Mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-05-29 23:03:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 6

video_core: Added copyGPUCmdBuffers option

Browse source
This commit is contained in:
Anton Kovalev 2024-08-27 23:16:14 +02:00
parent 3eb47226d9
commit 659e7a4675
4 changed files with 58 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

37
src/video_core/amdgpu/liverpool.cpp
View file

@ -5,6 +5,7 @@
#include "common/debug.h"
#include "common/polyfill_thread.h"
#include "common/thread.h"
#include "common/config.h"
#include "core/libraries/videoout/driver.h"
#include "video_core/amdgpu/liverpool.h"
#include "video_core/amdgpu/pm4_cmds.h"
@ -568,9 +569,45 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, int vqid) {
TracyFiberLeave;
}
void Liverpool::CopyCmdBuffers(std::span<const u32>& dcb, std::span<const u32>& ccb) {
auto& queue = mapped_queues[GfxQueueId];
// This is fine because resize doesn't reallocate the buffer on shrink
queue.dcb_buffer.resize(queue.dcb_buffer_offset + dcb.size());
queue.ccb_buffer.resize(queue.ccb_buffer_offset + dcb.size());
u32 prev_dcb_buffer_offset = queue.dcb_buffer_offset;
u32 prev_ccb_buffer_offset = queue.ccb_buffer_offset;
if (!dcb.empty()) {
std::memcpy(queue.dcb_buffer.data() + queue.dcb_buffer_offset, dcb.data(),
dcb.size_bytes());
queue.dcb_buffer_offset += dcb.size();
}
if (!ccb.empty()) {
std::memcpy(queue.ccb_buffer.data() + queue.ccb_buffer_offset, ccb.data(),
ccb.size_bytes());
queue.ccb_buffer_offset += dcb.size();
}
if (!queue.dcb_buffer.empty()) {
dcb = std::span<const u32>{queue.dcb_buffer.begin() + prev_dcb_buffer_offset,
queue.dcb_buffer.begin() + queue.dcb_buffer_offset};
}
if (!queue.ccb_buffer.empty()) {
ccb = std::span<const u32>{queue.ccb_buffer.begin() + prev_ccb_buffer_offset,
queue.ccb_buffer.begin() + queue.ccb_buffer_offset};
}
}
void Liverpool::SubmitGfx(std::span<const u32> dcb, std::span<const u32> ccb) {
auto& queue = mapped_queues[GfxQueueId];
if (Config::copyGPUCmdBuffers()) {
CopyCmdBuffers(dcb, ccb);
}
auto task = ProcessGraphics(dcb, ccb);
{
std::scoped_lock lock{queue.m_access};

8
src/video_core/amdgpu/liverpool.h
View file

@ -11,6 +11,7 @@
#include <span>
#include <thread>
#include <queue>
#include <vector>
#include "common/assert.h"
#include "common/bit_field.h"
@ -1047,6 +1048,8 @@ public:
void SubmitDone() noexcept {
std::scoped_lock lk{submit_mutex};
mapped_queues[GfxQueueId].ccb_buffer_offset = 0;
mapped_queues[GfxQueueId].dcb_buffer_offset = 0;
submit_done = true;
submit_cv.notify_one();
}
@ -1108,6 +1111,7 @@ private:
Handle handle;
};
void CopyCmdBuffers(std::span<const u32>& dcb, std::span<const u32>& ccb);
Task ProcessGraphics(std::span<const u32> dcb, std::span<const u32> ccb);
Task ProcessCeUpdate(std::span<const u32> ccb);
Task ProcessCompute(std::span<const u32> acb, int vqid);
@ -1116,6 +1120,10 @@ private:
struct GpuQueue {
std::mutex m_access{};
std::atomic_uint32_t dcb_buffer_offset;
std::atomic_uint32_t ccb_buffer_offset;
std::vector<u32> dcb_buffer;
std::vector<u32> ccb_buffer;
std::queue<Task::Handle> submits{};
ComputeProgram cs_state{};
};