This is an archive of the discontinued LLVM Phabricator instance.

Differential D55514

[OPENMP][NVPTX]Revert __kmpc_shuffle_int64 to its original form.
ClosedPublic

Authored by ABataev on Dec 10 2018, 8:39 AM.

Details

Reviewers
gtbercea
Commits
rG9056f1116d20: [OPENMP][NVPTX]Revert __kmpc_shuffle_int64 to its original form.
rL348772: [OPENMP][NVPTX]Revert __kmpc_shuffle_int64 to its original form.
rOMP348772: [OPENMP][NVPTX]Revert __kmpc_shuffle_int64 to its original form.
Summary

Use the original shuffle implementation for __kmpc_shuffle_int64 since
default implementation uses the same implementation.

Diff Detail

Repository
rOMP OpenMP

Event Timeline

ABataev created this revision.Dec 10 2018, 8:39 AM
Herald added a subscriber: guansong. · View Herald TranscriptDec 10 2018, 8:39 AM
gtbercea accepted this revision.Dec 10 2018, 8:40 AM

LG

This revision is now accepted and ready to land.Dec 10 2018, 8:40 AM
Harbormaster completed remote builds in B25890: Diff 177524.Dec 10 2018, 8:41 AM
Closed by commit rOMP348772: [OPENMP][NVPTX]Revert __kmpc_shuffle_int64 to its original form. (authored by ABataev). · Explain WhyDec 10 2018, 8:53 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
libomptarget/
deviceRTLs/
nvptx/
src/
5 lines

Diff 177536

libomptarget/deviceRTLs/nvptx/src/reduction.cu

Show First 20 Lines • Show All 70 Lines • ▼ Show 20 Lines
EXTERN EXTERN
void __kmpc_nvptx_end_reduce_nowait(int32_t global_tid) {} void __kmpc_nvptx_end_reduce_nowait(int32_t global_tid) {}
EXTERN int32_t __kmpc_shuffle_int32(int32_t val, int16_t delta, int16_t size) { EXTERN int32_t __kmpc_shuffle_int32(int32_t val, int16_t delta, int16_t size) {
return __SHFL_DOWN_SYNC(0xFFFFFFFF, val, delta, size); return __SHFL_DOWN_SYNC(0xFFFFFFFF, val, delta, size);
} }
EXTERN int64_t __kmpc_shuffle_int64(int64_t val, int16_t delta, int16_t size) { EXTERN int64_t __kmpc_shuffle_int64(int64_t val, int16_t delta, int16_t size) {
#if defined(CUDART_VERSION) && CUDART_VERSION >= 9000
return __SHFL_DOWN_SYNC(0xFFFFFFFFFFFFFFFFLL, (long long)val, (unsigned)delta,
(int)size);
#else
int lo, hi; int lo, hi;
asm volatile("mov.b64 {%0,%1}, %2;" : "=r"(lo), "=r"(hi) : "l"(val)); asm volatile("mov.b64 {%0,%1}, %2;" : "=r"(lo), "=r"(hi) : "l"(val));
hi = __SHFL_DOWN_SYNC(0xFFFFFFFF, hi, delta, size); hi = __SHFL_DOWN_SYNC(0xFFFFFFFF, hi, delta, size);
lo = __SHFL_DOWN_SYNC(0xFFFFFFFF, lo, delta, size); lo = __SHFL_DOWN_SYNC(0xFFFFFFFF, lo, delta, size);
asm volatile("mov.b64 %0, {%1,%2};" : "=l"(val) : "r"(lo), "r"(hi)); asm volatile("mov.b64 %0, {%1,%2};" : "=l"(val) : "r"(lo), "r"(hi));
return val; return val;
#endif
} }
static INLINE void gpu_regular_warp_reduce(void *reduce_data, static INLINE void gpu_regular_warp_reduce(void *reduce_data,
kmp_ShuffleReductFctPtr shflFct) { kmp_ShuffleReductFctPtr shflFct) {
for (uint32_t mask = WARPSIZE / 2; mask > 0; mask /= 2) { for (uint32_t mask = WARPSIZE / 2; mask > 0; mask /= 2) {
shflFct(reduce_data, /*LaneId - not used= */ 0, shflFct(reduce_data, /*LaneId - not used= */ 0,
/*Offset = */ mask, /*AlgoVersion=*/0); /*Offset = */ mask, /*AlgoVersion=*/0);
} }
▲ Show 20 Lines • Show All 358 Lines • Show Last 20 Lines