xref: /illumos-gate/usr/src/uts/sparc/fpu/addsub.c (revision 7c478bd95313f5f23a4c958a745db2134aa03244)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright (c) 1988 by Sun Microsystems, Inc.
24  */
25 
26 #ident	"%Z%%M%	%I%	%E% SMI"	/* SunOS-4.1 1.8 88/12/06 */
27 
28 #include <sys/fpu/fpu_simulator.h>
29 #include <sys/fpu/globals.h>
30 
31 static void
true_add(px,py,pz)32 true_add(px, py, pz)
33 	unpacked	*px, *py, *pz;
34 {
35 	unsigned 	c;
36 	unpacked	*pt;
37 
38 	if ((int) px->fpclass <= (int) py->fpclass) {	/* Reverse. */
39 		pt = py;
40 		py = px;
41 		px = pt;
42 	}
43 	/* Now class(x) >= class(y). */
44 	switch (px->fpclass) {
45 	case fp_quiet:		/* NaN + x -> NaN */
46 	case fp_signaling:	/* NaN + x -> NaN */
47 	case fp_infinity:	/* Inf + x -> Inf */
48 	case fp_zero:		/* 0 + 0 -> 0 */
49 		*pz = *px;
50 		return;
51 	default:
52 		if (py->fpclass == fp_zero) {
53 			*pz = *px;
54 			return;
55 		}
56 	}
57 	/* Now z is normal or subnormal. */
58 	/* Now y is normal or subnormal. */
59 	if (px->exponent < py->exponent) {	/* Reverse. */
60 		pt = py;
61 		py = px;
62 		px = pt;
63 	}
64 	/* Now class(x) >= class(y). */
65 	pz->fpclass = px->fpclass;
66 	pz->sign = px->sign;
67 	pz->exponent = px->exponent;
68 	pz->rounded = pz->sticky  = 0;
69 
70 	if (px->exponent != py->exponent) {	/* pre-alignment required */
71 		fpu_rightshift(py, pz->exponent - py->exponent);
72 		pz->rounded = py->rounded;
73 		pz->sticky  = py->sticky;
74 	}
75 	c = 0;
76 	c = fpu_add3wc(&(pz->significand[3]), px->significand[3],
77 						py->significand[3], c);
78 	c = fpu_add3wc(&(pz->significand[2]), px->significand[2],
79 						py->significand[2], c);
80 	c = fpu_add3wc(&(pz->significand[1]), px->significand[1],
81 						py->significand[1], c);
82 	c = fpu_add3wc(&(pz->significand[0]), px->significand[0],
83 						py->significand[0], c);
84 
85 	/* Handle carry out of msb. */
86 	if (pz->significand[0] >= 0x20000) {
87 		fpu_rightshift(pz, 1);	/* Carried out bit. */
88 		pz->exponent++;		/* Renormalize. */
89 	}
90 }
91 
92 static void
true_sub(pfpsd,px,py,pz)93 true_sub(pfpsd, px, py, pz)
94 	fp_simd_type	*pfpsd;		/* Pointer to simulator data */
95 	unpacked	*px, *py, *pz;
96 {
97 	unsigned	*z, g, s, r, c;
98 	unpacked	*pt;
99 
100 	if ((int) px->fpclass <= (int) py->fpclass) {	/* Reverse. */
101 		pt = py;
102 		py = px;
103 		px = pt;
104 	}
105 	/* Now class(x) >= class(y). */
106 	*pz = *px;		/* Tentative difference: x. */
107 	switch (pz->fpclass) {
108 	case fp_quiet:		/* NaN - x -> NaN */
109 	case fp_signaling:	/* NaN - x -> NaN */
110 		return;
111 	case fp_infinity:	/* Inf - x -> Inf */
112 		if (py->fpclass == fp_infinity) {
113 			fpu_error_nan(pfpsd, pz);	/* Inf - Inf -> NaN */
114 			pz->fpclass = fp_quiet;
115 		}
116 		return;
117 	case fp_zero:		/* 0 - 0 -> 0 */
118 		pz->sign = (pfpsd->fp_direction == fp_negative);
119 		return;
120 	default:
121 		if (py->fpclass == fp_zero)
122 			return;
123 	}
124 
125 	/* x and y are both normal or subnormal. */
126 
127 	if (px->exponent < py->exponent) { /* Reverse. */
128 		pt = py;
129 		py = px;
130 		px = pt;
131 	}
132 	/* Now exp(x) >= exp(y). */
133 	pz->fpclass = px->fpclass;
134 	pz->sign = px->sign;
135 	pz->exponent = px->exponent;
136 	pz->rounded = 0;
137 	pz->sticky = 0;
138 	z = pz->significand;
139 
140 	if (px->exponent == py->exponent) {	/* no pre-alignment required */
141 		c = 0;
142 		c = fpu_sub3wc(&z[3], px->significand[3],
143 				py->significand[3], c);
144 		c = fpu_sub3wc(&z[2], px->significand[2],
145 				py->significand[2], c);
146 		c = fpu_sub3wc(&z[1], px->significand[1],
147 				py->significand[1], c);
148 		c = fpu_sub3wc(&z[0], px->significand[0],
149 				py->significand[0], c);
150 		if ((z[0]|z[1]|z[2]|z[3]) == 0) {	/* exact zero result */
151 			pz->sign = (pfpsd->fp_direction == fp_negative);
152 			pz->fpclass = fp_zero;
153 			return;
154 		}
155 		if (z[0] >= 0x20000) {	/* sign reversal occurred */
156 			pz->sign = py->sign;
157 			c = 0;
158 			c = fpu_neg2wc(&z[3], z[3], c);
159 			c = fpu_neg2wc(&z[2], z[2], c);
160 			c = fpu_neg2wc(&z[1], z[1], c);
161 			c = fpu_neg2wc(&z[0], z[0], c);
162 		}
163 		fpu_normalize(pz);
164 		return;
165 	} else {		/* pre-alignment required */
166 		fpu_rightshift(py, pz->exponent - py->exponent - 1);
167 		r = py->rounded; 	/* rounded bit */
168 		s = py->sticky;		/* sticky bit */
169 		fpu_rightshift(py, 1);
170 		g = py->rounded;	/* guard bit */
171 		if (s != 0) r = (r == 0);
172 		if ((r|s) != 0) g = (g == 0); /* guard and rounded bits of z */
173 		c = ((g|r|s) != 0);
174 		c = fpu_sub3wc(&z[3], px->significand[3],
175 				py->significand[3], c);
176 		c = fpu_sub3wc(&z[2], px->significand[2],
177 				py->significand[2], c);
178 		c = fpu_sub3wc(&z[1], px->significand[1],
179 				py->significand[1], c);
180 		c = fpu_sub3wc(&z[0], px->significand[0],
181 				py->significand[0], c);
182 
183 		if (z[0] >= 0x10000) { 	/* don't need post-shifted */
184 			pz->sticky = s|r;
185 			pz->rounded = g;
186 		} else {		/* post-shifted left 1 bit */
187 			pz->sticky = s;
188 			pz->rounded = r;
189 			pz->significand[0] = (z[0]<<1)|((z[1]&0x80000000)>>31);
190 			pz->significand[1] = (z[1]<<1)|((z[2]&0x80000000)>>31);
191 			pz->significand[2] = (z[2]<<1)|((z[3]&0x80000000)>>31);
192 			pz->significand[3] = (z[3]<<1)|g;
193 			pz->exponent -= 1;
194 			if (z[0] < 0x10000) fpu_normalize(pz);
195 		}
196 		return;
197 	}
198 }
199 
200 void
_fp_add(pfpsd,px,py,pz)201 _fp_add(pfpsd, px, py, pz)
202 	fp_simd_type	*pfpsd;
203 	unpacked	*px, *py, *pz;
204 {
205 	if (px->sign == py->sign)
206 		true_add(px, py, pz);
207 	else
208 		true_sub(pfpsd, px, py, pz);
209 }
210 
211 void
_fp_sub(pfpsd,px,py,pz)212 _fp_sub(pfpsd, px, py, pz)
213 	fp_simd_type	*pfpsd;
214 	unpacked	*px, *py, *pz;
215 {
216 	if (py->fpclass < fp_quiet) py->sign = 1 - py->sign;
217 	if (px->sign == py->sign)
218 		true_add(px, py, pz);
219 	else
220 		true_sub(pfpsd, px, py, pz);
221 }
222