Skip to content

v4.1.x: ompi/coll/cuda: implement reduce_local #12752

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 30, 2024
Merged

v4.1.x: ompi/coll/cuda: implement reduce_local #12752

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions ompi/mca/coll/cuda/coll_cuda.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
/*
* Copyright (c) 2024 NVIDIA Corporation. All rights reserved.
* Copyright (c) 2014 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
Expand Down Expand Up @@ -45,6 +46,11 @@ mca_coll_cuda_allreduce(const void *sbuf, void *rbuf, int count,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module);

int mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
mca_coll_base_module_t *module);

int mca_coll_cuda_reduce(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
Expand Down
2 changes: 2 additions & 0 deletions ompi/mca/coll/cuda/coll_cuda_module.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
/*
* Copyright (c) 2024 NVIDIA Corporation. All rights reserved.
* Copyright (c) 2014-2017 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
Expand Down Expand Up @@ -104,6 +105,7 @@ mca_coll_cuda_comm_query(struct ompi_communicator_t *comm,
cuda_module->super.coll_gather = NULL;
cuda_module->super.coll_gatherv = NULL;
cuda_module->super.coll_reduce = mca_coll_cuda_reduce;
cuda_module->super.coll_reduce_local = mca_coll_cuda_reduce_local;
cuda_module->super.coll_reduce_scatter = NULL;
cuda_module->super.coll_reduce_scatter_block = mca_coll_cuda_reduce_scatter_block;
cuda_module->super.coll_scan = mca_coll_cuda_scan;
Expand Down
48 changes: 48 additions & 0 deletions ompi/mca/coll/cuda/coll_cuda_reduce.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
/*
* Copyright (c) 2024 NVIDIA Corporation. All rights reserved.
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
Expand Down Expand Up @@ -78,3 +79,50 @@ mca_coll_cuda_reduce(const void *sbuf, void *rbuf, int count,
}
return rc;
}

int
mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
mca_coll_base_module_t *module)
{
ptrdiff_t gap;
char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
size_t bufsize;
int rc;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why do you need rc ? Just return OMPI_SUCCESS.


bufsize = opal_datatype_span(&dtype->super, count, &gap);

if ((MPI_IN_PLACE != sbuf) && (opal_cuda_check_bufs((char *)sbuf, NULL))) {
sbuf1 = (char*)malloc(bufsize);
if (NULL == sbuf1) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
opal_cuda_memcpy_sync(sbuf1, sbuf, bufsize);
sbuf = sbuf1 - gap;
}

if (opal_cuda_check_bufs((char *)rbuf, NULL)) {
rbuf1 = (char*)malloc(bufsize);
if (NULL == rbuf1) {
if (NULL != sbuf1) free(sbuf1);
return OMPI_ERR_OUT_OF_RESOURCE;
}
opal_cuda_memcpy_sync(rbuf1, rbuf, bufsize);
rbuf2 = rbuf; /* save away original buffer */
rbuf = rbuf1 - gap;
}

ompi_op_reduce(op, (void *)sbuf, rbuf, count, dtype);
rc = OMPI_SUCCESS;

if (NULL != sbuf1) {
free(sbuf1);
}
if (NULL != rbuf1) {
rbuf = rbuf2;
opal_cuda_memcpy_sync(rbuf, rbuf1, bufsize);
free(rbuf1);
}
return rc;
}
Loading