Return to submul_1.asm CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / power

Annotation of OpenXM_contrib/gmp/mpn/power/submul_1.asm, Revision 1.1.1.1

1.1       ohara       1: dnl  IBM POWER mpn_submul_1 -- Multiply a limb vector with a limb and subtract
                      2: dnl  the result from a second limb vector.
                      3:
                      4: dnl  Copyright 1992, 1994, 1999, 2000, 2001 Free Software Foundation, Inc.
                      5:
                      6: dnl  This file is part of the GNU MP Library.
                      7:
                      8: dnl  The GNU MP Library is free software; you can redistribute it and/or modify
                      9: dnl  it under the terms of the GNU Lesser General Public License as published
                     10: dnl  by the Free Software Foundation; either version 2.1 of the License, or (at
                     11: dnl  your option) any later version.
                     12:
                     13: dnl  The GNU MP Library is distributed in the hope that it will be useful, but
                     14: dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
                     15: dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
                     16: dnl  License for more details.
                     17:
                     18: dnl  You should have received a copy of the GNU Lesser General Public License
                     19: dnl  along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
                     20: dnl  the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
                     21: dnl  MA 02111-1307, USA.
                     22:
                     23:
                     24: dnl  INPUT PARAMETERS
                     25: dnl  res_ptr   r3
                     26: dnl  s1_ptr    r4
                     27: dnl  size      r5
                     28: dnl  s2_limb   r6
                     29:
                     30: dnl  The POWER architecture has no unsigned 32x32->64 bit multiplication
                     31: dnl  instruction.  To obtain that operation, we have to use the 32x32->64
                     32: dnl  signed multiplication instruction, and add the appropriate compensation to
                     33: dnl  the high limb of the result.  We add the multiplicand if the multiplier
                     34: dnl  has its most significant bit set, and we add the multiplier if the
                     35: dnl  multiplicand has its most significant bit set.  We need to preserve the
                     36: dnl  carry flag between each iteration, so we have to compute the compensation
                     37: dnl  carefully (the natural, srai+and doesn't work).  Since all POWER can
                     38: dnl  branch in zero cycles, we use conditional branches to for the additions.
                     39:
                     40: include(`../config.m4')
                     41:
                     42: ASM_START()
                     43: PROLOGUE(mpn_submul_1)
                     44:        cal     3,-4(3)
                     45:        l       0,0(4)
                     46:        cmpi    0,6,0
                     47:        mtctr   5
                     48:        mul     9,0,6
                     49:        srai    7,0,31
                     50:        and     7,7,6
                     51:        mfmq    11
                     52:        cax     9,9,7
                     53:        l       7,4(3)
                     54:        sf      8,11,7          C add res_limb
                     55:        a       11,8,11         C invert cy (r11 is junk)
                     56:        blt     Lneg
                     57: Lpos:  bdz     Lend
                     58:
                     59: Lploop:        lu      0,4(4)
                     60:        stu     8,4(3)
                     61:        cmpi    0,0,0
                     62:        mul     10,0,6
                     63:        mfmq    0
                     64:        ae      11,0,9          C low limb + old_cy_limb + old cy
                     65:        l       7,4(3)
                     66:        aze     10,10           C propagate cy to new cy_limb
                     67:        sf      8,11,7          C add res_limb
                     68:        a       11,8,11         C invert cy (r11 is junk)
                     69:        bge     Lp0
                     70:        cax     10,10,6         C adjust high limb for negative limb from s1
                     71: Lp0:   bdz     Lend0
                     72:        lu      0,4(4)
                     73:        stu     8,4(3)
                     74:        cmpi    0,0,0
                     75:        mul     9,0,6
                     76:        mfmq    0
                     77:        ae      11,0,10
                     78:        l       7,4(3)
                     79:        aze     9,9
                     80:        sf      8,11,7
                     81:        a       11,8,11         C invert cy (r11 is junk)
                     82:        bge     Lp1
                     83:        cax     9,9,6           C adjust high limb for negative limb from s1
                     84: Lp1:   bdn     Lploop
                     85:
                     86:        b       Lend
                     87:
                     88: Lneg:  cax     9,9,0
                     89:        bdz     Lend
                     90: Lnloop:        lu      0,4(4)
                     91:        stu     8,4(3)
                     92:        cmpi    0,0,0
                     93:        mul     10,0,6
                     94:        mfmq    7
                     95:        ae      11,7,9
                     96:        l       7,4(3)
                     97:        ae      10,10,0         C propagate cy to new cy_limb
                     98:        sf      8,11,7          C add res_limb
                     99:        a       11,8,11         C invert cy (r11 is junk)
                    100:        bge     Ln0
                    101:        cax     10,10,6         C adjust high limb for negative limb from s1
                    102: Ln0:   bdz     Lend0
                    103:        lu      0,4(4)
                    104:        stu     8,4(3)
                    105:        cmpi    0,0,0
                    106:        mul     9,0,6
                    107:        mfmq    7
                    108:        ae      11,7,10
                    109:        l       7,4(3)
                    110:        ae      9,9,0           C propagate cy to new cy_limb
                    111:        sf      8,11,7          C add res_limb
                    112:        a       11,8,11         C invert cy (r11 is junk)
                    113:        bge     Ln1
                    114:        cax     9,9,6           C adjust high limb for negative limb from s1
                    115: Ln1:   bdn     Lnloop
                    116:        b       Lend
                    117:
                    118: Lend0: cal     9,0(10)
                    119: Lend:  st      8,4(3)
                    120:        aze     3,9
                    121:        br
                    122: EPILOGUE(mpn_submul_1)