|
Packit |
5c3484 |
dnl SPARC v9 mpn_sub_n for T3/T4.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by David Miller.
|
|
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 UltraSPARC T3: 8
|
|
Packit |
5c3484 |
C UltraSPARC T4: 3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`rp', `%i0')
|
|
Packit |
5c3484 |
define(`up', `%i1')
|
|
Packit |
5c3484 |
define(`vp', `%i2')
|
|
Packit |
5c3484 |
define(`n', `%i3')
|
|
Packit |
5c3484 |
define(`cy', `%i4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`u0_off', `%l0')
|
|
Packit |
5c3484 |
define(`u1_off', `%l1')
|
|
Packit |
5c3484 |
define(`v0_off', `%l2')
|
|
Packit |
5c3484 |
define(`v1_off', `%l3')
|
|
Packit |
5c3484 |
define(`r0_off', `%l4')
|
|
Packit |
5c3484 |
define(`r1_off', `%l5')
|
|
Packit |
5c3484 |
define(`loop_n', `%l6')
|
|
Packit |
5c3484 |
define(`tmp', `%l7')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
REGISTER(%g2,#scratch)
|
|
Packit |
5c3484 |
REGISTER(%g3,#scratch)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_sub_nc)
|
|
Packit |
5c3484 |
save %sp, -176, %sp
|
|
Packit |
5c3484 |
ba,pt %xcc, L(ent)
|
|
Packit |
5c3484 |
xor cy, 1, cy
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
Packit |
5c3484 |
PROLOGUE(mpn_sub_n)
|
|
Packit |
5c3484 |
save %sp, -176, %sp
|
|
Packit |
5c3484 |
mov 1, cy
|
|
Packit |
5c3484 |
L(ent):
|
|
Packit |
5c3484 |
subcc n, 1, n
|
|
Packit |
5c3484 |
be L(final_one)
|
|
Packit |
5c3484 |
cmp %g0, cy
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ldx [up + 0], %o4
|
|
Packit |
5c3484 |
sllx n, 3, tmp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ldx [vp + 0], %o5
|
|
Packit |
5c3484 |
add up, tmp, u0_off
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ldx [up + 8], %g5
|
|
Packit |
5c3484 |
add vp, tmp, v0_off
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ldx [vp + 8], %g1
|
|
Packit |
5c3484 |
add rp, tmp, r0_off
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
neg tmp, loop_n
|
|
Packit |
5c3484 |
add u0_off, 8, u1_off
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
add v0_off, 8, v1_off
|
|
Packit |
5c3484 |
sub loop_n, -(2 * 8), loop_n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
sub r0_off, 16, r0_off
|
|
Packit |
5c3484 |
brgez,pn loop_n, L(loop_tail)
|
|
Packit |
5c3484 |
sub r0_off, 8, r1_off
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
b,a L(top)
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(top):
|
|
Packit |
5c3484 |
xnor %o5, 0, tmp
|
|
Packit |
5c3484 |
ldx [loop_n + v0_off], %o5
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addxccc(%o4, tmp, %g3)
|
|
Packit |
5c3484 |
ldx [loop_n + u0_off], %o4
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
xnor %g1, 0, %g1
|
|
Packit |
5c3484 |
stx %g3, [loop_n + r0_off]
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addxccc(%g5, %g1, tmp)
|
|
Packit |
5c3484 |
ldx [loop_n + v1_off], %g1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ldx [loop_n + u1_off], %g5
|
|
Packit |
5c3484 |
sub loop_n, -(2 * 8), loop_n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
brlz loop_n, L(top)
|
|
Packit |
5c3484 |
stx tmp, [loop_n + r1_off]
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(loop_tail):
|
|
Packit |
5c3484 |
xnor %o5, 0, tmp
|
|
Packit |
5c3484 |
xnor %g1, 0, %g1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addxccc(%o4, tmp, %g3)
|
|
Packit |
5c3484 |
add loop_n, u0_off, up
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addxccc(%g5, %g1, %g5)
|
|
Packit |
5c3484 |
add loop_n, r0_off, rp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
stx %g3, [rp + 0]
|
|
Packit |
5c3484 |
add loop_n, v0_off, vp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
brgz,pt loop_n, L(done)
|
|
Packit |
5c3484 |
stx %g5, [rp + 8]
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
add rp, (2 * 8), rp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(final_one):
|
|
Packit |
5c3484 |
ldx [up+0], %o4
|
|
Packit |
5c3484 |
ldx [vp+0], %o5
|
|
Packit |
5c3484 |
xnor %o5, %g0, %o5
|
|
Packit |
5c3484 |
addxccc(%o4, %o5, %g3)
|
|
Packit |
5c3484 |
stx %g3, [rp+0]
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(done):
|
|
Packit |
5c3484 |
clr %i0
|
|
Packit |
5c3484 |
movcc %xcc, 1, %i0
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
restore
|
|
Packit |
5c3484 |
EPILOGUE()
|