fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 18:10:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

d3d12: Video Encode - Do not flush on direct buffer maps

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/38578>
This commit is contained in:
Silvio Vilerino 2025-11-20 15:48:29 -05:00 • committed by Pohsiang (John) Hsu
parent f7c306c654
commit d9c2ccd543
1 changed files with 32 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
src/gallium/drivers/d3d12/d3d12_video_enc.cpp
View file

@ -3041,6 +3041,14 @@ d3d12_video_encoder_calculate_max_slices_count_in_output(
return maxSlices; return maxSlices;
} }
static inline bool
d3d12_buffer_maps_directly(pipe_resource *buffer)
{
return buffer->target == PIPE_BUFFER &&
buffer->usage != PIPE_USAGE_DEFAULT &&
buffer->usage != PIPE_USAGE_IMMUTABLE;
}
void void
d3d12_video_encoder_get_slice_bitstream_data(struct pipe_video_codec *codec, d3d12_video_encoder_get_slice_bitstream_data(struct pipe_video_codec *codec,
void *feedback, void *feedback,
@ -3183,7 +3191,8 @@ d3d12_video_encoder_get_slice_bitstream_data(struct pipe_video_codec *codec,
} }
// If we uploaded new slice headers, flush and wait for the context to upload them // If we uploaded new slice headers, flush and wait for the context to upload them
if (pD3D12Enc->m_spEncodedFrameMetadata[current_metadata_slot].pSliceHeaders[slice_idx].size() > 0) if ((pD3D12Enc->m_spEncodedFrameMetadata[current_metadata_slot].pSliceHeaders[slice_idx].size() > 0) &&
!d3d12_buffer_maps_directly(pD3D12Enc->m_spEncodedFrameMetadata[current_metadata_slot].comp_bit_destinations[slice_idx])) // If the buffer maps directly, the buffer_subdata is synchronous on unmap, no need to flush
{ {
struct pipe_fence_handle *pUploadGPUCompletionFence = NULL; struct pipe_fence_handle *pUploadGPUCompletionFence = NULL;
pD3D12Enc->base.context->flush(pD3D12Enc->base.context, pD3D12Enc->base.context->flush(pD3D12Enc->base.context,
@ -3437,12 +3446,19 @@ d3d12_video_encoder_encode_bitstream_impl(struct pipe_video_codec *codec,
static_cast<unsigned int>(pD3D12Enc->m_BitstreamHeadersBuffer.size()), static_cast<unsigned int>(pD3D12Enc->m_BitstreamHeadersBuffer.size()),
pD3D12Enc->m_BitstreamHeadersBuffer.data()); pD3D12Enc->m_BitstreamHeadersBuffer.data());
if (!d3d12_buffer_maps_directly(&pD3D12Enc->m_pOutputBitstreamBuffers[0/*first slice buffer*/]->base.b)) // If the buffer maps directly, the buffer_subdata is synchronous on unmap, no need to flush
{
// If the destination buffer doesn't map directly (eg. DEFAULT usage), we need to flush
// and set a fence to ensure the upload is finished before EncodeFrame reads from it
debug_printf("[d3d12_video_encoder] d3d12_video_encoder_encode_bitstream_impl - Flushing pD3D12Enc->m_BitstreamHeadersBuffer data upload.\n"); debug_printf("[d3d12_video_encoder] d3d12_video_encoder_encode_bitstream_impl - Flushing pD3D12Enc->m_BitstreamHeadersBuffer data upload.\n");
pD3D12Enc->base.context->flush( pD3D12Enc->base.context->flush(
pD3D12Enc->base.context, pD3D12Enc->base.context,
&pD3D12Enc->m_inflightResourcesPool[d3d12_video_encoder_pool_current_index(pD3D12Enc)].headers_upload_completion_fence, &pD3D12Enc->m_inflightResourcesPool[d3d12_video_encoder_pool_current_index(pD3D12Enc)].headers_upload_completion_fence,
PIPE_FLUSH_ASYNC | PIPE_FLUSH_HINT_FINISH); PIPE_FLUSH_ASYNC | PIPE_FLUSH_HINT_FINISH);
assert(pD3D12Enc->m_inflightResourcesPool[d3d12_video_encoder_pool_current_index(pD3D12Enc)].headers_upload_completion_fence); assert(pD3D12Enc->m_inflightResourcesPool[d3d12_video_encoder_pool_current_index(pD3D12Enc)].headers_upload_completion_fence);
}
} }
} }
else else
@ -4670,18 +4686,21 @@ d3d12_video_encoder_get_feedback(struct pipe_video_codec *codec,
// //
// Flush copies in batch and wait on this CPU thread for GPU work completion // Flush copies in batch and wait on this CPU thread for GPU work completion
// //
struct pipe_fence_handle *pUploadGPUCompletionFence = NULL;
pD3D12Enc->base.context->flush(pD3D12Enc->base.context, if (!d3d12_buffer_maps_directly(pD3D12Enc->m_spEncodedFrameMetadata[current_metadata_slot].comp_bit_destinations[0/*first slice*/])) { // If the buffer maps directly, the buffer_subdata is synchronous on unmap, no need to flush
&pUploadGPUCompletionFence, struct pipe_fence_handle *pUploadGPUCompletionFence = NULL;
PIPE_FLUSH_ASYNC | PIPE_FLUSH_HINT_FINISH); pD3D12Enc->base.context->flush(pD3D12Enc->base.context,
assert(pUploadGPUCompletionFence); &pUploadGPUCompletionFence,
pD3D12Enc->m_pD3D12Screen->base.fence_finish(&pD3D12Enc->m_pD3D12Screen->base, PIPE_FLUSH_ASYNC | PIPE_FLUSH_HINT_FINISH);
NULL, assert(pUploadGPUCompletionFence);
pUploadGPUCompletionFence, pD3D12Enc->m_pD3D12Screen->base.fence_finish(&pD3D12Enc->m_pD3D12Screen->base,
OS_TIMEOUT_INFINITE); NULL,
pD3D12Enc->m_pD3D12Screen->base.fence_reference(&pD3D12Enc->m_pD3D12Screen->base, pUploadGPUCompletionFence,
&pUploadGPUCompletionFence, OS_TIMEOUT_INFINITE);
NULL); pD3D12Enc->m_pD3D12Screen->base.fence_reference(&pD3D12Enc->m_pD3D12Screen->base,
&pUploadGPUCompletionFence,
NULL);
}
} }
} }