Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Vulkan state optimizations #9899

Merged
merged 3 commits into from
Aug 14, 2017
Merged

Vulkan state optimizations #9899

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5a65334
Use dirty-flags more in Vulkan state setting
hrydgard Aug 14, 2017
a1fd2d1
Dirty-track the current pipeline to avoid many vkBindPipeline calls.
hrydgard Aug 14, 2017
000723f
Fix minor blendstate oversight, restore a disabled optimization
hrydgard Aug 14, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 4 additions & 5 deletions GPU/D3D11/StateMappingD3D11.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -457,10 +457,6 @@ void DrawEngineD3D11::ApplyDrawStateLate(bool applyStencilRef, uint8_t stencilRe
textureCache_->ApplyTexture();
}

// Need to do this AFTER ApplyTexture because the process of depallettization can ruin the blend state.
float blendColor[4];
Uint8x4ToFloat4(blendColor, dynState_.blendColor);

// we go through Draw here because it automatically handles screen rotation, as needed in UWP on mobiles.
if (gstate_c.IsDirty(DIRTY_VIEWPORTSCISSOR_STATE)) {
draw_->SetViewports(1, &dynState_.viewport);
Expand All @@ -470,13 +466,16 @@ void DrawEngineD3D11::ApplyDrawStateLate(bool applyStencilRef, uint8_t stencilRe
context_->RSSetState(rasterState_);
}
if (gstate_c.IsDirty(DIRTY_BLEND_STATE)) {
// Need to do this AFTER ApplyTexture because the process of depallettization can ruin the blend state.
float blendColor[4];
Uint8x4ToFloat4(blendColor, dynState_.blendColor);
if (device1_) {
context1_->OMSetBlendState(blendState1_, blendColor, 0xFFFFFFFF);
} else {
context_->OMSetBlendState(blendState_, blendColor, 0xFFFFFFFF);
}
}
if (gstate_c.IsDirty(DIRTY_DEPTHSTENCIL_STATE)) {
if (gstate_c.IsDirty(DIRTY_DEPTHSTENCIL_STATE) || applyStencilRef) {
context_->OMSetDepthStencilState(depthStencilState_, applyStencilRef ? stencilRef : dynState_.stencilRef);
}
gstate_c.Clean(DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_DEPTHSTENCIL_STATE | DIRTY_RASTER_STATE | DIRTY_BLEND_STATE);
Expand Down
44 changes: 13 additions & 31 deletions GPU/Vulkan/DrawEngineVulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -569,6 +569,12 @@ void DrawEngineVulkan::DoFlush() {
gpuStats.numFlushes++;

VkCommandBuffer cmd = (VkCommandBuffer)draw_->GetNativeObject(Draw::NativeObject::RENDERPASS_COMMANDBUFFER);
if (cmd != lastCmd_) {
lastPipeline_ = nullptr;
lastCmd_ = cmd;
gstate_c.Dirty(DIRTY_VIEWPORTSCISSOR_STATE);
}

VkRenderPass rp = (VkRenderPass)draw_->GetNativeObject(Draw::NativeObject::CURRENT_RENDERPASS);
if (!rp)
Crash();
Expand Down Expand Up @@ -650,19 +656,7 @@ void DrawEngineVulkan::DoFlush() {
}

ConvertStateToVulkanKey(*framebufferManager_, shaderManager_, prim, pipelineKey_, dynState_);
// TODO: Dirty-flag these.
vkCmdSetScissor(cmd, 0, 1, &dynState_.scissor);
vkCmdSetViewport(cmd, 0, 1, &dynState_.viewport);
if (dynState_.useStencil) {
vkCmdSetStencilWriteMask(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilWriteMask);
vkCmdSetStencilCompareMask(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilCompareMask);
vkCmdSetStencilReference(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilRef);
}
if (dynState_.useBlendColor) {
float bc[4];
Uint8x4ToFloat4(bc, dynState_.blendColor);
vkCmdSetBlendConstants(cmd, bc);
}
ApplyDrawStateLate(cmd, false, 0);

dirtyUniforms_ |= shaderManager_->UpdateUniforms();

Expand All @@ -672,7 +666,10 @@ void DrawEngineVulkan::DoFlush() {
// Already logged, let's bail out.
return;
}
vkCmdBindPipeline(cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline->pipeline); // TODO: Avoid if same as last draw.
if (pipeline != lastPipeline_) {
vkCmdBindPipeline(cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline->pipeline); // TODO: Avoid if same as last draw.
lastPipeline_ = pipeline;
}

UpdateUBOs(frame);

Expand Down Expand Up @@ -751,23 +748,8 @@ void DrawEngineVulkan::DoFlush() {
}

ConvertStateToVulkanKey(*framebufferManager_, shaderManager_, prim, pipelineKey_, dynState_);
// TODO: Dirty-flag these.
vkCmdSetScissor(cmd, 0, 1, &dynState_.scissor);
vkCmdSetViewport(cmd, 0, 1, &dynState_.viewport);
if (dynState_.useStencil) {
vkCmdSetStencilWriteMask(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilWriteMask);
vkCmdSetStencilCompareMask(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilCompareMask);
}
if (result.setStencil) {
vkCmdSetStencilReference(cmd, VK_STENCIL_FRONT_AND_BACK, result.stencilValue);
} else if (dynState_.useStencil) {
vkCmdSetStencilReference(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilRef);
}
if (dynState_.useBlendColor) {
float bc[4];
Uint8x4ToFloat4(bc, dynState_.blendColor);
vkCmdSetBlendConstants(cmd, bc);
}
ApplyDrawStateLate(cmd, result.setStencil, result.stencilValue);

dirtyUniforms_ |= shaderManager_->UpdateUniforms();

shaderManager_->GetShaders(prim, lastVType_, &vshader, &fshader, useHWTransform);
Expand Down
5 changes: 4 additions & 1 deletion GPU/Vulkan/DrawEngineVulkan.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,6 +60,7 @@ typedef u32 ReliableHashType;

class VulkanContext;
class VulkanPushBuffer;
struct VulkanPipeline;

struct DrawEngineVulkanStats {
int pushUBOSpaceUsed;
Expand Down Expand Up @@ -127,7 +128,7 @@ class DrawEngineVulkan : public DrawEngineCommon {

private:
struct FrameData;
void ApplyDrawStateLate();
void ApplyDrawStateLate(VkCommandBuffer cmd, bool applyStencilRef, uint8_t stencilRef);
void ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManager, ShaderManagerVulkan *shaderManager, int prim, VulkanPipelineRasterStateKey &key, VulkanDynamicState &dynState);

void InitDeviceObjects();
Expand All @@ -146,6 +147,8 @@ class DrawEngineVulkan : public DrawEngineCommon {
// We use a single descriptor set layout for all PSP draws.
VkDescriptorSetLayout descriptorSetLayout_;
VkPipelineLayout pipelineLayout_;
VkCommandBuffer lastCmd_ = VK_NULL_HANDLE;
VulkanPipeline *lastPipeline_;

struct DescriptorSetKey {
VkImageView imageView_;
Expand Down
30 changes: 25 additions & 5 deletions GPU/Vulkan/StateMappingVulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@

#include "Common/Vulkan/VulkanLoader.h"

#include "math/dataconv.h"
#include "GPU/Math3D.h"
#include "GPU/GPUState.h"
#include "GPU/ge_constants.h"
Expand Down Expand Up @@ -132,7 +133,6 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;

if (gstate_c.IsDirty(DIRTY_BLEND_STATE)) {
gstate_c.Clean(DIRTY_BLEND_STATE);
// Unfortunately, this isn't implemented yet.
gstate_c.SetAllowShaderBlend(false);
if (gstate.isModeClear()) {
Expand Down Expand Up @@ -235,7 +235,6 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
}

if (gstate_c.IsDirty(DIRTY_RASTER_STATE)) {
gstate_c.Clean(DIRTY_RASTER_STATE);
if (gstate.isModeClear()) {
key.cullMode = VK_CULL_MODE_NONE;
} else {
Expand All @@ -246,7 +245,6 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
}

if (gstate_c.IsDirty(DIRTY_DEPTHSTENCIL_STATE)) {
gstate_c.Clean(DIRTY_DEPTHSTENCIL_STATE);
if (gstate.isModeClear()) {
key.depthTestEnable = true;
key.depthCompareOp = VK_COMPARE_OP_ALWAYS;
Expand Down Expand Up @@ -328,7 +326,6 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
}

if (gstate_c.IsDirty(DIRTY_VIEWPORTSCISSOR_STATE)) {
gstate_c.Clean(DIRTY_VIEWPORTSCISSOR_STATE);
ViewportAndScissor vpAndScissor;
ConvertViewportAndScissor(useBufferedRendering,
fbManager.GetRenderWidth(), fbManager.GetRenderHeight(),
Expand Down Expand Up @@ -373,7 +370,7 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
}


void DrawEngineVulkan::ApplyDrawStateLate() {
void DrawEngineVulkan::ApplyDrawStateLate(VkCommandBuffer cmd, bool applyStencilRef, uint8_t stencilRef) {
// At this point, we know if the vertices are full alpha or not.
// TODO: Set the nearest/linear here (since we correctly know if alpha/color tests are needed)?
if (!gstate.isModeClear()) {
Expand All @@ -388,4 +385,27 @@ void DrawEngineVulkan::ApplyDrawStateLate() {
}
*/
}

if (gstate_c.IsDirty(DIRTY_VIEWPORTSCISSOR_STATE)) {
vkCmdSetScissor(cmd, 0, 1, &dynState_.scissor);
vkCmdSetViewport(cmd, 0, 1, &dynState_.viewport);
}
if (gstate_c.IsDirty(DIRTY_DEPTHSTENCIL_STATE)) {
if (dynState_.useStencil) {
vkCmdSetStencilWriteMask(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilWriteMask);
vkCmdSetStencilCompareMask(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilCompareMask);
if (!applyStencilRef) {
vkCmdSetStencilReference(cmd, VK_STENCIL_FRONT_AND_BACK, dynState_.stencilRef);
}
}
}
if (applyStencilRef) {
vkCmdSetStencilReference(cmd, VK_STENCIL_FRONT_AND_BACK, stencilRef);
}
if (gstate_c.IsDirty(DIRTY_BLEND_STATE)) {
float bc[4];
Uint8x4ToFloat4(bc, dynState_.blendColor);
vkCmdSetBlendConstants(cmd, bc);
}
gstate_c.Clean(DIRTY_BLEND_STATE | DIRTY_DEPTHSTENCIL_STATE | DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE);
}
14 changes: 7 additions & 7 deletions GPU/ge_constants.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,11 +78,11 @@ enum GECommand {
GE_CMD_TGENMATRIXNUMBER = 0x40,
GE_CMD_TGENMATRIXDATA = 0x41,
GE_CMD_VIEWPORTXSCALE = 0x42,
GE_CMD_VIEWPORTYSCALE,
GE_CMD_VIEWPORTZSCALE,
GE_CMD_VIEWPORTXCENTER,
GE_CMD_VIEWPORTYCENTER,
GE_CMD_VIEWPORTZCENTER,
GE_CMD_VIEWPORTYSCALE = 0x43,
GE_CMD_VIEWPORTZSCALE = 0x44,
GE_CMD_VIEWPORTXCENTER = 0x45,
GE_CMD_VIEWPORTYCENTER = 0x46,
GE_CMD_VIEWPORTZCENTER = 0x47,
GE_CMD_TEXSCALEU = 0x48,
GE_CMD_TEXSCALEV = 0x49,
GE_CMD_TEXOFFSETU = 0x4A,
Expand Down Expand Up @@ -182,8 +182,8 @@ enum GECommand {
GE_CMD_TEXBUFWIDTH5,
GE_CMD_TEXBUFWIDTH6,
GE_CMD_TEXBUFWIDTH7,
GE_CMD_CLUTADDR=0xB0,
GE_CMD_CLUTADDRUPPER,
GE_CMD_CLUTADDR = 0xB0,
GE_CMD_CLUTADDRUPPER = 0xB1,
GE_CMD_TRANSFERSRC,
GE_CMD_TRANSFERSRCW,
GE_CMD_TRANSFERDST,
Expand Down
2 changes: 1 addition & 1 deletion ext/native/math/dataconv.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
#pragma once

#include <inttypes.h>
#include <cstdint>
#include <cstring>

#ifdef _M_SSE
Expand Down