diff options
author | liamwhite <liamwhite@users.noreply.github.com> | 2023-12-18 00:59:49 +0100 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-12-18 00:59:49 +0100 |
commit | 65e646eeba53fe1067b7d69044ddabaee18d868f (patch) | |
tree | 9cd2c5d264e623060186bd51e9c07433e9629234 /src/video_core | |
parent | Merge pull request #12378 from liamwhite/offsetof (diff) | |
download | yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.tar yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.tar.gz yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.tar.bz2 yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.tar.lz yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.tar.xz yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.tar.zst yuzu-65e646eeba53fe1067b7d69044ddabaee18d868f.zip |
Diffstat (limited to 'src/video_core')
-rw-r--r-- | src/video_core/buffer_cache/word_manager.h | 2 | ||||
-rw-r--r-- | src/video_core/rasterizer_accelerated.cpp | 99 | ||||
-rw-r--r-- | src/video_core/rasterizer_accelerated.h | 29 | ||||
-rw-r--r-- | src/video_core/rasterizer_interface.h | 2 | ||||
-rw-r--r-- | src/video_core/shader_cache.cpp | 4 | ||||
-rw-r--r-- | src/video_core/texture_cache/texture_cache.h | 10 |
6 files changed, 76 insertions, 70 deletions
diff --git a/src/video_core/buffer_cache/word_manager.h b/src/video_core/buffer_cache/word_manager.h index 95b752055..a336bde41 100644 --- a/src/video_core/buffer_cache/word_manager.h +++ b/src/video_core/buffer_cache/word_manager.h @@ -473,7 +473,7 @@ private: VAddr addr = cpu_addr + word_index * BYTES_PER_WORD; IteratePages(changed_bits, [&](size_t offset, size_t size) { rasterizer->UpdatePagesCachedCount(addr + offset * BYTES_PER_PAGE, - size * BYTES_PER_PAGE, add_to_rasterizer); + size * BYTES_PER_PAGE, add_to_rasterizer ? 1 : -1); }); } diff --git a/src/video_core/rasterizer_accelerated.cpp b/src/video_core/rasterizer_accelerated.cpp index 3c9477f6e..f200a650f 100644 --- a/src/video_core/rasterizer_accelerated.cpp +++ b/src/video_core/rasterizer_accelerated.cpp @@ -3,7 +3,6 @@ #include <atomic> -#include "common/alignment.h" #include "common/assert.h" #include "common/common_types.h" #include "common/div_ceil.h" @@ -12,65 +11,61 @@ namespace VideoCore { -static constexpr u16 IdentityValue = 1; - using namespace Core::Memory; -RasterizerAccelerated::RasterizerAccelerated(Memory& cpu_memory_) : map{}, cpu_memory{cpu_memory_} { - // We are tracking CPU memory, which cannot map more than 39 bits. - const VAddr start_address = 0; - const VAddr end_address = (1ULL << 39); - const IntervalType address_space_interval(start_address, end_address); - const auto value = std::make_pair(address_space_interval, IdentityValue); - - map.add(value); -} +RasterizerAccelerated::RasterizerAccelerated(Memory& cpu_memory_) + : cached_pages(std::make_unique<CachedPages>()), cpu_memory{cpu_memory_} {} RasterizerAccelerated::~RasterizerAccelerated() = default; -void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, bool cache) { - std::scoped_lock lk{map_lock}; - - // Align sizes. - addr = Common::AlignDown(addr, YUZU_PAGESIZE); - size = Common::AlignUp(size, YUZU_PAGESIZE); - - // Declare the overall interval we are going to operate on. - const VAddr start_address = addr; - const VAddr end_address = addr + size; - const IntervalType modification_range(start_address, end_address); - - // Find the boundaries of where to iterate. - const auto lower = map.lower_bound(modification_range); - const auto upper = map.upper_bound(modification_range); - - // Iterate over the contained intervals. - for (auto it = lower; it != upper; it++) { - // Intersect interval range with modification range. - const auto current_range = modification_range & it->first; - - // Calculate the address and size to operate over. - const auto current_addr = current_range.lower(); - const auto current_size = current_range.upper() - current_addr; - - // Get the current value of the range. - const auto value = it->second; +void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) { + u64 uncache_begin = 0; + u64 cache_begin = 0; + u64 uncache_bytes = 0; + u64 cache_bytes = 0; + + std::atomic_thread_fence(std::memory_order_acquire); + const u64 page_end = Common::DivCeil(addr + size, YUZU_PAGESIZE); + for (u64 page = addr >> YUZU_PAGEBITS; page != page_end; ++page) { + std::atomic_uint16_t& count = cached_pages->at(page >> 2).Count(page); + + if (delta > 0) { + ASSERT_MSG(count.load(std::memory_order::relaxed) < UINT16_MAX, "Count may overflow!"); + } else if (delta < 0) { + ASSERT_MSG(count.load(std::memory_order::relaxed) > 0, "Count may underflow!"); + } else { + ASSERT_MSG(false, "Delta must be non-zero!"); + } - if (cache && value == IdentityValue) { - // If we are going to cache, and the value is not yet referenced, then cache this range. - cpu_memory.RasterizerMarkRegionCached(current_addr, current_size, true); - } else if (!cache && value == IdentityValue + 1) { - // If we are going to uncache, and this is the last reference, then uncache this range. - cpu_memory.RasterizerMarkRegionCached(current_addr, current_size, false); + // Adds or subtracts 1, as count is a unsigned 8-bit value + count.fetch_add(static_cast<u16>(delta), std::memory_order_release); + + // Assume delta is either -1 or 1 + if (count.load(std::memory_order::relaxed) == 0) { + if (uncache_bytes == 0) { + uncache_begin = page; + } + uncache_bytes += YUZU_PAGESIZE; + } else if (uncache_bytes > 0) { + cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, + false); + uncache_bytes = 0; + } + if (count.load(std::memory_order::relaxed) == 1 && delta > 0) { + if (cache_bytes == 0) { + cache_begin = page; + } + cache_bytes += YUZU_PAGESIZE; + } else if (cache_bytes > 0) { + cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true); + cache_bytes = 0; } } - - // Update the set. - const auto value = std::make_pair(modification_range, IdentityValue); - if (cache) { - map.add(value); - } else { - map.subtract(value); + if (uncache_bytes > 0) { + cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, false); + } + if (cache_bytes > 0) { + cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true); } } diff --git a/src/video_core/rasterizer_accelerated.h b/src/video_core/rasterizer_accelerated.h index f1968f186..e6c0ea87a 100644 --- a/src/video_core/rasterizer_accelerated.h +++ b/src/video_core/rasterizer_accelerated.h @@ -3,8 +3,8 @@ #pragma once -#include <mutex> -#include <boost/icl/interval_map.hpp> +#include <array> +#include <atomic> #include "common/common_types.h" #include "video_core/rasterizer_interface.h" @@ -21,17 +21,28 @@ public: explicit RasterizerAccelerated(Core::Memory::Memory& cpu_memory_); ~RasterizerAccelerated() override; - void UpdatePagesCachedCount(VAddr addr, u64 size, bool cache) override; + void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override; private: - using PageIndex = VAddr; - using PageReferenceCount = u16; + class CacheEntry final { + public: + CacheEntry() = default; - using IntervalMap = boost::icl::interval_map<PageIndex, PageReferenceCount>; - using IntervalType = IntervalMap::interval_type; + std::atomic_uint16_t& Count(std::size_t page) { + return values[page & 3]; + } - IntervalMap map; - std::mutex map_lock; + const std::atomic_uint16_t& Count(std::size_t page) const { + return values[page & 3]; + } + + private: + std::array<std::atomic_uint16_t, 4> values{}; + }; + static_assert(sizeof(CacheEntry) == 8, "CacheEntry should be 8 bytes!"); + + using CachedPages = std::array<CacheEntry, 0x2000000>; + std::unique_ptr<CachedPages> cached_pages; Core::Memory::Memory& cpu_memory; }; diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index fd42d26b5..af1469147 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -162,7 +162,7 @@ public: } /// Increase/decrease the number of object in pages touching the specified region - virtual void UpdatePagesCachedCount(VAddr addr, u64 size, bool cache) {} + virtual void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {} /// Initialize disk cached resources for the game being emulated virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading, diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp index a109f9cbe..e81cd031b 100644 --- a/src/video_core/shader_cache.cpp +++ b/src/video_core/shader_cache.cpp @@ -132,7 +132,7 @@ void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t storage.push_back(std::move(data)); - rasterizer.UpdatePagesCachedCount(addr, size, true); + rasterizer.UpdatePagesCachedCount(addr, size, 1); } void ShaderCache::InvalidatePagesInRegion(VAddr addr, size_t size) { @@ -209,7 +209,7 @@ void ShaderCache::UnmarkMemory(Entry* entry) { const VAddr addr = entry->addr_start; const size_t size = entry->addr_end - addr; - rasterizer.UpdatePagesCachedCount(addr, size, false); + rasterizer.UpdatePagesCachedCount(addr, size, -1); } void ShaderCache::RemoveShadersFromStorage(std::span<ShaderInfo*> removed_shaders) { diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index d7941f6a4..0d5a1709f 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -2080,7 +2080,7 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) { ASSERT(False(image.flags & ImageFlagBits::Tracked)); image.flags |= ImageFlagBits::Tracked; if (False(image.flags & ImageFlagBits::Sparse)) { - rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, true); + rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1); return; } if (True(image.flags & ImageFlagBits::Registered)) { @@ -2091,13 +2091,13 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) { const auto& map = slot_map_views[map_view_id]; const VAddr cpu_addr = map.cpu_addr; const std::size_t size = map.size; - rasterizer.UpdatePagesCachedCount(cpu_addr, size, true); + rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1); } return; } ForEachSparseSegment(image, [this]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) { - rasterizer.UpdatePagesCachedCount(cpu_addr, size, true); + rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1); }); } @@ -2106,7 +2106,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) { ASSERT(True(image.flags & ImageFlagBits::Tracked)); image.flags &= ~ImageFlagBits::Tracked; if (False(image.flags & ImageFlagBits::Sparse)) { - rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, false); + rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1); return; } ASSERT(True(image.flags & ImageFlagBits::Registered)); @@ -2117,7 +2117,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) { const auto& map = slot_map_views[map_view_id]; const VAddr cpu_addr = map.cpu_addr; const std::size_t size = map.size; - rasterizer.UpdatePagesCachedCount(cpu_addr, size, false); + rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1); } } |