🚀 go-pugleaf

In article <9470@uqcspe.cs.uq.oz.au> warwick@cs.uq.oz.au writes:
>psteffn@hal.gnu.ai.mit.edu (Paul Steffen) writes:
>>Inline assembly code is pure hell!!
>Obviously someone who has never used the "asm" extension in GNU C.

I am aware of the compiler option for reducing the number of used
registers.  I did say something to this effect in the message you
replied to.

>With GNU C, you specify the output values, input values, and destroyed
>registers, and it gives you registers to use in the asm function, and
>takes care of getting the values into those regs, not using them for
>optimisations across the asm inline, etc.  eg:
>
>void swap(int* first, int* second)
>/* Really fast swap :-) */
>{
>    printf("Now executing the Really Fast swap routine!!!");
>    printf("This program must be running very fast now");
>    printf("that you are using this fast routine!");
>
>    asm("   movel  %0@,d0
>            movel  %1@,%0@
>            movel  d0,%1@
>       " :             /* no outputs */
>         : (a) first,  /* %0 will be an adr. reg (a), with the value of first */
>           (a) second  /* %1 will be an address reg with the value of second */
>         : "d0"        /* d0 will be trashed by this asm code, so make sure */
>    );
>    printf("Sorry, I chatted so much, you missed the routine happening!!");
>

Sure, but if you had coded completely in assembler, couldn't you have done
an 'exg Dx,Dx' which is approximately 5 times faster?  With assembler, you
know which registers are being used by what at any given time.

>Warwick

STratoHAK/TPPi