arch/m68k-all/utility/umult64.s

   1 /*
   2     Copyright © 1995-2001, The AROS Development Team. All rights reserved.
   3     $Id$
   4
   5     Desc: Utility 64 bit multiplication routines. m68k version.
   6     Lang: english
   7 */
   8
   9 /* SMult64()/UMult64():
  10     These are the signed/unsigned 64 bit multiplication routines.
  11     There are two possibilities here because as of the 060 the
  12     32*32->64 bit result instructions are not supported, and I haven't
  13     quite figured out how to do this using the 32 bit ops yet (can't be
  14     that hard though).
  15
  16     Still, emulating is faster than a unsup integer instruction except.
  17
  18 */
  19
  20     #include "aros/m68k/asm.h"
  21
  22     .text
  23     .balign 4
  24
  25     .globl  AROS_SLIB_ENTRY(UMult64,Utility,34)
  26     .globl  AROS_SLIB_ENTRY(UMult64_020,Utility,34)
  27
  28     .type   AROS_SLIB_ENTRY(UMult64,Utility,34),@function
  29     .type   AROS_SLIB_ENTRY(UMult64_020,Utility,34),@function
  30
  31
  32 AROS_SLIB_ENTRY(UMult64_020,Utility,34):
  33     mulu.l  %d0,%d0:%d1
  34     rts
  35
  36 /* How do I do this, again consider:
  37       (a^16 + b) * (c^16 + d)
  38     = ac^32 + (ad + bc)^16 + bd
  39
  40     I tried to think of a way of doing this with the mulu.l instr,
  41     but I couldn't so I'll just use the mulu.w. Its quicker than
  42     an unsupp integer instruction anyway :)
  43 */
  44
  45     .balign 4
  46 AROS_SLIB_ENTRY(UMult64,Utility,34):
  47     movem.l %d2-%d5,-(%sp)
  48     /* Set up some registers */
  49     move.l  %d0,%d2
  50     move.l  %d1,%d3
  51     move.l  %d0,%d4   /* d */
  52     move.l  %d1,%d5   /* b */
  53     swap    %d2      /* a */
  54     swap    %d3      /* c */
  55
  56
  57     /* Firstly, find the product bd */
  58     mulu    %d5,%d1   /* d1 = bd */
  59     swap    %d1      /* d1 = (bd)^16 */
  60
  61     /* Then find ac, put in d0 */
  62     mulu    %d3,%d0   /* d0 = ac */
  63
  64     /* Next find ad, bc, and add together */
  65     mulu    %d2,%d4
  66     mulu    %d3,%d5
  67     add.l   %d5,%d4
  68
  69     /*
  70         Add the low 16 bits to d1, then add upper 16 bits to d0
  71         But make sure we carry the 1...
  72
  73         Apparently swap doesn't affect the X bit.
  74     */
  75     add.w   %d4,%d1
  76     swap    %d4
  77     addx.w  %d4,%d0
  78
  79     /* All that remains to do is to flip d1 around the right way */
  80     swap    %d1
  81     movem.l (%sp)+,%d2-%d5
  82
  83     rts