[GitHub] [incubator-tvm] t-vi commented on issue #4485: [WIP] Clean up AMD Code Gen

[GitBox <gi...@apache.org>]

t-vi commented on issue #4485: [WIP] Clean up AMD Code Gen
URL: https://github.com/apache/incubator-tvm/pull/4485#issuecomment-597788333
 
 
   I wonder whether the cleanup we're doing here is in fact worth it.
   So what happens is the PR wants to remove LLVM < 6 compatibility because we effectively cannot use it with ROCm.
   If we wanted to make this cleanup, we could make the AMDGPU codegen conditional on
   - LLVM >= 6 (which doesn't really make sense given that we want to clean up this condition),
   - make it dependent on USE_ROCM, but that would reduce the AMDGPU/ROCm coverage of "at least it copiles CI", which I would find regrettable (in an ideal world, I'd rather start a discussion on how to include ROCM in the CI, at least keeping it building should be possible even without hardware, but we might also want to see how much hardware we would need and how to see about this).
   
   BTW: With the same argument, we could also drop the < 9 and only keep the >= 9 bits because LLVM < 8 will crash with NAVI bitcode.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services