[Bug tree-optimization/100922] CSE leads to fully redundant (back to back) zero-extending loads of the same thing in a loop, or a register copy

[peter at cordes dot ca]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100922

--- Comment #2 from Peter Cordes <peter at cordes dot ca> ---
Possibly also related:

With different surrounding code, this loop can compile to asm which has two
useless movz / mov register copies in the loop at -O2 
(https://godbolt.org/z/PTcqzM6q7).  (To set up for entry into the next loop in
over-complicated ways, and doing this in the loop is unnecessary.)


  while( lut[(unsigned char)*str] == 0 ){  // also catches terminating 0
    str++;
  }


.L19:
        movzbl  1(%rdi), %edx
        addq    $1, %rdi
        movzbl  %dl, %ecx
        movl    %edx, %eax
        cmpb    $0, -120(%rsp,%rcx)
        je      .L19

from source

void remove_chars(char *restrict str, const char *restrict remove)
{
  char lut[256] = {0};
  do {
    lut[(unsigned char)*remove] = -1;
  }while(*remove++);

/*******   Over complicated asm in this loop *************/
  while( lut[(unsigned char)*str] == 0 ){  // also catches terminating 0
    str++;
  }
  // str points at first char to *not* keep (or the terminating 0)
  const char *in = str;
  char *out = str;
  while (*in)
    {
      char mask = lut[(unsigned char)*in];
        unsigned char cin = *in, cout = *out;
        *out = mask ? cout : cin;
      out += mask + 1;
      in++;
    }
  *out = *in;
}