Home | History | Annotate | Line # | Download | only in vax
n_sqrt.S revision 1.8.40.1
      1  1.8.40.1     tls /*	$NetBSD: n_sqrt.S,v 1.8.40.1 2014/08/20 00:02:18 tls Exp $	*/
      2       1.1   ragge /*
      3       1.1   ragge  * Copyright (c) 1985, 1993
      4       1.1   ragge  *	The Regents of the University of California.  All rights reserved.
      5       1.1   ragge  *
      6       1.1   ragge  * Redistribution and use in source and binary forms, with or without
      7       1.1   ragge  * modification, are permitted provided that the following conditions
      8       1.1   ragge  * are met:
      9       1.1   ragge  * 1. Redistributions of source code must retain the above copyright
     10       1.1   ragge  *    notice, this list of conditions and the following disclaimer.
     11       1.1   ragge  * 2. Redistributions in binary form must reproduce the above copyright
     12       1.1   ragge  *    notice, this list of conditions and the following disclaimer in the
     13       1.1   ragge  *    documentation and/or other materials provided with the distribution.
     14       1.6     agc  * 3. Neither the name of the University nor the names of its contributors
     15       1.1   ragge  *    may be used to endorse or promote products derived from this software
     16       1.1   ragge  *    without specific prior written permission.
     17       1.1   ragge  *
     18       1.1   ragge  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     19       1.1   ragge  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     20       1.1   ragge  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     21       1.1   ragge  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     22       1.1   ragge  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     23       1.1   ragge  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     24       1.1   ragge  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     25       1.1   ragge  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     26       1.1   ragge  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     27       1.1   ragge  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     28       1.1   ragge  * SUCH DAMAGE.
     29       1.1   ragge  *
     30       1.1   ragge  *	@(#)sqrt.s	8.1 (Berkeley) 6/4/93
     31       1.1   ragge  */
     32       1.1   ragge 
     33       1.3    matt #include <machine/asm.h>
     34       1.3    matt 
     35  1.8.40.1     tls #ifdef WEAK_ALIAS
     36  1.8.40.1     tls WEAK_ALIAS(sqrtl, sqrt)
     37  1.8.40.1     tls #endif
     38  1.8.40.1     tls 
     39       1.1   ragge /*
     40       1.1   ragge  * double sqrt(arg)   revised August 15,1982
     41       1.1   ragge  * double arg;
     42       1.1   ragge  * if(arg<0.0) { _errno = EDOM; return(<a reserved operand>); }
     43       1.1   ragge  * if arg is a reserved operand it is returned as it is
     44       1.1   ragge  * W. Kahan's magic square root
     45       1.1   ragge  * coded by Heidi Stettner and revised by Emile LeBlanc 8/18/82
     46       1.1   ragge  *
     47       1.1   ragge  * entry points:_d_sqrt		address of double arg is on the stack
     48       1.1   ragge  *		_sqrt		double arg is on the stack
     49       1.1   ragge  */
     50       1.1   ragge 	.set	EDOM,33
     51       1.1   ragge 
     52       1.5    matt ENTRY(d_sqrt, 0x003c)		# save %r5,%r4,%r3,%r2
     53       1.5    matt 	movq	*4(%ap),%r0
     54       1.3    matt 	jbr  	dsqrt2
     55       1.3    matt 
     56       1.5    matt ENTRY(sqrt, 0x003c)		# save %r5,%r4,%r3,%r2
     57       1.5    matt 	movq    4(%ap),%r0
     58       1.3    matt 
     59       1.5    matt dsqrt2:	bicw3	$0x807f,%r0,%r2	# check exponent of input
     60       1.1   ragge 	jeql	noexp		# biased exponent is zero -> 0.0 or reserved
     61       1.8    matt 	bsbb	__libm_dsqrt_r5_lcl
     62       1.1   ragge noexp:	ret
     63       1.1   ragge 
     64       1.1   ragge /* **************************** internal procedure */
     65       1.1   ragge 
     66       1.8    matt 	.hidden __libm_dsqrt_r5
     67       1.8    matt ALTENTRY(__libm_dsqrt_r5)
     68       1.8    matt 	halt
     69       1.8    matt 	halt
     70       1.4    matt __libm_dsqrt_r5_lcl:
     71       1.3    matt 				/* ENTRY POINT FOR cdabs and cdsqrt	*/
     72       1.1   ragge 				/* returns double square root scaled by	*/
     73       1.5    matt 				/* 2^%r6	*/
     74       1.1   ragge 
     75       1.5    matt 	movd	%r0,%r4
     76       1.1   ragge 	jleq	nonpos		# argument is not positive
     77       1.5    matt 	movzwl	%r4,%r2
     78       1.5    matt 	ashl	$-1,%r2,%r0
     79       1.5    matt 	addw2	$0x203c,%r0	# %r0 has magic initial approximation
     80       1.1   ragge /*
     81       1.1   ragge  * Do two steps of Heron's rule
     82       1.1   ragge  * ((arg/guess) + guess) / 2 = better guess
     83       1.1   ragge  */
     84       1.5    matt 	divf3	%r0,%r4,%r2
     85       1.5    matt 	addf2	%r2,%r0
     86       1.5    matt 	subw2	$0x80,%r0	# divide by two
     87       1.5    matt 
     88       1.5    matt 	divf3	%r0,%r4,%r2
     89       1.5    matt 	addf2	%r2,%r0
     90       1.5    matt 	subw2	$0x80,%r0	# divide by two
     91       1.1   ragge 
     92       1.1   ragge /* Scale argument and approximation to prevent over/underflow */
     93       1.1   ragge 
     94       1.5    matt 	bicw3	$0x807f,%r4,%r1
     95       1.5    matt 	subw2	$0x4080,%r1		# %r1 contains scaling factor
     96       1.5    matt 	subw2	%r1,%r4
     97       1.5    matt 	movl	%r0,%r2
     98       1.5    matt 	subw2	%r1,%r2
     99       1.1   ragge 
    100       1.1   ragge /* Cubic step
    101       1.1   ragge  *
    102       1.1   ragge  * b = a + 2*a*(n-a*a)/(n+3*a*a) where b is better approximation,
    103       1.1   ragge  * a is approximation, and n is the original argument.
    104       1.1   ragge  * (let s be scale factor in the following comments)
    105       1.1   ragge  */
    106       1.5    matt 	clrl	%r1
    107       1.5    matt 	clrl	%r3
    108       1.5    matt 	muld2	%r0,%r2			# %r2:%r3 = a*a/s
    109       1.5    matt 	subd2	%r2,%r4			# %r4:%r5 = n/s - a*a/s
    110       1.5    matt 	addw2	$0x100,%r2		# %r2:%r3 = 4*a*a/s
    111       1.5    matt 	addd2	%r4,%r2			# %r2:%r3 = n/s + 3*a*a/s
    112       1.5    matt 	muld2	%r0,%r4			# %r4:%r5 = a*n/s - a*a*a/s
    113       1.5    matt 	divd2	%r2,%r4			# %r4:%r5 = a*(n-a*a)/(n+3*a*a)
    114       1.5    matt 	addw2	$0x80,%r4		# %r4:%r5 = 2*a*(n-a*a)/(n+3*a*a)
    115       1.5    matt 	addd2	%r4,%r0			# %r0:%r1 = a + 2*a*(n-a*a)/(n+3*a*a)
    116       1.1   ragge 	rsb				# DONE!
    117       1.1   ragge nonpos:
    118       1.1   ragge 	jneq	negarg
    119       1.3    matt 	ret				# argument and root are zero
    120       1.1   ragge negarg:
    121       1.1   ragge 	pushl	$EDOM
    122       1.3    matt 	calls	$1,_C_LABEL(infnan)	# generate the reserved op fault
    123       1.1   ragge 	ret
    124       1.7  mhitch 
    125       1.7  mhitch ENTRY(sqrtf, 0)
    126       1.7  mhitch 	cvtfd	4(%ap),-(%sp)
    127       1.7  mhitch 	calls	$2,_C_LABEL(sqrt)
    128       1.7  mhitch 	cvtdf	%r0,%r0
    129       1.7  mhitch 	ret
    130