Merge branch 'master' into feature_openxr_vulkan

Common/Data/Collections/TinySet.h

@@ -47,12 +47,12 @@ struct TinySet {
 		size_t otherSize = other.size();
 		if (size() + otherSize <= MaxFastSize) {
 			// Fast case
-			for (int i = 0; i < otherSize; i++) {
+			for (size_t i = 0; i < otherSize; i++) {
 				fastLookup_[fastCount + i] = other.fastLookup_[i];
 			}
 			fastCount += other.fastCount;
 		} else {
-			for (int i = 0; i < otherSize; i++) {
+			for (size_t i = 0; i < otherSize; i++) {
 				push_back(other[i]);
 			}
 		}

Common/GPU/OpenGL/thin3d_gl.cpp

@@ -1219,7 +1219,7 @@ bool OpenGLPipeline::LinkShaders() {
 	}
 	std::vector<GLRProgram::Initializer> initialize;
 	for (int i = 0; i < MAX_TEXTURE_SLOTS; ++i) {
-		if (i < samplers_.size()) {
+		if (i < (int)samplers_.size()) {
 			initialize.push_back({ &samplerLocs_[i], 0, i });
 		} else {
 			samplerLocs_[i] = -1;

Common/GPU/Vulkan/thin3d_vulkan.cpp

@@ -1086,7 +1086,7 @@ Pipeline *VKContext::CreateGraphicsPipeline(const PipelineDesc &desc, const char
 	_dbg_assert_((int)input->attributes.size() == (int)input->visc.vertexAttributeDescriptionCount);
 
 	gDesc.ibd = input->bindings[0];
-	for (int i = 0; i < input->attributes.size(); i++) {
+	for (size_t i = 0; i < input->attributes.size(); i++) {
 		gDesc.attrs[i] = input->attributes[i];
 	}
 	gDesc.vis.vertexAttributeDescriptionCount = input->visc.vertexAttributeDescriptionCount;

Core/Dialog/PSPSaveDialog.cpp

@@ -1149,7 +1149,9 @@ void PSPSaveDialog::ExecuteNotVisibleIOAction() {
 	case SCE_UTILITY_SAVEDATA_TYPE_READDATA:
 	case SCE_UTILITY_SAVEDATA_TYPE_READDATASECURE:
 		result = param.Load(param.GetPspParam(), GetSelectedSaveDirName(), currentSelectedSave, param.GetPspParam()->mode == SCE_UTILITY_SAVEDATA_TYPE_READDATASECURE);
-		if(result == SCE_UTILITY_SAVEDATA_ERROR_LOAD_NO_DATA)
+		if (result == SCE_UTILITY_SAVEDATA_ERROR_LOAD_DATA_BROKEN)
+			result = SCE_UTILITY_SAVEDATA_ERROR_RW_DATA_BROKEN;
+		if (result == SCE_UTILITY_SAVEDATA_ERROR_LOAD_NO_DATA)
 			result = SCE_UTILITY_SAVEDATA_ERROR_RW_NO_DATA;
 		break;
 	case SCE_UTILITY_SAVEDATA_TYPE_ERASE:

Core/Dialog/SavedataParam.cpp

@@ -589,15 +589,22 @@ int SavedataParam::Load(SceUtilitySavedataParam *param, const std::string &saveD
 		return isRWMode ? SCE_UTILITY_SAVEDATA_ERROR_RW_NO_DATA : SCE_UTILITY_SAVEDATA_ERROR_LOAD_NO_DATA;
 	}
 
+	if (fileName != "" && !pspFileSystem.GetFileInfo(filePath).exists) {
+		return isRWMode ? SCE_UTILITY_SAVEDATA_ERROR_RW_FILE_NOT_FOUND : SCE_UTILITY_SAVEDATA_ERROR_LOAD_FILE_NOT_FOUND;
+	}
+
+	// If it wasn't zero, force to zero before loading and especially in case of error.
+	// This isn't reset if the path doesn't even exist.
+	param->dataSize = 0;
+	int result = LoadSaveData(param, saveDirName, dirPath, secureMode);
+	if (result != 0)
+		return result;
+
 	// Load sfo
 	if (!LoadSFO(param, dirPath)) {
 		return isRWMode ? SCE_UTILITY_SAVEDATA_ERROR_RW_DATA_BROKEN : SCE_UTILITY_SAVEDATA_ERROR_LOAD_DATA_BROKEN;
 	}
 
-	if (fileName != "" && !pspFileSystem.GetFileInfo(filePath).exists) {
-		return isRWMode ? SCE_UTILITY_SAVEDATA_ERROR_RW_FILE_NOT_FOUND : SCE_UTILITY_SAVEDATA_ERROR_LOAD_FILE_NOT_FOUND;
-	}
-
 	// Don't know what it is, but PSP always respond this and this unlock some game
 	param->bind = 1021;
 
@@ -612,15 +619,6 @@ int SavedataParam::Load(SceUtilitySavedataParam *param, const std::string &saveD
 	// Load SND0.AT3
 	LoadFile(dirPath, SND0_FILENAME, &param->snd0FileData);
 
-	if (fileName == "") {
-		// Don't load savedata but return success.
-		return 0;
-	}
-
-	int result = LoadSaveData(param, saveDirName, dirPath, secureMode);
-	if (result != 0)
-		return result;
-
 	return 0;
 }
 
@@ -638,6 +636,10 @@ int SavedataParam::LoadSaveData(SceUtilitySavedataParam *param, const std::strin
 
 	std::string filename = GetFileName(param);
 	std::string filePath = dirPath + "/" + filename;
+	// Blank filename always means success, if secureVersion was correct.
+	if (filename == "")
+		return 0;
+
 	s64 readSize;
 	INFO_LOG(SCEUTILITY, "Loading file with size %u in %s", param->dataBufSize, filePath.c_str());
 	u8 *saveData = nullptr;
@@ -667,14 +669,18 @@ int SavedataParam::LoadSaveData(SceUtilitySavedataParam *param, const std::strin
 	if (!saveDone) {
 		loadedSize = LoadNotCryptedSave(param, param->dataBuf, saveData, saveSize);
 	}
-	param->dataSize = (SceSize)saveSize;
 	delete[] saveData;
 
-	if (loadedSize != 0) {
+	// Ignore error codes.
+	if (loadedSize != 0 && (loadedSize & 0x80000000) == 0) {
 		std::string tag = "LoadSaveData/" + filePath;
 		NotifyMemInfo(MemBlockFlags::WRITE, param->dataBuf.ptr, loadedSize, tag.c_str(), tag.size());
 	}
 
+	if ((loadedSize & 0x80000000) != 0)
+		return loadedSize;
+
+	param->dataSize = (SceSize)saveSize;
 	return 0;
 }
 
@@ -760,8 +766,12 @@ u32 SavedataParam::LoadCryptedSave(SceUtilitySavedataParam *param, u8 *data, con
 	u32 sz = 0;
 	if (err == 0) {
 		if (param->dataBuf.IsValid()) {
-			sz = std::min((u32)saveSize, (u32)param->dataBufSize);
-			memcpy(data, data_base, sz);
+			if ((u32)saveSize > param->dataBufSize || !Memory::IsValidRange(param->dataBuf.ptr, saveSize)) {
+				sz = SCE_UTILITY_SAVEDATA_ERROR_LOAD_DATA_BROKEN;
+			} else {
+				sz = (u32)saveSize;
+				memcpy(data, data_base, sz);
+			}
 		}
 		saveDone = true;
 	}
@@ -773,9 +783,11 @@ u32 SavedataParam::LoadCryptedSave(SceUtilitySavedataParam *param, u8 *data, con
 
 u32 SavedataParam::LoadNotCryptedSave(SceUtilitySavedataParam *param, u8 *data, u8 *saveData, int &saveSize) {
 	if (param->dataBuf.IsValid()) {
-		u32 sz = std::min((u32)saveSize, (u32)param->dataBufSize);
-		memcpy(data, saveData, sz);
-		return sz;
+		if ((u32)saveSize > param->dataBufSize || !Memory::IsValidRange(param->dataBuf.ptr, saveSize)) {
+			return SCE_UTILITY_SAVEDATA_ERROR_LOAD_DATA_BROKEN;
+		}
+		memcpy(data, saveData, saveSize);
+		return saveSize;
 	}
 	return 0;
 }

GPU/Common/DepalettizeShaderCommon.cpp

@@ -132,10 +132,7 @@ void GenerateDepalShader300(ShaderWriter &writer, const DepalConfig &config) {
 		break;
 	}
 
-	float texturePixels = 256.0f;
-	if (config.clutFormat != GE_CMODE_32BIT_ABGR8888) {
-		texturePixels = 512.0f;
-	}
+	float texturePixels = 512.0f;
 
 	if (shift) {
 		writer.F("  index = (int(uint(index) >> uint(%d)) & 0x%02x)", shift, mask);
@@ -278,11 +275,9 @@ void GenerateDepalShaderFloat(ShaderWriter &writer, const DepalConfig &config) {
 		break;
 	}
 
-	float texturePixels = 256.f;
-	if (config.clutFormat != GE_CMODE_32BIT_ABGR8888) {
-		texturePixels = 512.f;
-		index_multiplier *= 0.5f;
-	}
+	// We always use 512-sized textures now.
+	float texturePixels = 512.f;
+	index_multiplier *= 0.5f;
 
 	// Adjust index_multiplier, similar to the use of 15.99 instead of 16 in the ES 3 path.
 	// index_multiplier -= 0.01f / texturePixels;
@@ -326,11 +321,7 @@ void GenerateDepalSmoothed(ShaderWriter &writer, const DepalConfig &config) {
 	}
 
 	writer.C("  float index = ").SampleTexture2D("tex", "v_texcoord").F(".%s * %0.1f;\n", sourceChannel, indexMultiplier);
-	float texturePixels = 256.f;
-	if (config.clutFormat != GE_CMODE_32BIT_ABGR8888) {
-		texturePixels = 512.f;
-	}
-
+	float texturePixels = 512.f;
 	writer.F("  float coord = (index + 0.5) * %f;\n", 1.0 / texturePixels);
 	writer.C("  vec4 outColor = ").SampleTexture2D("pal", "vec2(coord, 0.0)").C(";\n");
 }

GPU/Common/FragmentShaderGenerator.cpp

@@ -640,7 +640,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
 				p.C("    if (depalShift == 5u) { index0 = t.g; }\n");
 				p.C("    else if (depalShift == 10u) { index0 = t.b; }\n");
 				p.C("  }\n");
-				p.F("  t = ").SampleTexture2D("pal", "vec2(index0 * factor, 0.0)").C(";\n");
+				p.F("  t = ").SampleTexture2D("pal", "vec2(index0 * factor * 0.5, 0.0)").C(";\n");  // 0.5 for 512-entry CLUT.
 				break;
 			case ShaderDepalMode::NORMAL:
 				if (doTextureProjection) {

GPU/Common/FramebufferManagerCommon.cpp

@@ -666,6 +666,10 @@ static const char *reinterpretStrings[4][4] = {
 
 // Call this after the target has been bound for rendering. For color, raster is probably always going to win over blits/copies.
 void FramebufferManagerCommon::CopyToColorFromOverlappingFramebuffers(VirtualFramebuffer *dst) {
+	if (!useBufferedRendering_) {
+		return;
+	}
+
 	std::vector<CopySource> sources;
 	for (auto src : vfbs_) {
 		// Discard old and equal potential inputs.
@@ -773,30 +777,17 @@ void FramebufferManagerCommon::CopyToColorFromOverlappingFramebuffers(VirtualFra
 					WARN_LOG_ONCE(bta, G3D, "WARNING: Reinterpret encountered with BlueToAlpha on");
 				}
 
-				if (IsBufferFormat16Bit(src->fb_format) && !IsBufferFormat16Bit(dst->fb_format)) {
-					// We halve the X coordinates in the destination framebuffer.
-					// The shader will collect two pixels worth of input data and merge into one.
-					dstX1 *= 0.5f;
-					dstX2 *= 0.5f;
-				} else if (!IsBufferFormat16Bit(src->fb_format) && IsBufferFormat16Bit(dst->fb_format)) {
-					// We double the X coordinates in the destination framebuffer.
-					// The shader will sample and depending on the X coordinate & 1, use the upper or lower bits.
-					dstX1 *= 2.0f;
-					dstX2 *= 2.0f;
-				}
-
 				// Reinterpret!
 				WARN_LOG_N_TIMES(reint, 5, G3D, "Reinterpret detected from %08x_%s to %08x_%s",
 					src->fb_address, GeBufferFormatToString(src->fb_format),
 					dst->fb_address, GeBufferFormatToString(dst->fb_format));
-				pipeline = reinterpretFromTo_[(int)src->fb_format][(int)dst->fb_format];
+
+				float scaleFactorX = 1.0f;
+				pipeline = GetReinterpretPipeline(src->fb_format, dst->fb_format, &scaleFactorX);
+				dstX1 *= scaleFactorX;
+				dstX2 *= scaleFactorX;
+
 				pass_name = reinterpretStrings[(int)src->fb_format][(int)dst->fb_format];
-				if (!pipeline) {
-					pipeline = draw2D_.Create2DPipeline([=](ShaderWriter &shaderWriter) -> Draw2DPipelineInfo {
-						return GenerateReinterpretFragmentShader(shaderWriter, src->fb_format, dst->fb_format);
-					});
-					reinterpretFromTo_[(int)src->fb_format][(int)dst->fb_format] = pipeline;
-				}
 
 				gpuStats.numReinterpretCopies++;
 			}
@@ -819,6 +810,27 @@ void FramebufferManagerCommon::CopyToColorFromOverlappingFramebuffers(VirtualFra
 	textureCache_->ForgetLastTexture();
 }
 
+Draw2DPipeline *FramebufferManagerCommon::GetReinterpretPipeline(GEBufferFormat from, GEBufferFormat to, float *scaleFactorX) {
+	if (IsBufferFormat16Bit(from) && !IsBufferFormat16Bit(to)) {
+		// We halve the X coordinates in the destination framebuffer.
+		// The shader will collect two pixels worth of input data and merge into one.
+		*scaleFactorX = 0.5f;
+	} else if (!IsBufferFormat16Bit(from) && IsBufferFormat16Bit(to)) {
+		// We double the X coordinates in the destination framebuffer.
+		// The shader will sample and depending on the X coordinate & 1, use the upper or lower bits.
+		*scaleFactorX = 2.0f;
+	}
+
+	Draw2DPipeline *pipeline = reinterpretFromTo_[(int)from][(int)to];
+	if (!pipeline) {
+		pipeline = draw2D_.Create2DPipeline([=](ShaderWriter &shaderWriter) -> Draw2DPipelineInfo {
+			return GenerateReinterpretFragmentShader(shaderWriter, from, to);
+		});
+		reinterpretFromTo_[(int)from][(int)to] = pipeline;
+	}
+	return pipeline;
+}
+
 void FramebufferManagerCommon::DestroyFramebuf(VirtualFramebuffer *v) {
 	// Notify the texture cache of both the color and depth buffers.
 	textureCache_->NotifyFramebuffer(v, NOTIFY_FB_DESTROYED);
@@ -1069,7 +1081,7 @@ bool FramebufferManagerCommon::BindFramebufferAsColorTexture(int stage, VirtualF
 
 	// currentRenderVfb_ will always be set when this is called, except from the GE debugger.
 	// Let's just not bother with the copy in that case.
-	bool skipCopy = !(flags & BINDFBCOLOR_MAY_COPY) || GPUStepping::IsStepping();
+	bool skipCopy = !(flags & BINDFBCOLOR_MAY_COPY);
 
 	// Currently rendering to this framebuffer. Need to make a copy.
 	if (!skipCopy && framebuffer == currentRenderVfb_) {

GPU/Common/FramebufferManagerCommon.h

@@ -421,14 +421,15 @@ class FramebufferManagerCommon {
 	// Returns the resolved framebuffer.
 	VirtualFramebuffer *ResolveFramebufferColorToFormat(VirtualFramebuffer *vfb, GEBufferFormat newFormat);
 
+	Draw2DPipeline *Get2DPipeline(Draw2DShader shader);
+	Draw2DPipeline *GetReinterpretPipeline(GEBufferFormat from, GEBufferFormat to, float *scaleFactorX);
+
 protected:
 	virtual void PackFramebufferSync(VirtualFramebuffer *vfb, int x, int y, int w, int h, RasterChannel channel);
 	void SetViewport2D(int x, int y, int w, int h);
 	Draw::Texture *MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
 	void DrawActiveTexture(float x, float y, float w, float h, float destW, float destH, float u0, float v0, float u1, float v1, int uvRotation, int flags);
 
-	Draw2DPipeline *Get2DPipeline(Draw2DShader shader);
-
 	void CopyToColorFromOverlappingFramebuffers(VirtualFramebuffer *dest);
 	void CopyToDepthFromOverlappingFramebuffers(VirtualFramebuffer *dest);