[Mvapich-discuss] Contributing to OSU Micro Benchmarks

[Robison, Luke]

Hari,

Here is a rather large patch for one-sided MPI communication data validation in CAS, FOP and ACC latency tests, enabled with the -c option.  This should work for both GPU and non-GPU memory, although I only have CUDA testing available.

The general approach is to perform a one-sided operation, then mock the operation and compare the mocked operation to the resulting values.  In situations where the remote is not explicitly synchronizing in each iteration we just perform one validation instead of many.

Additionally I included loops to test many data types, although I did not find a good way to parameterize this.

I hope you find the changes helpful.

Luke Robison

From: Subramoni, Hari <subramoni.1 at osu.edu>
Sent: Friday, January 13, 2023 09:47
To: Robison, Luke <lrbison at amazon.com>; mvapich-discuss at lists.osu.edu
Cc: Subramoni, Hari <subramoni.1 at osu.edu>
Subject: RE: [EXTERNAL]Contributing to OSU Micro Benchmarks


CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.


Dear, Luke.

Thanks for trying out OMB and enhancing it.

We have not added data validation to the one-sided benchmarks yet. Kindly send over your patch to us. We will attempt to generalize it so that it is applicable to all the one-sided benchmarks and take it into the code.

Best,
Hari.

From: Mvapich-discuss <mvapich-discuss-bounces at lists.osu.edu<mailto:mvapich-discuss-bounces at lists.osu.edu>> On Behalf Of Robison, Luke via Mvapich-discuss
Sent: Wednesday, January 11, 2023 5:41 PM
To: mvapich-discuss at lists.osu.edu<mailto:mvapich-discuss at lists.osu.edu>
Subject: [Mvapich-discuss] Contributing to OSU Micro Benchmarks

OMB Team, I found the need to do some data validation of one-sided MPI operations. Rather than build my tests from scratch, I've made modifications to OSU's osu_fop_latency and osu_cas_latency, adding both a loop over datatype (and

OMB Team,

I found the need to do some data validation of one-sided MPI operations.  Rather than build my tests from scratch, I've made modifications to OSU's osu_fop_latency and osu_cas_latency, adding both a loop over datatype (and op, when appropriate).  Would your team be interested in taking my patch to incorporate an option to verify one-sided communications?  What is the best way to submit a patch?  (Is it to this group?)


Thank you,
Luke
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osu.edu/pipermail/mvapich-discuss/attachments/20230123/cd01e89b/attachment-0010.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: osc_validation.patch
Type: application/octet-stream
Size: 129892 bytes
Desc: osc_validation.patch
URL: <http://lists.osu.edu/pipermail/mvapich-discuss/attachments/20230123/cd01e89b/attachment-0010.obj>