Re: [avr-gcc-list] Efficient I/O handling w. bitfields codegen problem

[Albert Seward]

On Thu, 2 Dec 2004, E. Weddington wrote:

> Albert Seward wrote:
>
> >On Thu, 2 Dec 2004, E. Weddington wrote:
> >
> >
> >
> >>Albert Seward wrote:
> >>
> >>
> >>
> >>>I have some problems with code generation from GCC 3.3. Read my question
> >>>in the main block.
> >>>
> >>>/*
> >>>* bitfields.c
> >>>*
> >>>* GCC bitfields optimization test
> >>>*/
> >>>typedef union {
> >>>       unsigned char byte;
> >>>       struct {
> >>>               unsigned char pin0 : 1;
> >>>               unsigned char pin1 : 1;
> >>>               unsigned char pin2 : 1;
> >>>               unsigned char pin3 : 1;
> >>>               unsigned char pin4 : 1;
> >>>               unsigned char pin5 : 1;
> >>>               unsigned char pin6 : 1;
> >>>               unsigned char pin7 : 1;
> >>>       };
> >>>} volatile * const port_t;
> >>>
> >>>port_t port[] = {(port_t)0x3d, (port_t)0x38, (port_t)0x35, (port_t)0x32};
> >>>
> >>>// Another way...
> >>>struct  {
> >>>       port_t portA;
> >>>       port_t portB;
> >>>       port_t portC;
> >>>       port_t portD;
> >>>} ports = {(port_t)0x3d, (port_t)0x38, (port_t)0x35, (port_t)0x32};
> >>>
> >>>
> >>>int main(void) {
> >>>       /*  This works!
> >>>        *  It will give asm:
> >>>        *  cbi     0x18, 7
> >>>        */ sbi     0x18, 7
> >>>       port[1]->pin7 = 0;
> >>>       port[1]->pin7 = 1;
> >>>
> >>>       /*  Why dont I get the same asm as above?
> >>>        *  Now I do get this:
> >>>        *
> >>>        *  lds     r30, 0x0068
> >>>        *  lds     r31, 0x0069
> >>>        *  ld      r24, Z
> >>>        *  andi    r24, 0x7F       ; 127
> >>>        *  st      Z, r24
> >>>        *
> >>>        *  lds     r30, 0x0068
> >>>        *  lds     r31, 0x0069
> >>>        *  ld      r24, Z
> >>>        *  ori     r24, 0x80       ; 128
> >>>        *  st      Z, r24
> >>>        *
> >>>        *  This is not as efficient, and it does not do the job
> >>>        *  if I for example want to change GIMSK this way.
> >>>        */
> >>>       ports.portA->pin7 = 0;
> >>>       ports.portA->pin7 = 1;
> >>>
> >>>       for(;;)
> >>>
> >>>       return 0;
> >>>}
> >>>
> >>>
> >>>Hope anyone has the time to help me. I am really stuck.
> >>>
> >>>
> >>>
> >>>
> >>>
> >>The canonical way to change bits in ports in C language is not through
> >>the use of bitfields. You need to learn to use the C language's bitwise
> >>operators, in conjunction with a simple macro include in avr-libc:
> >>
> >>#include <avr/io.h>
> >>PORTA |= _BV(1);   // Set bit 1
> >>PORTA &= ~_BV(1);    // Clear bit 1
> >>
> >>See this thread on the AVR Freaks website on how to use these operators:
> >><http://www.avrfreaks.net/phpBB2/viewtopic.php?t=8764&highlight=programming+101>
> >>Go to the post that starts with "Programming 101", which should be
> >>highlighted.
> >>
> >>
> >>
> >
> >If I wanted to do it the "canonical way" would I have ask for it. But I
> >didn't, did I?
> >
> >I do wonder how your brain work. I wish I would have taken a degree in
> >philosophy instead of my master degree in computer science.
> >
> >But hey, thanks for the link. Here is a link so you can learn some
> >"Reading 101"
> >http://www.fisher-price.com/us/ms5/learn-to-read.asp
> >
> >Enjoy.
> >
> >Albert Seward
> >Uppsala University Sweden
> >
> >
> >
> >
>
>
> Albert Seward also wrote earlier:
>
> > Hope anyone has the time to help me. I am really stuck.
>
>
> Awww.  Isn't it nice that you were so polite when asking for help. And
> now look how rude you are. How were we to know your educational level,
> when you didn't mention it before? ;-) For that matter you didn't
> mention how overly sensitive you were to (incorrectly) assumed
> condescension.
>
> Since you have a Master's in CompSci, you should be able to figure it
> out from looking at the compiler sources.
>
> And next time, keep your personal rants off the list.
>
>
>
>
>

Yes you are right. Sorry. I apologize for by mail.

Albert