[PATCH] D45585: [DAGCombiner][X86] When promoting loads don't use ZEXTLOAD even its legal
Craig Topper via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 24 15:20:45 PDT 2018
craig.topper added inline comments.
================
Comment at: test/CodeGen/X86/mul-constant-i16.ll:75
; X86: # %bb.0:
-; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: leal (%eax,%eax,4), %eax
+; X86-NEXT: imull $5, {{[0-9]+}}(%esp), %eax
; X86-NEXT: # kill: def $ax killed $ax killed $eax
----------------
RKSimon wrote:
> These are likely to be performance regressions on most targets - the special case x86 'MulConstantOptimization' patterns in combineMul should still be used where possible.
Agreed. But that issue wasn't caused by this patch. It's pretty easy to hit with 32-bit instructions too.
https://godbolt.org/g/9Pa83o
https://reviews.llvm.org/D45585
More information about the llvm-commits
mailing list