- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I have a coarray program which runs fine on a small 4-core desktop, also runs oversubscribed well beyond the number of cores; this is using parallel_studio_xe_2017.0.035.
On a KNL single node (64 core) using the cluster edition on the same ifort above, it runs perfectly with up to 15 images, but it hangs forever at 16 or more images without running any code. I'm compiling with -coarray=shared and also setting the number of images with, say, -coarray-num-images=16 for 16 images. I'm unable to figure out what configuration detail is missing to hang at 16 or more images. Any insight would be appreciated!
Many thanks, Ted
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I had missed this thread URL below because it was in intel-visual-fortran-compiler-for-windows , sorry.
https://software.intel.com/en-us/forums/intel-visual-fortran-compiler-for-windows/topic/734154
The problem is the same for Linux intel cluster edition 17 using ifort coarrays, hangs when images>=16, and the same suggested fix in the meanwhile (setting I_MPI_COLL_INTRANODE=pt2pt) works. I guess this means that shared memory mode fails when images>=16 but MPI is fine.
Ted
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have reproduced the error and escalated it MPI team for fixing. The bug ID is MPI-2889
Thanks,
Xiaoping Duan
Intel Customer Support
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page