junrushao opened a new pull request, #16403: URL: https://github.com/apache/tvm/pull/16403
PR #16345 introduced a TIR intrinsic `tvm_thread_invariant` as a no-op intrinsic that indicates that a condition is an invariant across threads. It is a necessary change to support MoE group GEMM operations. However, we noticed that the LLVM-ROCm backend has been broken since then because this no-op intrinsic hasn't been properly handled in this particular backend. This PR fixes this behavior by checking this no-op in the LLVM backend. CC: @spectrometerHBH @yzhliu @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]
