Home | History | Annotate | Line # | Download | only in alpha
      1  1.1  mrg dnl  Alpha mpn_addlsh2_n/mpn_sublsh2_n -- rp[] = up[] +- (vp[] << 2).
      2  1.1  mrg 
      3  1.1  mrg dnl  Copyright 2003, 2013 Free Software Foundation, Inc.
      4  1.1  mrg 
      5  1.1  mrg dnl  This file is part of the GNU MP Library.
      6  1.1  mrg dnl
      7  1.1  mrg dnl  The GNU MP Library is free software; you can redistribute it and/or modify
      8  1.1  mrg dnl  it under the terms of either:
      9  1.1  mrg dnl
     10  1.1  mrg dnl    * the GNU Lesser General Public License as published by the Free
     11  1.1  mrg dnl      Software Foundation; either version 3 of the License, or (at your
     12  1.1  mrg dnl      option) any later version.
     13  1.1  mrg dnl
     14  1.1  mrg dnl  or
     15  1.1  mrg dnl
     16  1.1  mrg dnl    * the GNU General Public License as published by the Free Software
     17  1.1  mrg dnl      Foundation; either version 2 of the License, or (at your option) any
     18  1.1  mrg dnl      later version.
     19  1.1  mrg dnl
     20  1.1  mrg dnl  or both in parallel, as here.
     21  1.1  mrg dnl
     22  1.1  mrg dnl  The GNU MP Library is distributed in the hope that it will be useful, but
     23  1.1  mrg dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
     24  1.1  mrg dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
     25  1.1  mrg dnl  for more details.
     26  1.1  mrg dnl
     27  1.1  mrg dnl  You should have received copies of the GNU General Public License and the
     28  1.1  mrg dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
     29  1.1  mrg dnl  see https://www.gnu.org/licenses/.
     30  1.1  mrg 
     31  1.1  mrg include(`../config.m4')
     32  1.1  mrg 
     33  1.1  mrg C      cycles/limb
     34  1.1  mrg C EV4:     ?
     35  1.1  mrg C EV5:     6
     36  1.1  mrg C EV6:     3.75
     37  1.1  mrg 
     38  1.1  mrg C TODO
     39  1.1  mrg C  * Tune to reach 3.5 c/l on ev6 and 5.75 c/l on ev5.
     40  1.1  mrg 
     41  1.1  mrg define(`rp',`r16')
     42  1.1  mrg define(`up',`r17')
     43  1.1  mrg define(`vp',`r18')
     44  1.1  mrg define(`n', `r19')
     45  1.1  mrg 
     46  1.1  mrg define(`u0', `r8')
     47  1.1  mrg define(`u1', `r1')
     48  1.1  mrg define(`v0', `r4')
     49  1.1  mrg define(`v1', `r5')
     50  1.1  mrg 
     51  1.1  mrg define(`cy0', `r0')
     52  1.1  mrg define(`cy1', `r20')
     53  1.1  mrg define(`cy', `r22')
     54  1.1  mrg define(`rr', `r24')
     55  1.1  mrg define(`ps', `r25')
     56  1.1  mrg define(`sl', `r28')
     57  1.1  mrg 
     58  1.1  mrg ifdef(`OPERATION_addlsh2_n',`
     59  1.1  mrg   define(ADDSUB,       addq)
     60  1.1  mrg   define(CARRY,       `cmpult $1,$2,$3')
     61  1.1  mrg   define(func, mpn_addlsh2_n)
     62  1.1  mrg ')
     63  1.1  mrg ifdef(`OPERATION_sublsh2_n',`
     64  1.1  mrg   define(ADDSUB,       subq)
     65  1.1  mrg   define(CARRY,       `cmpult $2,$1,$3')
     66  1.1  mrg   define(func, mpn_sublsh2_n)
     67  1.1  mrg ')
     68  1.1  mrg 
     69  1.1  mrg MULFUNC_PROLOGUE(mpn_addlsh2_n mpn_sublsh2_n)
     70  1.1  mrg 
     71  1.1  mrg ASM_START()
     72  1.1  mrg PROLOGUE(func)
     73  1.1  mrg 	and	n, 2, cy0
     74  1.1  mrg 	blbs	n, L(bx1)
     75  1.1  mrg L(bx0):	ldq	v1, 0(vp)
     76  1.1  mrg 	ldq	u1, 0(up)
     77  1.1  mrg 	bis	r31, r31, r2
     78  1.1  mrg 	bne	cy0, L(b10)
     79  1.1  mrg 
     80  1.1  mrg L(b00):	lda	vp, 48(vp)
     81  1.1  mrg 	lda	up, -16(up)
     82  1.1  mrg 	lda	rp, -8(rp)
     83  1.1  mrg 	s4addq	v1, r31, sl
     84  1.1  mrg 	br	r31, L(lo0)
     85  1.1  mrg 
     86  1.1  mrg L(b10):	lda	vp, 32(vp)
     87  1.1  mrg 	lda	rp, 8(rp)
     88  1.1  mrg 	lda	cy0, 0(r31)
     89  1.1  mrg 	br	r31, L(lo2)
     90  1.1  mrg 
     91  1.1  mrg L(bx1):	ldq	v0, 0(vp)
     92  1.1  mrg 	ldq	u0, 0(up)
     93  1.1  mrg 	lda	cy1, 0(r31)
     94  1.1  mrg 	bis	r31, r31, r3
     95  1.1  mrg 	nop
     96  1.1  mrg 	beq	cy0, L(b01)
     97  1.1  mrg 
     98  1.1  mrg L(b11):	lda	vp, 40(vp)
     99  1.1  mrg 	lda	up, -24(up)
    100  1.1  mrg 	lda	rp, 16(rp)
    101  1.1  mrg 	br	r31, L(lo3)
    102  1.1  mrg 
    103  1.1  mrg L(b01):	lda	n, -4(n)
    104  1.1  mrg 	ble	n, L(end)
    105  1.1  mrg 	lda	vp, 24(vp)
    106  1.1  mrg 	lda	up, -8(up)
    107  1.1  mrg 
    108  1.1  mrg 	ALIGN(16)
    109  1.1  mrg L(top):	s4addq	v0, r3, sl	C combined vlimb
    110  1.1  mrg 	ldq	v1, -16(vp)
    111  1.1  mrg 	ADDSUB	u0, sl, ps	C ulimb + (vlimb << 1)
    112  1.1  mrg 	ldq	u1, 16(up)
    113  1.1  mrg 	srl	v0, 62, r2	C high v bits
    114  1.1  mrg 	ADDSUB	ps, cy1, rr	C consume carry from previous operation
    115  1.1  mrg 	CARRY(	ps, u0, cy0)	C carry out #2
    116  1.1  mrg 	stq	rr, 0(rp)
    117  1.1  mrg 	CARRY(	rr, ps, cy)	C carry out #3
    118  1.1  mrg 	lda	vp, 32(vp)	C bookkeeping
    119  1.1  mrg 	addq	cy, cy0, cy0	C final carry out
    120  1.1  mrg 	s4addq	v1, r2, sl
    121  1.1  mrg L(lo0):	ldq	v0, -40(vp)
    122  1.1  mrg 	ADDSUB	u1, sl, ps
    123  1.1  mrg 	ldq	u0, 24(up)
    124  1.1  mrg 	srl	v1, 62, r3
    125  1.1  mrg 	ADDSUB	ps, cy0, rr
    126  1.1  mrg 	CARRY(	ps, u1, cy1)
    127  1.1  mrg 	stq	rr, 8(rp)
    128  1.1  mrg 	CARRY(	rr, ps, cy)
    129  1.1  mrg 	lda	rp, 32(rp)	C bookkeeping
    130  1.1  mrg 	addq	cy, cy1, cy1
    131  1.1  mrg L(lo3):	s4addq	v0, r3, sl
    132  1.1  mrg 	ldq	v1, -32(vp)
    133  1.1  mrg 	ADDSUB	u0, sl, ps
    134  1.1  mrg 	ldq	u1, 32(up)
    135  1.1  mrg 	srl	v0, 62, r2
    136  1.1  mrg 	ADDSUB	ps, cy1, rr
    137  1.1  mrg 	CARRY(	ps, u0, cy0)
    138  1.1  mrg 	stq	rr, -16(rp)
    139  1.1  mrg 	CARRY(	rr, ps, cy)
    140  1.1  mrg 	lda	up, 32(up)	C bookkeeping
    141  1.1  mrg 	addq	cy, cy0, cy0
    142  1.1  mrg L(lo2):	s4addq	v1, r2, sl
    143  1.1  mrg 	ldq	v0, -24(vp)
    144  1.1  mrg 	ADDSUB	u1, sl, ps
    145  1.1  mrg 	ldq	u0, 8(up)
    146  1.1  mrg 	srl	v1, 62, r3
    147  1.1  mrg 	ADDSUB	ps, cy0, rr
    148  1.1  mrg 	CARRY(	ps, u1, cy1)
    149  1.1  mrg 	stq	rr, -8(rp)
    150  1.1  mrg 	CARRY(	rr, ps, cy)
    151  1.1  mrg 	lda	n, -4(n)	C bookkeeping
    152  1.1  mrg 	addq	cy, cy1, cy1
    153  1.1  mrg 	bgt	n, L(top)
    154  1.1  mrg 
    155  1.1  mrg L(end):	s4addq	v0, r3, sl
    156  1.1  mrg 	ADDSUB	u0, sl, ps
    157  1.1  mrg 	srl	v0, 62, r2
    158  1.1  mrg 	ADDSUB	ps, cy1, rr
    159  1.1  mrg 	CARRY(	ps, u0, cy0)
    160  1.1  mrg 	stq	rr, 0(rp)
    161  1.1  mrg 	CARRY(	rr, ps, cy)
    162  1.1  mrg 	addq	cy, cy0, cy0
    163  1.1  mrg 	addq	cy0, r2, r0
    164  1.1  mrg 
    165  1.1  mrg 	ret	r31,(r26),1
    166  1.1  mrg EPILOGUE()
    167  1.1  mrg ASM_END()
    168