1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
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 *	$NetBSD: fpu_add.c,v 1.3 1996/03/14 19:41:52 christos Exp $
42 */
43
44#include <sys/cdefs.h>
45
46/*
47 * Perform an FPU add (return x + y).
48 *
49 * To subtract, negate y and call add.
50 */
51
52#include <sys/param.h>
53#include <stdint.h>
54
55#include "fsr.h"
56#include "instr.h"
57
58#include "fpu_arith.h"
59#include "fpu_emu.h"
60#include "fpu_extern.h"
61
62struct fpn *
63__fpu_add(fe)
64	struct fpemu *fe;
65{
66	struct fpn *x = &fe->fe_f1, *y = &fe->fe_f2, *r;
67	uint32_t r0, r1, r2, r3;
68	int rd;
69
70	/*
71	 * Put the `heavier' operand on the right (see fpu_emu.h).
72	 * Then we will have one of the following cases, taken in the
73	 * following order:
74	 *
75	 *  - y = NaN.  Implied: if only one is a signalling NaN, y is.
76	 *	The result is y.
77	 *  - y = Inf.  Implied: x != NaN (is 0, number, or Inf: the NaN
78	 *    case was taken care of earlier).
79	 *	If x = -y, the result is NaN.  Otherwise the result
80	 *	is y (an Inf of whichever sign).
81	 *  - y is 0.  Implied: x = 0.
82	 *	If x and y differ in sign (one positive, one negative),
83	 *	the result is +0 except when rounding to -Inf.  If same:
84	 *	+0 + +0 = +0; -0 + -0 = -0.
85	 *  - x is 0.  Implied: y != 0.
86	 *	Result is y.
87	 *  - other.  Implied: both x and y are numbers.
88	 *	Do addition a la Hennessey & Patterson.
89	 */
90	ORDER(x, y);
91	if (ISNAN(y))
92		return (y);
93	if (ISINF(y)) {
94		if (ISINF(x) && x->fp_sign != y->fp_sign)
95			return (__fpu_newnan(fe));
96		return (y);
97	}
98	rd = FSR_GET_RD(fe->fe_fsr);
99	if (ISZERO(y)) {
100		if (rd != FSR_RD_NINF)	/* only -0 + -0 gives -0 */
101			y->fp_sign &= x->fp_sign;
102		else			/* any -0 operand gives -0 */
103			y->fp_sign |= x->fp_sign;
104		return (y);
105	}
106	if (ISZERO(x))
107		return (y);
108	/*
109	 * We really have two numbers to add, although their signs may
110	 * differ.  Make the exponents match, by shifting the smaller
111	 * number right (e.g., 1.011 => 0.1011) and increasing its
112	 * exponent (2^3 => 2^4).  Note that we do not alter the exponents
113	 * of x and y here.
114	 */
115	r = &fe->fe_f3;
116	r->fp_class = FPC_NUM;
117	if (x->fp_exp == y->fp_exp) {
118		r->fp_exp = x->fp_exp;
119		r->fp_sticky = 0;
120	} else {
121		if (x->fp_exp < y->fp_exp) {
122			/*
123			 * Try to avoid subtract case iii (see below).
124			 * This also guarantees that x->fp_sticky = 0.
125			 */
126			SWAP(x, y);
127		}
128		/* now x->fp_exp > y->fp_exp */
129		r->fp_exp = x->fp_exp;
130		r->fp_sticky = __fpu_shr(y, x->fp_exp - y->fp_exp);
131	}
132	r->fp_sign = x->fp_sign;
133	if (x->fp_sign == y->fp_sign) {
134		FPU_DECL_CARRY
135
136		/*
137		 * The signs match, so we simply add the numbers.  The result
138		 * may be `supernormal' (as big as 1.111...1 + 1.111...1, or
139		 * 11.111...0).  If so, a single bit shift-right will fix it
140		 * (but remember to adjust the exponent).
141		 */
142		/* r->fp_mant = x->fp_mant + y->fp_mant */
143		FPU_ADDS(r->fp_mant[3], x->fp_mant[3], y->fp_mant[3]);
144		FPU_ADDCS(r->fp_mant[2], x->fp_mant[2], y->fp_mant[2]);
145		FPU_ADDCS(r->fp_mant[1], x->fp_mant[1], y->fp_mant[1]);
146		FPU_ADDC(r0, x->fp_mant[0], y->fp_mant[0]);
147		if ((r->fp_mant[0] = r0) >= FP_2) {
148			(void) __fpu_shr(r, 1);
149			r->fp_exp++;
150		}
151	} else {
152		FPU_DECL_CARRY
153
154		/*
155		 * The signs differ, so things are rather more difficult.
156		 * H&P would have us negate the negative operand and add;
157		 * this is the same as subtracting the negative operand.
158		 * This is quite a headache.  Instead, we will subtract
159		 * y from x, regardless of whether y itself is the negative
160		 * operand.  When this is done one of three conditions will
161		 * hold, depending on the magnitudes of x and y:
162		 *   case i)   |x| > |y|.  The result is just x - y,
163		 *	with x's sign, but it may need to be normalized.
164		 *   case ii)  |x| = |y|.  The result is 0 (maybe -0)
165		 *	so must be fixed up.
166		 *   case iii) |x| < |y|.  We goofed; the result should
167		 *	be (y - x), with the same sign as y.
168		 * We could compare |x| and |y| here and avoid case iii,
169		 * but that would take just as much work as the subtract.
170		 * We can tell case iii has occurred by an overflow.
171		 *
172		 * N.B.: since x->fp_exp >= y->fp_exp, x->fp_sticky = 0.
173		 */
174		/* r->fp_mant = x->fp_mant - y->fp_mant */
175		FPU_SET_CARRY(y->fp_sticky);
176		FPU_SUBCS(r3, x->fp_mant[3], y->fp_mant[3]);
177		FPU_SUBCS(r2, x->fp_mant[2], y->fp_mant[2]);
178		FPU_SUBCS(r1, x->fp_mant[1], y->fp_mant[1]);
179		FPU_SUBC(r0, x->fp_mant[0], y->fp_mant[0]);
180		if (r0 < FP_2) {
181			/* cases i and ii */
182			if ((r0 | r1 | r2 | r3) == 0) {
183				/* case ii */
184				r->fp_class = FPC_ZERO;
185				r->fp_sign = rd == FSR_RD_NINF;
186				return (r);
187			}
188		} else {
189			/*
190			 * Oops, case iii.  This can only occur when the
191			 * exponents were equal, in which case neither
192			 * x nor y have sticky bits set.  Flip the sign
193			 * (to y's sign) and negate the result to get y - x.
194			 */
195#ifdef DIAGNOSTIC
196			if (x->fp_exp != y->fp_exp || r->fp_sticky)
197				__utrap_panic("fpu_add");
198#endif
199			r->fp_sign = y->fp_sign;
200			FPU_SUBS(r3, 0, r3);
201			FPU_SUBCS(r2, 0, r2);
202			FPU_SUBCS(r1, 0, r1);
203			FPU_SUBC(r0, 0, r0);
204		}
205		r->fp_mant[3] = r3;
206		r->fp_mant[2] = r2;
207		r->fp_mant[1] = r1;
208		r->fp_mant[0] = r0;
209		if (r0 < FP_1)
210			__fpu_norm(r);
211	}
212	return (r);
213}
214