Skip to content

Commit

Permalink
teams: avoid divide-by-zero if stride=0
Browse files Browse the repository at this point in the history 
Also, if split size is 1, internally set stride to 0
  • Loading branch information
@davidozog
davidozog committed Jun 17, 2024
1 parent 28b6361 commit 02aa405
Show file tree
Hide file tree
Showing 2 changed files with 17 additions and 13 deletions.
26 changes: 15 additions & 11 deletions src/collectives.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,8 @@ shmem_internal_build_kary_tree(int radix, int PE_start, int stride,

/* my_id is the index in a theoretical 0...N-1 array of
participating tasks. where the 0th entry is the root */
int my_id = (((shmem_internal_my_pe - PE_start) / stride) + PE_size - PE_root) % PE_size;
int my_id = (stride == 0) ? 0 :
(((shmem_internal_my_pe - PE_start) / stride) + PE_size - PE_root) % PE_size;

/* We shift PE_root to index 0, resulting in a PE active set layout of (for
example radix 2): 0 [ 1 2 ] [ 3 4 ] [ 5 6 ] ... The first group [ 1 2 ]
Expand Down Expand Up @@ -372,7 +373,7 @@ shmem_internal_sync_dissem(int PE_start, int PE_stride, int PE_size, long *pSync
{
int one = 1, neg_one = -1;
int distance, to, i;
int coll_rank = (shmem_internal_my_pe - PE_start) / PE_stride;
int coll_rank = (PE_stride == 0) ? 0 : (shmem_internal_my_pe - PE_start) / PE_stride;
int *pSync_ints = (int*) pSync;

/* need log2(num_procs) int slots. max_num_procs is
Expand Down Expand Up @@ -637,7 +638,7 @@ shmem_internal_op_to_all_ring(void *target, const void *source, size_t count, si
void *pWrk, long *pSync,
shm_internal_op_t op, shm_internal_datatype_t datatype)
{
int group_rank = (shmem_internal_my_pe - PE_start) / PE_stride;
int group_rank = (PE_stride == 0) ? 0 : (shmem_internal_my_pe - PE_start) / PE_stride;
long zero = 0, one = 1;

int peer = PE_start + ((group_rank + 1) % PE_size) * PE_stride;
Expand Down Expand Up @@ -840,7 +841,7 @@ shmem_internal_op_to_all_recdbl_sw(void *target, const void *source, size_t coun
void *pWrk, long *pSync,
shm_internal_op_t op, shm_internal_datatype_t datatype)
{
int my_id = ((shmem_internal_my_pe - PE_start) / PE_stride);
int my_id = (PE_stride == 0) ? 0 : ((shmem_internal_my_pe - PE_start) / PE_stride);
int log2_proc = 1, pow2_proc = 2;
int i = PE_size >> 1;
size_t wrk_size = type_size*count;
Expand Down Expand Up @@ -1072,7 +1073,8 @@ shmem_internal_fcollect_linear(void *target, const void *source, size_t len,
SHMEM_WAIT_UNTIL(pSync, SHMEM_CMP_EQ, 0);
} else {
/* Push data into the target */
size_t offset = ((shmem_internal_my_pe - PE_start) / PE_stride) * len;
size_t offset = (PE_stride == 0) ? 0 :
((shmem_internal_my_pe - PE_start) / PE_stride) * len;
shmem_internal_put_nb(SHMEM_CTX_DEFAULT, (char*) target + offset, source, len, PE_start,
&completion);
shmem_internal_put_wait(SHMEM_CTX_DEFAULT, &completion);
Expand Down Expand Up @@ -1104,7 +1106,7 @@ shmem_internal_fcollect_ring(void *target, const void *source, size_t len,
int i;
/* my_id is the index in a theoretical 0...N-1 array of
participating tasks */
int my_id = ((shmem_internal_my_pe - PE_start) / PE_stride);
int my_id = (PE_stride == 0) ? 0 : ((shmem_internal_my_pe - PE_start) / PE_stride);
int next_proc = PE_start + ((my_id + 1) % PE_size) * PE_stride;
long completion = 0;
long zero = 0, one = 1;
Expand Down Expand Up @@ -1157,7 +1159,7 @@ void
shmem_internal_fcollect_recdbl(void *target, const void *source, size_t len,
int PE_start, int PE_stride, int PE_size, long *pSync)
{
int my_id = ((shmem_internal_my_pe - PE_start) / PE_stride);
int my_id = (PE_stride == 0) ? 0 : ((shmem_internal_my_pe - PE_start) / PE_stride);
int i;
long completion = 0;
size_t curr_offset;
Expand Down Expand Up @@ -1214,7 +1216,7 @@ void
shmem_internal_alltoall(void *dest, const void *source, size_t len,
int PE_start, int PE_stride, int PE_size, long *pSync)
{
const int my_as_rank = (shmem_internal_my_pe - PE_start) / PE_stride;
const int my_as_rank = (PE_stride == 0) ? 0 : (shmem_internal_my_pe - PE_start) / PE_stride;
const void *dest_ptr = (uint8_t *) dest + my_as_rank * len;
int peer, start_pe, i;

Expand All @@ -1229,7 +1231,8 @@ shmem_internal_alltoall(void *dest, const void *source, size_t len,
PE_size);
peer = start_pe;
do {
int peer_as_rank = (peer - PE_start) / PE_stride; /* Peer's index in active set */
/* Peer's index in active set: */
int peer_as_rank = (PE_stride == 0) ? 0 : (peer - PE_start) / PE_stride;

shmem_internal_put_nbi(SHMEM_CTX_DEFAULT, (void *) dest_ptr, (uint8_t *) source + peer_as_rank * len,
len, peer);
Expand All @@ -1249,7 +1252,7 @@ shmem_internal_alltoalls(void *dest, const void *source, ptrdiff_t dst,
ptrdiff_t sst, size_t elem_size, size_t nelems,
int PE_start, int PE_stride, int PE_size, long *pSync)
{
const int my_as_rank = (shmem_internal_my_pe - PE_start) / PE_stride;
const int my_as_rank = (PE_stride == 0) ? 0 : (shmem_internal_my_pe - PE_start) / PE_stride;
const void *dest_base = (uint8_t *) dest + my_as_rank * nelems * dst * elem_size;
int peer, start_pe, i;

Expand All @@ -1273,7 +1276,8 @@ shmem_internal_alltoalls(void *dest, const void *source, ptrdiff_t dst,
peer = start_pe;
do {
size_t i;
int peer_as_rank = (peer - PE_start) / PE_stride; /* Peer's index in active set */
/* Peer's index in active set: */
int peer_as_rank = (PE_stride == 0) ? 0 : (peer - PE_start) / PE_stride;
uint8_t *dest_ptr = (uint8_t *) dest_base;
uint8_t *source_ptr = (uint8_t *) source + peer_as_rank * nelems * sst * elem_size;

Expand Down
4 changes: 2 additions & 2 deletions src/shmem_team.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -332,7 +332,7 @@ int shmem_internal_team_split_strided(shmem_internal_team_t *parent_team, int PE

myteam->my_pe = my_pe;
myteam->start = global_PE_start;
myteam->stride = PE_stride;
myteam->stride = (PE_size == 1) ? 0 : PE_stride;
myteam->size = PE_size;

if (config_mask == 0) {
Expand Down Expand Up @@ -424,7 +424,7 @@ int shmem_internal_team_split_strided(shmem_internal_team_t *parent_team, int PE
/* If no team was available, print some team triplet info and return nonzero. */
if (my_pe >= 0 && myteam != NULL && myteam->psync_idx == -1) {
RAISE_WARN_MSG("Team split strided failed: child <%d, %d, %d>, parent <%d, %d, %d>\n",
global_PE_start, PE_stride, PE_size,
myteam->start, myteam->stride, myteam->size,
parent_team->start, parent_team->stride, parent_team->size);
}
return *team_ret_val_reduced;
Expand Down

0 comments on commit 02aa405

Please sign in to comment.