|
Packit |
5c3484 |
dnl ARM64 Neon mpn_sec_tabselect.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by Torbjörn Granlund.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2011-2014 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 |
|
|
Packit |
5c3484 |
C cycles/limb
|
|
Packit |
5c3484 |
C Cortex-A53 ?
|
|
Packit |
5c3484 |
C Cortex-A57 ?
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C void
|
|
Packit |
5c3484 |
C mpn_sec_tabselect (mp_ptr rp, mp_srcptr *tab,
|
|
Packit |
5c3484 |
C mp_size_t n, mp_size_t nents, mp_size_t which)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
changecom(@&*$)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`rp', `x0')
|
|
Packit |
5c3484 |
define(`tp', `x1')
|
|
Packit |
5c3484 |
define(`n', `x2')
|
|
Packit |
5c3484 |
define(`nents', `x3')
|
|
Packit |
5c3484 |
define(`which', `x4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`i', `x5')
|
|
Packit |
5c3484 |
define(`j', `x6')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`maskq', `v4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
PROLOGUE(mpn_sec_tabselect)
|
|
Packit |
5c3484 |
dup v7.2d, x4 C 2 `which' copies
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mov x10, #1
|
|
Packit |
5c3484 |
dup v6.2d, x10 C 2 copies of 1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
subs j, n, #4
|
|
Packit |
5c3484 |
b.mi L(outer_end)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(outer_top):
|
|
Packit |
5c3484 |
mov i, nents
|
|
Packit |
5c3484 |
mov x12, tp C preserve tp
|
|
Packit |
5c3484 |
movi v5.16b, #0 C zero 2 counter copies
|
|
Packit |
5c3484 |
movi v2.16b, #0
|
|
Packit |
5c3484 |
movi v3.16b, #0
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(tp4): cmeq maskq.2d, v5.2d, v7.2d C compare idx copies to `which' copies
|
|
Packit |
5c3484 |
ld1 {v0.2d,v1.2d}, [tp]
|
|
Packit |
5c3484 |
add v5.2d, v5.2d, v6.2d
|
|
Packit |
5c3484 |
bit v2.16b, v0.16b, maskq.16b
|
|
Packit |
5c3484 |
bit v3.16b, v1.16b, maskq.16b
|
|
Packit |
5c3484 |
add tp, tp, n, lsl #3
|
|
Packit |
5c3484 |
sub i, i, #1
|
|
Packit |
5c3484 |
cbnz i, L(tp4)
|
|
Packit |
5c3484 |
st1 {v2.2d,v3.2d}, [rp], #32
|
|
Packit |
5c3484 |
add tp, x12, #32 C restore tp, point to next slice
|
|
Packit |
5c3484 |
subs j, j, #4
|
|
Packit |
5c3484 |
b.pl L(outer_top)
|
|
Packit |
5c3484 |
L(outer_end):
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
tbz n, #1, L(b0x)
|
|
Packit |
5c3484 |
mov i, nents
|
|
Packit |
5c3484 |
mov x12, tp
|
|
Packit |
5c3484 |
movi v5.16b, #0 C zero 2 counter copies
|
|
Packit |
5c3484 |
movi v2.16b, #0
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(tp2): cmeq maskq.2d, v5.2d, v7.2d
|
|
Packit |
5c3484 |
ld1 {v0.2d}, [tp]
|
|
Packit |
5c3484 |
add v5.2d, v5.2d, v6.2d
|
|
Packit |
5c3484 |
bit v2.16b, v0.16b, maskq.16b
|
|
Packit |
5c3484 |
add tp, tp, n, lsl #3
|
|
Packit |
5c3484 |
sub i, i, #1
|
|
Packit |
5c3484 |
cbnz i, L(tp2)
|
|
Packit |
5c3484 |
st1 {v2.2d}, [rp], #16
|
|
Packit |
5c3484 |
add tp, x12, #16
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b0x): tbz n, #0, L(b00)
|
|
Packit |
5c3484 |
mov i, nents
|
|
Packit |
5c3484 |
mov x12, tp
|
|
Packit |
5c3484 |
movi v5.16b, #0 C zero 2 counter copies
|
|
Packit |
5c3484 |
movi v2.16b, #0
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(tp1): cmeq maskq.2d, v5.2d, v7.2d
|
|
Packit |
5c3484 |
ld1 {v0.1d}, [tp]
|
|
Packit |
5c3484 |
add v5.2d, v5.2d, v6.2d C FIXME size should be `1d'
|
|
Packit |
5c3484 |
bit v2.8b, v0.8b, maskq.8b
|
|
Packit |
5c3484 |
add tp, tp, n, lsl #3
|
|
Packit |
5c3484 |
sub i, i, #1
|
|
Packit |
5c3484 |
cbnz i, L(tp1)
|
|
Packit |
5c3484 |
st1 {v2.1d}, [rp], #8
|
|
Packit |
5c3484 |
add tp, x12, #8
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b00): ret
|
|
Packit |
5c3484 |
EPILOGUE()
|