[llvm-bugs] [Bug 46138] New: Failure to avoid emitting a useless instruction when doing combined shits and rotates with a smaller result

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=46138

            Bug ID: 46138
           Summary: Failure to avoid emitting a useless instruction when
                    doing combined shits and rotates with a smaller result
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: gabravier at gmail.com
                CC: llvm-bugs at lists.llvm.org

uint32_t f() 
{
    static uint64_t x = 0;
    x++;
    x = (x>>32) | (x<<32);
    return x;
}

With -O3, GCC outputs this : 

f():
  mov rax, QWORD PTR msws()::x[rip]
  inc rax
  rorx rax, rax, 32
  mov QWORD PTR msws()::x[rip], rax
  ret

LLVM outputs this :

msws(): # @msws()
  mov rax, qword ptr [rip + msws()::x]
  inc rax
  rorx rcx, rax, 32
  shr rax, 32
  mov qword ptr [rip + msws()::x], rcx
  ret

The `shr` could be elimiated and replaced with part of the result of the
preceding rotate

-- 
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/20200529/bcdf3d69/attachment.html>