arch/m68k-all/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 /* 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(SMult64,Utility,33)
  26     .globl  AROS_SLIB_ENTRY(SMult64_020,Utility,33)
  27
  28     /* Required by SMult64() */
  29     .globl  AROS_SLIB_ENTRY(UMult64,Utility,34)
  30
  31     .type   AROS_SLIB_ENTRY(SMult64,Utility,33),@function
  32     .type   AROS_SLIB_ENTRY(SMult64_020,Utility,33),@function
  33
  34
  35 AROS_SLIB_ENTRY(SMult64_020,Utility,33):
  36     muls.l  %d0,%d0:%d1
  37     rts
  38
  39 /* How do I do this, again consider:
  40       (a^16 + b) * (c^16 + d)
  41     = ac^32 + (ad + bc)^16 + bd
  42
  43     I tried to think of a way of doing this with the mulu.l instr,
  44     but I couldn't so I'll just use the mulu.w. Its quicker than
  45     an unsupp integer instruction anyway :)
  46 */
  47
  48 /* Have to change the sign... */
  49     .balign 4
  50 AROS_SLIB_ENTRY(SMult64,Utility,33):
  51     move.l  %d2,-(%sp)
  52     moveq   #0,%d2
  53     tst.l   %d0
  54     bpl.s   .Lispos1
  55     neg.l   %d0
  56     addq.l  #1,%d2
  57 .Lispos1:
  58     tst.l   %d1
  59     bpl.s   .Lispos2
  60     neg.l   %d1
  61     subq.l  #1,%d2
  62
  63     /* Ok, so if d2 != 0, then the sign was changed */
  64 .Lispos2:
  65     jsr     AROS_SLIB_ENTRY(UMult64,Utility,34)
  66     tst.l   %d0
  67     beq.s   .Lispos
  68     moveq   #0,%d2
  69
  70     /* Ok we have to change the sign, 2's comp = 1's comp + 1 */
  71     not.l   %d0
  72     not.l   %d1
  73     /* Add 1 to low order 32 bits */
  74     addq.l  #1,%d1
  75     /* Add 0 and the carry to the high 32 bits */
  76     addx.l  %d2,%d0
  77 .Lispos:
  78     move.l  (%sp)+,%d2
  79     rts
  80