summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorFernando Sahmkow <fsahmkow27@gmail.com>2019-05-10 23:59:18 +0200
committerReinUsesLisp <reinuseslisp@airmail.cc>2019-06-21 02:36:12 +0200
commita4a58be2d46e95df4cead2916b6efbd658a0deaa (patch)
treefe1d66d69101467c47167af9c66f17a95d961ed7
parentvideo_core: Use un-shifted block sizes to avoid integer divisions (diff)
downloadyuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.tar
yuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.tar.gz
yuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.tar.bz2
yuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.tar.lz
yuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.tar.xz
yuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.tar.zst
yuzu-a4a58be2d46e95df4cead2916b6efbd658a0deaa.zip
Diffstat (limited to '')
-rw-r--r--src/video_core/texture_cache/texture_cache.h43
1 files changed, 30 insertions, 13 deletions
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index f35d0c88f..ad0fbd7ce 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -395,6 +395,26 @@ private:
const auto host_ptr{memory_manager->GetPointer(gpu_addr)};
const auto cache_addr{ToCacheAddr(host_ptr)};
+
+ if (l1_cache.count(cache_addr) > 0) {
+ TSurface current_surface = l1_cache[cache_addr];
+ if (!current_surface->MatchesTopology(params)) {
+ std::vector<TSurface> overlaps{current_surface};
+ return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, true);
+ }
+ MatchStructureResult s_result = current_surface->MatchesStructure(params);
+ if (s_result != MatchStructureResult::None &&
+ current_surface->GetGpuAddr() == gpu_addr &&
+ (params.target != SurfaceTarget::Texture3D ||
+ current_surface->MatchTarget(params.target))) {
+ if (s_result == MatchStructureResult::FullMatch) {
+ return ManageStructuralMatch(current_surface, params);
+ } else {
+ return RebuildSurface(current_surface, params);
+ }
+ }
+ }
+
const std::size_t candidate_size = params.GetGuestSizeInBytes();
auto overlaps{GetSurfacesInRegion(cache_addr, candidate_size)};
@@ -410,17 +430,6 @@ private:
if (overlaps.size() == 1) {
TSurface current_surface = overlaps[0];
- MatchStructureResult s_result = current_surface->MatchesStructure(params);
- if (s_result != MatchStructureResult::None &&
- current_surface->GetGpuAddr() == gpu_addr &&
- (params.target != SurfaceTarget::Texture3D ||
- current_surface->MatchTarget(params.target))) {
- if (s_result == MatchStructureResult::FullMatch) {
- return ManageStructuralMatch(current_surface, params);
- } else {
- return RebuildSurface(current_surface, params);
- }
- }
if (!current_surface->IsInside(gpu_addr, gpu_addr + candidate_size)) {
return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, false);
}
@@ -473,8 +482,10 @@ private:
}
void RegisterInnerCache(TSurface& surface) {
- CacheAddr start = surface->GetCacheAddr() >> registry_page_bits;
+ const CacheAddr cache_addr = surface->GetCacheAddr();
+ CacheAddr start = cache_addr >> registry_page_bits;
const CacheAddr end = (surface->GetCacheAddrEnd() - 1) >> registry_page_bits;
+ l1_cache[cache_addr] = surface;
while (start <= end) {
registry[start].push_back(surface);
start++;
@@ -482,8 +493,10 @@ private:
}
void UnregisterInnerCache(TSurface& surface) {
- CacheAddr start = surface->GetCacheAddr() >> registry_page_bits;
+ const CacheAddr cache_addr = surface->GetCacheAddr();
+ CacheAddr start = cache_addr >> registry_page_bits;
const CacheAddr end = (surface->GetCacheAddrEnd() - 1) >> registry_page_bits;
+ l1_cache.erase(cache_addr);
while (start <= end) {
auto& reg{registry[start]};
reg.erase(std::find(reg.begin(), reg.end(), surface));
@@ -559,6 +572,10 @@ private:
static constexpr u64 registry_page_size{1 << registry_page_bits};
std::unordered_map<CacheAddr, std::vector<TSurface>> registry;
+ // The L1 Cache is used for fast texture lookup before checking the overlaps
+ // This avoids calculating size and other stuffs.
+ std::unordered_map<CacheAddr, TSurface> l1_cache;
+
/// The surface reserve is a "backup" cache, this is where we put unique surfaces that have
/// previously been used. This is to prevent surfaces from being constantly created and
/// destroyed when used with different surface parameters.