[PATCH] D65351: [ARM] Lower "(x<<c) > 0x80000000U" to "lsls" on Thumb1.
Eli Friedman via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 30 16:33:35 PDT 2019
efriedma added a comment.
> Would this also be profitable for Thumb2?
I think it saves only one instruction on ARM/Thumb2 ("cmp rN, #0x80000000" is legal). Still worthwhile, I guess, assuming there isn't some sort of performance penalty for branching based on a shift. But I'd also have to figure out the right heuristics for D65175 <https://reviews.llvm.org/D65175>, since the transforms interact, and do a little more implementation work to lower ARMISD::LSLS. I'll leave a FIXME.
Repository:
rL LLVM
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D65351/new/
https://reviews.llvm.org/D65351
More information about the llvm-commits
mailing list