[Bug middle-end/109907] Missed optimization for bit extraction (uses shift instead of single bit-test)

["gjl at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109907

--- Comment #23 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
Thank you so much for looking into this.

For the test case from comment #21 though, the problem is somewhere in tree
optimizations.

> unsigned char lfsr32_mpp_ge0 (unsigned long number)
> {
>   unsigned char b = 0;
>   if (number >= 0) b--;
>   if (number & (1UL << 29)) b++;
>   if (number & (1UL << 13)) b++;
> 
>   return b;
> }

The -fdump-tree-optimized dump reads:

;; Function lfsr32_mpp_ge0 (lfsr32_mpp_ge0, funcdef_no=0, decl_uid=1880,
cgraph_uid=1, symbol_order=0)

unsigned char lfsr32_mpp_ge0 (long unsigned int number)
{
  unsigned char b;
  long unsigned int _1;
  long unsigned int _2;
  _Bool _3;
  unsigned char _8;
  _Bool _9;
  unsigned char _10;
  unsigned char _11;

  <bb 2> [local count: 1073741824]:
  _1 = number_5(D) & 536870912;
  _2 = number_5(D) & 8192;
  if (_2 != 0)
    goto <bb 4>; [50.00%]
  else
    goto <bb 3>; [50.00%]

  <bb 3> [local count: 536870912]:
  _9 = _1 == 0;
  _10 = (unsigned char) _9;
  _11 = -_10;
  goto <bb 5>; [100.00%]

  <bb 4> [local count: 536870913]:
  _3 = _1 != 0;
  _8 = (unsigned char) _3;

  <bb 5> [local count: 1073741824]:
  # b_4 = PHI <_11(3), _8(4)>
  return b_4;
}

The ANDs are expanded by expand_binop() and later passes have to deal with the
32-bit arithmnetic.  combine finds one combination of andsi3 into
"*sbrx_and_branch<mode>_split" with mode=si, but apart from that the mess still
lives in asm.