Return to mul_1.asm CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / x86 / pentium

Annotation of OpenXM_contrib/gmp/mpn/x86/pentium/mul_1.asm, Revision 1.1.1.2

1.1       maekawa     1: dnl  Intel Pentium mpn_mul_1 -- mpn by limb multiplication.
                      2:
1.1.1.2 ! ohara       3: dnl  Copyright 1992, 1994, 1996, 1999, 2000, 2002 Free Software Foundation,
1.1       maekawa     4: dnl  Inc.
                      5: dnl
                      6: dnl  This file is part of the GNU MP Library.
                      7: dnl
                      8: dnl  The GNU MP Library is free software; you can redistribute it and/or
                      9: dnl  modify it under the terms of the GNU Lesser General Public License as
                     10: dnl  published by the Free Software Foundation; either version 2.1 of the
                     11: dnl  License, or (at your option) any later version.
                     12: dnl
                     13: dnl  The GNU MP Library is distributed in the hope that it will be useful,
                     14: dnl  but WITHOUT ANY WARRANTY; without even the implied warranty of
                     15: dnl  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
                     16: dnl  Lesser General Public License for more details.
                     17: dnl
                     18: dnl  You should have received a copy of the GNU Lesser General Public
                     19: dnl  License along with the GNU MP Library; see the file COPYING.LIB.  If
                     20: dnl  not, write to the Free Software Foundation, Inc., 59 Temple Place -
                     21: dnl  Suite 330, Boston, MA 02111-1307, USA. */
                     22:
                     23: include(`../config.m4')
                     24:
                     25:
1.1.1.2 ! ohara      26: C P5: 12.0 cycles/limb
        !            27:
        !            28:
1.1       maekawa    29: C mp_limb_t mpn_mul_1 (mp_ptr dst, mp_srcptr src, mp_size_t size,
                     30: C                      mp_limb_t multiplier);
1.1.1.2 ! ohara      31: C mp_limb_t mpn_mul_1c (mp_ptr dst, mp_srcptr src, mp_size_t size,
        !            32: C                       mp_limb_t multiplier, mp_limb_t carry);
        !            33: C
1.1       maekawa    34:
1.1.1.2 ! ohara      35: defframe(PARAM_CARRY,     20)
1.1       maekawa    36: defframe(PARAM_MULTIPLIER,16)
                     37: defframe(PARAM_SIZE,      12)
                     38: defframe(PARAM_SRC,       8)
                     39: defframe(PARAM_DST,       4)
                     40:
1.1.1.2 ! ohara      41:        TEXT
        !            42:        ALIGN(8)
        !            43: PROLOGUE(mpn_mul_1c)
        !            44: deflit(`FRAME',0)
        !            45:
        !            46:        movl    PARAM_CARRY, %ecx
        !            47:        pushl   %esi            FRAME_pushl()
        !            48:
        !            49:        jmp     L(start_1c)
        !            50:
        !            51: EPILOGUE()
        !            52:
        !            53:
1.1       maekawa    54:        ALIGN(8)
                     55: PROLOGUE(mpn_mul_1)
1.1.1.2 ! ohara      56: deflit(`FRAME',0)
1.1       maekawa    57:
1.1.1.2 ! ohara      58:        xorl    %ecx, %ecx
        !            59:        pushl   %esi            FRAME_pushl()
1.1       maekawa    60:
1.1.1.2 ! ohara      61: L(start_1c):
1.1       maekawa    62:        movl    PARAM_SRC, %esi
1.1.1.2 ! ohara      63:        movl    PARAM_SIZE, %eax
        !            64:
        !            65:        shrl    %eax
        !            66:        jnz     L(two_or_more)
1.1       maekawa    67:
                     68:
1.1.1.2 ! ohara      69:        C one limb only
1.1       maekawa    70:
1.1.1.2 ! ohara      71:        movl    (%esi), %eax
        !            72:
        !            73:        mull    PARAM_MULTIPLIER
        !            74:
        !            75:        addl    %eax, %ecx
        !            76:        movl    PARAM_DST, %eax
        !            77:
        !            78:        adcl    $0, %edx
        !            79:        popl    %esi
        !            80:
        !            81:        movl    %ecx, (%eax)
        !            82:        movl    %edx, %eax
        !            83:
        !            84:        ret
        !            85:
        !            86:
        !            87: L(two_or_more):
        !            88:        C eax   size/2
        !            89:        C ebx
        !            90:        C ecx   carry
        !            91:        C edx
        !            92:        C esi   src
        !            93:        C edi
        !            94:        C ebp
        !            95:
        !            96:        pushl   %edi            FRAME_pushl()
        !            97:        pushl   %ebx            FRAME_pushl()
        !            98:
        !            99:        movl    PARAM_DST, %edi
        !           100:        leal    -1(%eax), %ebx          C size/2-1
1.1       maekawa   101:
1.1.1.2 ! ohara     102:        notl    %ebx                    C -size, preserve carry
1.1       maekawa   103:
1.1.1.2 ! ohara     104:        leal    (%esi,%eax,8), %esi     C src end
        !           105:        leal    (%edi,%eax,8), %edi     C dst end
1.1       maekawa   106:
1.1.1.2 ! ohara     107:        pushl   %ebp            FRAME_pushl()
        !           108:        jnc     L(top)
1.1       maekawa   109:
1.1.1.2 ! ohara     110:
        !           111:        C size was odd, process one limb separately
        !           112:
        !           113:        movl    (%esi,%ebx,8), %eax
        !           114:        addl    $4, %esi
        !           115:
        !           116:        mull    PARAM_MULTIPLIER
        !           117:
        !           118:        addl    %ecx, %eax
        !           119:        movl    %edx, %ecx
        !           120:
        !           121:        movl    %eax, (%edi,%ebx,8)
        !           122:        leal    4(%edi), %edi
        !           123:
        !           124:
        !           125: L(top):
        !           126:        C eax
        !           127:        C ebx   counter, negative
        !           128:        C ecx   carry
        !           129:        C edx
        !           130:        C esi   src end
        !           131:        C edi   dst end
        !           132:        C ebp
        !           133:
        !           134:        adcl    $0, %ecx
        !           135:        movl    (%esi,%ebx,8), %eax
        !           136:
        !           137:        mull    PARAM_MULTIPLIER
        !           138:
        !           139:        movl    %edx, %ebp
        !           140:        addl    %eax, %ecx
        !           141:
        !           142:        adcl    $0, %ebp
        !           143:        movl    4(%esi,%ebx,8), %eax
        !           144:
        !           145:        mull    PARAM_MULTIPLIER
        !           146:
        !           147:        movl    %ecx, (%edi,%ebx,8)
        !           148:        addl    %ebp, %eax
        !           149:
        !           150:        movl    %eax, 4(%edi,%ebx,8)
        !           151:        incl    %ebx
        !           152:
        !           153:        movl    %edx, %ecx
        !           154:        jnz     L(top)
        !           155:
        !           156:
        !           157:        adcl    $0, %ecx
1.1       maekawa   158:        popl    %ebp
1.1.1.2 ! ohara     159:
        !           160:        movl    %ecx, %eax
1.1       maekawa   161:        popl    %ebx
1.1.1.2 ! ohara     162:
1.1       maekawa   163:        popl    %edi
1.1.1.2 ! ohara     164:        popl    %esi
        !           165:
1.1       maekawa   166:        ret
                    167:
                    168: EPILOGUE()