[llvm] Minor post-commit review changes for #130988 (NFC) (PR #156895)
Frederik Harwath via llvm-commits
llvm-commits at lists.llvm.org
Tue Sep 9 02:56:59 PDT 2025
================
@@ -719,12 +719,12 @@ FUNCTION_PASS_WITH_PARAMS(
},
parseBoundsCheckingOptions,
"trap;rt;rt-abort;min-rt;min-rt-abort;merge;guard=N")
- FUNCTION_PASS_WITH_PARAMS(
+FUNCTION_PASS_WITH_PARAMS(
"expand-fp", "ExpandFpPass",
[TM = TM](CodeGenOptLevel OL) {
return ExpandFpPass(TM, OL);
},
- parseExpandFpOptions, "opt-level")
+ parseExpandFpOptions, "opt-level=N")
----------------
frederik-h wrote:
> > This opt-level argument is strange. What do other passes do for the opt level argument? This is only used in a binary way in the pass so does it really need the level?
>
> You are right, it is common to use the optimization level directly. That is, you would write `expand-fp<O0>` instead of `expand-fp<opt-level=0>`. Perhaps I was worried about other arguments that we might want to be able to use here, but that would simply work as in, e.g. `loop-unroll<partial;O2>`. I will change the argument parsing to use the same format.
See PR #157408 which changes `opt-level=0` to `O0` etc.
https://github.com/llvm/llvm-project/pull/156895
More information about the llvm-commits
mailing list