Skip to content

Commit 33bb26d

Browse files
authored
Get ImplicitLocalArgPos from DeviceKernelInfo instead of ProgramManager (#20621)
The performance regression was introduced in the #20316. We should get the position of the implicit local arg from the deviceKernelInfo, not from the ProgramManager.
1 parent b312f04 commit 33bb26d

File tree

1 file changed

+1
-2
lines changed

1 file changed

+1
-2
lines changed

sycl/source/detail/scheduler/commands.cpp

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2470,8 +2470,7 @@ static ur_result_t SetKernelParamsAndLaunch(
24702470
}
24712471

24722472
std::optional<int> ImplicitLocalArg =
2473-
ProgramManager::getInstance().kernelImplicitLocalArgPos(
2474-
DeviceKernelInfo.Name);
2473+
DeviceKernelInfo.getImplicitLocalArgPos();
24752474
// Set the implicit local memory buffer to support
24762475
// get_work_group_scratch_memory. This is for backend not supporting
24772476
// CUDA-style local memory setting. Note that we may have -1 as a position,

0 commit comments

Comments
 (0)