AW: [avr-gcc-list] Horrible code

[Haase Bjoern (PT-BEU/EMT)]

...probably the 4.1 mid-end generates slightly different RTL, so that the 
present combine patterns of gcc that identify possible uses of the sbrs/jmp 
instruction sequence are no longer triggered. No catastrophe. It should be easy 
to fix. Could you file a gcc bug report and mark the bug as "missing 
optimization" and 4.1 regression, so that it's not forgotten?

Bjoern.
 

-----Ursprüngliche Nachricht-----
Von: address@hidden [mailto:address@hidden Im Auftrag von James L. Evans
Gesendet: Mittwoch, 17. Mai 2006 00:47
An: address@hidden
Betreff: [avr-gcc-list] Horrible code

Could someone with a 4.1 version of the AVR-GCC compiler please compile
the following code (-Os) and see if it generates better code than the
horrible code (25 instructions) generated for Case 2 by version 3.4.3?

Thanks,

Jim Evans


#include <inttypes.h>
#include <avr/io.h>

void test7(void) {
    while (PINA & 0x40) PORTA = 0;                  // Case 1

    while ((PINA & 0x40) == 0) PORTA = 0;           // Case 2

    while (!(PINA & 0x40)) PORTA = 0;               // Case 3

    uint8_t b;
    while ((b = (PINA & 0x40)) == 0) PORTA = 0;     // Case 4
}


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