[llvm-bugs] [Bug 28968] New: [x86] missed chance to turn select (cmov) into logic ops

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=28968

            Bug ID: 28968
           Summary: [x86] missed chance to turn select (cmov) into logic
                    ops
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: spatel+llvm at rotateright.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Forking this off of bug 28964.

Notice that:
int goo(int x) {
  if (x == 1)
    x = 1;
  else
    x = -1;
  return x;
}

Could be rewritten as:
int hoo(int x) {
  if (x != 1)
    x = -1;
  return x;
}

Or as IR:

define i32 @hoo(i32 %x) {
  %cmp = icmp ne i32 %x, 1
  %sel = select i1 %cmp, i32 -1, i32 1
  ret i32 %sel
}

$ ./llc -o - cmpsel.ll 
    cmpl    $1, %edi
    movl    $-1, %eax
    cmovel    %edi, %eax
    retq


According to godbolt ( https://godbolt.org/g/TSTeIo ), gcc learned to avoid a
cmov on this at v6.1:

foo(int):
        xorl    %eax, %eax
        cmpl    $1, %edi
        setne   %al
        negl    %eax
        orl     $1, %eax
        ret

Note that gcc does this at -O2 but not -O1 or -Os.

Filing this as an x86 bug for now, but it might be useful for other targets
too.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160813/1841cb51/attachment-0001.html>