intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-22 10:17:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: flush ocl queue after staging transfer

Browse Source 
This is required for scenarios without direct submission
Otherwise, staging transfers are enqueued but not actually
flushed. If that happens, subsequent staging transfers can't
reuse previous staging buffers. It results in increased memory usage.

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
This commit is contained in:
Szymon Morek
2025-01-07 14:33:43 +00:00
committed by Compute-Runtime-Automation
parent dc0048661a
commit bd5ca4fed5
4 changed files with 14 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
opencl/source/command_queue/command_queue_staging.cpp
View File

@@ -34,7 +34,9 @@ cl_int CommandQueue::enqueueStagingBufferMemcpy(cl_bool blockingCopy, void *dstP
isSingleTransfer = isFirstTransfer && isLastTransfer;
cl_event *outEvent = assignEventForStaging(event, &profilingEvent, isFirstTransfer, isLastTransfer);
return this->enqueueSVMMemcpy(false, chunkDst, chunkSrc, chunkSize, 0, nullptr, outEvent, csr);
auto ret = this->enqueueSVMMemcpy(false, chunkDst, chunkSrc, chunkSize, 0, nullptr, outEvent, csr);
ret |= this->flush();
return ret;
};
auto stagingBufferManager = this->context->getStagingBufferManager();
@@ -55,7 +57,9 @@ cl_int CommandQueue::enqueueStagingWriteImage(Image *dstImage, cl_bool blockingC
isSingleTransfer = isFirstTransfer && isLastTransfer;
cl_event *outEvent = assignEventForStaging(event, &profilingEvent, isFirstTransfer, isLastTransfer);
return this->enqueueWriteImageImpl(dstImage, false, origin, region, inputRowPitch, inputSlicePitch, stagingBuffer, nullptr, 0, nullptr, outEvent, csr);
auto ret = this->enqueueWriteImageImpl(dstImage, false, origin, region, inputRowPitch, inputSlicePitch, stagingBuffer, nullptr, 0, nullptr, outEvent, csr);
ret |= this->flush();
return ret;
};
auto bytesPerPixel = dstImage->getSurfaceFormatInfo().surfaceFormat.imageElementSizeInBytes;
auto dstRowPitch = inputRowPitch ? inputRowPitch : globalRegion[0] * bytesPerPixel;
@@ -78,7 +82,9 @@ cl_int CommandQueue::enqueueStagingReadImage(Image *srcImage, cl_bool blockingCo
isSingleTransfer = isFirstTransfer && isLastTransfer;
cl_event *outEvent = assignEventForStaging(event, &profilingEvent, isFirstTransfer, isLastTransfer);
return this->enqueueReadImageImpl(srcImage, false, origin, region, inputRowPitch, inputSlicePitch, stagingBuffer, nullptr, 0, nullptr, outEvent, csr);
auto ret = this->enqueueReadImageImpl(srcImage, false, origin, region, inputRowPitch, inputSlicePitch, stagingBuffer, nullptr, 0, nullptr, outEvent, csr);
ret |= this->flush();
return ret;
};
auto bytesPerPixel = srcImage->getSurfaceFormatInfo().surfaceFormat.imageElementSizeInBytes;
auto dstRowPitch = inputRowPitch ? inputRowPitch : globalRegion[0] * bytesPerPixel;

4
opencl/test/unit_test/command_queue/enqueue_read_image_tests.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2018-2024 Intel Corporation
* Copyright (C) 2018-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -1127,7 +1127,7 @@ struct ReadImageStagingBufferTest : public EnqueueReadImageTest {
HWTEST_F(ReadImageStagingBufferTest, whenEnqueueStagingReadImageCalledThenReturnSuccess) {
MockCommandQueueHw<FamilyType> mockCommandQueueHw(context, device.get(), &props);
auto res = mockCommandQueueHw.enqueueStagingReadImage(srcImage, false, origin, region, MemoryConstants::megaByte, MemoryConstants::megaByte, ptr, nullptr);
EXPECT_TRUE(mockCommandQueueHw.flushCalled);
EXPECT_EQ(res, CL_SUCCESS);
EXPECT_EQ(4ul, mockCommandQueueHw.enqueueReadImageCounter);
auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();

3
opencl/test/unit_test/command_queue/enqueue_svm_tests.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2018-2024 Intel Corporation
* Copyright (C) 2018-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -2448,6 +2448,7 @@ HWTEST_F(StagingBufferTest, givenInOrderCmdQueueWhenEnqueueStagingBufferMemcpyNo
auto pEvent = (Event *)event;
auto numOfStagingBuffers = svmManager->getNumAllocs() - initialUsmAllocs;
EXPECT_EQ(CL_SUCCESS, retVal);
EXPECT_TRUE(myCmdQ.flushCalled);
EXPECT_EQ(1u, numOfStagingBuffers);
EXPECT_EQ(expectedNumOfCopies, myCmdQ.enqueueSVMMemcpyCalledCount);
EXPECT_EQ(0u, myCmdQ.finishCalledCount);

2
opencl/test/unit_test/command_queue/enqueue_write_image_tests.cpp
View File

@@ -847,7 +847,7 @@ struct WriteImageStagingBufferTest : public EnqueueWriteImageTest {
HWTEST_F(WriteImageStagingBufferTest, whenEnqueueStagingWriteImageCalledThenReturnSuccess) {
MockCommandQueueHw<FamilyType> mockCommandQueueHw(context, device.get(), &props);
auto res = mockCommandQueueHw.enqueueStagingWriteImage(dstImage, false, origin, region, MemoryConstants::megaByte, MemoryConstants::megaByte, ptr, nullptr);
EXPECT_TRUE(mockCommandQueueHw.flushCalled);
EXPECT_EQ(res, CL_SUCCESS);
EXPECT_EQ(4ul, mockCommandQueueHw.enqueueWriteImageCounter);
auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();