Commit | Line | Data |
---|---|---|
9655c439 AC |
1 | /* This is a software floating point library which can be used instead |
2 | of the floating point routines in libgcc1.c for targets without | |
3 | hardware floating point. */ | |
3971886a | 4 | |
f90b720b | 5 | /* Copyright (C) 1994,1997 Free Software Foundation, Inc. |
3971886a | 6 | |
f90b720b AC |
7 | This file is free software; you can redistribute it and/or modify it |
8 | under the terms of the GNU General Public License as published by the | |
9 | Free Software Foundation; either version 2, or (at your option) any | |
10 | later version. | |
3971886a | 11 | |
f90b720b AC |
12 | In addition to the permissions in the GNU General Public License, the |
13 | Free Software Foundation gives you unlimited permission to link the | |
14 | compiled version of this file with other programs, and to distribute | |
15 | those programs without any restriction coming from the use of this | |
16 | file. (The General Public License restrictions do apply in other | |
17 | respects; for example, they cover modification of the file, and | |
18 | distribution when not linked into another program.) | |
3971886a | 19 | |
f90b720b AC |
20 | This file is distributed in the hope that it will be useful, but |
21 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
22 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
23 | General Public License for more details. | |
3971886a | 24 | |
f90b720b AC |
25 | You should have received a copy of the GNU General Public License |
26 | along with this program; see the file COPYING. If not, write to | |
27 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */ | |
28 | ||
29 | /* As a special exception, if you link this library with other files, | |
30 | some of which are compiled with GCC, to produce an executable, | |
31 | this library does not by itself cause the resulting executable | |
32 | to be covered by the GNU General Public License. | |
33 | This exception does not however invalidate any other reasons why | |
34 | the executable file might be covered by the GNU General Public License. */ | |
35 | ||
36 | /* This implements IEEE 754 format arithmetic, but does not provide a | |
37 | mechanism for setting the rounding mode, or for generating or handling | |
38 | exceptions. | |
39 | ||
40 | The original code by Steve Chamberlain, hacked by Mark Eichin and Jim | |
41 | Wilson, all of Cygnus Support. */ | |
3971886a AC |
42 | |
43 | ||
07b4c0a6 AC |
44 | #ifndef SIM_FPU_C |
45 | #define SIM_FPU_C | |
3971886a | 46 | |
9655c439 | 47 | #include "sim-basics.h" |
3971886a | 48 | #include "sim-fpu.h" |
9655c439 AC |
49 | |
50 | #include "sim-io.h" | |
07b4c0a6 | 51 | #include "sim-assert.h" |
3971886a | 52 | |
07b4c0a6 | 53 | |
9655c439 | 54 | /* Debugging support. */ |
07b4c0a6 | 55 | |
9655c439 AC |
56 | static void |
57 | print_bits (unsigned64 x, | |
58 | int msbit, | |
59 | sim_fpu_print_func print, | |
60 | void *arg) | |
61 | { | |
62 | unsigned64 bit = LSBIT64 (msbit); | |
63 | int i = 4; | |
64 | while (bit) | |
65 | { | |
66 | if (i == 0) | |
67 | print (arg, ","); | |
68 | if ((x & bit)) | |
69 | print (arg, "1"); | |
70 | else | |
71 | print (arg, "0"); | |
72 | bit >>= 1; | |
73 | i = (i + 1) % 4; | |
74 | } | |
75 | } | |
07b4c0a6 | 76 | |
07b4c0a6 | 77 | |
07b4c0a6 | 78 | |
9655c439 | 79 | /* Quick and dirty conversion between a host double and host 64bit int */ |
07b4c0a6 | 80 | |
9655c439 AC |
81 | typedef union { |
82 | double d; | |
83 | unsigned64 i; | |
84 | } sim_fpu_map; | |
07b4c0a6 | 85 | |
07b4c0a6 | 86 | |
9655c439 AC |
87 | /* A packed IEEE floating point number. |
88 | ||
89 | Form is <SIGN:1><BIASEDEXP:NR_EXPBITS><FRAC:NR_FRACBITS> for both | |
90 | 32 and 64 bit numbers. This number is interpreted as: | |
91 | ||
92 | Normalized (0 < BIASEDEXP && BIASEDEXP < EXPMAX): | |
93 | (sign ? '-' : '+') 1.<FRAC> x 2 ^ (BIASEDEXP - EXPBIAS) | |
94 | ||
95 | Denormalized (0 == BIASEDEXP && FRAC != 0): | |
96 | (sign ? "-" : "+") 0.<FRAC> x 2 ^ (- EXPBIAS) | |
97 | ||
98 | Zero (0 == BIASEDEXP && FRAC == 0): | |
99 | (sign ? "-" : "+") 0.0 | |
100 | ||
101 | Infinity (BIASEDEXP == EXPMAX && FRAC == 0): | |
102 | (sign ? "-" : "+") "infinity" | |
103 | ||
104 | SignalingNaN (BIASEDEXP == EXPMAX && FRAC > 0 && FRAC < QUIET_NAN): | |
105 | SNaN.FRAC | |
106 | ||
107 | QuietNaN (BIASEDEXP == EXPMAX && FRAC > 0 && FRAC > QUIET_NAN): | |
108 | QNaN.FRAC | |
109 | ||
110 | */ | |
111 | ||
112 | #define NR_EXPBITS (is_double ? 11 : 8) | |
390ffa89 | 113 | #define NR_FRACBITS (is_double ? 52 : 23) |
9655c439 AC |
114 | #define SIGNBIT (is_double ? MSBIT64 (0) : MSBIT64 (32)) |
115 | ||
390ffa89 AC |
116 | #define EXPMAX32 (255) |
117 | #define EXMPAX64 (2047) | |
118 | #define EXPMAX ((unsigned) (is_double ? EXMPAX64 : EXPMAX32)) | |
119 | ||
120 | #define EXPBIAS32 (127) | |
121 | #define EXPBIAS64 (1023) | |
122 | #define EXPBIAS (is_double ? EXPBIAS64 : EXPBIAS32) | |
9655c439 AC |
123 | |
124 | #define QUIET_NAN LSBIT64 (NR_FRACBITS - 1) | |
125 | ||
126 | ||
127 | ||
128 | /* An unpacked floating point number. | |
129 | ||
130 | When unpacked, the fraction of both a 32 and 64 bit floating point | |
131 | number is stored using the same format: | |
132 | ||
133 | 64 bit - <IMPLICIT_1:1><FRACBITS:52><GUARDS:8><PAD:00> | |
134 | 32 bit - <IMPLICIT_1:1><FRACBITS:23><GUARDS:7><PAD:30> */ | |
135 | ||
390ffa89 AC |
136 | #define NR_PAD32 (30) |
137 | #define NR_PAD64 (0) | |
138 | #define NR_PAD (is_double ? NR_PAD64 : NR_PAD32) | |
139 | #define PADMASK (is_double ? 0 : LSMASK64 (NR_PAD32 - 1, 0)) | |
140 | ||
141 | #define NR_GUARDS32 (7 + NR_PAD32) | |
278bda40 | 142 | #define NR_GUARDS64 (8 + NR_PAD64) |
390ffa89 | 143 | #define NR_GUARDS (is_double ? NR_GUARDS64 : NR_GUARDS32) |
9655c439 | 144 | #define GUARDMASK LSMASK64 (NR_GUARDS - 1, 0) |
390ffa89 | 145 | |
9655c439 AC |
146 | #define GUARDMSB LSBIT64 (NR_GUARDS - 1) |
147 | #define GUARDLSB LSBIT64 (NR_PAD) | |
148 | #define GUARDROUND LSMASK64 (NR_GUARDS - 2, 0) | |
149 | ||
150 | #define NR_FRAC_GUARD (60) | |
151 | #define IMPLICIT_1 LSBIT64 (NR_FRAC_GUARD) | |
152 | #define IMPLICIT_2 LSBIT64 (NR_FRAC_GUARD + 1) | |
153 | #define IMPLICIT_4 LSBIT64 (NR_FRAC_GUARD + 2) | |
154 | #define NR_SPARE 2 | |
155 | ||
156 | #define FRAC32MASK LSMASK64 (63, NR_FRAC_GUARD - 32 + 1) | |
157 | ||
158 | #define NORMAL_EXPMIN (-(EXPBIAS)+1) | |
390ffa89 AC |
159 | |
160 | #define NORMAL_EXPMAX32 (EXPBIAS32) | |
161 | #define NORMAL_EXPMAX64 (EXPBIAS64) | |
9655c439 | 162 | #define NORMAL_EXPMAX (EXPBIAS) |
07b4c0a6 AC |
163 | |
164 | ||
9655c439 AC |
165 | /* Integer constants */ |
166 | ||
167 | #define MAX_INT32 ((signed64) LSMASK64 (30, 0)) | |
168 | #define MAX_UINT32 LSMASK64 (31, 0) | |
169 | #define MIN_INT32 ((signed64) LSMASK64 (63, 31)) | |
170 | ||
171 | #define MAX_INT64 ((signed64) LSMASK64 (62, 0)) | |
172 | #define MAX_UINT64 LSMASK64 (63, 0) | |
173 | #define MIN_INT64 ((signed64) LSMASK64 (63, 63)) | |
174 | ||
175 | #define MAX_INT (is_64bit ? MAX_INT64 : MAX_INT32) | |
176 | #define MIN_INT (is_64bit ? MIN_INT64 : MIN_INT32) | |
177 | #define MAX_UINT (is_64bit ? MAX_UINT64 : MAX_UINT32) | |
178 | #define NR_INTBITS (is_64bit ? 64 : 32) | |
179 | ||
bdfe5c04 | 180 | /* Squeese an unpacked sim_fpu struct into a 32/64 bit integer */ |
07b4c0a6 | 181 | STATIC_INLINE_SIM_FPU (unsigned64) |
9655c439 AC |
182 | pack_fpu (const sim_fpu *src, |
183 | int is_double) | |
07b4c0a6 | 184 | { |
07b4c0a6 | 185 | int sign; |
9655c439 AC |
186 | unsigned64 exp; |
187 | unsigned64 fraction; | |
188 | unsigned64 packed; | |
07b4c0a6 AC |
189 | |
190 | switch (src->class) | |
191 | { | |
07b4c0a6 AC |
192 | /* create a NaN */ |
193 | case sim_fpu_class_qnan: | |
9655c439 AC |
194 | sign = src->sign; |
195 | exp = EXPMAX; | |
196 | /* force fraction to correct class */ | |
197 | fraction = src->fraction; | |
198 | fraction >>= NR_GUARDS; | |
199 | fraction |= QUIET_NAN; | |
200 | break; | |
07b4c0a6 | 201 | case sim_fpu_class_snan: |
9655c439 | 202 | sign = src->sign; |
07b4c0a6 | 203 | exp = EXPMAX; |
9655c439 | 204 | /* force fraction to correct class */ |
07b4c0a6 | 205 | fraction = src->fraction; |
9655c439 AC |
206 | fraction >>= NR_GUARDS; |
207 | fraction &= ~QUIET_NAN; | |
07b4c0a6 AC |
208 | break; |
209 | case sim_fpu_class_infinity: | |
210 | sign = src->sign; | |
211 | exp = EXPMAX; | |
212 | fraction = 0; | |
213 | break; | |
214 | case sim_fpu_class_zero: | |
215 | sign = src->sign; | |
216 | exp = 0; | |
217 | fraction = 0; | |
218 | break; | |
219 | case sim_fpu_class_number: | |
bdfe5c04 | 220 | case sim_fpu_class_denorm: |
9655c439 AC |
221 | ASSERT (src->fraction >= IMPLICIT_1); |
222 | ASSERT (src->fraction < IMPLICIT_2); | |
07b4c0a6 AC |
223 | if (src->normal_exp < NORMAL_EXPMIN) |
224 | { | |
225 | /* This number's exponent is too low to fit into the bits | |
9655c439 AC |
226 | available in the number We'll denormalize the number by |
227 | storing zero in the exponent and shift the fraction to | |
228 | the right to make up for it. */ | |
229 | int nr_shift = NORMAL_EXPMIN - src->normal_exp; | |
230 | if (nr_shift > NR_FRACBITS) | |
07b4c0a6 | 231 | { |
9655c439 AC |
232 | /* underflow, just make the number zero */ |
233 | sign = src->sign; | |
234 | exp = 0; | |
07b4c0a6 AC |
235 | fraction = 0; |
236 | } | |
237 | else | |
238 | { | |
9655c439 AC |
239 | sign = src->sign; |
240 | exp = 0; | |
07b4c0a6 | 241 | /* Shift by the value */ |
9655c439 AC |
242 | fraction = src->fraction; |
243 | fraction >>= NR_GUARDS; | |
244 | fraction >>= nr_shift; | |
07b4c0a6 AC |
245 | } |
246 | } | |
9655c439 | 247 | else if (src->normal_exp > NORMAL_EXPMAX) |
07b4c0a6 AC |
248 | { |
249 | /* Infinity */ | |
250 | sign = src->sign; | |
251 | exp = EXPMAX; | |
252 | fraction = 0; | |
253 | } | |
254 | else | |
255 | { | |
07b4c0a6 | 256 | exp = (src->normal_exp + EXPBIAS); |
9655c439 AC |
257 | sign = src->sign; |
258 | fraction = src->fraction; | |
259 | /* FIXME: Need to round according to WITH_SIM_FPU_ROUNDING | |
260 | or some such */ | |
261 | /* Round to nearest: If the guard bits are the all zero, but | |
262 | the first, then we're half way between two numbers, | |
263 | choose the one which makes the lsb of the answer 0. */ | |
264 | if ((fraction & GUARDMASK) == GUARDMSB) | |
07b4c0a6 | 265 | { |
9655c439 AC |
266 | if ((fraction & (GUARDMSB << 1))) |
267 | fraction += (GUARDMSB << 1); | |
07b4c0a6 AC |
268 | } |
269 | else | |
270 | { | |
9655c439 AC |
271 | /* Add a one to the guards to force round to nearest */ |
272 | fraction += GUARDROUND; | |
07b4c0a6 | 273 | } |
9655c439 | 274 | if ((fraction & IMPLICIT_2)) /* rounding resulted in carry */ |
07b4c0a6 | 275 | { |
07b4c0a6 | 276 | exp += 1; |
9655c439 | 277 | fraction >>= 1; |
07b4c0a6 | 278 | } |
9655c439 AC |
279 | fraction >>= NR_GUARDS; |
280 | /* When exp == EXPMAX (overflow from carry) fraction must | |
281 | have been made zero */ | |
282 | ASSERT ((exp == EXPMAX) <= ((fraction & ~IMPLICIT_1) == 0)); | |
07b4c0a6 | 283 | } |
9655c439 AC |
284 | break; |
285 | default: | |
286 | abort (); | |
07b4c0a6 AC |
287 | } |
288 | ||
9655c439 AC |
289 | packed = ((sign ? SIGNBIT : 0) |
290 | | (exp << NR_FRACBITS) | |
291 | | LSMASKED64 (fraction, NR_FRACBITS - 1, 0)); | |
292 | ||
293 | /* trace operation */ | |
294 | #if 0 | |
295 | if (is_double) | |
296 | { | |
297 | } | |
298 | else | |
299 | { | |
300 | printf ("pack_fpu: "); | |
301 | printf ("-> %c%0lX.%06lX\n", | |
302 | LSMASKED32 (packed, 31, 31) ? '8' : '0', | |
303 | (long) LSEXTRACTED32 (packed, 30, 23), | |
304 | (long) LSEXTRACTED32 (packed, 23 - 1, 0)); | |
305 | } | |
306 | #endif | |
307 | ||
308 | return packed; | |
07b4c0a6 AC |
309 | } |
310 | ||
311 | ||
bdfe5c04 | 312 | /* Unpack a 32/64 bit integer into a sim_fpu structure */ |
07b4c0a6 | 313 | STATIC_INLINE_SIM_FPU (void) |
9655c439 | 314 | unpack_fpu (sim_fpu *dst, unsigned64 packed, int is_double) |
07b4c0a6 | 315 | { |
9655c439 AC |
316 | unsigned64 fraction = LSMASKED64 (packed, NR_FRACBITS - 1, 0); |
317 | unsigned exp = LSEXTRACTED64 (packed, NR_EXPBITS + NR_FRACBITS - 1, NR_FRACBITS); | |
318 | int sign = (packed & SIGNBIT) != 0; | |
07b4c0a6 AC |
319 | |
320 | if (exp == 0) | |
321 | { | |
322 | /* Hmm. Looks like 0 */ | |
323 | if (fraction == 0) | |
324 | { | |
325 | /* tastes like zero */ | |
326 | dst->class = sim_fpu_class_zero; | |
9655c439 | 327 | dst->sign = sign; |
07b4c0a6 AC |
328 | } |
329 | else | |
330 | { | |
331 | /* Zero exponent with non zero fraction - it's denormalized, | |
332 | so there isn't a leading implicit one - we'll shift it so | |
333 | it gets one. */ | |
334 | dst->normal_exp = exp - EXPBIAS + 1; | |
bdfe5c04 | 335 | dst->class = sim_fpu_class_denorm; |
9655c439 AC |
336 | dst->sign = sign; |
337 | fraction <<= NR_GUARDS; | |
07b4c0a6 AC |
338 | while (fraction < IMPLICIT_1) |
339 | { | |
340 | fraction <<= 1; | |
341 | dst->normal_exp--; | |
342 | } | |
9655c439 | 343 | dst->fraction = fraction; |
07b4c0a6 AC |
344 | } |
345 | } | |
346 | else if (exp == EXPMAX) | |
347 | { | |
348 | /* Huge exponent*/ | |
349 | if (fraction == 0) | |
350 | { | |
351 | /* Attached to a zero fraction - means infinity */ | |
352 | dst->class = sim_fpu_class_infinity; | |
9655c439 AC |
353 | dst->sign = sign; |
354 | /* dst->normal_exp = EXPBIAS; */ | |
355 | /* dst->fraction = 0; */ | |
07b4c0a6 AC |
356 | } |
357 | else | |
358 | { | |
9655c439 AC |
359 | /* Non zero fraction, means NaN */ |
360 | dst->sign = sign; | |
361 | dst->fraction = (fraction << NR_GUARDS); | |
362 | if (fraction >= QUIET_NAN) | |
363 | dst->class = sim_fpu_class_qnan; | |
07b4c0a6 | 364 | else |
9655c439 | 365 | dst->class = sim_fpu_class_snan; |
07b4c0a6 AC |
366 | } |
367 | } | |
368 | else | |
369 | { | |
370 | /* Nothing strange about this number */ | |
07b4c0a6 | 371 | dst->class = sim_fpu_class_number; |
9655c439 AC |
372 | dst->sign = sign; |
373 | dst->fraction = ((fraction << NR_GUARDS) | IMPLICIT_1); | |
374 | dst->normal_exp = exp - EXPBIAS; | |
07b4c0a6 AC |
375 | } |
376 | ||
9655c439 AC |
377 | /* trace operation */ |
378 | #if 0 | |
379 | if (is_double) | |
380 | { | |
381 | } | |
382 | else | |
383 | { | |
384 | printf ("unpack_fpu: %c%02lX.%06lX ->\n", | |
385 | LSMASKED32 (packed, 31, 31) ? '8' : '0', | |
386 | (long) LSEXTRACTED32 (packed, 30, 23), | |
387 | (long) LSEXTRACTED32 (packed, 23 - 1, 0)); | |
388 | } | |
389 | #endif | |
390 | ||
07b4c0a6 | 391 | /* sanity checks */ |
07b4c0a6 | 392 | { |
9655c439 AC |
393 | sim_fpu_map val; |
394 | val.i = pack_fpu (dst, 1); | |
07b4c0a6 AC |
395 | if (is_double) |
396 | { | |
9655c439 | 397 | ASSERT (val.i == packed); |
07b4c0a6 AC |
398 | } |
399 | else | |
400 | { | |
401 | unsigned32 val = pack_fpu (dst, 0); | |
9655c439 | 402 | unsigned32 org = packed; |
07b4c0a6 AC |
403 | ASSERT (val == org); |
404 | } | |
405 | } | |
406 | } | |
407 | ||
07b4c0a6 | 408 | |
bdfe5c04 | 409 | /* Convert a floating point into an integer */ |
07b4c0a6 | 410 | STATIC_INLINE_SIM_FPU (int) |
9655c439 AC |
411 | fpu2i (signed64 *i, |
412 | const sim_fpu *s, | |
413 | int is_64bit, | |
414 | sim_fpu_round round) | |
07b4c0a6 | 415 | { |
9655c439 AC |
416 | unsigned64 tmp; |
417 | int shift; | |
418 | int status = 0; | |
419 | if (sim_fpu_is_zero (s)) | |
07b4c0a6 | 420 | { |
9655c439 | 421 | *i = 0; |
07b4c0a6 AC |
422 | return 0; |
423 | } | |
9655c439 | 424 | if (sim_fpu_is_snan (s)) |
07b4c0a6 | 425 | { |
9655c439 AC |
426 | *i = MIN_INT; /* FIXME */ |
427 | return sim_fpu_status_invalid_cvi; | |
428 | } | |
429 | if (sim_fpu_is_qnan (s)) | |
430 | { | |
431 | *i = MIN_INT; /* FIXME */ | |
432 | return sim_fpu_status_invalid_cvi; | |
433 | } | |
434 | /* map infinity onto MAX_INT... */ | |
435 | if (sim_fpu_is_infinity (s)) | |
436 | { | |
437 | *i = s->sign ? MIN_INT : MAX_INT; | |
438 | return sim_fpu_status_invalid_cvi; | |
439 | } | |
440 | /* it is a number, but a small one */ | |
441 | if (s->normal_exp < 0) | |
442 | { | |
443 | *i = 0; | |
444 | return sim_fpu_status_inexact; | |
07b4c0a6 | 445 | } |
9655c439 AC |
446 | /* Is the floating point MIN_INT or just close? */ |
447 | if (s->sign && s->normal_exp == (NR_INTBITS - 1)) | |
448 | { | |
449 | *i = MIN_INT; | |
450 | ASSERT (s->fraction >= IMPLICIT_1); | |
451 | if (s->fraction == IMPLICIT_1) | |
452 | return 0; /* exact */ | |
453 | if (is_64bit) /* can't round */ | |
454 | return sim_fpu_status_invalid_cvi; /* must be overflow */ | |
455 | /* For a 32bit with MAX_INT, rounding is possible */ | |
456 | switch (round) | |
457 | { | |
458 | case sim_fpu_round_default: | |
459 | abort (); | |
460 | case sim_fpu_round_zero: | |
461 | if ((s->fraction & FRAC32MASK) != IMPLICIT_1) | |
462 | return sim_fpu_status_invalid_cvi; | |
463 | else | |
464 | return sim_fpu_status_inexact; | |
465 | break; | |
466 | case sim_fpu_round_near: | |
467 | { | |
468 | if ((s->fraction & FRAC32MASK) != IMPLICIT_1) | |
469 | return sim_fpu_status_invalid_cvi; | |
470 | else if ((s->fraction & !FRAC32MASK) >= (~FRAC32MASK >> 1)) | |
471 | return sim_fpu_status_invalid_cvi; | |
472 | else | |
473 | return sim_fpu_status_inexact; | |
474 | } | |
475 | case sim_fpu_round_up: | |
476 | if ((s->fraction & FRAC32MASK) == IMPLICIT_1) | |
477 | return sim_fpu_status_inexact; | |
478 | else | |
479 | return sim_fpu_status_invalid_cvi; | |
480 | case sim_fpu_round_down: | |
481 | return sim_fpu_status_invalid_cvi; | |
482 | } | |
483 | } | |
484 | /* Would right shifting result in the FRAC being shifted into | |
485 | (through) the integer's sign bit? */ | |
486 | if (s->normal_exp > (NR_INTBITS - 2)) | |
487 | { | |
488 | *i = s->sign ? MIN_INT : MAX_INT; | |
489 | return sim_fpu_status_invalid_cvi; | |
490 | } | |
491 | /* normal number shift it into place */ | |
492 | tmp = s->fraction; | |
493 | shift = (s->normal_exp - (NR_FRAC_GUARD)); | |
494 | if (shift > 0) | |
495 | { | |
496 | tmp <<= shift; | |
497 | } | |
498 | else | |
499 | { | |
500 | shift = -shift; | |
501 | if (tmp & ((SIGNED64 (1) << shift) - 1)) | |
502 | status |= sim_fpu_status_inexact; | |
503 | tmp >>= shift; | |
504 | } | |
505 | *i = s->sign ? (-tmp) : (tmp); | |
506 | return status; | |
07b4c0a6 AC |
507 | } |
508 | ||
bdfe5c04 | 509 | /* convert an integer into a floating point */ |
07b4c0a6 | 510 | STATIC_INLINE_SIM_FPU (int) |
9655c439 | 511 | i2fpu (sim_fpu *f, signed64 i, int is_64bit) |
07b4c0a6 | 512 | { |
9655c439 AC |
513 | int status = 0; |
514 | if (i == 0) | |
07b4c0a6 | 515 | { |
9655c439 AC |
516 | f->class = sim_fpu_class_zero; |
517 | f->sign = 0; | |
07b4c0a6 | 518 | } |
9655c439 AC |
519 | else |
520 | { | |
521 | f->class = sim_fpu_class_number; | |
522 | f->sign = (i < 0); | |
523 | f->normal_exp = NR_FRAC_GUARD; | |
07b4c0a6 | 524 | |
9655c439 AC |
525 | if (f->sign) |
526 | { | |
527 | /* Special case for minint, since there is no corresponding | |
528 | +ve integer representation for it */ | |
529 | if (i == MIN_INT) | |
530 | { | |
531 | f->fraction = IMPLICIT_1; | |
532 | f->normal_exp = NR_INTBITS - 1; | |
533 | } | |
534 | else | |
535 | f->fraction = (-i); | |
536 | } | |
537 | else | |
538 | f->fraction = i; | |
07b4c0a6 | 539 | |
9655c439 AC |
540 | if (f->fraction >= IMPLICIT_2) |
541 | { | |
542 | do | |
543 | { | |
544 | f->fraction >>= 1; | |
545 | f->normal_exp += 1; | |
546 | } | |
547 | while (f->fraction >= IMPLICIT_2); | |
548 | } | |
549 | else if (f->fraction < IMPLICIT_1) | |
550 | { | |
551 | do | |
552 | { | |
553 | f->fraction <<= 1; | |
554 | f->normal_exp -= 1; | |
555 | } | |
556 | while (f->fraction < IMPLICIT_1); | |
557 | } | |
07b4c0a6 | 558 | } |
07b4c0a6 | 559 | |
9655c439 AC |
560 | /* trace operation */ |
561 | #if 0 | |
562 | { | |
563 | printf ("i2fpu: 0x%08lX ->\n", (long) i); | |
564 | } | |
565 | #endif | |
566 | ||
567 | /* sanity check */ | |
568 | { | |
569 | signed64 val; | |
570 | fpu2i (&val, f, is_64bit, sim_fpu_round_zero); | |
571 | if (i >= MIN_INT32 && i <= MAX_INT32) | |
572 | { | |
573 | ASSERT (val == i); | |
574 | } | |
575 | } | |
07b4c0a6 | 576 | |
9655c439 | 577 | return status; |
07b4c0a6 AC |
578 | } |
579 | ||
580 | ||
bdfe5c04 | 581 | /* Convert a floating point into an integer */ |
9655c439 AC |
582 | STATIC_INLINE_SIM_FPU (int) |
583 | fpu2u (unsigned64 *u, const sim_fpu *s, int is_64bit) | |
07b4c0a6 | 584 | { |
9655c439 | 585 | const int is_double = 1; |
07b4c0a6 | 586 | unsigned64 tmp; |
9655c439 AC |
587 | int shift; |
588 | if (sim_fpu_is_zero (s)) | |
589 | { | |
590 | *u = 0; | |
591 | return 0; | |
592 | } | |
593 | if (sim_fpu_is_nan (s)) | |
594 | { | |
595 | *u = 0; | |
596 | return 0; | |
597 | } | |
598 | /* it is a negative number */ | |
599 | if (s->sign) | |
600 | { | |
601 | *u = 0; | |
602 | return 0; | |
603 | } | |
604 | /* get reasonable MAX_USI_INT... */ | |
605 | if (sim_fpu_is_infinity (s)) | |
606 | { | |
607 | *u = MAX_UINT; | |
608 | return 0; | |
609 | } | |
07b4c0a6 | 610 | /* it is a number, but a small one */ |
9655c439 AC |
611 | if (s->normal_exp < 0) |
612 | { | |
613 | *u = 0; | |
614 | return 0; | |
615 | } | |
616 | /* overflow */ | |
617 | if (s->normal_exp > (NR_INTBITS - 1)) | |
618 | { | |
619 | *u = MAX_UINT; | |
620 | return 0; | |
621 | } | |
622 | /* normal number */ | |
623 | tmp = (s->fraction & ~PADMASK); | |
624 | shift = (s->normal_exp - (NR_FRACBITS + NR_GUARDS)); | |
625 | if (shift > 0) | |
626 | { | |
627 | tmp <<= shift; | |
628 | } | |
07b4c0a6 | 629 | else |
9655c439 AC |
630 | { |
631 | shift = -shift; | |
632 | tmp >>= shift; | |
633 | } | |
634 | *u = tmp; | |
635 | return 0; | |
07b4c0a6 AC |
636 | } |
637 | ||
bdfe5c04 | 638 | /* Convert an unsigned integer into a floating point */ |
9655c439 AC |
639 | STATIC_INLINE_SIM_FPU (int) |
640 | u2fpu (sim_fpu *f, unsigned64 u, int is_64bit) | |
07b4c0a6 | 641 | { |
9655c439 AC |
642 | if (u == 0) |
643 | { | |
644 | f->class = sim_fpu_class_zero; | |
645 | f->sign = 0; | |
646 | } | |
07b4c0a6 | 647 | else |
9655c439 AC |
648 | { |
649 | f->class = sim_fpu_class_number; | |
650 | f->sign = 0; | |
651 | f->normal_exp = NR_FRAC_GUARD; | |
652 | f->fraction = u; | |
653 | ||
654 | while (f->fraction < IMPLICIT_1) | |
655 | { | |
656 | f->fraction <<= 1; | |
657 | f->normal_exp -= 1; | |
658 | } | |
659 | } | |
660 | return 0; | |
07b4c0a6 AC |
661 | } |
662 | ||
663 | ||
d24f06ee AC |
664 | /* register <-> sim_fpu */ |
665 | ||
9655c439 AC |
666 | INLINE_SIM_FPU (void) |
667 | sim_fpu_32to (sim_fpu *f, unsigned32 s) | |
d24f06ee | 668 | { |
9655c439 | 669 | unpack_fpu (f, s, 0); |
d24f06ee AC |
670 | } |
671 | ||
672 | ||
9655c439 AC |
673 | INLINE_SIM_FPU (void) |
674 | sim_fpu_232to (sim_fpu *f, unsigned32 h, unsigned32 l) | |
d24f06ee | 675 | { |
9655c439 AC |
676 | unsigned64 s = h; |
677 | s = (s << 32) | l; | |
678 | unpack_fpu (f, s, 1); | |
d24f06ee AC |
679 | } |
680 | ||
681 | ||
9655c439 AC |
682 | INLINE_SIM_FPU (void) |
683 | sim_fpu_64to (sim_fpu *f, unsigned64 s) | |
d24f06ee | 684 | { |
9655c439 | 685 | unpack_fpu (f, s, 1); |
d24f06ee AC |
686 | } |
687 | ||
688 | ||
9655c439 AC |
689 | INLINE_SIM_FPU (void) |
690 | sim_fpu_to32 (unsigned32 *s, | |
691 | const sim_fpu *f) | |
d24f06ee | 692 | { |
9655c439 | 693 | *s = pack_fpu (f, 0); |
d24f06ee AC |
694 | } |
695 | ||
696 | ||
9655c439 AC |
697 | INLINE_SIM_FPU (void) |
698 | sim_fpu_to232 (unsigned32 *h, unsigned32 *l, | |
699 | const sim_fpu *f) | |
3971886a | 700 | { |
9655c439 AC |
701 | unsigned64 s = pack_fpu (f, 1); |
702 | *l = s; | |
703 | *h = (s >> 32); | |
3971886a AC |
704 | } |
705 | ||
706 | ||
9655c439 AC |
707 | INLINE_SIM_FPU (void) |
708 | sim_fpu_to64 (unsigned64 *u, | |
709 | const sim_fpu *f) | |
3971886a | 710 | { |
9655c439 | 711 | *u = pack_fpu (f, 1); |
3971886a AC |
712 | } |
713 | ||
714 | ||
9655c439 | 715 | /* Rounding */ |
3971886a | 716 | |
9655c439 AC |
717 | STATIC_INLINE_SIM_FPU (int) |
718 | do_normal_overflow (sim_fpu *f, | |
719 | int is_double, | |
720 | sim_fpu_round round) | |
3971886a | 721 | { |
9655c439 | 722 | switch (round) |
07b4c0a6 | 723 | { |
9655c439 AC |
724 | case sim_fpu_round_default: |
725 | return 0; | |
726 | case sim_fpu_round_near: | |
727 | f->class = sim_fpu_class_infinity; | |
728 | break; | |
729 | case sim_fpu_round_up: | |
730 | if (!f->sign) | |
731 | f->class = sim_fpu_class_infinity; | |
732 | break; | |
733 | case sim_fpu_round_down: | |
734 | if (f->sign) | |
735 | f->class = sim_fpu_class_infinity; | |
736 | break; | |
737 | case sim_fpu_round_zero: | |
738 | break; | |
07b4c0a6 | 739 | } |
9655c439 AC |
740 | f->normal_exp = NORMAL_EXPMAX; |
741 | f->fraction = LSMASK64 (NR_FRAC_GUARD, NR_GUARDS); | |
742 | return (sim_fpu_status_overflow | sim_fpu_status_inexact); | |
743 | } | |
07b4c0a6 | 744 | |
9655c439 AC |
745 | STATIC_INLINE_SIM_FPU (int) |
746 | do_normal_underflow (sim_fpu *f, | |
747 | int is_double, | |
748 | sim_fpu_round round) | |
749 | { | |
750 | switch (round) | |
07b4c0a6 | 751 | { |
9655c439 AC |
752 | case sim_fpu_round_default: |
753 | return 0; | |
754 | case sim_fpu_round_near: | |
755 | f->class = sim_fpu_class_zero; | |
756 | break; | |
757 | case sim_fpu_round_up: | |
758 | if (f->sign) | |
759 | f->class = sim_fpu_class_zero; | |
760 | break; | |
761 | case sim_fpu_round_down: | |
762 | if (!f->sign) | |
763 | f->class = sim_fpu_class_zero; | |
764 | break; | |
765 | case sim_fpu_round_zero: | |
766 | f->class = sim_fpu_class_zero; | |
767 | break; | |
07b4c0a6 | 768 | } |
9655c439 AC |
769 | f->normal_exp = NORMAL_EXPMIN - NR_FRACBITS; |
770 | f->fraction = IMPLICIT_1; | |
771 | return (sim_fpu_status_inexact | sim_fpu_status_underflow); | |
772 | } | |
07b4c0a6 | 773 | |
07b4c0a6 | 774 | |
07b4c0a6 | 775 | |
9655c439 AC |
776 | /* Round a number using NR_GUARDS. |
777 | Will return the rounded number or F->FRACTION == 0 when underflow */ | |
778 | ||
779 | STATIC_INLINE_SIM_FPU (int) | |
780 | do_normal_round (sim_fpu *f, | |
781 | int nr_guards, | |
782 | sim_fpu_round round) | |
783 | { | |
784 | unsigned64 guardmask = LSMASK64 (nr_guards - 1, 0); | |
785 | unsigned64 guardmsb = LSBIT64 (nr_guards - 1); | |
786 | unsigned64 fraclsb = guardmsb << 1; | |
787 | if ((f->fraction & guardmask)) | |
788 | { | |
789 | int status = sim_fpu_status_inexact; | |
790 | switch (round) | |
791 | { | |
792 | case sim_fpu_round_default: | |
793 | return 0; | |
794 | case sim_fpu_round_near: | |
795 | if ((f->fraction & guardmsb)) | |
796 | { | |
797 | if ((f->fraction & fraclsb)) | |
798 | { | |
799 | status |= sim_fpu_status_rounded; | |
800 | } | |
801 | else if ((f->fraction & (guardmask >> 1))) | |
802 | { | |
803 | status |= sim_fpu_status_rounded; | |
804 | } | |
805 | } | |
806 | break; | |
807 | case sim_fpu_round_up: | |
808 | if (!f->sign) | |
809 | status |= sim_fpu_status_rounded; | |
810 | break; | |
811 | case sim_fpu_round_down: | |
812 | if (f->sign) | |
813 | status |= sim_fpu_status_rounded; | |
814 | break; | |
815 | case sim_fpu_round_zero: | |
816 | break; | |
817 | } | |
818 | f->fraction &= ~guardmask; | |
819 | /* round if needed, handle resulting overflow */ | |
820 | if ((status & sim_fpu_status_rounded)) | |
821 | { | |
822 | f->fraction += fraclsb; | |
823 | if ((f->fraction & IMPLICIT_2)) | |
824 | { | |
825 | f->fraction >>= 1; | |
826 | f->normal_exp += 1; | |
827 | } | |
828 | } | |
829 | return status; | |
830 | } | |
831 | else | |
832 | return 0; | |
833 | } | |
834 | ||
835 | ||
836 | STATIC_INLINE_SIM_FPU (int) | |
837 | do_round (sim_fpu *f, | |
838 | int is_double, | |
839 | sim_fpu_round round, | |
840 | sim_fpu_denorm denorm) | |
841 | { | |
842 | switch (f->class) | |
843 | { | |
844 | case sim_fpu_class_qnan: | |
845 | case sim_fpu_class_zero: | |
846 | case sim_fpu_class_infinity: | |
847 | return 0; | |
848 | break; | |
849 | case sim_fpu_class_snan: | |
850 | /* Quieten a SignalingNaN */ | |
851 | f->class = sim_fpu_class_qnan; | |
852 | return sim_fpu_status_invalid_snan; | |
853 | break; | |
854 | case sim_fpu_class_number: | |
bdfe5c04 | 855 | case sim_fpu_class_denorm: |
9655c439 AC |
856 | { |
857 | int status; | |
858 | ASSERT (f->fraction < IMPLICIT_2); | |
859 | ASSERT (f->fraction >= IMPLICIT_1); | |
860 | if (f->normal_exp < NORMAL_EXPMIN) | |
861 | { | |
862 | /* This number's exponent is too low to fit into the bits | |
863 | available in the number. Round off any bits that will be | |
864 | discarded as a result of denormalization. Edge case is | |
865 | the implicit bit shifted to GUARD0 and then rounded | |
866 | up. */ | |
867 | int shift = NORMAL_EXPMIN - f->normal_exp; | |
868 | if (shift + NR_GUARDS <= NR_FRAC_GUARD + 1 | |
869 | && !(denorm & sim_fpu_denorm_zero)) | |
870 | { | |
9655c439 AC |
871 | status = do_normal_round (f, shift + NR_GUARDS, round); |
872 | if (f->fraction == 0) /* rounding underflowed */ | |
bdfe5c04 AC |
873 | { |
874 | status |= do_normal_underflow (f, is_double, round); | |
875 | } | |
9655c439 AC |
876 | else if (f->normal_exp < NORMAL_EXPMIN) /* still underflow? */ |
877 | { | |
878 | status |= sim_fpu_status_denorm; | |
879 | /* Any loss of precision when denormalizing is | |
880 | underflow. Some processors check for underflow | |
881 | before rounding, some after! */ | |
882 | if (status & sim_fpu_status_inexact) | |
883 | status |= sim_fpu_status_underflow; | |
bdfe5c04 AC |
884 | /* Flag that resultant value has been denormalized */ |
885 | f->class = sim_fpu_class_denorm; | |
9655c439 AC |
886 | } |
887 | else if ((denorm & sim_fpu_denorm_underflow_inexact)) | |
888 | { | |
889 | if ((status & sim_fpu_status_inexact)) | |
890 | status |= sim_fpu_status_underflow; | |
891 | } | |
892 | } | |
893 | else | |
894 | { | |
895 | status = do_normal_underflow (f, is_double, round); | |
896 | } | |
897 | } | |
898 | else if (f->normal_exp > NORMAL_EXPMAX) | |
899 | { | |
900 | /* Infinity */ | |
901 | status = do_normal_overflow (f, is_double, round); | |
902 | } | |
903 | else | |
904 | { | |
905 | status = do_normal_round (f, NR_GUARDS, round); | |
906 | if (f->fraction == 0) | |
907 | /* f->class = sim_fpu_class_zero; */ | |
908 | status |= do_normal_underflow (f, is_double, round); | |
909 | else if (f->normal_exp > NORMAL_EXPMAX) | |
910 | /* oops! rounding caused overflow */ | |
911 | status |= do_normal_overflow (f, is_double, round); | |
912 | } | |
bdfe5c04 AC |
913 | ASSERT ((f->class == sim_fpu_class_number |
914 | || f->class == sim_fpu_class_denorm) | |
9655c439 AC |
915 | <= (f->fraction < IMPLICIT_2 && f->fraction >= IMPLICIT_1)); |
916 | return status; | |
917 | } | |
918 | } | |
919 | return 0; | |
920 | } | |
921 | ||
922 | INLINE_SIM_FPU (int) | |
923 | sim_fpu_round_32 (sim_fpu *f, | |
924 | sim_fpu_round round, | |
925 | sim_fpu_denorm denorm) | |
926 | { | |
927 | return do_round (f, 0, round, denorm); | |
928 | } | |
929 | ||
930 | INLINE_SIM_FPU (int) | |
931 | sim_fpu_round_64 (sim_fpu *f, | |
932 | sim_fpu_round round, | |
933 | sim_fpu_denorm denorm) | |
934 | { | |
935 | return do_round (f, 1, round, denorm); | |
936 | } | |
937 | ||
938 | ||
939 | ||
940 | /* Arithmetic ops */ | |
941 | ||
942 | INLINE_SIM_FPU (int) | |
943 | sim_fpu_add (sim_fpu *f, | |
944 | const sim_fpu *l, | |
945 | const sim_fpu *r) | |
946 | { | |
947 | if (sim_fpu_is_snan (l)) | |
948 | { | |
949 | *f = *l; | |
950 | f->class = sim_fpu_class_qnan; | |
951 | return sim_fpu_status_invalid_snan; | |
952 | } | |
953 | if (sim_fpu_is_snan (r)) | |
954 | { | |
955 | *f = *r; | |
956 | f->class = sim_fpu_class_qnan; | |
957 | return sim_fpu_status_invalid_snan; | |
958 | } | |
959 | if (sim_fpu_is_qnan (l)) | |
960 | { | |
961 | *f = *l; | |
962 | return 0; | |
963 | } | |
964 | if (sim_fpu_is_qnan (r)) | |
965 | { | |
966 | *f = *r; | |
967 | return 0; | |
968 | } | |
969 | if (sim_fpu_is_infinity (l)) | |
970 | { | |
971 | if (sim_fpu_is_infinity (r) | |
972 | && l->sign != r->sign) | |
973 | { | |
974 | *f = sim_fpu_qnan; | |
975 | return sim_fpu_status_invalid_isi; | |
976 | } | |
977 | *f = *l; | |
978 | return 0; | |
979 | } | |
980 | if (sim_fpu_is_infinity (r)) | |
981 | { | |
982 | *f = *r; | |
983 | return 0; | |
984 | } | |
985 | if (sim_fpu_is_zero (l)) | |
986 | { | |
987 | if (sim_fpu_is_zero (r)) | |
988 | { | |
989 | *f = sim_fpu_zero; | |
990 | f->sign = l->sign & r->sign; | |
991 | } | |
992 | else | |
993 | *f = *r; | |
994 | return 0; | |
995 | } | |
996 | if (sim_fpu_is_zero (r)) | |
997 | { | |
998 | *f = *l; | |
999 | return 0; | |
1000 | } | |
1001 | { | |
1002 | int status = 0; | |
1003 | int shift = l->normal_exp - r->normal_exp; | |
1004 | unsigned64 lfraction; | |
1005 | unsigned64 rfraction; | |
1006 | /* use exp of larger */ | |
1007 | if (shift >= NR_FRAC_GUARD) | |
1008 | { | |
1009 | /* left has much bigger magnitute */ | |
1010 | *f = *l; | |
1011 | return sim_fpu_status_inexact; | |
1012 | } | |
1013 | if (shift <= - NR_FRAC_GUARD) | |
1014 | { | |
1015 | /* right has much bigger magnitute */ | |
1016 | *f = *r; | |
1017 | return sim_fpu_status_inexact; | |
1018 | } | |
1019 | lfraction = l->fraction; | |
1020 | rfraction = r->fraction; | |
1021 | if (shift > 0) | |
1022 | { | |
1023 | f->normal_exp = l->normal_exp; | |
1024 | if (rfraction & LSMASK64 (shift - 1, 0)) | |
1025 | { | |
1026 | status |= sim_fpu_status_inexact; | |
1027 | rfraction |= LSBIT64 (shift); /* stick LSBit */ | |
1028 | } | |
1029 | rfraction >>= shift; | |
1030 | } | |
1031 | else if (shift < 0) | |
1032 | { | |
1033 | f->normal_exp = r->normal_exp; | |
1034 | if (lfraction & LSMASK64 (- shift - 1, 0)) | |
1035 | { | |
1036 | status |= sim_fpu_status_inexact; | |
1037 | lfraction |= LSBIT64 (- shift); /* stick LSBit */ | |
1038 | } | |
1039 | lfraction >>= -shift; | |
1040 | } | |
1041 | else | |
1042 | { | |
1043 | f->normal_exp = r->normal_exp; | |
1044 | } | |
1045 | ||
1046 | /* perform the addition */ | |
1047 | if (l->sign) | |
1048 | lfraction = - lfraction; | |
1049 | if (r->sign) | |
1050 | rfraction = - rfraction; | |
1051 | f->fraction = lfraction + rfraction; | |
1052 | ||
1053 | /* zero? */ | |
1054 | if (f->fraction == 0) | |
1055 | { | |
1056 | *f = sim_fpu_zero; | |
1057 | return 0; | |
1058 | } | |
1059 | ||
1060 | /* sign? */ | |
1061 | f->class = sim_fpu_class_number; | |
1062 | if ((signed64) f->fraction >= 0) | |
1063 | f->sign = 0; | |
1064 | else | |
1065 | { | |
1066 | f->sign = 1; | |
1067 | f->fraction = - f->fraction; | |
1068 | } | |
1069 | ||
1070 | /* normalize it */ | |
1071 | if ((f->fraction & IMPLICIT_2)) | |
1072 | { | |
1073 | f->fraction = (f->fraction >> 1) | (f->fraction & 1); | |
1074 | f->normal_exp ++; | |
1075 | } | |
1076 | else if (f->fraction < IMPLICIT_1) | |
1077 | { | |
1078 | do | |
1079 | { | |
1080 | f->fraction <<= 1; | |
1081 | f->normal_exp --; | |
1082 | } | |
1083 | while (f->fraction < IMPLICIT_1); | |
1084 | } | |
1085 | ASSERT (f->fraction >= IMPLICIT_1 && f->fraction < IMPLICIT_2); | |
1086 | return status; | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | ||
1091 | INLINE_SIM_FPU (int) | |
1092 | sim_fpu_sub (sim_fpu *f, | |
1093 | const sim_fpu *l, | |
1094 | const sim_fpu *r) | |
1095 | { | |
1096 | if (sim_fpu_is_snan (l)) | |
1097 | { | |
1098 | *f = *l; | |
1099 | f->class = sim_fpu_class_qnan; | |
1100 | return sim_fpu_status_invalid_snan; | |
1101 | } | |
1102 | if (sim_fpu_is_snan (r)) | |
1103 | { | |
1104 | *f = *r; | |
1105 | f->class = sim_fpu_class_qnan; | |
1106 | return sim_fpu_status_invalid_snan; | |
1107 | } | |
1108 | if (sim_fpu_is_qnan (l)) | |
1109 | { | |
1110 | *f = *l; | |
1111 | return 0; | |
1112 | } | |
1113 | if (sim_fpu_is_qnan (r)) | |
1114 | { | |
1115 | *f = *r; | |
1116 | return 0; | |
1117 | } | |
1118 | if (sim_fpu_is_infinity (l)) | |
1119 | { | |
1120 | if (sim_fpu_is_infinity (r) | |
1121 | && l->sign == r->sign) | |
1122 | { | |
1123 | *f = sim_fpu_qnan; | |
1124 | return sim_fpu_status_invalid_isi; | |
1125 | } | |
1126 | *f = *l; | |
1127 | return 0; | |
1128 | } | |
1129 | if (sim_fpu_is_infinity (r)) | |
1130 | { | |
1131 | *f = *r; | |
1132 | f->sign = !r->sign; | |
1133 | return 0; | |
1134 | } | |
1135 | if (sim_fpu_is_zero (l)) | |
1136 | { | |
1137 | if (sim_fpu_is_zero (r)) | |
1138 | { | |
1139 | *f = sim_fpu_zero; | |
1140 | f->sign = l->sign & !r->sign; | |
1141 | } | |
1142 | else | |
1143 | { | |
1144 | *f = *r; | |
1145 | f->sign = !r->sign; | |
1146 | } | |
1147 | return 0; | |
1148 | } | |
1149 | if (sim_fpu_is_zero (r)) | |
1150 | { | |
1151 | *f = *l; | |
1152 | return 0; | |
1153 | } | |
1154 | { | |
1155 | int status = 0; | |
1156 | int shift = l->normal_exp - r->normal_exp; | |
1157 | unsigned64 lfraction; | |
1158 | unsigned64 rfraction; | |
1159 | /* use exp of larger */ | |
1160 | if (shift >= NR_FRAC_GUARD) | |
1161 | { | |
1162 | /* left has much bigger magnitute */ | |
1163 | *f = *l; | |
1164 | return sim_fpu_status_inexact; | |
1165 | } | |
1166 | if (shift <= - NR_FRAC_GUARD) | |
1167 | { | |
1168 | /* right has much bigger magnitute */ | |
1169 | *f = *r; | |
1170 | f->sign = !r->sign; | |
1171 | return sim_fpu_status_inexact; | |
1172 | } | |
1173 | lfraction = l->fraction; | |
1174 | rfraction = r->fraction; | |
1175 | if (shift > 0) | |
1176 | { | |
1177 | f->normal_exp = l->normal_exp; | |
1178 | if (rfraction & LSMASK64 (shift - 1, 0)) | |
1179 | { | |
1180 | status |= sim_fpu_status_inexact; | |
1181 | rfraction |= LSBIT64 (shift); /* stick LSBit */ | |
1182 | } | |
1183 | rfraction >>= shift; | |
1184 | } | |
1185 | else if (shift < 0) | |
1186 | { | |
1187 | f->normal_exp = r->normal_exp; | |
1188 | if (lfraction & LSMASK64 (- shift - 1, 0)) | |
1189 | { | |
1190 | status |= sim_fpu_status_inexact; | |
1191 | lfraction |= LSBIT64 (- shift); /* stick LSBit */ | |
1192 | } | |
1193 | lfraction >>= -shift; | |
1194 | } | |
1195 | else | |
1196 | { | |
1197 | f->normal_exp = r->normal_exp; | |
1198 | } | |
1199 | ||
1200 | /* perform the subtraction */ | |
1201 | if (l->sign) | |
1202 | lfraction = - lfraction; | |
1203 | if (!r->sign) | |
1204 | rfraction = - rfraction; | |
1205 | f->fraction = lfraction + rfraction; | |
1206 | ||
1207 | /* zero? */ | |
1208 | if (f->fraction == 0) | |
1209 | { | |
1210 | *f = sim_fpu_zero; | |
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | /* sign? */ | |
1215 | f->class = sim_fpu_class_number; | |
1216 | if ((signed64) f->fraction >= 0) | |
1217 | f->sign = 0; | |
1218 | else | |
1219 | { | |
1220 | f->sign = 1; | |
1221 | f->fraction = - f->fraction; | |
1222 | } | |
1223 | ||
1224 | /* normalize it */ | |
1225 | if ((f->fraction & IMPLICIT_2)) | |
1226 | { | |
1227 | f->fraction = (f->fraction >> 1) | (f->fraction & 1); | |
1228 | f->normal_exp ++; | |
1229 | } | |
1230 | else if (f->fraction < IMPLICIT_1) | |
1231 | { | |
1232 | do | |
1233 | { | |
1234 | f->fraction <<= 1; | |
1235 | f->normal_exp --; | |
1236 | } | |
1237 | while (f->fraction < IMPLICIT_1); | |
1238 | } | |
1239 | ASSERT (f->fraction >= IMPLICIT_1 && f->fraction < IMPLICIT_2); | |
1240 | return status; | |
1241 | } | |
1242 | } | |
1243 | ||
1244 | ||
1245 | INLINE_SIM_FPU (int) | |
1246 | sim_fpu_mul (sim_fpu *f, | |
1247 | const sim_fpu *l, | |
1248 | const sim_fpu *r) | |
1249 | { | |
1250 | if (sim_fpu_is_snan (l)) | |
1251 | { | |
1252 | *f = *l; | |
1253 | f->class = sim_fpu_class_qnan; | |
1254 | return sim_fpu_status_invalid_snan; | |
1255 | } | |
1256 | if (sim_fpu_is_snan (r)) | |
1257 | { | |
1258 | *f = *r; | |
1259 | f->class = sim_fpu_class_qnan; | |
1260 | return sim_fpu_status_invalid_snan; | |
1261 | } | |
1262 | if (sim_fpu_is_qnan (l)) | |
1263 | { | |
1264 | *f = *l; | |
1265 | return 0; | |
1266 | } | |
1267 | if (sim_fpu_is_qnan (r)) | |
1268 | { | |
1269 | *f = *r; | |
1270 | return 0; | |
1271 | } | |
1272 | if (sim_fpu_is_infinity (l)) | |
1273 | { | |
1274 | if (sim_fpu_is_zero (r)) | |
1275 | { | |
1276 | *f = sim_fpu_qnan; | |
1277 | return sim_fpu_status_invalid_imz; | |
1278 | } | |
1279 | *f = *l; | |
1280 | f->sign = l->sign ^ r->sign; | |
1281 | return 0; | |
1282 | } | |
1283 | if (sim_fpu_is_infinity (r)) | |
1284 | { | |
1285 | if (sim_fpu_is_zero (l)) | |
1286 | { | |
1287 | *f = sim_fpu_qnan; | |
1288 | return sim_fpu_status_invalid_imz; | |
1289 | } | |
1290 | *f = *r; | |
1291 | f->sign = l->sign ^ r->sign; | |
1292 | return 0; | |
1293 | } | |
1294 | if (sim_fpu_is_zero (l) || sim_fpu_is_zero (r)) | |
1295 | { | |
1296 | *f = sim_fpu_zero; | |
1297 | f->sign = l->sign ^ r->sign; | |
1298 | return 0; | |
1299 | } | |
1300 | /* Calculate the mantissa by multiplying both 64bit numbers to get a | |
1301 | 128 bit number */ | |
1302 | { | |
1303 | unsigned64 low; | |
1304 | unsigned64 high; | |
1305 | unsigned64 nl = l->fraction & 0xffffffff; | |
1306 | unsigned64 nh = l->fraction >> 32; | |
1307 | unsigned64 ml = r->fraction & 0xffffffff; | |
1308 | unsigned64 mh = r->fraction >>32; | |
1309 | unsigned64 pp_ll = ml * nl; | |
1310 | unsigned64 pp_hl = mh * nl; | |
1311 | unsigned64 pp_lh = ml * nh; | |
1312 | unsigned64 pp_hh = mh * nh; | |
1313 | unsigned64 res2 = 0; | |
1314 | unsigned64 res0 = 0; | |
1315 | unsigned64 ps_hh__ = pp_hl + pp_lh; | |
1316 | if (ps_hh__ < pp_hl) | |
1317 | res2 += UNSIGNED64 (0x100000000); | |
1318 | pp_hl = (ps_hh__ << 32) & UNSIGNED64 (0xffffffff00000000); | |
1319 | res0 = pp_ll + pp_hl; | |
1320 | if (res0 < pp_ll) | |
1321 | res2++; | |
1322 | res2 += ((ps_hh__ >> 32) & 0xffffffff) + pp_hh; | |
1323 | high = res2; | |
1324 | low = res0; | |
1325 | ||
1326 | f->normal_exp = l->normal_exp + r->normal_exp; | |
1327 | f->sign = l->sign ^ r->sign; | |
1328 | f->class = sim_fpu_class_number; | |
1329 | ||
1330 | /* Input is bounded by [1,2) ; [2^60,2^61) | |
1331 | Output is bounded by [1,4) ; [2^120,2^122) */ | |
1332 | ||
1333 | /* Adjust the exponent according to where the decimal point ended | |
1334 | up in the high 64 bit word. In the source the decimal point | |
1335 | was at NR_FRAC_GUARD. */ | |
1336 | f->normal_exp += NR_FRAC_GUARD + 64 - (NR_FRAC_GUARD * 2); | |
1337 | ||
1338 | /* The high word is bounded according to the above. Consequently | |
1339 | it has never overflowed into IMPLICIT_2. */ | |
1340 | ASSERT (high < LSBIT64 (((NR_FRAC_GUARD + 1) * 2) - 64)); | |
1341 | ASSERT (high >= LSBIT64 ((NR_FRAC_GUARD * 2) - 64)); | |
1342 | ASSERT (LSBIT64 (((NR_FRAC_GUARD + 1) * 2) - 64) < IMPLICIT_1); | |
1343 | ||
1344 | #if 0 | |
1345 | printf ("\n"); | |
1346 | print_bits (high, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1347 | printf (";"); | |
1348 | print_bits (low, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1349 | printf ("\n"); | |
1350 | #endif | |
1351 | ||
1352 | /* normalize */ | |
1353 | do | |
1354 | { | |
1355 | f->normal_exp--; | |
1356 | high <<= 1; | |
1357 | if (low & LSBIT64 (63)) | |
1358 | high |= 1; | |
1359 | low <<= 1; | |
1360 | } | |
1361 | while (high < IMPLICIT_1); | |
1362 | ||
1363 | #if 0 | |
1364 | print_bits (high, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1365 | printf (";"); | |
1366 | print_bits (low, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1367 | printf ("\n"); | |
1368 | #endif | |
1369 | ||
1370 | ASSERT (high >= IMPLICIT_1 && high < IMPLICIT_2); | |
1371 | if (low != 0) | |
1372 | { | |
1373 | f->fraction = (high | 1); /* sticky */ | |
1374 | return sim_fpu_status_inexact; | |
1375 | } | |
1376 | else | |
1377 | { | |
1378 | f->fraction = high; | |
1379 | return 0; | |
1380 | } | |
1381 | return 0; | |
1382 | } | |
1383 | } | |
1384 | ||
1385 | INLINE_SIM_FPU (int) | |
1386 | sim_fpu_div (sim_fpu *f, | |
1387 | const sim_fpu *l, | |
1388 | const sim_fpu *r) | |
1389 | { | |
1390 | if (sim_fpu_is_snan (l)) | |
1391 | { | |
1392 | *f = *l; | |
1393 | f->class = sim_fpu_class_qnan; | |
1394 | return sim_fpu_status_invalid_snan; | |
1395 | } | |
1396 | if (sim_fpu_is_snan (r)) | |
1397 | { | |
1398 | *f = *r; | |
1399 | f->class = sim_fpu_class_qnan; | |
1400 | return sim_fpu_status_invalid_snan; | |
1401 | } | |
1402 | if (sim_fpu_is_qnan (l)) | |
1403 | { | |
1404 | *f = *l; | |
1405 | f->class = sim_fpu_class_qnan; | |
1406 | return 0; | |
1407 | } | |
1408 | if (sim_fpu_is_qnan (r)) | |
1409 | { | |
1410 | *f = *r; | |
1411 | f->class = sim_fpu_class_qnan; | |
1412 | return 0; | |
1413 | } | |
1414 | if (sim_fpu_is_infinity (l)) | |
1415 | { | |
1416 | if (sim_fpu_is_infinity (r)) | |
1417 | { | |
1418 | *f = sim_fpu_qnan; | |
1419 | return sim_fpu_status_invalid_idi; | |
1420 | } | |
1421 | else | |
1422 | { | |
1423 | *f = *l; | |
1424 | f->sign = l->sign ^ r->sign; | |
1425 | return 0; | |
1426 | } | |
1427 | } | |
1428 | if (sim_fpu_is_zero (l)) | |
1429 | { | |
1430 | if (sim_fpu_is_zero (r)) | |
1431 | { | |
1432 | *f = sim_fpu_qnan; | |
1433 | return sim_fpu_status_invalid_zdz; | |
1434 | } | |
1435 | else | |
1436 | { | |
1437 | *f = *l; | |
1438 | f->sign = l->sign ^ r->sign; | |
1439 | return 0; | |
1440 | } | |
1441 | } | |
1442 | if (sim_fpu_is_infinity (r)) | |
1443 | { | |
1444 | *f = sim_fpu_zero; | |
1445 | f->sign = l->sign ^ r->sign; | |
1446 | return 0; | |
1447 | } | |
1448 | if (sim_fpu_is_zero (r)) | |
1449 | { | |
1450 | f->class = sim_fpu_class_infinity; | |
1451 | f->sign = l->sign ^ r->sign; | |
1452 | return sim_fpu_status_invalid_div0; | |
1453 | } | |
1454 | ||
1455 | /* Calculate the mantissa by multiplying both 64bit numbers to get a | |
1456 | 128 bit number */ | |
1457 | { | |
1458 | /* quotient = ( ( numerator / denominator) | |
1459 | x 2^(numerator exponent - denominator exponent) | |
1460 | */ | |
1461 | unsigned64 numerator; | |
1462 | unsigned64 denominator; | |
1463 | unsigned64 quotient; | |
1464 | unsigned64 bit; | |
1465 | ||
1466 | f->class = sim_fpu_class_number; | |
1467 | f->sign = l->sign ^ r->sign; | |
1468 | f->normal_exp = l->normal_exp - r->normal_exp; | |
1469 | ||
1470 | numerator = l->fraction; | |
1471 | denominator = r->fraction; | |
1472 | ||
1473 | /* Fraction will be less than 1.0 */ | |
1474 | if (numerator < denominator) | |
1475 | { | |
1476 | numerator <<= 1; | |
1477 | f->normal_exp--; | |
1478 | } | |
1479 | ASSERT (numerator >= denominator); | |
1480 | ||
1481 | /* Gain extra precision, already used one spare bit */ | |
1482 | numerator <<= NR_SPARE; | |
1483 | denominator <<= NR_SPARE; | |
1484 | ||
1485 | /* Does divide one bit at a time. Optimize??? */ | |
1486 | quotient = 0; | |
1487 | bit = (IMPLICIT_1 << NR_SPARE); | |
1488 | while (bit) | |
1489 | { | |
1490 | if (numerator >= denominator) | |
1491 | { | |
1492 | quotient |= bit; | |
07b4c0a6 AC |
1493 | numerator -= denominator; |
1494 | } | |
1495 | bit >>= 1; | |
9655c439 AC |
1496 | numerator <<= 1; |
1497 | } | |
1498 | ||
1499 | #if 0 | |
1500 | printf ("\n"); | |
1501 | print_bits (quotient, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1502 | printf ("\n"); | |
1503 | print_bits (numerator, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1504 | printf ("\n"); | |
1505 | print_bits (denominator, 63, (sim_fpu_print_func*)fprintf, stdout); | |
1506 | printf ("\n"); | |
1507 | #endif | |
1508 | ||
1509 | /* discard (but save) the extra bits */ | |
1510 | if ((quotient & LSMASK64 (NR_SPARE -1, 0))) | |
1511 | quotient = (quotient >> NR_SPARE) | 1; | |
1512 | else | |
1513 | quotient = (quotient >> NR_SPARE); | |
1514 | ||
1515 | f->fraction = quotient; | |
1516 | ASSERT (f->fraction >= IMPLICIT_1 && f->fraction < IMPLICIT_2); | |
1517 | if (numerator != 0) | |
1518 | { | |
1519 | f->fraction |= 1; /* stick remaining bits */ | |
1520 | return sim_fpu_status_inexact; | |
07b4c0a6 | 1521 | } |
9655c439 AC |
1522 | else |
1523 | return 0; | |
1524 | } | |
1525 | } | |
1526 | ||
1527 | ||
390ffa89 AC |
1528 | INLINE_SIM_FPU (int) |
1529 | sim_fpu_max (sim_fpu *f, | |
1530 | const sim_fpu *l, | |
1531 | const sim_fpu *r) | |
1532 | { | |
1533 | if (sim_fpu_is_snan (l)) | |
1534 | { | |
1535 | *f = *l; | |
1536 | f->class = sim_fpu_class_qnan; | |
1537 | return sim_fpu_status_invalid_snan; | |
1538 | } | |
1539 | if (sim_fpu_is_snan (r)) | |
1540 | { | |
1541 | *f = *r; | |
1542 | f->class = sim_fpu_class_qnan; | |
1543 | return sim_fpu_status_invalid_snan; | |
1544 | } | |
1545 | if (sim_fpu_is_qnan (l)) | |
1546 | { | |
1547 | *f = *l; | |
1548 | return 0; | |
1549 | } | |
1550 | if (sim_fpu_is_qnan (r)) | |
1551 | { | |
1552 | *f = *r; | |
1553 | return 0; | |
1554 | } | |
1555 | if (sim_fpu_is_infinity (l)) | |
1556 | { | |
1557 | if (sim_fpu_is_infinity (r) | |
1558 | && l->sign == r->sign) | |
1559 | { | |
1560 | *f = sim_fpu_qnan; | |
1561 | return sim_fpu_status_invalid_isi; | |
1562 | } | |
1563 | if (l->sign) | |
1564 | *f = *r; /* -inf < anything */ | |
1565 | else | |
1566 | *f = *l; /* +inf > anthing */ | |
1567 | return 0; | |
1568 | } | |
1569 | if (sim_fpu_is_infinity (r)) | |
1570 | { | |
1571 | if (r->sign) | |
1572 | *f = *l; /* anything > -inf */ | |
1573 | else | |
1574 | *f = *r; /* anthing < +inf */ | |
1575 | return 0; | |
1576 | } | |
1577 | if (l->sign > r->sign) | |
1578 | { | |
1579 | *f = *r; /* -ve < +ve */ | |
1580 | return 0; | |
1581 | } | |
1582 | if (l->sign < r->sign) | |
1583 | { | |
1584 | *f = *l; /* +ve > -ve */ | |
1585 | return 0; | |
1586 | } | |
1587 | ASSERT (l->sign == r->sign); | |
1588 | if (l->normal_exp > r->normal_exp | |
1589 | || (l->normal_exp == r->normal_exp && | |
1590 | l->fraction > r->fraction)) | |
1591 | { | |
1592 | /* |l| > |r| */ | |
1593 | if (l->sign) | |
1594 | *f = *r; /* -ve < -ve */ | |
1595 | else | |
1596 | *f = *l; /* +ve > +ve */ | |
1597 | return 0; | |
1598 | } | |
1599 | else | |
1600 | { | |
1601 | /* |l| <= |r| */ | |
1602 | if (l->sign) | |
1603 | *f = *l; /* -ve > -ve */ | |
1604 | else | |
1605 | *f = *r; /* +ve < +ve */ | |
1606 | return 0; | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | ||
1611 | INLINE_SIM_FPU (int) | |
1612 | sim_fpu_min (sim_fpu *f, | |
1613 | const sim_fpu *l, | |
1614 | const sim_fpu *r) | |
1615 | { | |
1616 | if (sim_fpu_is_snan (l)) | |
1617 | { | |
1618 | *f = *l; | |
1619 | f->class = sim_fpu_class_qnan; | |
1620 | return sim_fpu_status_invalid_snan; | |
1621 | } | |
1622 | if (sim_fpu_is_snan (r)) | |
1623 | { | |
1624 | *f = *r; | |
1625 | f->class = sim_fpu_class_qnan; | |
1626 | return sim_fpu_status_invalid_snan; | |
1627 | } | |
1628 | if (sim_fpu_is_qnan (l)) | |
1629 | { | |
1630 | *f = *l; | |
1631 | return 0; | |
1632 | } | |
1633 | if (sim_fpu_is_qnan (r)) | |
1634 | { | |
1635 | *f = *r; | |
1636 | return 0; | |
1637 | } | |
1638 | if (sim_fpu_is_infinity (l)) | |
1639 | { | |
1640 | if (sim_fpu_is_infinity (r) | |
1641 | && l->sign == r->sign) | |
1642 | { | |
1643 | *f = sim_fpu_qnan; | |
1644 | return sim_fpu_status_invalid_isi; | |
1645 | } | |
1646 | if (l->sign) | |
1647 | *f = *l; /* -inf < anything */ | |
1648 | else | |
1649 | *f = *r; /* +inf > anthing */ | |
1650 | return 0; | |
1651 | } | |
1652 | if (sim_fpu_is_infinity (r)) | |
1653 | { | |
1654 | if (r->sign) | |
1655 | *f = *r; /* anything > -inf */ | |
1656 | else | |
1657 | *f = *l; /* anything < +inf */ | |
1658 | return 0; | |
1659 | } | |
1660 | if (l->sign > r->sign) | |
1661 | { | |
1662 | *f = *l; /* -ve < +ve */ | |
1663 | return 0; | |
1664 | } | |
1665 | if (l->sign < r->sign) | |
1666 | { | |
1667 | *f = *r; /* +ve > -ve */ | |
1668 | return 0; | |
1669 | } | |
1670 | ASSERT (l->sign == r->sign); | |
1671 | if (l->normal_exp > r->normal_exp | |
1672 | || (l->normal_exp == r->normal_exp && | |
1673 | l->fraction > r->fraction)) | |
1674 | { | |
1675 | /* |l| > |r| */ | |
1676 | if (l->sign) | |
1677 | *f = *l; /* -ve < -ve */ | |
1678 | else | |
1679 | *f = *r; /* +ve > +ve */ | |
1680 | return 0; | |
1681 | } | |
1682 | else | |
1683 | { | |
1684 | /* |l| <= |r| */ | |
1685 | if (l->sign) | |
1686 | *f = *r; /* -ve > -ve */ | |
1687 | else | |
1688 | *f = *l; /* +ve < +ve */ | |
1689 | return 0; | |
1690 | } | |
1691 | } | |
1692 | ||
1693 | ||
9655c439 AC |
1694 | INLINE_SIM_FPU (int) |
1695 | sim_fpu_neg (sim_fpu *f, | |
1696 | const sim_fpu *r) | |
1697 | { | |
1698 | if (sim_fpu_is_snan (r)) | |
1699 | { | |
1700 | *f = *r; | |
1701 | f->class = sim_fpu_class_qnan; | |
1702 | return sim_fpu_status_invalid_snan; | |
1703 | } | |
1704 | if (sim_fpu_is_qnan (r)) | |
1705 | { | |
1706 | *f = *r; | |
1707 | return 0; | |
1708 | } | |
1709 | *f = *r; | |
1710 | f->sign = !r->sign; | |
1711 | return 0; | |
1712 | } | |
1713 | ||
1714 | ||
1715 | INLINE_SIM_FPU (int) | |
1716 | sim_fpu_abs (sim_fpu *f, | |
1717 | const sim_fpu *r) | |
1718 | { | |
1719 | if (sim_fpu_is_snan (r)) | |
1720 | { | |
1721 | *f = *r; | |
1722 | f->class = sim_fpu_class_qnan; | |
1723 | return sim_fpu_status_invalid_snan; | |
1724 | } | |
1725 | if (sim_fpu_is_qnan (r)) | |
1726 | { | |
1727 | *f = *r; | |
1728 | return 0; | |
1729 | } | |
1730 | *f = *r; | |
1731 | f->sign = 0; | |
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | ||
1736 | INLINE_SIM_FPU (int) | |
1737 | sim_fpu_inv (sim_fpu *f, | |
1738 | const sim_fpu *r) | |
1739 | { | |
1740 | if (sim_fpu_is_snan (r)) | |
1741 | { | |
1742 | *f = *r; | |
1743 | f->class = sim_fpu_class_qnan; | |
1744 | return sim_fpu_status_invalid_snan; | |
1745 | } | |
1746 | if (sim_fpu_is_qnan (r)) | |
1747 | { | |
1748 | *f = *r; | |
1749 | f->class = sim_fpu_class_qnan; | |
1750 | return 0; | |
1751 | } | |
1752 | if (sim_fpu_is_infinity (r)) | |
1753 | { | |
1754 | *f = sim_fpu_zero; | |
1755 | f->sign = r->sign; | |
1756 | return 0; | |
1757 | } | |
1758 | if (sim_fpu_is_zero (r)) | |
1759 | { | |
1760 | f->class = sim_fpu_class_infinity; | |
1761 | f->sign = r->sign; | |
1762 | return sim_fpu_status_invalid_div0; | |
1763 | } | |
1764 | *f = *r; | |
1765 | f->normal_exp = - r->normal_exp; | |
1766 | return 0; | |
1767 | } | |
1768 | ||
1769 | ||
1770 | INLINE_SIM_FPU (int) | |
1771 | sim_fpu_sqrt (sim_fpu *f, | |
1772 | const sim_fpu *r) | |
1773 | { | |
1774 | if (sim_fpu_is_snan (r)) | |
1775 | { | |
1776 | *f = sim_fpu_qnan; | |
1777 | return sim_fpu_status_invalid_snan; | |
1778 | } | |
1779 | if (sim_fpu_is_qnan (r)) | |
1780 | { | |
1781 | *f = sim_fpu_qnan; | |
1782 | return 0; | |
1783 | } | |
1784 | if (sim_fpu_is_zero (r)) | |
1785 | { | |
1786 | f->class = sim_fpu_class_zero; | |
1787 | f->sign = r->sign; | |
1788 | return 0; | |
1789 | } | |
1790 | if (sim_fpu_is_infinity (r)) | |
1791 | { | |
1792 | if (r->sign) | |
1793 | { | |
1794 | *f = sim_fpu_qnan; | |
1795 | return sim_fpu_status_invalid_sqrt; | |
1796 | } | |
1797 | else | |
1798 | { | |
1799 | f->class = sim_fpu_class_infinity; | |
1800 | f->sign = 0; | |
1801 | f->sign = 0; | |
1802 | return 0; | |
1803 | } | |
1804 | } | |
1805 | if (r->sign) | |
1806 | { | |
1807 | *f = sim_fpu_qnan; | |
1808 | return sim_fpu_status_invalid_sqrt; | |
1809 | } | |
07b4c0a6 | 1810 | |
9655c439 AC |
1811 | /* @(#)e_sqrt.c 5.1 93/09/24 */ |
1812 | /* | |
1813 | * ==================================================== | |
1814 | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. | |
1815 | * | |
1816 | * Developed at SunPro, a Sun Microsystems, Inc. business. | |
1817 | * Permission to use, copy, modify, and distribute this | |
1818 | * software is freely granted, provided that this notice | |
1819 | * is preserved. | |
1820 | * ==================================================== | |
1821 | */ | |
1822 | ||
1823 | /* __ieee754_sqrt(x) | |
1824 | * Return correctly rounded sqrt. | |
1825 | * ------------------------------------------ | |
1826 | * | Use the hardware sqrt if you have one | | |
1827 | * ------------------------------------------ | |
1828 | * Method: | |
1829 | * Bit by bit method using integer arithmetic. (Slow, but portable) | |
1830 | * 1. Normalization | |
1831 | * Scale x to y in [1,4) with even powers of 2: | |
1832 | * find an integer k such that 1 <= (y=x*2^(2k)) < 4, then | |
1833 | * sqrt(x) = 2^k * sqrt(y) | |
1834 | - | |
1835 | - Since: | |
1836 | - sqrt ( x*2^(2m) ) = sqrt(x).2^m ; m even | |
1837 | - sqrt ( x*2^(2m + 1) ) = sqrt(2.x).2^m ; m odd | |
1838 | - Define: | |
1839 | - y = ((m even) ? x : 2.x) | |
1840 | - Then: | |
1841 | - y in [1, 4) ; [IMPLICIT_1,IMPLICIT_4) | |
1842 | - And: | |
1843 | - sqrt (y) in [1, 2) ; [IMPLICIT_1,IMPLICIT_2) | |
1844 | - | |
1845 | * 2. Bit by bit computation | |
1846 | * Let q = sqrt(y) truncated to i bit after binary point (q = 1), | |
1847 | * i 0 | |
1848 | * i+1 2 | |
1849 | * s = 2*q , and y = 2 * ( y - q ). (1) | |
1850 | * i i i i | |
1851 | * | |
1852 | * To compute q from q , one checks whether | |
1853 | * i+1 i | |
1854 | * | |
1855 | * -(i+1) 2 | |
1856 | * (q + 2 ) <= y. (2) | |
1857 | * i | |
1858 | * -(i+1) | |
1859 | * If (2) is false, then q = q ; otherwise q = q + 2 . | |
1860 | * i+1 i i+1 i | |
1861 | * | |
1862 | * With some algebric manipulation, it is not difficult to see | |
1863 | * that (2) is equivalent to | |
1864 | * -(i+1) | |
1865 | * s + 2 <= y (3) | |
1866 | * i i | |
1867 | * | |
1868 | * The advantage of (3) is that s and y can be computed by | |
1869 | * i i | |
1870 | * the following recurrence formula: | |
1871 | * if (3) is false | |
1872 | * | |
1873 | * s = s , y = y ; (4) | |
1874 | * i+1 i i+1 i | |
1875 | * | |
1876 | - | |
1877 | - NOTE: y = 2*y | |
1878 | - i+1 i | |
1879 | - | |
1880 | * otherwise, | |
1881 | * -i -(i+1) | |
1882 | * s = s + 2 , y = y - s - 2 (5) | |
1883 | * i+1 i i+1 i i | |
1884 | * | |
1885 | - | |
1886 | - -(i+1) | |
1887 | - NOTE: y = 2 (y - s - 2 ) | |
1888 | - i+1 i i | |
1889 | - | |
1890 | * One may easily use induction to prove (4) and (5). | |
1891 | * Note. Since the left hand side of (3) contain only i+2 bits, | |
1892 | * it does not necessary to do a full (53-bit) comparison | |
1893 | * in (3). | |
1894 | * 3. Final rounding | |
1895 | * After generating the 53 bits result, we compute one more bit. | |
1896 | * Together with the remainder, we can decide whether the | |
1897 | * result is exact, bigger than 1/2ulp, or less than 1/2ulp | |
1898 | * (it will never equal to 1/2ulp). | |
1899 | * The rounding mode can be detected by checking whether | |
1900 | * huge + tiny is equal to huge, and whether huge - tiny is | |
1901 | * equal to huge for some floating point number "huge" and "tiny". | |
1902 | * | |
1903 | * Special cases: | |
1904 | * sqrt(+-0) = +-0 ... exact | |
1905 | * sqrt(inf) = inf | |
1906 | * sqrt(-ve) = NaN ... with invalid signal | |
1907 | * sqrt(NaN) = NaN ... with invalid signal for signaling NaN | |
1908 | * | |
1909 | * Other methods : see the appended file at the end of the program below. | |
1910 | *--------------- | |
1911 | */ | |
1912 | ||
1913 | { | |
1914 | /* generate sqrt(x) bit by bit */ | |
1915 | unsigned64 y; | |
1916 | unsigned64 q; | |
1917 | unsigned64 s; | |
1918 | unsigned64 b; | |
1919 | ||
1920 | f->class = sim_fpu_class_number; | |
1921 | f->sign = 0; | |
1922 | y = r->fraction; | |
1923 | f->normal_exp = (r->normal_exp >> 1); /* exp = [exp/2] */ | |
1924 | ||
1925 | /* odd exp, double x to make it even */ | |
1926 | ASSERT (y >= IMPLICIT_1 && y < IMPLICIT_4); | |
1927 | if ((r->normal_exp & 1)) | |
07b4c0a6 | 1928 | { |
9655c439 AC |
1929 | y += y; |
1930 | } | |
1931 | ASSERT (y >= IMPLICIT_1 && y < (IMPLICIT_2 << 1)); | |
1932 | ||
1933 | /* Let loop determine first value of s (either 1 or 2) */ | |
1934 | b = IMPLICIT_1; | |
1935 | q = 0; | |
1936 | s = 0; | |
1937 | ||
1938 | while (b) | |
1939 | { | |
1940 | unsigned64 t = s + b; | |
1941 | if (t <= y) | |
07b4c0a6 | 1942 | { |
9655c439 AC |
1943 | s |= (b << 1); |
1944 | y -= t; | |
1945 | q |= b; | |
07b4c0a6 | 1946 | } |
9655c439 AC |
1947 | y <<= 1; |
1948 | b >>= 1; | |
07b4c0a6 AC |
1949 | } |
1950 | ||
9655c439 AC |
1951 | ASSERT (q >= IMPLICIT_1 && q < IMPLICIT_2); |
1952 | f->fraction = q; | |
1953 | if (y != 0) | |
1954 | { | |
1955 | f->fraction |= 1; /* stick remaining bits */ | |
1956 | return sim_fpu_status_inexact; | |
1957 | } | |
1958 | else | |
1959 | return 0; | |
07b4c0a6 | 1960 | } |
3971886a AC |
1961 | } |
1962 | ||
1963 | ||
9655c439 AC |
1964 | /* int/long <-> sim_fpu */ |
1965 | ||
1966 | INLINE_SIM_FPU (int) | |
1967 | sim_fpu_i32to (sim_fpu *f, | |
1968 | signed32 i, | |
1969 | sim_fpu_round round) | |
3971886a | 1970 | { |
9655c439 AC |
1971 | i2fpu (f, i, 0); |
1972 | return 0; | |
3971886a AC |
1973 | } |
1974 | ||
9655c439 AC |
1975 | INLINE_SIM_FPU (int) |
1976 | sim_fpu_u32to (sim_fpu *f, | |
1977 | unsigned32 u, | |
1978 | sim_fpu_round round) | |
1979 | { | |
1980 | u2fpu (f, u, 0); | |
1981 | return 0; | |
1982 | } | |
3971886a | 1983 | |
9655c439 AC |
1984 | INLINE_SIM_FPU (int) |
1985 | sim_fpu_i64to (sim_fpu *f, | |
1986 | signed64 i, | |
1987 | sim_fpu_round round) | |
3971886a | 1988 | { |
9655c439 AC |
1989 | i2fpu (f, i, 1); |
1990 | return 0; | |
3971886a AC |
1991 | } |
1992 | ||
9655c439 AC |
1993 | INLINE_SIM_FPU (int) |
1994 | sim_fpu_u64to (sim_fpu *f, | |
1995 | unsigned64 u, | |
1996 | sim_fpu_round round) | |
1997 | { | |
1998 | u2fpu (f, u, 1); | |
1999 | return 0; | |
2000 | } | |
3971886a | 2001 | |
d24f06ee | 2002 | |
9655c439 AC |
2003 | INLINE_SIM_FPU (int) |
2004 | sim_fpu_to32i (signed32 *i, | |
2005 | const sim_fpu *f, | |
2006 | sim_fpu_round round) | |
3971886a | 2007 | { |
9655c439 AC |
2008 | signed64 i64; |
2009 | int status = fpu2i (&i64, f, 0, round); | |
2010 | *i = i64; | |
2011 | return status; | |
3971886a AC |
2012 | } |
2013 | ||
9655c439 AC |
2014 | INLINE_SIM_FPU (int) |
2015 | sim_fpu_to32u (unsigned32 *u, | |
2016 | const sim_fpu *f, | |
2017 | sim_fpu_round round) | |
2018 | { | |
2019 | unsigned64 u64; | |
2020 | int status = fpu2u (&u64, f, 0); | |
2021 | *u = u64; | |
2022 | return status; | |
2023 | } | |
3971886a | 2024 | |
9655c439 AC |
2025 | INLINE_SIM_FPU (int) |
2026 | sim_fpu_to64i (signed64 *i, | |
2027 | const sim_fpu *f, | |
2028 | sim_fpu_round round) | |
07b4c0a6 | 2029 | { |
9655c439 | 2030 | return fpu2i (i, f, 1, round); |
07b4c0a6 AC |
2031 | } |
2032 | ||
2033 | ||
9655c439 AC |
2034 | INLINE_SIM_FPU (int) |
2035 | sim_fpu_to64u (unsigned64 *u, | |
2036 | const sim_fpu *f, | |
2037 | sim_fpu_round round) | |
3971886a | 2038 | { |
9655c439 | 2039 | return fpu2u (u, f, 1); |
3971886a AC |
2040 | } |
2041 | ||
2042 | ||
9655c439 AC |
2043 | |
2044 | /* sim_fpu -> host format */ | |
2045 | ||
2046 | #if 0 | |
2047 | INLINE_SIM_FPU (float) | |
2048 | sim_fpu_2f (const sim_fpu *f) | |
07b4c0a6 | 2049 | { |
9655c439 | 2050 | return fval.d; |
07b4c0a6 | 2051 | } |
9655c439 | 2052 | #endif |
07b4c0a6 AC |
2053 | |
2054 | ||
9655c439 AC |
2055 | INLINE_SIM_FPU (double) |
2056 | sim_fpu_2d (const sim_fpu *s) | |
3971886a | 2057 | { |
9655c439 AC |
2058 | sim_fpu_map val; |
2059 | val.i = pack_fpu (s, 1); | |
2060 | return val.d; | |
3971886a AC |
2061 | } |
2062 | ||
2063 | ||
9655c439 AC |
2064 | #if 0 |
2065 | INLINE_SIM_FPU (void) | |
2066 | sim_fpu_f2 (sim_fpu *f, | |
2067 | float s) | |
07b4c0a6 | 2068 | { |
9655c439 AC |
2069 | sim_fpu_map val; |
2070 | val.d = s; | |
2071 | unpack_fpu (f, val.i, 1); | |
07b4c0a6 | 2072 | } |
9655c439 | 2073 | #endif |
07b4c0a6 AC |
2074 | |
2075 | ||
9655c439 AC |
2076 | INLINE_SIM_FPU (void) |
2077 | sim_fpu_d2 (sim_fpu *f, | |
2078 | double d) | |
3971886a | 2079 | { |
9655c439 AC |
2080 | sim_fpu_map val; |
2081 | val.d = d; | |
2082 | unpack_fpu (f, val.i, 1); | |
3971886a AC |
2083 | } |
2084 | ||
2085 | ||
9655c439 AC |
2086 | /* General */ |
2087 | ||
2088 | INLINE_SIM_FPU (int) | |
2089 | sim_fpu_is_nan (const sim_fpu *d) | |
07b4c0a6 | 2090 | { |
9655c439 AC |
2091 | switch (d->class) |
2092 | { | |
2093 | case sim_fpu_class_qnan: | |
2094 | case sim_fpu_class_snan: | |
2095 | return 1; | |
2096 | default: | |
2097 | return 0; | |
2098 | } | |
07b4c0a6 AC |
2099 | } |
2100 | ||
9655c439 AC |
2101 | INLINE_SIM_FPU (int) |
2102 | sim_fpu_is_qnan (const sim_fpu *d) | |
2103 | { | |
2104 | switch (d->class) | |
2105 | { | |
2106 | case sim_fpu_class_qnan: | |
2107 | return 1; | |
2108 | default: | |
2109 | return 0; | |
2110 | } | |
2111 | } | |
07b4c0a6 | 2112 | |
9655c439 AC |
2113 | INLINE_SIM_FPU (int) |
2114 | sim_fpu_is_snan (const sim_fpu *d) | |
2115 | { | |
2116 | switch (d->class) | |
2117 | { | |
2118 | case sim_fpu_class_snan: | |
2119 | return 1; | |
2120 | default: | |
2121 | return 0; | |
2122 | } | |
2123 | } | |
d24f06ee | 2124 | |
9655c439 AC |
2125 | INLINE_SIM_FPU (int) |
2126 | sim_fpu_is_zero (const sim_fpu *d) | |
3971886a | 2127 | { |
9655c439 AC |
2128 | switch (d->class) |
2129 | { | |
2130 | case sim_fpu_class_zero: | |
2131 | return 1; | |
2132 | default: | |
2133 | return 0; | |
2134 | } | |
3971886a AC |
2135 | } |
2136 | ||
9655c439 AC |
2137 | INLINE_SIM_FPU (int) |
2138 | sim_fpu_is_infinity (const sim_fpu *d) | |
2139 | { | |
2140 | switch (d->class) | |
2141 | { | |
2142 | case sim_fpu_class_infinity: | |
2143 | return 1; | |
2144 | default: | |
2145 | return 0; | |
2146 | } | |
2147 | } | |
3971886a | 2148 | |
9655c439 AC |
2149 | INLINE_SIM_FPU (int) |
2150 | sim_fpu_is_number (const sim_fpu *d) | |
2151 | { | |
2152 | switch (d->class) | |
2153 | { | |
bdfe5c04 | 2154 | case sim_fpu_class_denorm: |
9655c439 AC |
2155 | case sim_fpu_class_number: |
2156 | return 1; | |
2157 | default: | |
2158 | return 0; | |
2159 | } | |
2160 | } | |
2161 | ||
bdfe5c04 AC |
2162 | INLINE_SIM_FPU (int) |
2163 | sim_fpu_is_denorm (const sim_fpu *d) | |
2164 | { | |
2165 | switch (d->class) | |
2166 | { | |
2167 | case sim_fpu_class_denorm: | |
2168 | return 1; | |
2169 | default: | |
2170 | return 0; | |
2171 | } | |
2172 | } | |
2173 | ||
390ffa89 AC |
2174 | |
2175 | INLINE_SIM_FPU (int) | |
2176 | sim_fpu_sign (const sim_fpu *d) | |
2177 | { | |
2178 | return d->sign; | |
2179 | } | |
2180 | ||
2181 | ||
2182 | INLINE_SIM_FPU (int) | |
2183 | sim_fpu_exp (const sim_fpu *d) | |
2184 | { | |
2185 | return d->normal_exp; | |
2186 | } | |
2187 | ||
2188 | ||
9655c439 AC |
2189 | INLINE_SIM_FPU (int) |
2190 | sim_fpu_is (const sim_fpu *d) | |
3971886a | 2191 | { |
9655c439 AC |
2192 | switch (d->class) |
2193 | { | |
2194 | case sim_fpu_class_qnan: | |
2195 | return SIM_FPU_IS_QNAN; | |
2196 | case sim_fpu_class_snan: | |
2197 | return SIM_FPU_IS_SNAN; | |
2198 | case sim_fpu_class_infinity: | |
2199 | return SIM_FPU_IS_NINF; | |
2200 | return SIM_FPU_IS_PINF; | |
2201 | case sim_fpu_class_number: | |
2202 | if (d->sign) | |
bdfe5c04 | 2203 | return SIM_FPU_IS_NNUMBER; |
9655c439 | 2204 | else |
bdfe5c04 AC |
2205 | return SIM_FPU_IS_PNUMBER; |
2206 | case sim_fpu_class_denorm: | |
9655c439 AC |
2207 | if (d->sign) |
2208 | return SIM_FPU_IS_NDENORM; | |
2209 | else | |
2210 | return SIM_FPU_IS_PDENORM; | |
9655c439 AC |
2211 | case sim_fpu_class_zero: |
2212 | if (d->sign) | |
2213 | return SIM_FPU_IS_NZERO; | |
2214 | else | |
2215 | return SIM_FPU_IS_PZERO; | |
2216 | default: | |
2217 | return -1; | |
2218 | abort (); | |
2219 | } | |
3971886a AC |
2220 | } |
2221 | ||
9655c439 AC |
2222 | INLINE_SIM_FPU (int) |
2223 | sim_fpu_cmp (const sim_fpu *l, const sim_fpu *r) | |
2224 | { | |
2225 | sim_fpu res; | |
2226 | sim_fpu_sub (&res, l, r); | |
2227 | return sim_fpu_is (&res); | |
2228 | } | |
3971886a | 2229 | |
9655c439 AC |
2230 | INLINE_SIM_FPU (int) |
2231 | sim_fpu_is_lt (const sim_fpu *l, const sim_fpu *r) | |
3971886a | 2232 | { |
9655c439 AC |
2233 | int status; |
2234 | sim_fpu_lt (&status, l, r); | |
2235 | return status; | |
3971886a AC |
2236 | } |
2237 | ||
9655c439 AC |
2238 | INLINE_SIM_FPU (int) |
2239 | sim_fpu_is_le (const sim_fpu *l, const sim_fpu *r) | |
2240 | { | |
2241 | int is; | |
2242 | sim_fpu_le (&is, l, r); | |
2243 | return is; | |
2244 | } | |
3971886a | 2245 | |
9655c439 AC |
2246 | INLINE_SIM_FPU (int) |
2247 | sim_fpu_is_eq (const sim_fpu *l, const sim_fpu *r) | |
3971886a | 2248 | { |
9655c439 AC |
2249 | int is; |
2250 | sim_fpu_eq (&is, l, r); | |
2251 | return is; | |
3971886a | 2252 | } |
d24f06ee | 2253 | |
9655c439 AC |
2254 | INLINE_SIM_FPU (int) |
2255 | sim_fpu_is_ne (const sim_fpu *l, const sim_fpu *r) | |
2256 | { | |
2257 | int is; | |
2258 | sim_fpu_ne (&is, l, r); | |
2259 | return is; | |
2260 | } | |
3971886a | 2261 | |
9655c439 AC |
2262 | INLINE_SIM_FPU (int) |
2263 | sim_fpu_is_ge (const sim_fpu *l, const sim_fpu *r) | |
2264 | { | |
2265 | int is; | |
2266 | sim_fpu_ge (&is, l, r); | |
2267 | return is; | |
2268 | } | |
3971886a AC |
2269 | |
2270 | INLINE_SIM_FPU (int) | |
9655c439 | 2271 | sim_fpu_is_gt (const sim_fpu *l, const sim_fpu *r) |
3971886a | 2272 | { |
9655c439 AC |
2273 | int is; |
2274 | sim_fpu_gt (&is, l, r); | |
2275 | return is; | |
3971886a AC |
2276 | } |
2277 | ||
2278 | ||
d24f06ee AC |
2279 | /* Compare operators */ |
2280 | ||
3971886a | 2281 | INLINE_SIM_FPU (int) |
9655c439 AC |
2282 | sim_fpu_lt (int *is, |
2283 | const sim_fpu *l, | |
2284 | const sim_fpu *r) | |
d24f06ee | 2285 | { |
9655c439 AC |
2286 | if (!sim_fpu_is_nan (l) && !sim_fpu_is_nan (r)) |
2287 | { | |
2288 | sim_fpu_map lval; | |
2289 | sim_fpu_map rval; | |
2290 | lval.i = pack_fpu (l, 1); | |
2291 | rval.i = pack_fpu (r, 1); | |
2292 | (*is) = (lval.d < rval.d); | |
2293 | return 0; | |
2294 | } | |
2295 | else if (sim_fpu_is_snan (l) || sim_fpu_is_snan (r)) | |
2296 | { | |
2297 | *is = 0; | |
2298 | return sim_fpu_status_invalid_snan; | |
2299 | } | |
07b4c0a6 | 2300 | else |
9655c439 AC |
2301 | { |
2302 | *is = 0; | |
2303 | return sim_fpu_status_invalid_qnan; | |
2304 | } | |
d24f06ee AC |
2305 | } |
2306 | ||
2307 | INLINE_SIM_FPU (int) | |
9655c439 AC |
2308 | sim_fpu_le (int *is, |
2309 | const sim_fpu *l, | |
2310 | const sim_fpu *r) | |
d24f06ee | 2311 | { |
9655c439 AC |
2312 | if (!sim_fpu_is_nan (l) && !sim_fpu_is_nan (r)) |
2313 | { | |
2314 | sim_fpu_map lval; | |
2315 | sim_fpu_map rval; | |
2316 | lval.i = pack_fpu (l, 1); | |
2317 | rval.i = pack_fpu (r, 1); | |
2318 | *is = (lval.d <= rval.d); | |
2319 | return 0; | |
2320 | } | |
2321 | else if (sim_fpu_is_snan (l) || sim_fpu_is_snan (r)) | |
2322 | { | |
2323 | *is = 0; | |
2324 | return sim_fpu_status_invalid_snan; | |
2325 | } | |
07b4c0a6 | 2326 | else |
9655c439 AC |
2327 | { |
2328 | *is = 0; | |
2329 | return sim_fpu_status_invalid_qnan; | |
2330 | } | |
d24f06ee AC |
2331 | } |
2332 | ||
2333 | INLINE_SIM_FPU (int) | |
9655c439 AC |
2334 | sim_fpu_eq (int *is, |
2335 | const sim_fpu *l, | |
2336 | const sim_fpu *r) | |
d24f06ee | 2337 | { |
9655c439 AC |
2338 | if (!sim_fpu_is_nan (l) && !sim_fpu_is_nan (r)) |
2339 | { | |
2340 | sim_fpu_map lval; | |
2341 | sim_fpu_map rval; | |
2342 | lval.i = pack_fpu (l, 1); | |
2343 | rval.i = pack_fpu (r, 1); | |
2344 | (*is) = (lval.d == rval.d); | |
2345 | return 0; | |
2346 | } | |
2347 | else if (sim_fpu_is_snan (l) || sim_fpu_is_snan (r)) | |
2348 | { | |
2349 | *is = 0; | |
2350 | return sim_fpu_status_invalid_snan; | |
2351 | } | |
07b4c0a6 | 2352 | else |
9655c439 AC |
2353 | { |
2354 | *is = 0; | |
2355 | return sim_fpu_status_invalid_qnan; | |
2356 | } | |
d24f06ee AC |
2357 | } |
2358 | ||
2359 | INLINE_SIM_FPU (int) | |
9655c439 AC |
2360 | sim_fpu_ne (int *is, |
2361 | const sim_fpu *l, | |
2362 | const sim_fpu *r) | |
d24f06ee | 2363 | { |
9655c439 AC |
2364 | if (!sim_fpu_is_nan (l) && !sim_fpu_is_nan (r)) |
2365 | { | |
2366 | sim_fpu_map lval; | |
2367 | sim_fpu_map rval; | |
2368 | lval.i = pack_fpu (l, 1); | |
2369 | rval.i = pack_fpu (r, 1); | |
2370 | (*is) = (lval.d != rval.d); | |
2371 | return 0; | |
2372 | } | |
2373 | else if (sim_fpu_is_snan (l) || sim_fpu_is_snan (r)) | |
2374 | { | |
2375 | *is = 0; | |
2376 | return sim_fpu_status_invalid_snan; | |
2377 | } | |
07b4c0a6 | 2378 | else |
9655c439 AC |
2379 | { |
2380 | *is = 0; | |
2381 | return sim_fpu_status_invalid_qnan; | |
2382 | } | |
d24f06ee AC |
2383 | } |
2384 | ||
2385 | INLINE_SIM_FPU (int) | |
9655c439 AC |
2386 | sim_fpu_ge (int *is, |
2387 | const sim_fpu *l, | |
2388 | const sim_fpu *r) | |
d24f06ee | 2389 | { |
9655c439 | 2390 | return sim_fpu_le (is, r, l); |
d24f06ee AC |
2391 | } |
2392 | ||
2393 | INLINE_SIM_FPU (int) | |
9655c439 AC |
2394 | sim_fpu_gt (int *is, |
2395 | const sim_fpu *l, | |
2396 | const sim_fpu *r) | |
3971886a | 2397 | { |
9655c439 AC |
2398 | return sim_fpu_lt (is, r, l); |
2399 | } | |
2400 | ||
2401 | ||
2402 | /* A number of useful constants */ | |
2403 | ||
278bda40 AC |
2404 | EXTERN_SIM_FPU (const sim_fpu) sim_fpu_zero = { |
2405 | sim_fpu_class_zero, | |
2406 | }; | |
2407 | EXTERN_SIM_FPU (const sim_fpu) sim_fpu_qnan = { | |
2408 | sim_fpu_class_qnan, | |
2409 | }; | |
2410 | EXTERN_SIM_FPU (const sim_fpu) sim_fpu_one = { | |
2411 | sim_fpu_class_number, 0, IMPLICIT_1, 1 | |
2412 | }; | |
2413 | EXTERN_SIM_FPU (const sim_fpu) sim_fpu_two = { | |
2414 | sim_fpu_class_number, 0, IMPLICIT_1, 2 | |
2415 | }; | |
2416 | EXTERN_SIM_FPU (const sim_fpu) sim_fpu_max32 = { | |
2417 | sim_fpu_class_number, 0, LSMASK64 (NR_FRAC_GUARD, NR_GUARDS32), NORMAL_EXPMAX32 | |
2418 | }; | |
2419 | EXTERN_SIM_FPU (const sim_fpu) sim_fpu_max64 = { | |
2420 | sim_fpu_class_number, 0, LSMASK64 (NR_FRAC_GUARD, NR_GUARDS64), NORMAL_EXPMAX64 | |
2421 | }; | |
9655c439 AC |
2422 | |
2423 | ||
2424 | /* For debugging */ | |
2425 | ||
2426 | INLINE_SIM_FPU (void) | |
2427 | sim_fpu_print_fpu (const sim_fpu *f, | |
2428 | sim_fpu_print_func *print, | |
2429 | void *arg) | |
2430 | { | |
2431 | print (arg, "%s", f->sign ? "-" : "+"); | |
2432 | switch (f->class) | |
2433 | { | |
2434 | case sim_fpu_class_qnan: | |
2435 | print (arg, "0."); | |
2436 | print_bits (f->fraction, NR_FRAC_GUARD - 1, print, arg); | |
2437 | print (arg, "*QuietNaN"); | |
2438 | break; | |
2439 | case sim_fpu_class_snan: | |
2440 | print (arg, "0."); | |
2441 | print_bits (f->fraction, NR_FRAC_GUARD - 1, print, arg); | |
2442 | print (arg, "*SignalNaN"); | |
2443 | break; | |
2444 | case sim_fpu_class_zero: | |
2445 | print (arg, "0.0"); | |
2446 | break; | |
2447 | case sim_fpu_class_infinity: | |
2448 | print (arg, "INF"); | |
2449 | break; | |
2450 | case sim_fpu_class_number: | |
bdfe5c04 | 2451 | case sim_fpu_class_denorm: |
9655c439 AC |
2452 | print (arg, "1."); |
2453 | print_bits (f->fraction, NR_FRAC_GUARD - 1, print, arg); | |
2454 | print (arg, "*2^%+-5d", f->normal_exp); | |
2455 | ASSERT (f->fraction >= IMPLICIT_1); | |
2456 | ASSERT (f->fraction < IMPLICIT_2); | |
2457 | } | |
2458 | } | |
2459 | ||
2460 | ||
2461 | INLINE_SIM_FPU (void) | |
2462 | sim_fpu_print_status (int status, | |
2463 | sim_fpu_print_func *print, | |
2464 | void *arg) | |
2465 | { | |
2466 | int i = 1; | |
2467 | char *prefix = ""; | |
2468 | while (status >= i) | |
2469 | { | |
2470 | switch ((sim_fpu_status) (status & i)) | |
2471 | { | |
2472 | case sim_fpu_status_denorm: | |
2473 | print (arg, "%sD", prefix); | |
2474 | break; | |
2475 | case sim_fpu_status_invalid_snan: | |
2476 | print (arg, "%sSNaN", prefix); | |
2477 | break; | |
2478 | case sim_fpu_status_invalid_qnan: | |
2479 | print (arg, "%sQNaN", prefix); | |
2480 | break; | |
2481 | case sim_fpu_status_invalid_isi: | |
2482 | print (arg, "%sISI", prefix); | |
2483 | break; | |
2484 | case sim_fpu_status_invalid_idi: | |
2485 | print (arg, "%sIDI", prefix); | |
2486 | break; | |
2487 | case sim_fpu_status_invalid_zdz: | |
2488 | print (arg, "%sZDZ", prefix); | |
2489 | break; | |
2490 | case sim_fpu_status_invalid_imz: | |
2491 | print (arg, "%sIMZ", prefix); | |
2492 | break; | |
2493 | case sim_fpu_status_invalid_cvi: | |
2494 | print (arg, "%sCVI", prefix); | |
2495 | break; | |
2496 | case sim_fpu_status_invalid_cmp: | |
2497 | print (arg, "%sCMP", prefix); | |
2498 | break; | |
2499 | case sim_fpu_status_invalid_sqrt: | |
2500 | print (arg, "%sSQRT", prefix); | |
2501 | break; | |
2502 | break; | |
2503 | case sim_fpu_status_inexact: | |
2504 | print (arg, "%sX", prefix); | |
2505 | break; | |
2506 | break; | |
2507 | case sim_fpu_status_overflow: | |
2508 | print (arg, "%sO", prefix); | |
2509 | break; | |
2510 | break; | |
2511 | case sim_fpu_status_underflow: | |
2512 | print (arg, "%sU", prefix); | |
2513 | break; | |
2514 | break; | |
2515 | case sim_fpu_status_invalid_div0: | |
2516 | print (arg, "%s/", prefix); | |
2517 | break; | |
2518 | break; | |
2519 | case sim_fpu_status_rounded: | |
2520 | print (arg, "%sR", prefix); | |
2521 | break; | |
2522 | break; | |
2523 | } | |
2524 | i <<= 1; | |
2525 | prefix = ","; | |
2526 | } | |
3971886a AC |
2527 | } |
2528 | ||
2529 | #endif |