Merge pull request #8092 from yuzu-emu/revert-8080-yo-momma-so-fat-that

Revert "Memory GPU <-> CPU: reduce infighting in the texture cache by adding CPU Cached memory."

src/core/memory.cpp

@@ -322,7 +322,7 @@ struct Memory::Impl {
         }
 
         if (Settings::IsFastmemEnabled()) {
-            const bool is_read_enable = !Settings::IsGPULevelExtreme() || !cached;
+            const bool is_read_enable = Settings::IsGPULevelHigh() || !cached;
             system.DeviceMemory().buffer.Protect(vaddr, size, is_read_enable, !cached);
         }
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -352,7 +352,7 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
     shader_cache.OnCPUWrite(addr, size);
     {
         std::scoped_lock lock{texture_cache.mutex};
-        texture_cache.CachedWriteMemory(addr, size);
+        texture_cache.WriteMemory(addr, size);
     }
     {
         std::scoped_lock lock{buffer_cache.mutex};
@@ -363,10 +363,6 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
 void RasterizerOpenGL::SyncGuestHost() {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     shader_cache.SyncGuestHost();
-    {
-        std::scoped_lock lock{texture_cache.mutex};
-        texture_cache.FlushCachedWrites();
-    }
     {
         std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.FlushCachedWrites();

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -408,7 +408,7 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
     pipeline_cache.OnCPUWrite(addr, size);
     {
         std::scoped_lock lock{texture_cache.mutex};
-        texture_cache.CachedWriteMemory(addr, size);
+        texture_cache.WriteMemory(addr, size);
     }
     {
         std::scoped_lock lock{buffer_cache.mutex};
@@ -418,10 +418,6 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
 
 void RasterizerVulkan::SyncGuestHost() {
     pipeline_cache.SyncGuestHost();
-    {
-        std::scoped_lock lock{texture_cache.mutex};
-        texture_cache.FlushCachedWrites();
-    }
     {
         std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.FlushCachedWrites();

src/video_core/texture_cache/image_base.h

@@ -39,9 +39,6 @@ enum class ImageFlagBits : u32 {
     Rescaled = 1 << 13,
     CheckingRescalable = 1 << 14,
     IsRescalable = 1 << 15,
-
-    // Cached CPU
-    CachedCpuModified = 1 << 16, ///< Contents have been modified from the CPU
 };
 DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)
 

src/video_core/texture_cache/texture_cache.h

@@ -437,23 +437,6 @@ void TextureCache<P>::WriteMemory(VAddr cpu_addr, size_t size) {
     });
 }
 
-template <class P>
-void TextureCache<P>::CachedWriteMemory(VAddr cpu_addr, size_t size) {
-    const VAddr new_cpu_addr = Common::AlignDown(cpu_addr, CPU_PAGE_SIZE);
-    const size_t new_size = Common::AlignUp(size + cpu_addr - new_cpu_addr, CPU_PAGE_SIZE);
-    ForEachImageInRegion(new_cpu_addr, new_size, [this](ImageId image_id, Image& image) {
-        if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
-            return;
-        }
-        image.flags |= ImageFlagBits::CachedCpuModified;
-        cached_cpu_invalidate.insert(image_id);
-
-        if (True(image.flags & ImageFlagBits::Tracked)) {
-            UntrackImage(image, image_id);
-        }
-    });
-}
-
 template <class P>
 void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
     std::vector<ImageId> images;
@@ -511,18 +494,6 @@ void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) {
     }
 }
 
-template <class P>
-void TextureCache<P>::FlushCachedWrites() {
-    for (ImageId image_id : cached_cpu_invalidate) {
-        Image& image = slot_images[image_id];
-        if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
-            image.flags &= ~ImageFlagBits::CachedCpuModified;
-            image.flags |= ImageFlagBits::CpuModified;
-        }
-    }
-    cached_cpu_invalidate.clear();
-}
-
 template <class P>
 void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
                                 const Tegra::Engines::Fermi2D::Surface& src,
@@ -1589,9 +1560,6 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
 template <class P>
 void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
     ASSERT(False(image.flags & ImageFlagBits::Tracked));
-    if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
-        return;
-    }
     image.flags |= ImageFlagBits::Tracked;
     if (False(image.flags & ImageFlagBits::Sparse)) {
         rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
@@ -1648,9 +1616,6 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
         tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
     }
     total_used_memory -= Common::AlignUp(tentative_size, 1024);
-    if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
-        cached_cpu_invalidate.erase(image_id);
-    }
     const GPUVAddr gpu_addr = image.gpu_addr;
     const auto alloc_it = image_allocs_table.find(gpu_addr);
     if (alloc_it == image_allocs_table.end()) {
@@ -1817,11 +1782,7 @@ template <class P>
 void TextureCache<P>::PrepareImage(ImageId image_id, bool is_modification, bool invalidate) {
     Image& image = slot_images[image_id];
     if (invalidate) {
-        if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
+        image.flags &= ~(ImageFlagBits::CpuModified | ImageFlagBits::GpuModified);
-            cached_cpu_invalidate.erase(image_id);
-        }
-        image.flags &= ~(ImageFlagBits::CpuModified | ImageFlagBits::GpuModified |
-                         ImageFlagBits::CachedCpuModified);
         if (False(image.flags & ImageFlagBits::Tracked)) {
             TrackImage(image, image_id);
         }

src/video_core/texture_cache/texture_cache_base.h

@@ -8,7 +8,6 @@
 #include <span>
 #include <type_traits>
 #include <unordered_map>
-#include <unordered_set>
 #include <vector>
 #include <queue>
 
@@ -51,9 +50,6 @@ class TextureCache {
     /// Address shift for caching images into a hash table
     static constexpr u64 PAGE_BITS = 20;
 
-    static constexpr u64 CPU_PAGE_BITS = 12;
-    static constexpr u64 CPU_PAGE_SIZE = 1ULL << CPU_PAGE_BITS;
-
     /// Enables debugging features to the texture cache
     static constexpr bool ENABLE_VALIDATION = P::ENABLE_VALIDATION;
     /// Implement blits as copies between framebuffers
@@ -140,9 +136,6 @@ public:
     /// Mark images in a range as modified from the CPU
     void WriteMemory(VAddr cpu_addr, size_t size);
 
-    /// Mark images in a range as modified from the CPU
-    void CachedWriteMemory(VAddr cpu_addr, size_t size);
-
     /// Download contents of host images to guest memory in a region
     void DownloadMemory(VAddr cpu_addr, size_t size);
 
@@ -152,8 +145,6 @@ public:
     /// Remove images in a region
     void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size);
 
-    void FlushCachedWrites();
-
     /// Blit an image with the given parameters
     void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
                    const Tegra::Engines::Fermi2D::Surface& src,
@@ -375,8 +366,6 @@ private:
 
     std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views;
 
-    std::unordered_set<ImageId> cached_cpu_invalidate;
-
     VAddr virtual_invalid_space{};
 
     bool has_deleted_images = false;