Update readme.md

.github/workflows/build.yml

@@ -4,6 +4,8 @@ on:
   push:
     branches:
     - master
+    # For testing.
+    - actions
     paths-ignore:
     - '*.{txt,md}'
     - 'Tools/**'
@@ -176,9 +178,19 @@ jobs:
       with:
         submodules: recursive
 
+    - name: Cache Qt
+      uses: actions/cache@v1
+      if: matrix.extra == 'qt'
+      id: cache-qt
+      with:
+        path: ${{ runner.workspace }}/Qt
+        key: ${{ runner.os }}-QtCache
+
     - name: Install Qt
       uses: jurplel/install-qt-action@v2
       if: matrix.extra == 'qt'
+      with:
+        cached: ${{ steps.cache-qt.outputs.cache-hit }}
 
     - uses: nttld/setup-ndk@v1
       if: matrix.extra == 'android'

.gitignore

@@ -116,6 +116,3 @@ debian/ppsspp/
 
 # RenderDoc
 *.rdc
-
-# bad output from libretro. don't want to accidentally add it
-nul

.gitlab-ci.yml

@@ -11,26 +11,49 @@ variables:
   variables:
     EXTRA_PATH: lib
 
+.windows-defs:
+  variables:
+    MAKEFILE_PATH: libretro
+
 include:
-  - template: Jobs/Code-Quality.gitlab-ci.yml
   - project: 'libretro-infrastructure/ci-templates'
     file: '/android-cmake.yml'
   - project: 'libretro-infrastructure/ci-templates'
     file: '/linux-cmake.yml'
+  - project: 'libretro-infrastructure/ci-templates'
+    file: '/windows-x64-msvc19-msys2.yml'
+  - project: 'libretro-infrastructure/ci-templates'
+    file: '/windows-i686-msvc19-msys2.yml'
 
 stages:
   - build-prepare
   - build-shared
-  - build-static
-  - test
 
-#Desktop
-libretro-build-linux-x86_64:
+# Desktop
+libretro-build-linux-x64:
   extends:
     - .libretro-linux-cmake-x86_64
     - .core-defs
     - .linux-defs
 
+libretro-build-linux-i686:
+  extends:
+    - .libretro-linux-cmake-x86
+    - .core-defs
+    - .linux-defs
+
+libretro-build-windows-x64:
+  extends:
+    - .libretro-windows-x64-msvc19-msys2-make-default
+    - .core-defs
+    - .windows-defs
+
+libretro-build-windows-i686:
+  extends:
+    - .libretro-windows-i686-msvc19-msys2-make-default
+    - .core-defs
+    - .windows-defs
+
 # Android
 libretro-build-android-armeabi-v7a:
   extends:

CMakeLists.txt

@@ -134,7 +134,8 @@ option(USE_MINIUPNPC "Build with miniUPnPc support" ON)
 option(USE_SYSTEM_SNAPPY "Dynamically link against system snappy" ${USE_SYSTEM_SNAPPY})
 option(USE_SYSTEM_FFMPEG "Dynamically link against system FFMPEG" ${USE_SYSTEM_FFMPEG})
 option(USE_SYSTEM_LIBZIP "Dynamically link against system libzip" ${USE_SYSTEM_LIBZIP})
-option(USE_ADDRESS_SANITIZER "Use Clang memory sanitizer" ${USE_ADDRESS_SANITIZER})
+option(USE_ASAN "Use address sanitizer" OFF)
+option(USE_UBSAN "Use undefined behaviour sanitizer" OFF)
 
 if(UNIX AND NOT (APPLE OR ANDROID) AND VULKAN)
 	if(USING_X11_VULKAN)
@@ -272,11 +273,24 @@ if(NOT MSVC)
 		set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-switch -Wno-uninitialized")
 	endif()
 
-	if(USE_ADDRESS_SANITIZER)
+	if(USE_ASAN)
 		message("Address sanitizer enabled (DEBUG only)")
-		set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS} -fsanitize=address")
+		set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fsanitize=address")
 		set(CMAKE_LINKER_FLAGS_DEBUG "${CMAKE_LINKER_FLAGS_DEBUG} -fsanitize=address")
-		add_definitions(-DUSE_ADDRESS_SANITIZER)
+		add_definitions(-DUSE_ASAN)
+	endif()
+	if(USE_UBSAN)
+		message("Undefined behaviour sanitizer enabled (DEBUG only)")
+		set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fsanitize=undefined")
+		set(CMAKE_LINKER_FLAGS_DEBUG "${CMAKE_LINKER_FLAGS_DEBUG} -fsanitize=undefined")
+
+		# UBSAN is a collection of sanitizers, including vtpr, which reqiuires RTTI.
+		# ext/glslang disables RTTI by default using the `ENABLE_RTTI` option.
+		# If RTTI is disabled, we must also disable the vtpr sanitizer.
+		if(NOT ENABLE_RTTI)
+			set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-sanitize=vptr")
+			set(CMAKE_LINKER_FLAGS_DEBUG "${CMAKE_LINKER_FLAGS_DEBUG} -fno-sanitize=vptr")
+		endif()
 	endif()
 
 	set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -g -D_DEBUG")
@@ -1492,6 +1506,10 @@ add_library(${CoreLibName} ${CoreLinkType}
 	Core/Debugger/WebSocket/GPURecordSubscriber.h
 	Core/Debugger/WebSocket/HLESubscriber.cpp
 	Core/Debugger/WebSocket/HLESubscriber.h
+	Core/Debugger/WebSocket/InputBroadcaster.cpp
+	Core/Debugger/WebSocket/InputBroadcaster.h
+	Core/Debugger/WebSocket/InputSubscriber.cpp
+	Core/Debugger/WebSocket/InputSubscriber.h
 	Core/Debugger/WebSocket/LogBroadcaster.cpp
 	Core/Debugger/WebSocket/LogBroadcaster.h
 	Core/Debugger/WebSocket/MemorySubscriber.cpp
@@ -2186,6 +2204,7 @@ if(IOS)
 		RESOURCE "ios/Settings.bundle"
 		RESOURCE "MoltenVK/iOS/Frameworks"
 		XCODE_ATTRIBUTE_IPHONEOS_DEPLOYMENT_TARGET ${DEPLOYMENT_TARGET}
+		XCODE_ATTRIBUTE_TARGETED_DEVICE_FAMILY "iPhone/iPad"
 		XCODE_ATTRIBUTE_CLANG_ENABLE_OBJC_ARC YES
 		XCODE_ATTRIBUTE_ENABLE_BITCODE NO
 		XCODE_ATTRIBUTE_CODE_SIGN_IDENTITY "-"

Common/GPU/OpenGL/GLCommon.h

@@ -8,9 +8,6 @@
 #elif defined(USING_GLES2)
 #include <GLES2/gl2.h>
 #include <GLES2/gl2ext.h>
-// At least Nokia platforms need the three below
-#include <KHR/khrplatform.h>
-typedef char GLchar;
 #define GL_BGRA_EXT 0x80E1
 #else // OpenGL
 #include "GL/glew.h"

Common/GPU/OpenGL/GLFeatures.cpp

@@ -181,7 +181,7 @@ void CheckGLExtensions() {
 			// Just for reference: Galaxy Y has renderer == "VideoCore IV HW"
 		} else if (vendor == "Vivante Corporation") {
 			gl_extensions.gpuVendor = GPU_VENDOR_VIVANTE;
-		} else if (vendor == "Apple Inc.") {
+		} else if (vendor == "Apple Inc." || vendor == "Apple") {
 			gl_extensions.gpuVendor = GPU_VENDOR_APPLE;
 		} else {
 			WARN_LOG(G3D, "Unknown GL vendor: '%s'", vendor.c_str());
@@ -594,6 +594,8 @@ std::string ApplyGLSLPrelude(const std::string &source, uint32_t stage) {
 	if (!gl_extensions.IsGLES && gl_extensions.IsCoreContext) {
 		// We need to add a corresponding #version.  Apple drivers fail without an exact match.
 		version = StringFromFormat("#version %d\n", gl_extensions.GLSLVersion());
+	} else if (gl_extensions.IsGLES && gl_extensions.GLES3) {
+		version = StringFromFormat("#version %d es\n", gl_extensions.GLSLVersion());
 	}
 	if (stage == GL_FRAGMENT_SHADER) {
 		temp = version + glsl_fragment_prelude + source;

Common/GPU/OpenGL/GLQueueRunner.cpp

@@ -1369,12 +1369,14 @@ void GLQueueRunner::PerformCopy(const GLRStep &step) {
 
 #if defined(USING_GLES2)
 #ifndef IOS
+	_assert_msg_(gl_extensions.OES_copy_image || gl_extensions.NV_copy_image || gl_extensions.EXT_copy_image, "Image copy extension expected");
 	glCopyImageSubDataOES(
 		srcTex, target, srcLevel, srcRect.x, srcRect.y, srcZ,
 		dstTex, target, dstLevel, dstPos.x, dstPos.y, dstZ,
 		srcRect.w, srcRect.h, depth);
 #endif
 #else
+	_assert_msg_(gl_extensions.ARB_copy_image || gl_extensions.NV_copy_image, "Image copy extension expected");
 	if (gl_extensions.ARB_copy_image) {
 		glCopyImageSubData(
 			srcTex, target, srcLevel, srcRect.x, srcRect.y, srcZ,

Common/GPU/OpenGL/thin3d_gl.cpp

@@ -302,7 +302,7 @@ class OpenGLPipeline : public Pipeline {
 
 	// TODO: Optimize by getting the locations first and putting in a custom struct
 	UniformBufferDesc dynamicUniforms;
-	GLint samplerLocs_[8];
+	GLint samplerLocs_[8]{};
 	std::vector<GLint> dynamicUniformLocs_;
 	GLRProgram *program_ = nullptr;
 
@@ -1038,8 +1038,8 @@ Pipeline *OpenGLContext::CreateGraphicsPipeline(const PipelineDesc &desc) {
 		ERROR_LOG(G3D,  "Pipeline requires at least one shader");
 		return nullptr;
 	}
-	if ((int)desc.prim >= (int)Primitive::PRIMITIVE_TYPE_COUNT) {
-		ERROR_LOG(G3D,  "Invalid primitive type");
+	if ((uint32_t)desc.prim >= (uint32_t)Primitive::PRIMITIVE_TYPE_COUNT) {
+		ERROR_LOG(G3D, "Invalid primitive type");
 		return nullptr;
 	}
 	if (!desc.depthStencil || !desc.blend || !desc.raster) {

Common/GPU/Vulkan/VulkanDebug.cpp

@@ -37,6 +37,11 @@ VKAPI_ATTR VkBool32 VKAPI_CALL VulkanDebugUtilsCallback(
 		// UNASSIGNED-CoreValidation-Shader-OutputNotConsumed - benign perf warning
 		return false;
 	}
+	if (messageCode == 1303270965) {
+		// Benign perf warning, image blit using GENERAL layout.
+		// UNASSIGNED
+		return false;
+	}
 
 	const char *pLayerPrefix = "";
 	if (messageSeverity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) {

Common/GPU/Vulkan/VulkanImage.cpp

@@ -31,6 +31,11 @@ static bool IsDepthStencilFormat(VkFormat format) {
 }
 
 bool VulkanTexture::CreateDirect(VkCommandBuffer cmd, VulkanDeviceAllocator *allocator, int w, int h, int numMips, VkFormat format, VkImageLayout initialLayout, VkImageUsageFlags usage, const VkComponentMapping *mapping) {
+	if (w == 0 || h == 0 || numMips == 0) {
+		ERROR_LOG(G3D, "Can't create a zero-size VulkanTexture");
+		return false;
+	}
+
 	Wipe();
 
 	width_ = w;
@@ -196,7 +201,8 @@ void VulkanTexture::ClearMip(VkCommandBuffer cmd, int mip, uint32_t value) {
 	vkCmdClearColorImage(cmd, image_, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, &clearVal, 1, &range);
 }
 
-void VulkanTexture::GenerateMip(VkCommandBuffer cmd, int mip) {
+// Low-quality mipmap generation by bilinear blit, but works okay.
+void VulkanTexture::GenerateMip(VkCommandBuffer cmd, int mip, VkImageLayout imageLayout) {
 	_assert_msg_(mip != 0, "Cannot generate the first level");
 	_assert_msg_(mip < numMips_, "Cannot generate mipmaps past the maximum created (%d vs %d)", mip, numMips_);
 	VkImageBlit blit{};
@@ -214,16 +220,20 @@ void VulkanTexture::GenerateMip(VkCommandBuffer cmd, int mip) {
 	blit.dstOffsets[1].y = height_ >> mip;
 	blit.dstOffsets[1].z = 1;
 
+	// TODO: We could do better with the image transitions - would be enough with one per level
+	// for the memory barrier, then one final one for the whole stack when done. This function
+	// currently doesn't have a global enough view, though.
+	// We should also coalesce barriers across multiple texture uploads in a frame and all kinds of other stuff, but...
+
 	TransitionImageLayout2(cmd, image_, mip - 1, 1, VK_IMAGE_ASPECT_COLOR_BIT,
-		VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+		imageLayout, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
 		VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
 		VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT);
 
-	// Low-quality mipmap generation, but works okay.
-	vkCmdBlitImage(cmd, image_, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, image_, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &blit, VK_FILTER_LINEAR);
+	vkCmdBlitImage(cmd, image_, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, image_, imageLayout, 1, &blit, VK_FILTER_LINEAR);
 
 	TransitionImageLayout2(cmd, image_, mip - 1, 1, VK_IMAGE_ASPECT_COLOR_BIT,
-		VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+		VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, imageLayout,
 		VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
 		VK_ACCESS_TRANSFER_READ_BIT, VK_ACCESS_TRANSFER_WRITE_BIT);
 }

Common/GPU/Vulkan/VulkanImage.h

@@ -21,7 +21,7 @@ class VulkanTexture {
 	bool CreateDirect(VkCommandBuffer cmd, VulkanDeviceAllocator *allocator, int w, int h, int numMips, VkFormat format, VkImageLayout initialLayout, VkImageUsageFlags usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT, const VkComponentMapping *mapping = nullptr);
 	void ClearMip(VkCommandBuffer cmd, int mip, uint32_t value);
 	void UploadMip(VkCommandBuffer cmd, int mip, int mipWidth, int mipHeight, VkBuffer buffer, uint32_t offset, size_t rowLength);  // rowLength is in pixels
-	void GenerateMip(VkCommandBuffer cmd, int mip);
+	void GenerateMip(VkCommandBuffer cmd, int mip, VkImageLayout imageLayout);
 	void EndCreate(VkCommandBuffer cmd, bool vertexTexture = false, VkImageLayout layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
 
 	// When loading mips from compute shaders, you need to pass VK_IMAGE_LAYOUT_GENERAL to the above function.

Common/GPU/Vulkan/thin3d_vulkan.cpp

@@ -215,7 +215,7 @@ bool VKShaderModule::Compile(VulkanContext *vulkan, ShaderLanguage language, con
 	std::vector<uint32_t> spirv;
 	std::string errorMessage;
 	if (!GLSLtoSPV(vkstage_, source_.c_str(), GLSLVariant::VULKAN, spirv, &errorMessage)) {
-		INFO_LOG(G3D, "Shader compile to module failed: %s", errorMessage.c_str());
+		WARN_LOG(G3D, "Shader compile to module failed: %s", errorMessage.c_str());
 		return false;
 	}
 
@@ -231,6 +231,7 @@ bool VKShaderModule::Compile(VulkanContext *vulkan, ShaderLanguage language, con
 	if (vulkan->CreateShaderModule(spirv, &module_)) {
 		ok_ = true;
 	} else {
+		WARN_LOG(G3D, "vkCreateShaderModule failed");
 		ok_ = false;
 	}
 	return ok_;
@@ -754,7 +755,7 @@ bool VKTexture::Create(VkCommandBuffer cmd, VulkanPushBuffer *push, const Textur
 		}
 		// Generate the rest of the mips automatically.
 		for (; i < mipLevels_; i++) {
-			vkTex_->GenerateMip(cmd, i);
+			vkTex_->GenerateMip(cmd, i, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
 		}
 	}
 	vkTex_->EndCreate(cmd, false);
@@ -1281,7 +1282,7 @@ ShaderModule *VKContext::CreateShaderModule(ShaderStage stage, ShaderLanguage la
 	if (shader->Compile(vulkan_, language, data, size)) {
 		return shader;
 	} else {
-		ERROR_LOG(G3D,  "Failed to compile shader: %s", (const char *)data);
+		ERROR_LOG(G3D,  "Failed to compile shader:\n%s", (const char *)data);
 		shader->Release();
 		return nullptr;
 	}

Common/MemArenaPosix.cpp

@@ -107,7 +107,7 @@ void MemArena::ReleaseView(void* view, size_t size) {
 
 u8* MemArena::Find4GBBase() {
 	// Now, create views in high memory where there's plenty of space.
-#if PPSSPP_ARCH(64BIT) && !defined(USE_ADDRESS_SANITIZER)
+#if PPSSPP_ARCH(64BIT) && !defined(USE_ASAN)
 	// We should probably just go look in /proc/self/maps for some free space.
 	// But let's try the anonymous mmap trick, just like on 32-bit, but bigger and
 	// aligned to 4GB for the movk trick. We can ensure that we get an aligned 4GB

Common/Render/DrawBuffer.cpp

@@ -224,6 +224,20 @@ void DrawBuffer::DrawImage(ImageID atlas_image, float x, float y, float scale, C
 	DrawImageStretch(atlas_image, x, y, x + w, y + h, color);
 }
 
+void DrawBuffer::DrawImageCenterTexel(ImageID atlas_image, float x1, float y1, float x2, float y2, Color color) {
+	const AtlasImage *image = atlas->getImage(atlas_image);
+	if (!image)
+		return;
+	float centerU = (image->u1 + image->u2) * 0.5f;
+	float centerV = (image->v1 + image->v2) * 0.5f;
+	V(x1,	y1, color, centerU, centerV);
+	V(x2,	y1, color, centerU, centerV);
+	V(x2,	y2, color, centerU, centerV);
+	V(x1,	y1, color, centerU, centerV);
+	V(x2,	y2, color, centerU, centerV);
+	V(x1,	y2, color, centerU, centerV);
+}
+
 void DrawBuffer::DrawImageStretch(ImageID atlas_image, float x1, float y1, float x2, float y2, Color color) {
 	const AtlasImage *image = atlas->getImage(atlas_image);
 	if (!image)
@@ -236,6 +250,18 @@ void DrawBuffer::DrawImageStretch(ImageID atlas_image, float x1, float y1, float
 	V(x1,	y2, color, image->u1, image->v2);
 }
 
+void DrawBuffer::DrawImageStretchVGradient(ImageID atlas_image, float x1, float y1, float x2, float y2, Color color1, Color color2) {
+	const AtlasImage *image = atlas->getImage(atlas_image);
+	if (!image)
+		return;
+	V(x1, y1, color1, image->u1, image->v1);
+	V(x2, y1, color1, image->u2, image->v1);
+	V(x2, y2, color2, image->u2, image->v2);
+	V(x1, y1, color1, image->u1, image->v1);
+	V(x2, y2, color2, image->u2, image->v2);
+	V(x1, y2, color2, image->u1, image->v2);
+}
+
 inline void rot(float *v, float angle, float xc, float yc) {
 	const float x = v[0] - xc;
 	const float y = v[1] - yc;

Common/Render/DrawBuffer.h

@@ -107,7 +107,11 @@ class DrawBuffer {
 	const Atlas *GetAtlas() const { return atlas; }
 	bool MeasureImage(ImageID atlas_image, float *w, float *h);
 	void DrawImage(ImageID atlas_image, float x, float y, float scale, Color color = COLOR(0xFFFFFF), int align = ALIGN_TOPLEFT);
+
+	// Good for stretching out a white image without edge artifacts that I'm getting on iOS.
+	void DrawImageCenterTexel(ImageID atlas_image, float x1, float y1, float x2, float y2, Color color = COLOR(0xFFFFFF));
 	void DrawImageStretch(ImageID atlas_image, float x1, float y1, float x2, float y2, Color color = COLOR(0xFFFFFF));
+	void DrawImageStretchVGradient(ImageID atlas_image, float x1, float y1, float x2, float y2, Color color1, Color color2);
 	void DrawImageStretch(ImageID atlas_image, const Bounds &bounds, Color color = COLOR(0xFFFFFF)) {
 		DrawImageStretch(atlas_image, bounds.x, bounds.y, bounds.x2(), bounds.y2(), color);
 	}

Common/Serialize/Serializer.cpp

@@ -174,7 +174,7 @@ void PointerWrap::DoMarker(const char *prevName, u32 arbitraryNumber) {
 	u32 cookie = arbitraryNumber;
 	Do(*this, cookie);
 	if (mode == PointerWrap::MODE_READ && cookie != arbitraryNumber) {
-		_assert_msg_(false, "Error: After \"%s\", found %d (0x%X) instead of save marker %d (0x%X). Aborting savestate load...", prevName, cookie, cookie, arbitraryNumber, arbitraryNumber);
+		ERROR_LOG(SAVESTATE, "Error: After \"%s\", found %d (0x%X) instead of save marker %d (0x%X). Aborting savestate load...", prevName, cookie, cookie, arbitraryNumber, arbitraryNumber);
 		SetError(ERROR_FAILURE);
 	}
 }