raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 11:26:56 +02:00
Code Issues Packages Projects Releases Wiki Activity

renderer: use LRU buffer cache collection

Browse Source
This commit is contained in:
nmzik
2026-08-24 00:17:02 +02:00
parent 11d4c0d2a2
commit 186621c6ff
3 changed files with 43 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/cache/bufferCache.cpp
Vendored
+21 -28
View File
@@ -51,7 +51,7 @@ struct BufferCache::DownloadCopy {
}; };
void BufferCache::Register(BufferId id) { void BufferCache::Register(BufferId id) {
const auto& buffer = m_slot_buffers[id]; auto& buffer = m_slot_buffers[id];
PageTable::PageRange pages {}; PageTable::PageRange pages {};
EXIT_IF(!PageTable::TryGetPageRange(buffer.CpuAddress(), buffer.Size(), pages)); EXIT_IF(!PageTable::TryGetPageRange(buffer.CpuAddress(), buffer.Size(), pages));
for (size_t page = pages.first; page < pages.last_exclusive; ++page) { for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
@@ -60,6 +60,7 @@ void BufferCache::Register(BufferId id) {
const auto [it, inserted] = m_buffers.emplace(buffer.CpuAddress(), id); const auto [it, inserted] = m_buffers.emplace(buffer.CpuAddress(), id);
(void)it; (void)it;
EXIT_IF(!inserted); EXIT_IF(!inserted);
buffer.lru_id = m_lru_cache.Insert(id, m_gc_tick);
m_total_used_memory += buffer.Size(); m_total_used_memory += buffer.Size();
} }
@@ -83,9 +84,16 @@ void BufferCache::Unregister(BufferId id) {
EXIT("BufferCache: allocation accounting underflow\n"); EXIT("BufferCache: allocation accounting underflow\n");
} }
m_total_used_memory -= buffer.Size(); m_total_used_memory -= buffer.Size();
m_lru_cache.Free(buffer.lru_id);
buffer.is_deleted = true; buffer.is_deleted = true;
} }
void BufferCache::TouchBuffer(const Buffer& buffer) {
if (!buffer.is_deleted) {
m_lru_cache.Touch(buffer.lru_id, m_gc_tick);
}
}
void BufferCache::DeleteBuffer(BufferId id) { void BufferCache::DeleteBuffer(BufferId id) {
auto* buffer = m_slot_buffers.try_get(id); auto* buffer = m_slot_buffers.try_get(id);
if (buffer == nullptr || buffer->is_deleted) { if (buffer == nullptr || buffer->is_deleted) {
@@ -364,7 +372,7 @@ BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
if (owner != nullptr && *owner) { if (owner != nullptr && *owner) {
auto& buffer = m_slot_buffers[*owner]; auto& buffer = m_slot_buffers[*owner];
if (buffer.IsInBounds(vaddr, size)) { if (buffer.IsInBounds(vaddr, size)) {
buffer.tick_accessed_last = m_gc_tick; TouchBuffer(buffer);
return *owner; return *owner;
} }
} }
@@ -393,8 +401,7 @@ BufferId BufferCache::CreateBuffer(uint64_t vaddr, uint64_t size) {
const auto id = m_slot_buffers.insert(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, begin, const auto id = m_slot_buffers.insert(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, begin,
AllFlags, end - begin); AllFlags, end - begin);
auto& buffer = m_slot_buffers[id]; auto& buffer = m_slot_buffers[id];
buffer.tick_accessed_last = m_gc_tick;
SetVulkanObjectNameF(m_graphics.device, buffer.Handle(), SetVulkanObjectNameF(m_graphics.device, buffer.Handle(),
"Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", begin, end - begin); "Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", begin, end - begin);
for (auto overlap = first; overlap != last;) { for (auto overlap = first; overlap != last;) {
@@ -509,6 +516,7 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBuffer(uint64_t vaddr, uint64_t
id = FindBuffer(vaddr, size); id = FindBuffer(vaddr, size);
buffer = &m_slot_buffers[id]; buffer = &m_slot_buffers[id];
} }
TouchBuffer(*buffer);
(void)SynchronizeBuffer(*buffer, vaddr, size, is_written, is_texel_buffer); (void)SynchronizeBuffer(*buffer, vaddr, size, is_written, is_texel_buffer);
if (is_written) { if (is_written) {
m_gpu_modified_ranges.Add(vaddr, size); m_gpu_modified_ranges.Add(vaddr, size);
@@ -545,7 +553,7 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, u
owner = &m_slot_buffers[FindBuffer(vaddr, size)]; owner = &m_slot_buffers[FindBuffer(vaddr, size)];
} }
if (owner != nullptr && !cpu_modified && (!gpu_modified || has_dirty_buffer_source)) { if (owner != nullptr && !cpu_modified && (!gpu_modified || has_dirty_buffer_source)) {
owner->tick_accessed_last = m_gc_tick; TouchBuffer(*owner);
return {owner, owner->Offset(vaddr)}; return {owner, owner->Offset(vaddr)};
} }
if (has_dirty_buffer_source && owner == nullptr) { if (has_dirty_buffer_source && owner == nullptr) {
@@ -570,7 +578,7 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, u
return {&m_staging_buffer, stage_offset}; return {&m_staging_buffer, stage_offset};
} }
owner->tick_accessed_last = m_gc_tick; TouchBuffer(*owner);
std::vector<std::pair<uint64_t, uint64_t>> uploads; std::vector<std::pair<uint64_t, uint64_t>> uploads;
m_memory_tracker.ForEachUploadRange( m_memory_tracker.ForEachUploadRange(
vaddr, size, false, vaddr, size, false,
@@ -603,7 +611,7 @@ void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data)
continue; continue;
} }
WriteDataBuffer(buffer, begin, data.data() + begin - vaddr, range_end - begin); WriteDataBuffer(buffer, begin, data.data() + begin - vaddr, range_end - begin);
buffer.tick_accessed_last = m_gc_tick; TouchBuffer(buffer);
} }
} }
@@ -743,30 +751,17 @@ void BufferCache::RunGarbageCollector() {
const uint64_t age = std::min<uint64_t>(aggressive ? 80 : 160, tick); const uint64_t age = std::min<uint64_t>(aggressive ? 80 : 160, tick);
const size_t limit = aggressive ? 64 : 32; const size_t limit = aggressive ? 64 : 32;
std::vector<uint64_t> candidates;
for (const auto& [address, id]: m_buffers) {
const auto& buffer = m_slot_buffers[id];
if (tick - std::min(tick, buffer.tick_accessed_last) >= age) {
candidates.push_back(address);
}
}
std::ranges::sort(candidates, [&](uint64_t left, uint64_t right) {
return m_slot_buffers[m_buffers.at(left)].tick_accessed_last <
m_slot_buffers[m_buffers.at(right)].tick_accessed_last;
});
std::vector<BufferId> dirty_buffers; std::vector<BufferId> dirty_buffers;
std::vector<DownloadCopy> copies; std::vector<DownloadCopy> copies;
size_t retire_count = 0; size_t retire_count = 0;
for (const auto address: candidates) { m_lru_cache.ForEachItemBelow(tick - age, [&](BufferId id) {
auto found = m_buffers.find(address); auto& buffer = m_slot_buffers[id];
EXIT_IF(found == m_buffers.end()); EXIT_IF(buffer.is_deleted);
const auto id = found->second;
auto& buffer = m_slot_buffers[id];
m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, buffer.CpuAddress(), m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, buffer.CpuAddress(),
buffer.Size(), "garbage collection"); buffer.Size(), "garbage collection");
const bool dirty = m_memory_tracker.IsRegionGpuModified(buffer.CpuAddress(), buffer.Size()); const bool dirty = m_memory_tracker.IsRegionGpuModified(buffer.CpuAddress(), buffer.Size());
if (dirty && !aggressive) { if (dirty && !aggressive) {
continue; return false;
} }
if (dirty) { if (dirty) {
m_memory_tracker.ForEachDownloadRange<false>( m_memory_tracker.ForEachDownloadRange<false>(
@@ -787,10 +782,8 @@ void BufferCache::RunGarbageCollector() {
m_memory_tracker.UntrackMemory(buffer.CpuAddress(), buffer.Size()); m_memory_tracker.UntrackMemory(buffer.CpuAddress(), buffer.Size());
DeleteBuffer(id); DeleteBuffer(id);
} }
if (++retire_count == limit) { return ++retire_count == limit;
break; });
}
}
if (dirty_buffers.empty()) { if (dirty_buffers.empty()) {
return; return;
} }
src/graphics/host_gpu/renderer/cache/bufferCache.h
Vendored
+20 -17
View File
@@ -3,6 +3,7 @@
#include "common/abi.h" #include "common/abi.h"
#include "common/common.h" #include "common/common.h"
#include "common/lruCache.h"
#include "common/slotVector.h" #include "common/slotVector.h"
#include "graphics/host_gpu/memoryTracker.h" #include "graphics/host_gpu/memoryTracker.h"
#include "graphics/host_gpu/rangeSet.h" #include "graphics/host_gpu/rangeSet.h"
@@ -74,6 +75,7 @@ private:
} }
[[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy); [[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy);
void WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source, uint64_t size); void WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source, uint64_t size);
void TouchBuffer(const Buffer& buffer);
[[nodiscard]] BufferId CreateBuffer(uint64_t vaddr, uint64_t size); [[nodiscard]] BufferId CreateBuffer(uint64_t vaddr, uint64_t size);
void Register(BufferId id); void Register(BufferId id);
void Unregister(BufferId id); void Unregister(BufferId id);
@@ -87,23 +89,24 @@ private:
void WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data); void WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data);
void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write); void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write);
GraphicContext& m_graphics; GraphicContext& m_graphics;
CommandScheduler& m_scheduler; CommandScheduler& m_scheduler;
Buffer m_gds_buffer; Buffer m_gds_buffer;
Common::SlotVector<Buffer> m_slot_buffers; Common::SlotVector<Buffer> m_slot_buffers;
std::map<uint64_t, BufferId> m_buffers; Common::LeastRecentlyUsedCache<BufferId, uint64_t> m_lru_cache;
PageTable m_page_table; std::map<uint64_t, BufferId> m_buffers;
RangeSet m_gpu_modified_ranges; PageTable m_page_table;
MemoryTracker m_memory_tracker; RangeSet m_gpu_modified_ranges;
StreamBuffer m_staging_buffer; MemoryTracker m_memory_tracker;
StreamBuffer m_stream_buffer; StreamBuffer m_staging_buffer;
StreamBuffer m_download_buffer; StreamBuffer m_stream_buffer;
StreamBuffer m_device_buffer; StreamBuffer m_download_buffer;
TextureCache& m_texture_cache; StreamBuffer m_device_buffer;
uint64_t m_total_used_memory = 0; TextureCache& m_texture_cache;
uint64_t m_trigger_gc_memory = 1ull * 1024 * 1024 * 1024; uint64_t m_total_used_memory = 0;
uint64_t m_critical_gc_memory = 2ull * 1024 * 1024 * 1024; uint64_t m_trigger_gc_memory = 1ull * 1024 * 1024 * 1024;
uint64_t m_gc_tick = 0; uint64_t m_critical_gc_memory = 2ull * 1024 * 1024 * 1024;
uint64_t m_gc_tick = 0;
}; };
} // namespace Libs::Graphics } // namespace Libs::Graphics
src/graphics/host_gpu/renderer/cache/streamBuffer.h
Vendored
+2 -2
View File
@@ -66,8 +66,8 @@ public:
void Fill(uint64_t offset, uint64_t size, uint32_t value); void Fill(uint64_t offset, uint64_t size, uint32_t value);
// BufferCache state lives directly on the resource. // BufferCache state lives directly on the resource.
bool is_deleted = false; bool is_deleted = false;
uint64_t tick_accessed_last = 0; size_t lru_id = 0;
protected: protected:
[[nodiscard]] GraphicContext& Graphics() const noexcept { return *m_graphics; } [[nodiscard]] GraphicContext& Graphics() const noexcept { return *m_graphics; }