Re: [avr-gcc-list] What's wrong with this preprocessor statement

[Denis Chertykov]

address@hidden (J Wunsch) writes:

> Denis Chertykov <address@hidden> wrote:
> 
> > Today it's not a true.
> > Now GCC can optimize out such loops.
> 
> Out of curiosity: when does gcc optimize them out?  All versions of
> gcc i tried (AVR, FreeBSD/i386, Linux/i386, Solaris/Sparc,
> HP-UX/PA-RISC, all of them were 2.95.2 versions) don't optimize it
> away in the following file, not even with -O3 (which implies
> -fstrength-reduce, which is documented to perform the loop variable
> elimination):
> 
> void
> foo(void)
> {
>       int i;
> 
>       for (i = 0; i < 10000; i++) ;
> }

Current avr-gcc (CVS version) generates:

.global foo
        .type   foo,@function
foo:
/* prologue: frame size=0 */
/* prologue end (size=0) */
        ldi r24,lo8(9999)
        ldi r25,hi8(9999)
.L7:
        subi r24,lo8(-(-625))  <---------- Will be executed only 16 times
        sbci r25,hi8(-(-625))  <----------
        brpl .L7
/* epilogue: frame size=0 */
        ret
/* epilogue end (size=1) */
/* function foo size 7 (6) */
.Lfe2:
        .size   foo,.Lfe2-foo