SHMEM_PUT8_NBI(3)
π shmem_put8_nbi(3) β the nonblocking put girlies π
okay so bestie, you're in the Open MPI manual, section 3, and this page is a whole squad of routines, not just one. Let me introduce everyone:
shmem_putmem_nbi(3),shmem_char_put_nbi(3),shmem_short_put_nbi(3),shmem_int_put_nbi(3),shmem_long_put_nbi(3),shmem_longlong_put_nbi(3),shmem_float_put_nbi(3),shmem_double_put_nbi(3),shmem_longdouble_put_nbi(3),shmem_put8_nbi(3),shmem_put16_nbi(3),shmem_put32_nbi(3),shmem_put64_nbi(3),shmem_put128_nbi(3)
and what do they all do? β¨ The nonblocking put routines provide a method for copying data from a contiguous local data object to a data object on a specified PE. that's the whole vibe. π«
β¨ SYNOPSIS β¨
π©΅ C or C++
#include <mpp/shmem.h>
void shmem_putmem_nbi(void *dest, const void *source,
size_t nelems, int pe)
void shmem_char_put(char *dest, const char *source,
size_t nelems, int pe)
void shmem_short_put(short *dest, const short *source,
size_t nelems, int pe)
void shmem_int_put(int *dest, const int *source,
size_t nelems, int pe)
void shmem_long_put(long *dest, const long *source,
size_t nelems, int pe)
void shmem_longlong_put(long long *dest, const long long *source,
size_t nelems, int pe)
void shmem_float_put(float *dest, const float *source,
size_t nelems, int pe)
void shmem_double_put(double *dest, const double *source,
size_t nelems, int pe)
void shmem_longdouble_put(long double *dest, const long double *source,
size_t nelems, int pe)
void shmem_put8(void *dest, const void *source,
size_t nelems, int pe)
void shmem_put16(void *dest, const void *source,
size_t nelems, int pe)
void shmem_put32(void *dest, const void *source,
size_t nelems, int pe)
void shmem_put64(void *dest, const void *source,
size_t nelems, int pe)
void shmem_put128(void *dest, const void *source,
size_t nelems, int pe)
(yes bb, type them exactly like that β the names as written, no edits, no "fixing", i don't make the rules i just enforce them π )
π©΅ Fortran
INCLUDE "mpp/shmem.fh"
INTEGER nelems, pe
CALL SHMEM_PUTMEM_NBI(dest, source, nelems, pe)
CALL SHMEM_CHARACTER_PUT_NBI(dest, source, nelems, pe)
CALL SHMEM_COMPLEX_PUT_NBI(dest, source, nelems, pe)
CALL SHMEM_DOUBLE_PUT_NBI(dest, source, nelems, pe)
CALL SHMEM_INTEGER_PUT_NBI(dest, source, nelems, pe)
CALL SHMEM_LOGICAL_PUT_NBI(dest, source, nelems, pe)
CALL SHMEM_REAL_PUT_NBI(dest, source, nelems, pe)
CALL SHMEM_PUT4_NBI(dest, source, nelems, pe)
CALL SHMEM_PUT8_NBI(dest, source, nelems, pe)
CALL SHMEM_PUT32_NBI(dest, source, nelems, pe)
CALL SHMEM_PUT64_NBI(dest, source, nelems, pe)
CALL SHMEM_PUT128_NBI(dest, source, nelems, pe)
πΈ DESCRIPTION πΈ
okay so here's the β¨teaβ¨. The routines return after posting the operation. like, they hand off the mission and dip β the work is not done, bestie, it's just scheduled. π«Ά
The operation is considered complete after a subsequent call to shmem_quiet. At the completion of shmem_quiet, the data has been copied into the dest array on the destination PE. so if you're printing before the quiet, you're printing lies. π
and here's the chaotic part: the delivery of data words into the data object on the destination PE may occur in any order. fr fr, any order. Furthermore, two successive put routines may deliver data out of order unless a call to shmem_fence is introduced between the two calls. so if order matters to you β and it does, bestie, it always does β you need that fence. no cap. π§
π The arguments π
destβ Data object to be updated on the remote PE. This data object must be remotely accessible. this one's the main character on the other side of the wire, she has to be reachable or the whole thing falls apart πsourceβ Data object containing the data to be copied. your local stash. the goods. π¦nelemsβ Number of elements in thedestandsourcearrays.nelemsmust be of typesize_tfor C. If you are using Fortran, it must be a constant, variable, or array element of default integer type. (C girlies:size_t. Fortran girlies: default integer, no fancy kinds. iykyk.) π’peβ PE number of the remote PE.pemust be of typeinteger. If you are using Fortran, it must be a constant, variable, or array element of default integer type. π―
and a little reminder bestie: if you are using Fortran, data types must be of default size. For example, arealvariable must be declared asREAL,REAL*4, orREAL(KIND=4). that's the vibe check, don't fail it. π
π¦ NOTES π¦
See intro_shmem(3) for a definition of the term remotely accessible. like, genuinely go read it, it's the lore episode. πβ¨
π EXAMPLES π
Consider this simple example for C. it's giving "small but iconic" and honestly? ate. π
#include <stdio.h>
#include <mpp/shmem.h>
main()
{
long source[10] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 };
static long target[10];
shmem_init();
if (shmem_my_pe() == 0) {
/* put 10 words into target on PE 1 */
shmem_long_put_nbi(target, source, 10, 1);
shmem_quiet();
}
shmem_barrier_all(); /* sync sender and receiver */
if (shmem_my_pe() == 1)
shmem_udcflush(); /* not required on Altix systems */
printf("target[0] on PE %d is %d\n", shmem_my_pe(), target[0]);
}
see how PE 0 posts the put and then calls shmem_quiet() before the barrier? that's the move. the barrier syncs sender and receiver, and shmem_udcflush() there is not required on Altix systems β noted, don't @ me. π§
π SEE ALSO π
π COPYRIGHT π
2003-2026, The Open MPI Community
and that's the page, bestie!! π mwah π
this lil masterpiece dropped May 14, 2026, straight from the Open MPI girlies, page SHMEM_PUT8_NBI(3) β go forth and put nonblockingly, and remember to shmem_quiet before you come for the data πβ¨π¦
π¦ where this page comes from
- arch rolling: nvhpc-comm-libs 26.9-1 Β· LicenseRef-NVIDIA-HPC-SDK
- π girlified by deepseek-ai/DeepSeek-V4.1-Flash-1017 on : a language model's rewrite of the original, under its terms