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