fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 00:58:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

d3d12: Optimize d3d12_video_proc heap allocations

Browse source 
Reviewed-by: Pohsiang (John) Hsu <pohhsu@microsoft.com>
Reviewed-by: Yubo Xie <yuboxie@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38144>
This commit is contained in:
Silvio Vilerino 2025-10-25 19:22:46 -04:00 • committed by Marge Bot
parent 484eeb762e
commit 80627e3579
2 changed files with 23 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

36
src/gallium/drivers/d3d12/d3d12_video_proc.cpp
View file

@ -132,8 +132,9 @@ d3d12_video_processor_end_frame(struct pipe_video_codec * codec,
/* P709= */ true,
/* StudioYUV= */ true);
std::vector<DXGI_FORMAT> InputFormats;
for(D3D12_VIDEO_PROCESS_INPUT_STREAM_ARGUMENTS1 curInput : pD3D12Proc->m_ProcessInputs)
auto& InputFormats = pD3D12Proc->m_inputFormatsScratch;
InputFormats.clear();
for(const auto& curInput : pD3D12Proc->m_ProcessInputs)
{
InputFormats.push_back(GetDesc(curInput.InputStream[0].pTexture2D).Format);
}
@ -154,14 +155,14 @@ d3d12_video_processor_end_frame(struct pipe_video_codec * codec,
}
}
// Schedule barrier transitions
std::vector<D3D12_RESOURCE_BARRIER> barrier_transitions;
auto& barrier_transitions = pD3D12Proc->m_barrierTransitionsScratch;
barrier_transitions.clear();
barrier_transitions.push_back(CD3DX12_RESOURCE_BARRIER::Transition(
pD3D12Proc->m_OutputArguments.args.OutputStream[0].pTexture2D,
D3D12_RESOURCE_STATE_COMMON,
D3D12_RESOURCE_STATE_VIDEO_PROCESS_WRITE));
for(D3D12_VIDEO_PROCESS_INPUT_STREAM_ARGUMENTS1 curInput : pD3D12Proc->m_ProcessInputs)
for(const auto& curInput : pD3D12Proc->m_ProcessInputs)
barrier_transitions.push_back(CD3DX12_RESOURCE_BARRIER::Transition(
curInput.InputStream[0].pTexture2D,
D3D12_RESOURCE_STATE_COMMON,
@ -280,8 +281,8 @@ d3d12_video_processor_process_frame(struct pipe_video_codec *codec,
(int)InputArguments.Transform.SourceRectangle.top, (int)InputArguments.Transform.SourceRectangle.left, (int)InputArguments.Transform.SourceRectangle.right, (int)InputArguments.Transform.SourceRectangle.bottom,
(int)InputArguments.Transform.DestinationRectangle.top, (int)InputArguments.Transform.DestinationRectangle.left, (int)InputArguments.Transform.DestinationRectangle.right, (int)InputArguments.Transform.DestinationRectangle.bottom);
pD3D12Proc->m_ProcessInputs.push_back(InputArguments);
pD3D12Proc->m_InputBuffers.push_back(pInputVideoBuffer);
pD3D12Proc->m_ProcessInputs.emplace_back(std::move(InputArguments));
pD3D12Proc->m_InputBuffers.emplace_back(pInputVideoBuffer);
///
/// Flush work to the GPU and blocking wait until GPU finishes
@ -349,7 +350,9 @@ d3d12_video_processor_flush(struct pipe_video_codec * codec)
d3d12_promote_to_permanent_residency(pD3D12Proc->m_pD3D12Screen, &curInput->texture, 1, pD3D12Proc->m_spResidencyFence.Get(), &pD3D12Proc->m_ResidencyFenceValue);
}
HRESULT hr = pD3D12Proc->m_pD3D12Screen->dev->GetDeviceRemovedReason();
HRESULT hr = S_OK;
#if MESA_DEBUG
hr = pD3D12Proc->m_pD3D12Screen->dev->GetDeviceRemovedReason();
if (hr != S_OK) {
debug_printf("[d3d12_video_processor] d3d12_video_processor_flush"
" - D3D12Device was removed BEFORE commandlist "
@ -357,16 +360,11 @@ d3d12_video_processor_flush(struct pipe_video_codec * codec)
(unsigned)hr);
goto flush_fail;
}
#endif // MESA_DEBUG
// Close and execute command list and wait for idle on CPU blocking
// this method before resetting list and allocator for next submission.
if (pD3D12Proc->m_transitionsBeforeCloseCmdList.size() > 0) {
pD3D12Proc->m_spCommandList->ResourceBarrier(static_cast<UINT>(pD3D12Proc->m_transitionsBeforeCloseCmdList.size()),
pD3D12Proc->m_transitionsBeforeCloseCmdList.data());
pD3D12Proc->m_transitionsBeforeCloseCmdList.clear();
}
hr = pD3D12Proc->m_spCommandList->Close();
if (FAILED(hr)) {
debug_printf("[d3d12_video_processor] d3d12_video_processor_flush - Can't close command list with HR %x\n", (unsigned)hr);
@ -391,6 +389,7 @@ d3d12_video_processor_flush(struct pipe_video_codec * codec)
pD3D12Proc->m_spCommandQueue->ExecuteCommandLists(1, ppCommandLists);
pD3D12Proc->m_spCommandQueue->Signal(pD3D12Proc->m_spFence.Get(), pD3D12Proc->m_fenceValue);
#if MESA_DEBUG
// Validate device was not removed
hr = pD3D12Proc->m_pD3D12Screen->dev->GetDeviceRemovedReason();
if (hr != S_OK) {
@ -400,6 +399,7 @@ d3d12_video_processor_flush(struct pipe_video_codec * codec)
(unsigned)hr);
goto flush_fail;
}
#endif // MESA_DEBUG
debug_printf(
"[d3d12_video_processor] d3d12_video_processor_flush - GPU signaled execution finalized for fenceValue: %d\n",
@ -410,7 +410,6 @@ d3d12_video_processor_flush(struct pipe_video_codec * codec)
}
pD3D12Proc->m_ProcessInputs.clear();
pD3D12Proc->m_InputBuffers.clear();
// Free the fence after completion finished
return;
@ -485,6 +484,13 @@ d3d12_video_processor_create(struct pipe_context *context, const struct pipe_vid
goto failed;
}
// Performance optimization: Pre-allocate all vectors with reasonable capacities
// This prevents reallocations during processing operations
pD3D12Proc->m_ProcessInputs.reserve(pD3D12Proc->m_vpMaxInputStreams.MaxInputStreams);
pD3D12Proc->m_InputBuffers.reserve(pD3D12Proc->m_vpMaxInputStreams.MaxInputStreams);
pD3D12Proc->m_barrierTransitionsScratch.reserve(pD3D12Proc->m_vpMaxInputStreams.MaxInputStreams + 1); // +1 for output barrier
pD3D12Proc->m_inputFormatsScratch.reserve(pD3D12Proc->m_vpMaxInputStreams.MaxInputStreams);
debug_printf("[d3d12_video_processor] d3d12_video_create_processor - Created successfully!\n");
if (pD3D12Ctx->priority_manager)

4
src/gallium/drivers/d3d12/d3d12_video_proc.h
View file

@ -115,8 +115,6 @@ struct d3d12_video_processor
std::vector<d3d12_unique_fence> m_PendingFences;
ComPtr<ID3D12VideoProcessCommandList1> m_spCommandList;
std::vector<D3D12_RESOURCE_BARRIER> m_transitionsBeforeCloseCmdList;
// Current state between begin and end frame
d3d12_video_processor_output_context m_OutputArguments;
std::vector<D3D12_VIDEO_PROCESS_INPUT_STREAM_ARGUMENTS1> m_ProcessInputs;
@ -129,6 +127,8 @@ struct d3d12_video_processor
struct d3d12_fence* input_surface_fence = NULL;
uint64_t input_surface_fence_value;
std::vector<D3D12_RESOURCE_BARRIER> m_barrierTransitionsScratch;
std::vector<DXGI_FORMAT> m_inputFormatsScratch;
};
struct pipe_video_codec *