xref: /linux/arch/mips/math-emu/dp_mul.c (revision 02680c23d7b3febe45ea3d4f9818c2b2dc89020a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* IEEE754 floating point arithmetic
3  * double precision: common utilities
4  */
5 /*
6  * MIPS floating point support
7  * Copyright (C) 1994-2000 Algorithmics Ltd.
8  */
9 
10 #include "ieee754dp.h"
11 
12 union ieee754dp ieee754dp_mul(union ieee754dp x, union ieee754dp y)
13 {
14 	int re;
15 	int rs;
16 	u64 rm;
17 	unsigned int lxm;
18 	unsigned int hxm;
19 	unsigned int lym;
20 	unsigned int hym;
21 	u64 lrm;
22 	u64 hrm;
23 	u64 t;
24 	u64 at;
25 
26 	COMPXDP;
27 	COMPYDP;
28 
29 	EXPLODEXDP;
30 	EXPLODEYDP;
31 
32 	ieee754_clearcx();
33 
34 	FLUSHXDP;
35 	FLUSHYDP;
36 
37 	switch (CLPAIR(xc, yc)) {
38 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
39 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
40 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
41 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
42 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
43 		return ieee754dp_nanxcpt(y);
44 
45 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
46 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
47 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
48 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
49 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
50 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
51 		return ieee754dp_nanxcpt(x);
52 
53 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
54 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
55 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
56 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
57 		return y;
58 
59 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
60 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
61 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
62 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
63 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
64 		return x;
65 
66 
67 	/*
68 	 * Infinity handling
69 	 */
70 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
71 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
72 		ieee754_setcx(IEEE754_INVALID_OPERATION);
73 		return ieee754dp_indef();
74 
75 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
76 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
77 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
78 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
79 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
80 		return ieee754dp_inf(xs ^ ys);
81 
82 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
83 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
84 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
85 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
86 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
87 		return ieee754dp_zero(xs ^ ys);
88 
89 
90 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
91 		DPDNORMX;
92 		fallthrough;
93 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
94 		DPDNORMY;
95 		break;
96 
97 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
98 		DPDNORMX;
99 		break;
100 
101 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
102 		break;
103 	}
104 	/* rm = xm * ym, re = xe+ye basically */
105 	assert(xm & DP_HIDDEN_BIT);
106 	assert(ym & DP_HIDDEN_BIT);
107 
108 	re = xe + ye;
109 	rs = xs ^ ys;
110 
111 	/* shunt to top of word */
112 	xm <<= 64 - (DP_FBITS + 1);
113 	ym <<= 64 - (DP_FBITS + 1);
114 
115 	/*
116 	 * Multiply 64 bits xm, ym to give high 64 bits rm with stickness.
117 	 */
118 
119 	lxm = xm;
120 	hxm = xm >> 32;
121 	lym = ym;
122 	hym = ym >> 32;
123 
124 	lrm = DPXMULT(lxm, lym);
125 	hrm = DPXMULT(hxm, hym);
126 
127 	t = DPXMULT(lxm, hym);
128 
129 	at = lrm + (t << 32);
130 	hrm += at < lrm;
131 	lrm = at;
132 
133 	hrm = hrm + (t >> 32);
134 
135 	t = DPXMULT(hxm, lym);
136 
137 	at = lrm + (t << 32);
138 	hrm += at < lrm;
139 	lrm = at;
140 
141 	hrm = hrm + (t >> 32);
142 
143 	rm = hrm | (lrm != 0);
144 
145 	/*
146 	 * Sticky shift down to normal rounding precision.
147 	 */
148 	if ((s64) rm < 0) {
149 		rm = (rm >> (64 - (DP_FBITS + 1 + 3))) |
150 		     ((rm << (DP_FBITS + 1 + 3)) != 0);
151 		re++;
152 	} else {
153 		rm = (rm >> (64 - (DP_FBITS + 1 + 3 + 1))) |
154 		     ((rm << (DP_FBITS + 1 + 3 + 1)) != 0);
155 	}
156 	assert(rm & (DP_HIDDEN_BIT << 3));
157 
158 	return ieee754dp_format(rs, re, rm);
159 }
160