yxsamliu wrote: > @yxsamliu please see https://github.com/yxsamliu/llvm-project/pull/14/changes > where I refactored this to eliminate regressions. The trick was to only run > this optimization after memcpyopt. I wrote a lot of comments explaining why > in the code. There might be some small nitpicks or cleanup after the fact, > but I think what I did is directionally right. > > See > [dtcxzyw/llvm-opt-benchmark-nightly#306](https://github.com/dtcxzyw/llvm-opt-benchmark-nightly/pull/306) > for the benchmark results with no regressions.
@YonahGoldberg Thanks for the refactor. The design looks right. I ran your branch locally — SROA lit passes, #306 looks clean, and the AMDGPU case that motivated this PR still benefits. Happy to land it either way. I can cherry-pick your commit onto #165159 (you stay as author), fix the three failing tests, and squash with you as Co-authored-by. Or you land #197087 and we close this one. My priority is getting the fix in. Which do you prefer? https://github.com/llvm/llvm-project/pull/165159 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
