Blame mpn/sparc64/sec_tabselect.asm

Packit 5c3484
dnl  SPARC v9 mpn_sec_tabselect.
Packit 5c3484
Packit 5c3484
dnl  Contributed to the GNU project by Torbjörn Granlund and 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 1&2:	 2 hopefully
Packit 5c3484
C UltraSPARC 3:		 3
Packit 5c3484
C UltraSPARC T1:	17
Packit 5c3484
C UltraSPARC T3:	 ?
Packit 5c3484
C UltraSPARC T4/T5:	 2.25 hopefully
Packit 5c3484
Packit 5c3484
C INPUT PARAMETERS
Packit 5c3484
define(`rp',     `%i0')
Packit 5c3484
define(`tp',     `%i1')
Packit 5c3484
define(`n',      `%i2')
Packit 5c3484
define(`nents',  `%i3')
Packit 5c3484
define(`which',  `%i4')
Packit 5c3484
Packit 5c3484
define(`i',      `%g1')
Packit 5c3484
define(`j',      `%g3')
Packit 5c3484
define(`stride', `%g4')
Packit 5c3484
define(`tporig', `%g5')
Packit 5c3484
define(`mask',   `%o0')
Packit 5c3484
Packit 5c3484
define(`data0',  `%l0')
Packit 5c3484
define(`data1',  `%l1')
Packit 5c3484
define(`data2',  `%l2')
Packit 5c3484
define(`data3',  `%l3')
Packit 5c3484
define(`t0',     `%l4')
Packit 5c3484
define(`t1',     `%l5')
Packit 5c3484
define(`t2',     `%l6')
Packit 5c3484
define(`t3',     `%l7')
Packit 5c3484
Packit 5c3484
ASM_START()
Packit 5c3484
	REGISTER(%g2,#scratch)
Packit 5c3484
	REGISTER(%g3,#scratch)
Packit 5c3484
PROLOGUE(mpn_sec_tabselect)
Packit 5c3484
	save	%sp, -176, %sp
Packit 5c3484
Packit 5c3484
	sllx	n, 3, stride
Packit 5c3484
	sub	n, 4, j
Packit 5c3484
	brlz	j, L(outer_end)
Packit 5c3484
	 mov	tp, tporig
Packit 5c3484
Packit 5c3484
L(outer_loop):
Packit 5c3484
	clr	data0
Packit 5c3484
	clr	data1
Packit 5c3484
	clr	data2
Packit 5c3484
	clr	data3
Packit 5c3484
	mov	tporig, tp
Packit 5c3484
	mov	nents, i
Packit 5c3484
	mov	which, %o1
Packit 5c3484
Packit 5c3484
L(top):	subcc	%o1, 1, %o1		C set carry iff o1 = 0
Packit 5c3484
	ldx	[tp + 0], t0
Packit 5c3484
	subc	%g0, %g0, mask
Packit 5c3484
	ldx	[tp + 8], t1
Packit 5c3484
	sub	i, 1, i
Packit 5c3484
	ldx	[tp + 16], t2
Packit 5c3484
	ldx	[tp + 24], t3
Packit 5c3484
	add	tp, stride, tp
Packit 5c3484
	and	t0, mask, t0
Packit 5c3484
	and	t1, mask, t1
Packit 5c3484
	or	t0, data0, data0
Packit 5c3484
	and	t2, mask, t2
Packit 5c3484
	or	t1, data1, data1
Packit 5c3484
	and	t3, mask, t3
Packit 5c3484
	or	t2, data2, data2
Packit 5c3484
	brnz	i, L(top)
Packit 5c3484
	 or	t3, data3, data3
Packit 5c3484
Packit 5c3484
	stx	data0, [rp + 0]
Packit 5c3484
	subcc	j, 4, j
Packit 5c3484
	stx	data1, [rp + 8]
Packit 5c3484
	stx	data2, [rp + 16]
Packit 5c3484
	stx	data3, [rp + 24]
Packit 5c3484
	add	tporig, (4 * 8), tporig
Packit 5c3484
Packit 5c3484
	brgez	j, L(outer_loop)
Packit 5c3484
	 add	rp, (4 * 8), rp
Packit 5c3484
L(outer_end):
Packit 5c3484
Packit 5c3484
Packit 5c3484
	andcc	n, 2, %g0
Packit 5c3484
	be	L(b0x)
Packit 5c3484
	 nop
Packit 5c3484
L(b1x):	clr	data0
Packit 5c3484
	clr	data1
Packit 5c3484
	mov	tporig, tp
Packit 5c3484
	mov	nents, i
Packit 5c3484
	mov	which, %o1
Packit 5c3484
Packit 5c3484
L(tp2):	subcc	%o1, 1, %o1
Packit 5c3484
	ldx	[tp + 0], t0
Packit 5c3484
	subc	%g0, %g0, mask
Packit 5c3484
	ldx	[tp + 8], t1
Packit 5c3484
	sub	i, 1, i
Packit 5c3484
	add	tp, stride, tp
Packit 5c3484
	and	t0, mask, t0
Packit 5c3484
	and	t1, mask, t1
Packit 5c3484
	or	t0, data0, data0
Packit 5c3484
	brnz	i, L(tp2)
Packit 5c3484
	 or	t1, data1, data1
Packit 5c3484
Packit 5c3484
	stx	data0, [rp + 0]
Packit 5c3484
	stx	data1, [rp + 8]
Packit 5c3484
	add	tporig, (2 * 8), tporig
Packit 5c3484
	add	rp, (2 * 8), rp
Packit 5c3484
Packit 5c3484
Packit 5c3484
L(b0x):	andcc	n, 1, %g0
Packit 5c3484
	be	L(b00)
Packit 5c3484
	 nop
Packit 5c3484
L(b01):	clr	data0
Packit 5c3484
	mov	tporig, tp
Packit 5c3484
	mov	nents, i
Packit 5c3484
	mov	which, %o1
Packit 5c3484
Packit 5c3484
L(tp1):	subcc	%o1, 1, %o1
Packit 5c3484
	ldx	[tp + 0], t0
Packit 5c3484
	subc	%g0, %g0, mask
Packit 5c3484
	sub	i, 1, i
Packit 5c3484
	add	tp, stride, tp
Packit 5c3484
	and	t0, mask, t0
Packit 5c3484
	brnz	i, L(tp1)
Packit 5c3484
	 or	t0, data0, data0
Packit 5c3484
Packit 5c3484
	stx	data0, [rp + 0]
Packit 5c3484
Packit 5c3484
L(b00):	 ret
Packit 5c3484
	  restore
Packit 5c3484
EPILOGUE()