diff options
Diffstat (limited to 'src/video_core')
23 files changed, 238 insertions, 60 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 58a45ab67..b5ed3380f 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -115,7 +115,34 @@ void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) { template <class P> void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) { - memory_tracker.CachedCpuWrite(cpu_addr, size); + const bool is_dirty = IsRegionRegistered(cpu_addr, size); + if (!is_dirty) { + return; + } + VAddr aligned_start = Common::AlignDown(cpu_addr, YUZU_PAGESIZE); + VAddr aligned_end = Common::AlignUp(cpu_addr + size, YUZU_PAGESIZE); + if (!IsRegionGpuModified(aligned_start, aligned_end - aligned_start)) { + WriteMemory(cpu_addr, size); + return; + } + + tmp_buffer.resize_destructive(size); + cpu_memory.ReadBlockUnsafe(cpu_addr, tmp_buffer.data(), size); + + InlineMemoryImplementation(cpu_addr, size, tmp_buffer); +} + +template <class P> +bool BufferCache<P>::OnCPUWrite(VAddr cpu_addr, u64 size) { + const bool is_dirty = IsRegionRegistered(cpu_addr, size); + if (!is_dirty) { + return false; + } + if (memory_tracker.IsRegionGpuModified(cpu_addr, size)) { + return true; + } + WriteMemory(cpu_addr, size); + return false; } template <class P> @@ -1553,6 +1580,14 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size, return false; } + InlineMemoryImplementation(dest_address, copy_size, inlined_buffer); + + return true; +} + +template <class P> +void BufferCache<P>::InlineMemoryImplementation(VAddr dest_address, size_t copy_size, + std::span<const u8> inlined_buffer) { const IntervalType subtract_interval{dest_address, dest_address + copy_size}; ClearDownload(subtract_interval); common_ranges.subtract(subtract_interval); @@ -1574,8 +1609,6 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size, } else { buffer.ImmediateUpload(buffer.Offset(dest_address), inlined_buffer.first(copy_size)); } - - return true; } template <class P> diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index fe6068cfe..460fc7551 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -245,6 +245,8 @@ public: void CachedWriteMemory(VAddr cpu_addr, u64 size); + bool OnCPUWrite(VAddr cpu_addr, u64 size); + void DownloadMemory(VAddr cpu_addr, u64 size); std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size); @@ -543,6 +545,9 @@ private: void ClearDownload(IntervalType subtract_interval); + void InlineMemoryImplementation(VAddr dest_address, size_t copy_size, + std::span<const u8> inlined_buffer); + VideoCore::RasterizerInterface& rasterizer; Core::Memory::Memory& cpu_memory; diff --git a/src/video_core/compatible_formats.cpp b/src/video_core/compatible_formats.cpp index ab4f4d407..87d69ebc5 100644 --- a/src/video_core/compatible_formats.cpp +++ b/src/video_core/compatible_formats.cpp @@ -272,6 +272,9 @@ constexpr Table MakeNonNativeBgrCopyTable() { bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b, bool broken_views, bool native_bgr) { + if (format_a == format_b) { + return true; + } if (broken_views) { // If format views are broken, only accept formats that are identical. return format_a == format_b; @@ -282,6 +285,9 @@ bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b, bool broken_vi } bool IsCopyCompatible(PixelFormat format_a, PixelFormat format_b, bool native_bgr) { + if (format_a == format_b) { + return true; + } static constexpr Table BGR_TABLE = MakeNativeBgrCopyTable(); static constexpr Table NO_BGR_TABLE = MakeNonNativeBgrCopyTable(); return IsSupported(native_bgr ? BGR_TABLE : NO_BGR_TABLE, format_a, format_b); diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index 35d699bbf..ab20ff30f 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -69,7 +69,6 @@ public: } void SignalFence(std::function<void()>&& func) { - rasterizer.InvalidateGPUCache(); bool delay_fence = Settings::IsGPULevelHigh(); if constexpr (!can_async_check) { TryReleasePendingFences<false>(); @@ -96,6 +95,7 @@ public: guard.unlock(); cv.notify_all(); } + rasterizer.InvalidateGPUCache(); } void SignalSyncPoint(u32 value) { diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index db385076d..c192e33b2 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -95,7 +95,9 @@ struct GPU::Impl { /// Synchronizes CPU writes with Host GPU memory. void InvalidateGPUCache() { - rasterizer->InvalidateGPUCache(); + std::function<void(VAddr, size_t)> callback_writes( + [this](VAddr address, size_t size) { rasterizer->OnCacheInvalidation(address, size); }); + system.GatherGPUDirtyMemory(callback_writes); } /// Signal the ending of command list. @@ -299,6 +301,10 @@ struct GPU::Impl { gpu_thread.InvalidateRegion(addr, size); } + bool OnCPUWrite(VAddr addr, u64 size) { + return rasterizer->OnCPUWrite(addr, size); + } + /// Notify rasterizer that any caches of the specified region should be flushed and invalidated void FlushAndInvalidateRegion(VAddr addr, u64 size) { gpu_thread.FlushAndInvalidateRegion(addr, size); @@ -561,6 +567,10 @@ void GPU::InvalidateRegion(VAddr addr, u64 size) { impl->InvalidateRegion(addr, size); } +bool GPU::OnCPUWrite(VAddr addr, u64 size) { + return impl->OnCPUWrite(addr, size); +} + void GPU::FlushAndInvalidateRegion(VAddr addr, u64 size) { impl->FlushAndInvalidateRegion(addr, size); } diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index e49c40cf2..ba2838b89 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -250,6 +250,10 @@ public: /// Notify rasterizer that any caches of the specified region should be invalidated void InvalidateRegion(VAddr addr, u64 size); + /// Notify rasterizer that CPU is trying to write this area. It returns true if the area is + /// sensible, false otherwise + bool OnCPUWrite(VAddr addr, u64 size); + /// Notify rasterizer that any caches of the specified region should be flushed and invalidated void FlushAndInvalidateRegion(VAddr addr, u64 size); diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index 889144f38..2f0f9f593 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -47,7 +47,7 @@ static void RunThread(std::stop_token stop_token, Core::System& system, } else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) { rasterizer->FlushRegion(flush->addr, flush->size); } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) { - rasterizer->OnCPUWrite(invalidate->addr, invalidate->size); + rasterizer->OnCacheInvalidation(invalidate->addr, invalidate->size); } else { ASSERT(false); } @@ -102,12 +102,12 @@ void ThreadManager::TickGPU() { } void ThreadManager::InvalidateRegion(VAddr addr, u64 size) { - rasterizer->OnCPUWrite(addr, size); + rasterizer->OnCacheInvalidation(addr, size); } void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) { // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important - rasterizer->OnCPUWrite(addr, size); + rasterizer->OnCacheInvalidation(addr, size); } u64 ThreadManager::PushCommand(CommandData&& command_data, bool block) { diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 7566a8c4e..cb8029a4f 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -109,7 +109,9 @@ public: } /// Notify rasterizer that any caches of the specified region are desync with guest - virtual void OnCPUWrite(VAddr addr, u64 size) = 0; + virtual void OnCacheInvalidation(VAddr addr, u64 size) = 0; + + virtual bool OnCPUWrite(VAddr addr, u64 size) = 0; /// Sync memory between guest and host. virtual void InvalidateGPUCache() = 0; diff --git a/src/video_core/renderer_null/null_rasterizer.cpp b/src/video_core/renderer_null/null_rasterizer.cpp index bf2ce4c49..92ecf6682 100644 --- a/src/video_core/renderer_null/null_rasterizer.cpp +++ b/src/video_core/renderer_null/null_rasterizer.cpp @@ -47,7 +47,10 @@ bool RasterizerNull::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheTyp return false; } void RasterizerNull::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType) {} -void RasterizerNull::OnCPUWrite(VAddr addr, u64 size) {} +bool RasterizerNull::OnCPUWrite(VAddr addr, u64 size) { + return false; +} +void RasterizerNull::OnCacheInvalidation(VAddr addr, u64 size) {} VideoCore::RasterizerDownloadArea RasterizerNull::GetFlushArea(VAddr addr, u64 size) { VideoCore::RasterizerDownloadArea new_area{ .start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE), diff --git a/src/video_core/renderer_null/null_rasterizer.h b/src/video_core/renderer_null/null_rasterizer.h index a8d35d2c1..93b9a6971 100644 --- a/src/video_core/renderer_null/null_rasterizer.h +++ b/src/video_core/renderer_null/null_rasterizer.h @@ -53,7 +53,8 @@ public: VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; - void OnCPUWrite(VAddr addr, u64 size) override; + void OnCacheInvalidation(VAddr addr, u64 size) override; + bool OnCPUWrite(VAddr addr, u64 size) override; VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index edf527f2d..aadd6967c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -485,12 +485,33 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache } } -void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { +bool RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + if (addr == 0 || size == 0) { + return false; + } + + { + std::scoped_lock lock{buffer_cache.mutex}; + if (buffer_cache.OnCPUWrite(addr, size)) { + return true; + } + } + + { + std::scoped_lock lock{texture_cache.mutex}; + texture_cache.WriteMemory(addr, size); + } + + shader_cache.InvalidateRegion(addr, size); + return false; +} + +void RasterizerOpenGL::OnCacheInvalidation(VAddr addr, u64 size) { MICROPROFILE_SCOPE(OpenGL_CacheManagement); if (addr == 0 || size == 0) { return; } - shader_cache.OnCPUWrite(addr, size); { std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); @@ -499,15 +520,11 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.CachedWriteMemory(addr, size); } + shader_cache.InvalidateRegion(addr, size); } void RasterizerOpenGL::InvalidateGPUCache() { - MICROPROFILE_SCOPE(OpenGL_CacheManagement); - shader_cache.SyncGuestHost(); - { - std::scoped_lock lock{buffer_cache.mutex}; - buffer_cache.FlushCachedWrites(); - } + gpu.InvalidateGPUCache(); } void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { @@ -519,7 +536,7 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.WriteMemory(addr, size); } - shader_cache.OnCPUWrite(addr, size); + shader_cache.OnCacheInvalidation(addr, size); } void RasterizerOpenGL::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index a73ad15c1..8eda2ddba 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -98,7 +98,8 @@ public: VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; - void OnCPUWrite(VAddr addr, u64 size) override; + void OnCacheInvalidation(VAddr addr, u64 size) override; + bool OnCPUWrite(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index f7c0d939a..456bb040e 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -566,11 +566,32 @@ void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<VAddr, std::s } } -void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { +bool RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { + if (addr == 0 || size == 0) { + return false; + } + + { + std::scoped_lock lock{buffer_cache.mutex}; + if (buffer_cache.OnCPUWrite(addr, size)) { + return true; + } + } + + { + std::scoped_lock lock{texture_cache.mutex}; + texture_cache.WriteMemory(addr, size); + } + + pipeline_cache.InvalidateRegion(addr, size); + return false; +} + +void RasterizerVulkan::OnCacheInvalidation(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } - pipeline_cache.OnCPUWrite(addr, size); + { std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); @@ -579,14 +600,11 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.CachedWriteMemory(addr, size); } + pipeline_cache.InvalidateRegion(addr, size); } void RasterizerVulkan::InvalidateGPUCache() { - pipeline_cache.SyncGuestHost(); - { - std::scoped_lock lock{buffer_cache.mutex}; - buffer_cache.FlushCachedWrites(); - } + gpu.InvalidateGPUCache(); } void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { @@ -598,7 +616,7 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.WriteMemory(addr, size); } - pipeline_cache.OnCPUWrite(addr, size); + pipeline_cache.OnCacheInvalidation(addr, size); } void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index b39710b3c..73257d964 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -96,7 +96,8 @@ public: void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) override; - void OnCPUWrite(VAddr addr, u64 size) override; + void OnCacheInvalidation(VAddr addr, u64 size) override; + bool OnCPUWrite(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 8385b5509..3aac3cfab 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -36,8 +36,10 @@ using VideoCommon::ImageFlagBits; using VideoCommon::ImageInfo; using VideoCommon::ImageType; using VideoCommon::SubresourceRange; +using VideoCore::Surface::BytesPerBlock; using VideoCore::Surface::IsPixelFormatASTC; using VideoCore::Surface::IsPixelFormatInteger; +using VideoCore::Surface::SurfaceType; namespace { constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { @@ -130,7 +132,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info) { const PixelFormat format = StorageFormat(info.format); const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format); - VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; + VkImageCreateFlags flags{}; if (info.type == ImageType::e2D && info.resources.layers >= 6 && info.size.width == info.size.height && !device.HasBrokenCubeImageCompability()) { flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT; @@ -163,11 +165,24 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { } [[nodiscard]] vk::Image MakeImage(const Device& device, const MemoryAllocator& allocator, - const ImageInfo& info) { + const ImageInfo& info, std::span<const VkFormat> view_formats) { if (info.type == ImageType::Buffer) { return vk::Image{}; } - return allocator.CreateImage(MakeImageCreateInfo(device, info)); + VkImageCreateInfo image_ci = MakeImageCreateInfo(device, info); + const VkImageFormatListCreateInfo image_format_list = { + .sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO, + .pNext = nullptr, + .viewFormatCount = static_cast<u32>(view_formats.size()), + .pViewFormats = view_formats.data(), + }; + if (view_formats.size() > 1) { + image_ci.flags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; + if (device.IsKhrImageFormatListSupported()) { + image_ci.pNext = &image_format_list; + } + } + return allocator.CreateImage(image_ci); } [[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) { @@ -806,6 +821,23 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue, memory_allocator); } + if (!device.IsKhrImageFormatListSupported()) { + return; + } + for (size_t index_a = 0; index_a < VideoCore::Surface::MaxPixelFormat; index_a++) { + const auto image_format = static_cast<PixelFormat>(index_a); + if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) { + view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32); + } + for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) { + const auto view_format = static_cast<PixelFormat>(index_b); + if (VideoCore::Surface::IsViewCompatible(image_format, view_format, false, true)) { + const auto view_info = + MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, view_format); + view_formats[index_a].push_back(view_info.format); + } + } + } } void TextureCacheRuntime::Finish() { @@ -1265,8 +1297,8 @@ void TextureCacheRuntime::TickFrame() {} Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_) : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime_.scheduler}, - runtime{&runtime_}, - original_image(MakeImage(runtime_.device, runtime_.memory_allocator, info)), + runtime{&runtime_}, original_image(MakeImage(runtime_.device, runtime_.memory_allocator, info, + runtime->ViewFormats(info.format))), aspect_mask(ImageAspectMask(info.format)) { if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) { if (Settings::values.async_astc.GetValue()) { @@ -1471,7 +1503,8 @@ bool Image::ScaleUp(bool ignore) { auto scaled_info = info; scaled_info.size.width = scaled_width; scaled_info.size.height = scaled_height; - scaled_image = MakeImage(runtime->device, runtime->memory_allocator, scaled_info); + scaled_image = MakeImage(runtime->device, runtime->memory_allocator, scaled_info, + runtime->ViewFormats(info.format)); ignore = false; } current_image = *scaled_image; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 220943116..6621210ea 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -103,6 +103,10 @@ public: [[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size); + std::span<const VkFormat> ViewFormats(PixelFormat format) { + return view_formats[static_cast<std::size_t>(format)]; + } + void BarrierFeedbackLoop(); const Device& device; @@ -113,6 +117,7 @@ public: RenderPassCache& render_pass_cache; std::optional<ASTCDecoderPass> astc_decoder_pass; const Settings::ResolutionScalingInfo& resolution; + std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats; static constexpr size_t indexing_slots = 8 * sizeof(size_t); std::array<vk::Buffer, indexing_slots> buffers{}; diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp index 4db948b6d..01701201d 100644 --- a/src/video_core/shader_cache.cpp +++ b/src/video_core/shader_cache.cpp @@ -24,7 +24,7 @@ void ShaderCache::InvalidateRegion(VAddr addr, size_t size) { RemovePendingShaders(); } -void ShaderCache::OnCPUWrite(VAddr addr, size_t size) { +void ShaderCache::OnCacheInvalidation(VAddr addr, size_t size) { std::scoped_lock lock{invalidation_mutex}; InvalidatePagesInRegion(addr, size); } diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h index f3cc4c70b..de8e08002 100644 --- a/src/video_core/shader_cache.h +++ b/src/video_core/shader_cache.h @@ -62,7 +62,7 @@ public: /// @brief Unmarks a memory region as cached and marks it for removal /// @param addr Start address of the CPU write operation /// @param size Number of bytes of the CPU write operation - void OnCPUWrite(VAddr addr, size_t size); + void OnCacheInvalidation(VAddr addr, size_t size); /// @brief Flushes delayed removal operations void SyncGuestHost(); diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 8190f3ba1..3a859139c 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -598,6 +598,10 @@ void TextureCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t siz [&](ImageId id, Image&) { deleted_images.push_back(id); }); for (const ImageId id : deleted_images) { Image& image = slot_images[id]; + if (True(image.flags & ImageFlagBits::CpuModified)) { + continue; + } + image.flags |= ImageFlagBits::CpuModified; if (True(image.flags & ImageFlagBits::Remapped)) { continue; } @@ -865,11 +869,15 @@ void TextureCache<P>::PopAsyncFlushes() { template <class P> ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload) { const ImageInfo dst_info(operand); - const ImageId image_id = FindDMAImage(dst_info, operand.address); - if (!image_id) { + const ImageId dst_id = FindDMAImage(dst_info, operand.address); + if (!dst_id) { + return NULL_IMAGE_ID; + } + auto& image = slot_images[dst_id]; + if (False(image.flags & ImageFlagBits::GpuModified)) { + // No need to waste time on an image that's synced with guest return NULL_IMAGE_ID; } - auto& image = slot_images[image_id]; if (image.info.type == ImageType::e3D) { // Don't accelerate 3D images. return NULL_IMAGE_ID; @@ -883,7 +891,7 @@ ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, boo if (!base) { return NULL_IMAGE_ID; } - return image_id; + return dst_id; } template <class P> diff --git a/src/video_core/texture_cache/types.h b/src/video_core/texture_cache/types.h index a0e10643f..0453456b4 100644 --- a/src/video_core/texture_cache/types.h +++ b/src/video_core/texture_cache/types.h @@ -54,7 +54,6 @@ enum class RelaxedOptions : u32 { Format = 1 << 1, Samples = 1 << 2, ForceBrokenViews = 1 << 3, - FormatBpp = 1 << 4, }; DECLARE_ENUM_FLAG_OPERATORS(RelaxedOptions) diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index 9a618a57a..0de6ed09d 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -1201,8 +1201,7 @@ std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate, const // Format checking is relaxed, but we still have to check for matching bytes per block. // This avoids creating a view for blits on UE4 titles where formats with different bytes // per block are aliased. - if (BytesPerBlock(existing.format) != BytesPerBlock(candidate.format) && - False(options & RelaxedOptions::FormatBpp)) { + if (BytesPerBlock(existing.format) != BytesPerBlock(candidate.format)) { return std::nullopt; } } else { @@ -1233,11 +1232,7 @@ std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate, const } const bool strict_size = False(options & RelaxedOptions::Size); if (!IsBlockLinearSizeCompatible(existing, candidate, base->level, 0, strict_size)) { - if (False(options & RelaxedOptions::FormatBpp)) { - return std::nullopt; - } else if (!IsBlockLinearSizeCompatibleBPPRelaxed(existing, candidate, base->level, 0)) { - return std::nullopt; - } + return std::nullopt; } // TODO: compare block sizes return base; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 421e71e5a..e04852e01 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -485,7 +485,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); } } - if (extensions.extended_dynamic_state2 && (is_radv || is_qualcomm)) { + if (extensions.extended_dynamic_state2 && is_radv) { const u32 version = (properties.properties.driverVersion << 3) >> 3; if (version < VK_MAKE_API_VERSION(0, 22, 3, 1)) { LOG_WARNING( @@ -498,6 +498,20 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); } } + if (extensions.extended_dynamic_state2 && is_qualcomm) { + const u32 version = (properties.properties.driverVersion << 3) >> 3; + if (version >= VK_MAKE_API_VERSION(0, 0, 676, 0) && + version < VK_MAKE_API_VERSION(0, 0, 680, 0)) { + // Qualcomm Adreno 7xx drivers do not properly support extended_dynamic_state2. + LOG_WARNING(Render_Vulkan, + "Qualcomm Adreno 7xx drivers have broken VK_EXT_extended_dynamic_state2"); + features.extended_dynamic_state2.extendedDynamicState2 = false; + features.extended_dynamic_state2.extendedDynamicState2LogicOp = false; + features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false; + extensions.extended_dynamic_state2 = false; + loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); + } + } if (extensions.extended_dynamic_state3 && is_radv) { LOG_WARNING(Render_Vulkan, "RADV has broken extendedDynamicState3ColorBlendEquation"); features.extended_dynamic_state3.extendedDynamicState3ColorBlendEnable = false; @@ -512,8 +526,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR dynamic_state3_enables = false; } } - if (extensions.vertex_input_dynamic_state && (is_radv || is_qualcomm)) { - // Qualcomm S8gen2 drivers do not properly support vertex_input_dynamic_state. + if (extensions.vertex_input_dynamic_state && is_radv) { // TODO(ameerj): Blacklist only offending driver versions // TODO(ameerj): Confirm if RDNA1 is affected const bool is_rdna2 = @@ -526,6 +539,19 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); } } + if (extensions.vertex_input_dynamic_state && is_qualcomm) { + const u32 version = (properties.properties.driverVersion << 3) >> 3; + if (version >= VK_MAKE_API_VERSION(0, 0, 676, 0) && + version < VK_MAKE_API_VERSION(0, 0, 680, 0)) { + // Qualcomm Adreno 7xx drivers do not properly support vertex_input_dynamic_state. + LOG_WARNING( + Render_Vulkan, + "Qualcomm Adreno 7xx drivers have broken VK_EXT_vertex_input_dynamic_state"); + features.vertex_input_dynamic_state.vertexInputDynamicState = false; + extensions.vertex_input_dynamic_state = false; + loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); + } + } sets_per_pool = 64; if (extensions.extended_dynamic_state3 && is_amd_driver && @@ -774,6 +800,17 @@ bool Device::ShouldBoostClocks() const { return validated_driver && !is_steam_deck && !is_debugging; } +bool Device::HasTimelineSemaphore() const { + if (GetDriverID() == VK_DRIVER_ID_QUALCOMM_PROPRIETARY || + GetDriverID() == VK_DRIVER_ID_MESA_TURNIP) { + // Timeline semaphores do not work properly on all Qualcomm drivers. + // They generally work properly with Turnip drivers, but are problematic on some devices + // (e.g. ZTE handsets with Snapdragon 870). + return false; + } + return features.timeline_semaphore.timelineSemaphore; +} + bool Device::GetSuitability(bool requires_swapchain) { // Assume we will be suitable. bool suitable = true; diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 1f17265d5..be3ed45ff 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -77,6 +77,7 @@ VK_DEFINE_HANDLE(VmaAllocator) EXTENSION(KHR, SPIRV_1_4, spirv_1_4) \ EXTENSION(KHR, SWAPCHAIN, swapchain) \ EXTENSION(KHR, SWAPCHAIN_MUTABLE_FORMAT, swapchain_mutable_format) \ + EXTENSION(KHR, IMAGE_FORMAT_LIST, image_format_list) \ EXTENSION(NV, DEVICE_DIAGNOSTICS_CONFIG, device_diagnostics_config) \ EXTENSION(NV, GEOMETRY_SHADER_PASSTHROUGH, geometry_shader_passthrough) \ EXTENSION(NV, VIEWPORT_ARRAY2, viewport_array2) \ @@ -408,6 +409,11 @@ public: return extensions.workgroup_memory_explicit_layout; } + /// Returns true if the device supports VK_KHR_image_format_list. + bool IsKhrImageFormatListSupported() const { + return extensions.image_format_list || instance_version >= VK_API_VERSION_1_2; + } + /// Returns true if the device supports VK_EXT_primitive_topology_list_restart. bool IsTopologyListPrimitiveRestartSupported() const { return features.primitive_topology_list_restart.primitiveTopologyListRestart; @@ -522,13 +528,7 @@ public: return extensions.shader_atomic_int64; } - bool HasTimelineSemaphore() const { - if (GetDriverID() == VK_DRIVER_ID_QUALCOMM_PROPRIETARY) { - // Timeline semaphores do not work properly on all Qualcomm drivers. - return false; - } - return features.timeline_semaphore.timelineSemaphore; - } + bool HasTimelineSemaphore() const; /// Returns the minimum supported version of SPIR-V. u32 SupportedSpirvVersion() const { |