.\" Man page generated from reStructuredText.
.
.
.nr rst2man-indent-level 0
.
.de1 rstReportMargin
\\$1 \\n[an-margin]
level \\n[rst2man-indent-level]
level margin: \\n[rst2man-indent\\n[rst2man-indent-level]]
-
\\n[rst2man-indent0]
\\n[rst2man-indent1]
\\n[rst2man-indent2]
..
.de1 INDENT
.\" .rstReportMargin pre:
. RS \\$1
. nr rst2man-indent\\n[rst2man-indent-level] \\n[an-margin]
. nr rst2man-indent-level +1
.\" .rstReportMargin post:
..
.de UNINDENT
. RE
.\" indent \\n[an-margin]
.\" old: \\n[rst2man-indent\\n[rst2man-indent-level]]
.nr rst2man-indent-level -1
.\" new: \\n[rst2man-indent\\n[rst2man-indent-level]]
.in \\n[rst2man-indent\\n[rst2man-indent-level]]u
..
.TH "INTRO_SHMEM" "3" "Apr 11, 2024" "" "Open MPI"
.sp
intro_shmem \- Introduction to the OpenSHMEM programming model
.SH DESCRIPTION
.sp
The SHMEM programming model consists of library routines that provide
low\-latency, high\-bandwidth communication for use in highly parallelized
scalable programs. The routines in the OpenSHMEM application programming
interface (API) provide a programming model for exchanging data between
cooperating parallel processes. The resulting programs are similar in
style to Message Passing Interface (MPI) programs. The SHMEM API can be
used either alone or in combination with MPI routines in the same
parallel program.
.sp
An OpenSHMEM program is SPMD (single program, multiple data) in style.
The SHMEM processes, called processing elements or PEs, all start at the
same time and they all run the same program. Usually the PEs perform
computation on their own subdomains of the larger problem and
periodically communicate with other PEs to exchange information on which
the next computation phase depends.
.sp
The OpenSHMEM routines minimize the overhead associated with data
transfer requests, maximize bandwidth and minimize data latency. Data
latency is the period of time that starts when a PE initiates a transfer
of data and ends when a PE can use the data. OpenSHMEM routines support
remote data transfer through put operations, which transfer data to a
different PE, get operations, which transfer data from a different PE,
and remote pointers, which allow direct references to data objects owned
by another PE. Other operations supported are collective broadcast and
reduction, barrier synchronization, and atomic memory operations. An
atomic memory operation is an atomic read\-and\-update operation, such as
a fetch\-and\-increment, on a remote or local data object.
.SH OPENSHMEM ROUTINES
.sp
This section lists the significant OpenSHMEM message\-passing routines.
.sp
PE queries
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI_num_pes\fP(3)
.IP \(bu 2
\fI_my_pe\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fINUM_PES\fP(3)
.IP \(bu 2
\fIMY_PE\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Elemental data put routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_p\fP(3)
.IP \(bu 2
\fI\%shmem_float_p\fP(3)
.IP \(bu 2
\fI\%shmem_int_p\fP(3)
.IP \(bu 2
\fI\%shmem_long_p\fP(3)
.IP \(bu 2
\fI\%shmem_short_p\fP\&.*(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Block data put routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_put32\fP(3)
.IP \(bu 2
\fI\%shmem_put64\fP(3)
.IP \(bu 2
\fI\%shmem_put128\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_put\fP(3)
.IP \(bu 2
\fI\%shmem_float_put\fP(3)
.IP \(bu 2
\fI\%shmem_int_put\fP(3)
.IP \(bu 2
\fI\%shmem_long_put\fP(3)
.IP \(bu 2
\fI\%shmem_short_put\fP\&.*(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_complex_put(3)
.IP \(bu 2
shmem_integer_put(3)
.IP \(bu 2
shmem_logical_put(3)
.IP \(bu 2
shmem_real_put(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Elemental data get routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_g\fP(3)
.IP \(bu 2
\fI\%shmem_float_g\fP(3)
.IP \(bu 2
\fI\%shmem_int_g\fP(3)
.IP \(bu 2
\fI\%shmem_long_g\fP(3)
.IP \(bu 2
\fI\%shmem_short_g\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Block data get routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_get32\fP(3)
.IP \(bu 2
\fI\%shmem_get64\fP(3)
.IP \(bu 2
\fI\%shmem_get128\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_get\fP(3)
.IP \(bu 2
\fI\%shmem_float_get\fP(3)
.IP \(bu 2
\fI\%shmem_int_get\fP(3)
.IP \(bu 2
\fI\%shmem_long_get\fP(3)
.IP \(bu 2
\fI\%shmem_short_get\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_complex_get(3)
.IP \(bu 2
shmem_integer_get(3)
.IP \(bu 2
shmem_logical_get(3)
.IP \(bu 2
shmem_real_get(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Strided put routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_iput32\fP(3)
.IP \(bu 2
\fI\%shmem_iput64\fP(3)
.IP \(bu 2
\fI\%shmem_iput128\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_iput\fP(3)
.IP \(bu 2
\fI\%shmem_float_iput\fP(3)
.IP \(bu 2
\fI\%shmem_int_iput\fP(3)
.IP \(bu 2
\fI\%shmem_long_iput\fP(3)
.IP \(bu 2
\fI\%shmem_short_iput\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_complex_iput(3)
.IP \(bu 2
shmem_integer_iput(3)
.IP \(bu 2
shmem_logical_iput(3)
.IP \(bu 2
shmem_real_iput(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Strided get routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_iget32\fP(3)
.IP \(bu 2
\fI\%shmem_iget64\fP(3)
.IP \(bu 2
\fI\%shmem_iget128\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_iget\fP(3)
.IP \(bu 2
\fI\%shmem_float_iget\fP(3)
.IP \(bu 2
\fI\%shmem_int_iget\fP(3)
.IP \(bu 2
\fI\%shmem_long_iget\fP(3)
.IP \(bu 2
\fI\%shmem_short_iget\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_complex_iget(3)
.IP \(bu 2
shmem_integer_iget(3)
.IP \(bu 2
shmem_logical_iget(3)
.IP \(bu 2
shmem_real_iget(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Point\-to\-point synchronization routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_int_wait\fP(3)
.IP \(bu 2
\fI\%shmem_int_wait_until\fP(3)
.IP \(bu 2
\fI\%shmem_long_wait\fP(3)
.IP \(bu 2
\fI\%shmem_long_wait_until\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_wait\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_wait_until\fP(3)
.IP \(bu 2
\fI\%shmem_short_wait\fP(3)
.IP \(bu 2
\fI\%shmem_short_wait_until\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_int4_wait(3)
.IP \(bu 2
shmem_int4_wait_until(3)
.IP \(bu 2
shmem_int8_wait(3)
.IP \(bu 2
shmem_int8_wait_until(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Barrier synchronization routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_barrier_all\fP(3)
.IP \(bu 2
\fI\%shmem_barrier\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Atomic memory fetch\-and\-operate (fetch\-op) routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_swap\fP
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Reduction routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_int_and_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_and_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_and_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_and_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_double_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_float_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_int_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_double_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_float_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_int_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_double_sum_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_float_sum_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_int_sum_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_sum_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_sum_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_sum_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_double_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_float_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_int_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_int_or_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_or_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_or_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_or_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_int_xor_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_long_xor_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_xor_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_short_xor_to_all\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_int4_and_to_all(3)
.IP \(bu 2
shmem_int8_and_to_all(3)
.IP \(bu 2
shmem_real4_max_to_all(3)
.IP \(bu 2
shmem_real8_max_to_all(3)
.IP \(bu 2
shmem_int4_max_to_all(3)
.IP \(bu 2
shmem_int8_max_to_all(3)
.IP \(bu 2
shmem_real4_min_to_all(3)
.IP \(bu 2
shmem_real8_min_to_all(3)
.IP \(bu 2
shmem_int4_min_to_all(3)
.IP \(bu 2
shmem_int8_min_to_all(3)
.IP \(bu 2
shmem_real4_sum_to_all(3)
.IP \(bu 2
shmem_real8_sum_to_all(3)
.IP \(bu 2
shmem_int4_sum_to_all(3)
.IP \(bu 2
shmem_int8_sum_to_all(3)
.IP \(bu 2
shmem_real4_prod_to_all(3)
.IP \(bu 2
shmem_real8_prod_to_all(3)
.IP \(bu 2
shmem_int4_prod_to_all(3)
.IP \(bu 2
shmem_int8_prod_to_all(3)
.IP \(bu 2
shmem_int4_or_to_all(3)
.IP \(bu 2
shmem_int8_or_to_all(3)
.IP \(bu 2
shmem_int4_xor_to_all(3)
.IP \(bu 2
shmem_int8_xor_to_all(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Broadcast routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_broadcast32\fP(3)
.IP \(bu 2
\fI\%shmem_broadcast64\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Cache management routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_udcflush\fP(3)
.IP \(bu 2
\fI\%shmem_udcflush_line\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Byte\-granularity block put routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_putmem\fP(3)
.IP \(bu 2
\fI\%shmem_getmem\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_character_put(3)
.IP \(bu 2
shmem_character_get(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Collect routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_collect32\fP(3)
.IP \(bu 2
\fI\%shmem_collect64\fP(3)
.IP \(bu 2
\fI\%shmem_fcollect32\fP(3)
.IP \(bu 2
\fI\%shmem_fcollect64\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Atomic memory fetch\-and\-operate (fetch\-op) routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_double_swap\fP(3)
.IP \(bu 2
\fI\%shmem_float_swap\fP(3)
.IP \(bu 2
\fI\%shmem_int_cswap\fP(3)
.IP \(bu 2
\fI\%shmem_int_fadd\fP(3)
.IP \(bu 2
\fI\%shmem_int_finc\fP(3)
.IP \(bu 2
\fI\%shmem_int_swap\fP(3)
.IP \(bu 2
\fI\%shmem_long_cswap\fP(3)
.IP \(bu 2
\fI\%shmem_long_fadd\fP(3)
.IP \(bu 2
\fI\%shmem_long_finc\fP(3)
.IP \(bu 2
\fI\%shmem_long_swap\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_cswap\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_fadd\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_finc\fP(3)
.IP \(bu 2
\fI\%shmem_longlong_swap\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_int4_cswap(3)
.IP \(bu 2
shmem_int4_fadd(3)
.IP \(bu 2
shmem_int4_finc(3)
.IP \(bu 2
shmem_int4_swap(3)
.IP \(bu 2
shmem_int8_swap(3)
.IP \(bu 2
shmem_real4_swap(3)
.IP \(bu 2
shmem_real8_swap(3)
.IP \(bu 2
shmem_int8_cswap(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Atomic memory operation routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_int4_add(3)
.IP \(bu 2
shmem_int4_inc(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Remote memory pointer function
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_ptr\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Reduction routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_longdouble_max_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longdouble_min_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longdouble_prod_to_all\fP(3)
.IP \(bu 2
\fI\%shmem_longdouble_sum_to_all\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.IP \(bu 2
Fortran only:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
shmem_real16_max_to_all(3)
.IP \(bu 2
shmem_real16_min_to_all(3)
.IP \(bu 2
shmem_real16_prod_to_all(3)
.IP \(bu 2
shmem_real16_sum_to_all(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Accessibility query routines
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
C/C++ and Fortran:
.INDENT 2.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
\fI\%shmem_pe_accessible\fP(3)
.IP \(bu 2
\fI\%shmem_addr_accessible\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.UNINDENT
.sp
Symmetric Data Objects
.sp
Consistent with the SPMD nature of the OpenSHMEM programming model is
the concept of symmetric data objects. These are arrays or variables
that exist with the same size, type, and relative address on all PEs.
Another term for symmetric data objects is “remotely accessible data
objects”. In the interface definitions for OpenSHMEM data transfer
routines, one or more of the parameters are typically required to be
symmetric or remotely accessible.
.sp
The following kinds of data objects are symmetric:
.INDENT 0.0
.INDENT 3.5
.INDENT 0.0
.IP \(bu 2
Fortran data objects in common blocks or with the SAVE
attribute. These data objects must not be defined in a dynamic
shared object (DSO).
.IP \(bu 2
Non\-stack C and C++ variables. These data objects must not be
defined in a DSO.
.IP \(bu 2
Fortran arrays allocated with \fIshpalloc\fP(3)
.IP \(bu 2
C and C++ data allocated by \fIshmalloc\fP(3)
.UNINDENT
.UNINDENT
.UNINDENT
.INDENT 0.0
.TP
.B Collective Routines
Some SHMEM routines, for example, shmem_broadcast(3) and
\fI\%shmem_float_sum_to_all\fP(3), are classified as collective routines
because they distribute work across a set of PEs. They must be called
concurrently by all PEs in the active set defined by the PE_start,
logPE_stride, PE_size argument triplet. The following man pages
describe the OpenSHMEM collective routines:
.INDENT 7.0
.IP \(bu 2
shmem_and(3)
.IP \(bu 2
\fI\%shmem_barrier\fP(3)
.IP \(bu 2
shmem_broadcast(3)
.IP \(bu 2
shmem_collect(3)
.IP \(bu 2
shmem_max(3)
.IP \(bu 2
shmem_min(3)
.IP \(bu 2
shmem_or(3)
.IP \(bu 2
shmem_prod(3)
.IP \(bu 2
shmem_sum(3)
.IP \(bu 2
shmem_xor(3)
.UNINDENT
.UNINDENT
.SH USING THE SYMMETRIC WORK ARRAY, PSYNC
.sp
Multiple pSync arrays are often needed if a particular PE calls as
OpenSHMEM collective routine twice without intervening barrier
synchronization. Problems would occur if some PEs in the active set for
call 2 arrive at call 2 before processing of call 1 is complete by all
PEs in the call 1 active set. You can use \fI\%shmem_barrier\fP(3) or
\fI\%shmem_barrier_all\fP(3) to perform a barrier synchronization between
consecutive calls to OpenSHMEM collective routines.
.sp
There are two special cases:
.INDENT 0.0
.IP \(bu 2
The \fI\%shmem_barrier\fP(3) routine allows the same pSync array to be
used on consecutive calls as long as the active PE set does not
change.
.IP \(bu 2
If the same collective routine is called multiple times with the same
active set, the calls may alternate between two pSync arrays. The
SHMEM routines guarantee that a first call is completely finished by
all PEs by the time processing of a third call begins on any PE.
.UNINDENT
.sp
Because the SHMEM routines restore pSync to its original contents,
multiple calls that use the same pSync array do not require that pSync
be reinitialized after the first call.
.SH SHMEM ENVIRONMENT VARIABLES
.sp
This section lists the significant SHMEM environment variables.
.INDENT 0.0
.IP \(bu 2
\fBSMA_VERSION\fP print the library version at start\-up.
.IP \(bu 2
\fBSMA_INFO\fP print helpful text about all these environment
variables.
.IP \(bu 2
\fBSMA_SYMMETRIC_SIZE\fP number of bytes to allocate for the symmetric
heap.
.IP \(bu 2
\fBSMA_DEBUG\fP enable debugging messages.
.UNINDENT
.sp
The first call to SHMEM must be \fIstart_pes\fP(3). This routines
initialize the SHMEM runtime.
.sp
Calling any other SHMEM routines beforehand has undefined behavior.
Multiple calls to this routine is not allowed.
.SH COMPILING AND RUNNING OPENSHMEM PROGRAMS
.sp
The OpenSHMEM specification is silent regarding how OpenSHMEM programs
are compiled, linked and run. This section shows some examples of how
wrapper programs could be utilized to compile and launch applications.
The commands are styled after wrapper programs found in many MPI
implementations.
.sp
The following sample command line demonstrates running an OpenSHMEM
Program using a wrapper script (\fBoshrun\fP in this case):
.INDENT 0.0
.IP \(bu 2
C/C++:
.UNINDENT
.INDENT 0.0
.INDENT 3.5
.sp
.EX
oshcc c_program.c
.EE
.UNINDENT
.UNINDENT
.INDENT 0.0
.IP \(bu 2
FORTRAN:
.UNINDENT
.INDENT 0.0
.INDENT 3.5
.sp
.EX
oshfort fortran_program.f
.EE
.UNINDENT
.UNINDENT
.sp
The following sample command line demonstrates running an OpenSHMEM
Program assuming that the library provides a wrapper script for such
purpose (named \fBoshrun\fP for this example):
.INDENT 0.0
.INDENT 3.5
.sp
.EX
oshrun \-n 32 ./a.out
.EE
.UNINDENT
.UNINDENT
.SH EXAMPLES
.sp
\fBExample 1\fP: The following Fortran OpenSHMEM program directs all PEs
to sum simultaneously the numbers in the VALUES variable across all PEs:
.INDENT 0.0
.INDENT 3.5
.sp
.EX
PROGRAM REDUCTION
  REAL VALUES, SUM
  COMMON /C/ VALUES
  REAL WORK

  CALL START_PES(0)
  VALUES = MY_PE()
  CALL SHMEM_BARRIER_ALL ! Synchronize all PEs
  SUM = 0.0
  DO I = 0, NUM_PES()\-1
    CALL SHMEM_REAL_GET(WORK, VALUES, 1, I) ! Get next value
    SUM = SUM + WORK                ! Sum it
  ENDDO
  PRINT *, \(aqPE \(aq, MY_PE(), \(aq COMPUTED SUM=\(aq, SUM
  CALL SHMEM_BARRIER_ALL
END
.EE
.UNINDENT
.UNINDENT
.sp
\fBExample 2\fP: The following C OpenSHMEM program transfers an array of
10 longs from PE 0 to PE 1:
.INDENT 0.0
.INDENT 3.5
.sp
.EX
#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 elements into target on PE 1 */
    shmem_long_put(target, source, 10, 1);
  }
  shmem_barrier_all(); /* sync sender and receiver */
  if (shmem_my_pe() == 1)
    printf(\(dqtarget[0] on PE %d is %d\en\(dq, shmem_my_pe(), target[0]);
}
.EE
.UNINDENT
.UNINDENT
.sp
\fBSEE ALSO:\fP
.INDENT 0.0
.INDENT 3.5
The following man pages also contain information on OpenSHMEM routines.
See the specific man pages for implementation information.
.sp
shmem_add(3) shmem_and(3) \fI\%shmem_barrier\fP(3)
\fI\%shmem_barrier_all\fP(3) shmem_broadcast(3) shmem_cache(3)
shmem_collect(3) shmem_cswap(3) shmem_fadd(3)
\fI\%shmem_fence\fP(3) shmem_finc(3) shmem_get(3) shmem_iget(3)
shmem_inc(3) shmem_iput(3) shmem_lock(3) shmem_max(3)
shmem_min(3) \fI\%shmem_my_pe\fP(3) shmem_or(3) shmem_prod(3)
shmem_put(3) \fI\%shmem_quiet\fP(3) \fI\%shmem_short_g\fP(3)
\fI\%shmem_short_p\fP(3) shmem_sum(3) \fI\%shmem_swap\fP(3)
\fI\%shmem_wait\fP(3) shmem_xor(3) \fI\%shmem_pe_accessible\fP(3)
\fI\%shmem_addr_accessible\fP(3) \fI\%shmem_init\fP(3) \fI\%shmem_malloc\fP(3)
\fIshmem_my_pe\fP(3) \fIshmem_n_pes\fP(3)
.UNINDENT
.UNINDENT
.SH COPYRIGHT
2003-2024, The Open MPI Community
.\" Generated by docutils manpage writer.
.