OpenPunk/yuzu
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge pull request #10942 from FernandoS27/android-is-a-pain-in-the-a--

Browse Source 
Memory Tracking: Add mechanism to register small writes when gpu page is contested by GPU
This commit is contained in:
liamwhite
2023-07-02 11:29:01 -04:00
committed by GitHub
parent 87080e71c5 0e6b559c98
commit eaa62aee98
20 changed files with 329 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
src/core/core.cpp
View File

@@ -27,6 +27,7 @@
#include "core/file_sys/savedata_factory.h"
#include "core/file_sys/vfs_concat.h"
#include "core/file_sys/vfs_real.h"
#include "core/gpu_dirty_memory_manager.h"
#include "core/hid/hid_core.h"
#include "core/hle/kernel/k_memory_manager.h"
#include "core/hle/kernel/k_process.h"
@@ -130,7 +131,10 @@ FileSys::VirtualFile GetGameFileFromPath(const FileSys::VirtualFilesystem& vfs,
struct System::Impl {
explicit Impl(System& system)
: kernel{system}, fs_controller{system}, memory{system}, hid_core{}, room_network{},
cpu_manager{system}, reporter{system}, applet_manager{system}, time_manager{system} {}
cpu_manager{system}, reporter{system}, applet_manager{system}, time_manager{system},
gpu_dirty_memory_write_manager{} {
memory.SetGPUDirtyManagers(gpu_dirty_memory_write_manager);
}
void Initialize(System& system) {
device_memory = std::make_unique<Core::DeviceMemory>();
@@ -234,6 +238,8 @@ struct System::Impl {
// Setting changes may require a full system reinitialization (e.g., disabling multicore).
ReinitializeIfNecessary(system);
memory.SetGPUDirtyManagers(gpu_dirty_memory_write_manager);
kernel.Initialize();
cpu_manager.Initialize();
@@ -540,6 +546,9 @@ struct System::Impl {
std::array<u64, Core::Hardware::NUM_CPU_CORES> dynarmic_ticks{};
std::array<MicroProfileToken, Core::Hardware::NUM_CPU_CORES> microprofile_cpu{};
std::array<Core::GPUDirtyMemoryManager, Core::Hardware::NUM_CPU_CORES>
gpu_dirty_memory_write_manager{};
};
System::System() : impl{std::make_unique<Impl>(*this)} {}
@@ -629,10 +638,31 @@ void System::PrepareReschedule(const u32 core_index) {
impl->kernel.PrepareReschedule(core_index);
}
Core::GPUDirtyMemoryManager& System::CurrentGPUDirtyMemoryManager() {
const std::size_t core = impl->kernel.GetCurrentHostThreadID();
return impl->gpu_dirty_memory_write_manager[core < Core::Hardware::NUM_CPU_CORES
? core
: Core::Hardware::NUM_CPU_CORES - 1];
}
/// Provides a constant reference to the current gou dirty memory manager.
const Core::GPUDirtyMemoryManager& System::CurrentGPUDirtyMemoryManager() const {
const std::size_t core = impl->kernel.GetCurrentHostThreadID();
return impl->gpu_dirty_memory_write_manager[core < Core::Hardware::NUM_CPU_CORES
? core
: Core::Hardware::NUM_CPU_CORES - 1];
}
size_t System::GetCurrentHostThreadID() const {
return impl->kernel.GetCurrentHostThreadID();
}
void System::GatherGPUDirtyMemory(std::function<void(VAddr, size_t)>& callback) {
for (auto& manager : impl->gpu_dirty_memory_write_manager) {
manager.Gather(callback);
}
}
PerfStatsResults System::GetAndResetPerfStats() {
return impl->GetAndResetPerfStats();
}

11
src/core/core.h
View File

@@ -108,9 +108,10 @@ class CpuManager;
class Debugger;
class DeviceMemory;
class ExclusiveMonitor;
class SpeedLimiter;
class GPUDirtyMemoryManager;
class PerfStats;
class Reporter;
class SpeedLimiter;
class TelemetrySession;
struct PerfStatsResults;
@@ -225,6 +226,14 @@ public:
/// Prepare the core emulation for a reschedule
void PrepareReschedule(u32 core_index);
/// Provides a reference to the gou dirty memory manager.
[[nodiscard]] Core::GPUDirtyMemoryManager& CurrentGPUDirtyMemoryManager();
/// Provides a constant reference to the current gou dirty memory manager.
[[nodiscard]] const Core::GPUDirtyMemoryManager& CurrentGPUDirtyMemoryManager() const;
void GatherGPUDirtyMemory(std::function<void(VAddr, size_t)>& callback);
[[nodiscard]] size_t GetCurrentHostThreadID() const;
/// Gets and resets core performance statistics

122
src/core/gpu_dirty_memory_manager.h Normal file
View File

@@ -0,0 +1,122 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <atomic>
#include <bit>
#include <functional>
#include <mutex>
#include <utility>
#include <vector>
#include "core/memory.h"
namespace Core {
class GPUDirtyMemoryManager {
public:
GPUDirtyMemoryManager() : current{default_transform} {
back_buffer.reserve(256);
front_buffer.reserve(256);
}
~GPUDirtyMemoryManager() = default;
void Collect(VAddr address, size_t size) {
TransformAddress t = BuildTransform(address, size);
TransformAddress tmp, original;
do {
tmp = current.load(std::memory_order_acquire);
original = tmp;
if (tmp.address != t.address) {
if (IsValid(tmp.address)) {
std::scoped_lock lk(guard);
back_buffer.emplace_back(tmp);
current.exchange(t, std::memory_order_relaxed);
return;
}
tmp.address = t.address;
tmp.mask = 0;
}
if ((tmp.mask | t.mask) == tmp.mask) {
return;
}
tmp.mask |= t.mask;
} while (!current.compare_exchange_weak(original, tmp, std::memory_order_release,
std::memory_order_relaxed));
}
void Gather(std::function<void(VAddr, size_t)>& callback) {
{
std::scoped_lock lk(guard);
TransformAddress t = current.exchange(default_transform, std::memory_order_relaxed);
front_buffer.swap(back_buffer);
if (IsValid(t.address)) {
front_buffer.emplace_back(t);
}
}
for (auto& transform : front_buffer) {
size_t offset = 0;
u64 mask = transform.mask;
while (mask != 0) {
const size_t empty_bits = std::countr_zero(mask);
offset += empty_bits << align_bits;
mask = mask >> empty_bits;
const size_t continuous_bits = std::countr_one(mask);
callback((static_cast<VAddr>(transform.address) << page_bits) + offset,
continuous_bits << align_bits);
mask = continuous_bits < align_size ? (mask >> continuous_bits) : 0;
offset += continuous_bits << align_bits;
}
}
front_buffer.clear();
}
private:
struct alignas(8) TransformAddress {
u32 address;
u32 mask;
};
constexpr static size_t page_bits = Memory::YUZU_PAGEBITS - 1;
constexpr static size_t page_size = 1ULL << page_bits;
constexpr static size_t page_mask = page_size - 1;
constexpr static size_t align_bits = 6U;
constexpr static size_t align_size = 1U << align_bits;
constexpr static size_t align_mask = align_size - 1;
constexpr static TransformAddress default_transform = {.address = ~0U, .mask = 0U};
bool IsValid(VAddr address) {
return address < (1ULL << 39);
}
template <typename T>
T CreateMask(size_t top_bit, size_t minor_bit) {
T mask = ~T(0);
mask <<= (sizeof(T) * 8 - top_bit);
mask >>= (sizeof(T) * 8 - top_bit);
mask >>= minor_bit;
mask <<= minor_bit;
return mask;
}
TransformAddress BuildTransform(VAddr address, size_t size) {
const size_t minor_address = address & page_mask;
const size_t minor_bit = minor_address >> align_bits;
const size_t top_bit = (minor_address + size + align_mask) >> align_bits;
TransformAddress result{};
result.address = static_cast<u32>(address >> page_bits);
result.mask = CreateMask<u32>(top_bit, minor_bit);
return result;
}
std::atomic<TransformAddress> current{};
std::mutex guard;
std::vector<TransformAddress> back_buffer;
std::vector<TransformAddress> front_buffer;
};
} // namespace Core

40
src/core/memory.cpp
View File

@@ -3,6 +3,7 @@
#include <algorithm>
#include <cstring>
#include <span>
#include "common/assert.h"
#include "common/atomic_ops.h"
@@ -13,6 +14,7 @@
#include "common/swap.h"
#include "core/core.h"
#include "core/device_memory.h"
#include "core/gpu_dirty_memory_manager.h"
#include "core/hardware_properties.h"
#include "core/hle/kernel/k_page_table.h"
#include "core/hle/kernel/k_process.h"
@@ -678,7 +680,7 @@ struct Memory::Impl {
LOG_ERROR(HW_Memory, "Unmapped Write{} @ 0x{:016X} = 0x{:016X}", sizeof(T) * 8,
GetInteger(vaddr), static_cast<u64>(data));
},
[&]() { system.GPU().InvalidateRegion(GetInteger(vaddr), sizeof(T)); });
[&]() { HandleRasterizerWrite(GetInteger(vaddr), sizeof(T)); });
if (ptr) {
std::memcpy(ptr, &data, sizeof(T));
}
@@ -692,7 +694,7 @@ struct Memory::Impl {
LOG_ERROR(HW_Memory, "Unmapped WriteExclusive{} @ 0x{:016X} = 0x{:016X}",
sizeof(T) * 8, GetInteger(vaddr), static_cast<u64>(data));
},
[&]() { system.GPU().InvalidateRegion(GetInteger(vaddr), sizeof(T)); });
[&]() { HandleRasterizerWrite(GetInteger(vaddr), sizeof(T)); });
if (ptr) {
const auto volatile_pointer = reinterpret_cast<volatile T*>(ptr);
return Common::AtomicCompareAndSwap(volatile_pointer, data, expected);
@@ -707,7 +709,7 @@ struct Memory::Impl {
LOG_ERROR(HW_Memory, "Unmapped WriteExclusive128 @ 0x{:016X} = 0x{:016X}{:016X}",
GetInteger(vaddr), static_cast<u64>(data[1]), static_cast<u64>(data[0]));
},
[&]() { system.GPU().InvalidateRegion(GetInteger(vaddr), sizeof(u128)); });
[&]() { HandleRasterizerWrite(GetInteger(vaddr), sizeof(u128)); });
if (ptr) {
const auto volatile_pointer = reinterpret_cast<volatile u64*>(ptr);
return Common::AtomicCompareAndSwap(volatile_pointer, data, expected);
@@ -717,7 +719,7 @@ struct Memory::Impl {
void HandleRasterizerDownload(VAddr address, size_t size) {
const size_t core = system.GetCurrentHostThreadID();
auto& current_area = rasterizer_areas[core];
auto& current_area = rasterizer_read_areas[core];
const VAddr end_address = address + size;
if (current_area.start_address <= address && end_address <= current_area.end_address)
[[likely]] {
@@ -726,9 +728,31 @@ struct Memory::Impl {
current_area = system.GPU().OnCPURead(address, size);
}
Common::PageTable* current_page_table = nullptr;
std::array<VideoCore::RasterizerDownloadArea, Core::Hardware::NUM_CPU_CORES> rasterizer_areas{};
void HandleRasterizerWrite(VAddr address, size_t size) {
const size_t core = system.GetCurrentHostThreadID();
auto& current_area = rasterizer_write_areas[core];
VAddr subaddress = address >> YUZU_PAGEBITS;
bool do_collection = current_area.last_address == subaddress;
if (!do_collection) [[unlikely]] {
do_collection = system.GPU().OnCPUWrite(address, size);
if (!do_collection) {
return;
}
current_area.last_address = subaddress;
}
gpu_dirty_managers[core].Collect(address, size);
}
struct GPUDirtyState {
VAddr last_address;
};
Core::System& system;
Common::PageTable* current_page_table = nullptr;
std::array<VideoCore::RasterizerDownloadArea, Core::Hardware::NUM_CPU_CORES>
rasterizer_read_areas{};
std::array<GPUDirtyState, Core::Hardware::NUM_CPU_CORES> rasterizer_write_areas{};
std::span<Core::GPUDirtyMemoryManager> gpu_dirty_managers;
};
Memory::Memory(Core::System& system_) : system{system_} {
@@ -876,6 +900,10 @@ void Memory::ZeroBlock(Common::ProcessAddress dest_addr, const std::size_t size)
impl->ZeroBlock(*system.ApplicationProcess(), dest_addr, size);
}
void Memory::SetGPUDirtyManagers(std::span<Core::GPUDirtyMemoryManager> managers) {
impl->gpu_dirty_managers = managers;
}
Result Memory::InvalidateDataCache(Common::ProcessAddress dest_addr, const std::size_t size) {
return impl->InvalidateDataCache(*system.ApplicationProcess(), dest_addr, size);
}

6
src/core/memory.h
View File

@@ -5,6 +5,7 @@
#include <cstddef>
#include <memory>
#include <span>
#include <string>
#include "common/typed_address.h"
#include "core/hle/result.h"
@@ -15,7 +16,8 @@ struct PageTable;
namespace Core {
class System;
}
class GPUDirtyMemoryManager;
} // namespace Core
namespace Kernel {
class PhysicalMemory;
@@ -458,6 +460,8 @@ public:
*/
void MarkRegionDebug(Common::ProcessAddress vaddr, u64 size, bool debug);
void SetGPUDirtyManagers(std::span<Core::GPUDirtyMemoryManager> managers);
private:
Core::System& system;

39
src/video_core/buffer_cache/buffer_cache.h
View File

@@ -115,7 +115,34 @@ void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) {
template <class P>
void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) {
memory_tracker.CachedCpuWrite(cpu_addr, size);
const bool is_dirty = IsRegionRegistered(cpu_addr, size);
if (!is_dirty) {
return;
}
VAddr aligned_start = Common::AlignDown(cpu_addr, YUZU_PAGESIZE);
VAddr aligned_end = Common::AlignUp(cpu_addr + size, YUZU_PAGESIZE);
if (!IsRegionGpuModified(aligned_start, aligned_end - aligned_start)) {
WriteMemory(cpu_addr, size);
return;
}
tmp_buffer.resize_destructive(size);
cpu_memory.ReadBlockUnsafe(cpu_addr, tmp_buffer.data(), size);
InlineMemoryImplementation(cpu_addr, size, tmp_buffer);
}
template <class P>
bool BufferCache<P>::OnCPUWrite(VAddr cpu_addr, u64 size) {
const bool is_dirty = IsRegionRegistered(cpu_addr, size);
if (!is_dirty) {
return false;
}
if (memory_tracker.IsRegionGpuModified(cpu_addr, size)) {
return true;
}
WriteMemory(cpu_addr, size);
return false;
}
template <class P>
@@ -1553,6 +1580,14 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size,
return false;
}
InlineMemoryImplementation(dest_address, copy_size, inlined_buffer);
return true;
}
template <class P>
void BufferCache<P>::InlineMemoryImplementation(VAddr dest_address, size_t copy_size,
std::span<const u8> inlined_buffer) {
const IntervalType subtract_interval{dest_address, dest_address + copy_size};
ClearDownload(subtract_interval);
common_ranges.subtract(subtract_interval);
@@ -1574,8 +1609,6 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size,
} else {
buffer.ImmediateUpload(buffer.Offset(dest_address), inlined_buffer.first(copy_size));
}
return true;
}
template <class P>

5
src/video_core/buffer_cache/buffer_cache_base.h
View File

@@ -245,6 +245,8 @@ public:
void CachedWriteMemory(VAddr cpu_addr, u64 size);
bool OnCPUWrite(VAddr cpu_addr, u64 size);
void DownloadMemory(VAddr cpu_addr, u64 size);
std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size);
@@ -543,6 +545,9 @@ private:
void ClearDownload(IntervalType subtract_interval);
void InlineMemoryImplementation(VAddr dest_address, size_t copy_size,
std::span<const u8> inlined_buffer);
VideoCore::RasterizerInterface& rasterizer;
Core::Memory::Memory& cpu_memory;

2
src/video_core/fence_manager.h
View File

@@ -69,7 +69,6 @@ public:
}
void SignalFence(std::function<void()>&& func) {
rasterizer.InvalidateGPUCache();
bool delay_fence = Settings::IsGPULevelHigh();
if constexpr (!can_async_check) {
TryReleasePendingFences<false>();
@@ -96,6 +95,7 @@ public:
guard.unlock();
cv.notify_all();
}
rasterizer.InvalidateGPUCache();
}
void SignalSyncPoint(u32 value) {

12
src/video_core/gpu.cpp
View File

@@ -95,7 +95,9 @@ struct GPU::Impl {
/// Synchronizes CPU writes with Host GPU memory.
void InvalidateGPUCache() {
rasterizer->InvalidateGPUCache();
std::function<void(VAddr, size_t)> callback_writes(
[this](VAddr address, size_t size) { rasterizer->OnCacheInvalidation(address, size); });
system.GatherGPUDirtyMemory(callback_writes);
}
/// Signal the ending of command list.
@@ -299,6 +301,10 @@ struct GPU::Impl {
gpu_thread.InvalidateRegion(addr, size);
}
bool OnCPUWrite(VAddr addr, u64 size) {
return rasterizer->OnCPUWrite(addr, size);
}
/// Notify rasterizer that any caches of the specified region should be flushed and invalidated
void FlushAndInvalidateRegion(VAddr addr, u64 size) {
gpu_thread.FlushAndInvalidateRegion(addr, size);
@@ -561,6 +567,10 @@ void GPU::InvalidateRegion(VAddr addr, u64 size) {
impl->InvalidateRegion(addr, size);
}
bool GPU::OnCPUWrite(VAddr addr, u64 size) {
return impl->OnCPUWrite(addr, size);
}
void GPU::FlushAndInvalidateRegion(VAddr addr, u64 size) {
impl->FlushAndInvalidateRegion(addr, size);
}

4
src/video_core/gpu.h
View File

@@ -250,6 +250,10 @@ public:
/// Notify rasterizer that any caches of the specified region should be invalidated
void InvalidateRegion(VAddr addr, u64 size);
/// Notify rasterizer that CPU is trying to write this area. It returns true if the area is
/// sensible, false otherwise
bool OnCPUWrite(VAddr addr, u64 size);
/// Notify rasterizer that any caches of the specified region should be flushed and invalidated
void FlushAndInvalidateRegion(VAddr addr, u64 size);

6
src/video_core/gpu_thread.cpp
View File

@@ -47,7 +47,7 @@ static void RunThread(std::stop_token stop_token, Core::System& system,
} else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) {
rasterizer->FlushRegion(flush->addr, flush->size);
} else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) {
rasterizer->OnCPUWrite(invalidate->addr, invalidate->size);
rasterizer->OnCacheInvalidation(invalidate->addr, invalidate->size);
} else {
ASSERT(false);
}
@@ -102,12 +102,12 @@ void ThreadManager::TickGPU() {
}
void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
rasterizer->OnCPUWrite(addr, size);
rasterizer->OnCacheInvalidation(addr, size);
}
void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
// Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important
rasterizer->OnCPUWrite(addr, size);
rasterizer->OnCacheInvalidation(addr, size);
}
u64 ThreadManager::PushCommand(CommandData&& command_data, bool block) {

4
src/video_core/rasterizer_interface.h
View File

@@ -109,7 +109,9 @@ public:
}
/// Notify rasterizer that any caches of the specified region are desync with guest
virtual void OnCPUWrite(VAddr addr, u64 size) = 0;
virtual void OnCacheInvalidation(VAddr addr, u64 size) = 0;
virtual bool OnCPUWrite(VAddr addr, u64 size) = 0;
/// Sync memory between guest and host.
virtual void InvalidateGPUCache() = 0;

5
src/video_core/renderer_null/null_rasterizer.cpp
View File

@@ -47,7 +47,10 @@ bool RasterizerNull::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheTyp
return false;
}
void RasterizerNull::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType) {}
void RasterizerNull::OnCPUWrite(VAddr addr, u64 size) {}
bool RasterizerNull::OnCPUWrite(VAddr addr, u64 size) {
return false;
}
void RasterizerNull::OnCacheInvalidation(VAddr addr, u64 size) {}
VideoCore::RasterizerDownloadArea RasterizerNull::GetFlushArea(VAddr addr, u64 size) {
VideoCore::RasterizerDownloadArea new_area{
.start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE),

3
src/video_core/renderer_null/null_rasterizer.h
View File

@@ -53,7 +53,8 @@ public:
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void InvalidateRegion(VAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void OnCPUWrite(VAddr addr, u64 size) override;
void OnCacheInvalidation(VAddr addr, u64 size) override;
bool OnCPUWrite(VAddr addr, u64 size) override;
VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;

35
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@@ -485,12 +485,33 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache
}
}
void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
bool RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
if (addr == 0 || size == 0) {
return false;
}
{
std::scoped_lock lock{buffer_cache.mutex};
if (buffer_cache.OnCPUWrite(addr, size)) {
return true;
}
}
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.WriteMemory(addr, size);
}
shader_cache.InvalidateRegion(addr, size);
return false;
}
void RasterizerOpenGL::OnCacheInvalidation(VAddr addr, u64 size) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
if (addr == 0 || size == 0) {
return;
}
shader_cache.OnCPUWrite(addr, size);
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.WriteMemory(addr, size);
@@ -499,15 +520,11 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.CachedWriteMemory(addr, size);
}
shader_cache.InvalidateRegion(addr, size);
}
void RasterizerOpenGL::InvalidateGPUCache() {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
shader_cache.SyncGuestHost();
{
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.FlushCachedWrites();
}
gpu.InvalidateGPUCache();
}
void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
@@ -519,7 +536,7 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.WriteMemory(addr, size);
}
shader_cache.OnCPUWrite(addr, size);
shader_cache.OnCacheInvalidation(addr, size);
}
void RasterizerOpenGL::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {

3
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@@ -98,7 +98,8 @@ public:
VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override;
void InvalidateRegion(VAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void OnCPUWrite(VAddr addr, u64 size) override;
void OnCacheInvalidation(VAddr addr, u64 size) override;
bool OnCPUWrite(VAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;
void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;

34
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -566,11 +566,32 @@ void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<VAddr, std::s
}
}
void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
bool RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return false;
}
{
std::scoped_lock lock{buffer_cache.mutex};
if (buffer_cache.OnCPUWrite(addr, size)) {
return true;
}
}
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.WriteMemory(addr, size);
}
pipeline_cache.InvalidateRegion(addr, size);
return false;
}
void RasterizerVulkan::OnCacheInvalidation(VAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return;
}
pipeline_cache.OnCPUWrite(addr, size);
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.WriteMemory(addr, size);
@@ -579,14 +600,11 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.CachedWriteMemory(addr, size);
}
pipeline_cache.InvalidateRegion(addr, size);
}
void RasterizerVulkan::InvalidateGPUCache() {
pipeline_cache.SyncGuestHost();
{
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.FlushCachedWrites();
}
gpu.InvalidateGPUCache();
}
void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) {
@@ -598,7 +616,7 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) {
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.WriteMemory(addr, size);
}
pipeline_cache.OnCPUWrite(addr, size);
pipeline_cache.OnCacheInvalidation(addr, size);
}
void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {

3
src/video_core/renderer_vulkan/vk_rasterizer.h
View File

@@ -96,7 +96,8 @@ public:
void InvalidateRegion(VAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) override;
void OnCPUWrite(VAddr addr, u64 size) override;
void OnCacheInvalidation(VAddr addr, u64 size) override;
bool OnCPUWrite(VAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;
void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;

2
src/video_core/shader_cache.cpp
View File

@@ -24,7 +24,7 @@ void ShaderCache::InvalidateRegion(VAddr addr, size_t size) {
RemovePendingShaders();
}
void ShaderCache::OnCPUWrite(VAddr addr, size_t size) {
void ShaderCache::OnCacheInvalidation(VAddr addr, size_t size) {
std::scoped_lock lock{invalidation_mutex};
InvalidatePagesInRegion(addr, size);
}

2
src/video_core/shader_cache.h
View File

@@ -62,7 +62,7 @@ public:
/// @brief Unmarks a memory region as cached and marks it for removal
/// @param addr Start address of the CPU write operation
/// @param size Number of bytes of the CPU write operation
void OnCPUWrite(VAddr addr, size_t size);
void OnCacheInvalidation(VAddr addr, size_t size);
/// @brief Flushes delayed removal operations
void SyncGuestHost();