Pruthvi Madugundu
Pruthvi Madugundu
The CUDA failure FAIL [2.540s]: test_MaxUnpool_index_errors_case1_cuda (__main__.TestNNDeviceTypeCUDA) FAIL [2.524s]: test_MaxUnpool_index_errors_case2_cuda (__main__.TestNNDeviceTypeCUDA) doesnt seem to be related to this change.
Seeing few performance regression when kernel asserts are enabled.
> Thanks. It looks good to me. Performance drop is expected if `assert()` is enabled. If this performance regression is okay for AMD (anyway the CUDA side enables this `assert()`),...
> I don't know why `assert()` is always enabled even in the release build, but my guess is to enhance debuggability. > > Is it okay for AMD to merge...
@shintaro-iwasaki Please review the latest changes, I have added a compile time flag to disable kernel assert for ROCm only.
Many Windows related jobs are failing and doesnt seem to be related to this PR.
> They seem not related to this PR, but we'd appreciate it if you could rebase this PR onto a stable one (perhaps later). > > So finally what does...
> Thanks for your explanation. Do you still want to merge this PR now if there's no particular use case, and this feature is disabled by default and will not...
@shintaro-iwasaki , The 2 failing test from periodic/ios jobs which are not related to this change and I find them failing in other PR also. So I believe this PR...
Documentation change is raised in PR - https://github.com/pytorch/pytorch.github.io/pull/1097