[llvm] [InstCombine] handle trunc to i1 in foldLogOpOfMaskedICmps. (PR #128861)

Yingwei Zheng via llvm-commits llvm-commits at lists.llvm.org
Mon Mar 31 03:11:51 PDT 2025


dtcxzyw wrote:

> sorry for the bad description what I was trying to say is that some of the regressions that this PR cause eg. https://github.com/dtcxzyw/llvm-opt-benchmark/pull/2173/files#diff-c88279c89225628a242defdd907f1444ed67e844ff858382dbea2edc36237e7fR167035-R167038 and https://github.com/dtcxzyw/llvm-opt-benchmark/pull/2173/files#diff-403632332e00c769093bc90d3e97457d65f8adf3b5824d6cd460e13c166cc164R1747-R1749 is solved by [ed57ea5](https://github.com/llvm/llvm-project/commit/ed57ea53190cadfb46deec4c5faefa47d91cdde4) but the fold from that commit have many regressions and some of those regressions is fixed by this PR.

This regression is still outstanding: https://alive2.llvm.org/ce/z/4tZwjg
Can you pre-commit this test?


https://github.com/llvm/llvm-project/pull/128861


More information about the llvm-commits mailing list