xref: /netbsd-src/lib/libc/arch/sparc/gen/modf.S (revision 76dfffe33547c37f8bdd446e3e4ab0f3c16cea4b)
1/*
2 * Copyright (c) 1992, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This software was developed by the Computer Systems Engineering group
6 * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
7 * contributed to Berkeley.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 *    must display the following acknowledgement:
19 *	This product includes software developed by the University of
20 *	California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 *    may be used to endorse or promote products derived from this software
23 *    without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * from: Header: modf.s,v 1.3 92/06/20 00:00:54 torek Exp
38 * $Id: modf.S,v 1.4 1996/11/16 20:53:20 pk Exp $
39 */
40
41#if defined(LIBC_SCCS) && !defined(lint)
42	.asciz "@(#)modf.s	8.1 (Berkeley) 6/4/93"
43#endif /* LIBC_SCCS and not lint */
44
45#include <machine/asm.h>
46#include <machine/fsr.h>
47
48/*
49 * double modf(double val, double *iptr)
50 *
51 * Returns the fractional part of `val', storing the integer part of
52 * `val' in *iptr.  Both *iptr and the return value have the same sign
53 * as `val'.
54 *
55 * Method:
56 *
57 * We use the fpu's normalization hardware to compute the integer portion
58 * of the double precision argument.  Sun IEEE double precision numbers
59 * have 52 bits of mantissa, 11 bits of exponent, and one bit of sign,
60 * with the sign occupying bit 31 of word 0, and the exponent bits 30:20
61 * of word 0.  Thus, values >= 2^52 are by definition integers.
62 *
63 * If we take a value that is in the range [+0..2^52) and add 2^52, all
64 * of the fractional bits fall out and all of the integer bits are summed
65 * with 2^52.  If we then subtract 2^52, we get those integer bits back.
66 * This must be done with rounding set to `towards 0' or `towards -inf'.
67 * `Toward -inf' fails when the value is 0 (we get -0 back)....
68 *
69 * Note that this method will work anywhere, but is machine dependent in
70 * various aspects.
71 *
72 * Stack usage:
73 *	4@[%fp - 4]	saved %fsr
74 *	4@[%fp - 8]	new %fsr with rounding set to `towards 0'
75 *	8@[%fp - 16]	space for moving between %i and %f registers
76 * Register usage:
77 *	%i0%i1		double val;
78 *	%l0		scratch
79 *	%l1		sign bit (0x80000000)
80 *	%i2		double *iptr;
81 *	%f2:f3		`magic number' 2^52, in fpu registers
82 *	%f4:f5		double v, in fpu registers
83 */
84
85	.align	8
86Lmagic:
87	.word	0x43300000	! sign = 0, exponent = 52 + 1023, mantissa = 0
88	.word	0		! (i.e., .double 0r4503599627370496e+00)
89
90L0:
91	.word	0		! 0.0
92	.word	0
93
94ENTRY(modf)
95	save	%sp, -64-16, %sp
96
97	/*
98	 * First, compute v = abs(val) by clearing sign bit,
99	 * and then set up the fpu registers.  This would be
100	 * much easier if we could do alu operations on fpu registers!
101	 */
102	sethi	%hi(0x80000000), %l1	! sign bit
103	andn	%i0, %l1, %l0
104	st	%l0, [%fp - 16]
105#ifdef PIC
106	PICCY_SET(Lmagic, %l0, %o7)
107	ldd	[%l0], %f2
108#else
109	sethi	%hi(Lmagic), %l0
110	ldd	[%l0 + %lo(Lmagic)], %f2
111#endif
112	st	%i1, [%fp - 12]
113	ldd	[%fp - 16], %f4		! %f4:f5 = v
114
115	/*
116	 * Is %f4:f5 >= %f2:f3 ?  If so, it is all integer bits.
117	 * It is probably less, though.
118	 */
119	fcmped	%f4, %f2
120	nop				! fpop2 delay
121	fbuge	Lbig			! if >= (or unordered), go out
122	nop
123
124	/*
125	 * v < 2^52, so add 2^52, then subtract 2^52, but do it all
126	 * with rounding set towards zero.  We leave any enabled
127	 * traps enabled, but change the rounding mode.  This might
128	 * not be so good.  Oh well....
129	 */
130	st	%fsr, [%fp - 4]		! %l5 = current FSR mode
131	set	FSR_RD, %l3		! %l3 = rounding direction mask
132	ld	[%fp - 4], %l5
133	set	FSR_RD_RZ << FSR_RD_SHIFT, %l4
134	andn	%l5, %l3, %l6
135	or	%l6, %l4, %l6		! round towards zero, please
136	and	%l5, %l3, %l5		! save original rounding mode
137	st	%l6, [%fp - 8]
138	ld	[%fp - 8], %fsr
139
140	faddd	%f4, %f2, %f4		! %f4:f5 += 2^52
141	fsubd	%f4, %f2, %f4		! %f4:f5 -= 2^52
142
143	/*
144	 * Restore %fsr, but leave exceptions accrued.
145	 */
146	st	%fsr, [%fp - 4]
147	ld	[%fp - 4], %l6
148	andn	%l6, %l3, %l6		! %l6 = %fsr & ~FSR_RD;
149	or	%l5, %l6, %l5		! %l5 |= %l6;
150	st	%l5, [%fp - 4]
151	ld	[%fp - 4], %fsr		! restore %fsr, leaving accrued stuff
152
153	/*
154	 * Now insert the original sign in %f4:f5.
155	 * This is a lot of work, so it is conditional here.
156	 */
157	btst	%l1, %i0
158	be	1f
159	nop
160	st	%f4, [%fp - 16]
161	ld	[%fp - 16], %g1
162	or	%l1, %g1, %g1
163	st	%g1, [%fp - 16]
164	ld	[%fp - 16], %f4
1651:
166
167	/*
168	 * The value in %f4:f5 is now the integer portion of the original
169	 * argument.  We need to store this in *ival (%i2), subtract it
170	 * from the original value argument (%i0:i1), and return the result.
171	 */
172	std	%f4, [%i2]		! *ival = %f4:f5;
173	std	%i0, [%fp - 16]
174	ldd	[%fp - 16], %f0		! %f0:f1 = val;
175	fsubd	%f0, %f4, %f0		! %f0:f1 -= %f4:f5;
176	ret
177	restore
178
179Lbig:
180	/*
181	 * We get here if the original comparison of %f4:f5 (v) to
182	 * %f2:f3 (2^52) came out `greater or unordered'.  In this
183	 * case the integer part is the original value, and the
184	 * fractional part is 0.
185	 */
186#ifdef PIC
187	PICCY_SET(L0, %l0, %o7)
188	std	%f0, [%i2]		! *ival = val;
189	ldd	[%l0], %f0		! return 0.0;
190#else
191	sethi	%hi(L0), %l0
192	std	%f0, [%i2]		! *ival = val;
193	ldd	[%l0 + %lo(L0)], %f0	! return 0.0;
194#endif
195	ret
196	restore
197