Skip to content

Commit

Permalink
src: Skip overlap check for RMA operations if dest_pe != source_pe
Browse files Browse the repository at this point in the history
  • Loading branch information
@philipmarshall21
philipmarshall21 committed Apr 12, 2024
1 parent 0734a51 commit 941a87d
Show file tree
Hide file tree
Showing 4 changed files with 351 additions and 333 deletions.
36 changes: 18 additions & 18 deletions src/collectives_c.c4
View file
Original file line number Diff line number Diff line change
Expand Up @@ -226,7 +226,7 @@ shmem_team_sync(shmem_team_t team)
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * \
SHMEM_REDUCE_SYNC_SIZE); \
SHMEM_ERR_CHECK_OVERLAP(target, source, sizeof(TYPE)*nreduce, \
sizeof(TYPE)*nreduce, 1); \
sizeof(TYPE)*nreduce, 1, 1); \
\
shmem_internal_op_to_all(target, source, nreduce, sizeof(TYPE), \
PE_start, 1 << logPE_stride, PE_size, \
Expand All @@ -244,7 +244,7 @@ shmem_team_sync(shmem_team_t team)
SHMEM_ERR_CHECK_SYMMETRIC(dest, sizeof(TYPE)*nreduce); \
SHMEM_ERR_CHECK_SYMMETRIC(source, sizeof(TYPE)*nreduce); \
SHMEM_ERR_CHECK_OVERLAP(dest, source, sizeof(TYPE)*nreduce, \
sizeof(TYPE)*nreduce, 1); \
sizeof(TYPE)*nreduce, 1, 1); \
TYPE *pWrk = NULL; \
\
shmem_internal_team_t *myteam = (shmem_internal_team_t *)team; \
Expand Down Expand Up @@ -290,7 +290,7 @@ shmem_broadcast32(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(target, nlong * 4);
SHMEM_ERR_CHECK_SYMMETRIC(source, nlong * 4);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long)*SHMEM_BCAST_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 4, nlong * 4, 1);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 4, nlong * 4, 1, 1);

shmem_internal_bcast(target, source, nlong * 4,
PE_root, PE_start, 1 << logPE_stride, PE_size,
Expand All @@ -309,7 +309,7 @@ shmem_broadcast64(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(target, nlong * 8);
SHMEM_ERR_CHECK_SYMMETRIC(source, nlong * 8);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long)*SHMEM_BCAST_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 8, nlong * 8, 1);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 8, nlong * 8, 1, 1);

shmem_internal_bcast(target, source, nlong * 8,
PE_root, PE_start, 1 << logPE_stride, PE_size,
Expand All @@ -325,7 +325,7 @@ shmem_broadcastmem(shmem_team_t team, void *dest, const void *source,
SHMEM_ERR_CHECK_TEAM_VALID(team);
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems);
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1, 1);

shmem_internal_team_t *myteam = (shmem_internal_team_t *)team;
long *psync = shmem_internal_team_choose_psync(myteam, BCAST);
Expand All @@ -351,7 +351,7 @@ shmem_broadcastmem(shmem_team_t team, void *dest, const void *source,
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * sizeof(TYPE), \
nelems * sizeof(TYPE), 1); \
nelems * sizeof(TYPE), 1, 1); \
\
shmem_internal_team_t *myteam = (shmem_internal_team_t *)team; \
long *psync = shmem_internal_team_choose_psync(myteam, BCAST); \
Expand All @@ -378,7 +378,7 @@ shmem_collect32(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(target, nlong * 4);
SHMEM_ERR_CHECK_SYMMETRIC(source, nlong * 4);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * SHMEM_COLLECT_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 4, nlong * 4, 1);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 4, nlong * 4, 1, 1);

shmem_internal_collect(target, source, nlong * 4,
PE_start, 1 << logPE_stride, PE_size, pSync);
Expand All @@ -394,7 +394,7 @@ shmem_collect64(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(target, nlong * 8);
SHMEM_ERR_CHECK_SYMMETRIC(source, nlong * 8);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * SHMEM_COLLECT_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 8, nlong * 8, 1);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 8, nlong * 8, 1, 1);

shmem_internal_collect(target, source, nlong * 8,
PE_start, 1 << logPE_stride, PE_size, pSync);
Expand All @@ -410,7 +410,7 @@ shmem_collect64(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * sizeof(TYPE), \
nelems * sizeof(TYPE), 1); \
nelems * sizeof(TYPE), 1, 1); \
\
shmem_internal_team_t *myteam = (shmem_internal_team_t *)team; \
long *psync = shmem_internal_team_choose_psync(myteam, \
Expand All @@ -432,7 +432,7 @@ shmem_collectmem(shmem_team_t team, void *dest, const void *source,
SHMEM_ERR_CHECK_TEAM_VALID(team);
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems);
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1, 1);

shmem_internal_team_t *myteam = (shmem_internal_team_t *)team;
long *psync = shmem_internal_team_choose_psync(myteam, COLLECT);
Expand All @@ -451,7 +451,7 @@ shmem_fcollect32(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(target, nlong * 4);
SHMEM_ERR_CHECK_SYMMETRIC(source, nlong * 4);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * SHMEM_COLLECT_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 4, nlong * 4, 1);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 4, nlong * 4, 1, 1);

shmem_internal_fcollect(target, source, nlong * 4,
PE_start, 1 << logPE_stride, PE_size, pSync);
Expand All @@ -467,7 +467,7 @@ shmem_fcollect64(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(target, nlong * 8);
SHMEM_ERR_CHECK_SYMMETRIC(source, nlong * 8);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * SHMEM_COLLECT_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 8, nlong * 8, 1);
SHMEM_ERR_CHECK_OVERLAP(target, source, nlong * 8, nlong * 8, 1, 1);

shmem_internal_fcollect(target, source, nlong * 8,
PE_start, 1 << logPE_stride, PE_size, pSync);
Expand All @@ -483,7 +483,7 @@ shmem_fcollect64(void *target, const void *source, size_t nlong,
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * sizeof(TYPE), \
nelems * sizeof(TYPE), 1); \
nelems * sizeof(TYPE), 1, 1); \
\
shmem_internal_team_t *myteam = (shmem_internal_team_t *)team; \
long *psync = shmem_internal_team_choose_psync(myteam, \
Expand All @@ -505,7 +505,7 @@ shmem_fcollectmem(shmem_team_t team, void *dest, const void *source,
SHMEM_ERR_CHECK_TEAM_VALID(team);
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems);
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1, 1);

shmem_internal_team_t *myteam = (shmem_internal_team_t *)team;
long *psync = shmem_internal_team_choose_psync(myteam, COLLECT);
Expand All @@ -524,7 +524,7 @@ shmem_alltoall32(void *dest, const void *source, size_t nelems, int PE_start,
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems * 4);
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems * 4);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * SHMEM_ALLTOALL_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * 4, nelems * 4, 1);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * 4, nelems * 4, 1, 1);

shmem_internal_alltoall(dest, source, nelems * 4,
PE_start, 1 << logPE_stride, PE_size, pSync);
Expand All @@ -540,7 +540,7 @@ shmem_alltoall64(void *dest, const void *source, size_t nelems, int PE_start,
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems * 8);
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems * 8);
SHMEM_ERR_CHECK_SYMMETRIC(pSync, sizeof(long) * SHMEM_ALLTOALL_SYNC_SIZE);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * 8, nelems * 8, 1);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * 8, nelems * 8, 1, 1);

shmem_internal_alltoall(dest, source, nelems * 8,
PE_start, 1 << logPE_stride, PE_size, pSync);
Expand All @@ -556,7 +556,7 @@ shmem_alltoall64(void *dest, const void *source, size_t nelems, int PE_start,
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems * sizeof(TYPE)); \
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems * sizeof(TYPE), \
nelems * sizeof(TYPE), 1); \
nelems * sizeof(TYPE), 1, 1); \
\
shmem_internal_team_t *myteam = (shmem_internal_team_t *)team; \
long *psync = shmem_internal_team_choose_psync(myteam, \
Expand All @@ -578,7 +578,7 @@ shmem_alltoallmem(shmem_team_t team, void *dest, const void *source,
SHMEM_ERR_CHECK_TEAM_VALID(team);
SHMEM_ERR_CHECK_SYMMETRIC(dest, nelems);
SHMEM_ERR_CHECK_SYMMETRIC(source, nelems);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1);
SHMEM_ERR_CHECK_OVERLAP(dest, source, nelems, nelems, 1, 1);

shmem_internal_team_t *myteam = (shmem_internal_team_t *)team;
long *psync = shmem_internal_team_choose_psync(myteam, ALLTOALL);
Expand Down
Loading

0 comments on commit 941a87d

Please sign in to comment.