|
Packit |
5c3484 |
dnl SPARC T3/T4/T5 mpn_bdiv_dbm1c.
|
|
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 UltraSPARC T3: 25
|
|
Packit |
5c3484 |
C UltraSPARC T4/T5: 4
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`qp', `%i0')
|
|
Packit |
5c3484 |
define(`ap', `%i1')
|
|
Packit |
5c3484 |
define(`n', `%i2')
|
|
Packit |
5c3484 |
define(`bd', `%i3')
|
|
Packit |
5c3484 |
define(`h', `%i4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`plo0',`%g4') define(`plo1',`%g5')
|
|
Packit |
5c3484 |
define(`phi0',`%l0') define(`phi1',`%l1')
|
|
Packit |
5c3484 |
define(`a0', `%g1') define(`a1', `%g3')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
REGISTER(%g2,#scratch)
|
|
Packit |
5c3484 |
REGISTER(%g3,#scratch)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_bdiv_dbm1c)
|
|
Packit |
5c3484 |
save %sp, -176, %sp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
and n, 3, %g5
|
|
Packit |
5c3484 |
ldx [ap + 0], %g2
|
|
Packit |
5c3484 |
add n, -5, n
|
|
Packit |
5c3484 |
brz %g5, L(b0)
|
|
Packit |
5c3484 |
cmp %g5, 2
|
|
Packit |
5c3484 |
bcs %xcc, L(b1)
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
be %xcc, L(b2)
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b3): ldx [ap + 8], a0
|
|
Packit |
5c3484 |
mulx bd, %g2, plo1
|
|
Packit |
5c3484 |
umulxhi(bd, %g2, phi1)
|
|
Packit |
5c3484 |
ldx [ap + 16], a1
|
|
Packit |
5c3484 |
add qp, -24, qp
|
|
Packit |
5c3484 |
b L(lo3)
|
|
Packit |
5c3484 |
add ap, -8, ap
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b2): ldx [ap + 8], a1
|
|
Packit |
5c3484 |
mulx bd, %g2, plo0
|
|
Packit |
5c3484 |
umulxhi(bd, %g2, phi0)
|
|
Packit |
5c3484 |
brlz,pt n, L(wd2)
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
L(gt2): ldx [ap + 16], a0
|
|
Packit |
5c3484 |
add ap, 16, ap
|
|
Packit |
5c3484 |
b L(lo2)
|
|
Packit |
5c3484 |
add n, -1, n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b1): mulx bd, %g2, plo1
|
|
Packit |
5c3484 |
umulxhi(bd, %g2, phi1)
|
|
Packit |
5c3484 |
brlz,pn n, L(wd1)
|
|
Packit |
5c3484 |
add qp, -8, qp
|
|
Packit |
5c3484 |
L(gt1): ldx [ap + 8], a0
|
|
Packit |
5c3484 |
ldx [ap + 16], a1
|
|
Packit |
5c3484 |
b L(lo1)
|
|
Packit |
5c3484 |
add ap, 8, ap
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b0): ldx [ap + 8], a1
|
|
Packit |
5c3484 |
mulx bd, %g2, plo0
|
|
Packit |
5c3484 |
umulxhi(bd, %g2, phi0)
|
|
Packit |
5c3484 |
ldx [ap + 16], a0
|
|
Packit |
5c3484 |
b L(lo0)
|
|
Packit |
5c3484 |
add qp, -16, qp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(top): ldx [ap + 0], a0
|
|
Packit |
5c3484 |
sub h, phi1, h
|
|
Packit |
5c3484 |
L(lo2): mulx bd, a1, plo1
|
|
Packit |
5c3484 |
umulxhi(bd, a1, phi1)
|
|
Packit |
5c3484 |
subcc h, plo0, h
|
|
Packit |
5c3484 |
addxc( phi0, %g0, phi0)
|
|
Packit |
5c3484 |
stx h, [qp + 0]
|
|
Packit |
5c3484 |
ldx [ap + 8], a1
|
|
Packit |
5c3484 |
sub h, phi0, h
|
|
Packit |
5c3484 |
L(lo1): mulx bd, a0, plo0
|
|
Packit |
5c3484 |
umulxhi(bd, a0, phi0)
|
|
Packit |
5c3484 |
subcc h, plo1, h
|
|
Packit |
5c3484 |
addxc( phi1, %g0, phi1)
|
|
Packit |
5c3484 |
stx h, [qp + 8]
|
|
Packit |
5c3484 |
ldx [ap + 16], a0
|
|
Packit |
5c3484 |
sub h, phi1, h
|
|
Packit |
5c3484 |
L(lo0): mulx bd, a1, plo1
|
|
Packit |
5c3484 |
umulxhi(bd, a1, phi1)
|
|
Packit |
5c3484 |
subcc h, plo0, h
|
|
Packit |
5c3484 |
addxc( phi0, %g0, phi0)
|
|
Packit |
5c3484 |
stx h, [qp + 16]
|
|
Packit |
5c3484 |
ldx [ap + 24], a1
|
|
Packit |
5c3484 |
sub h, phi0, h
|
|
Packit |
5c3484 |
L(lo3): mulx bd, a0, plo0
|
|
Packit |
5c3484 |
umulxhi(bd, a0, phi0)
|
|
Packit |
5c3484 |
subcc h, plo1, h
|
|
Packit |
5c3484 |
addxc( phi1, %g0, phi1)
|
|
Packit |
5c3484 |
stx h, [qp + 24]
|
|
Packit |
5c3484 |
add ap, 32, ap
|
|
Packit |
5c3484 |
add qp, 32, qp
|
|
Packit |
5c3484 |
brgz,pt n, L(top)
|
|
Packit |
5c3484 |
add n, -4, n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): sub h, phi1, h
|
|
Packit |
5c3484 |
L(wd2): mulx bd, a1, plo1
|
|
Packit |
5c3484 |
umulxhi(bd, a1, phi1)
|
|
Packit |
5c3484 |
subcc h, plo0, h
|
|
Packit |
5c3484 |
addxc( phi0, %g0, phi0)
|
|
Packit |
5c3484 |
stx h, [qp + 0]
|
|
Packit |
5c3484 |
sub h, phi0, h
|
|
Packit |
5c3484 |
L(wd1): subcc h, plo1, h
|
|
Packit |
5c3484 |
addxc( phi1, %g0, phi1)
|
|
Packit |
5c3484 |
stx h, [qp + 8]
|
|
Packit |
5c3484 |
sub h, phi1, %i0
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
restore
|
|
Packit |
5c3484 |
EPILOGUE()
|