[OPENMP][NVPTX]Fix parallel level counter in non-SPMD mode.

Summary:
In non-SPMD mode we may end up with the divergent threads when trying to
increment/decrement parallel level counter. It may lead to incorrect
calculations of the parallel level and wrong results when threads are
divergent. We need to reconverge the threads before trying to modify the
parallel level counter.

Reviewers: grokos, jdoerfert

Subscribers: guansong, openmp-commits, caomhin, kkwli0

Tags: #openmp

Differential Revision: https://reviews.llvm.org/D66802

llvm-svn: 370803
This commit is contained in:
Alexey Bataev 2019-09-03 18:11:50 +00:00
parent b187eef616
commit 4812941776
4 changed files with 45 additions and 16 deletions

View File

@ -311,7 +311,16 @@ EXTERN bool __kmpc_kernel_parallel(void **WorkFn,
(int)newTaskDescr->ThreadId(), (int)nThreads); (int)newTaskDescr->ThreadId(), (int)nThreads);
isActive = true; isActive = true;
IncParallelLevel(threadsInTeam != 1); // Reconverge the threads at the end of the parallel region to correctly
// handle parallel levels.
// In Cuda9+ in non-SPMD mode we have either 1 worker thread or the whole
// warp. If only 1 thread is active, not need to reconverge the threads.
// If we have the whole warp, reconverge all the threads in the warp before
// actually trying to change the parallel level. Otherwise, parallel level
// can be changed incorrectly because of threads divergence.
bool IsActiveParallelRegion = threadsInTeam != 1;
IncParallelLevel(IsActiveParallelRegion,
IsActiveParallelRegion ? 0xFFFFFFFF : 1u);
} }
return isActive; return isActive;
@ -329,7 +338,16 @@ EXTERN void __kmpc_kernel_end_parallel() {
omptarget_nvptx_threadPrivateContext->SetTopLevelTaskDescr( omptarget_nvptx_threadPrivateContext->SetTopLevelTaskDescr(
threadId, currTaskDescr->GetPrevTaskDescr()); threadId, currTaskDescr->GetPrevTaskDescr());
DecParallelLevel(threadsInTeam != 1); // Reconverge the threads at the end of the parallel region to correctly
// handle parallel levels.
// In Cuda9+ in non-SPMD mode we have either 1 worker thread or the whole
// warp. If only 1 thread is active, not need to reconverge the threads.
// If we have the whole warp, reconverge all the threads in the warp before
// actually trying to change the parallel level. Otherwise, parallel level can
// be changed incorrectly because of threads divergence.
bool IsActiveParallelRegion = threadsInTeam != 1;
DecParallelLevel(IsActiveParallelRegion,
IsActiveParallelRegion ? 0xFFFFFFFF : 1u);
} }
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
@ -339,7 +357,7 @@ EXTERN void __kmpc_kernel_end_parallel() {
EXTERN void __kmpc_serialized_parallel(kmp_Ident *loc, uint32_t global_tid) { EXTERN void __kmpc_serialized_parallel(kmp_Ident *loc, uint32_t global_tid) {
PRINT0(LD_IO, "call to __kmpc_serialized_parallel\n"); PRINT0(LD_IO, "call to __kmpc_serialized_parallel\n");
IncParallelLevel(/*ActiveParallel=*/false); IncParallelLevel(/*ActiveParallel=*/false, __kmpc_impl_activemask());
if (checkRuntimeUninitialized(loc)) { if (checkRuntimeUninitialized(loc)) {
ASSERT0(LT_FUSSY, checkSPMDMode(loc), ASSERT0(LT_FUSSY, checkSPMDMode(loc),
@ -378,7 +396,7 @@ EXTERN void __kmpc_end_serialized_parallel(kmp_Ident *loc,
uint32_t global_tid) { uint32_t global_tid) {
PRINT0(LD_IO, "call to __kmpc_end_serialized_parallel\n"); PRINT0(LD_IO, "call to __kmpc_end_serialized_parallel\n");
DecParallelLevel(/*ActiveParallel=*/false); DecParallelLevel(/*ActiveParallel=*/false, __kmpc_impl_activemask());
if (checkRuntimeUninitialized(loc)) { if (checkRuntimeUninitialized(loc)) {
ASSERT0(LT_FUSSY, checkSPMDMode(loc), ASSERT0(LT_FUSSY, checkSPMDMode(loc),

View File

@ -10,6 +10,7 @@
// //
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
#include "target_impl.h"
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
// Execution Parameters // Execution Parameters
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
@ -65,8 +66,8 @@ INLINE int GetNumberOfProcsInDevice(bool isSPMDExecutionMode);
INLINE int IsTeamMaster(int ompThreadId); INLINE int IsTeamMaster(int ompThreadId);
// Parallel level // Parallel level
INLINE void IncParallelLevel(bool ActiveParallel); INLINE void IncParallelLevel(bool ActiveParallel, __kmpc_impl_lanemask_t Mask);
INLINE void DecParallelLevel(bool ActiveParallel); INLINE void DecParallelLevel(bool ActiveParallel, __kmpc_impl_lanemask_t Mask);
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
// Memory // Memory

View File

@ -203,30 +203,28 @@ INLINE int IsTeamMaster(int ompThreadId) { return (ompThreadId == 0); }
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
// Parallel level // Parallel level
INLINE void IncParallelLevel(bool ActiveParallel) { INLINE void IncParallelLevel(bool ActiveParallel, __kmpc_impl_lanemask_t Mask) {
__kmpc_impl_lanemask_t Active = __kmpc_impl_activemask(); __kmpc_impl_syncwarp(Mask);
__kmpc_impl_syncwarp(Active);
__kmpc_impl_lanemask_t LaneMaskLt = __kmpc_impl_lanemask_lt(); __kmpc_impl_lanemask_t LaneMaskLt = __kmpc_impl_lanemask_lt();
unsigned Rank = __kmpc_impl_popc(Active & LaneMaskLt); unsigned Rank = __kmpc_impl_popc(Mask & LaneMaskLt);
if (Rank == 0) { if (Rank == 0) {
parallelLevel[GetWarpId()] += parallelLevel[GetWarpId()] +=
(1 + (ActiveParallel ? OMP_ACTIVE_PARALLEL_LEVEL : 0)); (1 + (ActiveParallel ? OMP_ACTIVE_PARALLEL_LEVEL : 0));
__threadfence(); __threadfence();
} }
__kmpc_impl_syncwarp(Active); __kmpc_impl_syncwarp(Mask);
} }
INLINE void DecParallelLevel(bool ActiveParallel) { INLINE void DecParallelLevel(bool ActiveParallel, __kmpc_impl_lanemask_t Mask) {
__kmpc_impl_lanemask_t Active = __kmpc_impl_activemask(); __kmpc_impl_syncwarp(Mask);
__kmpc_impl_syncwarp(Active);
__kmpc_impl_lanemask_t LaneMaskLt = __kmpc_impl_lanemask_lt(); __kmpc_impl_lanemask_t LaneMaskLt = __kmpc_impl_lanemask_lt();
unsigned Rank = __kmpc_impl_popc(Active & LaneMaskLt); unsigned Rank = __kmpc_impl_popc(Mask & LaneMaskLt);
if (Rank == 0) { if (Rank == 0) {
parallelLevel[GetWarpId()] -= parallelLevel[GetWarpId()] -=
(1 + (ActiveParallel ? OMP_ACTIVE_PARALLEL_LEVEL : 0)); (1 + (ActiveParallel ? OMP_ACTIVE_PARALLEL_LEVEL : 0));
__threadfence(); __threadfence();
} }
__kmpc_impl_syncwarp(Active); __kmpc_impl_syncwarp(Mask);
} }
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////

View File

@ -135,5 +135,17 @@ int main(int argc, char *argv[]) {
} }
} }
// Check for paraller level in non-SPMD kernels.
level = 0;
#pragma omp target teams distribute num_teams(1) thread_limit(32) reduction(+:level)
for (int i=0; i<5032; i+=32) {
int ub = (i+32 > 5032) ? 5032 : i+32;
#pragma omp parallel for schedule(dynamic)
for (int j=i ; j < ub; j++) ;
level += omp_get_level();
}
// CHECK: Integral level = 0.
printf("Integral level = %d.\n", level);
return 0; return 0;
} }