Return to mul_1.asm CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / powerpc32

Annotation of OpenXM_contrib/gmp/mpn/powerpc32/mul_1.asm, Revision 1.1.1.2

1.1       maekawa     1: dnl PowerPC-32 mpn_mul_1 -- Multiply a limb vector with a limb and store
                      2: dnl the result in a second limb vector.
                      3:
1.1.1.2 ! ohara       4: dnl Copyright 1995, 1997, 2000, 2002 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
                      9: dnl it under the terms of the GNU Lesser General Public License as published by
                     10: dnl the Free Software Foundation; either version 2.1 of the License, or (at your
                     11: dnl 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.
                     22:
                     23:
                     24: dnl INPUT PARAMETERS
                     25: dnl res_ptr    r3
                     26: dnl s1_ptr     r4
                     27: dnl size       r5
                     28: dnl s2_limb    r6
                     29:
                     30: dnl This is optimized for the PPC604 but it runs decently even on PPC601.  It
                     31: dnl has not been tested on a PPC603 since I don't have access to any such
                     32: dnl machines.
                     33:
                     34: include(`../config.m4')
                     35:
                     36: ASM_START()
                     37: PROLOGUE(mpn_mul_1)
                     38:        mtctr   r5
                     39:        addi    r3,r3,-4        C adjust res_ptr, it's offset before it's used
                     40:        li      r12,0           C clear upper product reg
                     41:        addic   r0,r0,0         C clear cy
                     42: C Start software pipeline
                     43:        lwz     r8,0(r4)
1.1.1.2 ! ohara      44:        bdz     L(end3)
1.1       maekawa    45:        lwzu    r9,4(r4)
                     46:        mullw   r11,r8,r6
                     47:        mulhwu  r0,r8,r6
1.1.1.2 ! ohara      48:        bdz     L(end1)
1.1       maekawa    49: C Software pipelined main loop
1.1.1.2 ! ohara      50: L(oop):        lwz     r8,4(r4)
1.1       maekawa    51:        mullw   r10,r9,r6
1.1.1.2 ! ohara      52:        adde    r5,r11,r12
1.1       maekawa    53:        mulhwu  r12,r9,r6
1.1.1.2 ! ohara      54:        stw     r5,4(r3)
        !            55:        bdz     L(end2)
1.1       maekawa    56:        lwzu    r9,8(r4)
                     57:        mullw   r11,r8,r6
1.1.1.2 ! ohara      58:        adde    r7,r10,r0
1.1       maekawa    59:        mulhwu  r0,r8,r6
1.1.1.2 ! ohara      60:        stwu    r7,8(r3)
        !            61:        bdnz    L(oop)
1.1       maekawa    62: C Finish software pipeline
1.1.1.2 ! ohara      63: L(end1):
        !            64:        mullw   r10,r9,r6
        !            65:        adde    r5,r11,r12
1.1       maekawa    66:        mulhwu  r12,r9,r6
1.1.1.2 ! ohara      67:        stw     r5,4(r3)
        !            68:        adde    r7,r10,r0
        !            69:        stwu    r7,8(r3)
1.1       maekawa    70:        addze   r3,r12
                     71:        blr
1.1.1.2 ! ohara      72: L(end2):
        !            73:        mullw   r11,r8,r6
        !            74:        adde    r7,r10,r0
1.1       maekawa    75:        mulhwu  r0,r8,r6
1.1.1.2 ! ohara      76:        stwu    r7,8(r3)
        !            77:        adde    r5,r11,r12
        !            78:        stw     r5,4(r3)
1.1       maekawa    79:        addze   r3,r0
                     80:        blr
1.1.1.2 ! ohara      81: L(end3):
        !            82:        mullw   r11,r8,r6
1.1       maekawa    83:        stw     r11,4(r3)
                     84:        mulhwu  r3,r8,r6
                     85:        blr
                     86: EPILOGUE(mpn_mul_1)