Return to add_n.asm CVS log

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

Annotation of OpenXM_contrib/gmp/mpn/alpha/add_n.asm, Revision 1.1.1.2

1.1       maekawa     1: dnl  Alpha mpn_add_n -- Add two limb vectors of the same length > 0 and
                      2: dnl  store sum in a third limb vector.
                      3:
1.1.1.2 ! ohara       4: dnl  Copyright 1995, 2000 Free Software Foundation, Inc.
1.1       maekawa     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
1.1.1.2 ! ohara       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.
1.1       maekawa    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: include(`../config.m4')
                     24:
                     25: dnl  INPUT PARAMETERS
                     26: dnl  res_ptr   r16
                     27: dnl  s1_ptr    r17
                     28: dnl  s2_ptr    r18
                     29: dnl  size      r19
                     30:
                     31: ASM_START()
                     32: PROLOGUE(mpn_add_n)
                     33:        ldq     r3,0(r17)
                     34:        ldq     r4,0(r18)
                     35:
                     36:        subq    r19,1,r19
                     37:        and     r19,4-1,r2      C number of limbs in first loop
                     38:        bis     r31,r31,r0
                     39:        beq     r2,$L0          C if multiple of 4 limbs, skip first loop
                     40:
                     41:        subq    r19,r2,r19
                     42:
                     43: $Loop0:        subq    r2,1,r2
                     44:        ldq     r5,8(r17)
                     45:        addq    r4,r0,r4
                     46:        ldq     r6,8(r18)
                     47:        cmpult  r4,r0,r1
                     48:        addq    r3,r4,r4
                     49:        cmpult  r4,r3,r0
                     50:        stq     r4,0(r16)
                     51:        bis     r0,r1,r0
                     52:
                     53:        addq    r17,8,r17
                     54:        addq    r18,8,r18
                     55:        bis     r5,r5,r3
                     56:        bis     r6,r6,r4
                     57:        addq    r16,8,r16
                     58:        bne     r2,$Loop0
                     59:
                     60: $L0:   beq     r19,$Lend
                     61:
                     62:        ALIGN(8)
                     63: $Loop: subq    r19,4,r19
                     64:
                     65:        ldq     r5,8(r17)
                     66:        addq    r4,r0,r4
                     67:        ldq     r6,8(r18)
                     68:        cmpult  r4,r0,r1
                     69:        addq    r3,r4,r4
                     70:        cmpult  r4,r3,r0
                     71:        stq     r4,0(r16)
                     72:        bis     r0,r1,r0
                     73:
                     74:        ldq     r3,16(r17)
                     75:        addq    r6,r0,r6
                     76:        ldq     r4,16(r18)
                     77:        cmpult  r6,r0,r1
                     78:        addq    r5,r6,r6
                     79:        cmpult  r6,r5,r0
                     80:        stq     r6,8(r16)
                     81:        bis     r0,r1,r0
                     82:
                     83:        ldq     r5,24(r17)
                     84:        addq    r4,r0,r4
                     85:        ldq     r6,24(r18)
                     86:        cmpult  r4,r0,r1
                     87:        addq    r3,r4,r4
                     88:        cmpult  r4,r3,r0
                     89:        stq     r4,16(r16)
                     90:        bis     r0,r1,r0
                     91:
                     92:        ldq     r3,32(r17)
                     93:        addq    r6,r0,r6
                     94:        ldq     r4,32(r18)
                     95:        cmpult  r6,r0,r1
                     96:        addq    r5,r6,r6
                     97:        cmpult  r6,r5,r0
                     98:        stq     r6,24(r16)
                     99:        bis     r0,r1,r0
                    100:
                    101:        addq    r17,32,r17
                    102:        addq    r18,32,r18
                    103:        addq    r16,32,r16
                    104:        bne     r19,$Loop
                    105:
                    106: $Lend: addq    r4,r0,r4
                    107:        cmpult  r4,r0,r1
                    108:        addq    r3,r4,r4
                    109:        cmpult  r4,r3,r0
                    110:        stq     r4,0(r16)
                    111:        bis     r0,r1,r0
                    112:        ret     r31,(r26),1
                    113: EPILOGUE(mpn_add_n)
                    114: ASM_END()