dnl Contributed to the GNU project by Torbjorn Granlund and Marco Bodrato.
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 This is an attempt at an addlsh1_n for x86-32, not relying on sse2 insns.
C The innerloop is 2*3-way unrolled, which is best we can do with the available
C registers. It seems tricky to use the same structure for rsblsh1_n, since we
C cannot feed carry between operations there.
C cycles/limb
C P5
C P6 model0-8,10-12
C P6 model9 (Banias)
C P6 model13 (Dothan) 5.4 (worse than add_n + lshift)
C P4 model0 (Willamette)
C P4 model1 (?)
C P4 model2 (Northwood)
C P4 model3 (Prescott)
C P4 model4 (Nocona)
C Intel Atom 6
C AMD K6 ?
C AMD K7 2.5
C AMD K8
C This is a basic addlsh1_n for k7, atom, and perhaps some other x86-32
C processors. It uses 2*3-way unrolling, for good reasons. Unfortunately,
C that means we need an initial magic multiply.
C
C It is not clear how to do sublsh1_n or rsblsh1_n using the same pattern. We
C cannot do rsblsh1_n since we feed carry from the shift blocks to the
C add/subtract blocks, which is right for addition but reversed for
C subtraction. We could perhaps do sublsh1_n, with some extra move insns,
C without losing any time, since we're not issue limited but carry recurrency
C latency.
C
C Breaking carry recurrency might be a good idea. We would then need separate
C registers for the shift carryandadd/subtract carry, which in turn would
C force us to 2*2-way unrolling.
push %ebx FRAME_pushl() mov PARAM_SIZE, %ebx C size
push rp FRAME_pushl() mov PARAM_DST, rp
mul %ebx
push up FRAME_pushl() mov PARAM_SRC, up
not %edx C count = -(size\8)-1 mov %edx, VAR_COUNT
push vp FRAME_pushl() mov PARAM_DBLD, vp
lea3(%edx,%edx,2), %ecx C count*3+3 = -(size\6)*3 xor %edx, %edx lea (%ebx,%ecx,2), %ebx C size + (count*3+3)*2 = size % 6 or %ebx, %ebx jz L(exact)
L(oop): ifdef(`CPU_P6',` shr %edx') C restore 2nd saved carry bit mov (vp), %eax adc %eax, %eax
rcr %edx C restore 1st saved carry bit lea4(vp), vp adc (up), %eax lea4(up), up adc %edx, %edx C save a carry bit in edx ifdef(`CPU_P6',` adc %edx, %edx') C save another carry bit in edx dec %ebx mov %eax, (rp) lea4(rp), rp jnz L(oop) mov vp, VAR_TMP
L(exact):
incl VAR_COUNT jz L(end)
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.