Re: [avr-gcc-list] interrupt optimization

[David Brown]

Perhaps you could try declaring your ISR as naked, with manual saving of
minimal registers, and make the called function a "signal" function.  That
way you'll have no extra pushes in the ISR, but the called function will
preserve everything as needed.

mvh.,

David


> Hi,
>
> I have an interrupt which is called a lot of times but only really do
> something very rarely:
>
> ISR(...)
> {
>      if (sampling-- == 0)
>      {
>          sampling = 0x7F;
>          callInlineFunction();
>      }
> }
>
> The problem is that I end up with a lot of push/pop that just eat up all
> my CPU cycles most of the time. Using an inline function helped removing
> all registers that were not used by the function at all (don't know why
> the optimization doesn't do this already) but I still get a lot of push
> not necessary by the "if" and that I would like to move before the
> function call inside the "if".
>
> What's the way to go in this situation? I thought of using a naked ISR and
> adding the necessary push by inline asm but what if I later change my code
> and the compiler use different registers? I'm afraid to end up with broken
> code.
>
> Thanks,
> David Bourgeois
>
> asm output of the ISR with inline function call:
>   334: 1f 92       push r1
>   336: 0f 92       push r0
>   338: 0f b6       in r0, 0x3f ; 63
>   33a: 0f 92       push r0
>   33c: 11 24       eor r1, r1
>   33e: 2f 93       push r18
>   340: 3f 93       push r19
>   342: 8f 93       push r24
>   344: 9f 93       push r25
>   346: ef 93       push r30
>   348: ff 93       push r31
>
>   34a: 80 91 0f 01 lds r24, 0x010F ; if (sampling-- == 0)
>   34e: 81 50       subi r24, 0x01 ; 1
>   350: 80 93 0f 01 sts 0x010F, r24
>   354: 88 23       and r24, r24
>   356: 51 f5       brne .+84     ; 0x3ac <__vector_16+0x78>
>
>
> _______________________________________________
> AVR-GCC-list mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/avr-gcc-list
>