intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-05 09:09:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

zeCommandListAppendMemoryFill fails to fill with 2 byte pattern

Browse Source 
zeCommandListAppendMemoryFill fails to fill sizes that are not
multiples of four when using a 2 byte pattern.

Fix the problem and add a blackbox test.

Signed-off-by: Troels Nielsen <bn.troels@gmail.com>
This commit is contained in:
Troels Nielsen
2022-03-22 20:02:06 +01:00
committed by Compute-Runtime-Automation
parent 4d4ccfd128
commit 1ee733e191
2 changed files with 61 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -1482,7 +1482,8 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryFill(void *ptr,
builtinFunction->setGroupSize(groupSizeX, groupSizeY, groupSizeZ);
uint32_t groups = static_cast<uint32_t>(adjustedSize) / groupSizeX;
uint32_t groupRemainderSizeX = static_cast<uint32_t>(size) % groupSizeX;
uint32_t remainingBytes = static_cast<uint32_t>((adjustedSize % groupSizeX) * middleElSize +
size % middleElSize);
size_t patternAllocationSize = alignUp(patternSize, MemoryConstants::cacheLineSize);
uint32_t patternSizeInEls = static_cast<uint32_t>(patternAllocationSize / middleElSize);
@@ -1523,7 +1524,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryFill(void *ptr,
return res;
}
if (groupRemainderSizeX) {
if (remainingBytes) {
uint32_t dstOffsetRemainder = groups * groupSizeX * static_cast<uint32_t>(middleElSize);
uint64_t patternOffsetRemainder = (groupSizeX * groups & (patternSizeInEls - 1)) * middleElSize;
@@ -1534,7 +1535,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryFill(void *ptr,
builtinFunctionRemainder = device->getBuiltinFunctionsLib()->getFunction(Builtin::FillBufferRightLeftover);
}
builtinFunctionRemainder->setGroupSize(groupRemainderSizeX, 1u, 1u);
builtinFunctionRemainder->setGroupSize(remainingBytes, 1u, 1u);
ze_group_count_t dispatchFuncArgs{1u, 1u, 1u};
builtinFunctionRemainder->setArgBufferWithAlloc(0,
@@ -1546,7 +1547,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryFill(void *ptr,
builtinFunctionRemainder->setArgBufferWithAlloc(2,
reinterpret_cast<uintptr_t>(patternGfxAllocPtr) + patternOffsetRemainder,
patternGfxAlloc);
builtinFunctionRemainder->setArgumentValue(3, sizeof(patternSizeInEls), &patternSizeInEls);
builtinFunctionRemainder->setArgumentValue(3, sizeof(patternAllocationSize), &patternAllocationSize);
res = appendLaunchKernelSplit(builtinFunctionRemainder->toHandle(), &dispatchFuncArgs, hSignalEvent);
if (res) {
return res;