Merge branch 'master' into TemporaryHackishBranch

.gitignore

@@ -62,7 +62,6 @@ build*/
 # Temp file used by jenkins windows build (TODO: remove)
 desc.txt
 
-
 Logs
 Memstick
 memstick
@@ -96,5 +95,8 @@ debian/files
 debian/ppsspp.substvars
 debian/ppsspp/
 
+# Libretro build
+*.o
+
 # YouCompleteMe file
 .ycm_extra_conf.pyc

Common/Vulkan/VulkanContext.cpp

@@ -431,6 +431,10 @@ void VulkanContext::ChooseDevice(int physical_device) {
 			break;
 		}
 	}
+	if (deviceInfo_.preferredDepthStencilFormat == VK_FORMAT_UNDEFINED) {
+		// WTF? This is bad.
+		ELOG("Could not find a usable depth stencil format.");
+	}
 
 	// This is as good a place as any to do this
 	vkGetPhysicalDeviceMemoryProperties(physical_devices_[physical_device_], &memory_properties);
@@ -708,9 +712,8 @@ bool VulkanContext::InitQueue() {
 
 bool VulkanContext::InitSwapchain() {
 	VkResult U_ASSERT_ONLY res;
-	VkSurfaceCapabilitiesKHR surfCapabilities;
 
-	res = vkGetPhysicalDeviceSurfaceCapabilitiesKHR(physical_devices_[physical_device_], surface_, &surfCapabilities);
+	res = vkGetPhysicalDeviceSurfaceCapabilitiesKHR(physical_devices_[physical_device_], surface_, &surfCapabilities_);
 	assert(res == VK_SUCCESS);
 
 	uint32_t presentModeCount;
@@ -723,7 +726,7 @@ bool VulkanContext::InitSwapchain() {
 
 	VkExtent2D swapChainExtent;
 	// width and height are either both -1, or both not -1.
-	if (surfCapabilities.currentExtent.width == (uint32_t)-1) {
+	if (surfCapabilities_.currentExtent.width == (uint32_t)-1) {
 		// If the surface size is undefined, the size is set to
 		// the size of the images requested.
 		ILOG("initSwapchain: %dx%d", width_, height_);
@@ -731,7 +734,7 @@ bool VulkanContext::InitSwapchain() {
 		swapChainExtent.height = height_;
 	} else {
 		// If the surface size is defined, the swap chain size must match
-		swapChainExtent = surfCapabilities.currentExtent;
+		swapChainExtent = surfCapabilities_.currentExtent;
 	}
 
 	// TODO: Find a better way to specify the prioritized present mode while being able
@@ -767,20 +770,20 @@ bool VulkanContext::InitSwapchain() {
 	// Determine the number of VkImage's to use in the swap chain (we desire to
 	// own only 1 image at a time, besides the images being displayed and
 	// queued for display):
-	uint32_t desiredNumberOfSwapChainImages = surfCapabilities.minImageCount + 1;
+	uint32_t desiredNumberOfSwapChainImages = surfCapabilities_.minImageCount + 1;
 	ILOG("numSwapChainImages: %d", desiredNumberOfSwapChainImages);
-	if ((surfCapabilities.maxImageCount > 0) &&
-		(desiredNumberOfSwapChainImages > surfCapabilities.maxImageCount))
+	if ((surfCapabilities_.maxImageCount > 0) &&
+		(desiredNumberOfSwapChainImages > surfCapabilities_.maxImageCount))
 	{
 		// Application must settle for fewer images than desired:
-		desiredNumberOfSwapChainImages = surfCapabilities.maxImageCount;
+		desiredNumberOfSwapChainImages = surfCapabilities_.maxImageCount;
 	}
 
 	VkSurfaceTransformFlagBitsKHR preTransform;
-	if (surfCapabilities.supportedTransforms & VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR) {
+	if (surfCapabilities_.supportedTransforms & VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR) {
 		preTransform = VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR;
 	} else {
-		preTransform = surfCapabilities.currentTransform;
+		preTransform = surfCapabilities_.currentTransform;
 	}
 
 	VkSwapchainCreateInfoKHR swap_chain_info = { VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR };
@@ -795,12 +798,22 @@ bool VulkanContext::InitSwapchain() {
 	swap_chain_info.presentMode = swapchainPresentMode;
 	swap_chain_info.oldSwapchain = VK_NULL_HANDLE;
 	swap_chain_info.clipped = true;
-	swap_chain_info.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+	swap_chain_info.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
+	if (surfCapabilities_.supportedUsageFlags & VK_IMAGE_USAGE_TRANSFER_DST_BIT)
+		swap_chain_info.imageUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+
+#ifndef ANDROID
+	// We don't support screenshots on Android
+	// Add more usage flags if they're supported.
+	if (surfCapabilities_.supportedUsageFlags & VK_IMAGE_USAGE_TRANSFER_SRC_BIT)
+		swap_chain_info.imageUsage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+#endif
+
 	swap_chain_info.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
 	swap_chain_info.queueFamilyIndexCount = 0;
 	swap_chain_info.pQueueFamilyIndices = NULL;
 	// OPAQUE is not supported everywhere.
-	if (surfCapabilities.supportedCompositeAlpha & VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR) {
+	if (surfCapabilities_.supportedCompositeAlpha & VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR) {
 		swap_chain_info.compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
 	} else {
 		// This should be supported anywhere, and is the only thing supported on the SHIELD TV, for example.

Common/Vulkan/VulkanContext.h

@@ -198,6 +198,7 @@ public:
 	const VkPhysicalDeviceFeatures &GetFeaturesAvailable() const { return featuresAvailable_; }
 	const VkPhysicalDeviceFeatures &GetFeaturesEnabled() const { return featuresEnabled_; }
 	const VulkanPhysicalDeviceInfo &GetDeviceInfo() const { return deviceInfo_; }
+	const VkSurfaceCapabilitiesKHR &GetSurfaceCapabilities() const { return surfCapabilities_; }
 
 	bool IsDeviceExtensionAvailable(const char *name) const {
 		for (auto &iter : device_extension_properties_) {
@@ -302,6 +303,8 @@ private:
 	VkPhysicalDeviceFeatures featuresAvailable_{};
 	VkPhysicalDeviceFeatures featuresEnabled_{};
 
+	VkSurfaceCapabilitiesKHR surfCapabilities_{};
+
 	std::vector<VkCommandBuffer> cmdQueue_;
 };
 

GPU/Common/DrawEngineCommon.cpp

@@ -140,9 +140,11 @@ void DrawEngineCommon::Resized() {
 	ClearTrackedVertexArrays();
 }
 
-u32 DrawEngineCommon::NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType) {
+u32 DrawEngineCommon::NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType, int *vertexSize) {
 	const u32 vertTypeID = (vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24);
 	VertexDecoder *dec = GetVertexDecoder(vertTypeID);
+	if (vertexSize)
+		*vertexSize = dec->VertexSize();
 	return DrawEngineCommon::NormalizeVertices(outPtr, bufPtr, inPtr, dec, lowerBound, upperBound, vertType);
 }
 
@@ -150,7 +152,7 @@ u32 DrawEngineCommon::NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr,
 //
 // It does the simplest and safest test possible: If all points of a bbox is outside a single of
 // our clipping planes, we reject the box. Tighter bounds would be desirable but would take more calculations.
-bool DrawEngineCommon::TestBoundingBox(void* control_points, int vertexCount, u32 vertType) {
+bool DrawEngineCommon::TestBoundingBox(void* control_points, int vertexCount, u32 vertType, int *bytesRead) {
 	SimpleVertex *corners = (SimpleVertex *)(decoded + 65536 * 12);
 	float *verts = (float *)(decoded + 65536 * 18);
 
@@ -158,25 +160,30 @@ bool DrawEngineCommon::TestBoundingBox(void* control_points, int vertexCount, u3
 	// and a large vertex format.
 	if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_FLOAT) {
 		verts = (float *)control_points;
+		*bytesRead = 3 * sizeof(float) * vertexCount;
 	} else if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_8BIT) {
 		const s8 *vtx = (const s8 *)control_points;
 		for (int i = 0; i < vertexCount * 3; i++) {
 			verts[i] = vtx[i] * (1.0f / 128.0f);
 		}
+		*bytesRead = 3 * sizeof(s8) * vertexCount;
 	} else if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_16BIT) {
 		const s16 *vtx = (const s16*)control_points;
 		for (int i = 0; i < vertexCount * 3; i++) {
 			verts[i] = vtx[i] * (1.0f / 32768.0f);
 		}
+		*bytesRead = 3 * sizeof(s16) * vertexCount;
 	} else {
 		// Simplify away bones and morph before proceeding
 		u8 *temp_buffer = decoded + 65536 * 24;
-		NormalizeVertices((u8 *)corners, temp_buffer, (u8 *)control_points, 0, vertexCount, vertType);
+		int vertexSize = 0;
+		NormalizeVertices((u8 *)corners, temp_buffer, (u8 *)control_points, 0, vertexCount, vertType, &vertexSize);
 		for (int i = 0; i < vertexCount; i++) {
 			verts[i * 3] = corners[i].pos.x;
 			verts[i * 3 + 1] = corners[i].pos.y;
 			verts[i * 3 + 2] = corners[i].pos.z;
 		}
+		*bytesRead = vertexSize * vertexCount;
 	}
 
 	Plane planes[6];

GPU/Common/DrawEngineCommon.h

@@ -49,8 +49,6 @@ public:
 	DrawEngineCommon();
 	virtual ~DrawEngineCommon();
 
-	bool TestBoundingBox(void* control_points, int vertexCount, u32 vertType);
-
 	bool GetCurrentSimpleVertices(int count, std::vector<GPUDebugVertex> &vertices, std::vector<u16> &indices);
 
 	static u32 NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, VertexDecoder *dec, int lowerBound, int upperBound, u32 vertType);
@@ -61,6 +59,7 @@ public:
 	// Same for SubmitPrim
 	virtual void DispatchSubmitPrim(void *verts, void *inds, GEPrimitiveType prim, int vertexCount, u32 vertType, int *bytesRead) = 0;
 
+	bool TestBoundingBox(void* control_points, int vertexCount, u32 vertType, int *bytesRead);
 	void SubmitSpline(const void *control_points, const void *indices, int tess_u, int tess_v, int count_u, int count_v, int type_u, int type_v, GEPatchPrimType prim_type, bool computeNormals, bool patchFacing, u32 vertType, int *bytesRead);
 	void SubmitBezier(const void *control_points, const void *indices, int tess_u, int tess_v, int count_u, int count_v, GEPatchPrimType prim_type, bool computeNormals, bool patchFacing, u32 vertType, int *bytesRead);
 
@@ -79,7 +78,7 @@ protected:
 	virtual void ClearTrackedVertexArrays() {}
 
 	// Preprocessing for spline/bezier
-	u32 NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType);
+	u32 NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType, int *vertexSize = nullptr);
 
 	// Utility for vertex caching
 	u32 ComputeMiniHash();

GPU/Common/FramebufferCommon.cpp

@@ -2106,6 +2106,16 @@ void FramebufferManagerCommon::DownloadFramebufferForClut(u32 fb_address, u32 lo
 	}
 }
 
+void FramebufferManagerCommon::RebindFramebuffer() {
+	if (currentRenderVfb_ && currentRenderVfb_->fbo) {
+		draw_->BindFramebufferAsRenderTarget(currentRenderVfb_->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
+	} else {
+		// Should this even happen?
+		draw_->BindFramebufferAsRenderTarget(nullptr, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
+	}
+	gstate_c.Dirty(DIRTY_VIEWPORTSCISSOR_STATE);
+}
+
 std::vector<FramebufferInfo> FramebufferManagerCommon::GetFramebufferList() {
 	std::vector<FramebufferInfo> list;
 

GPU/Common/FramebufferCommon.h

@@ -214,7 +214,7 @@ public:
 			return vfb;
 		}
 	}
-	virtual void RebindFramebuffer() = 0;
+	virtual void RebindFramebuffer();
 	std::vector<FramebufferInfo> GetFramebufferList();
 
 	void CopyDisplayToOutput();

GPU/Common/SplineCommon.cpp

@@ -855,6 +855,7 @@ void DrawEngineCommon::SubmitSpline(const void *control_points, const void *indi
 		GetIndexBounds(indices, count_u * count_v, vertType, &index_lower_bound, &index_upper_bound);
 
 	VertexDecoder *origVDecoder = GetVertexDecoder((vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24));
+	// TODO: If indexing is on, this may be wrong!
 	*bytesRead = count_u * count_v * origVDecoder->VertexSize();
 
 	// Real hardware seems to draw nothing when given < 4 either U or V.
@@ -971,6 +972,7 @@ void DrawEngineCommon::SubmitBezier(const void *control_points, const void *indi
 		GetIndexBounds(indices, count_u*count_v, vertType, &index_lower_bound, &index_upper_bound);
 
 	VertexDecoder *origVDecoder = GetVertexDecoder((vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24));
+	// TODO: If indexing is on, this may be wrong!
 	*bytesRead = count_u * count_v * origVDecoder->VertexSize();
 
 	// Real hardware seems to draw nothing when given < 4 either U or V.

GPU/D3D11/FramebufferManagerD3D11.cpp

@@ -459,8 +459,6 @@ void FramebufferManagerD3D11::BindPostShader(const PostShaderUniforms &uniforms)
 	context_->Unmap(postConstants_, 0);
 	context_->VSSetConstantBuffers(0, 1, &postConstants_);  // Probably not necessary
 	context_->PSSetConstantBuffers(0, 1, &postConstants_);
-
-	gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE);
 }
 
 void FramebufferManagerD3D11::RebindFramebuffer() {

GPU/Directx9/FramebufferDX9.cpp

@@ -304,15 +304,6 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 		}
 	}
 
-	void FramebufferManagerDX9::RebindFramebuffer() {
-		if (currentRenderVfb_ && currentRenderVfb_->fbo) {
-			draw_->BindFramebufferAsRenderTarget(currentRenderVfb_->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
-		} else {
-			// Should this even happen?
-			draw_->BindFramebufferAsRenderTarget(nullptr, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
-		}
-	}
-
 	void FramebufferManagerDX9::Bind2DShader() {
 		device_->SetVertexDeclaration(pFramebufferVertexDecl);
 		device_->SetPixelShader(pFramebufferPixelShader);

GPU/Directx9/FramebufferDX9.h

@@ -66,8 +66,6 @@ public:
 	bool GetStencilbuffer(u32 fb_address, int fb_stride, GPUDebugBuffer &buffer) override;
 	bool GetOutputFramebuffer(GPUDebugBuffer &buffer) override;
 
-	virtual void RebindFramebuffer() override;
-
 	LPDIRECT3DSURFACE9 GetOffscreenSurface(LPDIRECT3DSURFACE9 similarSurface, VirtualFramebuffer *vfb);
 	LPDIRECT3DSURFACE9 GetOffscreenSurface(D3DFORMAT fmt, u32 w, u32 h);
 

GPU/GLES/FramebufferManagerGLES.cpp

@@ -438,12 +438,7 @@ void FramebufferManagerGLES::DrawActiveTexture(float x, float y, float w, float
 }
 
 void FramebufferManagerGLES::RebindFramebuffer() {
-	if (currentRenderVfb_ && currentRenderVfb_->fbo) {
-		draw_->BindFramebufferAsRenderTarget(currentRenderVfb_->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
-	} else {
-		// Should this even happen?
-		draw_->BindFramebufferAsRenderTarget(nullptr, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
-	}
+	FramebufferManagerCommon::RebindFramebuffer();
 	if (g_Config.iRenderingMode == FB_NON_BUFFERED_MODE)
 		glstate.viewport.restore();
 }

GPU/GLES/FramebufferManagerGLES.h

@@ -81,7 +81,6 @@ public:
 	bool NotifyStencilUpload(u32 addr, int size, bool skipZero = false) override;
 
 	bool GetOutputFramebuffer(GPUDebugBuffer &buffer) override;
-
 	virtual void RebindFramebuffer() override;
 
 protected:

GPU/GPUCommon.cpp

@@ -1511,13 +1511,17 @@ void GPUCommon::Execute_Spline(u32 op, u32 diff) {
 
 void GPUCommon::Execute_BoundingBox(u32 op, u32 diff) {
 	// Just resetting, nothing to check bounds for.
-	const u32 data = op & 0x00FFFFFF;
-	if (data == 0) {
+	const u32 count = op & 0xFFFFFF;
+	if (count == 0) {
 		currentList->bboxResult = false;
 		return;
 	}
-	if (((data & 7) == 0) && data <= 64) {  // Sanity check
+	if (((count & 7) == 0) && count <= 64) {  // Sanity check
 		void *control_points = Memory::GetPointer(gstate_c.vertexAddr);
+		if (!control_points) {
+			return;
+		}
+
 		if (gstate.vertType & GE_VTYPE_IDX_MASK) {
 			ERROR_LOG_REPORT_ONCE(boundingbox, G3D, "Indexed bounding box data not supported.");
 			// Data seems invalid. Let's assume the box test passed.
@@ -1526,11 +1530,11 @@ void GPUCommon::Execute_BoundingBox(u32 op, u32 diff) {
 		}
 
 		// Test if the bounding box is within the drawing region.
-		if (control_points) {
-			currentList->bboxResult = drawEngineCommon_->TestBoundingBox(control_points, data, gstate.vertType);
-		}
+		int bytesRead;
+		currentList->bboxResult = drawEngineCommon_->TestBoundingBox(control_points, count, gstate.vertType, &bytesRead);
+		AdvanceVerts(gstate.vertType, count, bytesRead);
 	} else {
-		ERROR_LOG_REPORT_ONCE(boundingbox, G3D, "Bad bounding box data: %06x", data);
+		ERROR_LOG_REPORT_ONCE(boundingbox, G3D, "Bad bounding box data: %06x", count);
 		// Data seems invalid. Let's assume the box test passed.
 		currentList->bboxResult = true;
 	}

GPU/Null/NullGpu.cpp

@@ -15,7 +15,7 @@
 // Official git repository and contact information can be found at
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
-
+#include "GPU/Common/DrawEngineCommon.h"
 #include "GPU/Null/NullGpu.h"
 #include "GPU/GPUState.h"
 #include "GPU/ge_constants.h"
@@ -25,7 +25,17 @@
 #include "Core/HLE/sceKernelInterrupt.h"
 #include "Core/HLE/sceGe.h"
 
-NullGPU::NullGPU() : GPUCommon(nullptr, nullptr) { }
+class NullDrawEngine : public DrawEngineCommon {
+public:
+	void DispatchFlush() override {
+	}
+	void DispatchSubmitPrim(void *verts, void *inds, GEPrimitiveType prim, int vertexCount, u32 vertType, int *bytesRead) override {
+	}
+};
+
+NullGPU::NullGPU() : GPUCommon(nullptr, nullptr) {
+	drawEngineCommon_ = new NullDrawEngine();
+}
 NullGPU::~NullGPU() { }
 
 void NullGPU::FastRunLoop(DisplayList &list) {

GPU/Vulkan/DrawEngineVulkan.cpp

@@ -516,23 +516,23 @@ VkDescriptorSet DrawEngineVulkan::GetOrCreateDescriptorSet(VkImageView imageView
 	descAlloc.descriptorSetCount = 1;
 	VkResult result = vkAllocateDescriptorSets(vulkan_->GetDevice(), &descAlloc, &desc);
 	// Even in release mode, this is bad.
-	_assert_msg_(G3D, result == VK_SUCCESS, "Ran out of descriptors in pool. sz=%d", (int)frame->descSets.size());
+	_assert_msg_(G3D, result == VK_SUCCESS, "Ran out of descriptor space in pool. sz=%d", (int)frame->descSets.size());
 
 	// We just don't write to the slots we don't care about.
 	// We need 8 now that we support secondary texture bindings.
 	VkWriteDescriptorSet writes[8]{};
 	// Main texture
 	int n = 0;
-	VkDescriptorImageInfo tex{};
+	VkDescriptorImageInfo tex[2]{};
 	if (imageView) {
 		// TODO: Also support LAYOUT_GENERAL to be able to texture from framebuffers without transitioning them?
-		tex.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
-		tex.imageView = imageView;
-		tex.sampler = sampler;
+		tex[0].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
+		tex[0].imageView = imageView;
+		tex[0].sampler = sampler;
 		writes[n].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
 		writes[n].pNext = nullptr;
 		writes[n].dstBinding = DRAW_BINDING_TEXTURE;
-		writes[n].pImageInfo = &tex;
+		writes[n].pImageInfo = &tex[0];
 		writes[n].descriptorCount = 1;
 		writes[n].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		writes[n].dstSet = desc;
@@ -541,13 +541,13 @@ VkDescriptorSet DrawEngineVulkan::GetOrCreateDescriptorSet(VkImageView imageView
 
 	if (boundSecondary_) {
 		// TODO: Also support LAYOUT_GENERAL to be able to texture from framebuffers without transitioning them?
-		tex.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
-		tex.imageView = boundSecondary_;
-		tex.sampler = samplerSecondary_;
+		tex[1].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
+		tex[1].imageView = boundSecondary_;
+		tex[1].sampler = samplerSecondary_;
 		writes[n].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
 		writes[n].pNext = nullptr;
 		writes[n].dstBinding = DRAW_BINDING_2ND_TEXTURE;
-		writes[n].pImageInfo = &tex;
+		writes[n].pImageInfo = &tex[1];
 		writes[n].descriptorCount = 1;
 		writes[n].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		writes[n].dstSet = desc;
@@ -877,6 +877,7 @@ void DrawEngineVulkan::DoFlush() {
 				// Already logged, let's bail out.
 				return;
 			}
+			BindShaderBlendTex();  // This might cause copies so important to do before BindPipeline.
 			renderManager->BindPipeline(pipeline->pipeline);
 			if (pipeline != lastPipeline_) {
 				if (lastPipeline_ && !lastPipeline_->useBlendConstant && pipeline->useBlendConstant) {
@@ -976,6 +977,7 @@ void DrawEngineVulkan::DoFlush() {
 					// Already logged, let's bail out.
 					return;
 				}
+				BindShaderBlendTex();  // This might cause copies so super important to do before BindPipeline.
 				renderManager->BindPipeline(pipeline->pipeline);
 				if (pipeline != lastPipeline_) {
 					if (lastPipeline_ && !lastPipeline_->useBlendConstant && pipeline->useBlendConstant) {

GPU/Vulkan/DrawEngineVulkan.h

@@ -147,6 +147,15 @@ public:
 		DoFlush();
 	}
 
+	void FinishDeferred() {
+		if (!numDrawCalls)
+			return;
+		// Decode any pending vertices. And also flush while we're at it, for simplicity.
+		// It might be possible to only decode like in the other backends, but meh, it can't matter.
+		// Issue #10095 has a nice example of where this is required.
+		DoFlush();
+	}
+
 	void DispatchFlush() override { Flush(); }
 	void DispatchSubmitPrim(void *verts, void *inds, GEPrimitiveType prim, int vertexCount, u32 vertType, int *bytesRead) override {
 		SubmitPrim(verts, inds, prim, vertexCount, vertType, bytesRead);
@@ -179,6 +188,7 @@ private:
 	struct FrameData;
 	void ApplyDrawStateLate(VulkanRenderManager *renderManager, bool applyStencilRef, uint8_t stencilRef, bool useBlendConstant);
 	void ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManager, ShaderManagerVulkan *shaderManager, int prim, VulkanPipelineRasterStateKey &key, VulkanDynamicState &dynState);
+	void BindShaderBlendTex();
 	void ResetShaderBlending();
 
 	void InitDeviceObjects();
@@ -203,7 +213,7 @@ private:
 
 	// Secondary texture for shader blending
 	VkImageView boundSecondary_ = VK_NULL_HANDLE;
-	VkSampler samplerSecondary_ = VK_NULL_HANDLE;
+	VkSampler samplerSecondary_ = VK_NULL_HANDLE;  // This one is actually never used since we use fetch.
 
 	PrehashMap<VertexArrayInfoVulkan *, nullptr> vai_;
 	VulkanPushBuffer *vertexCache_;

GPU/Vulkan/FramebufferVulkan.cpp

@@ -367,16 +367,6 @@ void FramebufferManagerVulkan::BindPostShader(const PostShaderUniforms &uniforms
 	gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE);
 }
 
-void FramebufferManagerVulkan::RebindFramebuffer() {
-	if (currentRenderVfb_ && currentRenderVfb_->fbo) {
-		draw_->BindFramebufferAsRenderTarget(currentRenderVfb_->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
-	} else {
-		// Should this even happen?
-		draw_->BindFramebufferAsRenderTarget(nullptr, { Draw::RPAction::KEEP, Draw::RPAction::KEEP });
-	}
-	gstate_c.Dirty(DIRTY_VIEWPORTSCISSOR_STATE);
-}
-
 int FramebufferManagerVulkan::GetLineWidth() {
 	if (g_Config.iInternalResolution == 0) {
 		return std::max(1, (int)(renderWidth_ / 480));
@@ -699,4 +689,4 @@ void FramebufferManagerVulkan::CompilePostShader() {
 
 
 	usePostShader_ = true;
-}
+}

GPU/Vulkan/FramebufferVulkan.h

@@ -78,7 +78,6 @@ public:
 
 	bool NotifyStencilUpload(u32 addr, int size, bool skipZero = false) override;
 
-	virtual void RebindFramebuffer() override;
 	VkImageView BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags);
 
 	// If within a render pass, this will just issue a regular clear. If beginning a new render pass,

GPU/Vulkan/GPU_Vulkan.cpp

@@ -181,8 +181,13 @@ void GPU_Vulkan::CheckGPUFeatures() {
 	}
 	if (vulkan_->GetFeaturesEnabled().dualSrcBlend) {
 		switch (vulkan_->GetPhysicalDeviceProperties().vendorID) {
+		case VULKAN_VENDOR_NVIDIA:
+			// Workaround for Shield TV driver bug.
+			if (strcmp(vulkan_->GetPhysicalDeviceProperties().deviceName, "NVIDIA Tegra X1") != 0)
+				features |= GPU_SUPPORTS_DUALSOURCE_BLEND;
+			break;
 		case VULKAN_VENDOR_INTEL:
-			// Work around for Intel driver bug.
+			// Workaround for Intel driver bug.
 			break;
 		case VULKAN_VENDOR_AMD:
 			// See issue #10074, and also #10065 (AMD) and #10109 for the choice of the driver version to check for
@@ -205,6 +210,17 @@ void GPU_Vulkan::CheckGPUFeatures() {
 		features |= GPU_USE_CLEAR_RAM_HACK;
 	}
 
+	if (!g_Config.bHighQualityDepth && (features & GPU_SUPPORTS_ACCURATE_DEPTH) != 0) {
+		features |= GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT;
+	}
+	else if (PSP_CoreParameter().compat.flags().PixelDepthRounding) {
+		// Use fragment rounding on desktop and GLES3, most accurate.
+		features |= GPU_ROUND_FRAGMENT_DEPTH_TO_16BIT;
+	}
+	else if (PSP_CoreParameter().compat.flags().VertexDepthRounding) {
+		features |= GPU_ROUND_DEPTH_TO_16BIT;
+	}
+
 	// Mandatory features on Vulkan, which may be checked in "centralized" code
 	features |= GPU_SUPPORTS_ACCURATE_DEPTH;
 	features |= GPU_SUPPORTS_TEXTURE_LOD_CONTROL;
@@ -214,6 +230,9 @@ void GPU_Vulkan::CheckGPUFeatures() {
 	features |= GPU_SUPPORTS_OES_TEXTURE_NPOT;
 	features |= GPU_SUPPORTS_LARGE_VIEWPORTS;
 	features |= GPU_SUPPORTS_16BIT_FORMATS;
+	features |= GPU_SUPPORTS_INSTANCE_RENDERING;
+	features |= GPU_SUPPORTS_VERTEX_TEXTURE_FETCH;
+	features |= GPU_SUPPORTS_TEXTURE_FLOAT;
 	gstate_c.featureFlags = features;
 }
 
@@ -454,6 +473,7 @@ void GPU_Vulkan::FastRunLoop(DisplayList &list) {
 }
 
 void GPU_Vulkan::FinishDeferred() {
+	drawEngine_.FinishDeferred();
 }
 
 inline void GPU_Vulkan::CheckFlushOp(int cmd, u32 diff) {
@@ -525,12 +545,14 @@ void GPU_Vulkan::Execute_Prim(u32 op, u32 diff) {
 
 	void *verts = Memory::GetPointerUnchecked(gstate_c.vertexAddr);
 	void *inds = 0;
+	u32 vertexType = gstate.vertType;
 	if ((gstate.vertType & GE_VTYPE_IDX_MASK) != GE_VTYPE_IDX_NONE) {
-		if (!Memory::IsValidAddress(gstate_c.indexAddr)) {
-			ERROR_LOG_REPORT(G3D, "Bad index address %08x!", gstate_c.indexAddr);
+		u32 indexAddr = gstate_c.indexAddr;
+		if (!Memory::IsValidAddress(indexAddr)) {
+			ERROR_LOG_REPORT(G3D, "Bad index address %08x!", indexAddr);
 			return;
 		}
-		inds = Memory::GetPointerUnchecked(gstate_c.indexAddr);
+		inds = Memory::GetPointerUnchecked(indexAddr);
 	}
 
 #ifndef MOBILE_DEVICE
@@ -547,12 +569,12 @@ void GPU_Vulkan::Execute_Prim(u32 op, u32 diff) {
 
 	int bytesRead = 0;
 	UpdateUVScaleOffset();
-	drawEngine_.SubmitPrim(verts, inds, prim, count, gstate.vertType, &bytesRead);
+	drawEngine_.SubmitPrim(verts, inds, prim, count, vertexType, &bytesRead);
 
 	// After drawing, we advance the vertexAddr (when non indexed) or indexAddr (when indexed).
 	// Some games rely on this, they don't bother reloading VADDR and IADDR.
 	// The VADDR/IADDR registers are NOT updated.
-	AdvanceVerts(gstate.vertType, count, bytesRead);
+	AdvanceVerts(vertexType, count, bytesRead);
 }
 
 void GPU_Vulkan::Execute_LoadClut(u32 op, u32 diff) {

GPU/Vulkan/StateMappingVulkan.cpp

@@ -362,7 +362,7 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
 	}
 }
 
-void DrawEngineVulkan::ApplyDrawStateLate(VulkanRenderManager *renderManager, bool applyStencilRef, uint8_t stencilRef, bool useBlendConstant) {
+void DrawEngineVulkan::BindShaderBlendTex() {
 	// 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()) {
@@ -377,6 +377,9 @@ void DrawEngineVulkan::ApplyDrawStateLate(VulkanRenderManager *renderManager, bo
 		}
 	}
 
+}
+
+void DrawEngineVulkan::ApplyDrawStateLate(VulkanRenderManager *renderManager, bool applyStencilRef, uint8_t stencilRef, bool useBlendConstant) {
 	if (gstate_c.IsDirty(DIRTY_VIEWPORTSCISSOR_STATE)) {
 		renderManager->SetScissor(dynState_.scissor);
 		renderManager->SetViewport(dynState_.viewport);

Qt/mainwindow.cpp

@@ -391,12 +391,12 @@ void MainWindow::fullscrAct()
 
 void MainWindow::websiteAct()
 {
-	QDesktopServices::openUrl(QUrl("http://www.ppsspp.org/"));
+	QDesktopServices::openUrl(QUrl("https://www.ppsspp.org/"));
 }
 
 void MainWindow::forumAct()
 {
-	QDesktopServices::openUrl(QUrl("http://forums.ppsspp.org/"));
+	QDesktopServices::openUrl(QUrl("https://forums.ppsspp.org/"));
 }
 
 void MainWindow::gitAct() 

UI/GameSettingsScreen.cpp

@@ -694,8 +694,8 @@ void GameSettingsScreen::CreateViews() {
 
 #if PPSSPP_PLATFORM(ANDROID)
 	if (System_GetPropertyInt(SYSPROP_DEVICE_TYPE) == DEVICE_TYPE_MOBILE) {
-		static const char *screenRotation[] = {"Auto", "Landscape", "Portrait", "Landscape Reversed", "Portrait Reversed"};
-		PopupMultiChoice *rot = systemSettings->Add(new PopupMultiChoice(&g_Config.iScreenRotation, co->T("Screen Rotation"), screenRotation, 0, ARRAY_SIZE(screenRotation), co->GetName(), screenManager()));
+		static const char *screenRotation[] = {"Landscape", "Portrait", "Landscape Reversed", "Portrait Reversed"};
+		PopupMultiChoice *rot = systemSettings->Add(new PopupMultiChoice(&g_Config.iScreenRotation, co->T("Screen Rotation"), screenRotation, 1, ARRAY_SIZE(screenRotation), co->GetName(), screenManager()));
 		rot->OnChoice.Handle(this, &GameSettingsScreen::OnScreenRotation);
 
 		if (System_GetPropertyBool(SYSPROP_SUPPORTS_SUSTAINED_PERF_MODE)) {
@@ -866,7 +866,7 @@ static void RecreateActivity() {
 }
 
 UI::EventReturn GameSettingsScreen::OnAdhocGuides(UI::EventParams &e) {
-	LaunchBrowser("http://forums.ppsspp.org/forumdisplay.php?fid=34");
+	LaunchBrowser("https://forums.ppsspp.org/forumdisplay.php?fid=34");
 	return UI::EVENT_DONE;
 }
 
@@ -1136,6 +1136,7 @@ UI::EventReturn GameSettingsScreen::OnChangeNickname(UI::EventParams &e) {
 		g_Config.sNickName = StripSpaces(name);
 	}
 #elif defined(__ANDROID__)
+	// TODO: The return value is handled in NativeApp::inputbox_completed. This is horrific.
 	System_SendMessage("inputbox", ("nickname:" + g_Config.sNickName).c_str());
 #endif
 	return UI::EVENT_DONE;

UI/MainScreen.cpp

@@ -921,7 +921,7 @@ UI::EventReturn MainScreen::OnDownloadUpgrade(UI::EventParams &e) {
 	}
 #else
 	// Go directly to ppsspp.org and let the user sort it out
-	LaunchBrowser("http://www.ppsspp.org/downloads.html");
+	LaunchBrowser("https://www.ppsspp.org/downloads.html");
 #endif
 	return UI::EVENT_DONE;
 }
@@ -1132,18 +1132,18 @@ UI::EventReturn MainScreen::OnSupport(UI::EventParams &e) {
 #ifdef __ANDROID__
 	LaunchBrowser("market://details?id=org.ppsspp.ppssppgold");
 #else
-	LaunchBrowser("http://central.ppsspp.org/buygold");
+	LaunchBrowser("https://central.ppsspp.org/buygold");
 #endif
 	return UI::EVENT_DONE;
 }
 
 UI::EventReturn MainScreen::OnPPSSPPOrg(UI::EventParams &e) {
-	LaunchBrowser("http://www.ppsspp.org");
+	LaunchBrowser("https://www.ppsspp.org");
 	return UI::EVENT_DONE;
 }
 
 UI::EventReturn MainScreen::OnForums(UI::EventParams &e) {
-	LaunchBrowser("http://forums.ppsspp.org");
+	LaunchBrowser("https://forums.ppsspp.org");
 	return UI::EVENT_DONE;
 }
 

UI/MiscScreens.cpp

@@ -530,9 +530,9 @@ void CreditsScreen::CreateViews() {
 	Button *back = root_->Add(new Button(di->T("Back"), new AnchorLayoutParams(260, 64, NONE, NONE, 10, 10, false)));
 	back->OnClick.Handle(this, &CreditsScreen::OnOK);
 	root_->SetDefaultFocusView(back);
-#ifndef GOLD
-	root_->Add(new Button(cr->T("Buy Gold"), new AnchorLayoutParams(260, 64, 10, NONE, NONE, 10, false)))->OnClick.Handle(this, &CreditsScreen::OnSupport);
-#endif
+	if (!System_GetPropertyBool(SYSPROP_APP_GOLD)) {
+		root_->Add(new Button(cr->T("Buy Gold"), new AnchorLayoutParams(260, 64, 10, NONE, NONE, 10, false)))->OnClick.Handle(this, &CreditsScreen::OnSupport);
+	}
 	root_->Add(new Button(cr->T("PPSSPP Forums"), new AnchorLayoutParams(260, 64, 10, NONE, NONE, 84, false)))->OnClick.Handle(this, &CreditsScreen::OnForums);
 	root_->Add(new Button("www.ppsspp.org", new AnchorLayoutParams(260, 64, 10, NONE, NONE, 158, false)))->OnClick.Handle(this, &CreditsScreen::OnPPSSPPOrg);
 #ifdef __ANDROID__
@@ -550,7 +550,7 @@ UI::EventReturn CreditsScreen::OnSupport(UI::EventParams &e) {
 #ifdef __ANDROID__
 	LaunchBrowser("market://details?id=org.ppsspp.ppssppgold");
 #else
-	LaunchBrowser("http://central.ppsspp.org/buygold");
+	LaunchBrowser("https://central.ppsspp.org/buygold");
 #endif
 	return UI::EVENT_DONE;
 }
@@ -565,12 +565,12 @@ UI::EventReturn CreditsScreen::OnTwitter(UI::EventParams &e) {
 }
 
 UI::EventReturn CreditsScreen::OnPPSSPPOrg(UI::EventParams &e) {
-	LaunchBrowser("http://www.ppsspp.org");
+	LaunchBrowser("https://www.ppsspp.org");
 	return UI::EVENT_DONE;
 }
 
 UI::EventReturn CreditsScreen::OnForums(UI::EventParams &e) {
-	LaunchBrowser("http://forums.ppsspp.org");
+	LaunchBrowser("https://forums.ppsspp.org");
 	return UI::EVENT_DONE;
 }
 
@@ -689,7 +689,7 @@ void CreditsScreen::render() {
 		"",
 		"",
 		cr->T("check", "Also check out Dolphin, the best Wii/GC emu around:"),
-		"http://www.dolphin-emu.org",
+		"https://www.dolphin-emu.org",
 		"",
 		"",
 		cr->T("info1", "PPSSPP is only intended to play games you own."),

UI/NativeApp.cpp

@@ -302,12 +302,6 @@ bool CheckFontIsUsable(const wchar_t *fontFace) {
 #endif
 
 void NativeInit(int argc, const char *argv[], const char *savegame_dir, const char *external_dir, const char *cache_dir, bool fs) {
-#ifdef ANDROID_NDK_PROFILER
-	setenv("CPUPROFILE_FREQUENCY", "500", 1);
-	setenv("CPUPROFILE", "/sdcard/gmon.out", 1);
-	monstartup("ppsspp_jni.so");
-#endif
-
 	net::Init();  // This needs to happen before we load the config. So on Windows we also run it in Main. It's fine to call multiple times.
 
 	InitFastMath(cpu_info.bNEON);
@@ -442,6 +436,7 @@ void NativeInit(int argc, const char *argv[], const char *savegame_dir, const ch
 			}
 		} else {
 			if (boot_filename.empty()) {
+				ILOG("Boot filename found in args: %s", argv[i]);
 				boot_filename = argv[i];
 #ifdef _WIN32
 				boot_filename = ReplaceAll(boot_filename, "\\", "/");
@@ -460,7 +455,7 @@ void NativeInit(int argc, const char *argv[], const char *savegame_dir, const ch
 		}
 	}
 
-	if (fileToLog != NULL)
+	if (fileToLog)
 		LogManager::GetInstance()->ChangeFileLog(fileToLog);
 
 #ifndef _WIN32
@@ -874,8 +869,12 @@ void HandleGlobalMessage(const std::string &msg, const std::string &value) {
 		std::string setString = inputboxValue.size() > 1 ? inputboxValue[1] : "";
 		if (inputboxValue[0] == "IP")
 			g_Config.proAdhocServer = setString;
-		if (inputboxValue[0] == "nickname")
+		else if (inputboxValue[0] == "nickname")
 			g_Config.sNickName = setString;
+		else if (inputboxValue[0] == "remoteiso_subdir")
+			g_Config.sRemoteISOSubdir = setString;
+		else if (inputboxValue[0] == "remoteiso_server")
+			g_Config.sLastRemoteISOServer = setString;
 		inputboxValue.clear();
 	}
 	if (msg == "bgImage_updated") {

UI/RemoteISOScreen.cpp

@@ -675,13 +675,25 @@ void RemoteISOSettingsScreen::CreateViews() {
 
 	remoteisoSettings->Add(new ItemHeader(ri->T("Remote disc streaming")));
 	remoteisoSettings->Add(new CheckBox(&g_Config.bRemoteISOManual, ri->T("Manual Mode Client", "Manual Mode Client")));
+#if !defined(MOBILE_DEVICE)
 	PopupTextInputChoice *remoteServer = remoteisoSettings->Add(new PopupTextInputChoice(&g_Config.sLastRemoteISOServer, ri->T("Remote Server"), "", 255, screenManager()));
+#else
+	ChoiceWithValueDisplay *remoteServer = new ChoiceWithValueDisplay(&g_Config.sLastRemoteISOServer, ri->T("Remote Server"), nullptr);
+	remoteisoSettings->Add(remoteServer);
+	remoteServer->OnClick.Handle(this, &RemoteISOSettingsScreen::OnClickRemoteServer);
+#endif
 	remoteServer->SetEnabledPtr(&g_Config.bRemoteISOManual);
 	PopupSliderChoice *remotePort = remoteisoSettings->Add(new PopupSliderChoice(&g_Config.iLastRemoteISOPort, 0, 65535, ri->T("Remote Port", "Remote Port"), 100, screenManager()));
 	remotePort->SetEnabledPtr(&g_Config.bRemoteISOManual);
+#if !defined(MOBILE_DEVICE)
 	PopupTextInputChoice *remoteSubdir = remoteisoSettings->Add(new PopupTextInputChoice(&g_Config.sRemoteISOSubdir, ri->T("Remote Subdirectory"), "", 255, screenManager()));
-	remoteSubdir->SetEnabledPtr(&g_Config.bRemoteISOManual);
 	remoteSubdir->OnChange.Handle(this, &RemoteISOSettingsScreen::OnChangeRemoteISOSubdir);
+#else
+	ChoiceWithValueDisplay *remoteSubdir = remoteisoSettings->Add(
+			new ChoiceWithValueDisplay(&g_Config.sRemoteISOSubdir, ri->T("Remote Subdirectory"), nullptr));
+	remoteSubdir->OnClick.Handle(this, &RemoteISOSettingsScreen::OnClickRemoteISOSubdir);
+#endif
+	remoteSubdir->SetEnabledPtr(&g_Config.bRemoteISOManual);
 
 	PopupSliderChoice *portChoice = new PopupSliderChoice(&g_Config.iRemoteISOPort, 0, 65535, ri->T("Local Server Port", "Local Server Port"), 100, screenManager());
 	remoteisoSettings->Add(portChoice);
@@ -693,6 +705,16 @@ void RemoteISOSettingsScreen::CreateViews() {
 	AddStandardBack(root_);
 }
 
+UI::EventReturn RemoteISOSettingsScreen::OnClickRemoteServer(UI::EventParams &e) {
+	System_SendMessage("inputbox", ("remoteiso_server:" + g_Config.sLastRemoteISOServer).c_str());
+	return UI::EVENT_DONE;
+}
+
+UI::EventReturn RemoteISOSettingsScreen::OnClickRemoteISOSubdir(UI::EventParams &e) {
+	System_SendMessage("inputbox", ("remoteiso_subdir:" + g_Config.sRemoteISOSubdir).c_str());
+	return UI::EVENT_DONE;
+}
+
 UI::EventReturn RemoteISOSettingsScreen::OnChangeRemoteISOSubdir(UI::EventParams &e) {
 	//Conform to HTTP standards
 	ReplaceAll(g_Config.sRemoteISOSubdir, " ", "%20");

UI/RemoteISOScreen.h

@@ -89,7 +89,10 @@ class RemoteISOSettingsScreen : public UIDialogScreenWithBackground {
 public:
 	RemoteISOSettingsScreen();
 
+	UI::EventReturn OnClickRemoteISOSubdir(UI::EventParams &e);
+	UI::EventReturn OnClickRemoteServer(UI::EventParams &e);
 protected:
+
 	void update() override;
 	void CreateViews() override;
 

Windows/GPU/WindowsVulkanContext.cpp

@@ -173,6 +173,12 @@ bool WindowsVulkanContext::Init(HINSTANCE hInst, HWND hWnd, std::string *error_m
 
 	Version gitVer(PPSSPP_GIT_VERSION);
 	g_Vulkan = new VulkanContext();
+	if (g_Vulkan->InitError().size()) {
+		*error_message = g_Vulkan->InitError();
+		delete g_Vulkan;
+		g_Vulkan = nullptr;
+		return false;
+	}
 
 	// int vulkanFlags = VULKAN_FLAG_PRESENT_FIFO_RELAXED;
 	int vulkanFlags = VULKAN_FLAG_PRESENT_MAILBOX;
@@ -181,6 +187,8 @@ bool WindowsVulkanContext::Init(HINSTANCE hInst, HWND hWnd, std::string *error_m
 	}
 	if (VK_SUCCESS != g_Vulkan->CreateInstance("PPSSPP", gitVer.ToInteger(), vulkanFlags)) {
 		*error_message = g_Vulkan->InitError();
+		delete g_Vulkan;
+		g_Vulkan = nullptr;
 		return false;
 	}
 	g_Vulkan->ChooseDevice(g_Vulkan->GetBestPhysicalDevice());
@@ -189,6 +197,8 @@ bool WindowsVulkanContext::Init(HINSTANCE hInst, HWND hWnd, std::string *error_m
 	}
 	if (g_Vulkan->CreateDevice() != VK_SUCCESS) {
 		*error_message = g_Vulkan->InitError();
+		delete g_Vulkan;
+		g_Vulkan = nullptr;
 		return false;
 	}
 	if (g_validate_) {
@@ -197,6 +207,7 @@ bool WindowsVulkanContext::Init(HINSTANCE hInst, HWND hWnd, std::string *error_m
 	}
 	g_Vulkan->InitSurfaceWin32(hInst, hWnd);
 	if (!g_Vulkan->InitObjects()) {
+		*error_message = g_Vulkan->InitError();
 		Shutdown();
 		return false;
 	}

android/jni/app-android.cpp

@@ -291,6 +291,7 @@ bool AndroidVulkanContext::Init(ANativeWindow *wnd, int desiredBackbufferSizeX,
 	}
 	if (VK_SUCCESS != g_Vulkan->CreateInstance("PPSSPP", gitVer.ToInteger(), VULKAN_FLAG_PRESENT_MAILBOX | VULKAN_FLAG_PRESENT_FIFO_RELAXED)) {
 		ELOG("Failed to create vulkan context: %s", g_Vulkan->InitError().c_str());
+		System_SendMessage("toast", "No Vulkan compatible device found. Using OpenGL instead.");
 		delete g_Vulkan;
 		g_Vulkan = nullptr;
 		return false;
@@ -964,13 +965,16 @@ extern "C" void JNICALL Java_org_ppsspp_ppsspp_NativeApp_sendMessage(JNIEnv *env
 	if (msg == "moga") {
 		mogaVersion = prm;
 	} else if (msg == "permission_pending") {
+		ILOG("STORAGE PERMISSION: PENDING");
 		// TODO: Add support for other permissions
 		permissions[SYSTEM_PERMISSION_STORAGE] = PERMISSION_STATUS_PENDING;
 		NativePermissionStatus(SYSTEM_PERMISSION_STORAGE, PERMISSION_STATUS_PENDING);
 	} else if (msg == "permission_denied") {
+		ILOG("STORAGE PERMISSION: DENIED");
 		permissions[SYSTEM_PERMISSION_STORAGE] = PERMISSION_STATUS_DENIED;
 		NativePermissionStatus(SYSTEM_PERMISSION_STORAGE, PERMISSION_STATUS_PENDING);
 	} else if (msg == "permission_granted") {
+		ILOG("STORAGE PERMISSION: GRANTED");
 		permissions[SYSTEM_PERMISSION_STORAGE] = PERMISSION_STATUS_GRANTED;
 		NativePermissionStatus(SYSTEM_PERMISSION_STORAGE, PERMISSION_STATUS_PENDING);
 	} else if (msg == "sustained_perf_supported") {

android/src/org/ppsspp/ppsspp/NativeActivity.java

@@ -310,7 +310,7 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
 
 		// OK, config should be initialized, we can query for screen rotation.
 		if (Build.VERSION.SDK_INT >= 9) {
-			updateScreenRotation();
+			updateScreenRotation("Initialize");
 		}
 
 	    // Detect OpenGL support.
@@ -355,7 +355,7 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
 	}
 
 	@TargetApi(9)
-	private void updateScreenRotation() {
+	private void updateScreenRotation(String cause) {
 		// Query the native application on the desired rotation.
 		int rot = 0;
 		String rotString = NativeApp.queryConfig("screenRotation");
@@ -365,11 +365,12 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
 			Log.e(TAG, "Invalid rotation: " + rotString);
 			return;
 		}
-		Log.i(TAG, "Setting requested rotation: " + rot + " ('" + rotString + "')");
+		Log.i(TAG, "Setting requested rotation: " + rot + " ('" + rotString + "') (" + cause + ")");
 
 		switch (rot) {
 		case 0:
-			setRequestedOrientation(ActivityInfo.SCREEN_ORIENTATION_UNSPECIFIED);
+			// Auto is no longer supported.
+			setRequestedOrientation(ActivityInfo.SCREEN_ORIENTATION_LANDSCAPE);
 			break;
 		case 1:
 			setRequestedOrientation(ActivityInfo.SCREEN_ORIENTATION_LANDSCAPE);
@@ -462,7 +463,7 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
 		refreshRate = display.getRefreshRate();
 
 		// OK, config should be initialized, we can query for screen rotation.
-		updateScreenRotation();
+		updateScreenRotation("onCreate");
 		updateSustainedPerformanceMode();
 
 		// Keep the screen bright - very annoying if it goes dark when tilting away
@@ -519,12 +520,26 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
     }
 
 	private Point desiredSize = new Point();
-
+	private int badOrientationCount = 0;
 	@Override
 	public void surfaceCreated(SurfaceHolder holder) {
 		pixelWidth = holder.getSurfaceFrame().width();
 		pixelHeight = holder.getSurfaceFrame().height();
-		Log.d(TAG, "Surface created. pixelWidth=" + pixelWidth + ", pixelHeight=" + pixelHeight + " holder: " + holder.toString());
+
+		// Workaround for terrible bug when locking and unlocking the screen in landscape mode on Nexus 5X.
+		int requestedOr = getRequestedOrientation();
+		boolean requestedPortrait = requestedOr == ActivityInfo.SCREEN_ORIENTATION_PORTRAIT || requestedOr == ActivityInfo.SCREEN_ORIENTATION_REVERSE_PORTRAIT;
+		boolean detectedPortrait = pixelHeight > pixelWidth;
+		if (badOrientationCount < 3 && requestedPortrait != detectedPortrait) {
+			Log.e(TAG, "Bad orientation detected (w=" + pixelWidth + " h=" + pixelHeight + "! Recreating activity.");
+			badOrientationCount++;
+			recreate();;
+			return;
+		} else if (requestedPortrait == detectedPortrait) {
+			badOrientationCount = 0;
+		}
+
+		Log.d(TAG, "Surface created. pixelWidth=" + pixelWidth + ", pixelHeight=" + pixelHeight + " holder: " + holder.toString() + " or: " + requestedOr);
 		NativeApp.setDisplayParameters(pixelWidth, pixelHeight, (int)densityDpi, refreshRate);
 		getDesiredBackbufferSize(desiredSize);
 
@@ -687,7 +702,7 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
         }
 		// OK, config should be initialized, we can query for screen rotation.
 		if (javaGL || Build.VERSION.SDK_INT >= 9) {
-			updateScreenRotation();
+			updateScreenRotation("onResume");
 		}
 
 		Log.i(TAG, "onResume");
@@ -840,10 +855,10 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
 			return input.getDescriptor();
 		} else {
 			List<InputDevice.MotionRange> motions = input.getMotionRanges();
-			String fakeid = "";
+			StringBuilder fakeid = new StringBuilder();
 			for (InputDevice.MotionRange range : motions)
-				fakeid += range.getAxis();
-			return fakeid;
+				fakeid.append(range.getAxis());
+			return fakeid.toString();
 		}
 	}
 
@@ -999,7 +1014,7 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
     	input.selectAll();
 
     	// Lovely!
-    	AlertDialog.Builder bld = null;
+    	AlertDialog.Builder bld;
     	if (Build.VERSION.SDK_INT < Build.VERSION_CODES.HONEYCOMB)
     		bld = new AlertDialog.Builder(this);
     	else if (Build.VERSION.SDK_INT < Build.VERSION_CODES.ICE_CREAM_SANDWICH)
@@ -1181,13 +1196,13 @@ public abstract class NativeActivity extends Activity implements SurfaceHolder.C
 			finish();
 		} else if (command.equals("rotate")) {
 			if (javaGL) {
-				updateScreenRotation();
+				updateScreenRotation("rotate");
 				if (Build.VERSION.SDK_INT < Build.VERSION_CODES.ICE_CREAM_SANDWICH) {
 					Log.i(TAG, "Must recreate activity on rotation");
 				}
 			} else {
 				if (Build.VERSION.SDK_INT >= 9) {
-					updateScreenRotation();
+					updateScreenRotation("rotate");
 				}
 			}
 		} else if (command.equals("sustainedPerfMode")) {

android/src/org/ppsspp/ppsspp/PpssppActivity.java

@@ -75,23 +75,27 @@ public class PpssppActivity extends NativeActivity {
 		// using Intent extra string. Intent extra will be null if launch normal
 		// (from app drawer or file explorer).
 		Intent intent = getIntent();
-		String action = intent.getAction();
-		if (Intent.ACTION_VIEW.equals(action)) {
-			Uri data = intent.getData();
-			if (data != null) {
-				String path = intent.getData().getPath();
-				super.setShortcutParam(path);
-				Toast.makeText(getApplicationContext(), path, Toast.LENGTH_SHORT).show();
-			} else {
-				Log.e(TAG, "Got ACTION_VIEW without a valid uri");
-			}
+		// String action = intent.getAction();
+		Uri data = intent.getData();
+		if (data != null) {
+			String path = intent.getData().getPath();
+			Log.i(TAG, "Found Shortcut Parameter in data: " + path);
+			super.setShortcutParam(path);
+			// Toast.makeText(getApplicationContext(), path, Toast.LENGTH_SHORT).show();
 		} else {
-			super.setShortcutParam(getIntent().getStringExtra(SHORTCUT_EXTRA_KEY));
+			String param = getIntent().getStringExtra(SHORTCUT_EXTRA_KEY);
+			Log.e(TAG, "Got ACTION_VIEW without a valid uri, trying param");
+			if (param != null) {
+				Log.i(TAG, "Found Shortcut Parameter in extra-data: " + param);
+				super.setShortcutParam(getIntent().getStringExtra(SHORTCUT_EXTRA_KEY));
+			} else {
+				Log.e(TAG, "Shortcut missing parameter!");
+			}
 		}
-
 		super.onCreate(savedInstanceState);
 	}
 
+
 	// called by the C++ code through JNI. Dispatch anything we can't directly handle
 	// on the gfx thread to the UI thread.
 	public void postCommand(String command, String parameter) {

android/src/org/ppsspp/ppsspp/ShortcutActivity.java

@@ -6,15 +6,18 @@ import android.app.Activity;
 import android.app.AlertDialog;
 import android.content.Intent;
 import android.content.Intent.ShortcutIconResource;
+import android.net.Uri;
 import android.os.Bundle;
 import android.os.Environment;
 import android.os.Looper;
+import android.util.Log;
 
 /**
  * This class will respond to android.intent.action.CREATE_SHORTCUT intent from
  * launcher homescreen. Register this class in AndroidManifest.xml.
  */
 public class ShortcutActivity extends Activity {
+	private static final String TAG = "PPSSPP";
 
 	@Override
 	protected void onCreate(Bundle savedInstanceState) {
@@ -33,8 +36,12 @@ public class ShortcutActivity extends Activity {
 	private void respondToShortcutRequest(String path) {
 		// This is Intent that will be sent when user execute our shortcut on
 		// homescreen. Set our app as target Context. Set Main activity as
-		// target class. Add any parameter to extra.
+		// target class. Add any parameter as data.
 		Intent shortcutIntent = new Intent(this, PpssppActivity.class);
+		Uri uri = Uri.fromFile(new File(path));
+		Log.i(TAG, "Shortcut URI: " + uri.toString());
+		shortcutIntent.setData(uri);
+
 		shortcutIntent.putExtra(PpssppActivity.SHORTCUT_EXTRA_KEY, path);
 
 		PpssppActivity.CheckABIAndLoadLibrary();

assets/compat.ini

@@ -148,6 +148,7 @@ UCES00373 = true
 # Crash Tag Team Racing needs it to pass checking memory stick screen.
 ULUS10044 = true
 ULES00168 = true
+ULES00172 = true
 ULJM05036 = true
 # Gundam Battle Royale might need it to avoid crashes when certain Ace enemies shows up
 ULJS00083 = true

ext/native/thin3d/DataFormat.h

@@ -73,6 +73,7 @@ inline bool DataFormatIsColor(DataFormat fmt) {
 }
 
 void ConvertFromRGBA8888(uint8_t *dst, const uint8_t *src, uint32_t dstStride, uint32_t srcStride, uint32_t width, uint32_t height, DataFormat format);
+void ConvertFromBGRA8888(uint8_t *dst, const uint8_t *src, uint32_t dstStride, uint32_t srcStride, uint32_t width, uint32_t height, DataFormat format);
 void ConvertToD32F(uint8_t *dst, const uint8_t *src, uint32_t dstStride, uint32_t srcStride, uint32_t width, uint32_t height, DataFormat format);
 
 }  // namespace

ext/native/thin3d/VulkanQueueRunner.cpp

@@ -837,22 +837,6 @@ void VulkanQueueRunner::SetupTransitionToTransferDst(VKRImage &img, VkImageMemor
 }
 
 void VulkanQueueRunner::PerformReadback(const VKRStep &step, VkCommandBuffer cmd) {
-	VKRImage *srcImage;
-	if (step.readback.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
-		srcImage = &step.readback.src->color;
-	} else if (step.readback.aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
-		srcImage = &step.readback.src->depth;
-	} else {
-		assert(false);
-	}
-
-	VkImageMemoryBarrier barrier{ VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER };
-	VkPipelineStageFlags stage = 0;
-	if (srcImage->layout != VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL) {
-		SetupTransitionToTransferSrc(*srcImage, barrier, stage, step.readback.aspectMask);
-		vkCmdPipelineBarrier(cmd, stage, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &barrier);
-	}
-
 	ResizeReadbackBuffer(sizeof(uint32_t) * step.readback.srcRect.extent.width * step.readback.srcRect.extent.height);
 
 	VkBufferImageCopy region{};
@@ -863,9 +847,55 @@ void VulkanQueueRunner::PerformReadback(const VKRStep &step, VkCommandBuffer cmd
 	region.bufferOffset = 0;
 	region.bufferRowLength = step.readback.srcRect.extent.width;
 	region.bufferImageHeight = step.readback.srcRect.extent.height;
-	vkCmdCopyImageToBuffer(cmd, srcImage->image, srcImage->layout, readbackBuffer_, 1, &region);
+
+	VkImage image;
+	VkImageLayout copyLayout;
+	// Special case for backbuffer readbacks.
+	if (step.readback.src == nullptr) {
+		// We only take screenshots after the main render pass (anything else would be stupid) so we need to transition out of PRESENT,
+		// and then back into it.
+		TransitionImageLayout2(cmd, backbufferImage_, VK_IMAGE_ASPECT_COLOR_BIT,
+			VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+			VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+			0, VK_ACCESS_TRANSFER_READ_BIT);
+		copyLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
+		image = backbufferImage_;
+	} else {
+		VKRImage *srcImage;
+		if (step.readback.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
+			srcImage = &step.readback.src->color;
+		}
+		else if (step.readback.aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
+			srcImage = &step.readback.src->depth;
+		}
+		else {
+			assert(false);
+		}
+
+		VkImageMemoryBarrier barrier{ VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER };
+		VkPipelineStageFlags stage = 0;
+		if (srcImage->layout != VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL) {
+			SetupTransitionToTransferSrc(*srcImage, barrier, stage, step.readback.aspectMask);
+			vkCmdPipelineBarrier(cmd, stage, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &barrier);
+		}
+		image = srcImage->image;
+		copyLayout = srcImage->layout;
+	}
+
+	vkCmdCopyImageToBuffer(cmd, image, copyLayout, readbackBuffer_, 1, &region);
 
 	// NOTE: Can't read the buffer using the CPU here - need to sync first.
+
+	// If we copied from the backbuffer, transition it back.
+	if (step.readback.src == nullptr) {
+		// We only take screenshots after the main render pass (anything else would be stupid) so we need to transition out of PRESENT,
+		// and then back into it.
+		TransitionImageLayout2(cmd, backbufferImage_, VK_IMAGE_ASPECT_COLOR_BIT,
+			VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
+			VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+			VK_ACCESS_TRANSFER_READ_BIT, 0);
+		copyLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
+	}
 }
 
 void VulkanQueueRunner::PerformReadbackImage(const VKRStep &step, VkCommandBuffer cmd) {
@@ -911,6 +941,8 @@ void VulkanQueueRunner::CopyReadbackBuffer(int width, int height, Draw::DataForm
 	assert(res == VK_SUCCESS);
 	if (srcFormat == Draw::DataFormat::R8G8B8A8_UNORM) {
 		ConvertFromRGBA8888(pixels, (const uint8_t *)mappedData, pixelStride, width, width, height, destFormat);
+	} else if (srcFormat == Draw::DataFormat::B8G8R8A8_UNORM) {
+		ConvertFromBGRA8888(pixels, (const uint8_t *)mappedData, pixelStride, width, width, height, destFormat);
 	} else if (srcFormat == destFormat) {
 		uint8_t *dst = pixels;
 		const uint8_t *src = (const uint8_t *)mappedData;

ext/native/thin3d/VulkanQueueRunner.h

@@ -145,8 +145,9 @@ struct VKRStep {
 class VulkanQueueRunner {
 public:
 	VulkanQueueRunner(VulkanContext *vulkan) : vulkan_(vulkan) {}
-	void SetBackbuffer(VkFramebuffer fb) {
+	void SetBackbuffer(VkFramebuffer fb, VkImage img) {
 		backbuffer_ = fb;
+		backbufferImage_ = img;
 	}
 	void RunSteps(VkCommandBuffer cmd, const std::vector<VKRStep *> &steps);
 	void LogSteps(const std::vector<VKRStep *> &steps);
@@ -192,6 +193,7 @@ private:
 	VulkanContext *vulkan_;
 
 	VkFramebuffer backbuffer_;
+	VkImage backbufferImage_;
 	VkFramebuffer curFramebuffer_ = VK_NULL_HANDLE;
 
 	VkRenderPass backbufferRenderPass_ = VK_NULL_HANDLE;

ext/native/thin3d/VulkanRenderManager.cpp

@@ -11,9 +11,6 @@
 #define VLOG(...)
 #endif
 
-// This works great. Not much reason to disable so let's not even bother with an option.
-const bool useThread = true;
-
 #ifndef UINT64_MAX
 #define UINT64_MAX 0xFFFFFFFFFFFFFFFFULL
 #endif
@@ -127,6 +124,11 @@ VulkanRenderManager::VulkanRenderManager(VulkanContext *vulkan) : vulkan_(vulkan
 	}
 
 	queueRunner_.CreateDeviceObjects();
+
+	// Temporary AMD hack for issue #10097
+	if (vulkan_->GetPhysicalDeviceProperties().vendorID == VULKAN_VENDOR_AMD) {
+		useThread_ = false;
+	}
 }
 
 void VulkanRenderManager::CreateBackbuffers() {
@@ -188,7 +190,7 @@ void VulkanRenderManager::CreateBackbuffers() {
 	}
 
 	// Start the thread.
-	if (useThread && HasBackbuffers()) {
+	if (useThread_ && HasBackbuffers()) {
 		run_ = true;
 		// Won't necessarily be 0.
 		threadInitFrame_ = vulkan_->GetCurFrame();
@@ -198,7 +200,7 @@ void VulkanRenderManager::CreateBackbuffers() {
 }
 
 void VulkanRenderManager::StopThread() {
-	if (useThread && run_) {
+	if (useThread_ && run_) {
 		run_ = false;
 		// Stop the thread.
 		for (int i = 0; i < vulkan_->GetInflightFrames(); i++) {
@@ -337,7 +339,7 @@ void VulkanRenderManager::BeginFrame() {
 	FrameData &frameData = frameData_[curFrame];
 
 	// Make sure the very last command buffer from the frame before the previous has been fully executed.
-	if (useThread) {
+	if (useThread_) {
 		std::unique_lock<std::mutex> lock(frameData.push_mutex);
 		while (!frameData.readyForFence) {
 			VLOG("PUSH: Waiting for frame[%d].readyForFence = 1", curFrame);
@@ -413,7 +415,7 @@ void VulkanRenderManager::BindFramebufferAsRenderTarget(VKRFramebuffer *fb, VKRR
 	curHeight_ = fb ? fb->height : vulkan_->GetBackbufferHeight();
 }
 
-void VulkanRenderManager::CopyFramebufferToMemorySync(VKRFramebuffer *src, int aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride) {
+bool VulkanRenderManager::CopyFramebufferToMemorySync(VKRFramebuffer *src, int aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride) {
 	VKRStep *step = new VKRStep{ VKRStepType::READBACK };
 	step->readback.aspectMask = aspectBits;
 	step->readback.src = src;
@@ -427,9 +429,26 @@ void VulkanRenderManager::CopyFramebufferToMemorySync(VKRFramebuffer *src, int a
 
 	Draw::DataFormat srcFormat;
 	if (aspectBits & VK_IMAGE_ASPECT_COLOR_BIT) {
-		switch (src->color.format) {
-		case VK_FORMAT_R8G8B8A8_UNORM: srcFormat = Draw::DataFormat::R8G8B8A8_UNORM; break;
-		default: assert(false);
+		if (src) {
+			switch (src->color.format) {
+			case VK_FORMAT_R8G8B8A8_UNORM: srcFormat = Draw::DataFormat::R8G8B8A8_UNORM; break;
+			default: assert(false);
+			}
+		}
+		else {
+			// Backbuffer.
+			if (!(vulkan_->GetSurfaceCapabilities().supportedUsageFlags & VK_IMAGE_USAGE_TRANSFER_SRC_BIT)) {
+				ELOG("Copying from backbuffer not supported, can't take screenshots");
+				return false;
+			}
+			switch (vulkan_->GetSwapchainFormat()) {
+			case VK_FORMAT_B8G8R8A8_UNORM: srcFormat = Draw::DataFormat::B8G8R8A8_UNORM; break;
+			case VK_FORMAT_R8G8B8A8_UNORM: srcFormat = Draw::DataFormat::R8G8B8A8_UNORM; break;
+			// NOTE: If you add supported formats here, make sure to also support them in VulkanQueueRunner::CopyReadbackBuffer.
+			default:
+				ELOG("Unsupported backbuffer format for screenshots");
+				return false;
+			}
 		}
 	} else if (aspectBits & VK_IMAGE_ASPECT_STENCIL_BIT) {
 		// Copies from stencil are always S8.
@@ -446,6 +465,7 @@ void VulkanRenderManager::CopyFramebufferToMemorySync(VKRFramebuffer *src, int a
 	}
 	// Need to call this after FlushSync so the pixels are guaranteed to be ready in CPU-accessible VRAM.
 	queueRunner_.CopyReadbackBuffer(w, h, srcFormat, destFormat, pixelStride, pixels);
+	return true;
 }
 
 void VulkanRenderManager::CopyImageToMemorySync(VkImage image, int mipLevel, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride) {
@@ -675,7 +695,7 @@ void VulkanRenderManager::Finish() {
 	curRenderStep_ = nullptr;
 	int curFrame = vulkan_->GetCurFrame();
 	FrameData &frameData = frameData_[curFrame];
-	if (!useThread) {
+	if (!useThread_) {
 		frameData.steps = std::move(steps_);
 		frameData.type = VKRRunType::END;
 		Run(curFrame);
@@ -724,7 +744,7 @@ void VulkanRenderManager::BeginSubmitFrame(int frame) {
 
 		assert(res == VK_SUCCESS);
 
-		queueRunner_.SetBackbuffer(framebuffers_[frameData.curSwapchainImage]);
+		queueRunner_.SetBackbuffer(framebuffers_[frameData.curSwapchainImage], swapchainImages_[frameData.curSwapchainImage].image);
 
 		frameData.hasBegun = true;
 	}
@@ -782,7 +802,7 @@ void VulkanRenderManager::Submit(int frame, bool triggerFence) {
 	}
 
 	// When !triggerFence, we notify after syncing with Vulkan.
-	if (useThread && triggerFence) {
+	if (useThread_ && triggerFence) {
 		VLOG("PULL: Frame %d.readyForFence = true", frame);
 		std::unique_lock<std::mutex> lock(frameData.push_mutex);
 		frameData.readyForFence = true;
@@ -862,7 +882,7 @@ void VulkanRenderManager::EndSyncFrame(int frame) {
 	VkResult res = vkBeginCommandBuffer(frameData.mainCmd, &begin);
 	assert(res == VK_SUCCESS);
 
-	if (useThread) {
+	if (useThread_) {
 		std::unique_lock<std::mutex> lock(frameData.push_mutex);
 		frameData.readyForFence = true;
 		frameData.push_condVar.notify_all();
@@ -873,7 +893,7 @@ void VulkanRenderManager::FlushSync() {
 	// TODO: Reset curRenderStep_?
 	int curFrame = vulkan_->GetCurFrame();
 	FrameData &frameData = frameData_[curFrame];
-	if (!useThread) {
+	if (!useThread_) {
 		frameData.steps = std::move(steps_);
 		frameData.type = VKRRunType::SYNC;
 		Run(curFrame);
@@ -887,7 +907,7 @@ void VulkanRenderManager::FlushSync() {
 		frameData.pull_condVar.notify_all();
 	}
 
-	if (useThread) {
+	if (useThread_) {
 		std::unique_lock<std::mutex> lock(frameData.push_mutex);
 		// Wait for the flush to be hit, since we're syncing.
 		while (!frameData.readyForFence) {

ext/native/thin3d/VulkanRenderManager.h

@@ -94,7 +94,7 @@ public:
 
 	void BindFramebufferAsRenderTarget(VKRFramebuffer *fb, VKRRenderPassAction color, VKRRenderPassAction depth, uint32_t clearColor, float clearDepth, uint8_t clearStencil);
 	VkImageView BindFramebufferAsTexture(VKRFramebuffer *fb, int binding, int aspectBit, int attachment);
-	void CopyFramebufferToMemorySync(VKRFramebuffer *src, int aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride);
+	bool CopyFramebufferToMemorySync(VKRFramebuffer *src, int aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride);
 	void CopyImageToMemorySync(VkImage image, int mipLevel, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride);
 
 	void CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkOffset2D dstPos, int aspectMask);
@@ -283,4 +283,7 @@ private:
 		VkImageView view = VK_NULL_HANDLE;
 	};
 	DepthBufferInfo depth_;
+
+	// This works great - except see issue #10097. WTF?
+	bool useThread_ = true;
 };

ext/native/thin3d/thin3d.cpp

@@ -381,6 +381,25 @@ void ConvertFromRGBA8888(uint8_t *dst, const uint8_t *src, uint32_t dstStride, u
 	}
 }
 
+// TODO: SSE/NEON
+// Could also make C fake-simd for 64-bit, two 8888 pixels fit in a register :)
+void ConvertFromBGRA8888(uint8_t *dst, const uint8_t *src, uint32_t dstStride, uint32_t srcStride, uint32_t width, uint32_t height, DataFormat format) {
+	// Must skip stride in the cases below.  Some games pack data into the cracks, like MotoGP.
+	const uint32_t *src32 = (const uint32_t *)src;
+
+	if (format == Draw::DataFormat::R8G8B8A8_UNORM) {
+		uint32_t *dst32 = (uint32_t *)dst;
+		for (uint32_t y = 0; y < height; ++y) {
+			ConvertBGRA8888ToRGBA8888(dst32, src32, width);
+			src32 += srcStride;
+			dst32 += dstStride;
+		}
+	}
+	else {
+		// Don't even bother with these, this path only happens in screenshots and we don't save those to 16-bit.
+		assert(false);
+	}
+}
 void ConvertToD32F(uint8_t *dst, const uint8_t *src, uint32_t dstStride, uint32_t srcStride, uint32_t width, uint32_t height, DataFormat format) {
 	if (format == Draw::DataFormat::D32F) {
 		const float *src32 = (const float *)src;

ext/native/thin3d/thin3d_vulkan.cpp

@@ -1187,6 +1187,20 @@ void AddFeature(std::vector<std::string> &features, const char *name, VkBool32 a
 	features.push_back(buf);
 }
 
+// Limited to depth buffer formats as that's what we need right now.
+static const char *VulkanFormatToString(VkFormat fmt) {
+	switch (fmt) {
+	case VkFormat::VK_FORMAT_D24_UNORM_S8_UINT: return "D24S8";
+	case VkFormat::VK_FORMAT_D16_UNORM: return "D16";
+	case VkFormat::VK_FORMAT_D16_UNORM_S8_UINT: return "D16S8";
+	case VkFormat::VK_FORMAT_D32_SFLOAT: return "D32f";
+	case VkFormat::VK_FORMAT_D32_SFLOAT_S8_UINT: return "D32fS8";
+	case VkFormat::VK_FORMAT_S8_UINT: return "S8";
+	case VkFormat::VK_FORMAT_UNDEFINED: return "UNDEFINED (BAD!)";
+	default: return "UNKNOWN";
+	}
+}
+
 std::vector<std::string> VKContext::GetFeatureList() const {
 	const VkPhysicalDeviceFeatures &available = vulkan_->GetFeaturesAvailable();
 	const VkPhysicalDeviceFeatures &enabled = vulkan_->GetFeaturesEnabled();
@@ -1210,6 +1224,8 @@ std::vector<std::string> VKContext::GetFeatureList() const {
 	AddFeature(features, "occlusionQueryPrecise", available.occlusionQueryPrecise, enabled.occlusionQueryPrecise);
 	AddFeature(features, "multiDrawIndirect", available.multiDrawIndirect, enabled.multiDrawIndirect);
 
+	features.push_back(std::string("Preferred depth buffer format: ") + VulkanFormatToString(vulkan_->GetDeviceInfo().preferredDepthStencilFormat));
+
 	// Also list texture formats and their properties.
 	for (int i = VK_FORMAT_BEGIN_RANGE; i <= VK_FORMAT_END_RANGE; i++) {
 		// TODO
@@ -1321,8 +1337,7 @@ bool VKContext::CopyFramebufferToMemorySync(Framebuffer *srcfb, int channelBits,
 	if (channelBits & FBChannel::FB_DEPTH_BIT) aspectMask |= VK_IMAGE_ASPECT_DEPTH_BIT;
 	if (channelBits & FBChannel::FB_STENCIL_BIT) aspectMask |= VK_IMAGE_ASPECT_STENCIL_BIT;
 
-	renderManager_.CopyFramebufferToMemorySync(src->GetFB(), aspectMask, x, y, w, h, format, (uint8_t *)pixels, pixelStride);
-	return true;
+	return renderManager_.CopyFramebufferToMemorySync(src ? src->GetFB() : nullptr, aspectMask, x, y, w, h, format, (uint8_t *)pixels, pixelStride);
 }
 
 void VKContext::BindFramebufferAsRenderTarget(Framebuffer *fbo, const RenderPassInfo &rp) {

headless/Headless.cpp

@@ -161,6 +161,8 @@ bool RunAutoTest(HeadlessHost *headlessHost, CoreParameter &coreParameter, bool
 	Core_UpdateDebugStats(g_Config.bShowDebugStats || g_Config.bLogFrameDrops);
 
 	PSP_BeginHostFrame();
+	if (coreParameter.thin3d)
+		coreParameter.thin3d->BeginFrame();
 
 	coreState = CORE_RUNNING;
 	while (coreState == CORE_RUNNING)
@@ -184,9 +186,11 @@ bool RunAutoTest(HeadlessHost *headlessHost, CoreParameter &coreParameter, bool
 			Core_Stop();
 		}
 	}
-
 	PSP_EndHostFrame();
 
+	if (coreParameter.thin3d)
+		coreParameter.thin3d->EndFrame();
+
 	PSP_Shutdown();
 
 	headlessHost->FlushDebugOutput();
@@ -268,7 +272,7 @@ int main(int argc, const char* argv[])
 			else if (!strcasecmp(gpuName, "null"))
 				gpuCore = GPUCORE_NULL;
 			else
-				return printUsage(argv[0], "Unknown gpu backend specified after --graphics=");
+				return printUsage(argv[0], "Unknown gpu backend specified after --graphics=. Allowed: software, directx9, directx11, vulkan, gles, null.");
 		}
 		// Default to GLES if no value selected.
 		else if (!strcmp(argv[i], "--graphics"))

test.py

@@ -40,7 +40,7 @@ class Command(object):
   def run(self, timeout):
     def target():
       self.process = subprocess.Popen(self.cmd, bufsize=1, stdin=subprocess.PIPE, stdout=sys.stdout, stderr=subprocess.STDOUT)
-      self.process.stdin.write(self.data)
+      self.process.stdin.write(self.data.encode('utf-8'))
       self.process.stdin.close()
       self.process.communicate()
 
@@ -397,7 +397,7 @@ def run_tests(test_list, args):
     c = Command(cmdline, '\n'.join(test_filenames))
     c.run(TIMEOUT * len(test_filenames))
 
-    print("Ran " + PPSSPP_EXE)
+    print("Ran " + ' '.join(cmdline))
 
 
 def main():