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

gpu: queue AGC suspend points

Browse Source
This commit is contained in:
nmzik
2026-08-07 21:02:31 +02:00
parent d4f8338937
commit 90bf3c9236
4 changed files with 81 additions and 72 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/guest_gpu/graphicsRun.cpp
+29 -49
View File
@@ -97,7 +97,7 @@ public:
void SubmitCompute(uint32_t queue, uint32_t* cmd_buffer, uint32_t num_dw,
bool trigger_agc_interrupt_on_done);
void SubmitFlipPreparation(uint64_t request_id);
void Done();
void SuspendPoint();
void Shutdown();
[[nodiscard]] bool IsStopping();
void SendCommand(Common::UniqueFunction<void>&& command);
@@ -106,7 +106,7 @@ public:
[[nodiscard]] static bool IsGpuThread() noexcept { return g_gpu_thread; }
private:
enum class SubmissionType { Graphics, Compute, FlipPreparation };
enum class SubmissionType { Graphics, Compute, FlipPreparation, SuspendPoint };
struct Submission {
SubmissionType type = SubmissionType::Graphics;
@@ -124,9 +124,7 @@ private:
uint64_t flip_request_id = 0;
};
void WaitLocked();
void Enqueue(Submission submission);
void WaitForIdle();
void ProcessCommands();
bool Process(Submission& submission);
static void ThreadRun(void* data);
@@ -137,17 +135,16 @@ private:
Common::Mutex m_queue_mutex;
std::mutex m_shutdown_mutex;
Common::CondVar m_work_available;
Common::CondVar m_idle;
std::array<std::deque<Submission>, QueueCount> m_queues;
std::deque<Common::UniqueFunction<void>> m_commands;
std::atomic_uint32_t m_pending_commands {0};
uint32_t m_next_queue = 0;
uint32_t m_submission_count = 0;
bool m_processing = false;
bool m_graphics_done = true;
bool m_reset_graphics = true;
bool m_accepting = true;
bool m_stopping = false;
bool m_shutdown_complete = false;
std::binary_semaphore m_suspend_point_slot {1};
std::unique_ptr<CommandProcessor> m_gfx_cp;
std::array<std::unique_ptr<CommandProcessor>, ComputeQueueCount> m_compute_cp;
@@ -241,8 +238,8 @@ void GpuState::Submit(uint32_t* cmd_draw_buffer, uint32_t num_draw_dw, uint32_t*
submission.commands = OwnedCmdBuffer(cmd_draw_buffer, num_draw_dw);
submission.constant_commands = OwnedCmdBuffer(cmd_const_buffer, num_const_dw);
submission.trigger_agc_interrupt_on_done = trigger_agc_interrupt_on_done;
submission.reset_processor = m_graphics_done;
m_graphics_done = false;
submission.reset_processor = m_reset_graphics;
m_reset_graphics = false;
Enqueue(std::move(submission));
}
@@ -268,32 +265,27 @@ void GpuState::SubmitFlipPreparation(uint64_t request_id) {
Submission submission;
submission.type = SubmissionType::FlipPreparation;
submission.queue_id = 0;
submission.reset_processor = m_graphics_done;
submission.reset_processor = m_reset_graphics;
submission.flip_request_id = request_id;
m_graphics_done = false;
m_reset_graphics = false;
Enqueue(std::move(submission));
}
void GpuState::Done() {
void GpuState::SuspendPoint() {
EXIT_IF(IsGpuThread());
m_suspend_point_slot.acquire();
GpuMutexLock lock(m_submission_mutex);
if (IsGpuThread()) {
m_gfx_cp->BufferWait();
} else {
WaitLocked();
}
m_graphics_done = true;
m_done_num++;
Submission submission;
submission.type = SubmissionType::SuspendPoint;
submission.queue_id = 0;
m_reset_graphics = false;
Enqueue(std::move(submission));
}
int GpuState::GetFrameNum() {
return m_done_num;
}
void GpuState::WaitLocked() {
WaitForIdle();
SendCommandSync([this] { m_gfx_cp->BufferWait(); });
}
CommandProcessor& GpuState::GetProcessor(uint32_t queue_id) {
EXIT_IF(queue_id >= QueueCount);
if (queue_id == 0) {
@@ -316,9 +308,12 @@ void CommandProcessor::Reset() {
m_context_state_pushed = false;
m_index_type_and_size = 0;
m_index_buffer_size = 0;
m_index_base_addr = 0;
m_user_data_marker = HW::UserSgprType::Unknown;
m_draw_indirect_args_base_addr = 0;
m_dispatch_indirect_args_base_addr = 0;
m_num_instances = 1;
m_predicate_skip = false;
std::memset(m_const_ram, 0, sizeof(m_const_ram));
}
@@ -532,13 +527,6 @@ void GpuState::Enqueue(Submission submission) {
m_work_available.Signal();
}
void GpuState::WaitForIdle() {
Common::LockGuard lock(m_queue_mutex);
while (m_processing || !m_commands.empty() || m_submission_count != 0) {
m_idle.Wait(&m_queue_mutex);
}
}
void GpuState::ThreadRun(void* data) {
auto* gpu = static_cast<GpuState*>(data);
EXIT_IF(gpu == nullptr);
@@ -554,19 +542,14 @@ void GpuState::ThreadRun(void* data) {
{
Common::LockGuard lock(gpu->m_queue_mutex);
while (gpu->m_commands.empty() && gpu->m_submission_count == 0 && !gpu->m_stopping) {
gpu->m_processing = false;
gpu->m_idle.Signal();
gpu->m_work_available.Wait(&gpu->m_queue_mutex);
}
if (gpu->m_stopping && gpu->m_commands.empty() && gpu->m_submission_count == 0) {
gpu->m_processing = false;
gpu->m_idle.SignalAll();
should_stop = true;
} else if (!gpu->m_commands.empty()) {
command = std::move(gpu->m_commands.front());
gpu->m_commands.pop_front();
EXIT_IF(gpu->m_pending_commands.fetch_sub(1, std::memory_order_acq_rel) == 0);
gpu->m_processing = true;
} else {
int selected_queue = -1;
for (uint32_t offset = 0; offset < QueueCount; offset++) {
@@ -577,7 +560,6 @@ void GpuState::ThreadRun(void* data) {
}
}
if (selected_queue < 0) {
gpu->m_processing = false;
gpu->m_work_available.WaitFor(&gpu->m_queue_mutex, 100);
for (auto& queue: gpu->m_queues) {
if (!queue.empty()) {
@@ -591,7 +573,6 @@ void GpuState::ThreadRun(void* data) {
queue.pop_front();
gpu->m_submission_count--;
gpu->m_next_queue = (static_cast<uint32_t>(selected_queue) + 1) % QueueCount;
gpu->m_processing = true;
has_submission = true;
}
}
@@ -606,11 +587,6 @@ void GpuState::ThreadRun(void* data) {
EXIT_IF(g_current_processor != nullptr);
command();
Common::LockGuard lock(gpu->m_queue_mutex);
gpu->m_processing = false;
if (gpu->m_commands.empty() && gpu->m_submission_count == 0) {
gpu->m_idle.SignalAll();
}
continue;
}
@@ -629,16 +605,19 @@ void GpuState::ThreadRun(void* data) {
}
}
}
gpu->m_processing = false;
if (gpu->m_commands.empty() && gpu->m_submission_count == 0) {
gpu->m_idle.SignalAll();
}
}
}
bool GpuState::Process(Submission& submission) {
auto& cp = GetProcessor(submission.queue_id);
const bool first_slice = !submission.started;
if (submission.type == SubmissionType::SuspendPoint) {
EXIT_IF(!first_slice || submission.queue_id != 0);
cp.Reset();
m_done_num++;
m_suspend_point_slot.release();
return true;
}
if (first_slice && submission.reset_processor) {
cp.Reset();
@@ -726,6 +705,7 @@ bool GpuState::Process(Submission& submission) {
m_renderer.GetGpuResources().RunGarbageCollector();
cp.PrepareCpuFlip(submission.flip_request_id);
break;
case SubmissionType::SuspendPoint: EXIT("unreachable suspend point submission\n");
}
return complete;
@@ -1708,8 +1688,8 @@ void Gpu::SubmitFlipPreparation(uint64_t request_id) {
m_state->SubmitFlipPreparation(request_id);
}
void Gpu::Done() {
m_state->Done();
void Gpu::SuspendPoint() {
m_state->SuspendPoint();
}
int Gpu::GetFrameNum() const {