heiher wrote:

> `select-binop-combine.ll` already performs a similar transformation at the IR 
> level. After running `opt -O2`, the resulting code has the same effect as 
> this combine. Do we still need to handle this separately in the backend?

Thanks! I checked the assembly generated by `clang -O3` when building the 
llvm-test-suite, and the expected codegen improvement is present with this 
patch. So I think this backend combine complements the existing IR-level 
optimization?

https://gitlab.com/la-rs/llvm-opt-ci/-/commit/1f2459e50ac1922777319b3b6b645cc55dd40e9f?view=parallel

https://github.com/llvm/llvm-project/pull/215520
_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

Reply via email to