[llvm-bugs] [Bug 42189] New: Avoid copy to reg before cmp

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=42189

            Bug ID: 42189
           Summary: Avoid copy to reg before cmp
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: david.bolvansky at gmail.com
                CC: craig.topper at gmail.com, llvm-bugs at lists.llvm.org,
                    llvm-dev at redking.me.uk, spatel+llvm at rotateright.com

int g();
int f();

int shortmax(short c)
{
  if (c == 32767)
    return g();
  return f();
}

Clang:
shortmax(short):
  movzx eax, di
  cmp eax, 32767
  jne .LBB0_2
  jmp g()
.LBB0_2:
  jmp f()

But shouldn't we just use cmp di, 32767?

And I am just looking what -Os produces, and this is really interesting...

shortmax(short):
  movzx eax, di
  cmp eax, 32767
  jne f() // neat
  jmp g()

So for this pattern
if (x) call fn1();
else   call fn2();
Possibly -O2/O3 should do this tranformation too?

Maybe ideal code (what -O3 should produce) for my example is:
shortmax(short):
  cmp di, 32767
  jne f()
  jmp g()

?

-- 
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/20190607/bf7b3b7f/attachment.html>