Return to diveby3.asm CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / x86 / k7

Annotation of OpenXM_contrib/gmp/mpn/x86/k7/diveby3.asm, Revision 1.1.1.2

1.1       maekawa     1: dnl  AMD K7 mpn_divexact_by3 -- mpn division by 3, expecting no remainder.
                      2:
1.1.1.2 ! ohara       3: dnl  Copyright 2000, 2002 Free Software Foundation, Inc.
1.1       maekawa     4: dnl
                      5: dnl  This file is part of the GNU MP Library.
                      6: dnl
                      7: dnl  The GNU MP Library is free software; you can redistribute it and/or
                      8: dnl  modify it under the terms of the GNU Lesser General Public License as
                      9: dnl  published by the Free Software Foundation; either version 2.1 of the
                     10: dnl  License, or (at your option) any later version.
                     11: dnl
                     12: dnl  The GNU MP Library is distributed in the hope that it will be useful,
                     13: dnl  but WITHOUT ANY WARRANTY; without even the implied warranty of
                     14: dnl  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
                     15: dnl  Lesser General Public License for more details.
                     16: dnl
                     17: dnl  You should have received a copy of the GNU Lesser General Public
                     18: dnl  License along with the GNU MP Library; see the file COPYING.LIB.  If
                     19: dnl  not, write to the Free Software Foundation, Inc., 59 Temple Place -
                     20: dnl  Suite 330, Boston, MA 02111-1307, USA.
                     21:
                     22: include(`../config.m4')
                     23:
                     24:
1.1.1.2 ! ohara      25: C K7: 8.0 cycles/limb
        !            26:
        !            27:
1.1       maekawa    28: C mp_limb_t mpn_divexact_by3c (mp_ptr dst, mp_srcptr src, mp_size_t size,
                     29: C                              mp_limb_t carry);
                     30:
                     31: defframe(PARAM_CARRY,16)
                     32: defframe(PARAM_SIZE, 12)
                     33: defframe(PARAM_SRC,   8)
                     34: defframe(PARAM_DST,   4)
                     35:
                     36: dnl  multiplicative inverse of 3, modulo 2^32
                     37: deflit(INVERSE_3,        0xAAAAAAAB)
                     38:
                     39: dnl  ceil(b/3) and floor(b*2/3) where b=2^32
                     40: deflit(ONE_THIRD_CEIL,   0x55555556)
                     41: deflit(TWO_THIRDS_FLOOR, 0xAAAAAAAA)
                     42:
1.1.1.2 ! ohara      43:        TEXT
1.1       maekawa    44:        ALIGN(32)
                     45:
                     46: PROLOGUE(mpn_divexact_by3c)
                     47: deflit(`FRAME',0)
                     48:
                     49:        movl    PARAM_SRC, %ecx
                     50:        pushl   %ebx                    defframe_pushl(SAVE_EBX)
                     51:
                     52:        movl    PARAM_CARRY, %ebx
                     53:        pushl   %ebp                    defframe_pushl(SAVE_EBP)
                     54:
                     55:        movl    PARAM_SIZE, %ebp
                     56:        pushl   %edi                    defframe_pushl(SAVE_EDI)
                     57:
                     58:        movl    (%ecx), %eax            C src low limb
                     59:        pushl   %esi                    defframe_pushl(SAVE_ESI)
                     60:
                     61:        movl    PARAM_DST, %edi
                     62:        movl    $TWO_THIRDS_FLOOR, %esi
                     63:        leal    -4(%ecx,%ebp,4), %ecx   C &src[size-1]
                     64:
                     65:        subl    %ebx, %eax
                     66:
                     67:        setc    %bl
                     68:        decl    %ebp
                     69:        jz      L(last)
                     70:
                     71:        leal    (%edi,%ebp,4), %edi     C &dst[size-1]
                     72:        negl    %ebp
                     73:
                     74:
                     75:        ALIGN(16)
                     76: L(top):
                     77:        C eax   src limb, carry subtracted
                     78:        C ebx   carry limb (0 or 1)
                     79:        C ecx   &src[size-1]
                     80:        C edx   scratch
                     81:        C esi   TWO_THIRDS_FLOOR
                     82:        C edi   &dst[size-1]
                     83:        C ebp   counter, limbs, negative
                     84:
                     85:        imull   $INVERSE_3, %eax, %edx
                     86:
                     87:        movl    4(%ecx,%ebp,4), %eax    C next src limb
                     88:        cmpl    $ONE_THIRD_CEIL, %edx
                     89:
                     90:        sbbl    $-1, %ebx               C +1 if result>=ceil(b/3)
                     91:        cmpl    %edx, %esi
                     92:
                     93:        sbbl    %ebx, %eax              C and further 1 if result>=ceil(b*2/3)
                     94:        movl    %edx, (%edi,%ebp,4)
                     95:        incl    %ebp
                     96:
                     97:        setc    %bl                     C new carry
                     98:        jnz     L(top)
                     99:
                    100:
                    101:
                    102: L(last):
                    103:        C eax   src limb, carry subtracted
                    104:        C ebx   carry limb (0 or 1)
                    105:        C ecx   &src[size-1]
                    106:        C edx   scratch
                    107:        C esi   multiplier
                    108:        C edi   &dst[size-1]
                    109:        C ebp
                    110:
1.1.1.2 ! ohara     111:        imull   $INVERSE_3, %eax, %eax
1.1       maekawa   112:
                    113:        cmpl    $ONE_THIRD_CEIL, %eax
                    114:        movl    %eax, (%edi)
                    115:        movl    SAVE_EBP, %ebp
                    116:
                    117:        sbbl    $-1, %ebx               C +1 if eax>=ceil(b/3)
                    118:        cmpl    %eax, %esi
                    119:        movl    $0, %eax
                    120:
                    121:        adcl    %ebx, %eax              C further +1 if eax>=ceil(b*2/3)
                    122:        movl    SAVE_EDI, %edi
                    123:        movl    SAVE_ESI, %esi
                    124:
                    125:        movl    SAVE_EBX, %ebx
                    126:        addl    $FRAME, %esp
                    127:
                    128:        ret
                    129:
                    130: EPILOGUE()