Tyler,
can you please confirm you are using ompi v1.8 or v1.10 ?
this should be fixed in v2.x or master, and it seems a backport was missing.
can you please give a try to this patch and check whether it fixes your
problem ?
Cheers,
Gilles
On 2/22/2016 11:12 AM, Tyler Benster wrote:
I believe Open MPI is looking for a linux binary (.so) when it should
look for a Mach-O (.dylib) on OS X.
Steps to reproduce:
1. build Open MPI with CUDA-aware support on OS X (‘./configure
—with-cuda’)
2. run job that supports CUDA libraries
3. The following error is generated:
The library attempted to open the following supporting CUDA
libraries, but each of them failed. CUDA-aware support is
disabled. dlopen(libcuda.so.1, 9): image not found
dlopen(/usr/lib64/libcuda.so.1, 9): image not found If you are not
interested in CUDA-aware support, then run with –mca
mpi_cuda_support 0 to suppress this message. If you are interested
in CUDA-aware support, then try setting LD_LIBRARY_PATH to the
location of libcuda.so.1 to get passed this issue.
I do have libcuda.dylib in my path.
Any help is much appreciated!
_______________________________________________
users mailing list
us...@open-mpi.org
Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users
Link to this post:
http://www.open-mpi.org/community/lists/users/2016/02/28562.php
diff --git a/ompi/mca/common/cuda/common_cuda.c
b/ompi/mca/common/cuda/common_cuda.c
index 1ad679c..91fed98 100644
--- a/ompi/mca/common/cuda/common_cuda.c
+++ b/ompi/mca/common/cuda/common_cuda.c
@@ -252,7 +252,7 @@ static int
mca_common_cuda_init(opal_common_cuda_function_table_t *ftable)
int mca_common_cuda_stage_one_init(void)
{
int retval, i, j;
- char *cudalibs[] = {"libcuda.so.1", NULL};
+ char *cudalibs[] = {"libcuda.so.1", "libcuda.dylib", NULL};
char *searchpaths[] = {"", "/usr/lib64", NULL};
char **errmsgs = NULL;
char *errmsg = NULL;
