Texture Cache: Initial Implementation of Sparse Textures.

This commit is contained in:
Fernando Sahmkow 2021-06-12 15:52:27 +02:00
parent eb0e10cff2
commit 38165fb7e3
12 changed files with 310 additions and 23 deletions

View File

@ -127,8 +127,13 @@ void MemoryManager::SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::s
//// Lock the new page //// Lock the new page
// TryLockPage(page_entry, size); // TryLockPage(page_entry, size);
auto& current_page = page_table[PageEntryIndex(gpu_addr)];
if (current_page.IsValid() != page_entry.IsValid() ||
current_page.ToAddress() != page_entry.ToAddress()) {
rasterizer->ModifyGPUMemory(gpu_addr, size);
}
page_table[PageEntryIndex(gpu_addr)] = page_entry; current_page = page_entry;
} }
std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align, std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align,

View File

@ -87,6 +87,9 @@ public:
/// Unmap memory range /// Unmap memory range
virtual void UnmapMemory(VAddr addr, u64 size) = 0; virtual void UnmapMemory(VAddr addr, u64 size) = 0;
/// Unmap memory range
virtual void ModifyGPUMemory(GPUVAddr addr, u64 size) = 0;
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
/// and invalidated /// and invalidated
virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0; virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;

View File

@ -611,6 +611,13 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
shader_cache.OnCPUWrite(addr, size); shader_cache.OnCPUWrite(addr, size);
} }
void RasterizerOpenGL::ModifyGPUMemory(GPUVAddr addr, u64 size) {
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.UnmapGPUMemory(addr, size);
}
}
void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) { void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) {
if (!gpu.IsAsync()) { if (!gpu.IsAsync()) {
gpu_memory.Write<u32>(addr, value); gpu_memory.Write<u32>(addr, value);

View File

@ -80,6 +80,7 @@ public:
void OnCPUWrite(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override;
void SyncGuestHost() override; void SyncGuestHost() override;
void UnmapMemory(VAddr addr, u64 size) override; void UnmapMemory(VAddr addr, u64 size) override;
void ModifyGPUMemory(GPUVAddr addr, u64 size) override;
void SignalSemaphore(GPUVAddr addr, u32 value) override; void SignalSemaphore(GPUVAddr addr, u32 value) override;
void SignalSyncPoint(u32 value) override; void SignalSyncPoint(u32 value) override;
void ReleaseFences() override; void ReleaseFences() override;

View File

@ -557,6 +557,13 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) {
pipeline_cache.OnCPUWrite(addr, size); pipeline_cache.OnCPUWrite(addr, size);
} }
void RasterizerVulkan::ModifyGPUMemory(GPUVAddr addr, u64 size) {
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.UnmapGPUMemory(addr, size);
}
}
void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) { void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) {
if (!gpu.IsAsync()) { if (!gpu.IsAsync()) {
gpu_memory.Write<u32>(addr, value); gpu_memory.Write<u32>(addr, value);

View File

@ -72,6 +72,7 @@ public:
void OnCPUWrite(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override;
void SyncGuestHost() override; void SyncGuestHost() override;
void UnmapMemory(VAddr addr, u64 size) override; void UnmapMemory(VAddr addr, u64 size) override;
void ModifyGPUMemory(GPUVAddr addr, u64 size) override;
void SignalSemaphore(GPUVAddr addr, u32 value) override; void SignalSemaphore(GPUVAddr addr, u32 value) override;
void SignalSyncPoint(u32 value) override; void SignalSyncPoint(u32 value) override;
void ReleaseFences() override; void ReleaseFences() override;

View File

@ -69,6 +69,9 @@ ImageBase::ImageBase(const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_
} }
} }
ImageMapView::ImageMapView(GPUVAddr gpu_addr_, VAddr cpu_addr_, size_t size_, ImageId image_id_)
: gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_}, size{size_}, image_id{image_id_} {}
std::optional<SubresourceBase> ImageBase::TryFindBase(GPUVAddr other_addr) const noexcept { std::optional<SubresourceBase> ImageBase::TryFindBase(GPUVAddr other_addr) const noexcept {
if (other_addr < gpu_addr) { if (other_addr < gpu_addr) {
// Subresource address can't be lower than the base // Subresource address can't be lower than the base

View File

@ -57,6 +57,12 @@ struct ImageBase {
return cpu_addr < overlap_end && overlap_cpu_addr < cpu_addr_end; return cpu_addr < overlap_end && overlap_cpu_addr < cpu_addr_end;
} }
[[nodiscard]] bool OverlapsGPU(GPUVAddr overlap_gpu_addr, size_t overlap_size) const noexcept {
const VAddr overlap_end = overlap_gpu_addr + overlap_size;
const GPUVAddr gpu_addr_end = gpu_addr + guest_size_bytes;
return gpu_addr < overlap_end && overlap_gpu_addr < gpu_addr_end;
}
void CheckBadOverlapState(); void CheckBadOverlapState();
void CheckAliasState(); void CheckAliasState();
@ -84,6 +90,8 @@ struct ImageBase {
std::vector<AliasedImage> aliased_images; std::vector<AliasedImage> aliased_images;
std::vector<ImageId> overlapping_images; std::vector<ImageId> overlapping_images;
ImageMapId map_view_id{};
bool is_sparse{};
}; };
struct ImageAllocBase { struct ImageAllocBase {

View File

@ -152,6 +152,9 @@ public:
/// Remove images in a region /// Remove images in a region
void UnmapMemory(VAddr cpu_addr, size_t size); void UnmapMemory(VAddr cpu_addr, size_t size);
/// Remove images in a region
void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size);
/// Blit an image with the given parameters /// Blit an image with the given parameters
void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& src,
@ -190,7 +193,22 @@ public:
private: private:
/// Iterate over all page indices in a range /// Iterate over all page indices in a range
template <typename Func> template <typename Func>
static void ForEachPage(VAddr addr, size_t size, Func&& func) { static void ForEachCPUPage(VAddr addr, size_t size, Func&& func) {
static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
const u64 page_end = (addr + size - 1) >> PAGE_BITS;
for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
if constexpr (RETURNS_BOOL) {
if (func(page)) {
break;
}
} else {
func(page);
}
}
}
template <typename Func>
static void ForEachGPUPage(GPUVAddr addr, size_t size, Func&& func) {
static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>; static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
const u64 page_end = (addr + size - 1) >> PAGE_BITS; const u64 page_end = (addr + size - 1) >> PAGE_BITS;
for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) { for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
@ -269,6 +287,13 @@ private:
template <typename Func> template <typename Func>
void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func); void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func);
template <typename Func>
void ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func);
/// Iterates over all the images in a region calling func
template <typename Func>
void ForEachSparseSegment(ImageBase& image, Func&& func);
/// Find or create an image view in the given image with the passed parameters /// Find or create an image view in the given image with the passed parameters
[[nodiscard]] ImageViewId FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info); [[nodiscard]] ImageViewId FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info);
@ -340,7 +365,8 @@ private:
std::unordered_map<TSCEntry, SamplerId> samplers; std::unordered_map<TSCEntry, SamplerId> samplers;
std::unordered_map<RenderTargets, FramebufferId> framebuffers; std::unordered_map<RenderTargets, FramebufferId> framebuffers;
std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> page_table; std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table;
std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table;
bool has_deleted_images = false; bool has_deleted_images = false;
u64 total_used_memory = 0; u64 total_used_memory = 0;
@ -349,6 +375,7 @@ private:
u64 critical_memory; u64 critical_memory;
SlotVector<Image> slot_images; SlotVector<Image> slot_images;
SlotVector<ImageMapView> slot_map_views;
SlotVector<ImageView> slot_image_views; SlotVector<ImageView> slot_image_views;
SlotVector<ImageAlloc> slot_image_allocs; SlotVector<ImageAlloc> slot_image_allocs;
SlotVector<Sampler> slot_samplers; SlotVector<Sampler> slot_samplers;
@ -702,6 +729,21 @@ void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) {
} }
} }
template <class P>
void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) {
std::vector<ImageId> deleted_images;
ForEachImageInRegionGPU(gpu_addr, size,
[&](ImageId id, Image&) { deleted_images.push_back(id); });
for (const ImageId id : deleted_images) {
Image& image = slot_images[id];
if (True(image.flags & ImageFlagBits::Tracked)) {
UntrackImage(image);
}
UnregisterImage(id);
DeleteImage(id);
}
}
template <class P> template <class P>
void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& src,
@ -833,9 +875,10 @@ typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_ad
if (it == page_table.end()) { if (it == page_table.end()) {
return nullptr; return nullptr;
} }
const auto& image_ids = it->second; const auto& image_map_ids = it->second;
for (const ImageId image_id : image_ids) { for (const ImageMapId map_id : image_map_ids) {
const ImageBase& image = slot_images[image_id]; const ImageMapView& map = slot_map_views[map_id];
const ImageBase& image = slot_images[map.image_id];
if (image.cpu_addr != cpu_addr) { if (image.cpu_addr != cpu_addr) {
continue; continue;
} }
@ -958,7 +1001,7 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
template <class P> template <class P>
ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) { ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) {
if (!IsValidAddress(gpu_memory, config)) { if (!IsValidEntry(gpu_memory, config)) {
return NULL_IMAGE_VIEW_ID; return NULL_IMAGE_VIEW_ID;
} }
const auto [pair, is_new] = image_views.try_emplace(config); const auto [pair, is_new] = image_views.try_emplace(config);
@ -1026,7 +1069,7 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
} }
return false; return false;
}; };
ForEachImageInRegion(*cpu_addr, CalculateGuestSizeInBytes(info), lambda); ForEachImageInRegionGPU(gpu_addr, CalculateGuestSizeInBytes(info), lambda);
return image_id; return image_id;
} }
@ -1056,7 +1099,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
std::vector<ImageId> left_aliased_ids; std::vector<ImageId> left_aliased_ids;
std::vector<ImageId> right_aliased_ids; std::vector<ImageId> right_aliased_ids;
std::vector<ImageId> bad_overlap_ids; std::vector<ImageId> bad_overlap_ids;
ForEachImageInRegion(cpu_addr, size_bytes, [&](ImageId overlap_id, ImageBase& overlap) { ForEachImageInRegionGPU(gpu_addr, size_bytes, [&](ImageId overlap_id, ImageBase& overlap) {
if (info.type == ImageType::Linear) { if (info.type == ImageType::Linear) {
if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) { if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) {
// Alias linear images with the same pitch // Alias linear images with the same pitch
@ -1091,6 +1134,24 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr); const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr);
Image& new_image = slot_images[new_image_id]; Image& new_image = slot_images[new_image_id];
new_image.is_sparse = false;
if (new_image.info.type != ImageType::Linear && new_image.info.type != ImageType::Buffer) {
const LevelArray offsets = CalculateMipLevelOffsets(new_image.info);
size_t level;
const size_t levels = static_cast<size_t>(new_image.info.resources.levels);
VAddr n_cpu_addr = new_image.cpu_addr;
GPUVAddr n_gpu_addr = new_image.gpu_addr;
for (level = 0; level < levels; level++) {
n_gpu_addr += offsets[level];
n_cpu_addr += offsets[level];
std::optional<VAddr> cpu_addr_opt = gpu_memory.GpuToCpuAddress(n_gpu_addr);
if (!cpu_addr_opt || *cpu_addr_opt == 0 || n_cpu_addr != *cpu_addr_opt) {
new_image.is_sparse = true;
break;
}
}
}
// TODO: Only upload what we need // TODO: Only upload what we need
RefreshContents(new_image); RefreshContents(new_image);
@ -1239,7 +1300,8 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f
using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type; using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type;
static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>; static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
boost::container::small_vector<ImageId, 32> images; boost::container::small_vector<ImageId, 32> images;
ForEachPage(cpu_addr, size, [this, &images, cpu_addr, size, func](u64 page) { boost::container::small_vector<ImageMapId, 32> maps;
ForEachCPUPage(cpu_addr, size, [this, &images, &maps, cpu_addr, size, func](u64 page) {
const auto it = page_table.find(page); const auto it = page_table.find(page);
if (it == page_table.end()) { if (it == page_table.end()) {
if constexpr (BOOL_BREAK) { if constexpr (BOOL_BREAK) {
@ -1248,12 +1310,63 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f
return; return;
} }
} }
for (const ImageMapId map_id : it->second) {
ImageMapView& map = slot_map_views[map_id];
if (map.picked) {
continue;
}
if (!map.Overlaps(cpu_addr, size)) {
continue;
}
map.picked = true;
maps.push_back(map_id);
Image& image = slot_images[map.image_id];
if (True(image.flags & ImageFlagBits::Picked)) {
continue;
}
image.flags |= ImageFlagBits::Picked;
images.push_back(map.image_id);
if constexpr (BOOL_BREAK) {
if (func(map.image_id, image)) {
return true;
}
} else {
func(map.image_id, image);
}
}
if constexpr (BOOL_BREAK) {
return false;
}
});
for (const ImageId image_id : images) {
slot_images[image_id].flags &= ~ImageFlagBits::Picked;
}
for (const ImageMapId map_id : maps) {
slot_map_views[map_id].picked = false;
}
}
template <class P>
template <typename Func>
void TextureCache<P>::ForEachImageInRegionGPU(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 = gpu_page_table.find(page);
if (it == gpu_page_table.end()) {
if constexpr (BOOL_BREAK) {
return false;
} else {
return;
}
}
for (const ImageId image_id : it->second) { for (const ImageId image_id : it->second) {
Image& image = slot_images[image_id]; Image& image = slot_images[image_id];
if (True(image.flags & ImageFlagBits::Picked)) { if (True(image.flags & ImageFlagBits::Picked)) {
continue; continue;
} }
if (!image.Overlaps(cpu_addr, size)) { if (!image.OverlapsGPU(gpu_addr, size)) {
continue; continue;
} }
image.flags |= ImageFlagBits::Picked; image.flags |= ImageFlagBits::Picked;
@ -1275,6 +1388,30 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f
} }
} }
template <class P>
template <typename Func>
void TextureCache<P>::ForEachSparseSegment(ImageBase& image, Func&& func) {
using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, size_t>::type;
static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
GPUVAddr gpu_addr = image.gpu_addr;
const size_t levels = image.info.resources.levels;
const auto mipmap_sizes = CalculateMipLevelSizes(image.info);
for (size_t level = 0; level < levels; level++) {
const size_t size = mipmap_sizes[level];
std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
if (cpu_addr && *cpu_addr != 0) {
if constexpr (BOOL_BREAK) {
if (func(gpu_addr, *cpu_addr, size)) {
return true;
}
} else {
func(gpu_addr, *cpu_addr, size);
}
}
gpu_addr += size;
}
}
template <class P> template <class P>
ImageViewId TextureCache<P>::FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info) { ImageViewId TextureCache<P>::FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info) {
Image& image = slot_images[image_id]; Image& image = slot_images[image_id];
@ -1292,8 +1429,6 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
ASSERT_MSG(False(image.flags & ImageFlagBits::Registered), ASSERT_MSG(False(image.flags & ImageFlagBits::Registered),
"Trying to register an already registered image"); "Trying to register an already registered image");
image.flags |= ImageFlagBits::Registered; image.flags |= ImageFlagBits::Registered;
ForEachPage(image.cpu_addr, image.guest_size_bytes,
[this, image_id](u64 page) { page_table[page].push_back(image_id); });
u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes); u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
if ((IsPixelFormatASTC(image.info.format) && if ((IsPixelFormatASTC(image.info.format) &&
True(image.flags & ImageFlagBits::AcceleratedUpload)) || True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
@ -1301,6 +1436,21 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format); tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
} }
total_used_memory += Common::AlignUp(tentative_size, 1024); total_used_memory += Common::AlignUp(tentative_size, 1024);
ForEachGPUPage(image.gpu_addr, image.guest_size_bytes,
[this, image_id](u64 page) { gpu_page_table[page].push_back(image_id); });
if (!image.is_sparse) {
auto map_id =
slot_map_views.insert(image.gpu_addr, image.cpu_addr, image.guest_size_bytes, image_id);
ForEachCPUPage(image.cpu_addr, image.guest_size_bytes,
[this, map_id](u64 page) { page_table[page].push_back(map_id); });
image.map_view_id = map_id;
return;
}
ForEachSparseSegment(image, [this, image_id](GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
auto map_id = slot_map_views.insert(gpu_addr, cpu_addr, size, image_id);
ForEachCPUPage(cpu_addr, size,
[this, map_id](u64 page) { page_table[page].push_back(map_id); });
});
} }
template <class P> template <class P>
@ -1317,9 +1467,9 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format); tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
} }
total_used_memory -= Common::AlignUp(tentative_size, 1024); total_used_memory -= Common::AlignUp(tentative_size, 1024);
ForEachPage(image.cpu_addr, image.guest_size_bytes, [this, image_id](u64 page) { ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
const auto page_it = page_table.find(page); const auto page_it = gpu_page_table.find(page);
if (page_it == page_table.end()) { if (page_it == gpu_page_table.end()) {
UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS); UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
return; return;
} }
@ -1331,20 +1481,84 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
} }
image_ids.erase(vector_it); image_ids.erase(vector_it);
}); });
if (!image.is_sparse) {
const auto map_id = image.map_view_id;
ForEachCPUPage(image.cpu_addr, image.guest_size_bytes, [this, map_id](u64 page) {
const auto page_it = page_table.find(page);
if (page_it == page_table.end()) {
UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
return;
}
std::vector<ImageMapId>& image_map_ids = page_it->second;
const auto vector_it = std::ranges::find(image_map_ids, map_id);
if (vector_it == image_map_ids.end()) {
UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}",
page << PAGE_BITS);
return;
}
image_map_ids.erase(vector_it);
});
slot_map_views.erase(map_id);
return;
}
boost::container::small_vector<ImageMapId, 8> maps_to_delete;
ForEachSparseSegment(
image, [this, image_id, &maps_to_delete]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr,
size_t size) {
ForEachCPUPage(cpu_addr, size, [this, image_id, &maps_to_delete](u64 page) {
const auto page_it = page_table.find(page);
if (page_it == page_table.end()) {
UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
return;
}
std::vector<ImageMapId>& image_map_ids = page_it->second;
auto vector_it = image_map_ids.begin();
while (vector_it != image_map_ids.end()) {
ImageMapView& map = slot_map_views[*vector_it];
if (map.image_id != image_id) {
vector_it++;
continue;
}
if (!map.picked) {
maps_to_delete.push_back(*vector_it);
map.picked = true;
}
vector_it = image_map_ids.erase(vector_it);
}
});
});
for (const ImageMapId map_id : maps_to_delete) {
slot_map_views.erase(map_id);
}
} }
template <class P> template <class P>
void TextureCache<P>::TrackImage(ImageBase& image) { void TextureCache<P>::TrackImage(ImageBase& image) {
ASSERT(False(image.flags & ImageFlagBits::Tracked)); ASSERT(False(image.flags & ImageFlagBits::Tracked));
image.flags |= ImageFlagBits::Tracked; image.flags |= ImageFlagBits::Tracked;
if (!image.is_sparse) {
rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1); rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
return;
}
ForEachSparseSegment(image,
[this]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
});
} }
template <class P> template <class P>
void TextureCache<P>::UntrackImage(ImageBase& image) { void TextureCache<P>::UntrackImage(ImageBase& image) {
ASSERT(True(image.flags & ImageFlagBits::Tracked)); ASSERT(True(image.flags & ImageFlagBits::Tracked));
image.flags &= ~ImageFlagBits::Tracked; image.flags &= ~ImageFlagBits::Tracked;
if (!image.is_sparse) {
rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1); rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1);
return;
}
ForEachSparseSegment(image,
[this]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1);
});
} }
template <class P> template <class P>

View File

@ -16,6 +16,7 @@ constexpr size_t MAX_MIP_LEVELS = 14;
constexpr SlotId CORRUPT_ID{0xfffffffe}; constexpr SlotId CORRUPT_ID{0xfffffffe};
using ImageId = SlotId; using ImageId = SlotId;
using ImageMapId = SlotId;
using ImageViewId = SlotId; using ImageViewId = SlotId;
using ImageAllocId = SlotId; using ImageAllocId = SlotId;
using SamplerId = SlotId; using SamplerId = SlotId;

View File

@ -664,6 +664,16 @@ LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept {
return offsets; return offsets;
} }
LevelArray CalculateMipLevelSizes(const ImageInfo& info) noexcept {
const u32 num_levels = info.resources.levels;
const LevelInfo level_info = MakeLevelInfo(info);
LevelArray sizes{};
for (u32 level = 0; level < num_levels; ++level) {
sizes[level] = CalculateLevelSize(level_info, level);
}
return sizes;
}
std::vector<u32> CalculateSliceOffsets(const ImageInfo& info) { std::vector<u32> CalculateSliceOffsets(const ImageInfo& info) {
ASSERT(info.type == ImageType::e3D); ASSERT(info.type == ImageType::e3D);
std::vector<u32> offsets; std::vector<u32> offsets;
@ -776,14 +786,37 @@ std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageIn
return copies; return copies;
} }
bool IsValidAddress(const Tegra::MemoryManager& gpu_memory, const TICEntry& config) { bool IsValidAddress(const Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr) {
if (config.Address() == 0) { if (gpu_addr == 0) {
return false; return false;
} }
if (config.Address() > (u64(1) << 48)) { if (gpu_addr > (u64(1) << 48)) {
return false; return false;
} }
return gpu_memory.GpuToCpuAddress(config.Address()).has_value(); const auto cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
return cpu_addr.has_value() && *cpu_addr != 0;
}
bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config) {
const GPUVAddr gpu_addr = config.Address();
if (IsValidAddress(gpu_memory, gpu_addr)) {
return true;
}
if (!config.IsBlockLinear()) {
return false;
}
const size_t levels = config.max_mip_level + 1;
if (levels <= 1) {
return false;
}
const ImageInfo info{config};
const LevelArray offsets = CalculateMipLevelOffsets(info);
for (size_t level = 1; level < levels; level++) {
if (IsValidAddress(gpu_memory, static_cast<GPUVAddr>(gpu_addr + offsets[level]))) {
return true;
}
}
return false;
} }
std::vector<BufferImageCopy> UnswizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, std::vector<BufferImageCopy> UnswizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr,

View File

@ -40,6 +40,8 @@ struct OverlapResult {
[[nodiscard]] LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept; [[nodiscard]] LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept;
[[nodiscard]] LevelArray CalculateMipLevelSizes(const ImageInfo& info) noexcept;
[[nodiscard]] std::vector<u32> CalculateSliceOffsets(const ImageInfo& info); [[nodiscard]] std::vector<u32> CalculateSliceOffsets(const ImageInfo& info);
[[nodiscard]] std::vector<SubresourceBase> CalculateSliceSubresources(const ImageInfo& info); [[nodiscard]] std::vector<SubresourceBase> CalculateSliceSubresources(const ImageInfo& info);
@ -55,7 +57,9 @@ struct OverlapResult {
const ImageInfo& src, const ImageInfo& src,
SubresourceBase base); SubresourceBase base);
[[nodiscard]] bool IsValidAddress(const Tegra::MemoryManager& gpu_memory, const TICEntry& config); [[nodiscard]] bool IsValidAddress(const Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr);
[[nodiscard]] bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config);
[[nodiscard]] std::vector<BufferImageCopy> UnswizzleImage(Tegra::MemoryManager& gpu_memory, [[nodiscard]] std::vector<BufferImageCopy> UnswizzleImage(Tegra::MemoryManager& gpu_memory,
GPUVAddr gpu_addr, const ImageInfo& info, GPUVAddr gpu_addr, const ImageInfo& info,