Return to sqr_diagonal.asm CVS log

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

Annotation of OpenXM_contrib/gmp/mpn/powerpc32/sqr_diagonal.asm, Revision 1.1.1.1

1.1       ohara       1: dnl PowerPC-32 mpn_sqr_diagonal.
                      2:
                      3: dnl Copyright 2001, 2002 Free Software Foundation, Inc.
                      4:
                      5: dnl This file is part of the GNU MP Library.
                      6:
                      7: dnl The GNU MP Library is free software; you can redistribute it and/or modify
                      8: dnl it under the terms of the GNU Lesser General Public License as published by
                      9: dnl the Free Software Foundation; either version 2.1 of the License, or (at your
                     10: dnl option) any later version.
                     11:
                     12: dnl The GNU MP Library is distributed in the hope that it will be useful, but
                     13: dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
                     14: dnl or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
                     15: dnl License for more details.
                     16:
                     17: dnl You should have received a copy of the GNU Lesser General Public License
                     18: dnl along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
                     19: dnl the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
                     20: dnl MA 02111-1307, USA.
                     21:
                     22:
                     23: dnl INPUT PARAMETERS
                     24: dnl res_ptr    r3
                     25: dnl s1_ptr     r4
                     26: dnl size       r5
                     27:
                     28: dnl This runs at 4 cycles/limb on the PPC604.  That's saturating the multiply
                     29: dnl unit.
                     30:
                     31: include(`../config.m4')
                     32:
                     33: ASM_START()
                     34: PROLOGUE(mpn_sqr_diagonal)
                     35:        lwz     r6,0(r4)
                     36:        mtctr   r5
                     37:
                     38:        addi    r3,r3,-4
                     39:        bdz     L(end1)
                     40:
                     41:        lwzu    r7,4(r4)
                     42:        mullw   r9,r6,r6
                     43:        mulhwu  r11,r6,r6
                     44:        bdz     L(end2)
                     45:
                     46:        lwzu    r6,4(r4)
                     47:        mullw   r8,r7,r7
                     48:        mulhwu  r10,r7,r7
                     49:        bdz     L(ende)
                     50:
                     51: L(oop):        lwzu    r7,4(r4)
                     52:        stw     r9,4(r3)
                     53:        mullw   r9,r6,r6
                     54:        stwu    r11,8(r3)
                     55:        mulhwu  r11,r6,r6
                     56:        bdz     L(endo)
                     57:        lwzu    r6,4(r4)
                     58:        stw     r8,4(r3)
                     59:        mullw   r8,r7,r7
                     60:        stwu    r10,8(r3)
                     61:        mulhwu  r10,r7,r7
                     62:        bdnz    L(oop)
                     63:
                     64: L(ende):
                     65:        stw     r9,4(r3)
                     66:        mullw   r9,r6,r6
                     67:        stw     r11,8(r3)
                     68:        mulhwu  r11,r6,r6
                     69:        stw     r8,12(r3)
                     70:        stw     r10,16(r3)
                     71:        stw     r9,20(r3)
                     72:        stw     r11,24(r3)
                     73:        blr
                     74: L(endo):
                     75:        stw     r8,4(r3)
                     76:        mullw   r8,r7,r7
                     77:        stw     r10,8(r3)
                     78:        mulhwu  r10,r7,r7
                     79:        stw     r9,12(r3)
                     80:        stw     r11,16(r3)
                     81:        stw     r8,20(r3)
                     82:        stw     r10,24(r3)
                     83:        blr
                     84:
                     85: L(end2):
                     86:        mullw   r8,r7,r7
                     87:        stw     r9,4(r3)
                     88:        mulhwu  r10,r7,r7
                     89:        stw     r11,8(r3)
                     90:        stw     r8,12(r3)
                     91:        stw     r10,16(r3)
                     92:        blr
                     93: L(end1):
                     94:        mullw   r9,r6,r6
                     95:        mulhwu  r11,r6,r6
                     96:        stw     r9,4(r3)
                     97:        stw     r11,8(r3)
                     98:        blr
                     99: EPILOGUE(mpn_sqr_diagonal)