1 dnl AMD K7 mpn_copyd -- copy limb vector, decrementing.
3 dnl alignment dst/src, A=0mod8 N=4mod8
5 dnl K7 0.75 1.0 1.0 0.75
8 dnl Copyright (C) 1999, 2000 Free Software Foundation, Inc.
10 dnl This file is part of the GNU MP Library.
12 dnl The GNU MP Library is free software; you can redistribute it and/or
13 dnl modify it under the terms of the GNU Lesser General Public License as
14 dnl published by the Free Software Foundation; either version 2.1 of the
15 dnl License, or (at your option) any later version.
17 dnl The GNU MP Library is distributed in the hope that it will be useful,
18 dnl but WITHOUT ANY WARRANTY; without even the implied warranty of
19 dnl MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 dnl Lesser General Public License for more details.
22 dnl You should have received a copy of the GNU Lesser General Public
23 dnl License along with the GNU MP Library; see the file COPYING.LIB. If
24 dnl not, write to the Free Software Foundation, Inc., 59 Temple Place -
25 dnl Suite 330, Boston, MA 02111-1307, USA.
28 include(`../config.m4')
31 C void mpn_copyd (mp_ptr dst, mp_srcptr src, mp_size_t size);
33 C The various comments in mpn/x86/k7/copyi.asm apply here too.
35 defframe(PARAM_SIZE,12)
36 defframe(PARAM_SRC, 8)
37 defframe(PARAM_DST, 4)
40 dnl parameter space reused
41 define(SAVE_EBX,`PARAM_SIZE')
42 define(SAVE_ESI,`PARAM_SRC')
44 dnl minimum 5 since the unrolled code can't handle less than 5
45 deflit(UNROLL_THRESHOLD, 5)
57 cmpl $UNROLL_THRESHOLD, %ecx
69 C this loop is 2 cycles/limb
71 movl -4(%eax,%ecx,4), %ebx
72 movl %ebx, -4(%edx,%ecx,4)
83 leal (%eax,%ecx,4), %ebx
84 leal (%edx,%ecx,4), %esi
88 subl $4, %ecx C size-4
90 testl $4, %ebx C testl to pad code closer to 16 bytes for L(top)
93 C both src and dst unaligned, process one limb to align them
94 movl 12(%eax,%ecx,4), %ebx
95 movl %ebx, 12(%edx,%ecx,4)
107 movq 8(%eax,%ecx,4), %mm0
108 movq (%eax,%ecx,4), %mm1
110 movq %mm0, 16+8(%edx,%ecx,4)
111 movq %mm1, 16(%edx,%ecx,4)
115 C now %ecx is -4 to -1 representing respectively 0 to 3 limbs remaining
120 movq 8(%eax,%ecx,4), %mm0
121 movq %mm0, 8(%edx,%ecx,4)