yzh119 opened a new pull request, #14280: URL: https://github.com/apache/tvm/pull/14280
# Motivation The `LowerWarpMemory` pass cannot emit `shfl_sync` instructions because of an [internal check](https://github.com/apache/tvm/blob/e3c8f2bc2d9ab3f5ba1c0db0755e455bf59a39b8/src/tir/transforms/lower_warp_memory.cc#L334-L337) introduced in #9727 . Actually if we load value from another lane in the warp, the `local_index` would inevitably carry the warp index, and this case would be disabled by the check. This PR fix the issue by disabling the check and add an unit test for warp shuffling. The PR depends on #14279 , I'll rebase to upstream/main after that PR is merged. @Lunderberg @masahi @tqchen -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
