intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 05:56:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: use memory ptr from api when copy from/to image by blt

Browse Source 
Related-To: NEO-15934

Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka
2025-09-25 12:34:19 +00:00
committed by Compute-Runtime-Automation
parent 3c83d8f8c7
commit 6c5ac160a6
2 changed files with 89 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -831,7 +831,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendImageCopyFromMemoryExt(z
}
size_t imgRowPitch = image->getImageInfo().rowPitch;
size_t imgSlicePitch = image->getImageInfo().slicePitch;
auto status = appendCopyImageBlit(allocationStruct.alignedAllocationPtr, allocationStruct.alloc, image->getAllocation()->getGpuAddress(), image->getAllocation(),
this->commandContainer.addToResidencyContainer(allocationStruct.alloc);
auto ptr = ptrOffset(allocationStruct.alignedAllocationPtr, allocationStruct.offset);
auto status = appendCopyImageBlit(ptr, nullptr, image->getAllocation()->getGpuAddress(), image->getAllocation(),
{0, 0, 0}, {pDstRegion->originX, pDstRegion->originY, pDstRegion->originZ}, srcRowPitch, srcSlicePitch,
imgRowPitch, imgSlicePitch, bytesPerPixel, {pDstRegion->width, pDstRegion->height, pDstRegion->depth}, {pDstRegion->width, pDstRegion->height, pDstRegion->depth}, imgSize,
event, numWaitEvents, phWaitEvents, memoryCopyParams);
@@ -1034,7 +1036,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendImageCopyToMemoryExt(voi
}
size_t imgRowPitch = image->getImageInfo().rowPitch;
size_t imgSlicePitch = image->getImageInfo().slicePitch;
auto status = appendCopyImageBlit(image->getAllocation()->getGpuAddress(), image->getAllocation(), allocationStruct.alignedAllocationPtr, allocationStruct.alloc,
this->commandContainer.addToResidencyContainer(allocationStruct.alloc);
auto ptr = ptrOffset(allocationStruct.alignedAllocationPtr, allocationStruct.offset);
auto status = appendCopyImageBlit(image->getAllocation()->getGpuAddress(), image->getAllocation(), ptr, nullptr,
{pSrcRegion->originX, pSrcRegion->originY, pSrcRegion->originZ}, {0, 0, 0}, imgRowPitch, imgSlicePitch,
destRowPitch, destSlicePitch, bytesPerPixel, {pSrcRegion->width, pSrcRegion->height, pSrcRegion->depth},
imgSize, {pSrcRegion->width, pSrcRegion->height, pSrcRegion->depth}, event, numWaitEvents, phWaitEvents, memoryCopyParams);

83
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_2.cpp
View File

@@ -44,6 +44,11 @@ class MockCommandListHw : public WhiteBox<::L0::CommandListCoreFamily<gfxCoreFam
AlignedAllocationData getAlignedAllocationData(L0::Device *device, bool sharedSystemEnabled, const void *buffer, uint64_t bufferSize, bool allowHostCopy, bool copyOffload) override {
getAlignedAllocationCalledTimes++;
if (buffer && returnMockAllocationStruct) {
auto alignedPtr = reinterpret_cast<uintptr_t>(alignDown(buffer, sizeof(uint32_t)));
auto offset = reinterpret_cast<uintptr_t>(buffer) - alignedPtr;
return {alignedPtr, offset, &alignedAlloc, true};
}
if (buffer && !failAlignedAlloc) {
return {0, 0, &alignedAlloc, true};
}
@@ -161,6 +166,10 @@ class MockCommandListHw : public WhiteBox<::L0::CommandListCoreFamily<gfxCoreFam
appendImageRegionDstOrigin = dstOffsets;
appendCopyImageSrcSize = srcSize;
appendCopyImageDstSize = dstSize;
appendSrcAlloc = src;
appendDstAlloc = dst;
appendSrcPtr = srcPtr;
appendDstPtr = dstPtr;
if (signalEvent) {
useEvents = true;
} else {
@@ -179,6 +188,10 @@ class MockCommandListHw : public WhiteBox<::L0::CommandListCoreFamily<gfxCoreFam
Vec3<size_t> appendImageRegionDstOrigin = {9, 9, 9};
Vec3<size_t> appendCopyImageSrcSize{0, 0, 0};
Vec3<size_t> appendCopyImageDstSize{0, 0, 0};
uintptr_t appendSrcPtr;
uintptr_t appendDstPtr;
NEO::GraphicsAllocation *appendSrcAlloc;
NEO::GraphicsAllocation *appendDstAlloc;
void *alignedDataPtr = alignUp(mockAlignedAllocData, MemoryConstants::pageSize);
@@ -196,6 +209,7 @@ class MockCommandListHw : public WhiteBox<::L0::CommandListCoreFamily<gfxCoreFam
bool failOnFirstCopy = false;
bool useEvents = false;
bool failAlignedAlloc = false;
bool returnMockAllocationStruct = false;
};
HWTEST_F(CommandListAppend, givenCommandListWhenMemoryCopyCalledWithNullDstPtrThenAppendMemoryCopyWithappendMemoryCopyReturnsError) {
@@ -2804,5 +2818,74 @@ HWTEST2_F(PrimaryBatchBufferPreamblelessCmdListTest,
EXPECT_EQ(gpuReturnAddress, bbStartCmd->getBatchBufferStartAddress());
}
HWTEST_F(CommandListAppend, givenCopyCommandListWhenImageCopyFromFromMemoryExtThenNotAlignedSrcPtrPassedToBltDispatch) {
MockCommandListHw<FamilyType::gfxCoreFamily> cmdList;
cmdList.initialize(device, NEO::EngineGroupType::copy, 0u);
cmdList.returnMockAllocationStruct = true;
ze_image_desc_t zeDesc = {ZE_STRUCTURE_TYPE_IMAGE_DESC,
nullptr,
0,
ZE_IMAGE_TYPE_1D,
{ZE_IMAGE_FORMAT_LAYOUT_8_8_8_8, ZE_IMAGE_FORMAT_TYPE_UINT,
ZE_IMAGE_FORMAT_SWIZZLE_R, ZE_IMAGE_FORMAT_SWIZZLE_G,
ZE_IMAGE_FORMAT_SWIZZLE_B, ZE_IMAGE_FORMAT_SWIZZLE_A},
4,
1,
1,
0,
0};
zeDesc.stype = ZE_STRUCTURE_TYPE_IMAGE_DESC;
auto image = std::make_unique<WhiteBox<::L0::ImageCoreFamily<FamilyType::gfxCoreFamily>>>();
image->initialize(device, &zeDesc);
ze_image_region_t imgRegion = {0, 0, 0, static_cast<uint32_t>(zeDesc.width), 1, 1};
uint32_t rowPitch = static_cast<uint32_t>(image->getImageInfo().rowPitch);
uint32_t slicePitch = rowPitch;
void *data;
ze_host_mem_alloc_desc_t hostDesc = {};
context->allocHostMem(&hostDesc, 64u, 64u, &data);
void *srcPtr = ptrOffset(data, 3);
cmdList.appendImageCopyFromMemoryExt(image->toHandle(), srcPtr, &imgRegion, rowPitch, slicePitch, nullptr, 0, nullptr, copyParams);
EXPECT_EQ(cmdList.appendSrcPtr, reinterpret_cast<uintptr_t>(srcPtr));
EXPECT_EQ(cmdList.appendSrcAlloc, nullptr);
context->freeMem(data);
}
HWTEST_F(CommandListAppend, givenCopyCommandListWhenImageCopyFromToMemoryExtThenNotAlignedDstPtrPassedToBltDispatch) {
MockCommandListHw<FamilyType::gfxCoreFamily> cmdList;
cmdList.returnMockAllocationStruct = true;
cmdList.initialize(device, NEO::EngineGroupType::copy, 0u);
ze_image_desc_t zeDesc = {ZE_STRUCTURE_TYPE_IMAGE_DESC,
nullptr,
0,
ZE_IMAGE_TYPE_1D,
{ZE_IMAGE_FORMAT_LAYOUT_8_8_8_8, ZE_IMAGE_FORMAT_TYPE_UINT,
ZE_IMAGE_FORMAT_SWIZZLE_R, ZE_IMAGE_FORMAT_SWIZZLE_G,
ZE_IMAGE_FORMAT_SWIZZLE_B, ZE_IMAGE_FORMAT_SWIZZLE_A},
4,
1,
1,
0,
0};
zeDesc.stype = ZE_STRUCTURE_TYPE_IMAGE_DESC;
auto image = std::make_unique<WhiteBox<::L0::ImageCoreFamily<FamilyType::gfxCoreFamily>>>();
image->initialize(device, &zeDesc);
ze_image_region_t imgRegion = {0, 0, 0, static_cast<uint32_t>(zeDesc.width), 1, 1};
uint32_t rowPitch = static_cast<uint32_t>(image->getImageInfo().rowPitch);
uint32_t slicePitch = rowPitch;
void *data;
ze_host_mem_alloc_desc_t hostDesc = {};
context->allocHostMem(&hostDesc, 64u, 64u, &data);
void *dstPtr = ptrOffset(data, 3);
cmdList.appendImageCopyToMemoryExt(dstPtr, image->toHandle(), &imgRegion, rowPitch, slicePitch, nullptr, 0, nullptr, copyParams);
EXPECT_EQ(cmdList.appendDstPtr, reinterpret_cast<uintptr_t>(dstPtr));
EXPECT_EQ(cmdList.appendDstAlloc, nullptr);
context->freeMem(data);
}
} // namespace ult
} // namespace L0