Return to sub_n.asm CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / powerpc64

Annotation of OpenXM_contrib/gmp/mpn/powerpc64/sub_n.asm, Revision 1.1.1.2

1.1.1.2 ! ohara       1: dnl  PowerPC-64 mpn_sub_n -- Subtract two limb vectors of the same length > 0
        !             2: dnl  and store difference in a third limb vector.
1.1       maekawa     3:
1.1.1.2 ! ohara       4: dnl  Copyright 1999, 2000, 2001 Free Software Foundation, Inc.
1.1       maekawa     5:
1.1.1.2 ! ohara       6: dnl  This file is part of the GNU MP Library.
1.1       maekawa     7:
1.1.1.2 ! ohara       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.
1.1       maekawa    22:
                     23: include(`../config.m4')
                     24:
1.1.1.2 ! ohara      25: C INPUT PARAMETERS
        !            26: C res_ptr      r3
        !            27: C s1_ptr       r4
        !            28: C s2_ptr       r5
        !            29: C size         r6
        !            30:
        !            31: C This code runs at 1.5 cycles/limb on the PPC630.
        !            32:
1.1       maekawa    33: ASM_START()
                     34: PROLOGUE(mpn_sub_n)
1.1.1.2 ! ohara      35:        mtctr   r6              C copy size into CTR
        !            36:        addic   r0,r6,-1        C set cy
        !            37:        ld      r8,0(r4)        C load least significant s1 limb
        !            38:        ld      r0,0(r5)        C load least significant s2 limb
        !            39:        addi    r3,r3,-8        C offset res_ptr, it's updated before it's used
        !            40:        bdz     .Lend           C If done, skip loop
        !            41: .Loop: ld      r9,8(r4)        C load s1 limb
        !            42:        ld      r10,8(r5)       C load s2 limb
        !            43:        subfe   r7,r0,r8        C subtract limbs with cy, set cy
        !            44:        std     r7,8(r3)        C store result limb
        !            45:        bdz     .Lexit          C decrement CTR and exit if done
        !            46:        ldu     r8,16(r4)       C load s1 limb and update s1_ptr
        !            47:        ldu     r0,16(r5)       C load s2 limb and update s2_ptr
        !            48:        subfe   r7,r10,r9       C subtract limbs with cy, set cy
        !            49:        stdu    r7,16(r3)       C store result limb and update res_ptr
        !            50:        bdnz    .Loop           C decrement CTR and loop back
1.1       maekawa    51:
                     52: .Lend: subfe   r7,r0,r8
1.1.1.2 ! ohara      53:        std     r7,8(r3)        C store ultimate result limb
        !            54:        subfe   r3,r0,r0        C load !cy into ...
        !            55:        subfic  r3,r3,0         C ... return value register
1.1       maekawa    56:        blr
                     57: .Lexit:        subfe   r7,r10,r9
                     58:        std     r7,16(r3)
1.1.1.2 ! ohara      59:        subfe   r3,r0,r0        C load !cy into ...
        !            60:        subfic  r3,r3,0         C ... return value register
1.1       maekawa    61:        blr
                     62: EPILOGUE(mpn_sub_n)