1 dnl Intel Pentium-4 mpn_rsh1add_n -- mpn (x+y)/2
3 dnl Copyright 2001, 2002, 2003, 2004 Free Software Foundation, Inc.
5 dnl This file is part of the GNU MP Library.
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 3 of the
10 dnl License, or (at your option) any later version.
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.
17 dnl You should have received a copy of the GNU Lesser General Public License
18 dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/.
20 include(`../config.m4')
23 C cycles/limb (approx)
24 C dst!=src1,2 dst==src1 dst==src2
28 C mp_limb_t mpn_rsh1add_n (mp_ptr wp, mp_srcptr xp, mp_srcptr yp,
31 C The slightly strange combination of indexing and pointer incrementing
32 C that's used seems to work best. Not sure why, but for instance leal
33 C incrementing on %esi is a 1 or 2 cycle slowdown.
35 C The dependent chain is paddq combining the carry and next (shifted) part,
36 C plus psrlq to move the new carry down. That, and just 4 mmx instructions
37 C in total, makes 4 c/l the target speed, which is almost achieved for
38 C separate src/dst but when src==dst the write combining anomalies slow it
41 defframe(PARAM_SIZE, 16)
42 defframe(PARAM_YP, 12)
46 dnl re-use parameter space
47 define(SAVE_EBX,`PARAM_XP')
48 define(SAVE_ESI,`PARAM_YP')
53 PROLOGUE(mpn_rsh1add_n)
64 movd (%edx), %mm0 C xp[0]
66 movd (%ebx), %mm1 C yp[0]
69 movl (%edx), %eax C xp[0]
71 addl (%ebx), %eax C xp[0]+yp[0]
73 paddq %mm1, %mm0 C xp[0]+yp[0]
74 leal (%esi,%ecx,4), %esi C wp end
77 psrlq $1, %mm0 C (xp[0]+yp[0])/2
78 and $1, %eax C return value, rsh1 bit of xp[0]+yp[0]
79 addl $1, %ecx C -(size-1)
86 C ecx counter, limbs, -(size-1) to -1 inclusive
91 movd 4(%edx), %mm1 C xp[i+1]
92 movd 4(%ebx), %mm2 C yp[i+1]
95 paddq %mm2, %mm1 C xp[i+1]+yp[i+1]
96 psllq $31, %mm1 C low bit at 31, further 32 above
98 paddq %mm1, %mm0 C 31 and carry from prev add
99 movd %mm0, -4(%esi,%ecx,4) C low ready to store dst[i]
101 psrlq $32, %mm0 C high becomes new carry
108 movd %mm0, -4(%esi) C dst[size-1]