Return to copyi.asm CVS log

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

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

1.1       maekawa     1: dnl  x86 mpn_copyi -- copy limb vector, incrementing.
                      2:
1.1.1.2 ! ohara       3: dnl  Copyright 1999, 2000, 2001, 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     cycles/limb  startup (approx)
        !            26: C P5:     1.0         35
        !            27: C P6      0.75        45
        !            28: C K6      1.0         30
        !            29: C K7:     1.3         65
        !            30: C P4:     1.0        120
        !            31: C
        !            32: C (Startup time includes some function call overheads.)
        !            33:
        !            34:
1.1       maekawa    35: C void mpn_copyi (mp_ptr dst, mp_srcptr src, mp_size_t size);
                     36: C
                     37: C Copy src,size to dst,size, working from low to high addresses.
                     38: C
                     39: C The code here is very generic and can be expected to be reasonable on all
                     40: C the x86 family.
                     41: C
1.1.1.2 ! ohara      42: C P6 -  An MMX based copy was tried, but was found to be slower than a rep
        !            43: C       movs in all cases.  The fastest MMX found was 0.8 cycles/limb (when
        !            44: C       fully aligned).  A rep movs seems to have a startup time of about 15
        !            45: C       cycles, but doing something special for small sizes could lead to a
        !            46: C       branch misprediction that would destroy any saving.  For now a plain
        !            47: C       rep movs seems ok.
1.1       maekawa    48: C
1.1.1.2 ! ohara      49: C K62 - We used to have a big chunk of code doing an MMX copy at 0.56 c/l if
        !            50: C       aligned or a 1.0 rep movs if not.  But that seemed excessive since
        !            51: C       it only got an advantage half the time, and even then only showed it
        !            52: C       above 50 limbs or so.
1.1       maekawa    53:
                     54: defframe(PARAM_SIZE,12)
                     55: defframe(PARAM_SRC, 8)
                     56: defframe(PARAM_DST, 4)
                     57: deflit(`FRAME',0)
                     58:
1.1.1.2 ! ohara      59:        TEXT
1.1       maekawa    60:        ALIGN(32)
                     61:
                     62:        C eax   saved esi
                     63:        C ebx
                     64:        C ecx   counter
                     65:        C edx   saved edi
                     66:        C esi   src
                     67:        C edi   dst
                     68:        C ebp
                     69:
                     70: PROLOGUE(mpn_copyi)
                     71:
                     72:        movl    PARAM_SIZE, %ecx
                     73:        movl    %esi, %eax
                     74:
                     75:        movl    PARAM_SRC, %esi
                     76:        movl    %edi, %edx
                     77:
                     78:        movl    PARAM_DST, %edi
                     79:
1.1.1.2 ! ohara      80:        cld     C better safe than sorry, see mpn/x86/README
1.1       maekawa    81:
                     82:        rep
                     83:        movsl
                     84:
                     85:        movl    %eax, %esi
                     86:        movl    %edx, %edi
                     87:
                     88:        ret
                     89:
                     90: EPILOGUE()