This is an archive of the discontinued LLVM Phabricator instance.

[Libomptarget] Add unroll flag to shared variables loop
ClosedPublic

Authored by jhuber6 on Jul 23 2021, 11:51 AM.

Reviewers

jdoerfert
tianshilei1992

Commits

Summary

Unrolling this loop provides better performance in practice because it is
executed on the device and is likely to be very small.

jhuber6 requested review of this revision.Jul 23 2021, 11:51 AM

jhuber6 created this revision.

LGTM.

This revision is now accepted and ready to land.Jul 23 2021, 11:58 AM

This revision was automatically updated to reflect the committed changes.

Path

Size

openmp/

libomptarget/

deviceRTLs/

common/

src/

1 line

Show First 20 Lines • Show All 308 Lines • ▼ Show 20 Lines	if (num_threads != -1)
__kmpc_push_num_threads(ident, global_tid, num_threads);		__kmpc_push_num_threads(ident, global_tid, num_threads);

__kmpc_kernel_prepare_parallel((void *)wrapper_fn);		__kmpc_kernel_prepare_parallel((void *)wrapper_fn);

if (nargs) {		if (nargs) {
void **GlobalArgs;		void **GlobalArgs;
__kmpc_begin_sharing_variables(&GlobalArgs, nargs);		__kmpc_begin_sharing_variables(&GlobalArgs, nargs);
// TODO: faster memcpy?		// TODO: faster memcpy?
		#pragma unroll
for (int I = 0; I < nargs; I++)		for (int I = 0; I < nargs; I++)
GlobalArgs[I] = args[I];		GlobalArgs[I] = args[I];
}		}

// TODO: what if that's a parallel region with a single thread? this is		// TODO: what if that's a parallel region with a single thread? this is
// considered not active in the existing implementation.		// considered not active in the existing implementation.
bool IsActiveParallelRegion = threadsInTeam != 1;		bool IsActiveParallelRegion = threadsInTeam != 1;
int NumWarps =		int NumWarps =
Show All 32 Lines