Re: [avr-chat] Missed Optimisation ?

[Alex Eremeenkov]

Hello Bob,

Did you sure that compiler *must* determine that for OR operation only 
one byte changing and load only part of value instead of all 32 bits?
It's a simple optimization stuff according assemble set, but according 
initial logical operation with 32bit value it's a correct code. Compiler 
save at least 3 other 'ori' commands.
BTW, you simply could optimize your routie by doing OR command on MSB 
byte only by itself.


01.03.2011 13:48, bob writes:
> Hi,
>
> I have a very time-critcal routine where a 32-bit value is read serially
> (bit-shift).
> The code is very simple (runs in Port Change interrupt):
>
>      if ((PINA&  _BV(A2DDATA)) != 0)
>         result |= 0x80000000;
>
>      result>>= 1;
>
> The disassembly of the OR  however is:
>
>
>      if ((PINA&  _BV(A2DDATA)) != 0)
>   622:  01 9b           sbis    0x00, 1 ; 0
>   624:  11 c0           rjmp    .+34            ; 0x648<__vector_4+0x42>
>
>         result |= 0x80000000;
>   626:  80 91 5a 02     lds     r24, 0x025A
>   62a:  90 91 5b 02     lds     r25, 0x025B
>   62e:  a0 91 5c 02     lds     r26, 0x025C
>   632:  b0 91 5d 02     lds     r27, 0x025D
>   636:  b0 68           ori     r27, 0x80       ; 128
>   638:  80 93 5a 02     sts     0x025A, r24
>   63c:  90 93 5b 02     sts     0x025B, r25
>   640:  a0 93 5c 02     sts     0x025C, r26
>   644:  b0 93 5d 02     sts     0x025D, r27
>
> The compiler has loaded all 4 bytes of the uint32_t, but only operated
> on one of them, so it seems 'clever' enough to see that the immediate
> value is full of '0' and not bother to OR the lower bytes but still
> loads and saves them.
>
> I have:
>
> OPTIMIZE       = -Os -fno-inline
>
> override CFLAGS        = -g -Wall -Wno-main $(OPTIMIZE)
> -mmcu=$(MCU_TARGET) $(DEFS)
>
> in my Makefile.
>
> Am I missing something?
>
> Regards,
>
> Bob
>
> _______________________________________________
> AVR-chat mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/avr-chat