SHMEM_SHORT_WAIT(3)

πŸ“¦
version 5.0.10 Β· 4 versions
, in ubuntu stonking

πŸ’– SHMEM_SHORT_WAIT πŸ’–

okay bestie, buckle up, because this one is SHMEM_SHORT_WAIT(3) from the Open MPI manual and it's literally the "hey, wake up, something changed" family of calls. no cap. 🫢

the whole girlies in this squad are: shmem_int_wait, shmem_int_wait_until, shmem_int4_wait, shmem_int4_wait_until, shmem_int8_wait, shmem_int8_wait_until, shmem_long_wait, shmem_long_wait_until, shmem_longlong_wait, shmem_longlong_wait_until, shmem_short_wait, shmem_short_wait_until, shmem_wait, and shmem_wait_until.

and what do they all do? Waits for a variable on the local processing element (PE) to change. that's the whole job. iconic behavior honestly. πŸ’…


✨ SYNOPSIS ✨

C or C++ πŸ“

#include <mpp/shmem.h>

void shmem_int_wait(volatile int *var, int value)

void shmem_int_wait_until(volatile int *var, int cond, int value)

void shmem_long_wait(volatile long *var, long value)

void shmem_long_wait_until(volatile long *var, int cond, long value)

void shmem_longlong_wait(volatile long long *var, long long value)

void shmem_longlong_wait_until(volatile long long *var, int cond,
  long long value)

void shmem_short_wait(volatile short *var, short value)

void shmem_short_wait_until(volatile short *var, int cond,
  short value)

void shmem_wait(volatile long *ivar, long cmp_value)

void shmem_wait_until(volatile long *ivar, int cmp, long value)

Fortran 🌈

INCLUDE "mpp/shmem.fh"

CALL SHMEM_INT4_WAIT(ivar, cmp_value)

CALL SHMEM_INT4_WAIT_UNTIL(ivar, cmp, cmp_value)

CALL SHMEM_INT8_WAIT(ivar, cmp_value)

CALL SHMEM_INT8_WAIT_UNTIL(ivar, cmp, cmp_value)

CALL SHMEM_WAIT(ivar, cmp_value)

CALL SHMEM_WAIT_UNTIL(ivar, cmp, cmp_value)

🌸 DESCRIPTION 🌸

so here's the tea. shmem_wait and shmem_wait_until are just sitting there, vibing, waiting for ivar to get changed by a remote write or an atomic swap issued by a different processor. it's giving point-to-point directed synchronization, and honestly? we love a directed-sync moment. πŸ‘‘

  • a call to shmem_wait will not come back until some other processor writes a value into ivar on the waiting processor that is not equal to cmp_value. like, it does not return. it's committed. periodt.
  • a call to shmem_wait_until will not come back until some other processor changes ivar so it satisfies the condition implied by cmp and cmp_value. no shortcuts, no half-returns.

this whole mechanism is your bestie when a processor needs to tell another processor that it finished doing a thing. πŸ’Œ

the arguments are as follows:

the arguments, bestie πŸŽ€

  • target β€” the remotely accessible integer data object to be updated on the remote PE. if you're in C/C++ land, the type of target should match what the SYNOPSIS says. if you're on a Fortran compiler, it must be of type integer with an element size of 4 bytes for SHMEM_INT4_ADD and 8 bytes for SHMEM_INT8_ADD.
  • value β€” the value to be atomically added to target. in C/C++, the type of value should match what the SYNOPSIS says. in Fortran, it must be of type integer with an element size of target.
  • pe β€” an integer that says which PE number target is getting updated on. if you're in Fortran, it must be a default integer value.
  • ivar β€” a remotely accessible integer variable that is being updated by another PE. in C/C++, the type of ivar should match what the SYNOPSIS says. in Fortran, ivar must be a specific sized integer type depending on which function you called, like so:
    • shmem_wait, shmem_wait_until:** default INTEGER
    • shmem_int4_wait, shmem_int4_wait_until:** INTEGER*4
    • shmem_int8_wait, shmem_int8_wait_until:** INTEGER*8
  • cmp β€” the compare operator that compares ivar with cmp_value. cmp must be of type integer. if you're in Fortran, it must be of default kind. if you're in C/C++, the type of cmp should match what the SYNOPSIS says. these are the cmp values that are supported:
    • SHMEM_CMP_EQ β€” Equal
    • SHMEM_CMP_NE β€” Not equal
    • SHMEM_CMP_GT β€” Greater than
    • SHMEM_CMP_LE β€” Less than or equal to
    • SHMEM_CMP_LT β€” Less than
    • SHMEM_CMP_GE β€” Greater than or equal to
  • cmp_value β€” cmp_value must be of type integer. in C/C++, the type of cmp_value should match what the SYNOPSIS says (yes, that implied, we're keeping it real πŸ’€). in Fortran, cmp_value must be an integer of the same size and kind as ivar. the shmem_wait routines return when ivar is no longer equal to cmp_value. the shmem_wait_until routines return when the compare condition is true. and that compare condition? it's defined by the ivar argument compared with cmp_value using the comparison operator, cmp. simple as. ✨

πŸ¦‹ EXAMPLES πŸ¦‹

Example 1: this call returns when variable ivar is not equal to 100:

INTEGER*8 IVAR

CALL SHMEM_INT8_WAIT(IVAR, INT8(100))

Example 2: this call to SHMEM_INT8_WAIT_UNTIL is equivalent to the call to SHMEM_INT8_WAIT in example 1:

INTEGER*8 IVAR

CALL SHMEM_INT8_WAIT_UNTIL(IVAR, SHMEM_CMP_NE, INT8(100))

Example 3: this C/C++ call waits until the sign bit in ivar is set by a transfer from a remote PE:

int ivar;

shmem_int_wait_until(&ivar, SHMEM_CMP_LT, 0);

Example 4: this Fortran example is in the context of a subroutine:

SUBROUTINE EXAMPLE()
  INTEGER FLAG_VAR
  COMMON/FLAG/FLAG_VAR
  . . .
  FLAG_VAR = FLAG_VALUE ! initialize the event variable
  . . .
  IF (FLAG_VAR .EQ. FLAG_VALUE) THEN
    CALL SHMEM_WAIT(FLAG_VAR, FLAG_VALUE)
  ENDIF
  FLAG_VAR = FLAG_VALUE ! reset the event variable for next time
  . . .
END

πŸ‘€ SEE ALSO πŸ‘€


2003-2026, The Open MPI Community


and that's the whole page bestie πŸ’œ this Open MPI moment dropped on Sep 10, 2026, brought to you by the Open MPI girlies, copyright 2003-2026, The Open MPI Community. go forth and wait on your variables, you absolute main character. πŸŒŸπŸŽ€πŸ’–

πŸ“¦ where this page comes from