maxPrev = max(maxPrev, W); + } + } + // instead of computing the overlap: simply count how many samples are outside + // set reactive based on that + FFX_MIN16_F count = FFX_MIN16_F(0.f); + for (int y = -1; y < 2; ++y) + { + for (int x = -1; x < 2; ++x) + { + FfxFloat32x3 Y = LoadInputColor(uDispatchThreadId + FFX_MIN16_I2(x, y)); + +#if USE_YCOCG + Y = RGBToYCoCg(Y); +#endif + count += ((Y.x < minPrev.x) || (Y.x > maxPrev.x)) ? FFX_MIN16_F(1.f) : FFX_MIN16_F(0.f); + count += ((Y.y < minPrev.y) || (Y.y > maxPrev.y)) ? FFX_MIN16_F(1.f) : FFX_MIN16_F(0.f); + count += ((Y.z < minPrev.z) || (Y.z > maxPrev.z)) ? FFX_MIN16_F(1.f) : FFX_MIN16_F(0.f); + } + } + retVal = count / FFX_MIN16_F(27.f); + + return retVal; +} + + +// This function computes the Reactive mask: +// We want pixels marked where the alpha portion of the frame changes a lot between neighbours +// Those pixels are expected to change quickly between frames, too. (e.g. small particles, reflections on curved surfaces...) +// As a result history would not be trustworthy. +// On the other hand we don't want pixels marked where pre-alpha has a large differnce, since those would profit from accumulation +// For mirrors we may assume the pre-alpha is pretty uniform color. +// +// This works well generally, but also marks edge pixels +FFX_MIN16_F ComputeReactive(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx) +{ + // we only get here if alpha has a significant contribution and has changed since last frame. + FFX_MIN16_F retVal = FFX_MIN16_F(0.f); + + // mark pixels with huge variance in alpha as reactive + FFX_MIN16_F alphaEdge = FFX_MIN16_F(computeAlphaEdge(uDispatchThreadId, iPrevIdx)); + FFX_MIN16_F opaqueEdge = FFX_MIN16_F(computeSolidEdge(uDispatchThreadId, iPrevIdx)); + retVal = ffxSaturate(alphaEdge - opaqueEdge); + + // the above also marks edge pixels due to jitter, so we need to cancel those out + + + return retVal; +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl b/thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl new file mode 100644 index 000000000000..12b4b40e08ad --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl @@ -0,0 +1,122 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + + + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_INPUT_OPAQUE_ONLY 0 +#define FSR2_BIND_SRV_INPUT_COLOR 1 +#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 2 +#define FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR 3 +#define FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR 4 +#define FSR2_BIND_SRV_REACTIVE_MASK 5 +#define FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK 6 + +#define FSR2_BIND_UAV_AUTOREACTIVE 7 +#define FSR2_BIND_UAV_AUTOCOMPOSITION 8 +#define FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR 9 +#define FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR 10 + +#define FSR2_BIND_CB_FSR2 11 +#define FSR2_BIND_CB_REACTIVE 12 + +// -- GODOT start -- +#if FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS +#define FSR2_BIND_SRV_INPUT_DEPTH 13 +#endif +// -- GODOT end -- + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +#ifdef FSR2_BIND_CB_REACTIVE +layout (set = 1, binding = FSR2_BIND_CB_REACTIVE, std140) uniform cbGenerateReactive_t +{ + float fTcThreshold; // 0.1 is a good starting value, lower will result in more TC pixels + float fTcScale; + float fReactiveScale; + float fReactiveMax; +} cbGenerateReactive; + +float getTcThreshold() +{ + return cbGenerateReactive.fTcThreshold; +} + +#else + float getTcThreshold() + { + return 0.05f; + } +#endif + +#include "ffx_fsr2_tcr_autogen.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + FFX_MIN16_I2 uDispatchThreadId = FFX_MIN16_I2(gl_GlobalInvocationID.xy); + + // ToDo: take into account jitter (i.e. add delta of previous jitter and current jitter to previous UV + // fetch pre- and post-alpha color values + FFX_MIN16_F2 fUv = ( FFX_MIN16_F2(uDispatchThreadId) + FFX_MIN16_F2(0.5f, 0.5f) ) / FFX_MIN16_F2( RenderSize() ); + FFX_MIN16_F2 fPrevUV = fUv + FFX_MIN16_F2( LoadInputMotionVector(uDispatchThreadId) ); + FFX_MIN16_I2 iPrevIdx = FFX_MIN16_I2(fPrevUV * FFX_MIN16_F2(RenderSize()) - 0.5f); + + FFX_MIN16_F3 colorPreAlpha = FFX_MIN16_F3( LoadOpaqueOnly( uDispatchThreadId ) ); + FFX_MIN16_F3 colorPostAlpha = FFX_MIN16_F3( LoadInputColor( uDispatchThreadId ) ); + + FFX_MIN16_F2 outReactiveMask = FFX_MIN16_F2( 0.f, 0.f ); + + outReactiveMask.y = ComputeTransparencyAndComposition(uDispatchThreadId, iPrevIdx); + + if (outReactiveMask.y > 0.5f) + { + outReactiveMask.x = ComputeReactive(uDispatchThreadId, iPrevIdx); + outReactiveMask.x *= FFX_MIN16_F(cbGenerateReactive.fReactiveScale); + outReactiveMask.x = outReactiveMask.x < cbGenerateReactive.fReactiveMax ? outReactiveMask.x : FFX_MIN16_F( cbGenerateReactive.fReactiveMax ); + } + + outReactiveMask.y *= FFX_MIN16_F(cbGenerateReactive.fTcScale); + + outReactiveMask.x = ffxMax(outReactiveMask.x, FFX_MIN16_F(LoadReactiveMask(uDispatchThreadId))); + outReactiveMask.y = ffxMax(outReactiveMask.y, FFX_MIN16_F(LoadTransparencyAndCompositionMask(uDispatchThreadId))); + + StoreAutoReactive(uDispatchThreadId, outReactiveMask); + + StorePrevPreAlpha(uDispatchThreadId, colorPreAlpha); + StorePrevPostAlpha(uDispatchThreadId, colorPostAlpha); +} diff --git a/thirdparty/amd-fsr2/shaders/ffx_fsr2_upsample.h b/thirdparty/amd-fsr2/shaders/ffx_fsr2_upsample.h new file mode 100644 index 000000000000..abdb8888a9a4 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_fsr2_upsample.h @@ -0,0 +1,194 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_UPSAMPLE_H +#define FFX_FSR2_UPSAMPLE_H + +FFX_STATIC const FfxUInt32 iLanczos2SampleCount = 16; + +void Deringing(RectificationBox clippingBox, FFX_PARAMETER_INOUT FfxFloat32x3 fColor) +{ + fColor = clamp(fColor, clippingBox.aabbMin, clippingBox.aabbMax); +} +#if FFX_HALF +void Deringing(RectificationBoxMin16 clippingBox, FFX_PARAMETER_INOUT FFX_MIN16_F3 fColor) +{ + fColor = clamp(fColor, clippingBox.aabbMin, clippingBox.aabbMax); +} +#endif + +#ifndef FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE +#define FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE 2 // Approximate +#endif + +FfxFloat32 GetUpsampleLanczosWeight(FfxFloat32x2 fSrcSampleOffset, FfxFloat32 fKernelWeight) +{ + FfxFloat32x2 fSrcSampleOffsetBiased = fSrcSampleOffset * fKernelWeight.xx; +#if FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 0 // LANCZOS_TYPE_REFERENCE + FfxFloat32 fSampleWeight = Lanczos2(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 1 // LANCZOS_TYPE_LUT + FfxFloat32 fSampleWeight = Lanczos2_UseLUT(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 2 // LANCZOS_TYPE_APPROXIMATE + FfxFloat32 fSampleWeight = Lanczos2ApproxSq(dot(fSrcSampleOffsetBiased, fSrcSampleOffsetBiased)); +#else +#error "Invalid Lanczos type" +#endif + return fSampleWeight; +} + +#if FFX_HALF +FFX_MIN16_F GetUpsampleLanczosWeight(FFX_MIN16_F2 fSrcSampleOffset, FFX_MIN16_F fKernelWeight) +{ + FFX_MIN16_F2 fSrcSampleOffsetBiased = fSrcSampleOffset * fKernelWeight.xx; +#if FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 0 // LANCZOS_TYPE_REFERENCE + FFX_MIN16_F fSampleWeight = Lanczos2(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 1 // LANCZOS_TYPE_LUT + FFX_MIN16_F fSampleWeight = Lanczos2_UseLUT(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 2 // LANCZOS_TYPE_APPROXIMATE + FFX_MIN16_F fSampleWeight = Lanczos2ApproxSq(dot(fSrcSampleOffsetBiased, fSrcSampleOffsetBiased)); + + // To Test: Save reciproqual sqrt compute + // FfxFloat32 fSampleWeight = Lanczos2Sq_UseLUT(dot(fSrcSampleOffsetBiased, fSrcSampleOffsetBiased)); +#else +#error "Invalid Lanczos type" +#endif + return fSampleWeight; +} +#endif + +FfxFloat32 ComputeMaxKernelWeight() { + const FfxFloat32 fKernelSizeBias = 1.0f; + + FfxFloat32 fKernelWeight = FfxFloat32(1) + (FfxFloat32(1.0f) / FfxFloat32x2(DownscaleFactor()) - FfxFloat32(1)).x * FfxFloat32(fKernelSizeBias); + + return ffxMin(FfxFloat32(1.99f), fKernelWeight); +} + +FfxFloat32x4 ComputeUpsampledColorAndWeight(const AccumulationPassCommonParams params, + FFX_PARAMETER_INOUT RectificationBox clippingBox, FfxFloat32 fReactiveFactor) +{ + #if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF + #include "ffx_fsr2_force16_begin.h" + #endif + // We compute a sliced lanczos filter with 2 lobes (other slices are accumulated temporaly) + FfxFloat32x2 fDstOutputPos = FfxFloat32x2(params.iPxHrPos) + FFX_BROADCAST_FLOAT32X2(0.5f); // Destination resolution output pixel center position + FfxFloat32x2 fSrcOutputPos = fDstOutputPos * DownscaleFactor(); // Source resolution output pixel center position + FfxInt32x2 iSrcInputPos = FfxInt32x2(floor(fSrcOutputPos)); // TODO: what about weird upscale factors... + + #if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF + #include "ffx_fsr2_force16_end.h" + #endif + + FfxFloat32x3 fSamples[iLanczos2SampleCount]; + + FfxFloat32x2 fSrcUnjitteredPos = (FfxFloat32x2(iSrcInputPos) + FfxFloat32x2(0.5f, 0.5f)) - Jitter(); // This is the un-jittered position of the sample at offset 0,0 + + FfxInt32x2 offsetTL; + offsetTL.x = (fSrcUnjitteredPos.x > fSrcOutputPos.x) ? FfxInt32(-2) : FfxInt32(-1); + offsetTL.y = (fSrcUnjitteredPos.y > fSrcOutputPos.y) ? FfxInt32(-2) : FfxInt32(-1); + + //Load samples + // If fSrcUnjitteredPos.y > fSrcOutputPos.y, indicates offsetTL.y = -2, sample offset Y will be [-2, 1], clipbox will be rows [1, 3]. + // Flip row# for sampling offset in this case, so first 0~2 rows in the sampled array can always be used for computing the clipbox. + // This reduces branch or cmove on sampled colors, but moving this overhead to sample position / weight calculation time which apply to less values. + const FfxBoolean bFlipRow = fSrcUnjitteredPos.y > fSrcOutputPos.y; + const FfxBoolean bFlipCol = fSrcUnjitteredPos.x > fSrcOutputPos.x; + + FfxFloat32x2 fOffsetTL = FfxFloat32x2(offsetTL); + + FFX_UNROLL + for (FfxInt32 row = 0; row < 3; row++) { + + FFX_UNROLL + for (FfxInt32 col = 0; col < 3; col++) { + FfxInt32 iSampleIndex = col + (row << 2); + + FfxInt32x2 sampleColRow = FfxInt32x2(bFlipCol ? (3 - col) : col, bFlipRow ? (3 - row) : row); + FfxInt32x2 iSrcSamplePos = FfxInt32x2(iSrcInputPos) + offsetTL + sampleColRow; + + const FfxInt32x2 sampleCoord = ClampLoad(iSrcSamplePos, FfxInt32x2(0, 0), FfxInt32x2(RenderSize())); + + fSamples[iSampleIndex] = LoadPreparedInputColor(FfxInt32x2(sampleCoord)); + } + } + + FfxFloat32x4 fColorAndWeight = FfxFloat32x4(0.0f, 0.0f, 0.0f, 0.0f); + + FfxFloat32x2 fBaseSampleOffset = FfxFloat32x2(fSrcUnjitteredPos - fSrcOutputPos); + + // Identify how much of each upsampled color to be used for this frame + const FfxFloat32 fKernelReactiveFactor = ffxMax(fReactiveFactor, FfxFloat32(params.bIsNewSample)); + const FfxFloat32 fKernelBiasMax = ComputeMaxKernelWeight() * (1.0f - fKernelReactiveFactor); + + const FfxFloat32 fKernelBiasMin = ffxMax(1.0f, ((1.0f + fKernelBiasMax) * 0.3f)); + const FfxFloat32 fKernelBiasFactor = ffxMax(0.0f, ffxMax(0.25f * params.fDepthClipFactor, fKernelReactiveFactor)); + const FfxFloat32 fKernelBias = ffxLerp(fKernelBiasMax, fKernelBiasMin, fKernelBiasFactor); + + const FfxFloat32 fRectificationCurveBias = ffxLerp(-2.0f, -3.0f, ffxSaturate(params.fHrVelocity / 50.0f)); + + FFX_UNROLL + for (FfxInt32 row = 0; row < 3; row++) { + FFX_UNROLL + for (FfxInt32 col = 0; col < 3; col++) { + FfxInt32 iSampleIndex = col + (row << 2); + + const FfxInt32x2 sampleColRow = FfxInt32x2(bFlipCol ? (3 - col) : col, bFlipRow ? (3 - row) : row); + const FfxFloat32x2 fOffset = fOffsetTL + FfxFloat32x2(sampleColRow); + FfxFloat32x2 fSrcSampleOffset = fBaseSampleOffset + fOffset; + + FfxInt32x2 iSrcSamplePos = FfxInt32x2(iSrcInputPos) + FfxInt32x2(offsetTL) + sampleColRow; + + const FfxFloat32 fOnScreenFactor = FfxFloat32(IsOnScreen(FfxInt32x2(iSrcSamplePos), FfxInt32x2(RenderSize()))); + FfxFloat32 fSampleWeight = fOnScreenFactor * FfxFloat32(GetUpsampleLanczosWeight(fSrcSampleOffset, fKernelBias)); + + fColorAndWeight += FfxFloat32x4(fSamples[iSampleIndex] * fSampleWeight, fSampleWeight); + + // Update rectification box + { + const FfxFloat32 fSrcSampleOffsetSq = dot(fSrcSampleOffset, fSrcSampleOffset); + const FfxFloat32 fBoxSampleWeight = exp(fRectificationCurveBias * fSrcSampleOffsetSq); + + const FfxBoolean bInitialSample = (row == 0) && (col == 0); + RectificationBoxAddSample(bInitialSample, clippingBox, fSamples[iSampleIndex], fBoxSampleWeight); + } + } + } + + RectificationBoxComputeVarianceBoxData(clippingBox); + + fColorAndWeight.w *= FfxFloat32(fColorAndWeight.w > FSR2_EPSILON); + + if (fColorAndWeight.w > FSR2_EPSILON) { + // Normalize for deringing (we need to compare colors) + fColorAndWeight.xyz = fColorAndWeight.xyz / fColorAndWeight.w; + fColorAndWeight.w *= fUpsampleLanczosWeightScale; + + Deringing(clippingBox, fColorAndWeight.xyz); + } + + #if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF + #include "ffx_fsr2_force16_end.h" + #endif + + return fColorAndWeight; +} + +#endif //!defined( FFX_FSR2_UPSAMPLE_H ) diff --git a/thirdparty/amd-fsr2/shaders/ffx_spd.h b/thirdparty/amd-fsr2/shaders/ffx_spd.h new file mode 100644 index 000000000000..5ce24ec87cc3 --- /dev/null +++ b/thirdparty/amd-fsr2/shaders/ffx_spd.h @@ -0,0 +1,936 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifdef FFX_CPU +FFX_STATIC void SpdSetup(FfxUInt32x2 dispatchThreadGroupCountXY, // CPU side: dispatch thread group count xy + FfxUInt32x2 workGroupOffset, // GPU side: pass in as constant + FfxUInt32x2 numWorkGroupsAndMips, // GPU side: pass in as constant + FfxUInt32x4 rectInfo, // left, top, width, height + FfxInt32 mips) // optional: if -1, calculate based on rect width and height +{ + workGroupOffset[0] = rectInfo[0] / 64; // rectInfo[0] = left + workGroupOffset[1] = rectInfo[1] / 64; // rectInfo[1] = top + + FfxUInt32 endIndexX = (rectInfo[0] + rectInfo[2] - 1) / 64; // rectInfo[0] = left, rectInfo[2] = width + FfxUInt32 endIndexY = (rectInfo[1] + rectInfo[3] - 1) / 64; // rectInfo[1] = top, rectInfo[3] = height + + dispatchThreadGroupCountXY[0] = endIndexX + 1 - workGroupOffset[0]; + dispatchThreadGroupCountXY[1] = endIndexY + 1 - workGroupOffset[1]; + + numWorkGroupsAndMips[0] = (dispatchThreadGroupCountXY[0]) * (dispatchThreadGroupCountXY[1]); + + if (mips >= 0) + { + numWorkGroupsAndMips[1] = FfxUInt32(mips); + } + else + { + // calculate based on rect width and height + FfxUInt32 resolution = ffxMax(rectInfo[2], rectInfo[3]); + numWorkGroupsAndMips[1] = FfxUInt32((ffxMin(floor(log2(FfxFloat32(resolution))), FfxFloat32(12)))); + } +} + +FFX_STATIC void SpdSetup(FfxUInt32x2 dispatchThreadGroupCountXY, // CPU side: dispatch thread group count xy + FfxUInt32x2 workGroupOffset, // GPU side: pass in as constant + FfxUInt32x2 numWorkGroupsAndMips, // GPU side: pass in as constant + FfxUInt32x4 rectInfo) // left, top, width, height +{ + SpdSetup(dispatchThreadGroupCountXY, workGroupOffset, numWorkGroupsAndMips, rectInfo, -1); +} +#endif // #ifdef FFX_CPU + + +//============================================================================================================================== +// NON-PACKED VERSION +//============================================================================================================================== +#ifdef FFX_GPU +#ifdef SPD_PACKED_ONLY +// Avoid compiler error +FfxFloat32x4 SpdLoadSourceImage(FfxInt32x2 p, FfxUInt32 slice) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} + +FfxFloat32x4 SpdLoad(FfxInt32x2 p, FfxUInt32 slice) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} +void SpdStore(FfxInt32x2 p, FfxFloat32x4 value, FfxUInt32 mip, FfxUInt32 slice) +{ +} +FfxFloat32x4 SpdLoadIntermediate(FfxUInt32 x, FfxUInt32 y) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} +void SpdStoreIntermediate(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value) +{ +} +FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} +#endif // #ifdef SPD_PACKED_ONLY + +//_____________________________________________________________/\_______________________________________________________________ +#if defined(FFX_GLSL) && !defined(SPD_NO_WAVE_OPERATIONS) +#extension GL_KHR_shader_subgroup_quad:require +#endif + +void SpdWorkgroupShuffleBarrier() +{ +#ifdef FFX_GLSL + barrier(); +#endif +#ifdef FFX_HLSL + GroupMemoryBarrierWithGroupSync(); +#endif +} + +// Only last active workgroup should proceed +bool SpdExitWorkgroup(FfxUInt32 numWorkGroups, FfxUInt32 localInvocationIndex, FfxUInt32 slice) +{ + // global atomic counter + if (localInvocationIndex == 0) + { + SpdIncreaseAtomicCounter(slice); + } + + SpdWorkgroupShuffleBarrier(); + return (SpdGetAtomicCounter() != (numWorkGroups - 1)); +} + +// User defined: FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3); +FfxFloat32x4 SpdReduceQuad(FfxFloat32x4 v) +{ +#if defined(FFX_GLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + + FfxFloat32x4 v0 = v; + FfxFloat32x4 v1 = subgroupQuadSwapHorizontal(v); + FfxFloat32x4 v2 = subgroupQuadSwapVertical(v); + FfxFloat32x4 v3 = subgroupQuadSwapDiagonal(v); + return SpdReduce4(v0, v1, v2, v3); + +#elif defined(FFX_HLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + + // requires SM6.0 + FfxUInt32 quad = WaveGetLaneIndex() & (~0x3); + FfxFloat32x4 v0 = v; + FfxFloat32x4 v1 = WaveReadLaneAt(v, quad | 1); + FfxFloat32x4 v2 = WaveReadLaneAt(v, quad | 2); + FfxFloat32x4 v3 = WaveReadLaneAt(v, quad | 3); + return SpdReduce4(v0, v1, v2, v3); +/* + // if SM6.0 is not available, you can use the AMD shader intrinsics + // the AMD shader intrinsics are available in AMD GPU Services (AGS) library: + // https://gpuopen.com/amd-gpu-services-ags-library/ + // works for DX11 + FfxFloat32x4 v0 = v; + FfxFloat32x4 v1; + v1.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + FfxFloat32x4 v2; + v2.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + FfxFloat32x4 v3; + v3.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + return SpdReduce4(v0, v1, v2, v3); + */ +#endif + return v; +} + +FfxFloat32x4 SpdReduceIntermediate(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3) +{ + FfxFloat32x4 v0 = SpdLoadIntermediate(i0.x, i0.y); + FfxFloat32x4 v1 = SpdLoadIntermediate(i1.x, i1.y); + FfxFloat32x4 v2 = SpdLoadIntermediate(i2.x, i2.y); + FfxFloat32x4 v3 = SpdLoadIntermediate(i3.x, i3.y); + return SpdReduce4(v0, v1, v2, v3); +} + +FfxFloat32x4 SpdReduceLoad4(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat32x4 v0 = SpdLoad(FfxInt32x2(i0), slice); + FfxFloat32x4 v1 = SpdLoad(FfxInt32x2(i1), slice); + FfxFloat32x4 v2 = SpdLoad(FfxInt32x2(i2), slice); + FfxFloat32x4 v3 = SpdLoad(FfxInt32x2(i3), slice); + return SpdReduce4(v0, v1, v2, v3); +} + +FfxFloat32x4 SpdReduceLoad4(FfxUInt32x2 base, FfxUInt32 slice) +{ + return SpdReduceLoad4(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +} + +FfxFloat32x4 SpdReduceLoadSourceImage4(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat32x4 v0 = SpdLoadSourceImage(FfxInt32x2(i0), slice); + FfxFloat32x4 v1 = SpdLoadSourceImage(FfxInt32x2(i1), slice); + FfxFloat32x4 v2 = SpdLoadSourceImage(FfxInt32x2(i2), slice); + FfxFloat32x4 v3 = SpdLoadSourceImage(FfxInt32x2(i3), slice); + return SpdReduce4(v0, v1, v2, v3); +} + +FfxFloat32x4 SpdReduceLoadSourceImage(FfxUInt32x2 base, FfxUInt32 slice) +{ +#ifdef SPD_LINEAR_SAMPLER + return SpdLoadSourceImage(FfxInt32x2(base), slice); +#else + return SpdReduceLoadSourceImage4(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +#endif +} + +void SpdDownsampleMips_0_1_Intrinsics(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ + FfxFloat32x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[3], 0, slice); + + if (mip <= 1) + return; + + v[0] = SpdReduceQuad(v[0]); + v[1] = SpdReduceQuad(v[1]); + v[2] = SpdReduceQuad(v[2]); + v[3] = SpdReduceQuad(v[3]); + + if ((localInvocationIndex % 4) == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2), v[0], 1, slice); + SpdStoreIntermediate(x / 2, y / 2, v[0]); + + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2), v[1], 1, slice); + SpdStoreIntermediate(x / 2 + 8, y / 2, v[1]); + + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2 + 8), v[2], 1, slice); + SpdStoreIntermediate(x / 2, y / 2 + 8, v[2]); + + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2 + 8), v[3], 1, slice); + SpdStoreIntermediate(x / 2 + 8, y / 2 + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1_LDS(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ + FfxFloat32x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[3], 0, slice); + + if (mip <= 1) + return; + + for (FfxUInt32 i = 0; i < 4; i++) + { + SpdStoreIntermediate(x, y, v[i]); + SpdWorkgroupShuffleBarrier(); + if (localInvocationIndex < 64) + { + v[i] = SpdReduceIntermediate(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x + (i % 2) * 8, y + (i / 2) * 8), v[i], 1, slice); + } + SpdWorkgroupShuffleBarrier(); + } + + if (localInvocationIndex < 64) + { + SpdStoreIntermediate(x + 0, y + 0, v[0]); + SpdStoreIntermediate(x + 8, y + 0, v[1]); + SpdStoreIntermediate(x + 0, y + 8, v[2]); + SpdStoreIntermediate(x + 8, y + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + SpdDownsampleMips_0_1_LDS(x, y, workGroupID, localInvocationIndex, mip, slice); +#else + SpdDownsampleMips_0_1_Intrinsics(x, y, workGroupID, localInvocationIndex, mip, slice); +#endif +} + + +void SpdDownsampleMip_2(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 64) + { + FfxFloat32x4 v = SpdReduceIntermediate(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStore(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS, try to reduce bank conflicts + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 x + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // ... + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + SpdStoreIntermediate(x * 2 + y % 2, y * 2, v); + } +#else + FfxFloat32x4 v = SpdLoadIntermediate(x, y); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediate(x + (y / 2) % 2, y, v); + } +#endif +} + +void SpdDownsampleMip_3(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 16) + { + // x 0 x 0 + // 0 0 0 0 + // 0 x 0 x + // 0 0 0 0 + FfxFloat32x4 v = + SpdReduceIntermediate(FfxUInt32x2(x * 4 + 0 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 2 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 0 + 1, y * 4 + 2), FfxUInt32x2(x * 4 + 2 + 1, y * 4 + 2)); + SpdStore(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 + // ... + // 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 + // ... + // 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x + // ... + SpdStoreIntermediate(x * 4 + y, y * 4, v); + } +#else + if (localInvocationIndex < 64) + { + FfxFloat32x4 v = SpdLoadIntermediate(x * 2 + y % 2, y * 2); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediate(x * 2 + y / 2, y * 2, v); + } + } +#endif +} + +void SpdDownsampleMip_4(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 4) + { + // x 0 0 0 x 0 0 0 + // ... + // 0 x 0 0 0 x 0 0 + FfxFloat32x4 v = SpdReduceIntermediate(FfxUInt32x2(x * 8 + 0 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 4 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 0 + 1 + y * 2, y * 8 + 4), + FfxUInt32x2(x * 8 + 4 + 1 + y * 2, y * 8 + 4)); + SpdStore(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x x x x 0 ... + // 0 ... + SpdStoreIntermediate(x + y * 2, 0, v); + } +#else + if (localInvocationIndex < 16) + { + FfxFloat32x4 v = SpdLoadIntermediate(x * 4 + y, y * 4); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediate(x / 2 + y, 0, v); + } + } +#endif +} + +void SpdDownsampleMip_5(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 1) + { + // x x x x 0 ... + // 0 ... + FfxFloat32x4 v = SpdReduceIntermediate(FfxUInt32x2(0, 0), FfxUInt32x2(1, 0), FfxUInt32x2(2, 0), FfxUInt32x2(3, 0)); + SpdStore(FfxInt32x2(workGroupID.xy), v, mip, slice); + } +#else + if (localInvocationIndex < 4) + { + FfxFloat32x4 v = SpdLoadIntermediate(localInvocationIndex, 0); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy), v, mip, slice); + } + } +#endif +} + +void SpdDownsampleMips_6_7(FfxUInt32 x, FfxUInt32 y, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxInt32x2 tex = FfxInt32x2(x * 4 + 0, y * 4 + 0); + FfxInt32x2 pix = FfxInt32x2(x * 2 + 0, y * 2 + 0); + FfxFloat32x4 v0 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v0, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 0); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 0); + FfxFloat32x4 v1 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v1, 6, slice); + + tex = FfxInt32x2(x * 4 + 0, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 0, y * 2 + 1); + FfxFloat32x4 v2 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v2, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 1); + FfxFloat32x4 v3 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v3, 6, slice); + + if (mips <= 7) + return; + // no barrier needed, working on values only from the same thread + + FfxFloat32x4 v = SpdReduce4(v0, v1, v2, v3); + SpdStore(FfxInt32x2(x, y), v, 7, slice); + SpdStoreIntermediate(x, y, v); +} + +void SpdDownsampleNextFour(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 baseMip, FfxUInt32 mips, FfxUInt32 slice) +{ + if (mips <= baseMip) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_2(x, y, workGroupID, localInvocationIndex, baseMip, slice); + + if (mips <= baseMip + 1) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_3(x, y, workGroupID, localInvocationIndex, baseMip + 1, slice); + + if (mips <= baseMip + 2) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_4(x, y, workGroupID, localInvocationIndex, baseMip + 2, slice); + + if (mips <= baseMip + 3) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_5(workGroupID, localInvocationIndex, baseMip + 3, slice); +} + +void SpdDownsample(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice) +{ + FfxUInt32x2 sub_xy = ffxRemapForWaveReduction(localInvocationIndex % 64); + FfxUInt32 x = sub_xy.x + 8 * ((localInvocationIndex >> 6) % 2); + FfxUInt32 y = sub_xy.y + 8 * ((localInvocationIndex >> 7)); + SpdDownsampleMips_0_1(x, y, workGroupID, localInvocationIndex, mips, slice); + + SpdDownsampleNextFour(x, y, workGroupID, localInvocationIndex, 2, mips, slice); + + if (mips <= 6) + return; + + if (SpdExitWorkgroup(numWorkGroups, localInvocationIndex, slice)) + return; + + SpdResetAtomicCounter(slice); + + // After mip 6 there is only a single workgroup left that downsamples the remaining up to 64x64 texels. + SpdDownsampleMips_6_7(x, y, mips, slice); + + SpdDownsampleNextFour(x, y, FfxUInt32x2(0, 0), localInvocationIndex, 8, mips, slice); +} + +void SpdDownsample(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice, FfxUInt32x2 workGroupOffset) +{ + SpdDownsample(workGroupID + workGroupOffset, localInvocationIndex, mips, numWorkGroups, slice); +} + +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +//============================================================================================================================== +// PACKED VERSION +//============================================================================================================================== + +#if FFX_HALF + +#ifdef FFX_GLSL +#extension GL_EXT_shader_subgroup_extended_types_float16:require +#endif + +FfxFloat16x4 SpdReduceQuadH(FfxFloat16x4 v) +{ +#if defined(FFX_GLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + FfxFloat16x4 v0 = v; + FfxFloat16x4 v1 = subgroupQuadSwapHorizontal(v); + FfxFloat16x4 v2 = subgroupQuadSwapVertical(v); + FfxFloat16x4 v3 = subgroupQuadSwapDiagonal(v); + return SpdReduce4H(v0, v1, v2, v3); +#elif defined(FFX_HLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + // requires SM6.0 + FfxUInt32 quad = WaveGetLaneIndex() & (~0x3); + FfxFloat16x4 v0 = v; + FfxFloat16x4 v1 = WaveReadLaneAt(v, quad | 1); + FfxFloat16x4 v2 = WaveReadLaneAt(v, quad | 2); + FfxFloat16x4 v3 = WaveReadLaneAt(v, quad | 3); + return SpdReduce4H(v0, v1, v2, v3); +/* + // if SM6.0 is not available, you can use the AMD shader intrinsics + // the AMD shader intrinsics are available in AMD GPU Services (AGS) library: + // https://gpuopen.com/amd-gpu-services-ags-library/ + // works for DX11 + FfxFloat16x4 v0 = v; + FfxFloat16x4 v1; + v1.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + FfxFloat16x4 v2; + v2.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + FfxFloat16x4 v3; + v3.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + return SpdReduce4H(v0, v1, v2, v3); + */ +#endif + return FfxFloat16x4(0.0, 0.0, 0.0, 0.0); +} + +FfxFloat16x4 SpdReduceIntermediateH(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3) +{ + FfxFloat16x4 v0 = SpdLoadIntermediateH(i0.x, i0.y); + FfxFloat16x4 v1 = SpdLoadIntermediateH(i1.x, i1.y); + FfxFloat16x4 v2 = SpdLoadIntermediateH(i2.x, i2.y); + FfxFloat16x4 v3 = SpdLoadIntermediateH(i3.x, i3.y); + return SpdReduce4H(v0, v1, v2, v3); +} + +FfxFloat16x4 SpdReduceLoad4H(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat16x4 v0 = SpdLoadH(FfxInt32x2(i0), slice); + FfxFloat16x4 v1 = SpdLoadH(FfxInt32x2(i1), slice); + FfxFloat16x4 v2 = SpdLoadH(FfxInt32x2(i2), slice); + FfxFloat16x4 v3 = SpdLoadH(FfxInt32x2(i3), slice); + return SpdReduce4H(v0, v1, v2, v3); +} + +FfxFloat16x4 SpdReduceLoad4H(FfxUInt32x2 base, FfxUInt32 slice) +{ + return SpdReduceLoad4H(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +} + +FfxFloat16x4 SpdReduceLoadSourceImage4H(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat16x4 v0 = SpdLoadSourceImageH(FfxInt32x2(i0), slice); + FfxFloat16x4 v1 = SpdLoadSourceImageH(FfxInt32x2(i1), slice); + FfxFloat16x4 v2 = SpdLoadSourceImageH(FfxInt32x2(i2), slice); + FfxFloat16x4 v3 = SpdLoadSourceImageH(FfxInt32x2(i3), slice); + return SpdReduce4H(v0, v1, v2, v3); +} + +FfxFloat16x4 SpdReduceLoadSourceImageH(FfxUInt32x2 base, FfxUInt32 slice) +{ +#ifdef SPD_LINEAR_SAMPLER + return SpdLoadSourceImageH(FfxInt32x2(base), slice); +#else + return SpdReduceLoadSourceImage4H(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +#endif +} + +void SpdDownsampleMips_0_1_IntrinsicsH(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxFloat16x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[3], 0, slice); + + if (mips <= 1) + return; + + v[0] = SpdReduceQuadH(v[0]); + v[1] = SpdReduceQuadH(v[1]); + v[2] = SpdReduceQuadH(v[2]); + v[3] = SpdReduceQuadH(v[3]); + + if ((localInvocationIndex % 4) == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2), v[0], 1, slice); + SpdStoreIntermediateH(x / 2, y / 2, v[0]); + + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2), v[1], 1, slice); + SpdStoreIntermediateH(x / 2 + 8, y / 2, v[1]); + + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2 + 8), v[2], 1, slice); + SpdStoreIntermediateH(x / 2, y / 2 + 8, v[2]); + + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2 + 8), v[3], 1, slice); + SpdStoreIntermediateH(x / 2 + 8, y / 2 + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1_LDSH(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxFloat16x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[3], 0, slice); + + if (mips <= 1) + return; + + for (FfxInt32 i = 0; i < 4; i++) + { + SpdStoreIntermediateH(x, y, v[i]); + SpdWorkgroupShuffleBarrier(); + if (localInvocationIndex < 64) + { + v[i] = SpdReduceIntermediateH(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x + (i % 2) * 8, y + (i / 2) * 8), v[i], 1, slice); + } + SpdWorkgroupShuffleBarrier(); + } + + if (localInvocationIndex < 64) + { + SpdStoreIntermediateH(x + 0, y + 0, v[0]); + SpdStoreIntermediateH(x + 8, y + 0, v[1]); + SpdStoreIntermediateH(x + 0, y + 8, v[2]); + SpdStoreIntermediateH(x + 8, y + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + SpdDownsampleMips_0_1_LDSH(x, y, workGroupID, localInvocationIndex, mips, slice); +#else + SpdDownsampleMips_0_1_IntrinsicsH(x, y, workGroupID, localInvocationIndex, mips, slice); +#endif +} + + +void SpdDownsampleMip_2H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 64) + { + FfxFloat16x4 v = SpdReduceIntermediateH(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS, try to reduce bank conflicts + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 x + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // ... + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + SpdStoreIntermediateH(x * 2 + y % 2, y * 2, v); + } +#else + FfxFloat16x4 v = SpdLoadIntermediateH(x, y); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediateH(x + (y / 2) % 2, y, v); + } +#endif +} + +void SpdDownsampleMip_3H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 16) + { + // x 0 x 0 + // 0 0 0 0 + // 0 x 0 x + // 0 0 0 0 + FfxFloat16x4 v = + SpdReduceIntermediateH(FfxUInt32x2(x * 4 + 0 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 2 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 0 + 1, y * 4 + 2), FfxUInt32x2(x * 4 + 2 + 1, y * 4 + 2)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 + // ... + // 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 + // ... + // 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x + // ... + SpdStoreIntermediateH(x * 4 + y, y * 4, v); + } +#else + if (localInvocationIndex < 64) + { + FfxFloat16x4 v = SpdLoadIntermediateH(x * 2 + y % 2, y * 2); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediateH(x * 2 + y / 2, y * 2, v); + } + } +#endif +} + +void SpdDownsampleMip_4H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 4) + { + // x 0 0 0 x 0 0 0 + // ... + // 0 x 0 0 0 x 0 0 + FfxFloat16x4 v = SpdReduceIntermediateH(FfxUInt32x2(x * 8 + 0 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 4 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 0 + 1 + y * 2, y * 8 + 4), + FfxUInt32x2(x * 8 + 4 + 1 + y * 2, y * 8 + 4)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x x x x 0 ... + // 0 ... + SpdStoreIntermediateH(x + y * 2, 0, v); + } +#else + if (localInvocationIndex < 16) + { + FfxFloat16x4 v = SpdLoadIntermediateH(x * 4 + y, y * 4); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediateH(x / 2 + y, 0, v); + } + } +#endif +} + +void SpdDownsampleMip_5H(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 1) + { + // x x x x 0 ... + // 0 ... + FfxFloat16x4 v = SpdReduceIntermediateH(FfxUInt32x2(0, 0), FfxUInt32x2(1, 0), FfxUInt32x2(2, 0), FfxUInt32x2(3, 0)); + SpdStoreH(FfxInt32x2(workGroupID.xy), v, mip, slice); + } +#else + if (localInvocationIndex < 4) + { + FfxFloat16x4 v = SpdLoadIntermediateH(localInvocationIndex, 0); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy), v, mip, slice); + } + } +#endif +} + +void SpdDownsampleMips_6_7H(FfxUInt32 x, FfxUInt32 y, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxInt32x2 tex = FfxInt32x2(x * 4 + 0, y * 4 + 0); + FfxInt32x2 pix = FfxInt32x2(x * 2 + 0, y * 2 + 0); + FfxFloat16x4 v0 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v0, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 0); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 0); + FfxFloat16x4 v1 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v1, 6, slice); + + tex = FfxInt32x2(x * 4 + 0, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 0, y * 2 + 1); + FfxFloat16x4 v2 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v2, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 1); + FfxFloat16x4 v3 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v3, 6, slice); + + if (mips < 8) + return; + // no barrier needed, working on values only from the same thread + + FfxFloat16x4 v = SpdReduce4H(v0, v1, v2, v3); + SpdStoreH(FfxInt32x2(x, y), v, 7, slice); + SpdStoreIntermediateH(x, y, v); +} + +void SpdDownsampleNextFourH(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 baseMip, FfxUInt32 mips, FfxUInt32 slice) +{ + if (mips <= baseMip) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_2H(x, y, workGroupID, localInvocationIndex, baseMip, slice); + + if (mips <= baseMip + 1) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_3H(x, y, workGroupID, localInvocationIndex, baseMip + 1, slice); + + if (mips <= baseMip + 2) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_4H(x, y, workGroupID, localInvocationIndex, baseMip + 2, slice); + + if (mips <= baseMip + 3) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_5H(workGroupID, localInvocationIndex, baseMip + 3, slice); +} + +void SpdDownsampleH(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice) +{ + FfxUInt32x2 sub_xy = ffxRemapForWaveReduction(localInvocationIndex % 64); + FfxUInt32 x = sub_xy.x + 8 * ((localInvocationIndex >> 6) % 2); + FfxUInt32 y = sub_xy.y + 8 * ((localInvocationIndex >> 7)); + + SpdDownsampleMips_0_1H(x, y, workGroupID, localInvocationIndex, mips, slice); + + SpdDownsampleNextFourH(x, y, workGroupID, localInvocationIndex, 2, mips, slice); + + if (mips < 7) + return; + + if (SpdExitWorkgroup(numWorkGroups, localInvocationIndex, slice)) + return; + + SpdResetAtomicCounter(slice); + + // After mip 6 there is only a single workgroup left that downsamples the remaining up to 64x64 texels. + SpdDownsampleMips_6_7H(x, y, mips, slice); + + SpdDownsampleNextFourH(x, y, FfxUInt32x2(0, 0), localInvocationIndex, 8, mips, slice); +} + +void SpdDownsampleH(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice, FfxUInt32x2 workGroupOffset) +{ + SpdDownsampleH(workGroupID + workGroupOffset, localInvocationIndex, mips, numWorkGroups, slice); +} + +#endif // #if FFX_HALF +#endif // #ifdef FFX_GPU