fpu_add.c revision 1.5 1 /* $NetBSD: fpu_add.c,v 1.5 2003/08/07 16:29:36 agc Exp $ */
2
3 /*
4 * Copyright (c) 1992, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This software was developed by the Computer Systems Engineering group
8 * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
9 * contributed to Berkeley.
10 *
11 * All advertising materials mentioning features or use of this software
12 * must display the following acknowledgement:
13 * This product includes software developed by the University of
14 * California, Lawrence Berkeley Laboratory.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 3. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * SUCH DAMAGE.
39 *
40 * @(#)fpu_add.c 8.1 (Berkeley) 6/11/93
41 */
42
43 /*
44 * Perform an FPU add (return x + y).
45 *
46 * To subtract, negate y and call add.
47 */
48
49 #include <sys/cdefs.h>
50 __KERNEL_RCSID(0, "$NetBSD: fpu_add.c,v 1.5 2003/08/07 16:29:36 agc Exp $");
51
52 #include <sys/types.h>
53 #ifdef DIAGNOSTIC
54 #include <sys/systm.h>
55 #endif
56
57 #include <machine/reg.h>
58 #include <machine/instr.h>
59
60 #include <sparc/fpu/fpu_arith.h>
61 #include <sparc/fpu/fpu_emu.h>
62 #include <sparc/fpu/fpu_extern.h>
63
64 struct fpn *
65 fpu_add(fe)
66 register struct fpemu *fe;
67 {
68 register struct fpn *x = &fe->fe_f1, *y = &fe->fe_f2, *r;
69 register u_int r0, r1, r2, r3;
70 register int rd;
71
72 /*
73 * Put the `heavier' operand on the right (see fpu_emu.h).
74 * Then we will have one of the following cases, taken in the
75 * following order:
76 *
77 * - y = NaN. Implied: if only one is a signalling NaN, y is.
78 * The result is y.
79 * - y = Inf. Implied: x != NaN (is 0, number, or Inf: the NaN
80 * case was taken care of earlier).
81 * If x = -y, the result is NaN. Otherwise the result
82 * is y (an Inf of whichever sign).
83 * - y is 0. Implied: x = 0.
84 * If x and y differ in sign (one positive, one negative),
85 * the result is +0 except when rounding to -Inf. If same:
86 * +0 + +0 = +0; -0 + -0 = -0.
87 * - x is 0. Implied: y != 0.
88 * Result is y.
89 * - other. Implied: both x and y are numbers.
90 * Do addition a la Hennessey & Patterson.
91 */
92 ORDER(x, y);
93 if (ISNAN(y))
94 return (y);
95 if (ISINF(y)) {
96 if (ISINF(x) && x->fp_sign != y->fp_sign)
97 return (fpu_newnan(fe));
98 return (y);
99 }
100 rd = ((fe->fe_fsr >> FSR_RD_SHIFT) & FSR_RD_MASK);
101 if (ISZERO(y)) {
102 if (rd != FSR_RD_RM) /* only -0 + -0 gives -0 */
103 y->fp_sign &= x->fp_sign;
104 else /* any -0 operand gives -0 */
105 y->fp_sign |= x->fp_sign;
106 return (y);
107 }
108 if (ISZERO(x))
109 return (y);
110 /*
111 * We really have two numbers to add, although their signs may
112 * differ. Make the exponents match, by shifting the smaller
113 * number right (e.g., 1.011 => 0.1011) and increasing its
114 * exponent (2^3 => 2^4). Note that we do not alter the exponents
115 * of x and y here.
116 */
117 r = &fe->fe_f3;
118 r->fp_class = FPC_NUM;
119 if (x->fp_exp == y->fp_exp) {
120 r->fp_exp = x->fp_exp;
121 r->fp_sticky = 0;
122 } else {
123 if (x->fp_exp < y->fp_exp) {
124 /*
125 * Try to avoid subtract case iii (see below).
126 * This also guarantees that x->fp_sticky = 0.
127 */
128 SWAP(x, y);
129 }
130 /* now x->fp_exp > y->fp_exp */
131 r->fp_exp = x->fp_exp;
132 r->fp_sticky = fpu_shr(y, x->fp_exp - y->fp_exp);
133 }
134 r->fp_sign = x->fp_sign;
135 if (x->fp_sign == y->fp_sign) {
136 FPU_DECL_CARRY
137
138 /*
139 * The signs match, so we simply add the numbers. The result
140 * may be `supernormal' (as big as 1.111...1 + 1.111...1, or
141 * 11.111...0). If so, a single bit shift-right will fix it
142 * (but remember to adjust the exponent).
143 */
144 /* r->fp_mant = x->fp_mant + y->fp_mant */
145 FPU_ADDS(r->fp_mant[3], x->fp_mant[3], y->fp_mant[3]);
146 FPU_ADDCS(r->fp_mant[2], x->fp_mant[2], y->fp_mant[2]);
147 FPU_ADDCS(r->fp_mant[1], x->fp_mant[1], y->fp_mant[1]);
148 FPU_ADDC(r0, x->fp_mant[0], y->fp_mant[0]);
149 if ((r->fp_mant[0] = r0) >= FP_2) {
150 (void) fpu_shr(r, 1);
151 r->fp_exp++;
152 }
153 } else {
154 FPU_DECL_CARRY
155
156 /*
157 * The signs differ, so things are rather more difficult.
158 * H&P would have us negate the negative operand and add;
159 * this is the same as subtracting the negative operand.
160 * This is quite a headache. Instead, we will subtract
161 * y from x, regardless of whether y itself is the negative
162 * operand. When this is done one of three conditions will
163 * hold, depending on the magnitudes of x and y:
164 * case i) |x| > |y|. The result is just x - y,
165 * with x's sign, but it may need to be normalized.
166 * case ii) |x| = |y|. The result is 0 (maybe -0)
167 * so must be fixed up.
168 * case iii) |x| < |y|. We goofed; the result should
169 * be (y - x), with the same sign as y.
170 * We could compare |x| and |y| here and avoid case iii,
171 * but that would take just as much work as the subtract.
172 * We can tell case iii has occurred by an overflow.
173 *
174 * N.B.: since x->fp_exp >= y->fp_exp, x->fp_sticky = 0.
175 */
176 /* r->fp_mant = x->fp_mant - y->fp_mant */
177 FPU_SET_CARRY(y->fp_sticky);
178 FPU_SUBCS(r3, x->fp_mant[3], y->fp_mant[3]);
179 FPU_SUBCS(r2, x->fp_mant[2], y->fp_mant[2]);
180 FPU_SUBCS(r1, x->fp_mant[1], y->fp_mant[1]);
181 FPU_SUBC(r0, x->fp_mant[0], y->fp_mant[0]);
182 if (r0 < FP_2) {
183 /* cases i and ii */
184 if ((r0 | r1 | r2 | r3) == 0) {
185 /* case ii */
186 r->fp_class = FPC_ZERO;
187 r->fp_sign = rd == FSR_RD_RM;
188 return (r);
189 }
190 } else {
191 /*
192 * Oops, case iii. This can only occur when the
193 * exponents were equal, in which case neither
194 * x nor y have sticky bits set. Flip the sign
195 * (to y's sign) and negate the result to get y - x.
196 */
197 #ifdef DIAGNOSTIC
198 if (x->fp_exp != y->fp_exp || r->fp_sticky)
199 panic("fpu_add");
200 #endif
201 r->fp_sign = y->fp_sign;
202 FPU_SUBS(r3, 0, r3);
203 FPU_SUBCS(r2, 0, r2);
204 FPU_SUBCS(r1, 0, r1);
205 FPU_SUBC(r0, 0, r0);
206 }
207 r->fp_mant[3] = r3;
208 r->fp_mant[2] = r2;
209 r->fp_mant[1] = r1;
210 r->fp_mant[0] = r0;
211 if (r0 < FP_1)
212 fpu_norm(r);
213 }
214 return (r);
215 }
216