intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improve zeCommandListAppendMemoryFill Performance (2)

Browse Source 
Add missing kernel for remainder kernel when pattern size is 1.

Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
This commit is contained in:
Jaime Arteaga
2021-01-08 16:36:16 -08:00
committed by Compute-Runtime-Automation
parent a0db607083
commit 26b036ab97
2 changed files with 37 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -1182,12 +1182,25 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryFill(void *ptr,
uint32_t groups = static_cast<uint32_t>(size) / groupSizeX;
ze_group_count_t dispatchFuncArgs{groups, 1u, 1u};
res = CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernel(builtinFunction->toHandle(),
&dispatchFuncArgs, nullptr,
0, nullptr);
res = appendLaunchKernelSplit(builtinFunction->toHandle(), &dispatchFuncArgs, hSignalEvent);
if (res) {
return res;
}
uint32_t groupRemainderSizeX = static_cast<uint32_t>(size) % groupSizeX;
if (groupRemainderSizeX) {
builtinFunction->setGroupSize(groupRemainderSizeX, 1u, 1u);
ze_group_count_t dispatchFuncRemainderArgs{1u, 1u, 1u};
size_t dstOffset = dstAllocation.offset + (size - groupRemainderSizeX);
builtinFunction->setArgBufferWithAlloc(0, dstAllocation.alignedAllocationPtr, dstAllocation.alloc);
builtinFunction->setArgumentValue(1, sizeof(dstOffset), &dstOffset);
res = appendLaunchKernelSplit(builtinFunction->toHandle(), &dispatchFuncRemainderArgs, hSignalEvent);
if (res) {
return res;
}
}
} else {
auto builtinFunction = device->getBuiltinFunctionsLib()->getFunction(Builtin::FillBufferMiddle);