intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 00:58:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: add infrastructure for staging with 3D images

Browse Source 
Related-To: NEO-14026

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
This commit is contained in:
Szymon Morek
2025-03-31 12:32:11 +00:00
committed by Compute-Runtime-Automation
parent ed37a1e7ef
commit 3010af596e
6 changed files with 273 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
opencl/source/command_queue/command_queue_staging.cpp
View File

@@ -54,8 +54,10 @@ cl_int CommandQueue::enqueueStagingImageTransfer(cl_command_type commandType, Im
bool isSingleTransfer = false;
ChunkTransferImageFunc chunkWrite = [&](void *stagingBuffer, const size_t *origin, const size_t *region) -> int32_t {
auto isFirstTransfer = (globalOrigin[1] == origin[1]);
auto isLastTransfer = (globalOrigin[1] + globalRegion[1] == origin[1] + region[1]);
auto isFirstTransfer = (globalOrigin[1] == origin[1] && globalOrigin[2] == origin[2]);
auto isLastTransfer = (globalOrigin[1] + globalRegion[1] == origin[1] + region[1]) &&
(globalOrigin[2] + globalRegion[2] == origin[2] + region[2]);
isSingleTransfer = isFirstTransfer && isLastTransfer;
cl_event *outEvent = assignEventForStaging(event, &profilingEvent, isFirstTransfer, isLastTransfer);
cl_int ret = 0;
@@ -69,9 +71,10 @@ cl_int CommandQueue::enqueueStagingImageTransfer(cl_command_type commandType, Im
};
auto bytesPerPixel = image->getSurfaceFormatInfo().surfaceFormat.imageElementSizeInBytes;
auto dstRowPitch = inputRowPitch ? inputRowPitch : globalRegion[0] * bytesPerPixel;
auto dstSlicePitch = inputSlicePitch ? inputSlicePitch : globalRegion[1] * dstRowPitch;
auto stagingBufferManager = this->context->getStagingBufferManager();
auto ret = stagingBufferManager->performImageTransfer(ptr, globalOrigin, globalRegion, dstRowPitch, bytesPerPixel, chunkWrite, &csr, isRead);
auto ret = stagingBufferManager->performImageTransfer(ptr, globalOrigin, globalRegion, dstRowPitch, dstSlicePitch, bytesPerPixel, chunkWrite, &csr, isRead);
if (isRead && context->isProvidingPerformanceHints()) {
auto hostPtrSize = calculateHostPtrSizeForImage(globalRegion, inputRowPitch, inputSlicePitch, image);