[avr-gcc-list] How to use inversion in if() - might be a bug report ?

[Vaclav Peroutka]

Hello all,

I got portable WinAVR with avr-gcc version 4.3.2 and tried to compile simple 
code:

/////////////////////////////// code
#include <avr/io.h>
#define XTALFQ (8e+6)

volatile unsigned char var0x55, var0xaa;

void delay_ms(unsigned int ms)
{
  while(ms--) {
    unsigned short int i;
    for (i=100; i ; i-- ) {
      unsigned short int b;
      for(b=12*XTALFQ/12E6;b;b--) asm volatile("nop"::);
    }
  }
}

/************** main program ********************/
int main(void)
{
   var0x55 = 0x55;
   var0xaa = 0xaa;

  /* INITIALIZE */
   DDRC|= (1<<PC5);
   PORTC|= (1<<PC5);

  /* BLINK, BLINK ... */
  while (1) {
    /* invert PC5 */
    PORTC ^= (1<<PC5);

    // this code works
    if ( (var0xaa ^ 0xff) == var0x55) {
       delay_ms(100);
    } else {
       delay_ms(200);
    }
    // this code is buggy
    if ( (~var0xaa) == var0x55) {
       delay_ms(100);
    } else {
       delay_ms(200);
    }

  }
}
///////////////////////////////////////// end of code

The problem is with buggy code - when using '~', it gets expanded to 16 bits 
(however, in avr-gcc 3.4.6 it was the same problem as well) :
    // this code works
    if ( (var0xaa ^ 0xff) == var0x55) {
  84:   80 91 61 00     lds     r24, 0x0061
  88:   90 91 60 00     lds     r25, 0x0060
  8c:   80 95           com     r24
  8e:   89 17           cp      r24, r25
  90:   f1 f4           brne    .+60            ; 0xce <main+0x62>
  92:   09 c0           rjmp    .+18            ; 0xa6 <main+0x3a>
...
    // this code is buggy
    if ( (~var0xaa) == var0x55) {
  e4:   80 91 61 00     lds     r24, 0x0061
  e8:   20 91 60 00     lds     r18, 0x0060
  ec:   90 e0           ldi     r25, 0x00       ; 0
  ee:   80 95           com     r24
  f0:   90 95           com     r25
  f2:   30 e0           ldi     r19, 0x00       ; 0
  f4:   82 17           cp      r24, r18
  f6:   93 07           cpc     r25, r19
  f8:   f9 f4           brne    .+62            ; 0x138 <main+0xcc>
  fa:   09 c0           rjmp    .+18            ; 0x10e <main+0xa2>

Furthermore, the new .LST file looks really crazy and the resulting code is 
bigger ( 266 bytes from 4.3.2 compared to 130 bytes from 3.4.6).

Does anybody have a clue, what happened ?

Thank you in advance,
Vaclav