[avr-chat] Inline assembler warning: asm operand 2 probably doesn't match constraints

[Rick Mann]

Hi. I'm doing some experiments with writing C++ wrappers to access AVR 
hardware. I've got an AVRPort class that does some interesting things.

        http://pastebin.com/AwuY28F9

In particular, I'm able to write a method using inline assembler that produces 
code identical to the traditional C:

        PORTA |= 0x03;

  12:   82 b1           in      r24, 0x02       ; 2
  14:   83 60           ori     r24, 0x03       ; 3
  16:   82 b9           out     0x02, r24       ; 2


But when I try to write similar inline asm for the inverse operation, I get an 
error:

    void
    clear(uint8_t inPins)
    {
#if 0
        mPort &= ~inPins;
#else       
        uint8_t     inverse = ~inPins;
        uint8_t     temp;
        asm volatile (  "in     %0, %1"         "\n\t"
                        "andi   %0, %2"         "\n\t"
                        "out    %1, %0"         "\n\t"
                        : "=&r" (temp)
                        : "I" (&mPort), "I" (inverse)
                        );
#endif
    }

src/AVRPort.h: In member function 'virtual msg_t BlinkThread::entry()':
src/AVRPort.h:61: warning: asm operand 2 probably doesn't match constraints
src/AVRPort.h:61: error: impossible constraint in 'asm'


If instead of asm I use the expression

        mPort &= ~inPins;

the generated assembler looks like this:

  18:   80 91 02 00     lds     r24, 0x0002
  1c:   8b 7f           andi    r24, 0xFB       ; 251
  1e:   8a 83           std     Y+2, r24        ; 0x02

Which is certainly less space-efficient; I don't know if also runs slower. It 
clobbers more registers, too.

Ideally, I'd fix the inline asm, but I don't have much experience with it, and 
I'm not sure exactly what I'm doing wrong.

-- 
Rick