Bug in gcc-20011231

[Peter Barada]

>That's pretty weird.  3.0.1 on my VIA C3 box works fine without using an
>IMUL: looks like somebody patched the modulus code to perform the
>division using multiply by the reciprocal and got it wrong.  I don't
>happen to have gcc-20011231 
>
>If you change the code this way, does it perform the modulus using IDIV
>and thus get it right?
>
>main(int argc, char *argv[])
>{
>  ufix8 result;
>  ufix8 i;
>
>  for (i=2; i<7; i++)
>  {
>        result = val % i;
>        if (result != expected)
>         fprintf(stderr,"Huh? modulus is %d\n", result);
>  }
>  return(0);
>}
>

That runs fine, and the code generated is:

.L5:
	movzbl	val, %eax
	movzbl	%bl, %ecx
	movl	$0, %edx
	divl	%ecx
	cmpb	expected, %dl
	je	.L4

Whihc is what I'd expect.

I modified the code to look like:


#include <stdio.h>

typedef unsigned char ufix8;

ufix8 val = 1;
ufix8 expected = 1;

#define CHECK(val,i,result) \
  result = val % i; \
  if (result != expected) { \
   fprintf(stderr,"Huh? %d %% %d = %d\n", val, i, result); \
  }

main(int argc, char *argv[])
{
  ufix8 result;
  ufix8 i;

  CHECK(val, 2, result);
  CHECK(val, 3, result);
  CHECK(val, 4, result);
  CHECK(val, 5, result);
  CHECK(val, 6, result);
  CHECK(val, 7, result);
  CHECK(val, 8, result);
  CHECK(val, 9, result);
  CHECK(val, 10, result);
  CHECK(val, 11, result);
  CHECK(val, 12, result);
}


And the ouptut is:

Huh? 1 % 3 = 132
Huh? 1 % 5 = 198
Huh? 1 % 6 = 135
Huh? 1 % 10 = 203
Huh? 1 % 12 = 141

So modulus of 2,4,7,8,9,11 worked correctly(for these particular values).

If I change:

typedef unsigned char ufix8;

to:

typedef unsigned short ufix8;

then everything works fine, so the modulus replacement by imul plus
cleanup is broken for char types...


-- 
Peter Barada                                   Peter.Barada@motorola.com
Wizard                                         781-852-2768 (direct)
WaveMark Solutions(wholly owned by Motorola)   781-270-0193 (fax)