Michael Heinz

icon indicating a website link https://intel.com icon indicating an email [email protected]

icon company Intel icon location Phoenixville, PA icon location Antique code monkey who used to be a clown, but only for charity.

Results 4 issues of Michael Heinz

Hopefully a quick question about process layout across hosts in the hostfile...

1 comment

Hey, guys. I have a user who is used to handling process placement by simply listing hosts multiple times in the hostfile so that, for example, if he wanted 2...

Question about how OMPI handles OFI HMEM and completion semantics

10 comment

Hey, guys. A question came up about how the OMPI OFI MTL and BTL handle ensuring that CUDA/HMEM buffers are completely in sync at the end of a data transfer....

question

MTLs are inappropriately calling exit() or abort()

6 comment

As a part of discussing #7699, it was pointed out that there are several error paths in the OFI MTL that call exit() and several in the Portals MTL that...

bug
Target: main
Target: v5.0.x

Trouble using torch-ccl with the mlx provider

1 comment

We've had success using torch-ccl with resnet and other AI workloads to test with libfabric over psm3 but when we try to use libmlx-fi.so, torch-ccl does not seem to see...