Aphcity/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity Actions

texture_cache: Implement L1_Inner_cache

Browse source
This commit is contained in:
Fernando Sahmkow 2019-05-10 17:59:18 -04:00 committed by ReinUsesLisp
parent 345e73f2fe
commit a4a58be2d4
1 changed files with 30 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

43
src/video_core/texture_cache/texture_cache.h
View file

@ -395,6 +395,26 @@ private:
const auto host_ptr{memory_manager->GetPointer(gpu_addr)}; const auto host_ptr{memory_manager->GetPointer(gpu_addr)};
const auto cache_addr{ToCacheAddr(host_ptr)}; 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(); const std::size_t candidate_size = params.GetGuestSizeInBytes();
auto overlaps{GetSurfacesInRegion(cache_addr, candidate_size)}; auto overlaps{GetSurfacesInRegion(cache_addr, candidate_size)};
@ -410,17 +430,6 @@ private:
if (overlaps.size() == 1) { if (overlaps.size() == 1) {
TSurface current_surface = overlaps[0]; 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)) { if (!current_surface->IsInside(gpu_addr, gpu_addr + candidate_size)) {
return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, false); return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, false);
} }
@ -473,8 +482,10 @@ private:
} }
void RegisterInnerCache(TSurface& surface) { 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; const CacheAddr end = (surface->GetCacheAddrEnd() - 1) >> registry_page_bits;
l1_cache[cache_addr] = surface;
while (start <= end) { while (start <= end) {
registry[start].push_back(surface); registry[start].push_back(surface);
start++; start++;
@ -482,8 +493,10 @@ private:
} }
void UnregisterInnerCache(TSurface& surface) { 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; const CacheAddr end = (surface->GetCacheAddrEnd() - 1) >> registry_page_bits;
l1_cache.erase(cache_addr);
while (start <= end) { while (start <= end) {
auto& reg{registry[start]}; auto& reg{registry[start]};
reg.erase(std::find(reg.begin(), reg.end(), surface)); reg.erase(std::find(reg.begin(), reg.end(), surface));
@ -559,6 +572,10 @@ private:
static constexpr u64 registry_page_size{1 << registry_page_bits}; static constexpr u64 registry_page_size{1 << registry_page_bits};
std::unordered_map<CacheAddr, std::vector<TSurface>> registry; 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 /// 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 /// previously been used. This is to prevent surfaces from being constantly created and
/// destroyed when used with different surface parameters. /// destroyed when used with different surface parameters.