This is an archive of the discontinued LLVM Phabricator instance.

[CUDA] Revert defining __CUDA_ARCH__ for amdgcn targets
ClosedPublic

Authored by yaxunl on Apr 6 2018, 12:49 PM.

Details

Summary

amdgcn targets only support HIP, which does not define __CUDA_ARCH__.

__HIP_DEVICE_COMPILE__ depends on HIP language mode, therefore it needs to be added
in a different patch.

refer to discussions in https://reviews.llvm.org/D45277

Diff Detail

Event Timeline

yaxunl created this revision.Apr 6 2018, 12:49 PM
tra accepted this revision.Apr 6 2018, 1:02 PM

You may want to mention in the commit log that this is a partial unroll of r329232 / D45277.

This revision is now accepted and ready to land.Apr 6 2018, 1:02 PM
yaxunl added a comment.Apr 6 2018, 1:14 PM
In D45387#1060194, @tra wrote:

You may want to mention in the commit log that this is a partial unroll of r329232 / D45277.

Will do. Thanks.

This revision was automatically updated to reflect the committed changes.