intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-28 16:48:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Enable zero copy for enqueueImage r/w with hints

Browse Source 
Change-Id: I6d4379b4bebaca162f859ea790f6a77475f7e94e
This commit is contained in:
mplewka
2018-02-02 10:19:13 +01:00
committed by sys_ocldev
parent 1cfe5344fc
commit 21c1dce943
10 changed files with 659 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
runtime/command_queue/enqueue_read_image.h
View File

@@ -51,6 +51,31 @@ cl_int CommandQueueHw<GfxFamily>::enqueueReadImage(
MultiDispatchInfo di;
size_t hostOffset;
Image::calculateHostPtrOffset(&hostOffset, origin, region, inputRowPitch, inputSlicePitch, srcImage->getImageDesc().image_type, srcImage->getSurfaceFormatInfo().ImageElementSizeInBytes);
auto isMemTransferNeeded = srcImage->checkIfMemoryTransferIsRequired(hostOffset, 0, ptr, CL_COMMAND_READ_IMAGE);
if (!isMemTransferNeeded) {
NullSurface s;
Surface *surfaces[] = {&s};
enqueueHandler<CL_COMMAND_MARKER>(
surfaces,
blockingRead == CL_TRUE,
di,
numEventsInWaitList,
eventWaitList,
event);
if (event) {
auto pEvent = castToObjectOrAbort<Event>(*event);
pEvent->setCmdType(CL_COMMAND_READ_IMAGE);
}
if (context->isProvidingPerformanceHints()) {
context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_GOOD_INTEL, CL_ENQUEUE_READ_IMAGE_DOESNT_REQUIRES_COPY_DATA, static_cast<cl_mem>(srcImage));
}
return CL_SUCCESS;
}
auto &builder = BuiltIns::getInstance().getBuiltinDispatchInfoBuilder(EBuiltInOps::CopyImage3dToBuffer,
this->getContext(), this->getDevice());
@@ -86,4 +111,4 @@ cl_int CommandQueueHw<GfxFamily>::enqueueReadImage(
return CL_SUCCESS;
}
}
} // namespace OCLRT

25
runtime/command_queue/enqueue_write_image.h
View File

@@ -48,7 +48,30 @@ cl_int CommandQueueHw<GfxFamily>::enqueueWriteImage(
cl_event *event) {
MultiDispatchInfo di;
size_t hostOffset;
Image::calculateHostPtrOffset(&hostOffset, origin, region, inputRowPitch, inputSlicePitch, dstImage->getImageDesc().image_type, dstImage->getSurfaceFormatInfo().ImageElementSizeInBytes);
auto isMemTransferNeeded = dstImage->checkIfMemoryTransferIsRequired(hostOffset, 0, ptr, CL_COMMAND_WRITE_IMAGE);
if (!isMemTransferNeeded) {
NullSurface s;
Surface *surfaces[] = {&s};
enqueueHandler<CL_COMMAND_MARKER>(
surfaces,
blockingWrite == CL_TRUE,
di,
numEventsInWaitList,
eventWaitList,
event);
if (event) {
auto pEvent = castToObjectOrAbort<Event>(*event);
pEvent->setCmdType(CL_COMMAND_WRITE_IMAGE);
}
if (context->isProvidingPerformanceHints()) {
context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_GOOD_INTEL, CL_ENQUEUE_WRITE_IMAGE_DOESNT_REQUIRES_COPY_DATA, static_cast<cl_mem>(dstImage));
}
return CL_SUCCESS;
}
auto &builder = BuiltIns::getInstance().getBuiltinDispatchInfoBuilder(EBuiltInOps::CopyBufferToImage3d,
this->getContext(), this->getDevice());
@@ -80,4 +103,4 @@ cl_int CommandQueueHw<GfxFamily>::enqueueWriteImage(
return CL_SUCCESS;
}
}
} // namespace OCLRT