1 /*============================================================================
2 This C source fragment is part of the SoftFloat IEC/IEEE Floating-point
3 Arithmetic Package, Release 2b.
4
5 Written by John R. Hauser. This work was made possible in part by the
6 International Computer Science Institute, located at Suite 600, 1947 Center
7 Street, Berkeley, California 94704. Funding was partially provided by the
8 National Science Foundation under grant MIP-9311980. The original version
9 of this code was written as part of a project to build a fixed-point vector
10 processor in collaboration with the University of California at Berkeley,
11 overseen by Profs. Nelson Morgan and John Wawrzynek. More information
12 is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
13 arithmetic/SoftFloat.html'.
14
15 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
16 been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
17 RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
18 AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
19 COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
20 EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
21 INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES, COSTS, OR
22 OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
23
24 Derivative works are acceptable, even for commercial purposes, so long as
25 (1) the source code for the derivative work includes prominent notice that
26 the work is derivative, and (2) the source code includes prominent notice with
27 these four paragraphs for those parts of this code that are retained.
28 =============================================================================*/
29
30 #ifndef _SOFTFLOAT_SPECIALIZE_H_
31 #define _SOFTFLOAT_SPECIALIZE_H_
32
33 #include "softfloat.h"
34
35 /*============================================================================
36 * Adapted for Bochs (x86 achitecture simulator) by
37 * Stanislav Shwartsman [sshwarts at sourceforge net]
38 * ==========================================================================*/
39
40 #define int16_indefinite ((Bit16s)0x8000)
41 #define int32_indefinite ((Bit32s)0x80000000)
42 #define int64_indefinite BX_CONST64(0x8000000000000000)
43
44 #define uint16_indefinite (0xffff)
45 #define uint32_indefinite (0xffffffff)
46 #define uint64_indefinite BX_CONST64(0xffffffffffffffff)
47
48 /*----------------------------------------------------------------------------
49 | Internal canonical NaN format.
50 *----------------------------------------------------------------------------*/
51
52 typedef struct {
53 int sign;
54 Bit64u hi, lo;
55 } commonNaNT;
56
57 #ifdef FLOAT16
58
59 /*----------------------------------------------------------------------------
60 | The pattern for a default generated half-precision NaN.
61 *----------------------------------------------------------------------------*/
62 const float16 float16_default_nan = 0xFE00;
63
64 #define float16_fraction extractFloat16Frac
65 #define float16_exp extractFloat16Exp
66 #define float16_sign extractFloat16Sign
67
68 /*----------------------------------------------------------------------------
69 | Returns the fraction bits of the half-precision floating-point value `a'.
70 *----------------------------------------------------------------------------*/
71
extractFloat16Frac(float16 a)72 BX_CPP_INLINE Bit16u extractFloat16Frac(float16 a)
73 {
74 return a & 0x3FF;
75 }
76
77 /*----------------------------------------------------------------------------
78 | Returns the exponent bits of the half-precision floating-point value `a'.
79 *----------------------------------------------------------------------------*/
80
extractFloat16Exp(float16 a)81 BX_CPP_INLINE Bit16s extractFloat16Exp(float16 a)
82 {
83 return (a>>10) & 0x1F;
84 }
85
86 /*----------------------------------------------------------------------------
87 | Returns the sign bit of the half-precision floating-point value `a'.
88 *----------------------------------------------------------------------------*/
89
extractFloat16Sign(float16 a)90 BX_CPP_INLINE int extractFloat16Sign(float16 a)
91 {
92 return a>>15;
93 }
94
95 /*----------------------------------------------------------------------------
96 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
97 | single-precision floating-point value, returning the result. After being
98 | shifted into the proper positions, the three fields are simply added
99 | together to form the result. This means that any integer portion of `zSig'
100 | will be added into the exponent. Since a properly normalized significand
101 | will have an integer portion equal to 1, the `zExp' input should be 1 less
102 | than the desired result exponent whenever `zSig' is a complete, normalized
103 | significand.
104 *----------------------------------------------------------------------------*/
105
packFloat16(int zSign,int zExp,Bit16u zSig)106 BX_CPP_INLINE float16 packFloat16(int zSign, int zExp, Bit16u zSig)
107 {
108 return (((Bit16u) zSign)<<15) + (((Bit16u) zExp)<<10) + zSig;
109 }
110
111 /*----------------------------------------------------------------------------
112 | Returns 1 if the half-precision floating-point value `a' is a NaN;
113 | otherwise returns 0.
114 *----------------------------------------------------------------------------*/
115
float16_is_nan(float16 a)116 BX_CPP_INLINE int float16_is_nan(float16 a)
117 {
118 return (0xF800 < (Bit16u) (a<<1));
119 }
120
121 /*----------------------------------------------------------------------------
122 | Returns 1 if the half-precision floating-point value `a' is a signaling
123 | NaN; otherwise returns 0.
124 *----------------------------------------------------------------------------*/
125
float16_is_signaling_nan(float16 a)126 BX_CPP_INLINE int float16_is_signaling_nan(float16 a)
127 {
128 return (((a>>9) & 0x3F) == 0x3E) && (a & 0x1FF);
129 }
130
131 /*----------------------------------------------------------------------------
132 | Returns 1 if the half-precision floating-point value `a' is denormal;
133 | otherwise returns 0.
134 *----------------------------------------------------------------------------*/
135
float16_is_denormal(float16 a)136 BX_CPP_INLINE int float16_is_denormal(float16 a)
137 {
138 return (extractFloat16Exp(a) == 0) && (extractFloat16Frac(a) != 0);
139 }
140
141 /*----------------------------------------------------------------------------
142 | Convert float16 denormals to zero.
143 *----------------------------------------------------------------------------*/
144
float16_denormal_to_zero(float16 a)145 BX_CPP_INLINE float16 float16_denormal_to_zero(float16 a)
146 {
147 if (float16_is_denormal(a)) a &= 0x8000;
148 return a;
149 }
150
151 /*----------------------------------------------------------------------------
152 | Returns the result of converting the half-precision floating-point NaN
153 | `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
154 | exception is raised.
155 *----------------------------------------------------------------------------*/
156
float16ToCommonNaN(float16 a,float_status_t & status)157 BX_CPP_INLINE commonNaNT float16ToCommonNaN(float16 a, float_status_t &status)
158 {
159 commonNaNT z;
160 if (float16_is_signaling_nan(a)) float_raise(status, float_flag_invalid);
161 z.sign = a>>15;
162 z.lo = 0;
163 z.hi = ((Bit64u) a)<<54;
164 return z;
165 }
166
167 /*----------------------------------------------------------------------------
168 | Returns the result of converting the canonical NaN `a' to the half-
169 | precision floating-point format.
170 *----------------------------------------------------------------------------*/
171
commonNaNToFloat16(commonNaNT a)172 BX_CPP_INLINE float16 commonNaNToFloat16(commonNaNT a)
173 {
174 return (((Bit16u) a.sign)<<15) | 0x7E00 | (Bit16u)(a.hi>>54);
175 }
176
177 #endif
178
179 /*----------------------------------------------------------------------------
180 | Commonly used single-precision floating point constants
181 *----------------------------------------------------------------------------*/
182 const float32 float32_negative_inf = 0xff800000;
183 const float32 float32_positive_inf = 0x7f800000;
184 const float32 float32_negative_zero = 0x80000000;
185 const float32 float32_positive_zero = 0x00000000;
186 const float32 float32_negative_one = 0xbf800000;
187 const float32 float32_positive_one = 0x3f800000;
188 const float32 float32_max_float = 0x7f7fffff;
189 const float32 float32_min_float = 0xff7fffff;
190
191 /*----------------------------------------------------------------------------
192 | The pattern for a default generated single-precision NaN.
193 *----------------------------------------------------------------------------*/
194 const float32 float32_default_nan = 0xffc00000;
195
196 #define float32_fraction extractFloat32Frac
197 #define float32_exp extractFloat32Exp
198 #define float32_sign extractFloat32Sign
199
200 #define FLOAT32_EXP_BIAS 0x7F
201
202 /*----------------------------------------------------------------------------
203 | Returns the fraction bits of the single-precision floating-point value `a'.
204 *----------------------------------------------------------------------------*/
205
extractFloat32Frac(float32 a)206 BX_CPP_INLINE Bit32u extractFloat32Frac(float32 a)
207 {
208 return a & 0x007FFFFF;
209 }
210
211 /*----------------------------------------------------------------------------
212 | Returns the exponent bits of the single-precision floating-point value `a'.
213 *----------------------------------------------------------------------------*/
214
extractFloat32Exp(float32 a)215 BX_CPP_INLINE Bit16s extractFloat32Exp(float32 a)
216 {
217 return (a>>23) & 0xFF;
218 }
219
220 /*----------------------------------------------------------------------------
221 | Returns the sign bit of the single-precision floating-point value `a'.
222 *----------------------------------------------------------------------------*/
223
extractFloat32Sign(float32 a)224 BX_CPP_INLINE int extractFloat32Sign(float32 a)
225 {
226 return a>>31;
227 }
228
229 /*----------------------------------------------------------------------------
230 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
231 | single-precision floating-point value, returning the result. After being
232 | shifted into the proper positions, the three fields are simply added
233 | together to form the result. This means that any integer portion of `zSig'
234 | will be added into the exponent. Since a properly normalized significand
235 | will have an integer portion equal to 1, the `zExp' input should be 1 less
236 | than the desired result exponent whenever `zSig' is a complete, normalized
237 | significand.
238 *----------------------------------------------------------------------------*/
239
packFloat32(int zSign,Bit16s zExp,Bit32u zSig)240 BX_CPP_INLINE float32 packFloat32(int zSign, Bit16s zExp, Bit32u zSig)
241 {
242 return (((Bit32u) zSign)<<31) + (((Bit32u) zExp)<<23) + zSig;
243 }
244
245 /*----------------------------------------------------------------------------
246 | Returns 1 if the single-precision floating-point value `a' is a NaN;
247 | otherwise returns 0.
248 *----------------------------------------------------------------------------*/
249
float32_is_nan(float32 a)250 BX_CPP_INLINE int float32_is_nan(float32 a)
251 {
252 return (0xFF000000 < (Bit32u) (a<<1));
253 }
254
255 /*----------------------------------------------------------------------------
256 | Returns 1 if the single-precision floating-point value `a' is a signaling
257 | NaN; otherwise returns 0.
258 *----------------------------------------------------------------------------*/
259
float32_is_signaling_nan(float32 a)260 BX_CPP_INLINE int float32_is_signaling_nan(float32 a)
261 {
262 return (((a>>22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
263 }
264
265 /*----------------------------------------------------------------------------
266 | Returns 1 if the single-precision floating-point value `a' is denormal;
267 | otherwise returns 0.
268 *----------------------------------------------------------------------------*/
269
float32_is_denormal(float32 a)270 BX_CPP_INLINE int float32_is_denormal(float32 a)
271 {
272 return (extractFloat32Exp(a) == 0) && (extractFloat32Frac(a) != 0);
273 }
274
275 /*----------------------------------------------------------------------------
276 | Convert float32 denormals to zero.
277 *----------------------------------------------------------------------------*/
278
float32_denormal_to_zero(float32 a)279 BX_CPP_INLINE float32 float32_denormal_to_zero(float32 a)
280 {
281 if (float32_is_denormal(a)) a &= 0x80000000;
282 return a;
283 }
284
285 /*----------------------------------------------------------------------------
286 | Returns the result of converting the single-precision floating-point NaN
287 | `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
288 | exception is raised.
289 *----------------------------------------------------------------------------*/
290
float32ToCommonNaN(float32 a,float_status_t & status)291 BX_CPP_INLINE commonNaNT float32ToCommonNaN(float32 a, float_status_t &status)
292 {
293 commonNaNT z;
294 if (float32_is_signaling_nan(a)) float_raise(status, float_flag_invalid);
295 z.sign = a>>31;
296 z.lo = 0;
297 z.hi = ((Bit64u) a)<<41;
298 return z;
299 }
300
301 /*----------------------------------------------------------------------------
302 | Returns the result of converting the canonical NaN `a' to the single-
303 | precision floating-point format.
304 *----------------------------------------------------------------------------*/
305
commonNaNToFloat32(commonNaNT a)306 BX_CPP_INLINE float32 commonNaNToFloat32(commonNaNT a)
307 {
308 return (((Bit32u) a.sign)<<31) | 0x7FC00000 | (Bit32u)(a.hi>>41);
309 }
310
311 /*----------------------------------------------------------------------------
312 | Takes two single-precision floating-point values `a' and `b', one of which
313 | is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a
314 | signaling NaN, the invalid exception is raised.
315 *----------------------------------------------------------------------------*/
316
317 float32 propagateFloat32NaN(float32 a, float32 b, float_status_t &status);
318
319 /*----------------------------------------------------------------------------
320 | Takes single-precision floating-point NaN `a' and returns the appropriate
321 | NaN result. If `a' is a signaling NaN, the invalid exception is raised.
322 *----------------------------------------------------------------------------*/
323
propagateFloat32NaN(float32 a,float_status_t & status)324 BX_CPP_INLINE float32 propagateFloat32NaN(float32 a, float_status_t &status)
325 {
326 if (float32_is_signaling_nan(a))
327 float_raise(status, float_flag_invalid);
328
329 return a | 0x00400000;
330 }
331
332 /*----------------------------------------------------------------------------
333 | Commonly used single-precision floating point constants
334 *----------------------------------------------------------------------------*/
335 const float64 float64_negative_inf = BX_CONST64(0xfff0000000000000);
336 const float64 float64_positive_inf = BX_CONST64(0x7ff0000000000000);
337 const float64 float64_negative_zero = BX_CONST64(0x8000000000000000);
338 const float64 float64_positive_zero = BX_CONST64(0x0000000000000000);
339 const float64 float64_negative_one = BX_CONST64(0xbff0000000000000);
340 const float64 float64_positive_one = BX_CONST64(0x3ff0000000000000);
341 const float64 float64_max_float = BX_CONST64(0x7fefffffffffffff);
342 const float64 float64_min_float = BX_CONST64(0xffefffffffffffff);
343
344 /*----------------------------------------------------------------------------
345 | The pattern for a default generated double-precision NaN.
346 *----------------------------------------------------------------------------*/
347 const float64 float64_default_nan = BX_CONST64(0xFFF8000000000000);
348
349 #define float64_fraction extractFloat64Frac
350 #define float64_exp extractFloat64Exp
351 #define float64_sign extractFloat64Sign
352
353 #define FLOAT64_EXP_BIAS 0x3FF
354
355 /*----------------------------------------------------------------------------
356 | Returns the fraction bits of the double-precision floating-point value `a'.
357 *----------------------------------------------------------------------------*/
358
extractFloat64Frac(float64 a)359 BX_CPP_INLINE Bit64u extractFloat64Frac(float64 a)
360 {
361 return a & BX_CONST64(0x000FFFFFFFFFFFFF);
362 }
363
364 /*----------------------------------------------------------------------------
365 | Returns the exponent bits of the double-precision floating-point value `a'.
366 *----------------------------------------------------------------------------*/
367
extractFloat64Exp(float64 a)368 BX_CPP_INLINE Bit16s extractFloat64Exp(float64 a)
369 {
370 return (Bit16s)(a>>52) & 0x7FF;
371 }
372
373 /*----------------------------------------------------------------------------
374 | Returns the sign bit of the double-precision floating-point value `a'.
375 *----------------------------------------------------------------------------*/
376
extractFloat64Sign(float64 a)377 BX_CPP_INLINE int extractFloat64Sign(float64 a)
378 {
379 return (int)(a>>63);
380 }
381
382 /*----------------------------------------------------------------------------
383 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
384 | double-precision floating-point value, returning the result. After being
385 | shifted into the proper positions, the three fields are simply added
386 | together to form the result. This means that any integer portion of `zSig'
387 | will be added into the exponent. Since a properly normalized significand
388 | will have an integer portion equal to 1, the `zExp' input should be 1 less
389 | than the desired result exponent whenever `zSig' is a complete, normalized
390 | significand.
391 *----------------------------------------------------------------------------*/
392
packFloat64(int zSign,Bit16s zExp,Bit64u zSig)393 BX_CPP_INLINE float64 packFloat64(int zSign, Bit16s zExp, Bit64u zSig)
394 {
395 return (((Bit64u) zSign)<<63) + (((Bit64u) zExp)<<52) + zSig;
396 }
397
398 /*----------------------------------------------------------------------------
399 | Returns 1 if the double-precision floating-point value `a' is a NaN;
400 | otherwise returns 0.
401 *----------------------------------------------------------------------------*/
402
float64_is_nan(float64 a)403 BX_CPP_INLINE int float64_is_nan(float64 a)
404 {
405 return (BX_CONST64(0xFFE0000000000000) < (Bit64u) (a<<1));
406 }
407
408 /*----------------------------------------------------------------------------
409 | Returns 1 if the double-precision floating-point value `a' is a signaling
410 | NaN; otherwise returns 0.
411 *----------------------------------------------------------------------------*/
412
float64_is_signaling_nan(float64 a)413 BX_CPP_INLINE int float64_is_signaling_nan(float64 a)
414 {
415 return (((a>>51) & 0xFFF) == 0xFFE) && (a & BX_CONST64(0x0007FFFFFFFFFFFF));
416 }
417
418 /*----------------------------------------------------------------------------
419 | Returns 1 if the double-precision floating-point value `a' is denormal;
420 | otherwise returns 0.
421 *----------------------------------------------------------------------------*/
422
float64_is_denormal(float64 a)423 BX_CPP_INLINE int float64_is_denormal(float64 a)
424 {
425 return (extractFloat64Exp(a) == 0) && (extractFloat64Frac(a) != 0);
426 }
427
428 /*----------------------------------------------------------------------------
429 | Convert float64 denormals to zero.
430 *----------------------------------------------------------------------------*/
431
float64_denormal_to_zero(float64 a)432 BX_CPP_INLINE float64 float64_denormal_to_zero(float64 a)
433 {
434 if (float64_is_denormal(a)) a &= ((Bit64u)(1) << 63);
435 return a;
436 }
437
438 /*----------------------------------------------------------------------------
439 | Returns the result of converting the double-precision floating-point NaN
440 | `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
441 | exception is raised.
442 *----------------------------------------------------------------------------*/
443
float64ToCommonNaN(float64 a,float_status_t & status)444 BX_CPP_INLINE commonNaNT float64ToCommonNaN(float64 a, float_status_t &status)
445 {
446 commonNaNT z;
447 if (float64_is_signaling_nan(a)) float_raise(status, float_flag_invalid);
448 z.sign = (int)(a>>63);
449 z.lo = 0;
450 z.hi = a<<12;
451 return z;
452 }
453
454 /*----------------------------------------------------------------------------
455 | Returns the result of converting the canonical NaN `a' to the double-
456 | precision floating-point format.
457 *----------------------------------------------------------------------------*/
458
commonNaNToFloat64(commonNaNT a)459 BX_CPP_INLINE float64 commonNaNToFloat64(commonNaNT a)
460 {
461 return (((Bit64u) a.sign)<<63) | BX_CONST64(0x7FF8000000000000) | (a.hi>>12);
462 }
463
464 /*----------------------------------------------------------------------------
465 | Takes two double-precision floating-point values `a' and `b', one of which
466 | is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a
467 | signaling NaN, the invalid exception is raised.
468 *----------------------------------------------------------------------------*/
469
470 float64 propagateFloat64NaN(float64 a, float64 b, float_status_t &status);
471
472 /*----------------------------------------------------------------------------
473 | Takes double-precision floating-point NaN `a' and returns the appropriate
474 | NaN result. If `a' is a signaling NaN, the invalid exception is raised.
475 *----------------------------------------------------------------------------*/
476
propagateFloat64NaN(float64 a,float_status_t & status)477 BX_CPP_INLINE float64 propagateFloat64NaN(float64 a, float_status_t &status)
478 {
479 if (float64_is_signaling_nan(a))
480 float_raise(status, float_flag_invalid);
481
482 return a | BX_CONST64(0x0008000000000000);
483 }
484
485 #ifdef FLOATX80
486
487 /*----------------------------------------------------------------------------
488 | The pattern for a default generated extended double-precision NaN. The
489 | `high' and `low' values hold the most- and least-significant bits,
490 | respectively.
491 *----------------------------------------------------------------------------*/
492 #define floatx80_default_nan_exp 0xFFFF
493 #define floatx80_default_nan_fraction BX_CONST64(0xC000000000000000)
494
495 #define floatx80_fraction extractFloatx80Frac
496 #define floatx80_exp extractFloatx80Exp
497 #define floatx80_sign extractFloatx80Sign
498
499 #define FLOATX80_EXP_BIAS 0x3FFF
500
501 /*----------------------------------------------------------------------------
502 | Returns the fraction bits of the extended double-precision floating-point
503 | value `a'.
504 *----------------------------------------------------------------------------*/
505
extractFloatx80Frac(floatx80 a)506 BX_CPP_INLINE Bit64u extractFloatx80Frac(floatx80 a)
507 {
508 return a.fraction;
509 }
510
511 /*----------------------------------------------------------------------------
512 | Returns the exponent bits of the extended double-precision floating-point
513 | value `a'.
514 *----------------------------------------------------------------------------*/
515
extractFloatx80Exp(floatx80 a)516 BX_CPP_INLINE Bit32s extractFloatx80Exp(floatx80 a)
517 {
518 return a.exp & 0x7FFF;
519 }
520
521 /*----------------------------------------------------------------------------
522 | Returns the sign bit of the extended double-precision floating-point value
523 | `a'.
524 *----------------------------------------------------------------------------*/
525
extractFloatx80Sign(floatx80 a)526 BX_CPP_INLINE int extractFloatx80Sign(floatx80 a)
527 {
528 return a.exp>>15;
529 }
530
531 /*----------------------------------------------------------------------------
532 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an
533 | extended double-precision floating-point value, returning the result.
534 *----------------------------------------------------------------------------*/
535
packFloatx80(int zSign,Bit32s zExp,Bit64u zSig)536 BX_CPP_INLINE floatx80 packFloatx80(int zSign, Bit32s zExp, Bit64u zSig)
537 {
538 floatx80 z;
539 z.fraction = zSig;
540 z.exp = (zSign << 15) + zExp;
541 return z;
542 }
543
544 /*----------------------------------------------------------------------------
545 | Returns 1 if the extended double-precision floating-point value `a' is a
546 | NaN; otherwise returns 0.
547 *----------------------------------------------------------------------------*/
548
floatx80_is_nan(floatx80 a)549 BX_CPP_INLINE int floatx80_is_nan(floatx80 a)
550 {
551 return ((a.exp & 0x7FFF) == 0x7FFF) && (Bit64s) (a.fraction<<1);
552 }
553
554 /*----------------------------------------------------------------------------
555 | Returns 1 if the extended double-precision floating-point value `a' is a
556 | signaling NaN; otherwise returns 0.
557 *----------------------------------------------------------------------------*/
558
floatx80_is_signaling_nan(floatx80 a)559 BX_CPP_INLINE int floatx80_is_signaling_nan(floatx80 a)
560 {
561 Bit64u aLow = a.fraction & ~BX_CONST64(0x4000000000000000);
562 return ((a.exp & 0x7FFF) == 0x7FFF) &&
563 ((Bit64u) (aLow<<1)) && (a.fraction == aLow);
564 }
565
566 /*----------------------------------------------------------------------------
567 | Returns 1 if the extended double-precision floating-point value `a' is an
568 | unsupported; otherwise returns 0.
569 *----------------------------------------------------------------------------*/
570
floatx80_is_unsupported(floatx80 a)571 BX_CPP_INLINE int floatx80_is_unsupported(floatx80 a)
572 {
573 return ((a.exp & 0x7FFF) && !(a.fraction & BX_CONST64(0x8000000000000000)));
574 }
575
576 /*----------------------------------------------------------------------------
577 | Returns the result of converting the extended double-precision floating-
578 | point NaN `a' to the canonical NaN format. If `a' is a signaling NaN, the
579 | invalid exception is raised.
580 *----------------------------------------------------------------------------*/
581
floatx80ToCommonNaN(floatx80 a,float_status_t & status)582 BX_CPP_INLINE commonNaNT floatx80ToCommonNaN(floatx80 a, float_status_t &status)
583 {
584 commonNaNT z;
585 if (floatx80_is_signaling_nan(a)) float_raise(status, float_flag_invalid);
586 z.sign = a.exp >> 15;
587 z.lo = 0;
588 z.hi = a.fraction << 1;
589 return z;
590 }
591
592 /*----------------------------------------------------------------------------
593 | Returns the result of converting the canonical NaN `a' to the extended
594 | double-precision floating-point format.
595 *----------------------------------------------------------------------------*/
596
commonNaNToFloatx80(commonNaNT a)597 BX_CPP_INLINE floatx80 commonNaNToFloatx80(commonNaNT a)
598 {
599 floatx80 z;
600 z.fraction = BX_CONST64(0xC000000000000000) | (a.hi>>1);
601 z.exp = (((Bit16u) a.sign)<<15) | 0x7FFF;
602 return z;
603 }
604
605 /*----------------------------------------------------------------------------
606 | Takes two extended double-precision floating-point values `a' and `b', one
607 | of which is a NaN, and returns the appropriate NaN result. If either `a' or
608 | `b' is a signaling NaN, the invalid exception is raised.
609 *----------------------------------------------------------------------------*/
610
611 floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status_t &status);
612
613 /*----------------------------------------------------------------------------
614 | Takes extended double-precision floating-point NaN `a' and returns the
615 | appropriate NaN result. If `a' is a signaling NaN, the invalid exception
616 | is raised.
617 *----------------------------------------------------------------------------*/
618
propagateFloatx80NaN(floatx80 a,float_status_t & status)619 BX_CPP_INLINE floatx80 propagateFloatx80NaN(floatx80 a, float_status_t &status)
620 {
621 if (floatx80_is_signaling_nan(a))
622 float_raise(status, float_flag_invalid);
623
624 a.fraction |= BX_CONST64(0xC000000000000000);
625
626 return a;
627 }
628
629 /*----------------------------------------------------------------------------
630 | The pattern for a default generated extended double-precision NaN.
631 *----------------------------------------------------------------------------*/
632 extern const floatx80 floatx80_default_nan;
633
634 #endif /* FLOATX80 */
635
636 #ifdef FLOAT128
637
638 #include "softfloat-macros.h"
639
640 /*----------------------------------------------------------------------------
641 | The pattern for a default generated quadruple-precision NaN. The `high' and
642 | `low' values hold the most- and least-significant bits, respectively.
643 *----------------------------------------------------------------------------*/
644 #define float128_default_nan_hi BX_CONST64(0xFFFF800000000000)
645 #define float128_default_nan_lo BX_CONST64(0x0000000000000000)
646
647 #define float128_exp extractFloat128Exp
648
649 /*----------------------------------------------------------------------------
650 | Returns the least-significant 64 fraction bits of the quadruple-precision
651 | floating-point value `a'.
652 *----------------------------------------------------------------------------*/
653
extractFloat128Frac1(float128 a)654 BX_CPP_INLINE Bit64u extractFloat128Frac1(float128 a)
655 {
656 return a.lo;
657 }
658
659 /*----------------------------------------------------------------------------
660 | Returns the most-significant 48 fraction bits of the quadruple-precision
661 | floating-point value `a'.
662 *----------------------------------------------------------------------------*/
663
extractFloat128Frac0(float128 a)664 BX_CPP_INLINE Bit64u extractFloat128Frac0(float128 a)
665 {
666 return a.hi & BX_CONST64(0x0000FFFFFFFFFFFF);
667 }
668
669 /*----------------------------------------------------------------------------
670 | Returns the exponent bits of the quadruple-precision floating-point value
671 | `a'.
672 *----------------------------------------------------------------------------*/
673
extractFloat128Exp(float128 a)674 BX_CPP_INLINE Bit32s extractFloat128Exp(float128 a)
675 {
676 return ((Bit32s)(a.hi>>48)) & 0x7FFF;
677 }
678
679 /*----------------------------------------------------------------------------
680 | Returns the sign bit of the quadruple-precision floating-point value `a'.
681 *----------------------------------------------------------------------------*/
682
extractFloat128Sign(float128 a)683 BX_CPP_INLINE int extractFloat128Sign(float128 a)
684 {
685 return (int)(a.hi >> 63);
686 }
687
688 /*----------------------------------------------------------------------------
689 | Packs the sign `zSign', the exponent `zExp', and the significand formed
690 | by the concatenation of `zSig0' and `zSig1' into a quadruple-precision
691 | floating-point value, returning the result. After being shifted into the
692 | proper positions, the three fields `zSign', `zExp', and `zSig0' are simply
693 | added together to form the most significant 32 bits of the result. This
694 | means that any integer portion of `zSig0' will be added into the exponent.
695 | Since a properly normalized significand will have an integer portion equal
696 | to 1, the `zExp' input should be 1 less than the desired result exponent
697 | whenever `zSig0' and `zSig1' concatenated form a complete, normalized
698 | significand.
699 *----------------------------------------------------------------------------*/
700
packFloat128(int zSign,Bit32s zExp,Bit64u zSig0,Bit64u zSig1)701 BX_CPP_INLINE float128 packFloat128(int zSign, Bit32s zExp, Bit64u zSig0, Bit64u zSig1)
702 {
703 float128 z;
704 z.lo = zSig1;
705 z.hi = (((Bit64u) zSign)<<63) + (((Bit64u) zExp)<<48) + zSig0;
706 return z;
707 }
708
709 /*----------------------------------------------------------------------------
710 | Packs two 64-bit precision integers into into the quadruple-precision
711 | floating-point value, returning the result.
712 *----------------------------------------------------------------------------*/
713
packFloat128(Bit64u zHi,Bit64u zLo)714 BX_CPP_INLINE float128 packFloat128(Bit64u zHi, Bit64u zLo)
715 {
716 float128 z;
717 z.lo = zLo;
718 z.hi = zHi;
719 return z;
720 }
721
722 #ifdef _MSC_VER
723 #define PACK_FLOAT_128(hi,lo) { lo, hi }
724 #else
725 #define PACK_FLOAT_128(hi,lo) packFloat128(BX_CONST64(hi),BX_CONST64(lo))
726 #endif
727
728 /*----------------------------------------------------------------------------
729 | Returns 1 if the quadruple-precision floating-point value `a' is a NaN;
730 | otherwise returns 0.
731 *----------------------------------------------------------------------------*/
732
float128_is_nan(float128 a)733 BX_CPP_INLINE int float128_is_nan(float128 a)
734 {
735 return (BX_CONST64(0xFFFE000000000000) <= (Bit64u) (a.hi<<1))
736 && (a.lo || (a.hi & BX_CONST64(0x0000FFFFFFFFFFFF)));
737 }
738
739 /*----------------------------------------------------------------------------
740 | Returns 1 if the quadruple-precision floating-point value `a' is a
741 | signaling NaN; otherwise returns 0.
742 *----------------------------------------------------------------------------*/
743
float128_is_signaling_nan(float128 a)744 BX_CPP_INLINE int float128_is_signaling_nan(float128 a)
745 {
746 return (((a.hi>>47) & 0xFFFF) == 0xFFFE)
747 && (a.lo || (a.hi & BX_CONST64(0x00007FFFFFFFFFFF)));
748 }
749
750 /*----------------------------------------------------------------------------
751 | Returns the result of converting the quadruple-precision floating-point NaN
752 | `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
753 | exception is raised.
754 *----------------------------------------------------------------------------*/
755
float128ToCommonNaN(float128 a,float_status_t & status)756 BX_CPP_INLINE commonNaNT float128ToCommonNaN(float128 a, float_status_t &status)
757 {
758 commonNaNT z;
759 if (float128_is_signaling_nan(a)) float_raise(status, float_flag_invalid);
760 z.sign = (int)(a.hi>>63);
761 shortShift128Left(a.hi, a.lo, 16, &z.hi, &z.lo);
762 return z;
763 }
764
765 /*----------------------------------------------------------------------------
766 | Returns the result of converting the canonical NaN `a' to the quadruple-
767 | precision floating-point format.
768 *----------------------------------------------------------------------------*/
769
commonNaNToFloat128(commonNaNT a)770 BX_CPP_INLINE float128 commonNaNToFloat128(commonNaNT a)
771 {
772 float128 z;
773 shift128Right(a.hi, a.lo, 16, &z.hi, &z.lo);
774 z.hi |= (((Bit64u) a.sign)<<63) | BX_CONST64(0x7FFF800000000000);
775 return z;
776 }
777
778 /*----------------------------------------------------------------------------
779 | Takes two quadruple-precision floating-point values `a' and `b', one of
780 | which is a NaN, and returns the appropriate NaN result. If either `a' or
781 | `b' is a signaling NaN, the invalid exception is raised.
782 *----------------------------------------------------------------------------*/
783
784 float128 propagateFloat128NaN(float128 a, float128 b, float_status_t &status);
785
786 /*----------------------------------------------------------------------------
787 | The pattern for a default generated quadruple-precision NaN.
788 *----------------------------------------------------------------------------*/
789 extern const float128 float128_default_nan;
790
791 #endif /* FLOAT128 */
792
793 #endif
794