VERY experimental Nintendo Switch 2 emulator
C++
Texture Cache: make sparse texture table per channel
Fernando Sahmkow352f6d22e3f7
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.hindex 7398ed2ecf..a7400adfab 100644--- a/src/video_core/texture_cache/texture_cache.h+++ b/src/video_core/texture_cache/texture_cache.h@@ -1431,7 +1431,8 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DA}}};- ForEachSparseImageInRegion(gpu_addr, size_bytes, region_check_gpu);+ ForEachSparseImageInRegion(channel_state->gpu_memory.GetID(), gpu_addr, size_bytes,+ region_check_gpu);bool can_rescale = info.rescaleable;bool any_rescaled = false;@@ -1842,7 +1843,7 @@ void TextureCache<P>::ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, sif (!storage_id) {return;}- auto& gpu_page_table = gpu_page_table_storage[*storage_id];+ auto& gpu_page_table = gpu_page_table_storage[*storage_id * 2];ForEachGPUPage(gpu_addr, size,[this, &gpu_page_table, &images, gpu_addr, size, func](u64 page) {const auto it = gpu_page_table.find(page);@@ -1882,41 +1883,48 @@ void TextureCache<P>::ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, stemplate <class P>template <typename Func>-void TextureCache<P>::ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func) {+void TextureCache<P>::ForEachSparseImageInRegion(size_t as_id, GPUVAddr gpu_addr, size_t size,+ Func&& func) {using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type;static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;boost::container::small_vector<ImageId, 8> images;- ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) {- const auto it = sparse_page_table.find(page);- if (it == sparse_page_table.end()) {- if constexpr (BOOL_BREAK) {- return false;- } else {- return;- }- }- for (const ImageId image_id : it->second) {- Image& image = slot_images[image_id];- if (True(image.flags & ImageFlagBits::Picked)) {- continue;- }- if (!image.OverlapsGPU(gpu_addr, size)) {- continue;- }- image.flags |= ImageFlagBits::Picked;- images.push_back(image_id);- if constexpr (BOOL_BREAK) {- if (func(image_id, image)) {- return true;- }- } else {- func(image_id, image);- }- }- if constexpr (BOOL_BREAK) {- return false;- }- });+ auto storage_id = getStorageID(as_id);+ if (!storage_id) {+ return;+ }+ auto& sparse_page_table = gpu_page_table_storage[*storage_id * 2 + 1];+ ForEachGPUPage(gpu_addr, size,+ [this, &sparse_page_table, &images, gpu_addr, size, func](u64 page) {+ const auto it = sparse_page_table.find(page);+ if (it == sparse_page_table.end()) {+ if constexpr (BOOL_BREAK) {+ return false;+ } else {+ return;+ }+ }+ for (const ImageId image_id : it->second) {+ Image& image = slot_images[image_id];+ if (True(image.flags & ImageFlagBits::Picked)) {+ continue;+ }+ if (!image.OverlapsGPU(gpu_addr, size)) {+ continue;+ }+ image.flags |= ImageFlagBits::Picked;+ images.push_back(image_id);+ if constexpr (BOOL_BREAK) {+ if (func(image_id, image)) {+ return true;+ }+ } else {+ func(image_id, image);+ }+ }+ if constexpr (BOOL_BREAK) {+ return false;+ }+ });for (const ImageId image_id : images) {slot_images[image_id].flags &= ~ImageFlagBits::Picked;}@@ -1988,8 +1996,9 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {sparse_maps.push_back(map_id);});sparse_views.emplace(image_id, std::move(sparse_maps));- ForEachGPUPage(image.gpu_addr, image.guest_size_bytes,- [this, image_id](u64 page) { sparse_page_table[page].push_back(image_id); });+ ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {+ (*channel_state->sparse_page_table)[page].push_back(image_id);+ });}template <class P>@@ -2042,7 +2051,7 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {return;}ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, &clear_page_table](u64 page) {- clear_page_table(page, sparse_page_table);+ clear_page_table(page, (*channel_state->sparse_page_table));});auto it = sparse_views.find(image_id);ASSERT(it != sparse_views.end());@@ -2496,13 +2505,15 @@ void TextureCache<P>::CreateChannel(struct Tegra::Control::ChannelState& channelconst auto it = channel_map.find(channel.bind_id);auto* this_state = &channel_storage[it->second];const auto& this_as_ref = address_spaces[channel.memory_manager->GetID()];- this_state->gpu_page_table = &gpu_page_table_storage[this_as_ref.storage_id];+ this_state->gpu_page_table = &gpu_page_table_storage[this_as_ref.storage_id * 2];+ this_state->sparse_page_table = &gpu_page_table_storage[this_as_ref.storage_id * 2 + 1];}/// Bind a channel for execution.template <class P>void TextureCache<P>::OnGPUASRegister([[maybe_unused]] size_t map_id) {gpu_page_table_storage.emplace_back();+ gpu_page_table_storage.emplace_back();}} // namespace VideoCommondiff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.hindex 8699d40d4f..f9aebb293e 100644--- a/src/video_core/texture_cache/texture_cache_base.h+++ b/src/video_core/texture_cache/texture_cache_base.h@@ -86,6 +86,7 @@ public:std::unordered_map<TSCEntry, SamplerId> samplers;TextureCacheGPUMap* gpu_page_table;+ TextureCacheGPUMap* sparse_page_table;};template <class P>@@ -357,7 +358,7 @@ private:void ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, size_t size, Func&& func);template <typename Func>- void ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func);+ void ForEachSparseImageInRegion(size_t as_id, GPUVAddr gpu_addr, size_t size, Func&& func);/// Iterates over all the images in a region calling functemplate <typename Func>@@ -431,7 +432,6 @@ private:std::unordered_map<RenderTargets, FramebufferId> framebuffers;std::unordered_map<u64, std::vector<ImageMapId>, Common::IdentityHash<u64>> page_table;- std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>> sparse_page_table;std::unordered_map<ImageId, boost::container::small_vector<ImageViewId, 16>> sparse_views;DAddr virtual_invalid_space{};