xref: /original-bsd/lib/libc/mips/gen/ldexp.s (revision c3e32dec)
1/*-
2 * Copyright (c) 1991, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Ralph Campbell.
7 *
8 * %sccs.include.redist.c%
9 */
10
11#include <machine/machAsmDefs.h>
12
13#if defined(LIBC_SCCS) && !defined(lint)
14	ASMSTR("@(#)ldexp.s	8.1 (Berkeley) 06/04/93")
15#endif /* LIBC_SCCS and not lint */
16
17#define DEXP_INF	0x7ff
18#define DEXP_BIAS	1023
19#define DEXP_MIN	-1022
20#define DEXP_MAX	1023
21#define DFRAC_BITS	52
22#define DIMPL_ONE	0x00100000
23#define DLEAD_ZEROS	31 - 20
24#define STICKYBIT	1
25#define GUARDBIT	0x80000000
26#define DSIGNAL_NAN	0x00040000
27#define DQUIET_NAN0	0x0007ffff
28#define DQUIET_NAN1	0xffffffff
29
30/*
31 * double ldexp(x, N)
32 *	double x; int N;
33 *
34 * Return x * (2**N), for integer values N.
35 */
36LEAF(ldexp)
37	mfc1	v1, $f13		# get MSW of x
38	mfc1	t3, $f12		# get LSW of x
39	sll	t1, v1, 1		# get x exponent
40	srl	t1, t1, 32 - 11
41	beq	t1, DEXP_INF, 9f	# is it a NAN or infinity?
42	beq	t1, zero, 1f		# zero or denormalized number?
43	addu	t1, t1, a2		# scale exponent
44	sll	v0, a2, 20		# position N for addition
45	bge	t1, DEXP_INF, 8f	# overflow?
46	addu	v0, v0, v1		# multiply by (2**N)
47	ble	t1, zero, 4f		# underflow?
48	mtc1	v0, $f1			# save MSW of result
49	mtc1	t3, $f0			# save LSW of result
50	j	ra
511:
52	sll	t2, v1, 32 - 20		# get x fraction
53	srl	t2, t2, 32 - 20
54	srl	t0, v1, 31		# get x sign
55	bne	t2, zero, 1f
56	beq	t3, zero, 9f		# result is zero
571:
58/*
59 * Find out how many leading zero bits are in t2,t3 and put in t9.
60 */
61	move	v0, t2
62	move	t9, zero
63	bne	t2, zero, 1f
64	move	v0, t3
65	addu	t9, 32
661:
67	srl	t4, v0, 16
68	bne	t4, zero, 1f
69	addu	t9, 16
70	sll	v0, 16
711:
72	srl	t4, v0, 24
73	bne	t4, zero, 1f
74	addu	t9, 8
75	sll	v0, 8
761:
77	srl	t4, v0, 28
78	bne	t4, zero, 1f
79	addu	t9, 4
80	sll	v0, 4
811:
82	srl	t4, v0, 30
83	bne	t4, zero, 1f
84	addu	t9, 2
85	sll	v0, 2
861:
87	srl	t4, v0, 31
88	bne	t4, zero, 1f
89	addu	t9, 1
90/*
91 * Now shift t2,t3 the correct number of bits.
92 */
931:
94	subu	t9, t9, DLEAD_ZEROS	# dont count normal leading zeros
95	li	t1, DEXP_MIN + DEXP_BIAS
96	subu	t1, t1, t9		# adjust exponent
97	addu	t1, t1, a2		# scale exponent
98	li	v0, 32
99	blt	t9, v0, 1f
100	subu	t9, t9, v0		# shift fraction left >= 32 bits
101	sll	t2, t3, t9
102	move	t3, zero
103	b	2f
1041:
105	subu	v0, v0, t9		# shift fraction left < 32 bits
106	sll	t2, t2, t9
107	srl	t4, t3, v0
108	or	t2, t2, t4
109	sll	t3, t3, t9
1102:
111	bge	t1, DEXP_INF, 8f	# overflow?
112	ble	t1, zero, 4f		# underflow?
113	sll	t2, t2, 32 - 20		# clear implied one bit
114	srl	t2, t2, 32 - 20
1153:
116	sll	t1, t1, 31 - 11		# reposition exponent
117	sll	t0, t0, 31		# reposition sign
118	or	t0, t0, t1		# put result back together
119	or	t0, t0, t2
120	mtc1	t0, $f1			# save MSW of result
121	mtc1	t3, $f0			# save LSW of result
122	j	ra
1234:
124	li	v0, 0x80000000
125	ble	t1, -52, 7f		# is result too small for denorm?
126	sll	t2, v1, 31 - 20		# clear exponent, extract fraction
127	or	t2, t2, v0		# set implied one bit
128	blt	t1, -30, 2f		# will all bits in t3 be shifted out?
129	srl	t2, t2, 31 - 20		# shift fraction back to normal position
130	subu	t1, t1, 1
131	sll	t4, t2, t1		# shift right t2,t3 based on exponent
132	srl	t8, t3, t1		# save bits shifted out
133	negu	t1
134	srl	t3, t3, t1
135	or	t3, t3, t4
136	srl	t2, t2, t1
137	bge	t8, zero, 1f		# does result need to be rounded?
138	addu	t3, t3, 1		# round result
139	sltu	t4, t3, 1
140	sll	t8, t8, 1
141	addu	t2, t2, t4
142	bne	t8, zero, 1f		# round result to nearest
143	and	t3, t3, ~1
1441:
145	mtc1	t3, $f0			# save denormalized result (LSW)
146	mtc1	t2, $f1			# save denormalized result (MSW)
147	bge	v1, zero, 1f		# should result be negative?
148	neg.d	$f0, $f0		# negate result
1491:
150	j	ra
1512:
152	mtc1	zero, $f1		# exponent and upper fraction
153	addu	t1, t1, 20		# compute amount to shift right by
154	sll	t8, t2, t1		# save bits shifted out
155	negu	t1
156	srl	t3, t2, t1
157	bge	t8, zero, 1f		# does result need to be rounded?
158	addu	t3, t3, 1		# round result
159	sltu	t4, t3, 1
160	sll	t8, t8, 1
161	mtc1	t4, $f1			# exponent and upper fraction
162	bne	t8, zero, 1f		# round result to nearest
163	and	t3, t3, ~1
1641:
165	mtc1	t3, $f0
166	bge	v1, zero, 1f		# is result negative?
167	neg.d	$f0, $f0		# negate result
1681:
169	j	ra
1707:
171	mtc1	zero, $f0		# result is zero
172	mtc1	zero, $f1
173	beq	t0, zero, 1f		# is result positive?
174	neg.d	$f0, $f0		# negate result
1751:
176	j	ra
1778:
178	li	t1, 0x7ff00000		# result is infinity (MSW)
179	mtc1	t1, $f1
180	mtc1	zero, $f0		# result is infinity (LSW)
181	bge	v1, zero, 1f		# should result be negative infinity?
182	neg.d	$f0, $f0		# result is negative infinity
1831:
184	add.d	$f0, $f0		# cause overflow faults if enabled
185	j	ra
1869:
187	mov.d	$f0, $f12		# yes, result is just x
188	j	ra
189END(ldexp)
190