|
Packit |
5c3484 |
dnl ARM64 mpn_copyd.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 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 Cortex-A53 ?
|
|
Packit |
5c3484 |
C Cortex-A57 ?
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
changecom(@&*$)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`rp', `x0')
|
|
Packit |
5c3484 |
define(`up', `x1')
|
|
Packit |
5c3484 |
define(`n', `x2')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
PROLOGUE(mpn_copyd)
|
|
Packit |
5c3484 |
add rp, rp, n, lsl #3
|
|
Packit |
5c3484 |
add up, up, n, lsl #3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cmp n, #3
|
|
Packit |
5c3484 |
b.le L(bc)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C Copy until rp is 128-bit aligned
|
|
Packit |
5c3484 |
tbz rp, #3, L(al2)
|
|
Packit |
5c3484 |
sub up, up, #8
|
|
Packit |
5c3484 |
ld1 {v22.1d}, [up]
|
|
Packit |
5c3484 |
sub n, n, #1
|
|
Packit |
5c3484 |
sub rp, rp, #8
|
|
Packit |
5c3484 |
st1 {v22.1d}, [rp]
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(al2): sub up, up, #16
|
|
Packit |
5c3484 |
ld1 {v26.2d}, [up]
|
|
Packit |
5c3484 |
subs n, n, #6
|
|
Packit |
5c3484 |
sub rp, rp, #16 C offset rp for loop
|
|
Packit |
5c3484 |
b.lt L(end)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
sub up, up, #16 C offset up for loop
|
|
Packit |
5c3484 |
mov x12, #-16
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(top): ld1 {v22.2d}, [up], x12
|
|
Packit |
5c3484 |
st1 {v26.2d}, [rp], x12
|
|
Packit |
5c3484 |
ld1 {v26.2d}, [up], x12
|
|
Packit |
5c3484 |
st1 {v22.2d}, [rp], x12
|
|
Packit |
5c3484 |
subs n, n, #4
|
|
Packit |
5c3484 |
b.ge L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
add up, up, #16 C undo up offset
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): st1 {v26.2d}, [rp]
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C Copy last 0-3 limbs. Note that rp is aligned after loop, but not when we
|
|
Packit |
5c3484 |
C arrive here via L(bc)
|
|
Packit |
5c3484 |
L(bc): tbz n, #1, L(tl1)
|
|
Packit |
5c3484 |
sub up, up, #16
|
|
Packit |
5c3484 |
ld1 {v22.2d}, [up]
|
|
Packit |
5c3484 |
sub rp, rp, #16
|
|
Packit |
5c3484 |
st1 {v22.2d}, [rp]
|
|
Packit |
5c3484 |
L(tl1): tbz n, #0, L(tl2)
|
|
Packit |
5c3484 |
sub up, up, #8
|
|
Packit |
5c3484 |
ld1 {v22.1d}, [up]
|
|
Packit |
5c3484 |
sub rp, rp, #8
|
|
Packit |
5c3484 |
st1 {v22.1d}, [rp]
|
|
Packit |
5c3484 |
L(tl2): ret
|
|
Packit |
5c3484 |
EPILOGUE()
|