Merge pull request hrydgard#19748 from hrydgard/software-depth-proto

Render a software depth buffer in parallel with HW rendering

Author: hrydgard

CMakeLists.txt

@@ -1906,6 +1906,8 @@ set(GPU_SOURCES
 	GPU/Common/Draw2D.cpp
 	GPU/Common/Draw2D.h
 	GPU/Common/DepthBufferCommon.cpp
+	GPU/Common/DepthRaster.cpp
+	GPU/Common/DepthRaster.h
 	GPU/Common/TextureShaderCommon.cpp
 	GPU/Common/TextureShaderCommon.h
 	GPU/Common/DepalettizeShaderCommon.cpp

Common/Data/Convert/ColorConv.cpp

@@ -65,7 +65,7 @@ void ConvertBGRA8888ToRGB888(u8 *dst, const u32 *src, u32 numPixels) {
 }
 
 #if PPSSPP_ARCH(SSE2)
-// fp64's improved version, see #19751
+// fp64's improved SSE2 version, see #19751. SSE4 no longer required here.
 static inline void ConvertRGBA8888ToRGBA5551(__m128i *dstp, const __m128i *srcp, u32 sseChunks) {
 	const __m128i maskRB = _mm_set1_epi32(0x00F800F8);
 	const __m128i maskGA = _mm_set1_epi32(0x8000F800);
@@ -76,7 +76,7 @@ static inline void ConvertRGBA8888ToRGBA5551(__m128i *dstp, const __m128i *srcp,
 		__m128i c0 = _mm_load_si128(&srcp[i + 0]);
 		__m128i c1 = _mm_load_si128(&srcp[i + 1]);
 
-		__m128i rb0 = _mm_and_si128(c0, maskRB);              // 00000000bbbbb00000000000rrrrr000
+		__m128i rb0 = _mm_and_si128(c0, maskRB);              // 00000000bbbbb00000000000rrrrr000 (each 32-bit lane)
 		__m128i rb1 = _mm_and_si128(c1, maskRB);              // 00000000bbbbb00000000000rrrrr000
 		__m128i ga0 = _mm_and_si128(c0, maskGA);              // a000000000000000ggggg00000000000
 		__m128i ga1 = _mm_and_si128(c1, maskGA);              // a000000000000000ggggg00000000000

Common/GPU/DataFormat.h

@@ -74,6 +74,7 @@ bool DataFormatIsDepthStencil(DataFormat fmt);
 inline bool DataFormatIsColor(DataFormat fmt) {
 	return !DataFormatIsDepthStencil(fmt);
 }
+int DataFormatNumChannels(DataFormat fmt);
 bool DataFormatIsBlockCompressed(DataFormat fmt, int *blockSize);
 
 // Limited format support for now.

Common/GPU/Vulkan/thin3d_vulkan.cpp

@@ -803,9 +803,15 @@ bool VKTexture::Create(VkCommandBuffer cmd, VulkanBarrierBatch *postBarriers, Vu
 	}
 
 	VkComponentMapping r8AsAlpha[4] = { VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_ONE, VK_COMPONENT_SWIZZLE_R };
+	VkComponentMapping r8AsColor[4] = { VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_ONE };
 
+	VkComponentMapping *swizzle = nullptr;
+	switch (desc.swizzle) {
+	case TextureSwizzle::R8_AS_ALPHA: swizzle = r8AsAlpha; break;
+	case TextureSwizzle::R8_AS_GRAYSCALE: swizzle = r8AsColor; break;
+	}
 	VulkanBarrierBatch barrier;
-	if (!vkTex_->CreateDirect(width_, height_, 1, mipLevels_, vulkanFormat, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, usageBits, &barrier, desc.swizzle == TextureSwizzle::R8_AS_ALPHA ? r8AsAlpha : nullptr)) {
+	if (!vkTex_->CreateDirect(width_, height_, 1, mipLevels_, vulkanFormat, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, usageBits, &barrier, swizzle)) {
 		ERROR_LOG(Log::G3D,  "Failed to create VulkanTexture: %dx%dx%d fmt %d, %d levels", width_, height_, depth_, (int)vulkanFormat, mipLevels_);
 		return false;
 	}

Common/GPU/thin3d.cpp

@@ -118,6 +118,25 @@ bool DataFormatIsBlockCompressed(DataFormat fmt, int *blockSize) {
 	}
 }
 
+int DataFormatNumChannels(DataFormat fmt) {
+	switch (fmt) {
+	case DataFormat::D16:
+	case DataFormat::D32F:
+	case DataFormat::R8_UNORM:
+	case DataFormat::R16_UNORM:
+	case DataFormat::R16_FLOAT:
+	case DataFormat::R32_FLOAT:
+		return 1;
+	case DataFormat::R8G8B8A8_UNORM:
+	case DataFormat::R8G8B8A8_UNORM_SRGB:
+	case DataFormat::B8G8R8A8_UNORM:
+	case DataFormat::B8G8R8A8_UNORM_SRGB:
+		return 4;
+	default:
+		return 0;
+	}
+}
+
 RefCountedObject::~RefCountedObject() {
 	const int rc = refcount_.load();
 	_dbg_assert_msg_(rc == 0xDEDEDE, "Unexpected refcount %d in object of type '%s'", rc, name_);

Common/GPU/thin3d.h

@@ -643,6 +643,7 @@ typedef std::function<bool(uint8_t *data, const uint8_t *initData, uint32_t w, u
 enum class TextureSwizzle {
 	DEFAULT,
 	R8_AS_ALPHA,
+	R8_AS_GRAYSCALE,
 };
 
 struct TextureDesc {