vo_gpu: vulkan: initial implementation

This time based on ra/vo_gpu. 2017 is the year of the vulkan desktop!

Current problems / limitations / improvement opportunities:

1. The swapchain/flipping code violates the vulkan spec, by assuming
   that the presentation queue will be bounded (in cases where rendering
   is significantly faster than vsync). But apparently, there's simply
   no better way to do this right now, to the point where even the
   stupid cube.c examples from LunarG etc. do it wrong.
   (cf. KhronosGroup/Vulkan-Docs#370)

2. The memory allocator could be improved. (This is a universal
   constant)

3. Could explore using push descriptors instead of descriptor sets,
   especially since we expect to switch descriptors semi-often for some
   passes (like interpolation). Probably won't make a difference, but
   the synchronization overhead might be a factor. Who knows.

4. Parallelism across frames / async transfer is not well-defined, we
   either need to use a better semaphore / command buffer strategy or a
   resource pooling layer to safely handle cross-frame parallelism.
   (That said, I gave resource pooling a try and was not happy with the
   result at all - so I'm still exploring the semaphore strategy)

5. We aggressively use pipeline barriers where events would offer a much
   more fine-grained synchronization mechanism. As a result of this, we
   might be suffering from GPU bubbles due to too-short dependencies on
   objects. (That said, I'm also exploring the use of semaphores as a an
   ordering tactic which would allow cross-frame time slicing in theory)

Some minor changes to the vo_gpu and infrastructure, but nothing
consequential.

options/options.c

@@ -89,6 +89,7 @@ extern const struct m_obj_list vo_obj_list;
 extern const struct m_obj_list ao_obj_list;
 
 extern const struct m_sub_options opengl_conf;
+extern const struct m_sub_options vulkan_conf;
 extern const struct m_sub_options angle_conf;
 extern const struct m_sub_options cocoa_conf;
 
@@ -690,6 +691,10 @@ const m_option_t mp_opts[] = {
     OPT_SUBSTRUCT("", opengl_opts, opengl_conf, 0),
 #endif
 
+#if HAVE_VULKAN
+    OPT_SUBSTRUCT("", vulkan_opts, vulkan_conf, 0),
+#endif
+
 #if HAVE_EGL_ANGLE_WIN32
     OPT_SUBSTRUCT("", angle_opts, angle_conf, 0),
 #endif

options/options.h

@@ -329,6 +329,7 @@ typedef struct MPOpts {
     struct gl_video_opts *gl_video_opts;
     struct angle_opts *angle_opts;
     struct opengl_opts *opengl_opts;
+    struct vulkan_opts *vulkan_opts;
     struct cocoa_opts *cocoa_opts;
     struct dvd_opts *dvd_opts;
 

video/out/gpu/context.c

@@ -44,6 +44,7 @@ extern const struct ra_ctx_fns ra_ctx_dxgl;
 extern const struct ra_ctx_fns ra_ctx_rpi;
 extern const struct ra_ctx_fns ra_ctx_mali;
 extern const struct ra_ctx_fns ra_ctx_vdpauglx;
+extern const struct ra_ctx_fns ra_ctx_vulkan_xlib;
 
 static const struct ra_ctx_fns *contexts[] = {
 // OpenGL contexts:
@@ -83,6 +84,13 @@ static const struct ra_ctx_fns *contexts[] = {
 #if HAVE_VDPAU_GL_X11
     &ra_ctx_vdpauglx,
 #endif
+
+// Vulkan contexts:
+#if HAVE_VULKAN
+#if HAVE_X11
+    &ra_ctx_vulkan_xlib,
+#endif
+#endif
 };
 
 static bool get_help(struct mp_log *log, struct bstr param)

video/out/gpu/ra.h

@@ -146,6 +146,7 @@ enum ra_buf_type {
     RA_BUF_TYPE_TEX_UPLOAD,     // texture upload buffer (pixel buffer object)
     RA_BUF_TYPE_SHADER_STORAGE, // shader buffer (SSBO), for RA_VARTYPE_BUF_RW
     RA_BUF_TYPE_UNIFORM,        // uniform buffer (UBO), for RA_VARTYPE_BUF_RO
+    RA_BUF_TYPE_VERTEX,         // not publicly usable (RA-internal usage)
 };
 
 struct ra_buf_params {
@@ -369,10 +370,10 @@ struct ra_fns {
 
     void (*buf_destroy)(struct ra *ra, struct ra_buf *buf);
 
-    // Update the contents of a buffer, starting at a given offset and up to a
-    // given size, with the contents of *data. This is an extremely common
-    // operation. Calling this while the buffer is considered "in use" is an
-    // error. (See: buf_poll)
+    // Update the contents of a buffer, starting at a given offset (*must* be a
+    // multiple of 4) and up to a given size, with the contents of *data. This
+    // is an extremely common operation. Calling this while the buffer is
+    // considered "in use" is an error. (See: buf_poll)
     void (*buf_update)(struct ra *ra, struct ra_buf *buf, ptrdiff_t offset,
                        const void *data, size_t size);
 

video/out/vo_gpu.c

@@ -60,6 +60,7 @@ struct gpu_priv {
 static void resize(struct gpu_priv *p)
 {
     struct vo *vo = p->vo;
+    struct ra_swapchain *sw = p->ctx->swapchain;
 
     MP_VERBOSE(vo, "Resize: %dx%d\n", vo->dwidth, vo->dheight);
 
@@ -69,6 +70,11 @@ static void resize(struct gpu_priv *p)
 
     gl_video_resize(p->renderer, &src, &dst, &osd);
 
+    int fb_depth = sw->fns->color_depth ? sw->fns->color_depth(sw) : 0;
+    if (fb_depth)
+        MP_VERBOSE(p, "Reported display depth: %d\n", fb_depth);
+    gl_video_set_fb_depth(p->renderer, fb_depth);
+
     vo->want_redraw = true;
 }
 
@@ -289,7 +295,6 @@ static int preinit(struct vo *vo)
         goto err_out;
     assert(p->ctx->ra);
     assert(p->ctx->swapchain);
-    struct ra_swapchain *sw = p->ctx->swapchain;
 
     p->renderer = gl_video_init(p->ctx->ra, vo->log, vo->global);
     gl_video_set_osd_source(p->renderer, vo->osd);
@@ -305,11 +310,6 @@ static int preinit(struct vo *vo)
                              vo->hwdec_devs, vo->opts->gl_hwdec_interop);
     gl_video_set_hwdec(p->renderer, p->hwdec);
 
-    int fb_depth = sw->fns->color_depth ? sw->fns->color_depth(sw) : 0;
-    if (fb_depth)
-        MP_VERBOSE(p, "Reported display depth: %d\n", fb_depth);
-    gl_video_set_fb_depth(p->renderer, fb_depth);
-
     return 0;
 
 err_out:

video/out/vulkan/common.h

@@ -0,0 +1,52 @@
+#pragma once
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <assert.h>
+
+#include "config.h"
+
+#include "common/common.h"
+#include "common/msg.h"
+
+// We need to define all platforms we want to support. Since we have
+// our own mechanism for checking this, we re-define the right symbols
+#if HAVE_X11
+#define VK_USE_PLATFORM_XLIB_KHR
+#endif
+
+#include <vulkan/vulkan.h>
+
+// Vulkan allows the optional use of a custom allocator. We don't need one but
+// mark this parameter with a better name in case we ever decide to change this
+// in the future. (And to make the code more readable)
+#define MPVK_ALLOCATOR NULL
+
+// A lot of things depend on streaming resources across frames. Depending on
+// how many frames we render ahead of time, we need to pick enough to avoid
+// any conflicts, so make all of these tunable relative to this constant in
+// order to centralize them.
+#define MPVK_MAX_STREAMING_DEPTH 8
+
+// Shared struct used to hold vulkan context information
+struct mpvk_ctx {
+    struct mp_log *log;
+    VkInstance inst;
+    VkPhysicalDevice physd;
+    VkDebugReportCallbackEXT dbg;
+    VkDevice dev;
+
+    // Surface, must be initialized fter the context itself
+    VkSurfaceKHR surf;
+    VkSurfaceFormatKHR surf_format; // picked at surface initialization time
+
+    struct vk_malloc *alloc; // memory allocator for this device
+    struct vk_cmdpool *pool; // primary command pool for this device
+    struct vk_cmdpool *pool_transfer; // pool for async transfers (optional)
+    struct vk_cmd *last_cmd; // most recently submitted command
+
+    // Cached capabilities
+    VkPhysicalDeviceLimits limits;
+};