add support for request_get_status_any/all/some #13279

edgargabriel · 2025-05-26T20:44:51Z

Add support for the new MPI_Request_get_status_any/all/some functions introduced with MPI 4.1

This PR contains the C implementation for the functions as well as the man-pages.
There are however also some things missing:

in the C code generation python scripts, we need to introduce the ability to create const MPI_Request[], (I have currently removed the const from the function definitions in mpi.h.in in order to make the code compile).
Fortran interfaces

hppritcha · 2025-05-26T23:51:34Z

Let me know if you want help with the python binding infrastructure.

edgargabriel · 2025-05-27T13:20:25Z

Let me know if you want help with the python binding infrastructure.

@hppritcha yes, I would appreciate some help on that front (and the same applies for the fortran bindings, even if its just a conversation that I understand what needs to be done).

ompi/mpi/c/request_get_status_all.c.in

hppritcha · 2025-05-27T18:19:36Z

@edgargabriel added new data type to the bindings code to add this new use of const MPI_Request reqs[] in the standard.

hppritcha · 2025-05-27T18:20:02Z

I'll work on the fortran but that will have to wait till later in the week.

hppritcha · 2025-05-27T18:24:29Z

note that mpi4py is probably not testing this since it keys off our MPI_Get_version return values for major/minor.
@edgargabriel have you tested mpi4py with VERSION faked to return 4.1?

edgargabriel · 2025-05-28T13:25:27Z

@hppritcha thank you very much, I will test it later this week. I have not done the mpi4py test that you mentioned, will look into this as well.

Signed-off-by: Edgar Gabriel <edgar.gabriel@amd.com>

add the man pages for the newly implemented MPI_Request_get_status_all/any/some functions. Signed-off-by: Edgar Gabriel <edgar.gabriel@amd.com>

Three new functions were added to the MPI API as part of the 4.1 standard. These used an array of const MPI_Request s which the bindings code didn't support. This commit also fixes back the mpi.h prototypes to include the constants and required changes to the new template files. Signed-off-by: Howard Pritchard <howardp@lanl.gov>

edgargabriel · 2025-06-11T22:22:18Z

note that mpi4py is probably not testing this since it keys off our MPI_Get_version return values for major/minor. @edgargabriel have you tested mpi4py with VERSION faked to return 4.1?

faking MPI 4.1. for mpi4py doesn't work unfortunately, we fail the compilation of mpi4py due to other missing function (MPI_Buffer_flush etc.). I will write some tests for the new functions.

also switch MPI_Request_get_status to use new method for generating f08 bindings. Update fortran bindings interfaces generation code. f90/f77 interfaces will be added as another commit. Signed-off-by: Howard Pritchard <howardp@lanl.gov>

for new fortran methods starting with MPI_Request_get_status_(any/all/some) plus switch MPI_Request_get_status to use the binding infrastructure as well. Signed-off-by: Howard Pritchard <howardp@lanl.gov>

Signed-off-by: Howard Pritchard <howardp@lanl.gov>

hppritcha · 2025-06-19T19:55:26Z

@edgargabriel i pushed the remainder of the fortran related files to the PR - hopefully.

edgargabriel · 2025-06-19T20:28:58Z

@edgargabriel i pushed the remainder of the fortran related files to the PR - hopefully.

@hppritcha thank you! I hope to finish up adding tests to the ibm testsuite for the new function this weekend, and hopefully next week we could merge the PR if all tests pass.

edgargabriel · 2025-06-21T23:07:13Z

I added some tests for request_get_status_any/all/some to the ompi-tests private testsuite, and they pass. This is however only testing the C-interfaces of the function. The tests are also probably not entirely bullet-proof in terms of that they might miss some corner cases, but the base function is shown to be working.

jsquyres · 2025-07-07T19:22:27Z

ompi/mpi/bindings/bindings.py

+    parser_interface.add_argument('--fort-std', choices=('f90', 'f08'),
+                             help='fortran standard to use for fortran module generation')


@hppritcha I see --fort-std added in both parser_code and parser_interface -- is that intentional?

yes it is. i can look if it makes sense to put in the higher level "fortran parser".

jsquyres · 2025-07-07T19:23:00Z

ompi/mpi/bindings/bindings.py

@@ -37,12 +37,16 @@ def main():
    subparsers_fortran = parser_fortran.add_subparsers()
    parser_code = subparsers_fortran.add_parser('code', help='generate binding code')
    parser_code.set_defaults(handler=lambda args, out: fortran.generate_code(args, out))
-    parser_code.add_argument('--lang', choices=('fortran', 'c'),
+    parser_code.add_argument('--lang', choices=('fortran', 'c', 'f90'),


@hppritcha What is the f90 choice for? I don't see that being checked for in the python code -- am I missing it?

its implied in the check around line 297 in fortran.py but i'll make it explicit. its suppose to mean generate the "f90" aka use mpi fortran interfaces.

jsquyres · 2025-07-07T19:44:59Z

ompi/mpi/c/request_get_status_all.c.in

+    int i;
+    all_done = true;
+    for (i = 0; i < count; i++) {
+        one_done = false;


How is one_done ever set to true? I think the "if" conditional below is incorrect.

jsquyres · 2025-07-07T19:55:32Z

ompi/mpi/c/request_get_status_all.c.in

+    bool one_done;
+
+#if OPAL_ENABLE_PROGRESS_THREADS == 0
+    int do_it_once = 0;


Is there a reason to make this an int instead of bool?

jsquyres · 2025-07-07T19:58:41Z

docs/man-openmpi/man3/MPI_Request_get_status_all.3.rst

+:ref:`MPI_Request_get_status_all` |mdash| Access information associated with a
+request without freeing the request.


You're access information with all active requests in an array -- not just a request.

jsquyres · 2025-07-07T20:00:50Z

docs/man-openmpi/man3/MPI_Request_get_status_all.3.rst

+status entries are undefined.
+
+If your application does not need to examine the *status* field, you can
+save resources by using the predefined constant ``MPI_STATUS_IGNORE`` as a


Did you mean MPI_STATUSES_IGNORE?

jsquyres · 2025-07-07T20:06:30Z

ompi/mpi/c/request_get_status_all.c.in

@@ -54,7 +54,7 @@ PROTOTYPE ERROR_CLASS request_get_status_all(INT count, REQUEST_INOUT requests:c
        if (NULL == requests) {
            rc = MPI_ERR_REQUEST;
        } else {
-            if(!ompi_request_check_same_instance(requests, count) ) {
+            if(!ompi_request_check_same_instance((MPI_Request *)requests, count) ) {


Why cast? Is there a reason the 1st arg to ompi_request_check_same_instance isn't const?

jsquyres · 2025-07-07T20:10:21Z

ompi/mpi/fortran/use-mpi-f08/request_get_status_all.c.in

+    if (NULL != ierr) *ierr = OMPI_INT_2_FINT(c_ierr);
+
+    if (MPI_SUCCESS == c_ierr) {
+        *flag = OMPI_INT_2_FINT(c_flag);


flag is a LOGICAL.

same below. you need to look at the generated fortran code to see how this works.

jsquyres · 2025-07-07T20:11:05Z

ompi/mpi/fortran/use-mpi-f08/request_get_status_any.c.in

+    if (MPI_SUCCESS == c_ierr) {
+
+        *indx = OMPI_INT_2_FINT(c_indx);
+        *flag = OMPI_INT_2_FINT(c_flag);


flag is a LOGICAL.

nope it isn't there. See how the inner fortran binding works once you build the code out.

jsquyres · 2025-07-07T20:24:11Z

ompi/mpi/fortran/mpif-h/request_get_status_some_f.c

+
+    if (MPI_SUCCESS == c_ierr) {
+        OMPI_SINGLE_INT_2_FINT(outcount);
+        OMPI_ARRAY_INT_2_FINT(array_of_indices, *incount);


Should this be *outcount?

oh yes. it should be *outcount

edgargabriel added the ⚠️ WIP-DNM! label May 26, 2025

github-actions bot added the Target: main label May 26, 2025

edgargabriel force-pushed the topic/request-status-mult branch from d233c73 to 723b84e Compare May 26, 2025 20:46

jsquyres added this to MPI 4.1 compliance May 26, 2025

jsquyres moved this to In Progress in MPI 4.1 compliance May 26, 2025

jsquyres marked this pull request as draft May 26, 2025 20:47

jsquyres added this to the v6.0.0 milestone May 27, 2025

jsquyres reviewed May 27, 2025

View reviewed changes

ompi/mpi/c/request_get_status_all.c.in Outdated Show resolved Hide resolved

edgargabriel force-pushed the topic/request-status-mult branch 2 times, most recently from efd9cf5 to f21dff7 Compare May 27, 2025 13:32

edgargabriel and others added 3 commits June 11, 2025 21:30

mpi: add request_get_status_all/any/some

37d0a54

Signed-off-by: Edgar Gabriel <edgar.gabriel@amd.com>

docs: add man-pages for new functions

f636bc5

add the man pages for the newly implemented MPI_Request_get_status_all/any/some functions. Signed-off-by: Edgar Gabriel <edgar.gabriel@amd.com>

edgargabriel force-pushed the topic/request-status-mult branch from a2aa8f9 to 81e3492 Compare June 11, 2025 21:30

hppritcha added 3 commits June 15, 2025 08:59

request_get_status variants: add f08 interfaces

31a0d60

also switch MPI_Request_get_status to use new method for generating f08 bindings. Update fortran bindings interfaces generation code. f90/f77 interfaces will be added as another commit. Signed-off-by: Howard Pritchard <howardp@lanl.gov>

use-mpi-ignore: start using binding infrastructure

fb2e098

for new fortran methods starting with MPI_Request_get_status_(any/all/some) plus switch MPI_Request_get_status to use the binding infrastructure as well. Signed-off-by: Howard Pritchard <howardp@lanl.gov>

request_get_status_mult: add f77 functions

a391063

Signed-off-by: Howard Pritchard <howardp@lanl.gov>

edgargabriel removed the ⚠️ WIP-DNM! label Jun 22, 2025

edgargabriel marked this pull request as ready for review June 22, 2025 18:48

jsquyres reviewed Jul 7, 2025

View reviewed changes

		parser_interface.add_argument('--fort-std', choices=('f90', 'f08'),
		help='fortran standard to use for fortran module generation')

		:ref:`MPI_Request_get_status_all` \|mdash\| Access information associated with a
		request without freeing the request.

add support for request_get_status_any/all/some #13279

Are you sure you want to change the base?

add support for request_get_status_any/all/some #13279

Uh oh!

Conversation

edgargabriel commented May 26, 2025

Uh oh!

hppritcha commented May 26, 2025

Uh oh!

edgargabriel commented May 27, 2025

Uh oh!

Uh oh!

hppritcha commented May 27, 2025

Uh oh!

hppritcha commented May 27, 2025

Uh oh!

hppritcha commented May 27, 2025

Uh oh!

edgargabriel commented May 28, 2025

Uh oh!

edgargabriel commented Jun 11, 2025

Uh oh!

hppritcha commented Jun 19, 2025

Uh oh!

edgargabriel commented Jun 19, 2025

Uh oh!

edgargabriel commented Jun 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

edgargabriel commented Jun 21, 2025 •

edited

Loading