[mvapich-discuss] OSU Benchmarks 5.4

[Chu, Ching-Hsiang]

Hi, Sameh,

Thanks for contacting us. Please note that we have fixed the issues you mentioned in the recent release of OSU Micro-Benchmarks. We strongly encourage you to try the latest release, more details can be found in http://mvapich.cse.ohio-state.edu/benchmarks/.

Please feel free to let us know if you encounter any issues or have any further questions.
Thanks,

Ching-Hsiang Chu

________________________________
From: mvapich-discuss <mvapich-discuss-bounces at cse.ohio-state.edu> on behalf of Sameh S Sharkawi <sssharka at us.ibm.com>
Sent: Tuesday, May 21, 2019 4:20 PM
To: mvapich-discuss at cse.ohio-state.edu
Subject: [mvapich-discuss] OSU Benchmarks 5.4

This is my first email on this list. I want to bring up an issue I found with OSU benchmarks when sending from H to D or D to H.. Also the same issue occurs when sending from M to H or H to M. Basically, the 5.4 version calls cudaMemset if the access.option is set to CUDA or MANAGED regardless of the src/dst which might be a host memory. cudaMemset fails but the util.c file never checks for errors.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20190522/02fb894d/attachment-0001.html>