asm ("...") on PPC

[Geir Frode Raanes <geirfrs at invalid dot ed dot ntnu dot no>]

I am attemting to create the tightest loop possible on my
PowerPC (MBX860 PowerQUICC,) copying data from a memory 
mapped IO region to actual RAM.

For this I intend to use inline assembly as it seems that
GCC does not make use of the dedicated loop count register
(CRT.) I will copy from address "ioptr" to address "memptr."

In other words I try to write the following C code in assembly:

 unsigned int linjeteller = 0xFF, *ioptr, *memptr;

 do {
   *memptr++ = *ioptr++;
 } while (linjeteller--);  /* FF -> 00 == 256 Words == 1 KByte */



The equivalent of the above code should be something like:

    asm volatile ("mtctr %3         /* crt = linjeteller */
                   subi %0,4        /* ioptr--  */
                   subi %1,4        /* memptr-- */
        linjeloop:
                   lwzu r9,4(%0);   /* get source [++ioptr] */
                   stwu r9,4(%1);   /* store target [++memptr] */
                   bdnz+ linjeloop  /* loop until crt == 0 */ "
    : "=b" (ioptr), "=b" (memptr)
    :  "0" (ioptr),  "1" (memptr), "r" (linjeteller)
    : "r9", "ctr" );

Which produces the result I expect in a separate .c file, 
but is completely ignored inside the larger actual .c file,
even with optimalization turned off. The compiler knows it's
there because if I deliberately insert errors into the above
expression GCC complains. Oh well - I'll figure that one out later.


Now, I get the result I expect. Unfortunately, this code is
not accepteable for PPC/vxWorks/GAS. This is what I get:

          mtctr 4
          subi 3,4  <- "Error: missing operand" Fix: subi 3,0,4
          subi 4,4  <- "Error: missing operand" Fix: subi 4,0,4
linjeloop:
          lwzu r9,4(3)   <- "Error: unsupported relocation type" ???
          stwu r9,0,4(4) <- "Error: unsupported relocation type" ???
          bdnz+ linjeloop

"Error: unsupported relocation type" But this is supposed to be
correct code according to the PowerPC manual. What is GAS really
complaining about here? The following is what GCC produced:
(with: asm ("mfspr %0,%1" : "=r" (IMMREG) : "i" (638));)

        mfspr 0,638
        stw 0,0(11)        <- Almost same thing.
                              Apart from the [u]pdate flag;
                               i.e. EA[0(r11)] -> r11
        addis 9,0,ioptr@ha <- What's this for addressing mode
        addi 11,9,ioptr@l  <- And this? Lot's of "@ha" and "@l"

I really don't know what syntax GAS expects so I could use some
help here. Also, I would like someone to confirm that my 

 asm ("instruction" : OUT : IN : AFFECTED ) 

settings are correct for the load/store with Effective address
[u]pdate case above.


--
  ******************************************************
  Never ever underestimate the power of human stupidity.
  -Robert Anson Heinlein

		GeirFRS@invalid.and.so.forth
  ******************************************************


------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.cygnus.com