Re: [avr-gcc-list] Bad optimization AVR-GCC 4.1.1 compared to AVR-GCC 3.

On 10/6/06, Peter Dannegger <address@hidden> wrote:

Dear GCC experts,

following an example code, which was about 25% bigger on using the
AVR-GCC 4.1.1. instead AVR-GCC 3.4.6.

The code was bigger on 3 points:

1.
From a 16 bit variable, where only the high byte was used, also the low
byte was read from SRAM into a register (3 words more).

2.
A function was inlined nevertheless, the optimization for size -Os was
forced and no inline or static attribute was given. It seems, that the
-Os switch was ignored.

3.
A fully crazy behaviour was seen on the attempt to clear a variable
conditional.
Instead clearing the register (1 word instruction ) like on the AVR-GCC
3.4.6, a big sequence of code (23 words) was duplicated.

Following the code, it was compiled with this command line:

avr-gcc.exe -xc -Os -mmcu=attiny26 -Wall -g disptime.c

int main(void)
{
  for(;;){

  }
}

unsigned int sec_timer;

unsigned char minute, hour;

unsigned char digit[4];

unsigned char numbers[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };

unsigned char get7s( unsigned char val )
{
  return numbers[val];
}

void display_time( unsigned char set )
{
  unsigned char t = sec_timer >> 8;     // high byte only

  if( t < 10 )                          // 1/2 second
    set = 0;

  if( set == 1 ){
    digit[3] = 0;
    digit[2] = 0;
  }else{
    digit[3] = get7s( hour / 10);
    digit[2] = get7s( hour % 10);
  }
  if( set == 2 ){
    digit[1] = 0;
    digit[0] = 0;
  }else{
    digit[1] = get7s( minute / 10);
    digit[0] = get7s( minute % 10);
  }
}

--
        mit freundlichen Grüssen

        Peter Dannegger

        SPECS - COMPETENCE IN SURFACE SCIENCE
        -------------------------------------
        Peter Dannegger
        Specs GmbH
        Voltastr. 5
        13355 Berlin
        Germany
        phone +49 (30) 467 824-21
        fax   +49 (30) 464 20 83
         address@hidden
        www.specs.de

_______________________________________________
AVR-GCC-list mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/avr-gcc-list

From:	Tom Deutschman
Subject:	Re: [avr-gcc-list] Bad optimization AVR-GCC 4.1.1 compared to AVR-GCC 3.4.6
Date:	Fri, 6 Oct 2006 07:12:35 -0700