rendervulkan: add facilities for vulkan profiling w/ Tracy profiler

src/meson.build

@@ -43,7 +43,7 @@ libdecor_dep = dependency('libdecor-0')
 tracy_public_deps = []
 tracy_public_args = []
 if get_option('tracy_enable')
-	tracy_public_args += ['-DTRACY_ENABLE']
+	tracy_public_args += ['-DTRACY_ENABLE', '-DTRACY_VK_USE_SYMBOL_TABLE']
 
 	if get_option('tracy_on_demand')
 		tracy_public_args += ['-DTRACY_ON_DEMAND']
@@ -73,7 +73,7 @@ if get_option('tracy_enable')
 endif
 
 tracy_source = ['../thirdparty/tracy/public/TracyClient.cpp']
-tracy_include = include_directories(['../thirdparty/tracy/public'])
+tracy_include = include_directories(['../thirdparty/tracy/public', '../thirdparty/tracy/public/tracy'])
 
 tracy_dep = declare_dependency(
 	compile_args: ['-Wno-sign-compare'] + tracy_public_args,

src/rendervulkan.cpp

@@ -1196,9 +1196,26 @@ int32_t CVulkanDevice::findMemoryType( VkMemoryPropertyFlags properties, uint32_
 	return -1;
 }
 
-std::unique_ptr<CVulkanCmdBuffer> CVulkanDevice::commandBuffer()
+inline std::unique_ptr<CVulkanCmdBuffer> __attribute__((hot,visibility("internal"))) CVulkanDevice::commandBuffer([[maybe_unused]] const std::source_location& loc)
 {
-	std::unique_ptr<CVulkanCmdBuffer> cmdBuffer;
+	auto finalize_buf = [this, &loc]<bool bIsRecycled>(std::unique_ptr<CVulkanCmdBuffer> cmdBuf) {
+		//using this lambda allows for Named Return Value Optimization (NRVO) w/o duplicating code
+		if constexpr (bIsRecycled)
+			m_unusedCmdBufs.pop_back();
+
+		cmdBuf->begin();
+
+#ifdef TRACY_ENABLE
+		assert( !cmdBuf->gpuZoneHolder() );	
+		auto source_data = tracy::SourceLocationData(loc.function_name(), loc.function_name(), loc.file_name(), loc.line());
+
+		//we don't need to worry about dangling pointer to source_data,
+		//the gpu zone object held in gpuZoneHolder will only touch the pointer during its construction w/ emplace() 
+		cmdBuf->gpuZoneHolder().emplace(cmdBuf->tracyCtx(), &source_data, cmdBuf->rawBuffer(), true);
+#endif
+		return cmdBuf;
+	};
+
 	if (m_unusedCmdBufs.empty())
 	{
 		VkCommandBuffer rawCmdBuffer;
@@ -1216,16 +1233,12 @@ std::unique_ptr<CVulkanCmdBuffer> CVulkanDevice::commandBuffer()
 			return nullptr;
 		}
 
-		cmdBuffer = std::make_unique<CVulkanCmdBuffer>(this, rawCmdBuffer, queue(), queueFamily());
+		return finalize_buf.operator()<false>(std::make_unique<CVulkanCmdBuffer>(this, rawCmdBuffer, queue(), queueFamily()));
 	}
 	else
 	{
-		cmdBuffer = std::move(m_unusedCmdBufs.back());
-		m_unusedCmdBufs.pop_back();
+		return finalize_buf.operator()<true>(std::move(m_unusedCmdBufs.back()));
 	}
-
-	cmdBuffer->begin();
-	return cmdBuffer;
 }
 
 uint64_t CVulkanDevice::submitInternal( CVulkanCmdBuffer* cmdBuffer )
@@ -1318,13 +1331,22 @@ void CVulkanDevice::resetCmdBuffers(uint64_t sequence)
 	m_pendingCmdBufs.erase(m_pendingCmdBufs.begin(), ++last);
 }
 
+
+
 CVulkanCmdBuffer::CVulkanCmdBuffer(CVulkanDevice *parent, VkCommandBuffer cmdBuffer, VkQueue queue, uint32_t queueFamily)
 	: m_cmdBuffer(cmdBuffer), m_device(parent), m_queue(queue), m_queueFamily(queueFamily)
+#ifdef TRACY_ENABLE 
+	, m_tracyCtx{tracy::CreateVkContext(parent->instance(), parent->physDev(), parent->device(), queue, cmdBuffer, g_pfn_vkGetInstanceProcAddr, parent->vk.GetDeviceProcAddr)}
+#endif
 {
 }
 
 CVulkanCmdBuffer::~CVulkanCmdBuffer()
 {
+#ifdef TRACY_ENABLE
+	TracyVkDestroy(m_tracyCtx);
+#endif
+
 	m_device->vk.FreeCommandBuffers(m_device->device(), m_device->commandPool(), 1, &m_cmdBuffer);
 }
 
@@ -1333,6 +1355,9 @@ void CVulkanCmdBuffer::reset()
 	vk_check( m_device->vk.ResetCommandBuffer(m_cmdBuffer, 0) );
 	m_textureRefs.clear();
 	m_textureState.clear();
+#ifdef TRACY_ENABLE
+	m_gpuZoneHolder.reset();
+#endif
 }
 
 void CVulkanCmdBuffer::begin()
@@ -3349,7 +3374,6 @@ gamescope::OwningRc<CVulkanTexture> vulkan_create_texture_from_bits( uint32_t wi
 	memcpy( g_device.uploadBufferData(size), bits, size );
 
 	auto cmdBuffer = g_device.commandBuffer();
-
 	cmdBuffer->copyBufferToImage(g_device.uploadBuffer(), 0, 0, pTex.get());
 	// TODO: Sync this copyBufferToImage.
 

src/rendervulkan.hpp

@@ -10,6 +10,7 @@
 #include <bitset>
 #include <mutex>
 #include <optional>
+#include <source_location>
 
 #include "main.hpp"
 #include "color_helpers.h"
@@ -18,6 +19,11 @@
 
 #include "shaders/descriptor_set_constants.h"
 
+namespace tracy { //forward declaration
+	class VkCtx;
+	class VkCtxScope;
+}
+
 class CVulkanCmdBuffer;
 
 // 1: Fade Plane (Fade outs between switching focus)
@@ -720,7 +726,7 @@ class CVulkanDevice
 	VkSampler sampler(SamplerState key);
 	VkPipeline pipeline(ShaderType type, uint32_t layerCount = 1, uint32_t ycbcrMask = 0, uint32_t blur_layers = 0, uint32_t colorspace_mask = 0, uint32_t output_eotf = EOTF_Gamma22, bool itm_enable = false);
 	int32_t findMemoryType( VkMemoryPropertyFlags properties, uint32_t requiredTypeBits );
-	std::unique_ptr<CVulkanCmdBuffer> commandBuffer();
+	inline std::unique_ptr<CVulkanCmdBuffer> __attribute__((hot,visibility("internal"))) commandBuffer([[maybe_unused]] const std::source_location& loc = std::source_location::current());
 	uint64_t submit( std::unique_ptr<CVulkanCmdBuffer> cmdBuf);
 	uint64_t submitInternal( CVulkanCmdBuffer* cmdBuf );
 	void wait(uint64_t sequence, bool reset = true);
@@ -895,7 +901,14 @@ class CVulkanCmdBuffer
 
 	VkQueue queue() { return m_queue; }
 	uint32_t queueFamily() { return m_queueFamily; }
-
+
+protected:
+	friend class CVulkanDevice;
+#ifdef TRACY_ENABLE
+	inline std::optional<tracy::VkCtxScope>& gpuZoneHolder() {return m_gpuZoneHolder;}
+	inline tracy::VkCtx* tracyCtx() {return m_tracyCtx;}
+#endif
+
 private:
 	VkCommandBuffer m_cmdBuffer;
 	CVulkanDevice *m_device;
@@ -917,6 +930,10 @@ class CVulkanCmdBuffer
 	std::array<CVulkanTexture *, VKR_LUT3D_COUNT> m_lut3D;
 
 	uint32_t m_renderBufferOffset = 0;
+#ifdef TRACY_ENABLE
+	tracy::VkCtx* m_tracyCtx;
+	std::optional<tracy::VkCtxScope> m_gpuZoneHolder;
+#endif
 };
 
 uint32_t VulkanFormatToDRM( VkFormat vkFormat );

src/vulkan_include.h

@@ -2,3 +2,4 @@
 
 #define VK_NO_PROTOTYPES
 #include <vulkan/vulkan_core.h>
+#include "tracy/TracyVulkan.hpp"