raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

guest_gpu: remove memory-unmap submission deadlock + remove legacy agc buffering

Browse Source
This commit is contained in:
nmzik
2026-08-02 13:17:53 +02:00
parent 59b8fad341
commit fc8d2a3b83
6 changed files with 102 additions and 295 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/guest_gpu/graphicsRun.cpp
+5 -62
View File
@@ -32,11 +32,10 @@
namespace Libs::Graphics {
static thread_local CommandProcessor* g_current_processor = nullptr;
static thread_local Pm4Execution* g_current_execution = nullptr;
static thread_local uint32_t g_submission_pause_depth = 0;
static thread_local bool g_gpu_mutex_owned = false;
static thread_local bool g_gpu_thread = false;
static thread_local CommandProcessor* g_current_processor = nullptr;
static thread_local Pm4Execution* g_current_execution = nullptr;
static thread_local bool g_gpu_mutex_owned = false;
static thread_local bool g_gpu_thread = false;
class GpuMutexLock final {
public:
@@ -98,8 +97,6 @@ public:
bool trigger_agc_interrupt_on_done);
void SubmitFlipPreparation(uint64_t request_id);
void Done();
void PauseSubmissions();
void ResumeSubmissions();
void Shutdown();
[[nodiscard]] bool IsStopping();
void SendCommand(Common::UniqueFunction<void>&& command);
@@ -407,7 +404,7 @@ void CommandProcessor::WriteData(uint32_t* dst, const uint32_t* src, uint32_t dw
uint32_t write_control) {
const uint32_t dst_sel = ((write_control >> 30u) & 0x1u) | ((write_control >> 7u) & 0x1eu);
const uint32_t cache_policy = (write_control >> 25u) & 0x3u;
const uint32_t increment = (write_control >> 16u) & 0x1u;
const uint32_t increment = (write_control >> 16u) & 0x1u;
const uint32_t write_confirm = (write_control >> 20u) & 0x1u;
switch (dst_sel) {
@@ -696,26 +693,6 @@ bool GpuState::Process(Submission& submission) {
return complete;
}
void GpuState::PauseSubmissions() {
if (g_gpu_mutex_owned) {
EXIT("GPU submissions are already paused by this thread\n");
}
g_gpu_mutex_owned = true;
m_submission_mutex.Lock();
if (!IsGpuThread()) {
WaitLocked();
}
m_renderer.GetCommandScheduler().DrainPriorityOperations();
}
void GpuState::ResumeSubmissions() {
if (!g_gpu_mutex_owned) {
EXIT("GPU submissions resumed without an active pause\n");
}
m_submission_mutex.Unlock();
g_gpu_mutex_owned = false;
}
Pm4ProcessResult CommandProcessor::Process(Pm4Execution& execution, uint32_t* buffer,
uint32_t size_dw) {
KYTY_PROFILER_BLOCK("CommandProcessor::Process");
@@ -1693,32 +1670,6 @@ int Gpu::GetFrameNum() const {
return m_state->GetFrameNum();
}
void Gpu::PauseSubmissions() {
m_state->PauseSubmissions();
}
void Gpu::ResumeSubmissions() {
m_state->ResumeSubmissions();
}
Gpu::SubmissionLock::SubmissionLock(Gpu& gpu): m_gpu(gpu) {
if (g_current_processor != nullptr || g_submission_pause_depth == UINT32_MAX) {
EXIT("cannot acquire GPU submission lock in the current state\n");
}
if (g_submission_pause_depth++ == 0) {
m_gpu.PauseSubmissions();
}
}
Gpu::SubmissionLock::~SubmissionLock() {
if (g_submission_pause_depth == 0) {
EXIT("GPU submission lock released without ownership\n");
}
if (--g_submission_pause_depth == 0) {
m_gpu.ResumeSubmissions();
}
}
bool Gpu::IsCommandProcessorThread() noexcept {
return g_current_processor != nullptr;
}
@@ -1727,12 +1678,4 @@ CommandProcessor* Gpu::CurrentCommandProcessor() noexcept {
return g_current_processor;
}
bool Gpu::SubmissionLockHeld() noexcept {
return g_submission_pause_depth != 0;
}
bool Gpu::MutexHeld() noexcept {
return g_gpu_mutex_owned;
}
} // namespace Libs::Graphics
src/graphics/guest_gpu/graphicsRun.h
-17
View File
@@ -35,25 +35,8 @@ public:
[[nodiscard]] static bool IsCommandProcessorThread() noexcept;
[[nodiscard]] static CommandProcessor* CurrentCommandProcessor() noexcept;
[[nodiscard]] static bool SubmissionLockHeld() noexcept;
[[nodiscard]] static bool MutexHeld() noexcept;
class SubmissionLock final {
public:
explicit SubmissionLock(Gpu& gpu);
~SubmissionLock();
KYTY_CLASS_NO_COPY(SubmissionLock);
private:
Gpu& m_gpu;
};
private:
friend class SubmissionLock;
void PauseSubmissions();
void ResumeSubmissions();
std::unique_ptr<GpuState> m_state;
};
} // namespace Libs::Graphics
src/graphics/host_gpu/renderer/cache/gpuResourceManager.cpp
Vendored
+17 -5
View File
@@ -7,7 +7,8 @@
namespace Libs::Graphics {
GpuResourceManager::GpuResourceManager(GraphicContext& graphics, CommandScheduler& scheduler)
: m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache, m_resource_mutex),
: m_scheduler(scheduler),
m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache, m_resource_mutex),
m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache, m_resource_mutex) {}
GpuResourceManager::~GpuResourceManager() = default;
@@ -101,7 +102,22 @@ void GpuResourceManager::MapMemory(uint64_t vaddr, uint64_t size) {
}
void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
if (CommandScheduler::InDeferredOperation()) {
EXIT("unsupported memory unmap from an asynchronous GPU completion, "
"addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
vaddr, size);
}
if (m_resource_mutex.IsOwnedByCurrentThread()) {
EXIT("unsupported memory unmap from a pre-owned resource transaction, "
"addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
vaddr, size);
}
const auto unmap = [this, vaddr, size] {
if (m_scheduler.Active()) {
const auto tick = m_scheduler.CurrentTick();
m_scheduler.FinishCurrent();
m_scheduler.WaitPriorityOperations(tick);
}
m_buffer_cache.UnmapMemory(vaddr, size);
m_texture_cache.UnmapMemory(vaddr, size);
m_page_manager.OnGpuUnmap(vaddr, size);
@@ -109,13 +125,9 @@ void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
m_mapped_ranges.Subtract(vaddr, size);
};
if (m_gpu == nullptr) {
if (m_resource_mutex.IsOwnedByCurrentThread()) {
EXIT("cannot synchronously unmap from a resource transaction\n");
}
unmap();
return;
}
Gpu::SubmissionLock submissions(*m_gpu);
m_gpu->SendCommandSync(unmap);
}
src/graphics/host_gpu/renderer/cache/gpuResourceManager.h
Vendored
+1
View File
@@ -36,6 +36,7 @@ public:
private:
PageManager m_page_manager;
ResourceMutex m_resource_mutex;
CommandScheduler& m_scheduler;
BufferCache m_buffer_cache;
TextureCache m_texture_cache;
mutable std::shared_mutex m_mapped_ranges_mutex;