arch/.unmaintained/m68k-linux/utility/smult64.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     .text
  22     .balign 16
  23
  24     .globl  AROS_SLIB_ENTRY(SMult64,Utility)
  25     .globl  AROS_SLIB_ENTRY(SMult64_020,Utility)
  26
  27     /* Required by SMult64() */
  28     .globl  AROS_SLIB_ENTRY(UMult64,Utility)
  29
  30     .type   AROS_SLIB_ENTRY(SMult64,Utility),@function
  31     .type   AROS_SLIB_ENTRY(SMult64_020,Utility),@function
  32
  33
  34 AROS_SLIB_ENTRY(SMult64_020,Utility):
  35     muls.l  %d0,%d0:%d1
  36     rts
  37
  38 /* How do I do this, again consider:
  39       (a^16 + b) * (c^16 + d)
  40     = ac^32 + (ad + bc)^16 + bd
  41
  42     I tried to think of a way of doing this with the mulu.l instr,
  43     but I couldn't so I'll just use the mulu.w. Its quicker than
  44     an unsupp integer instruction anyway :)
  45 */
  46
  47 /* Have to change the sign... */
  48 AROS_SLIB_ENTRY(SMult64,Utility):
  49     move.l  %d2,-(%a7)
  50     moveq   #0,%d2
  51     tst.l   %d0
  52     jbpl    .ispos1
  53     neg.l   %d0
  54     addq.l  #1,%d2
  55 .ispos1:
  56     tst.l   %d1
  57     jbpl    .ispos2
  58     neg.l   %d1
  59     subq.l  #1,%d2
  60
  61     /* Ok, so if d2 != 0, then the sign was changed */
  62 .ispos2:
  63     bsr.s   AROS_SLIB_ENTRY(UMult64,Utility)
  64     tst.l   %d0
  65     jbeq    .ispos
  66     moveq   #0,%d2
  67
  68     /* Ok we have to change the sign, 2's comp = 1's comp + 1 */
  69     not.l   %d0
  70     not.l   %d1
  71     /* Add 1 to low order 32 bits */
  72     addq.l  #1,%d1
  73     /* Add 0 and the carry to the high 32 bits */
  74     addx.l  %d2,%d0
  75 .ispos:
  76     move.l  (%sp)+,%d2
  77     rts
  78