[Bug tree-optimization/94956] New: Unable to remove impossible ffs() test for zero

["steinar+gcc at gunderson dot no" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94956

            Bug ID: 94956
           Summary: Unable to remove impossible ffs() test for zero
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: steinar+gcc at gunderson dot no
  Target Milestone: ---

Hi,

Given GCC 10 x86-64, and this code:

#include <stdint.h>
#include <string.h>

int foo(uint32_t x) {
        if (x == 0) __builtin_unreachable();
        return ffs(x) - 1;
}

I get this assembler:

atum17:~> gcc-10 -O2 -c test.c 
atum17:~> objdump --disassemble test.o

The cmovne test is rather expensive for me due to high instruction latency,and
I can never have zero in my situation. (It costs ~10% in a much larger graph
algorithm.) I'm unable to get GCC to understand that it doesn't need it, save
for using an explicit asm statement.

By contrast, Clang 10 gets this right:

atum17:~> clang-10 -O2 -c test.c        
atum17:~> objdump --disassemble test.o

test.o:     file format elf64-x86-64


Disassembly of section .text:

0000000000000000 <foo>:
   0:   0f bc c7                bsf    %edi,%eax
   3:   c3                      retq   

Is it possible to get access to the raw instruction by some clever means? :-)