dnl Pentium-4 mpn_copyi -- copy limb vector, incrementing.
dnl Copyright 1999, 2000, 2001 Free Software Foundation, Inc.
dnl This file is part of the GNU MP Library.
dnl The GNU MP Library is free software; you can redistribute it and/or
dnl modify it under the terms of the GNU Lesser General Public License as
dnl published by the Free Software Foundation; either version 2.1 of the
dnl License, or (at your option) any later version.
dnl The GNU MP Library is distributed in the hope that it will be useful,
dnl but WITHOUT ANY WARRANTY; without even the implied warranty of
dnl MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
dnl Lesser General Public License for more details.
dnl You should have received a copy of the GNU Lesser General Public
dnl License along with the GNU MP Library; see the file COPYING.LIB. If
dnl not, write to the Free Software Foundation, Inc., 59 Temple Place -
dnl Suite 330, Boston, MA 02111-1307, USA.
dnl The rep/movsl is very slow for small blocks on pentium4. Its startup
dnl time seems to be about 110 cycles. It then copies at a rate of one
dnl limb per cycle. We therefore fall back to an open-coded 2 c/l copying
dnl loop for smaller sizes.
dnl Ultimately, we may want to use 64-bit movd or 128-bit movdqu in some
dnl nifty unrolled arrangement. Clearly, that could reach much higher
dnl speeds, at least for large blocks.
movl PARAM_SIZE, %ecx
cmpl $150, %ecx
movl PARAM_SRC, %eax
movl PARAM_DST, %edx
movl %ebx, PARAM_SIZE
testl %ecx, %ecx
movl (%eax), %ebx
leal 4(%eax), %eax
addl $-1, %ecx
movl %ebx, (%edx)
leal 4(%edx), %edx
movl PARAM_SIZE, %ebx
cld C better safe than sorry, see mpn/x86/README
movl %esi, %eax
movl PARAM_SRC, %esi
movl %edi, %edx
movl PARAM_DST, %edi
movl %eax, %esi
movl %edx, %edi