Blame mpn/arm64/mul_1.asm
|
Packit |
5c3484 |
dnl ARM64 mpn_mul_1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by Torbjörn Granlund.
|
|
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 |
define(`rp', `x0')
|
|
Packit |
5c3484 |
define(`up', `x1')
|
|
Packit |
5c3484 |
define(`n', `x2')
|
|
Packit |
5c3484 |
define(`v0', `x3')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
PROLOGUE(mpn_mul_1)
|
|
Packit |
5c3484 |
ldr x12, [up], #8
|
|
Packit |
5c3484 |
and x6, n, #3
|
|
Packit |
5c3484 |
and n, n, #-4
|
|
Packit |
5c3484 |
cbz x6, L(fi0)
|
|
Packit |
5c3484 |
cmp x6, #2
|
|
Packit |
5c3484 |
b.cc L(fi1)
|
|
Packit |
5c3484 |
b.eq L(fi2)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(fi3): mul x8, x12, v0
|
|
Packit |
5c3484 |
umulh x13, x12, v0
|
|
Packit |
5c3484 |
cmn xzr, xzr
|
|
Packit |
5c3484 |
b L(L3)
|
|
Packit |
5c3484 |
L(fi2): mul x7, x12, v0
|
|
Packit |
5c3484 |
umulh x5, x12, v0
|
|
Packit |
5c3484 |
cmn xzr, xzr
|
|
Packit |
5c3484 |
b L(L2)
|
|
Packit |
5c3484 |
L(fi0): mul x9, x12, v0
|
|
Packit |
5c3484 |
umulh x5, x12, v0
|
|
Packit |
5c3484 |
sub n, n, #4
|
|
Packit |
5c3484 |
cmn xzr, xzr
|
|
Packit |
5c3484 |
b L(L0)
|
|
Packit |
5c3484 |
L(fi1): mul x10, x12, v0
|
|
Packit |
5c3484 |
umulh x13, x12, v0
|
|
Packit |
5c3484 |
cmn xzr, xzr
|
|
Packit |
5c3484 |
cbz n, L(end)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(top): sub n, n, #4
|
|
Packit |
5c3484 |
ldr x12, [up], #8
|
|
Packit |
5c3484 |
mul x6, x12, v0
|
|
Packit |
5c3484 |
umulh x5, x12, v0
|
|
Packit |
5c3484 |
str x10, [rp], #8
|
|
Packit |
5c3484 |
adcs x9, x6, x13
|
|
Packit |
5c3484 |
L(L0): ldr x12, [up], #8
|
|
Packit |
5c3484 |
mul x6, x12, v0
|
|
Packit |
5c3484 |
umulh x13, x12, v0
|
|
Packit |
5c3484 |
str x9, [rp] ,#8
|
|
Packit |
5c3484 |
adcs x8, x6, x5
|
|
Packit |
5c3484 |
L(L3): ldr x12, [up], #8
|
|
Packit |
5c3484 |
mul x6, x12, v0
|
|
Packit |
5c3484 |
umulh x5, x12, v0
|
|
Packit |
5c3484 |
str x8, [rp], #8
|
|
Packit |
5c3484 |
adcs x7, x6, x13
|
|
Packit |
5c3484 |
L(L2): ldr x12, [up], #8
|
|
Packit |
5c3484 |
mul x6, x12, v0
|
|
Packit |
5c3484 |
umulh x13, x12, v0
|
|
Packit |
5c3484 |
str x7, [rp], #8
|
|
Packit |
5c3484 |
adcs x10, x6, x5
|
|
Packit |
5c3484 |
cbnz n, L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): str x10, [rp]
|
|
Packit |
5c3484 |
adc x0, x13, xzr
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
EPILOGUE()
|