Return to aorsmul_1.asm CVS log

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

Annotation of OpenXM_contrib/gmp/mpn/x86/pentium/aorsmul_1.asm, Revision 1.1.1.2

1.1       maekawa     1: dnl  Intel Pentium mpn_addmul_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: 14.0 cycles/limb
        !            27:
        !            28:
1.1       maekawa    29: ifdef(`OPERATION_addmul_1', `
                     30:       define(M4_inst,        addl)
                     31:       define(M4_function_1,  mpn_addmul_1)
1.1.1.2 ! ohara      32:       define(M4_function_1c, mpn_addmul_1c)
1.1       maekawa    33:
                     34: ',`ifdef(`OPERATION_submul_1', `
                     35:       define(M4_inst,        subl)
                     36:       define(M4_function_1,  mpn_submul_1)
1.1.1.2 ! ohara      37:       define(M4_function_1c, mpn_submul_1c)
1.1       maekawa    38:
                     39: ',`m4_error(`Need OPERATION_addmul_1 or OPERATION_submul_1
                     40: ')')')
                     41:
1.1.1.2 ! ohara      42: MULFUNC_PROLOGUE(mpn_addmul_1 mpn_addmul_1c mpn_submul_1 mpn_submul_1c)
1.1       maekawa    43:
                     44:
1.1.1.2 ! ohara      45: C mp_limb_t mpn_addmul_1 (mp_ptr dst, mp_srcptr src, mp_size_t size,
        !            46: C                         mp_limb_t mult);
        !            47: C mp_limb_t mpn_addmul_1c (mp_ptr dst, mp_srcptr src, mp_size_t size,
        !            48: C                          mp_limb_t mult, mp_limb_t carry);
        !            49: C
        !            50: C mp_limb_t mpn_submul_1 (mp_ptr dst, mp_srcptr src, mp_size_t size,
        !            51: C                         mp_limb_t mult);
        !            52: C mp_limb_t mpn_submul_1c (mp_ptr dst, mp_srcptr src, mp_size_t size,
        !            53: C                          mp_limb_t mult, mp_limb_t carry);
        !            54: C
1.1       maekawa    55:
1.1.1.2 ! ohara      56: defframe(PARAM_CARRY,     20)
1.1       maekawa    57: defframe(PARAM_MULTIPLIER,16)
                     58: defframe(PARAM_SIZE,      12)
                     59: defframe(PARAM_SRC,       8)
                     60: defframe(PARAM_DST,       4)
                     61:
1.1.1.2 ! ohara      62:        TEXT
        !            63:
1.1       maekawa    64:        ALIGN(8)
1.1.1.2 ! ohara      65: PROLOGUE(M4_function_1c)
        !            66: deflit(`FRAME',0)
1.1       maekawa    67:
1.1.1.2 ! ohara      68:        movl    PARAM_CARRY, %ecx
        !            69:        pushl   %esi            FRAME_pushl()
        !            70:
        !            71:        jmp     L(start_1c)
        !            72:
        !            73: EPILOGUE()
        !            74:
        !            75:
        !            76:        ALIGN(8)
1.1       maekawa    77: PROLOGUE(M4_function_1)
1.1.1.2 ! ohara      78: deflit(`FRAME',0)
1.1       maekawa    79:
1.1.1.2 ! ohara      80:        xorl    %ecx, %ecx
        !            81:        pushl   %esi            FRAME_pushl()
1.1       maekawa    82:
1.1.1.2 ! ohara      83: L(start_1c):
1.1       maekawa    84:        movl    PARAM_SRC, %esi
1.1.1.2 ! ohara      85:        movl    PARAM_SIZE, %eax
1.1       maekawa    86:
1.1.1.2 ! ohara      87:        pushl   %edi            FRAME_pushl()
        !            88:        pushl   %ebx            FRAME_pushl()
        !            89:
        !            90:        movl    PARAM_DST, %edi
        !            91:        leal    -1(%eax), %ebx          C size-1
        !            92:
        !            93:        leal    (%esi,%eax,4), %esi
        !            94:        xorl    $-1, %ebx               C -size, and clear carry
        !            95:
        !            96:        leal    (%edi,%eax,4), %edi
        !            97:
        !            98: L(top):
        !            99:        C eax
        !           100:        C ebx   counter, negative
        !           101:        C ecx   carry
        !           102:        C edx
        !           103:        C esi   src end
        !           104:        C edi   dst end
        !           105:        C ebp
1.1       maekawa   106:
1.1.1.2 ! ohara     107:        adcl    $0, %ecx
        !           108:        movl    (%esi,%ebx,4), %eax
1.1       maekawa   109:
1.1.1.2 ! ohara     110:        mull    PARAM_MULTIPLIER
1.1       maekawa   111:
1.1.1.2 ! ohara     112:        addl    %ecx, %eax
        !           113:        movl    (%edi,%ebx,4), %ecx
1.1       maekawa   114:
                    115:        adcl    $0, %edx
1.1.1.2 ! ohara     116:        M4_inst %eax, %ecx
1.1       maekawa   117:
1.1.1.2 ! ohara     118:        movl    %ecx, (%edi,%ebx,4)
        !           119:        incl    %ebx
1.1       maekawa   120:
1.1.1.2 ! ohara     121:        movl    %edx, %ecx
        !           122:        jnz     L(top)
1.1       maekawa   123:
1.1.1.2 ! ohara     124:
        !           125:        adcl    $0, %ecx
1.1       maekawa   126:        popl    %ebx
1.1.1.2 ! ohara     127:
        !           128:        movl    %ecx, %eax
1.1       maekawa   129:        popl    %edi
1.1.1.2 ! ohara     130:
        !           131:        popl    %esi
        !           132:
1.1       maekawa   133:        ret
                    134:
                    135: EPILOGUE()