summaryrefslogtreecommitdiffstats
path: root/arch/mips/math-emu/dp_maddf.c
blob: 4a2d03c72959cb5f1fb67da81e9f11b143086256 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
/*
 * IEEE754 floating point arithmetic
 * double precision: MADDF.f (Fused Multiply Add)
 * MADDF.fmt: FPR[fd] = FPR[fd] + (FPR[fs] x FPR[ft])
 *
 * MIPS floating point support
 * Copyright (C) 2015 Imagination Technologies, Ltd.
 * Author: Markos Chandras <markos.chandras@imgtec.com>
 *
 *  This program is free software; you can distribute it and/or modify it
 *  under the terms of the GNU General Public License as published by the
 *  Free Software Foundation; version 2 of the License.
 */

#include "ieee754dp.h"

enum maddf_flags {
	maddf_negate_product	= 1 << 0,
};

static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x,
				 union ieee754dp y, enum maddf_flags flags)
{
	int re;
	int rs;
	u64 rm;
	unsigned lxm;
	unsigned hxm;
	unsigned lym;
	unsigned hym;
	u64 lrm;
	u64 hrm;
	u64 t;
	u64 at;
	int s;

	COMPXDP;
	COMPYDP;
	COMPZDP;

	EXPLODEXDP;
	EXPLODEYDP;
	EXPLODEZDP;

	FLUSHXDP;
	FLUSHYDP;
	FLUSHZDP;

	ieee754_clearcx();

	switch (zc) {
	case IEEE754_CLASS_SNAN:
		ieee754_setcx(IEEE754_INVALID_OPERATION);
		return ieee754dp_nanxcpt(z);
	case IEEE754_CLASS_DNORM:
		DPDNORMZ;
	/* QNAN is handled separately below */
	}

	switch (CLPAIR(xc, yc)) {
	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
		return ieee754dp_nanxcpt(y);

	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
		return ieee754dp_nanxcpt(x);

	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
		return y;

	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
		return x;


	/*
	 * Infinity handling
	 */
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
		if (zc == IEEE754_CLASS_QNAN)
			return z;
		ieee754_setcx(IEEE754_INVALID_OPERATION);
		return ieee754dp_indef();

	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
		if (zc == IEEE754_CLASS_QNAN)
			return z;
		return ieee754dp_inf(xs ^ ys);

	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
		if (zc == IEEE754_CLASS_INF)
			return ieee754dp_inf(zs);
		/* Multiplication is 0 so just return z */
		return z;

	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
		DPDNORMX;

	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
		if (zc == IEEE754_CLASS_QNAN)
			return z;
		else if (zc == IEEE754_CLASS_INF)
			return ieee754dp_inf(zs);
		DPDNORMY;
		break;

	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
		if (zc == IEEE754_CLASS_QNAN)
			return z;
		else if (zc == IEEE754_CLASS_INF)
			return ieee754dp_inf(zs);
		DPDNORMX;
		break;

	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
		if (zc == IEEE754_CLASS_QNAN)
			return z;
		else if (zc == IEEE754_CLASS_INF)
			return ieee754dp_inf(zs);
		/* fall through to real computations */
	}

	/* Finally get to do some computation */

	/*
	 * Do the multiplication bit first
	 *
	 * rm = xm * ym, re = xe + ye basically
	 *
	 * At this point xm and ym should have been normalized.
	 */
	assert(xm & DP_HIDDEN_BIT);
	assert(ym & DP_HIDDEN_BIT);

	re = xe + ye;
	rs = xs ^ ys;
	if (flags & maddf_negate_product)
		rs ^= 1;

	/* shunt to top of word */
	xm <<= 64 - (DP_FBITS + 1);
	ym <<= 64 - (DP_FBITS + 1);

	/*
	 * Multiply 64 bits xm, ym to give high 64 bits rm with stickness.
	 */

	/* 32 * 32 => 64 */
#define DPXMULT(x, y)	((u64)(x) * (u64)y)

	lxm = xm;
	hxm = xm >> 32;
	lym = ym;
	hym = ym >> 32;

	lrm = DPXMULT(lxm, lym);
	hrm = DPXMULT(hxm, hym);

	t = DPXMULT(lxm, hym);

	at = lrm + (t << 32);
	hrm += at < lrm;
	lrm = at;

	hrm = hrm + (t >> 32);

	t = DPXMULT(hxm, lym);

	at = lrm + (t << 32);
	hrm += at < lrm;
	lrm = at;

	hrm = hrm + (t >> 32);

	rm = hrm | (lrm != 0);

	/*
	 * Sticky shift down to normal rounding precision.
	 */
	if ((s64) rm < 0) {
		rm = (rm >> (64 - (DP_FBITS + 1 + 3))) |
		     ((rm << (DP_FBITS + 1 + 3)) != 0);
		re++;
	} else {
		rm = (rm >> (64 - (DP_FBITS + 1 + 3 + 1))) |
		     ((rm << (DP_FBITS + 1 + 3 + 1)) != 0);
	}
	assert(rm & (DP_HIDDEN_BIT << 3));

	/* And now the addition */
	assert(zm & DP_HIDDEN_BIT);

	/*
	 * Provide guard,round and stick bit space.
	 */
	zm <<= 3;

	if (ze > re) {
		/*
		 * Have to shift y fraction right to align.
		 */
		s = ze - re;
		rm = XDPSRS(rm, s);
		re += s;
	} else if (re > ze) {
		/*
		 * Have to shift x fraction right to align.
		 */
		s = re - ze;
		zm = XDPSRS(zm, s);
		ze += s;
	}
	assert(ze == re);
	assert(ze <= DP_EMAX);

	if (zs == rs) {
		/*
		 * Generate 28 bit result of adding two 27 bit numbers
		 * leaving result in xm, xs and xe.
		 */
		zm = zm + rm;

		if (zm >> (DP_FBITS + 1 + 3)) { /* carry out */
			zm = XDPSRS1(zm);
			ze++;
		}
	} else {
		if (zm >= rm) {
			zm = zm - rm;
		} else {
			zm = rm - zm;
			zs = rs;
		}
		if (zm == 0)
			return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);

		/*
		 * Normalize to rounding precision.
		 */
		while ((zm >> (DP_FBITS + 3)) == 0) {
			zm <<= 1;
			ze--;
		}
	}

	return ieee754dp_format(zs, ze, zm);
}

union ieee754dp ieee754dp_maddf(union ieee754dp z, union ieee754dp x,
				union ieee754dp y)
{
	return _dp_maddf(z, x, y, 0);
}

union ieee754dp ieee754dp_msubf(union ieee754dp z, union ieee754dp x,
				union ieee754dp y)
{
	return _dp_maddf(z, x, y, maddf_negate_product);
}