Closed dmcdougall closed 2 months ago
@jxy Can you please confirm this fixes the staggered dslash test failure you were observing.
Hi Damon, that was an excellent detective work! Thanks a lot!
I believe it fixed all the issues as I no longer see all the r8 failures as before.
I got a single failure with half_r8_partition0, but I could never reproduce this failure after 20 more runs. So I guess that's probably because of some early hardware instabilities.
I tested with rocm 6.0.3 on tioga at LLNL.
Hi Damon, that was an excellent detective work! Thanks a lot!
That's very kind of you, but I can't credit for this one. I thought it was a compiler bug and one of our compiler engineers found it. I'll pass along your comments to them.
Appreciate you testing.
This addresses a correctness check failure in various staggered dslash tests.