intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 12:23:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: non-usm copy through staging buffers

Browse Source 
Related-To: NEO-11501

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
This commit is contained in:
Szymon Morek
2024-06-24 17:01:44 +00:00
committed by Compute-Runtime-Automation
parent 659075ffe5
commit 29e3eb512c
21 changed files with 738 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
opencl/source/api/api.cpp
View File

@@ -20,6 +20,7 @@
#include "shared/source/os_interface/device_factory.h"
#include "shared/source/utilities/buffer_pool_allocator.inl"
#include "shared/source/utilities/heap_allocator.h"
#include "shared/source/utilities/staging_buffer_manager.h"
#include "opencl/source/accelerators/intel_motion_estimation.h"
#include "opencl/source/api/additional_extensions.h"
@@ -4913,14 +4914,19 @@ cl_int CL_API_CALL clEnqueueSVMMemcpy(cl_command_queue commandQueue,
}
if (size != 0) {
retVal = pCommandQueue->enqueueSVMMemcpy(
blockingCopy,
dstPtr,
srcPtr,
size,
numEventsInWaitList,
eventWaitList,
event);
auto stagingBufferManager = pCommandQueue->getContext().getStagingBufferManager();
if (stagingBufferManager->isValidForCopy(device, dstPtr, srcPtr, numEventsInWaitList)) {
retVal = pCommandQueue->enqueueStagingBufferMemcpy(blockingCopy, dstPtr, srcPtr, size, event);
} else {
retVal = pCommandQueue->enqueueSVMMemcpy(
blockingCopy,
dstPtr,
srcPtr,
size,
numEventsInWaitList,
eventWaitList,
event);
}
} else {
retVal = pCommandQueue->enqueueMarkerWithWaitList(numEventsInWaitList, eventWaitList, event);
}