intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-21 09:14:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

Limit system memory flag in builtin kernels to destination argument

Browse Source 
Related-To: NEO-6959

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2022-07-15 19:47:17 +00:00
committed by Compute-Runtime-Automation
parent f89ea1685e
commit 3f8c19eec9
13 changed files with 341 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
opencl/source/built_ins/builtins_dispatch_builder.cpp
View File

@@ -61,14 +61,16 @@ class BuiltInOp<EBuiltInOps::CopyBufferToBuffer> : public BuiltinDispatchInfoBui
auto middleSizeEls = middleSizeBytes / middleElSize; // num work items in middle walker
uint32_t rootDeviceIndex = clDevice.getRootDeviceIndex();
// Set-up ISA
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Left, kernLeftLeftover->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Left, kernLeftLeftover->getKernel(rootDeviceIndex));
if (isSrcMisaligned) {
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Middle, kernMiddleMisaligned->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Middle, kernMiddleMisaligned->getKernel(rootDeviceIndex));
} else {
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Middle, kernMiddle->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Middle, kernMiddle->getKernel(rootDeviceIndex));
}
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Right, kernRightLeftover->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Right, kernRightLeftover->getKernel(rootDeviceIndex));
// Set-up common kernel args
if (operationParams.srcSvmAlloc) {
@@ -78,13 +80,19 @@ class BuiltInOp<EBuiltInOps::CopyBufferToBuffer> : public BuiltinDispatchInfoBui
} else {
kernelSplit1DBuilder.setArgSvm(0, operationParams.size.x + operationParams.srcOffset.x, operationParams.srcPtr, nullptr, CL_MEM_READ_ONLY);
}
bool isDestinationInSystem = false;
if (operationParams.dstSvmAlloc) {
kernelSplit1DBuilder.setArgSvmAlloc(1, operationParams.dstPtr, operationParams.dstSvmAlloc);
isDestinationInSystem = Kernel::graphicsAllocationTypeUseSystemMemory(operationParams.dstSvmAlloc->getAllocationType());
} else if (operationParams.dstMemObj) {
kernelSplit1DBuilder.setArg(1, operationParams.dstMemObj);
isDestinationInSystem = Kernel::graphicsAllocationTypeUseSystemMemory(operationParams.dstMemObj->getGraphicsAllocation(rootDeviceIndex)->getAllocationType());
} else {
kernelSplit1DBuilder.setArgSvm(1, operationParams.size.x + operationParams.dstOffset.x, operationParams.dstPtr, nullptr, 0u);
isDestinationInSystem = operationParams.dstPtr != nullptr;
}
kernelSplit1DBuilder.setKernelDestinationArgumentInSystem(isDestinationInSystem);
kernelSplit1DBuilder.setUnifiedMemorySyncRequirement(operationParams.unifiedMemoryArgsRequireMemSync);
@@ -185,9 +193,11 @@ class BuiltInOp<EBuiltInOps::CopyBufferRect> : public BuiltinDispatchInfoBuilder
}
}
uint32_t rootDeviceIndex = clDevice.getRootDeviceIndex();
// Set-up ISA
int dimensions = is3D ? 3 : 2;
kernelNoSplit3DBuilder.setKernel(kernelBytes[dimensions - 1]->getKernel(clDevice.getRootDeviceIndex()));
kernelNoSplit3DBuilder.setKernel(kernelBytes[dimensions - 1]->getKernel(rootDeviceIndex));
size_t srcOffsetFromAlignedPtr = 0;
size_t dstOffsetFromAlignedPtr = 0;
@@ -205,9 +215,11 @@ class BuiltInOp<EBuiltInOps::CopyBufferRect> : public BuiltinDispatchInfoBuilder
kernelNoSplit3DBuilder.setArgSvm(0, hostPtrSize, srcPtrToSet, nullptr, CL_MEM_READ_ONLY);
}
bool isDestinationInSystem = false;
// arg1 = dst
if (operationParams.dstMemObj) {
kernelNoSplit3DBuilder.setArg(1, operationParams.dstMemObj);
isDestinationInSystem = Kernel::graphicsAllocationTypeUseSystemMemory(operationParams.dstMemObj->getGraphicsAllocation(rootDeviceIndex)->getAllocationType());
} else {
void *dstPtrToSet = operationParams.dstPtr;
if (!is3D) {
@@ -216,7 +228,9 @@ class BuiltInOp<EBuiltInOps::CopyBufferRect> : public BuiltinDispatchInfoBuilder
dstOffsetFromAlignedPtr = ptrDiff(dstPtr, dstPtrToSet);
}
kernelNoSplit3DBuilder.setArgSvm(1, hostPtrSize, dstPtrToSet, nullptr, 0u);
isDestinationInSystem = operationParams.dstPtr != nullptr;
}
kernelNoSplit3DBuilder.setKernelDestinationArgumentInSystem(isDestinationInSystem);
// arg2 = srcOrigin
OffsetType kSrcOrigin[4] = {static_cast<OffsetType>(operationParams.srcOffset.x + srcOffsetFromAlignedPtr), static_cast<OffsetType>(operationParams.srcOffset.y), static_cast<OffsetType>(operationParams.srcOffset.z), 0};
@@ -302,20 +316,26 @@ class BuiltInOp<EBuiltInOps::FillBuffer> : public BuiltinDispatchInfoBuilder {
auto middleSizeEls = middleSizeBytes / middleElSize; // num work items in middle walker
uint32_t rootDeviceIndex = clDevice.getRootDeviceIndex();
// Set-up ISA
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Left, kernLeftLeftover->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Middle, kernMiddle->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Right, kernRightLeftover->getKernel(clDevice.getRootDeviceIndex()));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Left, kernLeftLeftover->getKernel(rootDeviceIndex));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Middle, kernMiddle->getKernel(rootDeviceIndex));
kernelSplit1DBuilder.setKernel(SplitDispatch::RegionCoordX::Right, kernRightLeftover->getKernel(rootDeviceIndex));
DEBUG_BREAK_IF((operationParams.srcMemObj == nullptr) || (operationParams.srcOffset != 0));
DEBUG_BREAK_IF((operationParams.dstMemObj == nullptr) && (operationParams.dstSvmAlloc == nullptr));
bool isDestinationInSystem = false;
// Set-up dstMemObj with buffer
if (operationParams.dstSvmAlloc) {
kernelSplit1DBuilder.setArgSvmAlloc(0, operationParams.dstPtr, operationParams.dstSvmAlloc);
isDestinationInSystem = Kernel::graphicsAllocationTypeUseSystemMemory(operationParams.dstSvmAlloc->getAllocationType());
} else {
kernelSplit1DBuilder.setArg(0, operationParams.dstMemObj);
isDestinationInSystem = Kernel::graphicsAllocationTypeUseSystemMemory(operationParams.dstMemObj->getGraphicsAllocation(rootDeviceIndex)->getAllocationType());
}
kernelSplit1DBuilder.setKernelDestinationArgumentInSystem(isDestinationInSystem);
// Set-up dstOffset
kernelSplit1DBuilder.setArg(SplitDispatch::RegionCoordX::Left, 1, static_cast<OffsetType>(operationParams.dstOffset.x));
@@ -545,20 +565,26 @@ class BuiltInOp<EBuiltInOps::CopyImage3dToBuffer> : public BuiltinDispatchInfoBu
size_t hostPtrSize = operationParams.dstPtr ? Image::calculateHostPtrSize(region, dstRowPitch, dstSlicePitch, bytesPerPixel, srcImage->getImageDesc().image_type) : 0;
hostPtrSize += operationParams.dstOffset.x;
uint32_t rootDeviceIndex = clDevice.getRootDeviceIndex();
// Set-up ISA
auto bytesExponent = Math::log2(bytesPerPixel);
DEBUG_BREAK_IF(bytesExponent >= 5);
kernelNoSplit3DBuilder.setKernel(kernelBytes[bytesExponent]->getKernel(clDevice.getRootDeviceIndex()));
kernelNoSplit3DBuilder.setKernel(kernelBytes[bytesExponent]->getKernel(rootDeviceIndex));
// Set-up source image
kernelNoSplit3DBuilder.setArg(0, srcImageRedescribed, operationParams.srcMipLevel);
bool isDestinationInSystem = false;
// Set-up destination host ptr / buffer
if (operationParams.dstPtr) {
kernelNoSplit3DBuilder.setArgSvm(1, hostPtrSize, operationParams.dstPtr, nullptr, 0u);
isDestinationInSystem = operationParams.dstPtr != nullptr;
} else {
kernelNoSplit3DBuilder.setArg(1, operationParams.dstMemObj);
isDestinationInSystem = Kernel::graphicsAllocationTypeUseSystemMemory(operationParams.dstMemObj->getGraphicsAllocation(rootDeviceIndex)->getAllocationType());
}
kernelNoSplit3DBuilder.setKernelDestinationArgumentInSystem(isDestinationInSystem);
// Set-up srcOrigin
{