Home | History | Annotate | Line # | Download | only in ultrasparct3
      1  1.1  mrg dnl  SPARC T3/T4/T5 mpn_bdiv_dbm1c.
      2  1.1  mrg 
      3  1.1  mrg dnl  Contributed to the GNU project by Torbjrn Granlund.
      4  1.1  mrg 
      5  1.1  mrg dnl  Copyright 2013 Free Software Foundation, Inc.
      6  1.1  mrg 
      7  1.1  mrg dnl  This file is part of the GNU MP Library.
      8  1.1  mrg dnl
      9  1.1  mrg dnl  The GNU MP Library is free software; you can redistribute it and/or modify
     10  1.1  mrg dnl  it under the terms of either:
     11  1.1  mrg dnl
     12  1.1  mrg dnl    * the GNU Lesser General Public License as published by the Free
     13  1.1  mrg dnl      Software Foundation; either version 3 of the License, or (at your
     14  1.1  mrg dnl      option) any later version.
     15  1.1  mrg dnl
     16  1.1  mrg dnl  or
     17  1.1  mrg dnl
     18  1.1  mrg dnl    * the GNU General Public License as published by the Free Software
     19  1.1  mrg dnl      Foundation; either version 2 of the License, or (at your option) any
     20  1.1  mrg dnl      later version.
     21  1.1  mrg dnl
     22  1.1  mrg dnl  or both in parallel, as here.
     23  1.1  mrg dnl
     24  1.1  mrg dnl  The GNU MP Library is distributed in the hope that it will be useful, but
     25  1.1  mrg dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
     26  1.1  mrg dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
     27  1.1  mrg dnl  for more details.
     28  1.1  mrg dnl
     29  1.1  mrg dnl  You should have received copies of the GNU General Public License and the
     30  1.1  mrg dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
     31  1.1  mrg dnl  see https://www.gnu.org/licenses/.
     32  1.1  mrg 
     33  1.1  mrg include(`../config.m4')
     34  1.1  mrg 
     35  1.1  mrg C                  cycles/limb
     36  1.1  mrg C UltraSPARC T3:	25
     37  1.1  mrg C UltraSPARC T4/T5:	 4
     38  1.1  mrg 
     39  1.1  mrg C INPUT PARAMETERS
     40  1.1  mrg define(`qp',  `%i0')
     41  1.1  mrg define(`ap',  `%i1')
     42  1.1  mrg define(`n',   `%i2')
     43  1.1  mrg define(`bd',  `%i3')
     44  1.1  mrg define(`h',   `%i4')
     45  1.1  mrg 
     46  1.1  mrg define(`plo0',`%g4')  define(`plo1',`%g5')
     47  1.1  mrg define(`phi0',`%l0')  define(`phi1',`%l1')
     48  1.1  mrg define(`a0',  `%g1')  define(`a1',  `%g3')
     49  1.1  mrg 
     50  1.1  mrg ASM_START()
     51  1.1  mrg 	REGISTER(%g2,#scratch)
     52  1.1  mrg 	REGISTER(%g3,#scratch)
     53  1.1  mrg PROLOGUE(mpn_bdiv_dbm1c)
     54  1.1  mrg 	save	%sp, -176, %sp
     55  1.1  mrg 
     56  1.1  mrg 	and	n, 3, %g5
     57  1.1  mrg 	ldx	[ap + 0], %g2
     58  1.1  mrg 	add	n, -5, n
     59  1.1  mrg 	brz	%g5, L(b0)
     60  1.1  mrg 	 cmp	%g5, 2
     61  1.1  mrg 	bcs	%xcc, L(b1)
     62  1.1  mrg 	 nop
     63  1.1  mrg 	be	%xcc, L(b2)
     64  1.1  mrg 	 nop
     65  1.1  mrg 
     66  1.1  mrg L(b3):	ldx	[ap + 8], a0
     67  1.1  mrg 	mulx	bd, %g2, plo1
     68  1.1  mrg 	umulxhi(bd, %g2, phi1)
     69  1.1  mrg 	ldx	[ap + 16], a1
     70  1.1  mrg 	add	qp, -24, qp
     71  1.1  mrg 	b	L(lo3)
     72  1.1  mrg 	 add	ap, -8, ap
     73  1.1  mrg 
     74  1.1  mrg L(b2):	ldx	[ap + 8], a1
     75  1.1  mrg 	mulx	bd, %g2, plo0
     76  1.1  mrg 	umulxhi(bd, %g2, phi0)
     77  1.1  mrg 	brlz,pt n, L(wd2)
     78  1.1  mrg 	 nop
     79  1.1  mrg L(gt2):	ldx	[ap + 16], a0
     80  1.1  mrg 	add	ap, 16, ap
     81  1.1  mrg 	b	L(lo2)
     82  1.1  mrg 	 add	n, -1, n
     83  1.1  mrg 
     84  1.1  mrg L(b1):	mulx	bd, %g2, plo1
     85  1.1  mrg 	 umulxhi(bd, %g2, phi1)
     86  1.1  mrg 	brlz,pn	n, L(wd1)
     87  1.1  mrg 	 add	qp, -8, qp
     88  1.1  mrg L(gt1):	ldx	[ap + 8], a0
     89  1.1  mrg 	ldx	[ap + 16], a1
     90  1.1  mrg 	b	L(lo1)
     91  1.1  mrg 	 add	ap, 8, ap
     92  1.1  mrg 
     93  1.1  mrg L(b0):	ldx	[ap + 8], a1
     94  1.1  mrg 	mulx	bd, %g2, plo0
     95  1.1  mrg 	umulxhi(bd, %g2, phi0)
     96  1.1  mrg 	ldx	[ap + 16], a0
     97  1.1  mrg 	b	L(lo0)
     98  1.1  mrg 	 add	qp, -16, qp
     99  1.1  mrg 
    100  1.1  mrg L(top):	ldx	[ap + 0], a0
    101  1.1  mrg 	sub	h, phi1, h
    102  1.1  mrg L(lo2):	mulx	bd, a1, plo1
    103  1.1  mrg 	umulxhi(bd, a1, phi1)
    104  1.1  mrg 	subcc	h, plo0, h
    105  1.1  mrg 	addxc(	phi0, %g0, phi0)
    106  1.1  mrg 	stx	h, [qp + 0]
    107  1.1  mrg 	ldx	[ap + 8], a1
    108  1.1  mrg 	sub	h, phi0, h
    109  1.1  mrg L(lo1):	mulx	bd, a0, plo0
    110  1.1  mrg 	umulxhi(bd, a0, phi0)
    111  1.1  mrg 	subcc	h, plo1, h
    112  1.1  mrg 	addxc(	phi1, %g0, phi1)
    113  1.1  mrg 	stx	h, [qp + 8]
    114  1.1  mrg 	ldx	[ap + 16], a0
    115  1.1  mrg 	sub	h, phi1, h
    116  1.1  mrg L(lo0):	mulx	bd, a1, plo1
    117  1.1  mrg 	umulxhi(bd, a1, phi1)
    118  1.1  mrg 	subcc	h, plo0, h
    119  1.1  mrg 	addxc(	phi0, %g0, phi0)
    120  1.1  mrg 	stx	h, [qp + 16]
    121  1.1  mrg 	ldx	[ap + 24], a1
    122  1.1  mrg 	sub	h, phi0, h
    123  1.1  mrg L(lo3):	mulx	bd, a0, plo0
    124  1.1  mrg 	umulxhi(bd, a0, phi0)
    125  1.1  mrg 	subcc	h, plo1, h
    126  1.1  mrg 	addxc(	phi1, %g0, phi1)
    127  1.1  mrg 	stx	h, [qp + 24]
    128  1.1  mrg 	add	ap, 32, ap
    129  1.1  mrg 	add	qp, 32, qp
    130  1.1  mrg 	brgz,pt	n, L(top)
    131  1.1  mrg 	 add	n, -4, n
    132  1.1  mrg 
    133  1.1  mrg L(end):	sub	h, phi1, h
    134  1.1  mrg L(wd2):	mulx	bd, a1, plo1
    135  1.1  mrg 	umulxhi(bd, a1, phi1)
    136  1.1  mrg 	subcc	h, plo0, h
    137  1.1  mrg 	addxc(	phi0, %g0, phi0)
    138  1.1  mrg 	stx	h, [qp + 0]
    139  1.1  mrg 	sub	h, phi0, h
    140  1.1  mrg L(wd1):	subcc	h, plo1, h
    141  1.1  mrg 	addxc(	phi1, %g0, phi1)
    142  1.1  mrg 	stx	h, [qp + 8]
    143  1.1  mrg 	sub	h, phi1, %i0
    144  1.1  mrg 
    145  1.1  mrg 	ret
    146  1.1  mrg 	 restore
    147  1.1  mrg EPILOGUE()
    148