dnl AMD64 mpn_com optimised for CPUs with fast SSE copying and SSSE3.
dnl Copyright 2012, 2013, 2015 Free Software Foundation, Inc.
dnl Contributed to the GNU project by Torbjorn Granlund.
dnl This file is part of the GNU MP Library.
dnl
dnl The GNU MP Library is free software; you can redistribute it and/or modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
C cycles/limb cycles/limb cycles/limb good
C aligned unaligned best seen forcpu?
C AMD K8,K9 2.0 illop 1.0/1.0 N
C AMD K10 0.85 illop Y/N
C AMD bd1 1.39 ? 1.45 Y/N
C AMD bd2 0.8-1.40.7-1.4 Y
C AMD bd3
C AMD bd4
C AMD bobcat 1.97 ? 8.171.5/1.5 N
C AMD jaguar 1.021.020.91/0.91 N
C Intel P4 2.26 illop Y/N
C Intel core 0.580.87 opt/0.74 Y
C Intel NHM 0.641.14 opt/bad Y
C Intel SBR 0.510.65 opt/opt Y
C Intel IBR 0.500.64 opt/0.57 Y
C Intel HWL 0.510.58 opt/opt Y
C Intel BWL 0.520.64 opt/opt Y
C Intel SKL 0.510.63 opt/opt Y
C Intel atom 1.161.70 opt/opt Y
C Intel SLM 1.021.52 N
C VIA nano 1.091.10 opt/opt Y
C We use only 16-byte operations, except for unaligned top-most and bottom-most
C limbs. We use the SSSE3 palignr instruction when rp - up = 8 (mod16). That
C instruction is better adapted to mpn_copyd's needs, we need to contort the
C code to use it here.
C
C For operands of < COM_SSE_THRESHOLD limbs, we use a plain 64-bit loop, taken
C from the x86_64 default code.
C INPUT PARAMETERS
define(`rp', `%rdi')
define(`up', `%rsi')
define(`n', `%rdx')
C There are three instructions for loading an aligned 128-bit quantity. We use
C movaps, since it has the shortest coding.
define(`movdqa', ``movaps'')
test $4, R8(n) jz1f
movdqa (up), %xmm0
movdqa 16(up), %xmm1 lea32(up), up
pxor %xmm5, %xmm0
pxor %xmm5, %xmm1
movdqa %xmm0, (rp)
movdqa %xmm1, 16(rp) lea32(rp), rp
1: test $2, R8(n) jz1f
movdqa (up), %xmm0 lea16(up), up
pxor %xmm5, %xmm0
movdqa %xmm0, (rp) lea16(rp), rp
1: test $1, R8(n) jz1f mov (up), %r8
not %r8 mov %r8, (rp)
1: FUNC_EXIT() ret
L(uent):
C Code handling up - rp = 8 (mod16)
C FIXME: The code below only handles overlap if it is close to complete, or
C quite separate: up-rp < 5or up-up > 15 limbs lea -40(up), %rax C 40 = 5 * GMP_LIMB_BYTES sub rp, %rax cmp $80, %rax C 80 = (15-5) * GMP_LIMB_BYTES
jbe L(bc) C deflect to plain loop
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.