early-access version 1287

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1286.
+This is the source code for early-access 1287.
 
 ## Legal Notice
 

src/common/logging/backend.cpp

@@ -145,10 +145,18 @@ void ColorConsoleBackend::Write(const Entry& entry) {
     PrintColoredMessage(entry);
 }
 
-// _SH_DENYWR allows read only access to the file for other programs.
+FileBackend::FileBackend(const std::string& filename) : bytes_written(0) {
-// It is #defined to 0 on other platforms
+    if (Common::FS::Exists(filename + ".old.txt")) {
-FileBackend::FileBackend(const std::string& filename)
+        Common::FS::Delete(filename + ".old.txt");
-    : file(filename, "w", _SH_DENYWR), bytes_written(0) {}
+    }
+    if (Common::FS::Exists(filename)) {
+        Common::FS::Rename(filename, filename + ".old.txt");
+    }
+
+    // _SH_DENYWR allows read only access to the file for other programs.
+    // It is #defined to 0 on other platforms
+    file = Common::FS::IOFile(filename, "w", _SH_DENYWR);
+}
 
 void FileBackend::Write(const Entry& entry) {
     // prevent logs from going over the maximum size (in case its spamming and the user doesn't

src/core/hle/service/nvflinger/buffer_queue.cpp

@@ -26,10 +26,10 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
     LOG_WARNING(Service, "Adding graphics buffer {}", slot);
 
     {
-        std::unique_lock lock{queue_mutex};
+        std::unique_lock lock{free_buffers_mutex};
         free_buffers.push_back(slot);
     }
-    condition.notify_one();
+    free_buffers_condition.notify_one();
 
     buffers[slot] = {
         .slot = slot,
@@ -48,8 +48,8 @@ std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::Dequeue
                                                                                        u32 height) {
     // Wait for first request before trying to dequeue
     {
-        std::unique_lock lock{queue_mutex};
+        std::unique_lock lock{free_buffers_mutex};
-        condition.wait(lock, [this] { return !free_buffers.empty() || !is_connect; });
+        free_buffers_condition.wait(lock, [this] { return !free_buffers.empty() || !is_connect; });
     }
 
     if (!is_connect) {
@@ -58,7 +58,7 @@ std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::Dequeue
         return std::nullopt;
     }
 
-    std::unique_lock lock{queue_mutex};
+    std::unique_lock lock{free_buffers_mutex};
 
     auto f_itr = free_buffers.begin();
     auto slot = buffers.size();
@@ -100,6 +100,7 @@ void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
     buffers[slot].crop_rect = crop_rect;
     buffers[slot].swap_interval = swap_interval;
     buffers[slot].multi_fence = multi_fence;
+    std::unique_lock lock{queue_sequence_mutex};
     queue_sequence.push_back(slot);
 }
 
@@ -113,15 +114,16 @@ void BufferQueue::CancelBuffer(u32 slot, const Service::Nvidia::MultiFence& mult
     buffers[slot].swap_interval = 0;
 
     {
-        std::unique_lock lock{queue_mutex};
+        std::unique_lock lock{free_buffers_mutex};
         free_buffers.push_back(slot);
     }
-    condition.notify_one();
+    free_buffers_condition.notify_one();
 
     buffer_wait_event.writable->Signal();
 }
 
 std::optional<std::reference_wrapper<const BufferQueue::Buffer>> BufferQueue::AcquireBuffer() {
+    std::unique_lock lock{queue_sequence_mutex};
     std::size_t buffer_slot = buffers.size();
     // Iterate to find a queued buffer matching the requested slot.
     while (buffer_slot == buffers.size() && !queue_sequence.empty()) {
@@ -147,25 +149,29 @@ void BufferQueue::ReleaseBuffer(u32 slot) {
 
     buffers[slot].status = Buffer::Status::Free;
     {
-        std::unique_lock lock{queue_mutex};
+        std::unique_lock lock{free_buffers_mutex};
         free_buffers.push_back(slot);
     }
-    condition.notify_one();
+    free_buffers_condition.notify_one();
 
     buffer_wait_event.writable->Signal();
 }
 
 void BufferQueue::Connect() {
+    std::unique_lock lock{queue_sequence_mutex};
     queue_sequence.clear();
     is_connect = true;
 }
 
 void BufferQueue::Disconnect() {
     buffers.fill({});
+    {
+        std::unique_lock lock{queue_sequence_mutex};
         queue_sequence.clear();
+    }
     buffer_wait_event.writable->Signal();
     is_connect = false;
-    condition.notify_one();
+    free_buffers_condition.notify_one();
 }
 
 u32 BufferQueue::Query(QueryType type) {

src/core/hle/service/nvflinger/buffer_queue.h

@@ -129,8 +129,10 @@ private:
     std::list<u32> queue_sequence;
     Kernel::EventPair buffer_wait_event;
 
-    std::mutex queue_mutex;
+    std::mutex free_buffers_mutex;
-    std::condition_variable condition;
+    std::condition_variable free_buffers_condition;
+
+    std::mutex queue_sequence_mutex;
 };
 
 } // namespace Service::NVFlinger

src/video_core/compatible_formats.cpp

@@ -10,9 +10,7 @@
 #include "video_core/surface.h"
 
 namespace VideoCore::Surface {
-
 namespace {
-
 using Table = std::array<std::array<u64, 2>, MaxPixelFormat>;
 
 // Compatibility table taken from Table 3.X.2 in:
@@ -233,10 +231,13 @@ constexpr Table MakeCopyTable() {
     EnableRange(copy, COPY_CLASS_64_BITS);
     return copy;
 }
-
 } // Anonymous namespace
 
-bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b) {
+bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b, bool broken_views) {
+    if (broken_views) {
+        // If format views are broken, only accept formats that are identical.
+        return format_a == format_b;
+    }
     static constexpr Table TABLE = MakeViewTable();
     return IsSupported(TABLE, format_a, format_b);
 }

src/video_core/compatible_formats.h

@@ -8,7 +8,7 @@
 
 namespace VideoCore::Surface {
 
-bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b);
+bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b, bool broken_views);
 
 bool IsCopyCompatible(PixelFormat format_a, PixelFormat format_b);
 

src/video_core/renderer_opengl/gl_device.cpp

@@ -208,6 +208,7 @@ Device::Device()
 
     const bool is_nvidia = vendor == "NVIDIA Corporation";
     const bool is_amd = vendor == "ATI Technologies Inc.";
+    const bool is_intel = vendor == "Intel";
 
     bool disable_fast_buffer_sub_data = false;
     if (is_nvidia && version == "4.6.0 NVIDIA 443.24") {
@@ -231,6 +232,7 @@ Device::Device()
     has_variable_aoffi = TestVariableAoffi();
     has_component_indexing_bug = is_amd;
     has_precise_bug = TestPreciseBug();
+    has_broken_texture_view_formats = is_amd || is_intel;
     has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2;
     has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory;
     has_debugging_tool_attached = IsDebugToolAttached(extensions);
@@ -248,6 +250,8 @@ Device::Device()
     LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi);
     LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug);
     LOG_INFO(Render_OpenGL, "Renderer_PreciseBug: {}", has_precise_bug);
+    LOG_INFO(Render_OpenGL, "Renderer_BrokenTextureViewFormats: {}",
+             has_broken_texture_view_formats);
 
     if (Settings::values.use_assembly_shaders.GetValue() && !use_assembly_shaders) {
         LOG_ERROR(Render_OpenGL, "Assembly shaders enabled but not supported");

src/video_core/renderer_opengl/gl_device.h

@@ -96,6 +96,10 @@ public:
         return has_precise_bug;
     }
 
+    bool HasBrokenTextureViewFormats() const {
+        return has_broken_texture_view_formats;
+    }
+
     bool HasFastBufferSubData() const {
         return has_fast_buffer_sub_data;
     }
@@ -137,6 +141,7 @@ private:
     bool has_variable_aoffi{};
     bool has_component_indexing_bug{};
     bool has_precise_bug{};
+    bool has_broken_texture_view_formats{};
     bool has_fast_buffer_sub_data{};
     bool has_nv_viewport_array2{};
     bool has_debugging_tool_attached{};

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -61,7 +61,6 @@ struct FormatTuple {
     GLenum internal_format;
     GLenum format = GL_NONE;
     GLenum type = GL_NONE;
-    GLenum store_format = internal_format;
 };
 
 constexpr std::array<FormatTuple, MaxPixelFormat> FORMAT_TABLE = {{
@@ -114,7 +113,7 @@ constexpr std::array<FormatTuple, MaxPixelFormat> FORMAT_TABLE = {{
     {GL_RG16I, GL_RG_INTEGER, GL_SHORT},                              // R16G16_SINT
     {GL_RG16_SNORM, GL_RG, GL_SHORT},                                 // R16G16_SNORM
     {GL_RGB32F, GL_RGB, GL_FLOAT},                                    // R32G32B32_FLOAT
-    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, GL_RGBA8}, // A8B8G8R8_SRGB
+    {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV},          // A8B8G8R8_SRGB
     {GL_RG8, GL_RG, GL_UNSIGNED_BYTE},                                // R8G8_UNORM
     {GL_RG8_SNORM, GL_RG, GL_BYTE},                                   // R8G8_SNORM
     {GL_RG8I, GL_RG_INTEGER, GL_BYTE},                                // R8G8_SINT
@@ -126,7 +125,7 @@ constexpr std::array<FormatTuple, MaxPixelFormat> FORMAT_TABLE = {{
     {GL_COMPRESSED_RGBA_ASTC_8x8_KHR},                                // ASTC_2D_8X8_UNORM
     {GL_COMPRESSED_RGBA_ASTC_8x5_KHR},                                // ASTC_2D_8X5_UNORM
     {GL_COMPRESSED_RGBA_ASTC_5x4_KHR},                                // ASTC_2D_5X4_UNORM
-    {GL_SRGB8_ALPHA8, GL_BGRA, GL_UNSIGNED_BYTE, GL_RGBA8},            // B8G8R8A8_UNORM
+    {GL_SRGB8_ALPHA8, GL_BGRA, GL_UNSIGNED_BYTE},                     // B8G8R8A8_UNORM
     {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT},                         // BC1_RGBA_SRGB
     {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT},                         // BC2_SRGB
     {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT},                         // BC3_SRGB
@@ -431,6 +430,8 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager&
             format_properties[i].emplace(format, properties);
         }
     }
+    has_broken_texture_view_formats = device.HasBrokenTextureViewFormats();
+
     null_image_1d_array.Create(GL_TEXTURE_1D_ARRAY);
     null_image_cube_array.Create(GL_TEXTURE_CUBE_MAP_ARRAY);
     null_image_3d.Create(GL_TEXTURE_3D);
@@ -651,13 +652,11 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_,
     if (IsConverted(runtime.device, info.format, info.type)) {
         flags |= ImageFlagBits::Converted;
         gl_internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8;
-        gl_store_format = GL_RGBA8;
         gl_format = GL_RGBA;
         gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
     } else {
         const auto& tuple = GetFormatTuple(info.format);
         gl_internal_format = tuple.internal_format;
-        gl_store_format = tuple.store_format;
         gl_format = tuple.format;
         gl_type = tuple.type;
     }
@@ -677,23 +676,23 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_,
     }
     switch (target) {
     case GL_TEXTURE_1D_ARRAY:
-        glTextureStorage2D(handle, num_levels, gl_store_format, width, num_layers);
+        glTextureStorage2D(handle, num_levels, gl_internal_format, width, num_layers);
         break;
     case GL_TEXTURE_2D_ARRAY:
-        glTextureStorage3D(handle, num_levels, gl_store_format, width, height, num_layers);
+        glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, num_layers);
         break;
     case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: {
         // TODO: Where should 'fixedsamplelocations' come from?
         const auto [samples_x, samples_y] = SamplesLog2(info.num_samples);
-        glTextureStorage3DMultisample(handle, num_samples, gl_store_format, width >> samples_x,
+        glTextureStorage3DMultisample(handle, num_samples, gl_internal_format, width >> samples_x,
                                       height >> samples_y, num_layers, GL_FALSE);
         break;
     }
     case GL_TEXTURE_RECTANGLE:
-        glTextureStorage2D(handle, num_levels, gl_store_format, width, height);
+        glTextureStorage2D(handle, num_levels, gl_internal_format, width, height);
         break;
     case GL_TEXTURE_3D:
-        glTextureStorage3D(handle, num_levels, gl_store_format, width, height, depth);
+        glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth);
         break;
     case GL_TEXTURE_BUFFER:
         buffer.Create();

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -96,6 +96,10 @@ public:
 
     FormatProperties FormatInfo(VideoCommon::ImageType type, GLenum internal_format) const;
 
+    bool HasBrokenTextureViewFormats() const noexcept {
+        return has_broken_texture_view_formats;
+    }
+
 private:
     struct StagingBuffers {
         explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_);
@@ -120,6 +124,7 @@ private:
     UtilShaders util_shaders;
 
     std::array<std::unordered_map<GLenum, FormatProperties>, 3> format_properties;
+    bool has_broken_texture_view_formats = false;
 
     StagingBuffers upload_buffers{GL_MAP_WRITE_BIT, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT};
     StagingBuffers download_buffers{GL_MAP_READ_BIT, GL_MAP_READ_BIT};
@@ -165,7 +170,6 @@ private:
     OGLTextureView store_view;
     OGLBuffer buffer;
     GLenum gl_internal_format = GL_NONE;
-    GLenum gl_store_format = GL_NONE;
     GLenum gl_format = GL_NONE;
     GLenum gl_type = GL_NONE;
 };

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -104,6 +104,11 @@ struct TextureCacheRuntime {
     }
 
     void InsertUploadMemoryBarrier() {}
+
+    bool HasBrokenTextureViewFormats() const noexcept {
+        // No known Vulkan driver has broken image views
+        return false;
+    }
 };
 
 class Image : public VideoCommon::ImageBase {

src/video_core/texture_cache/image_base.cpp

@@ -120,7 +120,9 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i
     if (lhs.info.type == ImageType::Linear) {
         base = SubresourceBase{.level = 0, .layer = 0};
     } else {
-        base = FindSubresource(rhs.info, lhs, rhs.gpu_addr, OPTIONS);
+        // We are passing relaxed formats as an option, having broken views or not won't matter
+        static constexpr bool broken_views = false;
+        base = FindSubresource(rhs.info, lhs, rhs.gpu_addr, OPTIONS, broken_views);
     }
     if (!base) {
         LOG_ERROR(HW_GPU, "Image alias should have been flipped");

src/video_core/texture_cache/image_view_base.cpp

@@ -24,7 +24,7 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i
           .height = std::max(image_info.size.height >> range.base.level, 1u),
           .depth = std::max(image_info.size.depth >> range.base.level, 1u),
       } {
-    ASSERT_MSG(VideoCore::Surface::IsViewCompatible(image_info.format, info.format),
+    ASSERT_MSG(VideoCore::Surface::IsViewCompatible(image_info.format, info.format, false),
                "Image view format {} is incompatible with image format {}", info.format,
                image_info.format);
     const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue();

src/video_core/texture_cache/texture_cache.h

@@ -883,6 +883,7 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
     if (!cpu_addr) {
         return ImageId{};
     }
+    const bool broken_views = runtime.HasBrokenTextureViewFormats();
     ImageId image_id;
     const auto lambda = [&](ImageId existing_image_id, ImageBase& existing_image) {
         if (info.type == ImageType::Linear || existing_image.info.type == ImageType::Linear) {
@@ -892,11 +893,11 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
             if (existing_image.gpu_addr == gpu_addr && existing.type == info.type &&
                 existing.pitch == info.pitch &&
                 IsPitchLinearSameSize(existing, info, strict_size) &&
-                IsViewCompatible(existing.format, info.format)) {
+                IsViewCompatible(existing.format, info.format, broken_views)) {
                 image_id = existing_image_id;
                 return true;
             }
-        } else if (IsSubresource(info, existing_image, gpu_addr, options)) {
+        } else if (IsSubresource(info, existing_image, gpu_addr, options, broken_views)) {
             image_id = existing_image_id;
             return true;
         }
@@ -926,6 +927,7 @@ template <class P>
 ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr) {
     ImageInfo new_info = info;
     const size_t size_bytes = CalculateGuestSizeInBytes(new_info);
+    const bool broken_views = runtime.HasBrokenTextureViewFormats();
     std::vector<ImageId> overlap_ids;
     std::vector<ImageId> left_aliased_ids;
     std::vector<ImageId> right_aliased_ids;
@@ -940,7 +942,9 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             }
             return;
         }
-        const auto solution = ResolveOverlap(new_info, gpu_addr, cpu_addr, overlap, true);
+        static constexpr bool strict_size = true;
+        const std::optional<OverlapResult> solution =
+            ResolveOverlap(new_info, gpu_addr, cpu_addr, overlap, strict_size, broken_views);
         if (solution) {
             gpu_addr = solution->gpu_addr;
             cpu_addr = solution->cpu_addr;
@@ -950,9 +954,10 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         }
         static constexpr auto options = RelaxedOptions::Size | RelaxedOptions::Format;
         const ImageBase new_image_base(new_info, gpu_addr, cpu_addr);
-        if (IsSubresource(new_info, overlap, gpu_addr, options)) {
+        if (IsSubresource(new_info, overlap, gpu_addr, options, broken_views)) {
             left_aliased_ids.push_back(overlap_id);
-        } else if (IsSubresource(overlap.info, new_image_base, overlap.gpu_addr, options)) {
+        } else if (IsSubresource(overlap.info, new_image_base, overlap.gpu_addr, options,
+                                 broken_views)) {
             right_aliased_ids.push_back(overlap_id);
         }
     });

src/video_core/texture_cache/util.cpp

@@ -1069,13 +1069,13 @@ bool IsPitchLinearSameSize(const ImageInfo& lhs, const ImageInfo& rhs, bool stri
 
 std::optional<OverlapResult> ResolveOverlap(const ImageInfo& new_info, GPUVAddr gpu_addr,
                                             VAddr cpu_addr, const ImageBase& overlap,
-                                            bool strict_size) {
+                                            bool strict_size, bool broken_views) {
     ASSERT(new_info.type != ImageType::Linear);
     ASSERT(overlap.info.type != ImageType::Linear);
     if (!IsLayerStrideCompatible(new_info, overlap.info)) {
         return std::nullopt;
     }
-    if (!IsViewCompatible(overlap.info.format, new_info.format)) {
+    if (!IsViewCompatible(overlap.info.format, new_info.format, broken_views)) {
         return std::nullopt;
     }
     if (gpu_addr == overlap.gpu_addr) {
@@ -1118,14 +1118,15 @@ bool IsLayerStrideCompatible(const ImageInfo& lhs, const ImageInfo& rhs) {
 }
 
 std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate, const ImageBase& image,
-                                               GPUVAddr candidate_addr, RelaxedOptions options) {
+                                               GPUVAddr candidate_addr, RelaxedOptions options,
+                                               bool broken_views) {
     const std::optional<SubresourceBase> base = image.TryFindBase(candidate_addr);
     if (!base) {
         return std::nullopt;
     }
     const ImageInfo& existing = image.info;
     if (False(options & RelaxedOptions::Format)) {
-        if (!IsViewCompatible(existing.format, candidate.format)) {
+        if (!IsViewCompatible(existing.format, candidate.format, broken_views)) {
             return std::nullopt;
         }
     }
@@ -1162,8 +1163,8 @@ std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate, const
 }
 
 bool IsSubresource(const ImageInfo& candidate, const ImageBase& image, GPUVAddr candidate_addr,
-                   RelaxedOptions options) {
+                   RelaxedOptions options, bool broken_views) {
-    return FindSubresource(candidate, image, candidate_addr, options).has_value();
+    return FindSubresource(candidate, image, candidate_addr, options, broken_views).has_value();
 }
 
 void DeduceBlitImages(ImageInfo& dst_info, ImageInfo& src_info, const ImageBase* dst,

src/video_core/texture_cache/util.h

@@ -87,17 +87,19 @@ void SwizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, const Ima
 [[nodiscard]] std::optional<OverlapResult> ResolveOverlap(const ImageInfo& new_info,
                                                           GPUVAddr gpu_addr, VAddr cpu_addr,
                                                           const ImageBase& overlap,
-                                                          bool strict_size);
+                                                          bool strict_size, bool broken_views);
 
 [[nodiscard]] bool IsLayerStrideCompatible(const ImageInfo& lhs, const ImageInfo& rhs);
 
 [[nodiscard]] std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate,
                                                              const ImageBase& image,
                                                              GPUVAddr candidate_addr,
-                                                             RelaxedOptions options);
+                                                             RelaxedOptions options,
+                                                             bool broken_views);
 
 [[nodiscard]] bool IsSubresource(const ImageInfo& candidate, const ImageBase& image,
-                                 GPUVAddr candidate_addr, RelaxedOptions options);
+                                 GPUVAddr candidate_addr, RelaxedOptions options,
+                                 bool broken_views);
 
 void DeduceBlitImages(ImageInfo& dst_info, ImageInfo& src_info, const ImageBase* dst,
                       const ImageBase* src);