Training courses

Kernel and Embedded Linux

Bootlin training courses

Embedded Linux, kernel,
Yocto Project, Buildroot, real-time,
graphics, boot time, debugging...

Bootlin logo

Elixir Cross Referencer

  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
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
/*	$NetBSD: fpu_implode.c,v 1.14 2022/08/28 22:09:26 rin Exp $ */

/*
 * Copyright (c) 1992, 1993
 *	The Regents of the University of California.  All rights reserved.
 *
 * This software was developed by the Computer Systems Engineering group
 * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
 * contributed to Berkeley.
 *
 * All advertising materials mentioning features or use of this software
 * must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Lawrence Berkeley Laboratory.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	@(#)fpu_implode.c	8.1 (Berkeley) 6/11/93
 */

/*
 * FPU subroutines: `implode' internal format numbers into the machine's
 * `packed binary' format.
 */

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: fpu_implode.c,v 1.14 2022/08/28 22:09:26 rin Exp $");

#if defined(_KERNEL_OPT)
#include "opt_sparc_arch.h"
#endif

#include <sys/types.h>
#include <sys/systm.h>

#include <machine/ieee.h>
#include <machine/instr.h>
#include <machine/reg.h>

#include <sparc/fpu/fpu_arith.h>
#include <sparc/fpu/fpu_emu.h>
#include <sparc/fpu/fpu_extern.h>

static int round(struct fpemu *, struct fpn *);
static int toinf(struct fpemu *, int);

/*
 * Round a number (algorithm from Motorola MC68882 manual, modified for
 * our internal format).  Set inexact exception if rounding is required.
 * Return true iff we rounded up.
 *
 * After rounding, we discard the guard and round bits by shifting right
 * 2 bits (a la fpu_shr(), but we do not bother with fp->fp_sticky).
 * This saves effort later.
 *
 * Note that we may leave the value 2.0 in fp->fp_mant; it is the caller's
 * responsibility to fix this if necessary.
 */
static int
round(struct fpemu *fe, struct fpn *fp)
{
	u_int m0, m1, m2, m3;
	int gr, s;

	m0 = fp->fp_mant[0];
	m1 = fp->fp_mant[1];
	m2 = fp->fp_mant[2];
	m3 = fp->fp_mant[3];
	gr = m3 & 3;
	s = fp->fp_sticky;

	/* mant >>= FP_NG */
	m3 = (m3 >> FP_NG) | (m2 << (32 - FP_NG));
	m2 = (m2 >> FP_NG) | (m1 << (32 - FP_NG));
	m1 = (m1 >> FP_NG) | (m0 << (32 - FP_NG));
	m0 >>= FP_NG;

	if ((gr | s) == 0)	/* result is exact: no rounding needed */
		goto rounddown;

	fe->fe_cx |= FSR_NX;	/* inexact */

	/* Go to rounddown to round down; break to round up. */
	switch ((fe->fe_fsr >> FSR_RD_SHIFT) & FSR_RD_MASK) {

	case FSR_RD_RN:
	default:
		/*
		 * Round only if guard is set (gr & 2).  If guard is set,
		 * but round & sticky both clear, then we want to round
		 * but have a tie, so round to even, i.e., add 1 iff odd.
		 */
		if ((gr & 2) == 0)
			goto rounddown;
		if ((gr & 1) || fp->fp_sticky || (m3 & 1))
			break;
		goto rounddown;

	case FSR_RD_RZ:
		/* Round towards zero, i.e., down. */
		goto rounddown;

	case FSR_RD_RM:
		/* Round towards -Inf: up if negative, down if positive. */
		if (fp->fp_sign)
			break;
		goto rounddown;

	case FSR_RD_RP:
		/* Round towards +Inf: up if positive, down otherwise. */
		if (!fp->fp_sign)
			break;
		goto rounddown;
	}

	/* Bump low bit of mantissa, with carry. */
	FPU_ADDS(m3, m3, 1);
	FPU_ADDCS(m2, m2, 0);
	FPU_ADDCS(m1, m1, 0);
	FPU_ADDC(m0, m0, 0);
	fp->fp_mant[0] = m0;
	fp->fp_mant[1] = m1;
	fp->fp_mant[2] = m2;
	fp->fp_mant[3] = m3;
	return (1);

rounddown:
	fp->fp_mant[0] = m0;
	fp->fp_mant[1] = m1;
	fp->fp_mant[2] = m2;
	fp->fp_mant[3] = m3;
	return (0);
}

/*
 * For overflow: return true if overflow is to go to +/-Inf, according
 * to the sign of the overflowing result.  If false, overflow is to go
 * to the largest magnitude value instead.
 */
static int
toinf(struct fpemu *fe, int sign)
{
	int inf;

	/* look at rounding direction */
	switch ((fe->fe_fsr >> FSR_RD_SHIFT) & FSR_RD_MASK) {

	default:
	case FSR_RD_RN:		/* the nearest value is always Inf */
		inf = 1;
		break;

	case FSR_RD_RZ:		/* toward 0 => never towards Inf */
		inf = 0;
		break;

	case FSR_RD_RP:		/* toward +Inf iff positive */
		inf = sign == 0;
		break;

	case FSR_RD_RM:		/* toward -Inf iff negative */
		inf = sign;
		break;
	}
	return (inf);
}

/*
 * fpn -> int (int value returned as return value).
 *
 * N.B.: this conversion always rounds towards zero (this is a peculiarity
 * of the SPARC instruction set).
 */
u_int
fpu_ftoi(struct fpemu *fe, struct fpn *fp)
{
	u_int i;
	int sign, exp;

	sign = fp->fp_sign;
	switch (fp->fp_class) {

	case FPC_ZERO:
		return (0);

	case FPC_NUM:
		/*
		 * If exp >= 2^32, overflow.  Otherwise shift value right
		 * into last mantissa word (this will not exceed 0xffffffff),
		 * shifting any guard and round bits out into the sticky
		 * bit.  Then ``round'' towards zero, i.e., just set an
		 * inexact exception if sticky is set (see round()).
		 * If the result is > 0x80000000, or is positive and equals
		 * 0x80000000, overflow; otherwise the last fraction word
		 * is the result.
		 */
		if ((exp = fp->fp_exp) >= 32)
			break;
		/* NB: the following includes exp < 0 cases */
		if (fpu_shr(fp, FP_NMANT - 1 - exp) != 0)
			fe->fe_cx |= FSR_NX;
		i = fp->fp_mant[3];
		if (i >= ((u_int)0x80000000 + sign))
			break;
		return (sign ? -i : i);

	default:		/* Inf, qNaN, sNaN */
		break;
	}
	/* overflow: replace any inexact exception with invalid */
	fe->fe_cx = (fe->fe_cx & ~FSR_NX) | FSR_NV;
	return (0x7fffffff + sign);
}

#ifdef SUN4U
/*
 * fpn -> extended int (high bits of int value returned as return value).
 *
 * N.B.: this conversion always rounds towards zero (this is a peculiarity
 * of the SPARC instruction set).
 */
u_int
fpu_ftox(struct fpemu *fe, struct fpn *fp, u_int *res)
{
	uint64_t i;
	int sign, exp;

	sign = fp->fp_sign;
	switch (fp->fp_class) {

	case FPC_ZERO:
		res[1] = 0;
		return (0);

	case FPC_NUM:
		/*
		 * If exp >= 2^64, overflow.  Otherwise shift value right
		 * into last mantissa word (this will not exceed 0xffffffffffffffff),
		 * shifting any guard and round bits out into the sticky
		 * bit.  Then ``round'' towards zero, i.e., just set an
		 * inexact exception if sticky is set (see round()).
		 * If the result is > 0x8000000000000000, or is positive and equals
		 * 0x8000000000000000, overflow; otherwise the last fraction word
		 * is the result.
		 */
		if ((exp = fp->fp_exp) >= 64)
			break;
		/* NB: the following includes exp < 0 cases */
		if (fpu_shr(fp, FP_NMANT - 1 - exp) != 0)
			fe->fe_cx |= FSR_NX;
		i = ((uint64_t)fp->fp_mant[2]<<32)|fp->fp_mant[3];
		if (i >= ((uint64_t)0x8000000000000000LL + sign))
			break;
		if (sign) i = -i;
		res[1] = (int)i;
		return (i>>32);

	default:		/* Inf, qNaN, sNaN */
		break;
	}
	/* overflow: replace any inexact exception with invalid */
	fe->fe_cx = (fe->fe_cx & ~FSR_NX) | FSR_NV;
	return (0x7fffffffffffffffLL + sign);
}
#endif /* SUN4U */

/*
 * fpn -> single (32 bit single returned as return value).
 * We assume <= 29 bits in a single-precision fraction (1.f part).
 */
u_int
fpu_ftos(struct fpemu *fe, struct fpn *fp)
{
	u_int sign = fp->fp_sign << 31;
	int exp;

#define	SNG_EXP(e)	((e) << SNG_FRACBITS)	/* makes e an exponent */
#define	SNG_MASK	(SNG_EXP(1) - 1)	/* mask for fraction */

	/* Take care of non-numbers first. */
	if (ISNAN(fp)) {
		/*
		 * Preserve upper bits of NaN, per SPARC V8 appendix N.
		 * Note that fp->fp_mant[0] has the quiet bit set,
		 * even if it is classified as a signalling NaN.
		 */
		(void) fpu_shr(fp, FP_NMANT - 1 - SNG_FRACBITS);
		exp = SNG_EXP_INFNAN;
		goto done;
	}
	if (ISINF(fp))
		return (sign | SNG_EXP(SNG_EXP_INFNAN));
	if (ISZERO(fp))
		return (sign);

	/*
	 * Normals (including subnormals).  Drop all the fraction bits
	 * (including the explicit ``implied'' 1 bit) down into the
	 * single-precision range.  If the number is subnormal, move
	 * the ``implied'' 1 into the explicit range as well, and shift
	 * right to introduce leading zeroes.  Rounding then acts
	 * differently for normals and subnormals: the largest subnormal
	 * may round to the smallest normal (1.0 x 2^minexp), or may
	 * remain subnormal.  In the latter case, signal an underflow
	 * if the result was inexact or if underflow traps are enabled.
	 *
	 * Rounding a normal, on the other hand, always produces another
	 * normal (although either way the result might be too big for
	 * single precision, and cause an overflow).  If rounding a
	 * normal produces 2.0 in the fraction, we need not adjust that
	 * fraction at all, since both 1.0 and 2.0 are zero under the
	 * fraction mask.
	 *
	 * Note that the guard and round bits vanish from the number after
	 * rounding.
	 */
	if ((exp = fp->fp_exp + SNG_EXP_BIAS) <= 0) {	/* subnormal */
		/* -NG for g,r; -SNG_FRACBITS-exp for fraction */
		(void) fpu_shr(fp, FP_NMANT - FP_NG - SNG_FRACBITS - exp);
		if (round(fe, fp) && fp->fp_mant[3] == SNG_EXP(1))
			return (sign | SNG_EXP(1) | 0);
		if ((fe->fe_cx & FSR_NX) ||
		    (fe->fe_fsr & (FSR_UF << FSR_TEM_SHIFT)))
			fe->fe_cx |= FSR_UF;
		return (sign | SNG_EXP(0) | fp->fp_mant[3]);
	}
	/* -FP_NG for g,r; -1 for implied 1; -SNG_FRACBITS for fraction */
	(void) fpu_shr(fp, FP_NMANT - FP_NG - 1 - SNG_FRACBITS);
#ifdef DIAGNOSTIC
	if ((fp->fp_mant[3] & SNG_EXP(1 << FP_NG)) == 0)
		panic("fpu_ftos");
#endif
	if (round(fe, fp) && fp->fp_mant[3] == SNG_EXP(2))
		exp++;
	if (exp >= SNG_EXP_INFNAN) {
		/* overflow to inf or to max single */
		fe->fe_cx |= FSR_OF | FSR_NX;
		if (toinf(fe, sign))
			return (sign | SNG_EXP(SNG_EXP_INFNAN));
		return (sign | SNG_EXP(SNG_EXP_INFNAN - 1) | SNG_MASK);
	}
done:
	/* phew, made it */
	return (sign | SNG_EXP(exp) | (fp->fp_mant[3] & SNG_MASK));
}

/*
 * fpn -> double (32 bit high-order result returned; 32-bit low order result
 * left in res[1]).  Assumes <= 61 bits in double precision fraction.
 *
 * This code mimics fpu_ftos; see it for comments.
 */
u_int
fpu_ftod(struct fpemu *fe, struct fpn *fp, u_int *res)
{
	u_int sign = fp->fp_sign << 31;
	int exp;

#define	DBL_EXP(e)	((e) << (DBL_FRACBITS & 31))
#define	DBL_MASK	(DBL_EXP(1) - 1)

	if (ISNAN(fp)) {
		(void) fpu_shr(fp, FP_NMANT - 1 - DBL_FRACBITS);
		exp = DBL_EXP_INFNAN;
		goto done;
	}
	if (ISINF(fp)) {
		sign |= DBL_EXP(DBL_EXP_INFNAN);
		goto zero;
	}
	if (ISZERO(fp)) {
zero:		res[1] = 0;
		return (sign);
	}

	if ((exp = fp->fp_exp + DBL_EXP_BIAS) <= 0) {
		(void) fpu_shr(fp, FP_NMANT - FP_NG - DBL_FRACBITS - exp);
		if (round(fe, fp) && fp->fp_mant[2] == DBL_EXP(1)) {
			res[1] = 0;
			return (sign | DBL_EXP(1) | 0);
		}
		if ((fe->fe_cx & FSR_NX) ||
		    (fe->fe_fsr & (FSR_UF << FSR_TEM_SHIFT)))
			fe->fe_cx |= FSR_UF;
		exp = 0;
		goto done;
	}
	(void) fpu_shr(fp, FP_NMANT - FP_NG - 1 - DBL_FRACBITS);
	if (round(fe, fp) && fp->fp_mant[2] == DBL_EXP(2))
		exp++;
	if (exp >= DBL_EXP_INFNAN) {
		fe->fe_cx |= FSR_OF | FSR_NX;
		if (toinf(fe, sign)) {
			res[1] = 0;
			return (sign | DBL_EXP(DBL_EXP_INFNAN) | 0);
		}
		res[1] = ~0;
		return (sign | DBL_EXP(DBL_EXP_INFNAN) | DBL_MASK);
	}
done:
	res[1] = fp->fp_mant[3];
	return (sign | DBL_EXP(exp) | (fp->fp_mant[2] & DBL_MASK));
}

/*
 * fpn -> extended (32 bit high-order result returned; low-order fraction
 * words left in res[1]..res[3]).  Like ftod, which is like ftos ... but
 * our internal format *is* extended precision, plus 2 bits for guard/round,
 * so we can avoid a small bit of work.
 */
u_int
fpu_ftoq(struct fpemu *fe, struct fpn *fp, u_int *res)
{
	u_int sign = fp->fp_sign << 31;
	int exp;

#define	EXT_EXP(e)	((e) << (EXT_FRACBITS & 31))
#define	EXT_MASK	(EXT_EXP(1) - 1)

	if (ISNAN(fp)) {
		(void) fpu_shr(fp, 2);	/* since we are not rounding */
		exp = EXT_EXP_INFNAN;
		goto done;
	}
	if (ISINF(fp)) {
		sign |= EXT_EXP(EXT_EXP_INFNAN);
		goto zero;
	}
	if (ISZERO(fp)) {
zero:		res[1] = res[2] = res[3] = 0;
		return (sign);
	}

	if ((exp = fp->fp_exp + EXT_EXP_BIAS) <= 0) {
		(void) fpu_shr(fp, FP_NMANT - FP_NG - EXT_FRACBITS - exp);
		if (round(fe, fp) && fp->fp_mant[0] == EXT_EXP(1)) {
			res[1] = res[2] = res[3] = 0;
			return (sign | EXT_EXP(1) | 0);
		}
		if ((fe->fe_cx & FSR_NX) ||
		    (fe->fe_fsr & (FSR_UF << FSR_TEM_SHIFT)))
			fe->fe_cx |= FSR_UF;
		exp = 0;
		goto done;
	}
	/* Since internal == extended, no need to shift here. */
	if (round(fe, fp) && fp->fp_mant[0] == EXT_EXP(2))
		exp++;
	if (exp >= EXT_EXP_INFNAN) {
		fe->fe_cx |= FSR_OF | FSR_NX;
		if (toinf(fe, sign)) {
			res[1] = res[2] = res[3] = 0;
			return (sign | EXT_EXP(EXT_EXP_INFNAN) | 0);
		}
		res[1] = res[2] = res[3] = ~0;
		return (sign | EXT_EXP(EXT_EXP_INFNAN) | EXT_MASK);
	}
done:
	res[1] = fp->fp_mant[1];
	res[2] = fp->fp_mant[2];
	res[3] = fp->fp_mant[3];
	return (sign | EXT_EXP(exp) | (fp->fp_mant[0] & EXT_MASK));
}

/*
 * Implode an fpn, writing the result into the given space.
 */
void
fpu_implode(struct fpemu *fe, struct fpn *fp, int type, u_int *space)
{

	DPRINTF(FPE_REG, ("\n imploding: "));
	DUMPFPN(FPE_REG, fp);
	DPRINTF(FPE_REG, ("\n"));

	switch (type) {

#ifdef SUN4U
	case FTYPE_LNG:
		space[0] = fpu_ftox(fe, fp, space);
		break;
#endif /* SUN4U */

	case FTYPE_INT:
		space[0] = fpu_ftoi(fe, fp);
		break;

	case FTYPE_SNG:
		space[0] = fpu_ftos(fe, fp);
		break;

	case FTYPE_DBL:
		space[0] = fpu_ftod(fe, fp, space);
		break;

	case FTYPE_EXT:
		/* funky rounding precision options ?? */
		space[0] = fpu_ftoq(fe, fp, space);
		break;

	default:
		panic("fpu_implode");
	}
#ifdef SUN4U
	DPRINTF(FPE_REG, ("fpu_implode: %x %x %x %x\n",
		space[0], space[1], space[2], space[3]));
#else
	DPRINTF(FPE_REG, ("fpu_implode: %x %x\n",
		space[0], space[1]));
#endif
}