|
Packit |
5c3484 |
dnl AMD64 mpn_sqr_diag_addlsh1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by Torbjörn Granlund.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2011-2013 Free Software Foundation, Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl This file is part of the GNU MP Library.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is free software; you can redistribute it and/or modify
|
|
Packit |
5c3484 |
dnl it under the terms of either:
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU Lesser General Public License as published by the Free
|
|
Packit |
5c3484 |
dnl Software Foundation; either version 3 of the License, or (at your
|
|
Packit |
5c3484 |
dnl option) any later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU General Public License as published by the Free Software
|
|
Packit |
5c3484 |
dnl Foundation; either version 2 of the License, or (at your option) any
|
|
Packit |
5c3484 |
dnl later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or both in parallel, as here.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is distributed in the hope that it will be useful, but
|
|
Packit |
5c3484 |
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
Packit |
5c3484 |
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
Packit |
5c3484 |
dnl for more details.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl You should have received copies of the GNU General Public License and the
|
|
Packit |
5c3484 |
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
|
|
Packit |
5c3484 |
dnl see https://www.gnu.org/licenses/.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
include(`../config.m4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C cycles/limb
|
|
Packit |
5c3484 |
C AMD K8,K9 2.5
|
|
Packit |
5c3484 |
C AMD K10 2.5
|
|
Packit |
5c3484 |
C AMD bull 3.6
|
|
Packit |
5c3484 |
C AMD pile 3.6
|
|
Packit |
5c3484 |
C AMD steam ?
|
|
Packit |
5c3484 |
C AMD bobcat 4
|
|
Packit |
5c3484 |
C AMD jaguar ?
|
|
Packit |
5c3484 |
C Intel P4 ?
|
|
Packit |
5c3484 |
C Intel core 4
|
|
Packit |
5c3484 |
C Intel NHM 3.6
|
|
Packit |
5c3484 |
C Intel SBR 3.15
|
|
Packit |
5c3484 |
C Intel IBR 3.2
|
|
Packit |
5c3484 |
C Intel HWL 2.6
|
|
Packit |
5c3484 |
C Intel BWL ?
|
|
Packit |
5c3484 |
C Intel atom 14
|
|
Packit |
5c3484 |
C VIA nano 3.5
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C When playing with pointers, set this to $2 to fall back to conservative
|
|
Packit |
5c3484 |
C indexing in wind-down code.
|
|
Packit |
5c3484 |
define(`I',`$1')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`rp', `%rdi')
|
|
Packit |
5c3484 |
define(`tp', `%rsi')
|
|
Packit |
5c3484 |
define(`up_arg', `%rdx')
|
|
Packit |
5c3484 |
define(`n', `%rcx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`up', `%r11')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ABI_SUPPORT(DOS64)
|
|
Packit |
5c3484 |
ABI_SUPPORT(STD64)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
TEXT
|
|
Packit |
5c3484 |
ALIGN(32)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_sqr_diag_addlsh1)
|
|
Packit |
5c3484 |
FUNC_ENTRY(4)
|
|
Packit |
5c3484 |
push %rbx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dec n
|
|
Packit |
5c3484 |
shl n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mov (up_arg), %rax
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
lea (rp,n,8), rp
|
|
Packit |
5c3484 |
lea (tp,n,8), tp
|
|
Packit |
5c3484 |
lea (up_arg,n,4), up
|
|
Packit |
5c3484 |
neg n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mul %rax
|
|
Packit |
5c3484 |
mov %rax, (rp,n,8)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
xor R32(%rbx), R32(%rbx)
|
|
Packit |
5c3484 |
jmp L(mid)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(top): add %r10, %r8
|
|
Packit |
5c3484 |
adc %rax, %r9
|
|
Packit |
5c3484 |
mov %r8, -8(rp,n,8)
|
|
Packit |
5c3484 |
mov %r9, (rp,n,8)
|
|
Packit |
5c3484 |
L(mid): mov 8(up,n,4), %rax
|
|
Packit |
5c3484 |
mov (tp,n,8), %r8
|
|
Packit |
5c3484 |
mov 8(tp,n,8), %r9
|
|
Packit |
5c3484 |
adc %r8, %r8
|
|
Packit |
5c3484 |
adc %r9, %r9
|
|
Packit |
5c3484 |
lea (%rdx,%rbx), %r10
|
|
Packit |
5c3484 |
setc R8(%rbx)
|
|
Packit |
5c3484 |
mul %rax
|
|
Packit |
5c3484 |
add $2, n
|
|
Packit |
5c3484 |
js L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): add %r10, %r8
|
|
Packit |
5c3484 |
adc %rax, %r9
|
|
Packit |
5c3484 |
mov %r8, I(-8(rp),-8(rp,n,8))
|
|
Packit |
5c3484 |
mov %r9, I((rp),(rp,n,8))
|
|
Packit |
5c3484 |
adc %rbx, %rdx
|
|
Packit |
5c3484 |
mov %rdx, I(8(rp),8(rp,n,8))
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
pop %rbx
|
|
Packit |
5c3484 |
FUNC_EXIT()
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
rpm-build |
c3cd4f |
CF_PROT
|