[PATCH] D54119: [X86] Avoid creating a critical edge during cmov expansion

[Matthias Braun via Phabricator via llvm-commits]

MatzeB added a comment.

> Unfortunately, it doesn't seem to fix the problem reported in that PR. But maybe it gets us closer to being able to fix it?

This is surprising to me. I didn't do anything else in my experiments in http://llvm.org/PR39534 except forcing phi elimination to split more...



================
Comment at: test/CodeGen/X86/atomic32.ll:325
 ; X86-NOCMOV-NEXT:    pushl %ebx
-; X86-NOCMOV-NEXT:    pushl %esi
 ; X86-NOCMOV-NEXT:    subl $24, %esp
 ; X86-NOCMOV-NEXT:    movl {{[0-9]+}}(%esp), %eax
----------------
on the bright side it saves a spill here... Probably needs some benchmarking to see whether it is a net progression or regression...


https://reviews.llvm.org/D54119