summaryrefslogtreecommitdiff
path: root/arch/mips/math-emu/dp_maddf.c
blob: 7ad79ed411f5b3a94695e0d85f55f511c2abf270 (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
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
/*
 * 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"


/* 128 bits shift right logical with rounding. */
void srl128(u64 *hptr, u64 *lptr, int count)
{
	u64 low;

	if (count >= 128) {
		*lptr = *hptr != 0 || *lptr != 0;
		*hptr = 0;
	} else if (count >= 64) {
		if (count == 64) {
			*lptr = *hptr | (*lptr != 0);
		} else {
			low = *lptr;
			*lptr = *hptr >> (count - 64);
			*lptr |= (*hptr << (128 - count)) != 0 || low != 0;
		}
		*hptr = 0;
	} else {
		low = *lptr;
		*lptr = low >> count | *hptr << (64 - count);
		*lptr |= (low << (64 - count)) != 0;
		*hptr = *hptr >> count;
	}
}

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

	COMPXDP;
	COMPYDP;
	COMPZDP;

	EXPLODEXDP;
	EXPLODEYDP;
	EXPLODEZDP;

	FLUSHXDP;
	FLUSHYDP;
	FLUSHZDP;

	ieee754_clearcx();

	/*
	 * Handle the cases when at least one of x, y or z is a NaN.
	 * Order of precedence is sNaN, qNaN and z, x, y.
	 */
	if (zc == IEEE754_CLASS_SNAN)
		return ieee754dp_nanxcpt(z);
	if (xc == IEEE754_CLASS_SNAN)
		return ieee754dp_nanxcpt(x);
	if (yc == IEEE754_CLASS_SNAN)
		return ieee754dp_nanxcpt(y);
	if (zc == IEEE754_CLASS_QNAN)
		return z;
	if (xc == IEEE754_CLASS_QNAN)
		return x;
	if (yc == IEEE754_CLASS_QNAN)
		return y;

	if (zc == IEEE754_CLASS_DNORM)
		DPDNORMZ;
	/* ZERO z cases are handled separately below */

	switch (CLPAIR(xc, yc)) {

	/*
	 * Infinity handling
	 */
	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
		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_INF) &&
		    ((!(flags & MADDF_NEGATE_PRODUCT) && (zs != (xs ^ ys))) ||
		     ((flags & MADDF_NEGATE_PRODUCT) && (zs == (xs ^ ys))))) {
			/*
			 * Cases of addition of infinities with opposite signs
			 * or subtraction of infinities with same signs.
			 */
			ieee754_setcx(IEEE754_INVALID_OPERATION);
			return ieee754dp_indef();
		}
		/*
		 * z is here either not an infinity, or an infinity having the
		 * same sign as product (x*y) (in case of MADDF.D instruction)
		 * or product -(x*y) (in MSUBF.D case). The result must be an
		 * infinity, and its sign is determined only by the value of
		 * (flags & MADDF_NEGATE_PRODUCT) and the signs of x and y.
		 */
		if (flags & MADDF_NEGATE_PRODUCT)
			return ieee754dp_inf(1 ^ (xs ^ ys));
		else
			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);
		if (zc == IEEE754_CLASS_ZERO) {
			/* Handle cases +0 + (-0) and similar ones. */
			if ((!(flags & MADDF_NEGATE_PRODUCT)
					&& (zs == (xs ^ ys))) ||
			    ((flags & MADDF_NEGATE_PRODUCT)
					&& (zs != (xs ^ ys))))
				/*
				 * Cases of addition of zeros of equal signs
				 * or subtraction of zeroes of opposite signs.
				 * The sign of the resulting zero is in any
				 * such case determined only by the sign of z.
				 */
				return z;

			return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
		}
		/* x*y is here 0, and z is not 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_INF)
			return ieee754dp_inf(zs);
		DPDNORMY;
		break;

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

	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
		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 and ym to give 128 bits result in hrm:lrm.
	 */

	/* 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);

	/* Put explicit bit at bit 126 if necessary */
	if ((int64_t)hrm < 0) {
		lrm = (hrm << 63) | (lrm >> 1);
		hrm = hrm >> 1;
		re++;
	}

	assert(hrm & (1 << 62));

	if (zc == IEEE754_CLASS_ZERO) {
		/*
		 * Move explicit bit from bit 126 to bit 55 since the
		 * ieee754dp_format code expects the mantissa to be
		 * 56 bits wide (53 + 3 rounding bits).
		 */
		srl128(&hrm, &lrm, (126 - 55));
		return ieee754dp_format(rs, re, lrm);
	}

	/* Move explicit bit from bit 52 to bit 126 */
	lzm = 0;
	hzm = zm << 10;
	assert(hzm & (1 << 62));

	/* Make the exponents the same */
	if (ze > re) {
		/*
		 * Have to shift y fraction right to align.
		 */
		s = ze - re;
		srl128(&hrm, &lrm, s);
		re += s;
	} else if (re > ze) {
		/*
		 * Have to shift x fraction right to align.
		 */
		s = re - ze;
		srl128(&hzm, &lzm, s);
		ze += s;
	}
	assert(ze == re);
	assert(ze <= DP_EMAX);

	/* Do the addition */
	if (zs == rs) {
		/*
		 * Generate 128 bit result by adding two 127 bit numbers
		 * leaving result in hzm:lzm, zs and ze.
		 */
		hzm = hzm + hrm + (lzm > (lzm + lrm));
		lzm = lzm + lrm;
		if ((int64_t)hzm < 0) {        /* carry out */
			srl128(&hzm, &lzm, 1);
			ze++;
		}
	} else {
		if (hzm > hrm || (hzm == hrm && lzm >= lrm)) {
			hzm = hzm - hrm - (lzm < lrm);
			lzm = lzm - lrm;
		} else {
			hzm = hrm - hzm - (lrm < lzm);
			lzm = lrm - lzm;
			zs = rs;
		}
		if (lzm == 0 && hzm == 0)
			return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);

		/*
		 * Put explicit bit at bit 126 if necessary.
		 */
		if (hzm == 0) {
			/* left shift by 63 or 64 bits */
			if ((int64_t)lzm < 0) {
				/* MSB of lzm is the explicit bit */
				hzm = lzm >> 1;
				lzm = lzm << 63;
				ze -= 63;
			} else {
				hzm = lzm;
				lzm = 0;
				ze -= 64;
			}
		}

		t = 0;
		while ((hzm >> (62 - t)) == 0)
			t++;

		assert(t <= 62);
		if (t) {
			hzm = hzm << t | lzm >> (64 - t);
			lzm = lzm << t;
			ze -= t;
		}
	}

	/*
	 * Move explicit bit from bit 126 to bit 55 since the
	 * ieee754dp_format code expects the mantissa to be
	 * 56 bits wide (53 + 3 rounding bits).
	 */
	srl128(&hzm, &lzm, (126 - 55));

	return ieee754dp_format(zs, ze, lzm);
}

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);
}