[GitHub] [incubator-tvm] roastduck edited a comment on issue #5307: [TIR] Make lower_warp_memory support extent(threadIdx.x) < warp_size

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

roastduck edited a comment on issue #5307: [TIR] Make lower_warp_memory support extent(threadIdx.x) < warp_size
URL: https://github.com/apache/incubator-tvm/pull/5307#issuecomment-612562955
 
 
   Let me sort out the two approaches.
   
   - In the current approach, the lowering pass directly detects shuffles on `threadIdx.x`, which has an arbitrary extent. It may be difficult to detect all the complex use cases, for example when there are mixed half-warp and full-warp shuffling.
   - In the alternative approach, we assume `threadIdx.x == warp size` when detecting shuffles. Therefore, either we require users to set the extent of `threadIdx.x` to the warp size (and then we close this PR), otherwise we may have to add an additional pass to convert `threadIdx.x` to match the warp size.

----------------------------------------------------------------
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