1 |
|
---|
2 | /*============================================================================
|
---|
3 |
|
---|
4 | This C source file is part of TestFloat, Release 3e, a package of programs for
|
---|
5 | testing the correctness of floating-point arithmetic complying with the IEEE
|
---|
6 | Standard for Floating-Point, by John R. Hauser.
|
---|
7 |
|
---|
8 | Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
|
---|
9 | University of California. All rights reserved.
|
---|
10 |
|
---|
11 | Redistribution and use in source and binary forms, with or without
|
---|
12 | modification, are permitted provided that the following conditions are met:
|
---|
13 |
|
---|
14 | 1. Redistributions of source code must retain the above copyright notice,
|
---|
15 | this list of conditions, and the following disclaimer.
|
---|
16 |
|
---|
17 | 2. Redistributions in binary form must reproduce the above copyright notice,
|
---|
18 | this list of conditions, and the following disclaimer in the documentation
|
---|
19 | and/or other materials provided with the distribution.
|
---|
20 |
|
---|
21 | 3. Neither the name of the University nor the names of its contributors may
|
---|
22 | be used to endorse or promote products derived from this software without
|
---|
23 | specific prior written permission.
|
---|
24 |
|
---|
25 | THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
|
---|
26 | EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
---|
27 | WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
|
---|
28 | DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
|
---|
29 | DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
---|
30 | (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
---|
31 | LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
|
---|
32 | ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
---|
33 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
---|
34 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
---|
35 |
|
---|
36 | =============================================================================*/
|
---|
37 |
|
---|
38 | #include <stdbool.h>
|
---|
39 | #include <stdint.h>
|
---|
40 | #include "platform.h"
|
---|
41 | #include "uint128.h"
|
---|
42 | #include "softfloat.h"
|
---|
43 | #include "slowfloat.h"
|
---|
44 |
|
---|
45 | uint_fast8_t slowfloat_roundingMode;
|
---|
46 | uint_fast8_t slowfloat_detectTininess;
|
---|
47 | uint_fast8_t slowfloat_exceptionFlags;
|
---|
48 | #ifdef EXTFLOAT80
|
---|
49 | uint_fast8_t slow_extF80_roundingPrecision;
|
---|
50 | #endif
|
---|
51 |
|
---|
52 | #ifdef FLOAT16
|
---|
53 | union ui16_f16 { uint16_t ui; float16_t f; };
|
---|
54 | #endif
|
---|
55 | union ui32_f32 { uint32_t ui; float32_t f; };
|
---|
56 | #ifdef FLOAT64
|
---|
57 | union ui64_f64 { uint64_t ui; float64_t f; };
|
---|
58 | #endif
|
---|
59 |
|
---|
60 | /*----------------------------------------------------------------------------
|
---|
61 | *----------------------------------------------------------------------------*/
|
---|
62 |
|
---|
63 | struct floatX {
|
---|
64 | bool isNaN;
|
---|
65 | bool isInf;
|
---|
66 | bool isZero;
|
---|
67 | bool sign;
|
---|
68 | int_fast32_t exp;
|
---|
69 | struct uint128 sig;
|
---|
70 | };
|
---|
71 |
|
---|
72 | static const struct floatX floatXNaN =
|
---|
73 | { true, false, false, false, 0, { 0, 0 } };
|
---|
74 | static const struct floatX floatXPositiveZero =
|
---|
75 | { false, false, true, false, 0, { 0, 0 } };
|
---|
76 | static const struct floatX floatXNegativeZero =
|
---|
77 | { false, false, true, true, 0, { 0, 0 } };
|
---|
78 |
|
---|
79 | static
|
---|
80 | void
|
---|
81 | roundFloatXTo11(
|
---|
82 | bool isTiny, struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
83 | {
|
---|
84 | uint_fast64_t roundBits, sigX64;
|
---|
85 |
|
---|
86 | sigX64 = xPtr->sig.v64;
|
---|
87 | roundBits = (sigX64 & UINT64_C( 0x1FFFFFFFFFFF )) | (xPtr->sig.v0 != 0);
|
---|
88 | if ( roundBits ) {
|
---|
89 | sigX64 &= UINT64_C( 0xFFFFE00000000000 );
|
---|
90 | if ( exact ) slowfloat_exceptionFlags |= softfloat_flag_inexact;
|
---|
91 | if ( isTiny ) slowfloat_exceptionFlags |= softfloat_flag_underflow;
|
---|
92 | switch ( roundingMode ) {
|
---|
93 | case softfloat_round_near_even:
|
---|
94 | if ( roundBits < UINT64_C( 0x100000000000 ) ) goto noIncrement;
|
---|
95 | if (
|
---|
96 | (roundBits == UINT64_C( 0x100000000000 ))
|
---|
97 | && !(sigX64 & UINT64_C( 0x200000000000 ))
|
---|
98 | ) {
|
---|
99 | goto noIncrement;
|
---|
100 | }
|
---|
101 | break;
|
---|
102 | case softfloat_round_minMag:
|
---|
103 | goto noIncrement;
|
---|
104 | case softfloat_round_min:
|
---|
105 | if ( !xPtr->sign ) goto noIncrement;
|
---|
106 | break;
|
---|
107 | case softfloat_round_max:
|
---|
108 | if ( xPtr->sign ) goto noIncrement;
|
---|
109 | break;
|
---|
110 | case softfloat_round_near_maxMag:
|
---|
111 | if ( roundBits < UINT64_C( 0x100000000000 ) ) goto noIncrement;
|
---|
112 | break;
|
---|
113 | #ifdef FLOAT_ROUND_ODD
|
---|
114 | case softfloat_round_odd:
|
---|
115 | sigX64 |= UINT64_C( 0x200000000000 );
|
---|
116 | goto noIncrement;
|
---|
117 | #endif
|
---|
118 | }
|
---|
119 | sigX64 += UINT64_C( 0x200000000000 );
|
---|
120 | if ( sigX64 == UINT64_C( 0x0100000000000000 ) ) {
|
---|
121 | ++xPtr->exp;
|
---|
122 | sigX64 = UINT64_C( 0x0080000000000000 );
|
---|
123 | }
|
---|
124 | noIncrement:
|
---|
125 | xPtr->sig.v64 = sigX64;
|
---|
126 | xPtr->sig.v0 = 0;
|
---|
127 | }
|
---|
128 |
|
---|
129 | }
|
---|
130 |
|
---|
131 | static
|
---|
132 | void
|
---|
133 | roundFloatXTo24(
|
---|
134 | bool isTiny, struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
135 | {
|
---|
136 | uint_fast64_t sigX64;
|
---|
137 | uint_fast32_t roundBits;
|
---|
138 |
|
---|
139 | sigX64 = xPtr->sig.v64;
|
---|
140 | roundBits = (uint32_t) sigX64 | (xPtr->sig.v0 != 0);
|
---|
141 | if ( roundBits ) {
|
---|
142 | sigX64 &= UINT64_C( 0xFFFFFFFF00000000 );
|
---|
143 | if ( exact ) slowfloat_exceptionFlags |= softfloat_flag_inexact;
|
---|
144 | if ( isTiny ) slowfloat_exceptionFlags |= softfloat_flag_underflow;
|
---|
145 | switch ( roundingMode ) {
|
---|
146 | case softfloat_round_near_even:
|
---|
147 | if ( roundBits < 0x80000000 ) goto noIncrement;
|
---|
148 | if (
|
---|
149 | (roundBits == 0x80000000)
|
---|
150 | && !(sigX64 & UINT64_C( 0x100000000 ))
|
---|
151 | ) {
|
---|
152 | goto noIncrement;
|
---|
153 | }
|
---|
154 | break;
|
---|
155 | case softfloat_round_minMag:
|
---|
156 | goto noIncrement;
|
---|
157 | case softfloat_round_min:
|
---|
158 | if ( !xPtr->sign ) goto noIncrement;
|
---|
159 | break;
|
---|
160 | case softfloat_round_max:
|
---|
161 | if ( xPtr->sign ) goto noIncrement;
|
---|
162 | break;
|
---|
163 | case softfloat_round_near_maxMag:
|
---|
164 | if ( roundBits < 0x80000000 ) goto noIncrement;
|
---|
165 | break;
|
---|
166 | #ifdef FLOAT_ROUND_ODD
|
---|
167 | case softfloat_round_odd:
|
---|
168 | sigX64 |= UINT64_C( 0x100000000 );
|
---|
169 | goto noIncrement;
|
---|
170 | #endif
|
---|
171 | }
|
---|
172 | sigX64 += UINT64_C( 0x100000000 );
|
---|
173 | if ( sigX64 == UINT64_C( 0x0100000000000000 ) ) {
|
---|
174 | ++xPtr->exp;
|
---|
175 | sigX64 = UINT64_C( 0x0080000000000000 );
|
---|
176 | }
|
---|
177 | noIncrement:
|
---|
178 | xPtr->sig.v64 = sigX64;
|
---|
179 | xPtr->sig.v0 = 0;
|
---|
180 | }
|
---|
181 |
|
---|
182 | }
|
---|
183 |
|
---|
184 | static
|
---|
185 | void
|
---|
186 | roundFloatXTo53(
|
---|
187 | bool isTiny, struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
188 | {
|
---|
189 | uint_fast64_t sigX64;
|
---|
190 | uint_fast8_t roundBits;
|
---|
191 |
|
---|
192 | sigX64 = xPtr->sig.v64;
|
---|
193 | roundBits = (sigX64 & 7) | (xPtr->sig.v0 != 0);
|
---|
194 | if ( roundBits ) {
|
---|
195 | sigX64 &= UINT64_C( 0xFFFFFFFFFFFFFFF8 );
|
---|
196 | if ( exact ) slowfloat_exceptionFlags |= softfloat_flag_inexact;
|
---|
197 | if ( isTiny ) slowfloat_exceptionFlags |= softfloat_flag_underflow;
|
---|
198 | switch ( roundingMode ) {
|
---|
199 | case softfloat_round_near_even:
|
---|
200 | if ( roundBits < 4 ) goto noIncrement;
|
---|
201 | if ( (roundBits == 4) && !(sigX64 & 8) ) goto noIncrement;
|
---|
202 | break;
|
---|
203 | case softfloat_round_minMag:
|
---|
204 | goto noIncrement;
|
---|
205 | case softfloat_round_min:
|
---|
206 | if ( !xPtr->sign ) goto noIncrement;
|
---|
207 | break;
|
---|
208 | case softfloat_round_max:
|
---|
209 | if ( xPtr->sign ) goto noIncrement;
|
---|
210 | break;
|
---|
211 | case softfloat_round_near_maxMag:
|
---|
212 | if ( roundBits < 4 ) goto noIncrement;
|
---|
213 | break;
|
---|
214 | #ifdef FLOAT_ROUND_ODD
|
---|
215 | case softfloat_round_odd:
|
---|
216 | sigX64 |= 8;
|
---|
217 | goto noIncrement;
|
---|
218 | #endif
|
---|
219 | }
|
---|
220 | sigX64 += 8;
|
---|
221 | if ( sigX64 == UINT64_C( 0x0100000000000000 ) ) {
|
---|
222 | ++xPtr->exp;
|
---|
223 | sigX64 = UINT64_C( 0x0080000000000000 );
|
---|
224 | }
|
---|
225 | noIncrement:
|
---|
226 | xPtr->sig.v64 = sigX64;
|
---|
227 | xPtr->sig.v0 = 0;
|
---|
228 | }
|
---|
229 |
|
---|
230 | }
|
---|
231 |
|
---|
232 | static
|
---|
233 | void
|
---|
234 | roundFloatXTo64(
|
---|
235 | bool isTiny, struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
236 | {
|
---|
237 | uint_fast64_t sigX0, roundBits, sigX64;
|
---|
238 |
|
---|
239 | sigX0 = xPtr->sig.v0;
|
---|
240 | roundBits = sigX0 & UINT64_C( 0x00FFFFFFFFFFFFFF );
|
---|
241 | if ( roundBits ) {
|
---|
242 | sigX0 &= UINT64_C( 0xFF00000000000000 );
|
---|
243 | if ( exact ) slowfloat_exceptionFlags |= softfloat_flag_inexact;
|
---|
244 | if ( isTiny ) slowfloat_exceptionFlags |= softfloat_flag_underflow;
|
---|
245 | switch ( roundingMode ) {
|
---|
246 | case softfloat_round_near_even:
|
---|
247 | if ( roundBits < UINT64_C( 0x0080000000000000 ) ) goto noIncrement;
|
---|
248 | if (
|
---|
249 | (roundBits == UINT64_C( 0x0080000000000000 ))
|
---|
250 | && !(sigX0 & UINT64_C( 0x0100000000000000 ))
|
---|
251 | ) {
|
---|
252 | goto noIncrement;
|
---|
253 | }
|
---|
254 | break;
|
---|
255 | case softfloat_round_minMag:
|
---|
256 | goto noIncrement;
|
---|
257 | case softfloat_round_min:
|
---|
258 | if ( !xPtr->sign ) goto noIncrement;
|
---|
259 | break;
|
---|
260 | case softfloat_round_max:
|
---|
261 | if ( xPtr->sign ) goto noIncrement;
|
---|
262 | break;
|
---|
263 | case softfloat_round_near_maxMag:
|
---|
264 | if ( roundBits < UINT64_C( 0x0080000000000000 ) ) goto noIncrement;
|
---|
265 | break;
|
---|
266 | #ifdef FLOAT_ROUND_ODD
|
---|
267 | case softfloat_round_odd:
|
---|
268 | sigX0 |= UINT64_C( 0x100000000000000 );
|
---|
269 | goto noIncrement;
|
---|
270 | #endif
|
---|
271 | }
|
---|
272 | sigX0 += UINT64_C( 0x100000000000000 );
|
---|
273 | sigX64 = xPtr->sig.v64 + !sigX0;
|
---|
274 | if ( sigX64 == UINT64_C( 0x0100000000000000 ) ) {
|
---|
275 | ++xPtr->exp;
|
---|
276 | sigX64 = UINT64_C( 0x0080000000000000 );
|
---|
277 | }
|
---|
278 | xPtr->sig.v64 = sigX64;
|
---|
279 | noIncrement:
|
---|
280 | xPtr->sig.v0 = sigX0;
|
---|
281 | }
|
---|
282 |
|
---|
283 | }
|
---|
284 |
|
---|
285 | static
|
---|
286 | void
|
---|
287 | roundFloatXTo113(
|
---|
288 | bool isTiny, struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
289 | {
|
---|
290 | uint_fast64_t sigX0;
|
---|
291 | uint_fast8_t roundBits;
|
---|
292 | uint_fast64_t sigX64;
|
---|
293 |
|
---|
294 | sigX0 = xPtr->sig.v0;
|
---|
295 | roundBits = sigX0 & 0x7F;
|
---|
296 | if ( roundBits ) {
|
---|
297 | sigX0 &= UINT64_C( 0xFFFFFFFFFFFFFF80 );
|
---|
298 | if ( exact ) slowfloat_exceptionFlags |= softfloat_flag_inexact;
|
---|
299 | if ( isTiny ) slowfloat_exceptionFlags |= softfloat_flag_underflow;
|
---|
300 | switch ( roundingMode ) {
|
---|
301 | case softfloat_round_near_even:
|
---|
302 | if ( roundBits < 0x40 ) goto noIncrement;
|
---|
303 | if ( (roundBits == 0x40) && !(sigX0 & 0x80) ) goto noIncrement;
|
---|
304 | break;
|
---|
305 | case softfloat_round_minMag:
|
---|
306 | goto noIncrement;
|
---|
307 | case softfloat_round_min:
|
---|
308 | if ( !xPtr->sign ) goto noIncrement;
|
---|
309 | break;
|
---|
310 | case softfloat_round_max:
|
---|
311 | if ( xPtr->sign ) goto noIncrement;
|
---|
312 | break;
|
---|
313 | case softfloat_round_near_maxMag:
|
---|
314 | if ( roundBits < 0x40 ) goto noIncrement;
|
---|
315 | break;
|
---|
316 | #ifdef FLOAT_ROUND_ODD
|
---|
317 | case softfloat_round_odd:
|
---|
318 | sigX0 |= 0x80;
|
---|
319 | goto noIncrement;
|
---|
320 | #endif
|
---|
321 | }
|
---|
322 | sigX0 += 0x80;
|
---|
323 | sigX64 = xPtr->sig.v64 + !sigX0;
|
---|
324 | if ( sigX64 == UINT64_C( 0x0100000000000000 ) ) {
|
---|
325 | ++xPtr->exp;
|
---|
326 | sigX64 = UINT64_C( 0x0080000000000000 );
|
---|
327 | }
|
---|
328 | xPtr->sig.v64 = sigX64;
|
---|
329 | noIncrement:
|
---|
330 | xPtr->sig.v0 = sigX0;
|
---|
331 | }
|
---|
332 |
|
---|
333 | }
|
---|
334 |
|
---|
335 | static void ui32ToFloatX( uint_fast32_t a, struct floatX *xPtr )
|
---|
336 | {
|
---|
337 | uint_fast64_t sig64;
|
---|
338 | int_fast32_t exp;
|
---|
339 |
|
---|
340 | xPtr->isNaN = false;
|
---|
341 | xPtr->isInf = false;
|
---|
342 | xPtr->sign = false;
|
---|
343 | sig64 = a;
|
---|
344 | if ( a ) {
|
---|
345 | xPtr->isZero = false;
|
---|
346 | exp = 31;
|
---|
347 | sig64 <<= 24;
|
---|
348 | while ( sig64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
349 | --exp;
|
---|
350 | sig64 <<= 1;
|
---|
351 | }
|
---|
352 | xPtr->exp = exp;
|
---|
353 | } else {
|
---|
354 | xPtr->isZero = true;
|
---|
355 | }
|
---|
356 | xPtr->sig.v64 = sig64;
|
---|
357 | xPtr->sig.v0 = 0;
|
---|
358 |
|
---|
359 | }
|
---|
360 |
|
---|
361 | static
|
---|
362 | uint_fast32_t
|
---|
363 | floatXToUI32(
|
---|
364 | const struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
365 | {
|
---|
366 | uint_fast8_t savedExceptionFlags;
|
---|
367 | struct floatX x;
|
---|
368 | int_fast32_t shiftDist;
|
---|
369 | uint_fast32_t z;
|
---|
370 |
|
---|
371 | if ( xPtr->isInf || xPtr->isNaN ) {
|
---|
372 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
373 | return (xPtr->isInf && xPtr->sign) ? 0 : 0xFFFFFFFF;
|
---|
374 | }
|
---|
375 | if ( xPtr->isZero ) return 0;
|
---|
376 | savedExceptionFlags = slowfloat_exceptionFlags;
|
---|
377 | x = *xPtr;
|
---|
378 | shiftDist = 52 - x.exp;
|
---|
379 | if ( 56 < shiftDist ) {
|
---|
380 | x.sig.v64 = 0;
|
---|
381 | x.sig.v0 = 1;
|
---|
382 | } else {
|
---|
383 | while ( 0 < shiftDist ) {
|
---|
384 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
385 | --shiftDist;
|
---|
386 | }
|
---|
387 | }
|
---|
388 | roundFloatXTo53( false, &x, roundingMode, exact );
|
---|
389 | x.sig = shortShiftRightJam128( x.sig, 3 );
|
---|
390 | z = x.sig.v64;
|
---|
391 | if ( (shiftDist < 0) || x.sig.v64>>32 || (x.sign && z) ) {
|
---|
392 | slowfloat_exceptionFlags =
|
---|
393 | savedExceptionFlags | softfloat_flag_invalid;
|
---|
394 | return x.sign ? 0 : 0xFFFFFFFF;
|
---|
395 | }
|
---|
396 | return z;
|
---|
397 |
|
---|
398 | }
|
---|
399 |
|
---|
400 | static void ui64ToFloatX( uint_fast64_t a, struct floatX *xPtr )
|
---|
401 | {
|
---|
402 | struct uint128 sig;
|
---|
403 | int_fast32_t exp;
|
---|
404 |
|
---|
405 | xPtr->isNaN = false;
|
---|
406 | xPtr->isInf = false;
|
---|
407 | xPtr->sign = false;
|
---|
408 | sig.v64 = 0;
|
---|
409 | sig.v0 = a;
|
---|
410 | if ( a ) {
|
---|
411 | xPtr->isZero = false;
|
---|
412 | exp = 63;
|
---|
413 | sig = shortShiftLeft128( sig, 56 );
|
---|
414 | while ( sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
415 | --exp;
|
---|
416 | sig = shortShiftLeft128( sig, 1 );
|
---|
417 | }
|
---|
418 | xPtr->exp = exp;
|
---|
419 | } else {
|
---|
420 | xPtr->isZero = true;
|
---|
421 | }
|
---|
422 | xPtr->sig = sig;
|
---|
423 |
|
---|
424 | }
|
---|
425 |
|
---|
426 | static
|
---|
427 | uint_fast64_t
|
---|
428 | floatXToUI64(
|
---|
429 | const struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
430 | {
|
---|
431 | uint_fast8_t savedExceptionFlags;
|
---|
432 | struct floatX x;
|
---|
433 | int_fast32_t shiftDist;
|
---|
434 | uint_fast64_t z;
|
---|
435 |
|
---|
436 | if ( xPtr->isInf || xPtr->isNaN ) {
|
---|
437 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
438 | return
|
---|
439 | (xPtr->isInf && xPtr->sign) ? 0 : UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
440 | }
|
---|
441 | if ( xPtr->isZero ) return 0;
|
---|
442 | savedExceptionFlags = slowfloat_exceptionFlags;
|
---|
443 | x = *xPtr;
|
---|
444 | shiftDist = 112 - x.exp;
|
---|
445 | if ( 116 < shiftDist ) {
|
---|
446 | x.sig.v64 = 0;
|
---|
447 | x.sig.v0 = 1;
|
---|
448 | } else {
|
---|
449 | while ( 0 < shiftDist ) {
|
---|
450 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
451 | --shiftDist;
|
---|
452 | }
|
---|
453 | }
|
---|
454 | roundFloatXTo113( false, &x, roundingMode, exact );
|
---|
455 | x.sig = shortShiftRightJam128( x.sig, 7 );
|
---|
456 | z = x.sig.v0;
|
---|
457 | if ( (shiftDist < 0) || x.sig.v64 || (x.sign && z) ) {
|
---|
458 | slowfloat_exceptionFlags =
|
---|
459 | savedExceptionFlags | softfloat_flag_invalid;
|
---|
460 | return x.sign ? 0 : UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
461 | }
|
---|
462 | return z;
|
---|
463 |
|
---|
464 | }
|
---|
465 |
|
---|
466 | static void i32ToFloatX( int_fast32_t a, struct floatX *xPtr )
|
---|
467 | {
|
---|
468 | bool sign;
|
---|
469 | uint_fast64_t sig64;
|
---|
470 | int_fast32_t exp;
|
---|
471 |
|
---|
472 | xPtr->isNaN = false;
|
---|
473 | xPtr->isInf = false;
|
---|
474 | sign = (a < 0);
|
---|
475 | xPtr->sign = sign;
|
---|
476 | sig64 = sign ? -(uint64_t) a : a;
|
---|
477 | if ( a ) {
|
---|
478 | xPtr->isZero = false;
|
---|
479 | exp = 31;
|
---|
480 | sig64 <<= 24;
|
---|
481 | while ( sig64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
482 | --exp;
|
---|
483 | sig64 <<= 1;
|
---|
484 | }
|
---|
485 | xPtr->exp = exp;
|
---|
486 | } else {
|
---|
487 | xPtr->isZero = true;
|
---|
488 | }
|
---|
489 | xPtr->sig.v64 = sig64;
|
---|
490 | xPtr->sig.v0 = 0;
|
---|
491 |
|
---|
492 | }
|
---|
493 |
|
---|
494 | static
|
---|
495 | int_fast32_t
|
---|
496 | floatXToI32(
|
---|
497 | const struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
498 | {
|
---|
499 | uint_fast8_t savedExceptionFlags;
|
---|
500 | struct floatX x;
|
---|
501 | int_fast32_t shiftDist;
|
---|
502 | union { uint32_t ui; int32_t i; } uZ;
|
---|
503 |
|
---|
504 | if ( xPtr->isInf || xPtr->isNaN ) {
|
---|
505 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
506 | return (xPtr->isInf && xPtr->sign) ? -0x7FFFFFFF - 1 : 0x7FFFFFFF;
|
---|
507 | }
|
---|
508 | if ( xPtr->isZero ) return 0;
|
---|
509 | savedExceptionFlags = slowfloat_exceptionFlags;
|
---|
510 | x = *xPtr;
|
---|
511 | shiftDist = 52 - x.exp;
|
---|
512 | if ( 56 < shiftDist ) {
|
---|
513 | x.sig.v64 = 0;
|
---|
514 | x.sig.v0 = 1;
|
---|
515 | } else {
|
---|
516 | while ( 0 < shiftDist ) {
|
---|
517 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
518 | --shiftDist;
|
---|
519 | }
|
---|
520 | }
|
---|
521 | roundFloatXTo53( false, &x, roundingMode, exact );
|
---|
522 | x.sig = shortShiftRightJam128( x.sig, 3 );
|
---|
523 | uZ.ui = x.sig.v64;
|
---|
524 | if ( x.sign ) uZ.ui = -uZ.ui;
|
---|
525 | if (
|
---|
526 | (shiftDist < 0) || x.sig.v64>>32
|
---|
527 | || ((uZ.i != 0) && (x.sign != (uZ.i < 0)))
|
---|
528 | ) {
|
---|
529 | slowfloat_exceptionFlags =
|
---|
530 | savedExceptionFlags | softfloat_flag_invalid;
|
---|
531 | return x.sign ? -0x7FFFFFFF - 1 : 0x7FFFFFFF;
|
---|
532 | }
|
---|
533 | return uZ.i;
|
---|
534 |
|
---|
535 | }
|
---|
536 |
|
---|
537 | static void i64ToFloatX( int_fast64_t a, struct floatX *xPtr )
|
---|
538 | {
|
---|
539 | bool sign;
|
---|
540 | struct uint128 sig;
|
---|
541 | int_fast32_t exp;
|
---|
542 |
|
---|
543 | xPtr->isNaN = false;
|
---|
544 | xPtr->isInf = false;
|
---|
545 | sign = (a < 0);
|
---|
546 | xPtr->sign = sign;
|
---|
547 | sig.v64 = 0;
|
---|
548 | sig.v0 = sign ? -(uint_fast64_t) a : a;
|
---|
549 | if ( a ) {
|
---|
550 | xPtr->isZero = false;
|
---|
551 | exp = 63;
|
---|
552 | sig = shortShiftLeft128( sig, 56 );
|
---|
553 | while ( sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
554 | --exp;
|
---|
555 | sig = shortShiftLeft128( sig, 1 );
|
---|
556 | }
|
---|
557 | xPtr->exp = exp;
|
---|
558 | } else {
|
---|
559 | xPtr->isZero = true;
|
---|
560 | }
|
---|
561 | xPtr->sig = sig;
|
---|
562 |
|
---|
563 | }
|
---|
564 |
|
---|
565 | static
|
---|
566 | int_fast64_t
|
---|
567 | floatXToI64(
|
---|
568 | const struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
569 | {
|
---|
570 | uint_fast8_t savedExceptionFlags;
|
---|
571 | struct floatX x;
|
---|
572 | int_fast32_t shiftDist;
|
---|
573 | union { uint64_t ui; int64_t i; } uZ;
|
---|
574 |
|
---|
575 | if ( xPtr->isInf || xPtr->isNaN ) {
|
---|
576 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
577 | return
|
---|
578 | (xPtr->isInf && xPtr->sign) ? -INT64_C( 0x7FFFFFFFFFFFFFFF ) - 1
|
---|
579 | : INT64_C( 0x7FFFFFFFFFFFFFFF );
|
---|
580 | }
|
---|
581 | if ( xPtr->isZero ) return 0;
|
---|
582 | savedExceptionFlags = slowfloat_exceptionFlags;
|
---|
583 | x = *xPtr;
|
---|
584 | shiftDist = 112 - x.exp;
|
---|
585 | if ( 116 < shiftDist ) {
|
---|
586 | x.sig.v64 = 0;
|
---|
587 | x.sig.v0 = 1;
|
---|
588 | } else {
|
---|
589 | while ( 0 < shiftDist ) {
|
---|
590 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
591 | --shiftDist;
|
---|
592 | }
|
---|
593 | }
|
---|
594 | roundFloatXTo113( false, &x, roundingMode, exact );
|
---|
595 | x.sig = shortShiftRightJam128( x.sig, 7 );
|
---|
596 | uZ.ui = x.sig.v0;
|
---|
597 | if ( x.sign ) uZ.ui = -uZ.ui;
|
---|
598 | if (
|
---|
599 | (shiftDist < 0) || x.sig.v64 || ((uZ.i != 0) && (x.sign != (uZ.i < 0)))
|
---|
600 | ) {
|
---|
601 | slowfloat_exceptionFlags =
|
---|
602 | savedExceptionFlags | softfloat_flag_invalid;
|
---|
603 | return
|
---|
604 | x.sign ? -INT64_C( 0x7FFFFFFFFFFFFFFF ) - 1
|
---|
605 | : INT64_C( 0x7FFFFFFFFFFFFFFF );
|
---|
606 | }
|
---|
607 | return uZ.i;
|
---|
608 |
|
---|
609 | }
|
---|
610 |
|
---|
611 | #ifdef FLOAT16
|
---|
612 |
|
---|
613 | static void f16ToFloatX( float16_t a, struct floatX *xPtr )
|
---|
614 | {
|
---|
615 | union ui16_f16 uA;
|
---|
616 | uint_fast16_t uiA;
|
---|
617 | int_fast8_t exp;
|
---|
618 | uint_fast64_t sig64;
|
---|
619 |
|
---|
620 | uA.f = a;
|
---|
621 | uiA = uA.ui;
|
---|
622 | xPtr->isNaN = false;
|
---|
623 | xPtr->isInf = false;
|
---|
624 | xPtr->isZero = false;
|
---|
625 | xPtr->sign = ((uiA & 0x8000) != 0);
|
---|
626 | exp = uiA>>10 & 0x1F;
|
---|
627 | sig64 = uiA & 0x03FF;
|
---|
628 | sig64 <<= 45;
|
---|
629 | if ( exp == 0x1F ) {
|
---|
630 | if ( sig64 ) {
|
---|
631 | xPtr->isNaN = true;
|
---|
632 | } else {
|
---|
633 | xPtr->isInf = true;
|
---|
634 | }
|
---|
635 | } else if ( !exp ) {
|
---|
636 | if ( !sig64 ) {
|
---|
637 | xPtr->isZero = true;
|
---|
638 | } else {
|
---|
639 | exp = 1 - 0xF;
|
---|
640 | do {
|
---|
641 | --exp;
|
---|
642 | sig64 <<= 1;
|
---|
643 | } while ( sig64 < UINT64_C( 0x0080000000000000 ) );
|
---|
644 | xPtr->exp = exp;
|
---|
645 | }
|
---|
646 | } else {
|
---|
647 | xPtr->exp = exp - 0xF;
|
---|
648 | sig64 |= UINT64_C( 0x0080000000000000 );
|
---|
649 | }
|
---|
650 | xPtr->sig.v64 = sig64;
|
---|
651 | xPtr->sig.v0 = 0;
|
---|
652 |
|
---|
653 | }
|
---|
654 |
|
---|
655 | static float16_t floatXToF16( const struct floatX *xPtr )
|
---|
656 | {
|
---|
657 | uint_fast16_t uiZ;
|
---|
658 | struct floatX x, savedX;
|
---|
659 | bool isTiny;
|
---|
660 | int_fast32_t exp;
|
---|
661 | union ui16_f16 uZ;
|
---|
662 |
|
---|
663 | if ( xPtr->isNaN ) {
|
---|
664 | uiZ = 0xFFFF;
|
---|
665 | goto uiZ;
|
---|
666 | }
|
---|
667 | if ( xPtr->isInf ) {
|
---|
668 | uiZ = xPtr->sign ? 0xFC00 : 0x7C00;
|
---|
669 | goto uiZ;
|
---|
670 | }
|
---|
671 | if ( xPtr->isZero ) {
|
---|
672 | uiZ = xPtr->sign ? 0x8000 : 0;
|
---|
673 | goto uiZ;
|
---|
674 | }
|
---|
675 | x = *xPtr;
|
---|
676 | while ( UINT64_C( 0x0100000000000000 ) <= x.sig.v64 ) {
|
---|
677 | ++x.exp;
|
---|
678 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
679 | }
|
---|
680 | while ( x.sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
681 | --x.exp;
|
---|
682 | x.sig = shortShiftLeft128( x.sig, 1 );
|
---|
683 | }
|
---|
684 | savedX = x;
|
---|
685 | isTiny =
|
---|
686 | (slowfloat_detectTininess == softfloat_tininess_beforeRounding)
|
---|
687 | && (x.exp + 0xF <= 0);
|
---|
688 | roundFloatXTo11( isTiny, &x, slowfloat_roundingMode, true );
|
---|
689 | exp = x.exp + 0xF;
|
---|
690 | if ( 0x1F <= exp ) {
|
---|
691 | slowfloat_exceptionFlags |=
|
---|
692 | softfloat_flag_overflow | softfloat_flag_inexact;
|
---|
693 | if ( x.sign ) {
|
---|
694 | switch ( slowfloat_roundingMode ) {
|
---|
695 | case softfloat_round_near_even:
|
---|
696 | case softfloat_round_min:
|
---|
697 | case softfloat_round_near_maxMag:
|
---|
698 | uiZ = 0xFC00;
|
---|
699 | break;
|
---|
700 | case softfloat_round_minMag:
|
---|
701 | case softfloat_round_max:
|
---|
702 | case softfloat_round_odd:
|
---|
703 | uiZ = 0xFBFF;
|
---|
704 | break;
|
---|
705 | }
|
---|
706 | } else {
|
---|
707 | switch ( slowfloat_roundingMode ) {
|
---|
708 | case softfloat_round_near_even:
|
---|
709 | case softfloat_round_max:
|
---|
710 | case softfloat_round_near_maxMag:
|
---|
711 | uiZ = 0x7C00;
|
---|
712 | break;
|
---|
713 | case softfloat_round_minMag:
|
---|
714 | case softfloat_round_min:
|
---|
715 | case softfloat_round_odd:
|
---|
716 | uiZ = 0x7BFF;
|
---|
717 | break;
|
---|
718 | }
|
---|
719 | }
|
---|
720 | goto uiZ;
|
---|
721 | }
|
---|
722 | if ( exp <= 0 ) {
|
---|
723 | isTiny = true;
|
---|
724 | x = savedX;
|
---|
725 | exp = x.exp + 0xF;
|
---|
726 | if ( exp < -14 ) {
|
---|
727 | x.sig.v0 = (x.sig.v64 != 0) || (x.sig.v0 != 0);
|
---|
728 | x.sig.v64 = 0;
|
---|
729 | } else {
|
---|
730 | while ( exp <= 0 ) {
|
---|
731 | ++exp;
|
---|
732 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
733 | }
|
---|
734 | }
|
---|
735 | roundFloatXTo11( isTiny, &x, slowfloat_roundingMode, true );
|
---|
736 | exp = (UINT64_C( 0x0080000000000000 ) <= x.sig.v64) ? 1 : 0;
|
---|
737 | }
|
---|
738 | uiZ = (uint_fast16_t) exp<<10;
|
---|
739 | if ( x.sign ) uiZ |= 0x8000;
|
---|
740 | uiZ |= x.sig.v64>>45 & 0x03FF;
|
---|
741 | uiZ:
|
---|
742 | uZ.ui = uiZ;
|
---|
743 | return uZ.f;
|
---|
744 |
|
---|
745 | }
|
---|
746 |
|
---|
747 | #endif
|
---|
748 |
|
---|
749 | static void f32ToFloatX( float32_t a, struct floatX *xPtr )
|
---|
750 | {
|
---|
751 | union ui32_f32 uA;
|
---|
752 | uint_fast32_t uiA;
|
---|
753 | int_fast16_t exp;
|
---|
754 | uint_fast64_t sig64;
|
---|
755 |
|
---|
756 | uA.f = a;
|
---|
757 | uiA = uA.ui;
|
---|
758 | xPtr->isNaN = false;
|
---|
759 | xPtr->isInf = false;
|
---|
760 | xPtr->isZero = false;
|
---|
761 | xPtr->sign = ((uiA & 0x80000000) != 0);
|
---|
762 | exp = uiA>>23 & 0xFF;
|
---|
763 | sig64 = uiA & 0x007FFFFF;
|
---|
764 | sig64 <<= 32;
|
---|
765 | if ( exp == 0xFF ) {
|
---|
766 | if ( sig64 ) {
|
---|
767 | xPtr->isNaN = true;
|
---|
768 | } else {
|
---|
769 | xPtr->isInf = true;
|
---|
770 | }
|
---|
771 | } else if ( !exp ) {
|
---|
772 | if ( !sig64 ) {
|
---|
773 | xPtr->isZero = true;
|
---|
774 | } else {
|
---|
775 | exp = 1 - 0x7F;
|
---|
776 | do {
|
---|
777 | --exp;
|
---|
778 | sig64 <<= 1;
|
---|
779 | } while ( sig64 < UINT64_C( 0x0080000000000000 ) );
|
---|
780 | xPtr->exp = exp;
|
---|
781 | }
|
---|
782 | } else {
|
---|
783 | xPtr->exp = exp - 0x7F;
|
---|
784 | sig64 |= UINT64_C( 0x0080000000000000 );
|
---|
785 | }
|
---|
786 | xPtr->sig.v64 = sig64;
|
---|
787 | xPtr->sig.v0 = 0;
|
---|
788 |
|
---|
789 | }
|
---|
790 |
|
---|
791 | static float32_t floatXToF32( const struct floatX *xPtr )
|
---|
792 | {
|
---|
793 | uint_fast32_t uiZ;
|
---|
794 | struct floatX x, savedX;
|
---|
795 | bool isTiny;
|
---|
796 | int_fast32_t exp;
|
---|
797 | union ui32_f32 uZ;
|
---|
798 |
|
---|
799 | if ( xPtr->isNaN ) {
|
---|
800 | uiZ = 0xFFFFFFFF;
|
---|
801 | goto uiZ;
|
---|
802 | }
|
---|
803 | if ( xPtr->isInf ) {
|
---|
804 | uiZ = xPtr->sign ? 0xFF800000 : 0x7F800000;
|
---|
805 | goto uiZ;
|
---|
806 | }
|
---|
807 | if ( xPtr->isZero ) {
|
---|
808 | uiZ = xPtr->sign ? 0x80000000 : 0;
|
---|
809 | goto uiZ;
|
---|
810 | }
|
---|
811 | x = *xPtr;
|
---|
812 | while ( UINT64_C( 0x0100000000000000 ) <= x.sig.v64 ) {
|
---|
813 | ++x.exp;
|
---|
814 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
815 | }
|
---|
816 | while ( x.sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
817 | --x.exp;
|
---|
818 | x.sig = shortShiftLeft128( x.sig, 1 );
|
---|
819 | }
|
---|
820 | savedX = x;
|
---|
821 | isTiny =
|
---|
822 | (slowfloat_detectTininess == softfloat_tininess_beforeRounding)
|
---|
823 | && (x.exp + 0x7F <= 0);
|
---|
824 | roundFloatXTo24( isTiny, &x, slowfloat_roundingMode, true );
|
---|
825 | exp = x.exp + 0x7F;
|
---|
826 | if ( 0xFF <= exp ) {
|
---|
827 | slowfloat_exceptionFlags |=
|
---|
828 | softfloat_flag_overflow | softfloat_flag_inexact;
|
---|
829 | if ( x.sign ) {
|
---|
830 | switch ( slowfloat_roundingMode ) {
|
---|
831 | case softfloat_round_near_even:
|
---|
832 | case softfloat_round_min:
|
---|
833 | case softfloat_round_near_maxMag:
|
---|
834 | uiZ = 0xFF800000;
|
---|
835 | break;
|
---|
836 | case softfloat_round_minMag:
|
---|
837 | case softfloat_round_max:
|
---|
838 | case softfloat_round_odd:
|
---|
839 | uiZ = 0xFF7FFFFF;
|
---|
840 | break;
|
---|
841 | }
|
---|
842 | } else {
|
---|
843 | switch ( slowfloat_roundingMode ) {
|
---|
844 | case softfloat_round_near_even:
|
---|
845 | case softfloat_round_max:
|
---|
846 | case softfloat_round_near_maxMag:
|
---|
847 | uiZ = 0x7F800000;
|
---|
848 | break;
|
---|
849 | case softfloat_round_minMag:
|
---|
850 | case softfloat_round_min:
|
---|
851 | case softfloat_round_odd:
|
---|
852 | uiZ = 0x7F7FFFFF;
|
---|
853 | break;
|
---|
854 | }
|
---|
855 | }
|
---|
856 | goto uiZ;
|
---|
857 | }
|
---|
858 | if ( exp <= 0 ) {
|
---|
859 | isTiny = true;
|
---|
860 | x = savedX;
|
---|
861 | exp = x.exp + 0x7F;
|
---|
862 | if ( exp < -27 ) {
|
---|
863 | x.sig.v0 = (x.sig.v64 != 0) || (x.sig.v0 != 0);
|
---|
864 | x.sig.v64 = 0;
|
---|
865 | } else {
|
---|
866 | while ( exp <= 0 ) {
|
---|
867 | ++exp;
|
---|
868 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
869 | }
|
---|
870 | }
|
---|
871 | roundFloatXTo24( isTiny, &x, slowfloat_roundingMode, true );
|
---|
872 | exp = (UINT64_C( 0x0080000000000000 ) <= x.sig.v64) ? 1 : 0;
|
---|
873 | }
|
---|
874 | uiZ = (uint_fast32_t) exp<<23;
|
---|
875 | if ( x.sign ) uiZ |= 0x80000000;
|
---|
876 | uiZ |= x.sig.v64>>32 & 0x007FFFFF;
|
---|
877 | uiZ:
|
---|
878 | uZ.ui = uiZ;
|
---|
879 | return uZ.f;
|
---|
880 |
|
---|
881 | }
|
---|
882 |
|
---|
883 | #ifdef FLOAT64
|
---|
884 |
|
---|
885 | static void f64ToFloatX( float64_t a, struct floatX *xPtr )
|
---|
886 | {
|
---|
887 | union ui64_f64 uA;
|
---|
888 | uint_fast64_t uiA;
|
---|
889 | int_fast16_t exp;
|
---|
890 | uint_fast64_t sig64;
|
---|
891 |
|
---|
892 | uA.f = a;
|
---|
893 | uiA = uA.ui;
|
---|
894 | xPtr->isNaN = false;
|
---|
895 | xPtr->isInf = false;
|
---|
896 | xPtr->isZero = false;
|
---|
897 | xPtr->sign = ((uiA & UINT64_C( 0x8000000000000000 )) != 0);
|
---|
898 | exp = uiA>>52 & 0x7FF;
|
---|
899 | sig64 = uiA & UINT64_C( 0x000FFFFFFFFFFFFF );
|
---|
900 | if ( exp == 0x7FF ) {
|
---|
901 | if ( sig64 ) {
|
---|
902 | xPtr->isNaN = true;
|
---|
903 | } else {
|
---|
904 | xPtr->isInf = true;
|
---|
905 | }
|
---|
906 | } else if ( !exp ) {
|
---|
907 | if ( !sig64 ) {
|
---|
908 | xPtr->isZero = true;
|
---|
909 | } else {
|
---|
910 | exp = 1 - 0x3FF;
|
---|
911 | do {
|
---|
912 | --exp;
|
---|
913 | sig64 <<= 1;
|
---|
914 | } while ( sig64 < UINT64_C( 0x0010000000000000 ) );
|
---|
915 | xPtr->exp = exp;
|
---|
916 | }
|
---|
917 | } else {
|
---|
918 | xPtr->exp = exp - 0x3FF;
|
---|
919 | sig64 |= UINT64_C( 0x0010000000000000 );
|
---|
920 | }
|
---|
921 | xPtr->sig.v64 = sig64<<3;
|
---|
922 | xPtr->sig.v0 = 0;
|
---|
923 |
|
---|
924 | }
|
---|
925 |
|
---|
926 | static float64_t floatXToF64( const struct floatX *xPtr )
|
---|
927 | {
|
---|
928 | uint_fast64_t uiZ;
|
---|
929 | struct floatX x, savedX;
|
---|
930 | bool isTiny;
|
---|
931 | int_fast32_t exp;
|
---|
932 | union ui64_f64 uZ;
|
---|
933 |
|
---|
934 | if ( xPtr->isNaN ) {
|
---|
935 | uiZ = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
936 | goto uiZ;
|
---|
937 | }
|
---|
938 | if ( xPtr->isInf ) {
|
---|
939 | uiZ =
|
---|
940 | xPtr->sign ? UINT64_C( 0xFFF0000000000000 )
|
---|
941 | : UINT64_C( 0x7FF0000000000000 );
|
---|
942 | goto uiZ;
|
---|
943 | }
|
---|
944 | if ( xPtr->isZero ) {
|
---|
945 | uiZ = xPtr->sign ? UINT64_C( 0x8000000000000000 ) : 0;
|
---|
946 | goto uiZ;
|
---|
947 | }
|
---|
948 | x = *xPtr;
|
---|
949 | while ( UINT64_C( 0x0100000000000000 ) <= x.sig.v64 ) {
|
---|
950 | ++x.exp;
|
---|
951 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
952 | }
|
---|
953 | while ( x.sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
954 | --x.exp;
|
---|
955 | x.sig = shortShiftLeft128( x.sig, 1 );
|
---|
956 | }
|
---|
957 | savedX = x;
|
---|
958 | isTiny =
|
---|
959 | (slowfloat_detectTininess == softfloat_tininess_beforeRounding)
|
---|
960 | && (x.exp + 0x3FF <= 0);
|
---|
961 | roundFloatXTo53( isTiny, &x, slowfloat_roundingMode, true );
|
---|
962 | exp = x.exp + 0x3FF;
|
---|
963 | if ( 0x7FF <= exp ) {
|
---|
964 | slowfloat_exceptionFlags |=
|
---|
965 | softfloat_flag_overflow | softfloat_flag_inexact;
|
---|
966 | if ( x.sign ) {
|
---|
967 | switch ( slowfloat_roundingMode ) {
|
---|
968 | case softfloat_round_near_even:
|
---|
969 | case softfloat_round_min:
|
---|
970 | case softfloat_round_near_maxMag:
|
---|
971 | uiZ = UINT64_C( 0xFFF0000000000000 );
|
---|
972 | break;
|
---|
973 | case softfloat_round_minMag:
|
---|
974 | case softfloat_round_max:
|
---|
975 | case softfloat_round_odd:
|
---|
976 | uiZ = UINT64_C( 0xFFEFFFFFFFFFFFFF );
|
---|
977 | break;
|
---|
978 | }
|
---|
979 | } else {
|
---|
980 | switch ( slowfloat_roundingMode ) {
|
---|
981 | case softfloat_round_near_even:
|
---|
982 | case softfloat_round_max:
|
---|
983 | case softfloat_round_near_maxMag:
|
---|
984 | uiZ = UINT64_C( 0x7FF0000000000000 );
|
---|
985 | break;
|
---|
986 | case softfloat_round_minMag:
|
---|
987 | case softfloat_round_min:
|
---|
988 | case softfloat_round_odd:
|
---|
989 | uiZ = UINT64_C( 0x7FEFFFFFFFFFFFFF );
|
---|
990 | break;
|
---|
991 | }
|
---|
992 | }
|
---|
993 | goto uiZ;
|
---|
994 | }
|
---|
995 | if ( exp <= 0 ) {
|
---|
996 | isTiny = true;
|
---|
997 | x = savedX;
|
---|
998 | exp = x.exp + 0x3FF;
|
---|
999 | if ( exp < -56 ) {
|
---|
1000 | x.sig.v0 = (x.sig.v64 != 0) || (x.sig.v0 != 0);
|
---|
1001 | x.sig.v64 = 0;
|
---|
1002 | } else {
|
---|
1003 | while ( exp <= 0 ) {
|
---|
1004 | ++exp;
|
---|
1005 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
1006 | }
|
---|
1007 | }
|
---|
1008 | roundFloatXTo53( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1009 | exp = (UINT64_C( 0x0080000000000000 ) <= x.sig.v64) ? 1 : 0;
|
---|
1010 | }
|
---|
1011 | uiZ = (uint_fast64_t) exp<<52;
|
---|
1012 | if ( x.sign ) uiZ |= UINT64_C( 0x8000000000000000 );
|
---|
1013 | uiZ |= x.sig.v64>>3 & UINT64_C( 0x000FFFFFFFFFFFFF );
|
---|
1014 | uiZ:
|
---|
1015 | uZ.ui = uiZ;
|
---|
1016 | return uZ.f;
|
---|
1017 |
|
---|
1018 | }
|
---|
1019 |
|
---|
1020 | #endif
|
---|
1021 |
|
---|
1022 | #ifdef EXTFLOAT80
|
---|
1023 |
|
---|
1024 | static void extF80MToFloatX( const extFloat80_t *aPtr, struct floatX *xPtr )
|
---|
1025 | {
|
---|
1026 | const struct extFloat80M *aSPtr;
|
---|
1027 | uint_fast16_t uiA64;
|
---|
1028 | int_fast32_t exp;
|
---|
1029 | struct uint128 sig;
|
---|
1030 |
|
---|
1031 | aSPtr = (const struct extFloat80M *) aPtr;
|
---|
1032 | xPtr->isNaN = false;
|
---|
1033 | xPtr->isInf = false;
|
---|
1034 | xPtr->isZero = false;
|
---|
1035 | uiA64 = aSPtr->signExp;
|
---|
1036 | xPtr->sign = ((uiA64 & 0x8000) != 0);
|
---|
1037 | exp = uiA64 & 0x7FFF;
|
---|
1038 | sig.v64 = 0;
|
---|
1039 | sig.v0 = aSPtr->signif;
|
---|
1040 | if ( exp == 0x7FFF ) {
|
---|
1041 | if ( sig.v0 & UINT64_C( 0x7FFFFFFFFFFFFFFF ) ) {
|
---|
1042 | xPtr->isNaN = true;
|
---|
1043 | } else {
|
---|
1044 | xPtr->isInf = true;
|
---|
1045 | }
|
---|
1046 | } else {
|
---|
1047 | if ( !exp ) ++exp;
|
---|
1048 | exp -= 0x3FFF;
|
---|
1049 | if ( !(sig.v0 & UINT64_C( 0x8000000000000000 )) ) {
|
---|
1050 | if ( !sig.v0 ) {
|
---|
1051 | xPtr->isZero = true;
|
---|
1052 | } else {
|
---|
1053 | do {
|
---|
1054 | --exp;
|
---|
1055 | sig.v0 <<= 1;
|
---|
1056 | } while ( sig.v0 < UINT64_C( 0x8000000000000000 ) );
|
---|
1057 | }
|
---|
1058 | }
|
---|
1059 | xPtr->exp = exp;
|
---|
1060 | }
|
---|
1061 | xPtr->sig = shortShiftLeft128( sig, 56 );
|
---|
1062 |
|
---|
1063 | }
|
---|
1064 |
|
---|
1065 | static void floatXToExtF80M( const struct floatX *xPtr, extFloat80_t *zPtr )
|
---|
1066 | {
|
---|
1067 | struct extFloat80M *zSPtr;
|
---|
1068 | struct floatX x, savedX;
|
---|
1069 | bool isTiny;
|
---|
1070 | int_fast32_t exp;
|
---|
1071 | uint_fast64_t uiZ0;
|
---|
1072 | uint_fast16_t uiZ64;
|
---|
1073 |
|
---|
1074 | zSPtr = (struct extFloat80M *) zPtr;
|
---|
1075 | if ( xPtr->isNaN ) {
|
---|
1076 | zSPtr->signExp = 0xFFFF;
|
---|
1077 | zSPtr->signif = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
1078 | return;
|
---|
1079 | }
|
---|
1080 | if ( xPtr->isInf ) {
|
---|
1081 | zSPtr->signExp = xPtr->sign ? 0xFFFF : 0x7FFF;
|
---|
1082 | zSPtr->signif = UINT64_C( 0x8000000000000000 );
|
---|
1083 | return;
|
---|
1084 | }
|
---|
1085 | if ( xPtr->isZero ) {
|
---|
1086 | zSPtr->signExp = xPtr->sign ? 0x8000 : 0;
|
---|
1087 | zSPtr->signif = 0;
|
---|
1088 | return;
|
---|
1089 | }
|
---|
1090 | x = *xPtr;
|
---|
1091 | while ( UINT64_C( 0x0100000000000000 ) <= x.sig.v64 ) {
|
---|
1092 | ++x.exp;
|
---|
1093 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
1094 | }
|
---|
1095 | while ( x.sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
1096 | --x.exp;
|
---|
1097 | x.sig = shortShiftLeft128( x.sig, 1 );
|
---|
1098 | }
|
---|
1099 | savedX = x;
|
---|
1100 | isTiny =
|
---|
1101 | (slowfloat_detectTininess == softfloat_tininess_beforeRounding)
|
---|
1102 | && (x.exp + 0x3FFF <= 0);
|
---|
1103 | switch ( slow_extF80_roundingPrecision ) {
|
---|
1104 | case 32:
|
---|
1105 | roundFloatXTo24( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1106 | break;
|
---|
1107 | case 64:
|
---|
1108 | roundFloatXTo53( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1109 | break;
|
---|
1110 | default:
|
---|
1111 | roundFloatXTo64( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1112 | break;
|
---|
1113 | }
|
---|
1114 | exp = x.exp + 0x3FFF;
|
---|
1115 | if ( 0x7FFF <= exp ) {
|
---|
1116 | slowfloat_exceptionFlags |=
|
---|
1117 | softfloat_flag_overflow | softfloat_flag_inexact;
|
---|
1118 | if ( x.sign ) {
|
---|
1119 | switch ( slowfloat_roundingMode ) {
|
---|
1120 | case softfloat_round_near_even:
|
---|
1121 | case softfloat_round_min:
|
---|
1122 | case softfloat_round_near_maxMag:
|
---|
1123 | zSPtr->signExp = 0xFFFF;
|
---|
1124 | zSPtr->signif = UINT64_C( 0x8000000000000000 );
|
---|
1125 | break;
|
---|
1126 | case softfloat_round_minMag:
|
---|
1127 | case softfloat_round_max:
|
---|
1128 | case softfloat_round_odd:
|
---|
1129 | switch ( slow_extF80_roundingPrecision ) {
|
---|
1130 | case 32:
|
---|
1131 | uiZ0 = UINT64_C( 0xFFFFFF0000000000 );
|
---|
1132 | break;
|
---|
1133 | case 64:
|
---|
1134 | uiZ0 = UINT64_C( 0xFFFFFFFFFFFFF800 );
|
---|
1135 | break;
|
---|
1136 | default:
|
---|
1137 | uiZ0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
1138 | break;
|
---|
1139 | }
|
---|
1140 | zSPtr->signExp = 0xFFFE;
|
---|
1141 | zSPtr->signif = uiZ0;
|
---|
1142 | break;
|
---|
1143 | }
|
---|
1144 | } else {
|
---|
1145 | switch ( slowfloat_roundingMode ) {
|
---|
1146 | case softfloat_round_near_even:
|
---|
1147 | case softfloat_round_max:
|
---|
1148 | case softfloat_round_near_maxMag:
|
---|
1149 | zSPtr->signExp = 0x7FFF;
|
---|
1150 | zSPtr->signif = UINT64_C( 0x8000000000000000 );
|
---|
1151 | break;
|
---|
1152 | case softfloat_round_minMag:
|
---|
1153 | case softfloat_round_min:
|
---|
1154 | case softfloat_round_odd:
|
---|
1155 | switch ( slow_extF80_roundingPrecision ) {
|
---|
1156 | case 32:
|
---|
1157 | uiZ0 = UINT64_C( 0xFFFFFF0000000000 );
|
---|
1158 | break;
|
---|
1159 | case 64:
|
---|
1160 | uiZ0 = UINT64_C( 0xFFFFFFFFFFFFF800 );
|
---|
1161 | break;
|
---|
1162 | default:
|
---|
1163 | uiZ0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
1164 | break;
|
---|
1165 | }
|
---|
1166 | zSPtr->signExp = 0x7FFE;
|
---|
1167 | zSPtr->signif = uiZ0;
|
---|
1168 | break;
|
---|
1169 | }
|
---|
1170 | }
|
---|
1171 | return;
|
---|
1172 | }
|
---|
1173 | if ( exp <= 0 ) {
|
---|
1174 | isTiny = true;
|
---|
1175 | x = savedX;
|
---|
1176 | exp = x.exp + 0x3FFF;
|
---|
1177 | if ( exp < -70 ) {
|
---|
1178 | x.sig.v0 = (x.sig.v64 != 0) || (x.sig.v0 != 0);
|
---|
1179 | x.sig.v64 = 0;
|
---|
1180 | } else {
|
---|
1181 | while ( exp <= 0 ) {
|
---|
1182 | ++exp;
|
---|
1183 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
1184 | }
|
---|
1185 | }
|
---|
1186 | switch ( slow_extF80_roundingPrecision ) {
|
---|
1187 | case 32:
|
---|
1188 | roundFloatXTo24( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1189 | break;
|
---|
1190 | case 64:
|
---|
1191 | roundFloatXTo53( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1192 | break;
|
---|
1193 | default:
|
---|
1194 | roundFloatXTo64( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1195 | break;
|
---|
1196 | }
|
---|
1197 | exp = (UINT64_C( 0x0080000000000000 ) <= x.sig.v64) ? 1 : 0;
|
---|
1198 | }
|
---|
1199 | uiZ64 = exp;
|
---|
1200 | if ( x.sign ) uiZ64 |= 0x8000;
|
---|
1201 | zSPtr->signExp = uiZ64;
|
---|
1202 | zSPtr->signif = shortShiftRightJam128( x.sig, 56 ).v0;
|
---|
1203 |
|
---|
1204 | }
|
---|
1205 |
|
---|
1206 | #endif
|
---|
1207 |
|
---|
1208 | #ifdef FLOAT128
|
---|
1209 |
|
---|
1210 | static void f128MToFloatX( const float128_t *aPtr, struct floatX *xPtr )
|
---|
1211 | {
|
---|
1212 | const struct uint128 *uiAPtr;
|
---|
1213 | uint_fast64_t uiA64;
|
---|
1214 | int_fast32_t exp;
|
---|
1215 | struct uint128 sig;
|
---|
1216 |
|
---|
1217 | uiAPtr = (const struct uint128 *) aPtr;
|
---|
1218 | xPtr->isNaN = false;
|
---|
1219 | xPtr->isInf = false;
|
---|
1220 | xPtr->isZero = false;
|
---|
1221 | uiA64 = uiAPtr->v64;
|
---|
1222 | xPtr->sign = ((uiA64 & UINT64_C( 0x8000000000000000 )) != 0);
|
---|
1223 | exp = uiA64>>48 & 0x7FFF;
|
---|
1224 | sig.v64 = uiA64 & UINT64_C( 0x0000FFFFFFFFFFFF );
|
---|
1225 | sig.v0 = uiAPtr->v0;
|
---|
1226 | if ( exp == 0x7FFF ) {
|
---|
1227 | if ( sig.v64 || sig.v0 ) {
|
---|
1228 | xPtr->isNaN = true;
|
---|
1229 | } else {
|
---|
1230 | xPtr->isInf = true;
|
---|
1231 | }
|
---|
1232 | } else if ( !exp ) {
|
---|
1233 | if ( !sig.v64 && !sig.v0 ) {
|
---|
1234 | xPtr->isZero = true;
|
---|
1235 | } else {
|
---|
1236 | exp = 1 - 0x3FFF;
|
---|
1237 | do {
|
---|
1238 | --exp;
|
---|
1239 | sig = shortShiftLeft128( sig, 1 );
|
---|
1240 | } while ( sig.v64 < UINT64_C( 0x0001000000000000 ) );
|
---|
1241 | xPtr->exp = exp;
|
---|
1242 | }
|
---|
1243 | } else {
|
---|
1244 | xPtr->exp = exp - 0x3FFF;
|
---|
1245 | sig.v64 |= UINT64_C( 0x0001000000000000 );
|
---|
1246 | }
|
---|
1247 | xPtr->sig = shortShiftLeft128( sig, 7 );
|
---|
1248 |
|
---|
1249 | }
|
---|
1250 |
|
---|
1251 | static void floatXToF128M( const struct floatX *xPtr, float128_t *zPtr )
|
---|
1252 | {
|
---|
1253 | struct uint128 *uiZPtr;
|
---|
1254 | struct floatX x, savedX;
|
---|
1255 | bool isTiny;
|
---|
1256 | int_fast32_t exp;
|
---|
1257 | uint_fast64_t uiZ64;
|
---|
1258 |
|
---|
1259 | uiZPtr = (struct uint128 *) zPtr;
|
---|
1260 | if ( xPtr->isNaN ) {
|
---|
1261 | uiZPtr->v64 = uiZPtr->v0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
1262 | return;
|
---|
1263 | }
|
---|
1264 | if ( xPtr->isInf ) {
|
---|
1265 | uiZPtr->v64 =
|
---|
1266 | xPtr->sign ? UINT64_C( 0xFFFF000000000000 )
|
---|
1267 | : UINT64_C( 0x7FFF000000000000 );
|
---|
1268 | uiZPtr->v0 = 0;
|
---|
1269 | return;
|
---|
1270 | }
|
---|
1271 | if ( xPtr->isZero ) {
|
---|
1272 | uiZPtr->v64 = xPtr->sign ? UINT64_C( 0x8000000000000000 ) : 0;
|
---|
1273 | uiZPtr->v0 = 0;
|
---|
1274 | return;
|
---|
1275 | }
|
---|
1276 | x = *xPtr;
|
---|
1277 | while ( UINT64_C( 0x0100000000000000 ) <= x.sig.v64 ) {
|
---|
1278 | ++x.exp;
|
---|
1279 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
1280 | }
|
---|
1281 | while ( x.sig.v64 < UINT64_C( 0x0080000000000000 ) ) {
|
---|
1282 | --x.exp;
|
---|
1283 | x.sig = shortShiftLeft128( x.sig, 1 );
|
---|
1284 | }
|
---|
1285 | savedX = x;
|
---|
1286 | isTiny =
|
---|
1287 | (slowfloat_detectTininess == softfloat_tininess_beforeRounding)
|
---|
1288 | && (x.exp + 0x3FFF <= 0);
|
---|
1289 | roundFloatXTo113( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1290 | exp = x.exp + 0x3FFF;
|
---|
1291 | if ( 0x7FFF <= exp ) {
|
---|
1292 | slowfloat_exceptionFlags |=
|
---|
1293 | softfloat_flag_overflow | softfloat_flag_inexact;
|
---|
1294 | if ( x.sign ) {
|
---|
1295 | switch ( slowfloat_roundingMode ) {
|
---|
1296 | case softfloat_round_near_even:
|
---|
1297 | case softfloat_round_min:
|
---|
1298 | case softfloat_round_near_maxMag:
|
---|
1299 | uiZPtr->v64 = UINT64_C( 0xFFFF000000000000 );
|
---|
1300 | uiZPtr->v0 = 0;
|
---|
1301 | break;
|
---|
1302 | case softfloat_round_minMag:
|
---|
1303 | case softfloat_round_max:
|
---|
1304 | case softfloat_round_odd:
|
---|
1305 | uiZPtr->v64 = UINT64_C( 0xFFFEFFFFFFFFFFFF );
|
---|
1306 | uiZPtr->v0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
1307 | break;
|
---|
1308 | }
|
---|
1309 | } else {
|
---|
1310 | switch ( slowfloat_roundingMode ) {
|
---|
1311 | case softfloat_round_near_even:
|
---|
1312 | case softfloat_round_max:
|
---|
1313 | case softfloat_round_near_maxMag:
|
---|
1314 | uiZPtr->v64 = UINT64_C( 0x7FFF000000000000 );
|
---|
1315 | uiZPtr->v0 = 0;
|
---|
1316 | break;
|
---|
1317 | case softfloat_round_minMag:
|
---|
1318 | case softfloat_round_min:
|
---|
1319 | case softfloat_round_odd:
|
---|
1320 | uiZPtr->v64 = UINT64_C( 0x7FFEFFFFFFFFFFFF );
|
---|
1321 | uiZPtr->v0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
---|
1322 | break;
|
---|
1323 | }
|
---|
1324 | }
|
---|
1325 | return;
|
---|
1326 | }
|
---|
1327 | if ( exp <= 0 ) {
|
---|
1328 | isTiny = true;
|
---|
1329 | x = savedX;
|
---|
1330 | exp = x.exp + 0x3FFF;
|
---|
1331 | if ( exp < -120 ) {
|
---|
1332 | x.sig.v0 = (x.sig.v64 != 0) || (x.sig.v0 != 0);
|
---|
1333 | x.sig.v64 = 0;
|
---|
1334 | } else {
|
---|
1335 | while ( exp <= 0 ) {
|
---|
1336 | ++exp;
|
---|
1337 | x.sig = shortShiftRightJam128( x.sig, 1 );
|
---|
1338 | }
|
---|
1339 | }
|
---|
1340 | roundFloatXTo113( isTiny, &x, slowfloat_roundingMode, true );
|
---|
1341 | exp = (UINT64_C( 0x0080000000000000 ) <= x.sig.v64) ? 1 : 0;
|
---|
1342 | }
|
---|
1343 | uiZ64 = (uint_fast64_t) exp<<48;
|
---|
1344 | if ( x.sign ) uiZ64 |= UINT64_C( 0x8000000000000000 );
|
---|
1345 | x.sig = shortShiftRightJam128( x.sig, 7 );
|
---|
1346 | uiZPtr->v64 = uiZ64 | (x.sig.v64 & UINT64_C( 0x0000FFFFFFFFFFFF ));
|
---|
1347 | uiZPtr->v0 = x.sig.v0;
|
---|
1348 |
|
---|
1349 | }
|
---|
1350 |
|
---|
1351 | #endif
|
---|
1352 |
|
---|
1353 | static void floatXInvalid( struct floatX *xPtr )
|
---|
1354 | {
|
---|
1355 |
|
---|
1356 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
1357 | *xPtr = floatXNaN;
|
---|
1358 |
|
---|
1359 | }
|
---|
1360 |
|
---|
1361 | static
|
---|
1362 | void
|
---|
1363 | floatXRoundToInt( struct floatX *xPtr, uint_fast8_t roundingMode, bool exact )
|
---|
1364 | {
|
---|
1365 | int_fast32_t exp, shiftDist;
|
---|
1366 | struct uint128 sig;
|
---|
1367 |
|
---|
1368 | if ( xPtr->isNaN || xPtr->isInf ) return;
|
---|
1369 | exp = xPtr->exp;
|
---|
1370 | shiftDist = 112 - exp;
|
---|
1371 | if ( shiftDist <= 0 ) return;
|
---|
1372 | if ( 119 < shiftDist ) {
|
---|
1373 | xPtr->exp = 112;
|
---|
1374 | xPtr->sig.v64 = 0;
|
---|
1375 | xPtr->sig.v0 = !xPtr->isZero;
|
---|
1376 | } else {
|
---|
1377 | sig = xPtr->sig;
|
---|
1378 | while ( 0 < shiftDist ) {
|
---|
1379 | ++exp;
|
---|
1380 | sig = shortShiftRightJam128( sig, 1 );
|
---|
1381 | --shiftDist;
|
---|
1382 | }
|
---|
1383 | xPtr->exp = exp;
|
---|
1384 | xPtr->sig = sig;
|
---|
1385 | }
|
---|
1386 | roundFloatXTo113( false, xPtr, roundingMode, exact );
|
---|
1387 | if ( !xPtr->sig.v64 && !xPtr->sig.v0 ) xPtr->isZero = true;
|
---|
1388 |
|
---|
1389 | }
|
---|
1390 |
|
---|
1391 | static void floatXAdd( struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1392 | {
|
---|
1393 | int_fast32_t expX, expY, expDiff;
|
---|
1394 | struct uint128 sigY;
|
---|
1395 |
|
---|
1396 | if ( xPtr->isNaN ) return;
|
---|
1397 | if ( yPtr->isNaN ) goto copyY;
|
---|
1398 | if ( xPtr->isInf && yPtr->isInf ) {
|
---|
1399 | if ( xPtr->sign != yPtr->sign ) floatXInvalid( xPtr );
|
---|
1400 | return;
|
---|
1401 | }
|
---|
1402 | if ( xPtr->isInf ) return;
|
---|
1403 | if ( yPtr->isInf ) goto copyY;
|
---|
1404 | if ( xPtr->isZero && yPtr->isZero ) {
|
---|
1405 | if ( xPtr->sign == yPtr->sign ) return;
|
---|
1406 | goto completeCancellation;
|
---|
1407 | }
|
---|
1408 | expX = xPtr->exp;
|
---|
1409 | expY = yPtr->exp;
|
---|
1410 | if (
|
---|
1411 | (xPtr->sign != yPtr->sign) && (expX == expY)
|
---|
1412 | && eq128( xPtr->sig, yPtr->sig )
|
---|
1413 | ) {
|
---|
1414 | completeCancellation:
|
---|
1415 | if (slowfloat_roundingMode == softfloat_round_min) {
|
---|
1416 | *xPtr = floatXNegativeZero;
|
---|
1417 | } else {
|
---|
1418 | *xPtr = floatXPositiveZero;
|
---|
1419 | }
|
---|
1420 | return;
|
---|
1421 | }
|
---|
1422 | if ( xPtr->isZero ) goto copyY;
|
---|
1423 | if ( yPtr->isZero ) return;
|
---|
1424 | expDiff = expX - expY;
|
---|
1425 | if ( expDiff < 0 ) {
|
---|
1426 | xPtr->exp = expY;
|
---|
1427 | if ( expDiff < -120 ) {
|
---|
1428 | xPtr->sig.v64 = 0;
|
---|
1429 | xPtr->sig.v0 = 1;
|
---|
1430 | } else {
|
---|
1431 | while ( expDiff < 0 ) {
|
---|
1432 | ++expDiff;
|
---|
1433 | xPtr->sig = shortShiftRightJam128( xPtr->sig, 1 );
|
---|
1434 | }
|
---|
1435 | }
|
---|
1436 | if ( xPtr->sign != yPtr->sign ) xPtr->sig = neg128( xPtr->sig );
|
---|
1437 | xPtr->sign = yPtr->sign;
|
---|
1438 | xPtr->sig = add128( xPtr->sig, yPtr->sig );
|
---|
1439 | } else {
|
---|
1440 | sigY = yPtr->sig;
|
---|
1441 | if ( 120 < expDiff ) {
|
---|
1442 | sigY.v64 = 0;
|
---|
1443 | sigY.v0 = 1;
|
---|
1444 | } else {
|
---|
1445 | while ( 0 < expDiff ) {
|
---|
1446 | --expDiff;
|
---|
1447 | sigY = shortShiftRightJam128( sigY, 1 );
|
---|
1448 | }
|
---|
1449 | }
|
---|
1450 | if ( xPtr->sign != yPtr->sign ) sigY = neg128( sigY );
|
---|
1451 | xPtr->sig = add128( xPtr->sig, sigY );
|
---|
1452 | }
|
---|
1453 | if ( xPtr->sig.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
---|
1454 | xPtr->sign = !xPtr->sign;
|
---|
1455 | xPtr->sig = neg128( xPtr->sig );
|
---|
1456 | }
|
---|
1457 | return;
|
---|
1458 | copyY:
|
---|
1459 | *xPtr = *yPtr;
|
---|
1460 |
|
---|
1461 | }
|
---|
1462 |
|
---|
1463 | static void floatXMul( struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1464 | {
|
---|
1465 | struct uint128 sig;
|
---|
1466 | int bitNum;
|
---|
1467 |
|
---|
1468 | if ( xPtr->isNaN ) return;
|
---|
1469 | if ( yPtr->isNaN ) {
|
---|
1470 | xPtr->isNaN = true;
|
---|
1471 | xPtr->isInf = false;
|
---|
1472 | xPtr->isZero = false;
|
---|
1473 | xPtr->sign = yPtr->sign;
|
---|
1474 | return;
|
---|
1475 | }
|
---|
1476 | if ( yPtr->sign ) xPtr->sign = !xPtr->sign;
|
---|
1477 | if ( xPtr->isInf ) {
|
---|
1478 | if ( yPtr->isZero ) floatXInvalid( xPtr );
|
---|
1479 | return;
|
---|
1480 | }
|
---|
1481 | if ( yPtr->isInf ) {
|
---|
1482 | if ( xPtr->isZero ) {
|
---|
1483 | floatXInvalid( xPtr );
|
---|
1484 | return;
|
---|
1485 | }
|
---|
1486 | xPtr->isInf = true;
|
---|
1487 | return;
|
---|
1488 | }
|
---|
1489 | if ( xPtr->isZero || yPtr->isZero ) {
|
---|
1490 | if ( xPtr->sign ) {
|
---|
1491 | *xPtr = floatXNegativeZero;
|
---|
1492 | } else {
|
---|
1493 | *xPtr = floatXPositiveZero;
|
---|
1494 | }
|
---|
1495 | return;
|
---|
1496 | }
|
---|
1497 | xPtr->exp += yPtr->exp;
|
---|
1498 | sig.v64 = 0;
|
---|
1499 | sig.v0 = 0;
|
---|
1500 | for ( bitNum = 0; bitNum < 120; ++bitNum ) {
|
---|
1501 | sig = shortShiftRightJam128( sig, 1 );
|
---|
1502 | if ( xPtr->sig.v0 & 1 ) sig = add128( sig, yPtr->sig );
|
---|
1503 | xPtr->sig = shortShiftRight128( xPtr->sig, 1 );
|
---|
1504 | }
|
---|
1505 | if ( UINT64_C( 0x0100000000000000 ) <= sig.v64 ) {
|
---|
1506 | ++xPtr->exp;
|
---|
1507 | sig = shortShiftRightJam128( sig, 1 );
|
---|
1508 | }
|
---|
1509 | xPtr->sig = sig;
|
---|
1510 |
|
---|
1511 | }
|
---|
1512 |
|
---|
1513 | static void floatXDiv( struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1514 | {
|
---|
1515 | struct uint128 sig, negSigY;
|
---|
1516 | int bitNum;
|
---|
1517 |
|
---|
1518 | if ( xPtr->isNaN ) return;
|
---|
1519 | if ( yPtr->isNaN ) {
|
---|
1520 | xPtr->isNaN = true;
|
---|
1521 | xPtr->isInf = false;
|
---|
1522 | xPtr->isZero = false;
|
---|
1523 | xPtr->sign = yPtr->sign;
|
---|
1524 | return;
|
---|
1525 | }
|
---|
1526 | if ( yPtr->sign ) xPtr->sign = !xPtr->sign;
|
---|
1527 | if ( xPtr->isInf ) {
|
---|
1528 | if ( yPtr->isInf ) floatXInvalid( xPtr );
|
---|
1529 | return;
|
---|
1530 | }
|
---|
1531 | if ( yPtr->isZero ) {
|
---|
1532 | if ( xPtr->isZero ) {
|
---|
1533 | floatXInvalid( xPtr );
|
---|
1534 | return;
|
---|
1535 | }
|
---|
1536 | slowfloat_exceptionFlags |= softfloat_flag_infinite;
|
---|
1537 | xPtr->isInf = true;
|
---|
1538 | return;
|
---|
1539 | }
|
---|
1540 | if ( xPtr->isZero || yPtr->isInf ) {
|
---|
1541 | if ( xPtr->sign ) {
|
---|
1542 | *xPtr = floatXNegativeZero;
|
---|
1543 | } else {
|
---|
1544 | *xPtr = floatXPositiveZero;
|
---|
1545 | }
|
---|
1546 | return;
|
---|
1547 | }
|
---|
1548 | xPtr->exp -= yPtr->exp + 1;
|
---|
1549 | sig.v64 = 0;
|
---|
1550 | sig.v0 = 0;
|
---|
1551 | negSigY = neg128( yPtr->sig );
|
---|
1552 | for ( bitNum = 0; bitNum < 120; ++bitNum ) {
|
---|
1553 | if ( le128( yPtr->sig, xPtr->sig ) ) {
|
---|
1554 | sig.v0 |= 1;
|
---|
1555 | xPtr->sig = add128( xPtr->sig, negSigY );
|
---|
1556 | }
|
---|
1557 | xPtr->sig = shortShiftLeft128( xPtr->sig, 1 );
|
---|
1558 | sig = shortShiftLeft128( sig, 1 );
|
---|
1559 | }
|
---|
1560 | if ( xPtr->sig.v64 || xPtr->sig.v0 ) sig.v0 |= 1;
|
---|
1561 | xPtr->sig = sig;
|
---|
1562 |
|
---|
1563 | }
|
---|
1564 |
|
---|
1565 | static void floatXRem( struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1566 | {
|
---|
1567 | int_fast32_t expX, expY;
|
---|
1568 | struct uint128 sigY, negSigY;
|
---|
1569 | bool lastQuotientBit;
|
---|
1570 | struct uint128 savedSigX;
|
---|
1571 |
|
---|
1572 | if ( xPtr->isNaN ) return;
|
---|
1573 | if ( yPtr->isNaN ) {
|
---|
1574 | xPtr->isNaN = true;
|
---|
1575 | xPtr->isInf = false;
|
---|
1576 | xPtr->isZero = false;
|
---|
1577 | xPtr->sign = yPtr->sign;
|
---|
1578 | return;
|
---|
1579 | }
|
---|
1580 | if ( xPtr->isInf || yPtr->isZero ) {
|
---|
1581 | floatXInvalid( xPtr );
|
---|
1582 | return;
|
---|
1583 | }
|
---|
1584 | if ( xPtr->isZero || yPtr->isInf ) return;
|
---|
1585 | expX = xPtr->exp;
|
---|
1586 | expY = yPtr->exp - 1;
|
---|
1587 | if ( expX < expY ) return;
|
---|
1588 | sigY = shortShiftLeft128( yPtr->sig, 1 );
|
---|
1589 | negSigY = neg128( sigY );
|
---|
1590 | while ( expY < expX ) {
|
---|
1591 | --expX;
|
---|
1592 | if ( le128( sigY, xPtr->sig ) ) {
|
---|
1593 | xPtr->sig = add128( xPtr->sig, negSigY );
|
---|
1594 | }
|
---|
1595 | xPtr->sig = shortShiftLeft128( xPtr->sig, 1 );
|
---|
1596 | }
|
---|
1597 | xPtr->exp = expX;
|
---|
1598 | lastQuotientBit = le128( sigY, xPtr->sig );
|
---|
1599 | if ( lastQuotientBit ) xPtr->sig = add128( xPtr->sig, negSigY );
|
---|
1600 | savedSigX = xPtr->sig;
|
---|
1601 | xPtr->sig = neg128( add128( xPtr->sig, negSigY ) );
|
---|
1602 | if ( lt128( xPtr->sig, savedSigX ) ) {
|
---|
1603 | xPtr->sign = !xPtr->sign;
|
---|
1604 | } else if ( lt128( savedSigX, xPtr->sig ) ) {
|
---|
1605 | goto restoreSavedSigX;
|
---|
1606 | } else {
|
---|
1607 | if ( lastQuotientBit ) {
|
---|
1608 | xPtr->sign = !xPtr->sign;
|
---|
1609 | } else {
|
---|
1610 | restoreSavedSigX:
|
---|
1611 | xPtr->sig = savedSigX;
|
---|
1612 | }
|
---|
1613 | }
|
---|
1614 | if ( !xPtr->sig.v64 && !xPtr->sig.v0 ) xPtr->isZero = true;
|
---|
1615 |
|
---|
1616 | }
|
---|
1617 |
|
---|
1618 | static void floatXSqrt( struct floatX *xPtr )
|
---|
1619 | {
|
---|
1620 | struct uint128 sig, bitSig;
|
---|
1621 | int bitNum;
|
---|
1622 | struct uint128 savedSigX;
|
---|
1623 |
|
---|
1624 | if ( xPtr->isNaN || xPtr->isZero ) return;
|
---|
1625 | if ( xPtr->sign ) {
|
---|
1626 | floatXInvalid( xPtr );
|
---|
1627 | return;
|
---|
1628 | }
|
---|
1629 | if ( xPtr->isInf ) return;
|
---|
1630 | if ( !(xPtr->exp & 1) ) xPtr->sig = shortShiftRightJam128( xPtr->sig, 1 );
|
---|
1631 | xPtr->exp >>= 1;
|
---|
1632 | sig.v64 = 0;
|
---|
1633 | sig.v0 = 0;
|
---|
1634 | bitSig.v64 = UINT64_C( 0x0080000000000000 );
|
---|
1635 | bitSig.v0 = 0;
|
---|
1636 | for ( bitNum = 0; bitNum < 120; ++bitNum ) {
|
---|
1637 | savedSigX = xPtr->sig;
|
---|
1638 | xPtr->sig = add128( xPtr->sig, neg128( sig ) );
|
---|
1639 | xPtr->sig = shortShiftLeft128( xPtr->sig, 1 );
|
---|
1640 | xPtr->sig = add128( xPtr->sig, neg128( bitSig ) );
|
---|
1641 | if ( xPtr->sig.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
---|
1642 | xPtr->sig = shortShiftLeft128( savedSigX, 1 );
|
---|
1643 | } else {
|
---|
1644 | sig.v64 |= bitSig.v64;
|
---|
1645 | sig.v0 |= bitSig.v0;
|
---|
1646 | }
|
---|
1647 | bitSig = shortShiftRightJam128( bitSig, 1 );
|
---|
1648 | }
|
---|
1649 | if ( xPtr->sig.v64 || xPtr->sig.v0 ) sig.v0 |= 1;
|
---|
1650 | xPtr->sig = sig;
|
---|
1651 |
|
---|
1652 | }
|
---|
1653 |
|
---|
1654 | static bool floatXEq( const struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1655 | {
|
---|
1656 |
|
---|
1657 | if ( xPtr->isNaN || yPtr->isNaN ) return false;
|
---|
1658 | if ( xPtr->isZero && yPtr->isZero ) return true;
|
---|
1659 | if ( xPtr->sign != yPtr->sign ) return false;
|
---|
1660 | if ( xPtr->isInf || yPtr->isInf ) return xPtr->isInf && yPtr->isInf;
|
---|
1661 | return ( xPtr->exp == yPtr->exp ) && eq128( xPtr->sig, yPtr->sig );
|
---|
1662 |
|
---|
1663 | }
|
---|
1664 |
|
---|
1665 | static bool floatXLe( const struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1666 | {
|
---|
1667 |
|
---|
1668 | if ( xPtr->isNaN || yPtr->isNaN ) return false;
|
---|
1669 | if ( xPtr->isZero && yPtr->isZero ) return true;
|
---|
1670 | if ( xPtr->sign != yPtr->sign ) return xPtr->sign;
|
---|
1671 | if ( xPtr->sign ) {
|
---|
1672 | if ( xPtr->isInf || yPtr->isZero ) return true;
|
---|
1673 | if ( yPtr->isInf || xPtr->isZero ) return false;
|
---|
1674 | if ( yPtr->exp < xPtr->exp ) return true;
|
---|
1675 | if ( xPtr->exp < yPtr->exp ) return false;
|
---|
1676 | return le128( yPtr->sig, xPtr->sig );
|
---|
1677 | } else {
|
---|
1678 | if ( yPtr->isInf || xPtr->isZero ) return true;
|
---|
1679 | if ( xPtr->isInf || yPtr->isZero ) return false;
|
---|
1680 | if ( xPtr->exp < yPtr->exp ) return true;
|
---|
1681 | if ( yPtr->exp < xPtr->exp ) return false;
|
---|
1682 | return le128( xPtr->sig, yPtr->sig );
|
---|
1683 | }
|
---|
1684 |
|
---|
1685 | }
|
---|
1686 |
|
---|
1687 | static bool floatXLt( const struct floatX *xPtr, const struct floatX *yPtr )
|
---|
1688 | {
|
---|
1689 |
|
---|
1690 | if ( xPtr->isNaN || yPtr->isNaN ) return false;
|
---|
1691 | if ( xPtr->isZero && yPtr->isZero ) return false;
|
---|
1692 | if ( xPtr->sign != yPtr->sign ) return xPtr->sign;
|
---|
1693 | if ( xPtr->isInf && yPtr->isInf ) return false;
|
---|
1694 | if ( xPtr->sign ) {
|
---|
1695 | if ( xPtr->isInf || yPtr->isZero ) return true;
|
---|
1696 | if ( yPtr->isInf || xPtr->isZero ) return false;
|
---|
1697 | if ( yPtr->exp < xPtr->exp ) return true;
|
---|
1698 | if ( xPtr->exp < yPtr->exp ) return false;
|
---|
1699 | return lt128( yPtr->sig, xPtr->sig );
|
---|
1700 | } else {
|
---|
1701 | if ( yPtr->isInf || xPtr->isZero ) return true;
|
---|
1702 | if ( xPtr->isInf || yPtr->isZero ) return false;
|
---|
1703 | if ( xPtr->exp < yPtr->exp ) return true;
|
---|
1704 | if ( yPtr->exp < xPtr->exp ) return false;
|
---|
1705 | return lt128( xPtr->sig, yPtr->sig );
|
---|
1706 | }
|
---|
1707 |
|
---|
1708 | }
|
---|
1709 |
|
---|
1710 | /*----------------------------------------------------------------------------
|
---|
1711 | *----------------------------------------------------------------------------*/
|
---|
1712 |
|
---|
1713 | #if defined EXTFLOAT80 || defined FLOAT128
|
---|
1714 |
|
---|
1715 | #ifdef LITTLEENDIAN
|
---|
1716 | struct uint256 { uint64_t v0, v64, v128, v192; };
|
---|
1717 | #else
|
---|
1718 | struct uint256 { uint64_t v192, v128, v64, v0; };
|
---|
1719 | #endif
|
---|
1720 |
|
---|
1721 | static bool eq256M( const struct uint256 *aPtr, const struct uint256 *bPtr )
|
---|
1722 | {
|
---|
1723 |
|
---|
1724 | return
|
---|
1725 | (aPtr->v192 == bPtr->v192) && (aPtr->v128 == bPtr->v128)
|
---|
1726 | && (aPtr->v64 == bPtr->v64) && (aPtr->v0 == bPtr->v0);
|
---|
1727 |
|
---|
1728 | }
|
---|
1729 |
|
---|
1730 | static void shiftLeft1256M( struct uint256 *ptr )
|
---|
1731 | {
|
---|
1732 | uint64_t dword1, dword2;
|
---|
1733 |
|
---|
1734 | dword1 = ptr->v128;
|
---|
1735 | ptr->v192 = ptr->v192<<1 | dword1>>63;
|
---|
1736 | dword2 = ptr->v64;
|
---|
1737 | ptr->v128 = dword1<<1 | dword2>>63;
|
---|
1738 | dword1 = ptr->v0;
|
---|
1739 | ptr->v64 = dword2<<1 | dword1>>63;
|
---|
1740 | ptr->v0 = dword1<<1;
|
---|
1741 |
|
---|
1742 | }
|
---|
1743 |
|
---|
1744 | static void shiftRight1256M( struct uint256 *ptr )
|
---|
1745 | {
|
---|
1746 | uint64_t dword1, dword2;
|
---|
1747 |
|
---|
1748 | dword1 = ptr->v64;
|
---|
1749 | ptr->v0 = dword1<<63 | ptr->v0>>1;
|
---|
1750 | dword2 = ptr->v128;
|
---|
1751 | ptr->v64 = dword2<<63 | dword1>>1;
|
---|
1752 | dword1 = ptr->v192;
|
---|
1753 | ptr->v128 = dword1<<63 | dword2>>1;
|
---|
1754 | ptr->v192 = dword1>>1;
|
---|
1755 |
|
---|
1756 | }
|
---|
1757 |
|
---|
1758 | static void shiftRight1Jam256M( struct uint256 *ptr )
|
---|
1759 | {
|
---|
1760 | int extra;
|
---|
1761 |
|
---|
1762 | extra = ptr->v0 & 1;
|
---|
1763 | shiftRight1256M( ptr );
|
---|
1764 | ptr->v0 |= extra;
|
---|
1765 |
|
---|
1766 | }
|
---|
1767 |
|
---|
1768 | static void neg256M( struct uint256 *ptr )
|
---|
1769 | {
|
---|
1770 | uint64_t v64, v0, v128;
|
---|
1771 |
|
---|
1772 | v64 = ptr->v64;
|
---|
1773 | v0 = ptr->v0;
|
---|
1774 | if ( v64 | v0 ) {
|
---|
1775 | ptr->v192 = ~ptr->v192;
|
---|
1776 | ptr->v128 = ~ptr->v128;
|
---|
1777 | if ( v0 ) {
|
---|
1778 | ptr->v64 = ~v64;
|
---|
1779 | ptr->v0 = -v0;
|
---|
1780 | } else {
|
---|
1781 | ptr->v64 = -v64;
|
---|
1782 | }
|
---|
1783 | } else {
|
---|
1784 | v128 = ptr->v128;
|
---|
1785 | if ( v128 ) {
|
---|
1786 | ptr->v192 = ~ptr->v192;
|
---|
1787 | ptr->v128 = -v128;
|
---|
1788 | } else {
|
---|
1789 | ptr->v192 = -ptr->v192;
|
---|
1790 | }
|
---|
1791 | }
|
---|
1792 |
|
---|
1793 | }
|
---|
1794 |
|
---|
1795 | static void add256M( struct uint256 *aPtr, const struct uint256 *bPtr )
|
---|
1796 | {
|
---|
1797 | uint64_t dwordA, dwordZ;
|
---|
1798 | unsigned int carry1, carry2;
|
---|
1799 |
|
---|
1800 | dwordA = aPtr->v0;
|
---|
1801 | dwordZ = dwordA + bPtr->v0;
|
---|
1802 | carry1 = (dwordZ < dwordA);
|
---|
1803 | aPtr->v0 = dwordZ;
|
---|
1804 | dwordA = aPtr->v64;
|
---|
1805 | dwordZ = dwordA + bPtr->v64;
|
---|
1806 | carry2 = (dwordZ < dwordA);
|
---|
1807 | dwordZ += carry1;
|
---|
1808 | carry2 += (dwordZ < carry1);
|
---|
1809 | aPtr->v64 = dwordZ;
|
---|
1810 | dwordA = aPtr->v128;
|
---|
1811 | dwordZ = dwordA + bPtr->v128;
|
---|
1812 | carry1 = (dwordZ < dwordA);
|
---|
1813 | dwordZ += carry2;
|
---|
1814 | carry1 += (dwordZ < carry2);
|
---|
1815 | aPtr->v128 = dwordZ;
|
---|
1816 | aPtr->v192 = aPtr->v192 + bPtr->v192 + carry1;
|
---|
1817 |
|
---|
1818 | }
|
---|
1819 |
|
---|
1820 | struct floatX256 {
|
---|
1821 | bool isNaN;
|
---|
1822 | bool isInf;
|
---|
1823 | bool isZero;
|
---|
1824 | bool sign;
|
---|
1825 | int_fast32_t exp;
|
---|
1826 | struct uint256 sig;
|
---|
1827 | };
|
---|
1828 |
|
---|
1829 | static const struct floatX256 floatX256NaN =
|
---|
1830 | { true, false, false, false, 0, { 0, 0, 0, 0 } };
|
---|
1831 | static const struct floatX256 floatX256PositiveZero =
|
---|
1832 | { false, false, true, false, 0, { 0, 0, 0, 0 } };
|
---|
1833 | static const struct floatX256 floatX256NegativeZero =
|
---|
1834 | { false, false, true, true, 0, { 0, 0, 0, 0 } };
|
---|
1835 |
|
---|
1836 | #ifdef FLOAT128
|
---|
1837 |
|
---|
1838 | static void f128MToFloatX256( const float128_t *aPtr, struct floatX256 *xPtr )
|
---|
1839 | {
|
---|
1840 | struct floatX x;
|
---|
1841 |
|
---|
1842 | f128MToFloatX( aPtr, &x );
|
---|
1843 | xPtr->isNaN = x.isNaN;
|
---|
1844 | xPtr->isInf = x.isInf;
|
---|
1845 | xPtr->isZero = x.isZero;
|
---|
1846 | xPtr->sign = x.sign;
|
---|
1847 | xPtr->exp = x.exp;
|
---|
1848 | xPtr->sig.v192 = x.sig.v64;
|
---|
1849 | xPtr->sig.v128 = x.sig.v0;
|
---|
1850 | xPtr->sig.v64 = 0;
|
---|
1851 | xPtr->sig.v0 = 0;
|
---|
1852 |
|
---|
1853 | }
|
---|
1854 |
|
---|
1855 | static void floatX256ToF128M( const struct floatX256 *xPtr, float128_t *zPtr )
|
---|
1856 | {
|
---|
1857 | struct floatX x;
|
---|
1858 | int_fast32_t expZ;
|
---|
1859 | struct uint256 sig;
|
---|
1860 |
|
---|
1861 | x.isNaN = xPtr->isNaN;
|
---|
1862 | x.isInf = xPtr->isInf;
|
---|
1863 | x.isZero = xPtr->isZero;
|
---|
1864 | x.sign = xPtr->sign;
|
---|
1865 | if ( !(x.isNaN | x.isInf | x.isZero) ) {
|
---|
1866 | expZ = xPtr->exp;
|
---|
1867 | sig = xPtr->sig;
|
---|
1868 | while ( !sig.v192 ) {
|
---|
1869 | expZ -= 64;
|
---|
1870 | sig.v192 = sig.v128;
|
---|
1871 | sig.v128 = sig.v64;
|
---|
1872 | sig.v64 = sig.v0;
|
---|
1873 | sig.v0 = 0;
|
---|
1874 | }
|
---|
1875 | while ( sig.v192 < UINT64_C( 0x0100000000000000 ) ) {
|
---|
1876 | --expZ;
|
---|
1877 | shiftLeft1256M( &sig );
|
---|
1878 | }
|
---|
1879 | x.exp = expZ;
|
---|
1880 | x.sig.v64 = sig.v192;
|
---|
1881 | x.sig.v0 = sig.v128 | ((sig.v64 | sig.v0) != 0);
|
---|
1882 | }
|
---|
1883 | floatXToF128M( &x, zPtr );
|
---|
1884 |
|
---|
1885 | }
|
---|
1886 |
|
---|
1887 | #endif
|
---|
1888 |
|
---|
1889 | static void floatX256Invalid( struct floatX256 *xPtr )
|
---|
1890 | {
|
---|
1891 |
|
---|
1892 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
1893 | *xPtr = floatX256NaN;
|
---|
1894 |
|
---|
1895 | }
|
---|
1896 |
|
---|
1897 | static
|
---|
1898 | void floatX256Add( struct floatX256 *xPtr, const struct floatX256 *yPtr )
|
---|
1899 | {
|
---|
1900 | int_fast32_t expX, expY, expDiff;
|
---|
1901 | struct uint256 sigY;
|
---|
1902 |
|
---|
1903 | if ( xPtr->isNaN ) return;
|
---|
1904 | if ( yPtr->isNaN ) goto copyY;
|
---|
1905 | if ( xPtr->isInf && yPtr->isInf ) {
|
---|
1906 | if ( xPtr->sign != yPtr->sign ) floatX256Invalid( xPtr );
|
---|
1907 | return;
|
---|
1908 | }
|
---|
1909 | if ( xPtr->isInf ) return;
|
---|
1910 | if ( yPtr->isInf ) goto copyY;
|
---|
1911 | if ( xPtr->isZero && yPtr->isZero ) {
|
---|
1912 | if ( xPtr->sign == yPtr->sign ) return;
|
---|
1913 | goto completeCancellation;
|
---|
1914 | }
|
---|
1915 | expX = xPtr->exp;
|
---|
1916 | expY = yPtr->exp;
|
---|
1917 | if (
|
---|
1918 | (xPtr->sign != yPtr->sign) && (expX == expY)
|
---|
1919 | && eq256M( &xPtr->sig, &yPtr->sig )
|
---|
1920 | ) {
|
---|
1921 | completeCancellation:
|
---|
1922 | if (slowfloat_roundingMode == softfloat_round_min) {
|
---|
1923 | *xPtr = floatX256NegativeZero;
|
---|
1924 | } else {
|
---|
1925 | *xPtr = floatX256PositiveZero;
|
---|
1926 | }
|
---|
1927 | return;
|
---|
1928 | }
|
---|
1929 | if ( xPtr->isZero ) goto copyY;
|
---|
1930 | if ( yPtr->isZero ) return;
|
---|
1931 | expDiff = expX - expY;
|
---|
1932 | if ( expDiff < 0 ) {
|
---|
1933 | xPtr->exp = expY;
|
---|
1934 | if ( expDiff < -248 ) {
|
---|
1935 | xPtr->sig.v192 = 0;
|
---|
1936 | xPtr->sig.v128 = 0;
|
---|
1937 | xPtr->sig.v64 = 0;
|
---|
1938 | xPtr->sig.v0 = 1;
|
---|
1939 | } else {
|
---|
1940 | while ( expDiff < 0 ) {
|
---|
1941 | ++expDiff;
|
---|
1942 | shiftRight1Jam256M( &xPtr->sig );
|
---|
1943 | }
|
---|
1944 | }
|
---|
1945 | if ( xPtr->sign != yPtr->sign ) neg256M( &xPtr->sig );
|
---|
1946 | xPtr->sign = yPtr->sign;
|
---|
1947 | add256M( &xPtr->sig, &yPtr->sig );
|
---|
1948 | } else {
|
---|
1949 | sigY = yPtr->sig;
|
---|
1950 | if ( 248 < expDiff ) {
|
---|
1951 | sigY.v192 = 0;
|
---|
1952 | sigY.v128 = 0;
|
---|
1953 | sigY.v64 = 0;
|
---|
1954 | sigY.v0 = 1;
|
---|
1955 | } else {
|
---|
1956 | while ( 0 < expDiff ) {
|
---|
1957 | --expDiff;
|
---|
1958 | shiftRight1Jam256M( &sigY );
|
---|
1959 | }
|
---|
1960 | }
|
---|
1961 | if ( xPtr->sign != yPtr->sign ) neg256M( &sigY );
|
---|
1962 | add256M( &xPtr->sig, &sigY );
|
---|
1963 | }
|
---|
1964 | if ( xPtr->sig.v192 & UINT64_C( 0x8000000000000000 ) ) {
|
---|
1965 | xPtr->sign = !xPtr->sign;
|
---|
1966 | neg256M( &xPtr->sig );
|
---|
1967 | }
|
---|
1968 | return;
|
---|
1969 | copyY:
|
---|
1970 | *xPtr = *yPtr;
|
---|
1971 |
|
---|
1972 | }
|
---|
1973 |
|
---|
1974 | static
|
---|
1975 | void floatX256Mul( struct floatX256 *xPtr, const struct floatX256 *yPtr )
|
---|
1976 | {
|
---|
1977 | struct uint256 sig;
|
---|
1978 | int bitNum;
|
---|
1979 |
|
---|
1980 | if ( xPtr->isNaN ) return;
|
---|
1981 | if ( yPtr->isNaN ) {
|
---|
1982 | xPtr->isNaN = true;
|
---|
1983 | xPtr->isInf = false;
|
---|
1984 | xPtr->isZero = false;
|
---|
1985 | xPtr->sign = yPtr->sign;
|
---|
1986 | return;
|
---|
1987 | }
|
---|
1988 | if ( yPtr->sign ) xPtr->sign = !xPtr->sign;
|
---|
1989 | if ( xPtr->isInf ) {
|
---|
1990 | if ( yPtr->isZero ) floatX256Invalid( xPtr );
|
---|
1991 | return;
|
---|
1992 | }
|
---|
1993 | if ( yPtr->isInf ) {
|
---|
1994 | if ( xPtr->isZero ) {
|
---|
1995 | floatX256Invalid( xPtr );
|
---|
1996 | return;
|
---|
1997 | }
|
---|
1998 | xPtr->isInf = true;
|
---|
1999 | return;
|
---|
2000 | }
|
---|
2001 | if ( xPtr->isZero || yPtr->isZero ) {
|
---|
2002 | if ( xPtr->sign ) {
|
---|
2003 | *xPtr = floatX256NegativeZero;
|
---|
2004 | } else {
|
---|
2005 | *xPtr = floatX256PositiveZero;
|
---|
2006 | }
|
---|
2007 | return;
|
---|
2008 | }
|
---|
2009 | xPtr->exp += yPtr->exp;
|
---|
2010 | sig.v192 = 0;
|
---|
2011 | sig.v128 = 0;
|
---|
2012 | sig.v64 = 0;
|
---|
2013 | sig.v0 = 0;
|
---|
2014 | for ( bitNum = 0; bitNum < 248; ++bitNum ) {
|
---|
2015 | shiftRight1Jam256M( &sig );
|
---|
2016 | if ( xPtr->sig.v0 & 1 ) add256M( &sig, &yPtr->sig );
|
---|
2017 | shiftRight1256M( &xPtr->sig );
|
---|
2018 | }
|
---|
2019 | if ( UINT64_C( 0x0100000000000000 ) <= sig.v192 ) {
|
---|
2020 | ++xPtr->exp;
|
---|
2021 | shiftRight1Jam256M( &sig );
|
---|
2022 | }
|
---|
2023 | xPtr->sig = sig;
|
---|
2024 |
|
---|
2025 | }
|
---|
2026 |
|
---|
2027 | #endif
|
---|
2028 |
|
---|
2029 | /*----------------------------------------------------------------------------
|
---|
2030 | *----------------------------------------------------------------------------*/
|
---|
2031 |
|
---|
2032 | #ifdef FLOAT16
|
---|
2033 |
|
---|
2034 | float16_t slow_ui32_to_f16( uint32_t a )
|
---|
2035 | {
|
---|
2036 | struct floatX x;
|
---|
2037 |
|
---|
2038 | ui32ToFloatX( a, &x );
|
---|
2039 | return floatXToF16( &x );
|
---|
2040 |
|
---|
2041 | }
|
---|
2042 |
|
---|
2043 | #endif
|
---|
2044 |
|
---|
2045 | float32_t slow_ui32_to_f32( uint32_t a )
|
---|
2046 | {
|
---|
2047 | struct floatX x;
|
---|
2048 |
|
---|
2049 | ui32ToFloatX( a, &x );
|
---|
2050 | return floatXToF32( &x );
|
---|
2051 |
|
---|
2052 | }
|
---|
2053 |
|
---|
2054 | #ifdef FLOAT64
|
---|
2055 |
|
---|
2056 | float64_t slow_ui32_to_f64( uint32_t a )
|
---|
2057 | {
|
---|
2058 | struct floatX x;
|
---|
2059 |
|
---|
2060 | ui32ToFloatX( a, &x );
|
---|
2061 | return floatXToF64( &x );
|
---|
2062 |
|
---|
2063 | }
|
---|
2064 |
|
---|
2065 | #endif
|
---|
2066 |
|
---|
2067 | #ifdef EXTFLOAT80
|
---|
2068 |
|
---|
2069 | void slow_ui32_to_extF80M( uint32_t a, extFloat80_t *zPtr )
|
---|
2070 | {
|
---|
2071 | struct floatX x;
|
---|
2072 |
|
---|
2073 | ui32ToFloatX( a, &x );
|
---|
2074 | floatXToExtF80M( &x, zPtr );
|
---|
2075 |
|
---|
2076 | }
|
---|
2077 |
|
---|
2078 | #endif
|
---|
2079 |
|
---|
2080 | #ifdef FLOAT128
|
---|
2081 |
|
---|
2082 | void slow_ui32_to_f128M( uint32_t a, float128_t *zPtr )
|
---|
2083 | {
|
---|
2084 | struct floatX x;
|
---|
2085 |
|
---|
2086 | ui32ToFloatX( a, &x );
|
---|
2087 | floatXToF128M( &x, zPtr );
|
---|
2088 |
|
---|
2089 | }
|
---|
2090 |
|
---|
2091 | #endif
|
---|
2092 |
|
---|
2093 | #ifdef FLOAT16
|
---|
2094 |
|
---|
2095 | float16_t slow_ui64_to_f16( uint64_t a )
|
---|
2096 | {
|
---|
2097 | struct floatX x;
|
---|
2098 |
|
---|
2099 | ui64ToFloatX( a, &x );
|
---|
2100 | return floatXToF16( &x );
|
---|
2101 |
|
---|
2102 | }
|
---|
2103 |
|
---|
2104 | #endif
|
---|
2105 |
|
---|
2106 | float32_t slow_ui64_to_f32( uint64_t a )
|
---|
2107 | {
|
---|
2108 | struct floatX x;
|
---|
2109 |
|
---|
2110 | ui64ToFloatX( a, &x );
|
---|
2111 | return floatXToF32( &x );
|
---|
2112 |
|
---|
2113 | }
|
---|
2114 |
|
---|
2115 | #ifdef FLOAT64
|
---|
2116 |
|
---|
2117 | float64_t slow_ui64_to_f64( uint64_t a )
|
---|
2118 | {
|
---|
2119 | struct floatX x;
|
---|
2120 |
|
---|
2121 | ui64ToFloatX( a, &x );
|
---|
2122 | return floatXToF64( &x );
|
---|
2123 |
|
---|
2124 | }
|
---|
2125 |
|
---|
2126 | #endif
|
---|
2127 |
|
---|
2128 | #ifdef EXTFLOAT80
|
---|
2129 |
|
---|
2130 | void slow_ui64_to_extF80M( uint64_t a, extFloat80_t *zPtr )
|
---|
2131 | {
|
---|
2132 | struct floatX x;
|
---|
2133 |
|
---|
2134 | ui64ToFloatX( a, &x );
|
---|
2135 | floatXToExtF80M( &x, zPtr );
|
---|
2136 |
|
---|
2137 | }
|
---|
2138 |
|
---|
2139 | #endif
|
---|
2140 |
|
---|
2141 | #ifdef FLOAT128
|
---|
2142 |
|
---|
2143 | void slow_ui64_to_f128M( uint64_t a, float128_t *zPtr )
|
---|
2144 | {
|
---|
2145 | struct floatX x;
|
---|
2146 |
|
---|
2147 | ui64ToFloatX( a, &x );
|
---|
2148 | floatXToF128M( &x, zPtr );
|
---|
2149 |
|
---|
2150 | }
|
---|
2151 |
|
---|
2152 | #endif
|
---|
2153 |
|
---|
2154 | #ifdef FLOAT16
|
---|
2155 |
|
---|
2156 | float16_t slow_i32_to_f16( int32_t a )
|
---|
2157 | {
|
---|
2158 | struct floatX x;
|
---|
2159 |
|
---|
2160 | i32ToFloatX( a, &x );
|
---|
2161 | return floatXToF16( &x );
|
---|
2162 |
|
---|
2163 | }
|
---|
2164 |
|
---|
2165 | #endif
|
---|
2166 |
|
---|
2167 | float32_t slow_i32_to_f32( int32_t a )
|
---|
2168 | {
|
---|
2169 | struct floatX x;
|
---|
2170 |
|
---|
2171 | i32ToFloatX( a, &x );
|
---|
2172 | return floatXToF32( &x );
|
---|
2173 |
|
---|
2174 | }
|
---|
2175 |
|
---|
2176 | #ifdef FLOAT64
|
---|
2177 |
|
---|
2178 | float64_t slow_i32_to_f64( int32_t a )
|
---|
2179 | {
|
---|
2180 | struct floatX x;
|
---|
2181 |
|
---|
2182 | i32ToFloatX( a, &x );
|
---|
2183 | return floatXToF64( &x );
|
---|
2184 |
|
---|
2185 | }
|
---|
2186 |
|
---|
2187 | #endif
|
---|
2188 |
|
---|
2189 | #ifdef EXTFLOAT80
|
---|
2190 |
|
---|
2191 | void slow_i32_to_extF80M( int32_t a, extFloat80_t *zPtr )
|
---|
2192 | {
|
---|
2193 | struct floatX x;
|
---|
2194 |
|
---|
2195 | i32ToFloatX( a, &x );
|
---|
2196 | floatXToExtF80M( &x, zPtr );
|
---|
2197 |
|
---|
2198 | }
|
---|
2199 |
|
---|
2200 | #endif
|
---|
2201 |
|
---|
2202 | #ifdef FLOAT128
|
---|
2203 |
|
---|
2204 | void slow_i32_to_f128M( int32_t a, float128_t *zPtr )
|
---|
2205 | {
|
---|
2206 | struct floatX x;
|
---|
2207 |
|
---|
2208 | i32ToFloatX( a, &x );
|
---|
2209 | floatXToF128M( &x, zPtr );
|
---|
2210 |
|
---|
2211 | }
|
---|
2212 |
|
---|
2213 | #endif
|
---|
2214 |
|
---|
2215 | #ifdef FLOAT16
|
---|
2216 |
|
---|
2217 | float16_t slow_i64_to_f16( int64_t a )
|
---|
2218 | {
|
---|
2219 | struct floatX x;
|
---|
2220 |
|
---|
2221 | i64ToFloatX( a, &x );
|
---|
2222 | return floatXToF16( &x );
|
---|
2223 |
|
---|
2224 | }
|
---|
2225 |
|
---|
2226 | #endif
|
---|
2227 |
|
---|
2228 | float32_t slow_i64_to_f32( int64_t a )
|
---|
2229 | {
|
---|
2230 | struct floatX x;
|
---|
2231 |
|
---|
2232 | i64ToFloatX( a, &x );
|
---|
2233 | return floatXToF32( &x );
|
---|
2234 |
|
---|
2235 | }
|
---|
2236 |
|
---|
2237 | #ifdef FLOAT64
|
---|
2238 |
|
---|
2239 | float64_t slow_i64_to_f64( int64_t a )
|
---|
2240 | {
|
---|
2241 | struct floatX x;
|
---|
2242 |
|
---|
2243 | i64ToFloatX( a, &x );
|
---|
2244 | return floatXToF64( &x );
|
---|
2245 |
|
---|
2246 | }
|
---|
2247 |
|
---|
2248 | #endif
|
---|
2249 |
|
---|
2250 | #ifdef EXTFLOAT80
|
---|
2251 |
|
---|
2252 | void slow_i64_to_extF80M( int64_t a, extFloat80_t *zPtr )
|
---|
2253 | {
|
---|
2254 | struct floatX x;
|
---|
2255 |
|
---|
2256 | i64ToFloatX( a, &x );
|
---|
2257 | floatXToExtF80M( &x, zPtr );
|
---|
2258 |
|
---|
2259 | }
|
---|
2260 |
|
---|
2261 | #endif
|
---|
2262 |
|
---|
2263 | #ifdef FLOAT128
|
---|
2264 |
|
---|
2265 | void slow_i64_to_f128M( int64_t a, float128_t *zPtr )
|
---|
2266 | {
|
---|
2267 | struct floatX x;
|
---|
2268 |
|
---|
2269 | i64ToFloatX( a, &x );
|
---|
2270 | floatXToF128M( &x, zPtr );
|
---|
2271 |
|
---|
2272 | }
|
---|
2273 |
|
---|
2274 | #endif
|
---|
2275 |
|
---|
2276 | #ifdef FLOAT16
|
---|
2277 |
|
---|
2278 | uint_fast32_t
|
---|
2279 | slow_f16_to_ui32( float16_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2280 | {
|
---|
2281 | struct floatX x;
|
---|
2282 |
|
---|
2283 | f16ToFloatX( a, &x );
|
---|
2284 | return floatXToUI32( &x, roundingMode, exact );
|
---|
2285 |
|
---|
2286 | }
|
---|
2287 |
|
---|
2288 | uint_fast64_t
|
---|
2289 | slow_f16_to_ui64( float16_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2290 | {
|
---|
2291 | struct floatX x;
|
---|
2292 |
|
---|
2293 | f16ToFloatX( a, &x );
|
---|
2294 | return floatXToUI64( &x, roundingMode, exact );
|
---|
2295 |
|
---|
2296 | }
|
---|
2297 |
|
---|
2298 | int_fast32_t
|
---|
2299 | slow_f16_to_i32( float16_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2300 | {
|
---|
2301 | struct floatX x;
|
---|
2302 |
|
---|
2303 | f16ToFloatX( a, &x );
|
---|
2304 | return floatXToI32( &x, roundingMode, exact );
|
---|
2305 |
|
---|
2306 | }
|
---|
2307 |
|
---|
2308 | int_fast64_t
|
---|
2309 | slow_f16_to_i64( float16_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2310 | {
|
---|
2311 | struct floatX x;
|
---|
2312 |
|
---|
2313 | f16ToFloatX( a, &x );
|
---|
2314 | return floatXToI64( &x, roundingMode, exact );
|
---|
2315 |
|
---|
2316 | }
|
---|
2317 |
|
---|
2318 | uint_fast32_t slow_f16_to_ui32_r_minMag( float16_t a, bool exact )
|
---|
2319 | {
|
---|
2320 | struct floatX x;
|
---|
2321 |
|
---|
2322 | f16ToFloatX( a, &x );
|
---|
2323 | return floatXToUI32( &x, softfloat_round_minMag, exact );
|
---|
2324 |
|
---|
2325 | }
|
---|
2326 |
|
---|
2327 | uint_fast64_t slow_f16_to_ui64_r_minMag( float16_t a, bool exact )
|
---|
2328 | {
|
---|
2329 | struct floatX x;
|
---|
2330 |
|
---|
2331 | f16ToFloatX( a, &x );
|
---|
2332 | return floatXToUI64( &x, softfloat_round_minMag, exact );
|
---|
2333 |
|
---|
2334 | }
|
---|
2335 |
|
---|
2336 | int_fast32_t slow_f16_to_i32_r_minMag( float16_t a, bool exact )
|
---|
2337 | {
|
---|
2338 | struct floatX x;
|
---|
2339 |
|
---|
2340 | f16ToFloatX( a, &x );
|
---|
2341 | return floatXToI32( &x, softfloat_round_minMag, exact );
|
---|
2342 |
|
---|
2343 | }
|
---|
2344 |
|
---|
2345 | int_fast64_t slow_f16_to_i64_r_minMag( float16_t a, bool exact )
|
---|
2346 | {
|
---|
2347 | struct floatX x;
|
---|
2348 |
|
---|
2349 | f16ToFloatX( a, &x );
|
---|
2350 | return floatXToI64( &x, softfloat_round_minMag, exact );
|
---|
2351 |
|
---|
2352 | }
|
---|
2353 |
|
---|
2354 | float32_t slow_f16_to_f32( float16_t a )
|
---|
2355 | {
|
---|
2356 | struct floatX x;
|
---|
2357 |
|
---|
2358 | f16ToFloatX( a, &x );
|
---|
2359 | return floatXToF32( &x );
|
---|
2360 |
|
---|
2361 | }
|
---|
2362 |
|
---|
2363 | #ifdef FLOAT64
|
---|
2364 |
|
---|
2365 | float64_t slow_f16_to_f64( float16_t a )
|
---|
2366 | {
|
---|
2367 | struct floatX x;
|
---|
2368 |
|
---|
2369 | f16ToFloatX( a, &x );
|
---|
2370 | return floatXToF64( &x );
|
---|
2371 |
|
---|
2372 | }
|
---|
2373 |
|
---|
2374 | #endif
|
---|
2375 |
|
---|
2376 | #ifdef EXTFLOAT80
|
---|
2377 |
|
---|
2378 | void slow_f16_to_extF80M( float16_t a, extFloat80_t *zPtr )
|
---|
2379 | {
|
---|
2380 | struct floatX x;
|
---|
2381 |
|
---|
2382 | f16ToFloatX( a, &x );
|
---|
2383 | floatXToExtF80M( &x, zPtr );
|
---|
2384 |
|
---|
2385 | }
|
---|
2386 |
|
---|
2387 | #endif
|
---|
2388 |
|
---|
2389 | #ifdef FLOAT128
|
---|
2390 |
|
---|
2391 | void slow_f16_to_f128M( float16_t a, float128_t *zPtr )
|
---|
2392 | {
|
---|
2393 | struct floatX x;
|
---|
2394 |
|
---|
2395 | f16ToFloatX( a, &x );
|
---|
2396 | floatXToF128M( &x, zPtr );
|
---|
2397 |
|
---|
2398 | }
|
---|
2399 |
|
---|
2400 | #endif
|
---|
2401 |
|
---|
2402 | float16_t
|
---|
2403 | slow_f16_roundToInt( float16_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2404 | {
|
---|
2405 | struct floatX x;
|
---|
2406 |
|
---|
2407 | f16ToFloatX( a, &x );
|
---|
2408 | floatXRoundToInt( &x, roundingMode, exact );
|
---|
2409 | return floatXToF16( &x );
|
---|
2410 |
|
---|
2411 | }
|
---|
2412 |
|
---|
2413 | float16_t slow_f16_add( float16_t a, float16_t b )
|
---|
2414 | {
|
---|
2415 | struct floatX x, y;
|
---|
2416 |
|
---|
2417 | f16ToFloatX( a, &x );
|
---|
2418 | f16ToFloatX( b, &y );
|
---|
2419 | floatXAdd( &x, &y );
|
---|
2420 | return floatXToF16( &x );
|
---|
2421 |
|
---|
2422 | }
|
---|
2423 |
|
---|
2424 | float16_t slow_f16_sub( float16_t a, float16_t b )
|
---|
2425 | {
|
---|
2426 | struct floatX x, y;
|
---|
2427 |
|
---|
2428 | f16ToFloatX( a, &x );
|
---|
2429 | f16ToFloatX( b, &y );
|
---|
2430 | y.sign = !y.sign;
|
---|
2431 | floatXAdd( &x, &y );
|
---|
2432 | return floatXToF16( &x );
|
---|
2433 |
|
---|
2434 |
|
---|
2435 | }
|
---|
2436 |
|
---|
2437 | float16_t slow_f16_mul( float16_t a, float16_t b )
|
---|
2438 | {
|
---|
2439 | struct floatX x, y;
|
---|
2440 |
|
---|
2441 | f16ToFloatX( a, &x );
|
---|
2442 | f16ToFloatX( b, &y );
|
---|
2443 | floatXMul( &x, &y );
|
---|
2444 | return floatXToF16( &x );
|
---|
2445 |
|
---|
2446 | }
|
---|
2447 |
|
---|
2448 | float16_t slow_f16_mulAdd( float16_t a, float16_t b, float16_t c )
|
---|
2449 | {
|
---|
2450 | struct floatX x, y;
|
---|
2451 |
|
---|
2452 | f16ToFloatX( a, &x );
|
---|
2453 | f16ToFloatX( b, &y );
|
---|
2454 | floatXMul( &x, &y );
|
---|
2455 | f16ToFloatX( c, &y );
|
---|
2456 | floatXAdd( &x, &y );
|
---|
2457 | return floatXToF16( &x );
|
---|
2458 |
|
---|
2459 | }
|
---|
2460 |
|
---|
2461 | float16_t slow_f16_div( float16_t a, float16_t b )
|
---|
2462 | {
|
---|
2463 | struct floatX x, y;
|
---|
2464 |
|
---|
2465 | f16ToFloatX( a, &x );
|
---|
2466 | f16ToFloatX( b, &y );
|
---|
2467 | floatXDiv( &x, &y );
|
---|
2468 | return floatXToF16( &x );
|
---|
2469 |
|
---|
2470 | }
|
---|
2471 |
|
---|
2472 | float16_t slow_f16_rem( float16_t a, float16_t b )
|
---|
2473 | {
|
---|
2474 | struct floatX x, y;
|
---|
2475 |
|
---|
2476 | f16ToFloatX( a, &x );
|
---|
2477 | f16ToFloatX( b, &y );
|
---|
2478 | floatXRem( &x, &y );
|
---|
2479 | return floatXToF16( &x );
|
---|
2480 |
|
---|
2481 | }
|
---|
2482 |
|
---|
2483 | float16_t slow_f16_sqrt( float16_t a )
|
---|
2484 | {
|
---|
2485 | struct floatX x;
|
---|
2486 |
|
---|
2487 | f16ToFloatX( a, &x );
|
---|
2488 | floatXSqrt( &x );
|
---|
2489 | return floatXToF16( &x );
|
---|
2490 |
|
---|
2491 | }
|
---|
2492 |
|
---|
2493 | bool slow_f16_eq( float16_t a, float16_t b )
|
---|
2494 | {
|
---|
2495 | struct floatX x, y;
|
---|
2496 |
|
---|
2497 | f16ToFloatX( a, &x );
|
---|
2498 | f16ToFloatX( b, &y );
|
---|
2499 | return floatXEq( &x, &y );
|
---|
2500 |
|
---|
2501 | }
|
---|
2502 |
|
---|
2503 | bool slow_f16_le( float16_t a, float16_t b )
|
---|
2504 | {
|
---|
2505 | struct floatX x, y;
|
---|
2506 |
|
---|
2507 | f16ToFloatX( a, &x );
|
---|
2508 | f16ToFloatX( b, &y );
|
---|
2509 | if ( x.isNaN || y.isNaN ) {
|
---|
2510 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
2511 | }
|
---|
2512 | return floatXLe( &x, &y );
|
---|
2513 |
|
---|
2514 | }
|
---|
2515 |
|
---|
2516 | bool slow_f16_lt( float16_t a, float16_t b )
|
---|
2517 | {
|
---|
2518 | struct floatX x, y;
|
---|
2519 |
|
---|
2520 | f16ToFloatX( a, &x );
|
---|
2521 | f16ToFloatX( b, &y );
|
---|
2522 | if ( x.isNaN || y.isNaN ) {
|
---|
2523 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
2524 | }
|
---|
2525 | return floatXLt( &x, &y );
|
---|
2526 |
|
---|
2527 | }
|
---|
2528 |
|
---|
2529 | bool slow_f16_eq_signaling( float16_t a, float16_t b )
|
---|
2530 | {
|
---|
2531 | struct floatX x, y;
|
---|
2532 |
|
---|
2533 | f16ToFloatX( a, &x );
|
---|
2534 | f16ToFloatX( b, &y );
|
---|
2535 | if ( x.isNaN || y.isNaN ) {
|
---|
2536 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
2537 | }
|
---|
2538 | return floatXEq( &x, &y );
|
---|
2539 |
|
---|
2540 | }
|
---|
2541 |
|
---|
2542 | bool slow_f16_le_quiet( float16_t a, float16_t b )
|
---|
2543 | {
|
---|
2544 | struct floatX x, y;
|
---|
2545 |
|
---|
2546 | f16ToFloatX( a, &x );
|
---|
2547 | f16ToFloatX( b, &y );
|
---|
2548 | return floatXLe( &x, &y );
|
---|
2549 |
|
---|
2550 | }
|
---|
2551 |
|
---|
2552 | bool slow_f16_lt_quiet( float16_t a, float16_t b )
|
---|
2553 | {
|
---|
2554 | struct floatX x, y;
|
---|
2555 |
|
---|
2556 | f16ToFloatX( a, &x );
|
---|
2557 | f16ToFloatX( b, &y );
|
---|
2558 | return floatXLt( &x, &y );
|
---|
2559 |
|
---|
2560 | }
|
---|
2561 |
|
---|
2562 | #endif
|
---|
2563 |
|
---|
2564 | uint_fast32_t
|
---|
2565 | slow_f32_to_ui32( float32_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2566 | {
|
---|
2567 | struct floatX x;
|
---|
2568 |
|
---|
2569 | f32ToFloatX( a, &x );
|
---|
2570 | return floatXToUI32( &x, roundingMode, exact );
|
---|
2571 |
|
---|
2572 | }
|
---|
2573 |
|
---|
2574 | uint_fast64_t
|
---|
2575 | slow_f32_to_ui64( float32_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2576 | {
|
---|
2577 | struct floatX x;
|
---|
2578 |
|
---|
2579 | f32ToFloatX( a, &x );
|
---|
2580 | return floatXToUI64( &x, roundingMode, exact );
|
---|
2581 |
|
---|
2582 | }
|
---|
2583 |
|
---|
2584 | int_fast32_t
|
---|
2585 | slow_f32_to_i32( float32_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2586 | {
|
---|
2587 | struct floatX x;
|
---|
2588 |
|
---|
2589 | f32ToFloatX( a, &x );
|
---|
2590 | return floatXToI32( &x, roundingMode, exact );
|
---|
2591 |
|
---|
2592 | }
|
---|
2593 |
|
---|
2594 | int_fast64_t
|
---|
2595 | slow_f32_to_i64( float32_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2596 | {
|
---|
2597 | struct floatX x;
|
---|
2598 |
|
---|
2599 | f32ToFloatX( a, &x );
|
---|
2600 | return floatXToI64( &x, roundingMode, exact );
|
---|
2601 |
|
---|
2602 | }
|
---|
2603 |
|
---|
2604 | uint_fast32_t slow_f32_to_ui32_r_minMag( float32_t a, bool exact )
|
---|
2605 | {
|
---|
2606 | struct floatX x;
|
---|
2607 |
|
---|
2608 | f32ToFloatX( a, &x );
|
---|
2609 | return floatXToUI32( &x, softfloat_round_minMag, exact );
|
---|
2610 |
|
---|
2611 | }
|
---|
2612 |
|
---|
2613 | uint_fast64_t slow_f32_to_ui64_r_minMag( float32_t a, bool exact )
|
---|
2614 | {
|
---|
2615 | struct floatX x;
|
---|
2616 |
|
---|
2617 | f32ToFloatX( a, &x );
|
---|
2618 | return floatXToUI64( &x, softfloat_round_minMag, exact );
|
---|
2619 |
|
---|
2620 | }
|
---|
2621 |
|
---|
2622 | int_fast32_t slow_f32_to_i32_r_minMag( float32_t a, bool exact )
|
---|
2623 | {
|
---|
2624 | struct floatX x;
|
---|
2625 |
|
---|
2626 | f32ToFloatX( a, &x );
|
---|
2627 | return floatXToI32( &x, softfloat_round_minMag, exact );
|
---|
2628 |
|
---|
2629 | }
|
---|
2630 |
|
---|
2631 | int_fast64_t slow_f32_to_i64_r_minMag( float32_t a, bool exact )
|
---|
2632 | {
|
---|
2633 | struct floatX x;
|
---|
2634 |
|
---|
2635 | f32ToFloatX( a, &x );
|
---|
2636 | return floatXToI64( &x, softfloat_round_minMag, exact );
|
---|
2637 |
|
---|
2638 | }
|
---|
2639 |
|
---|
2640 | #ifdef FLOAT16
|
---|
2641 |
|
---|
2642 | float16_t slow_f32_to_f16( float32_t a )
|
---|
2643 | {
|
---|
2644 | struct floatX x;
|
---|
2645 |
|
---|
2646 | f32ToFloatX( a, &x );
|
---|
2647 | return floatXToF16( &x );
|
---|
2648 |
|
---|
2649 | }
|
---|
2650 |
|
---|
2651 | #endif
|
---|
2652 |
|
---|
2653 | #ifdef FLOAT64
|
---|
2654 |
|
---|
2655 | float64_t slow_f32_to_f64( float32_t a )
|
---|
2656 | {
|
---|
2657 | struct floatX x;
|
---|
2658 |
|
---|
2659 | f32ToFloatX( a, &x );
|
---|
2660 | return floatXToF64( &x );
|
---|
2661 |
|
---|
2662 | }
|
---|
2663 |
|
---|
2664 | #endif
|
---|
2665 |
|
---|
2666 | #ifdef EXTFLOAT80
|
---|
2667 |
|
---|
2668 | void slow_f32_to_extF80M( float32_t a, extFloat80_t *zPtr )
|
---|
2669 | {
|
---|
2670 | struct floatX x;
|
---|
2671 |
|
---|
2672 | f32ToFloatX( a, &x );
|
---|
2673 | floatXToExtF80M( &x, zPtr );
|
---|
2674 |
|
---|
2675 | }
|
---|
2676 |
|
---|
2677 | #endif
|
---|
2678 |
|
---|
2679 | #ifdef FLOAT128
|
---|
2680 |
|
---|
2681 | void slow_f32_to_f128M( float32_t a, float128_t *zPtr )
|
---|
2682 | {
|
---|
2683 | struct floatX x;
|
---|
2684 |
|
---|
2685 | f32ToFloatX( a, &x );
|
---|
2686 | floatXToF128M( &x, zPtr );
|
---|
2687 |
|
---|
2688 | }
|
---|
2689 |
|
---|
2690 | #endif
|
---|
2691 |
|
---|
2692 | float32_t
|
---|
2693 | slow_f32_roundToInt( float32_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2694 | {
|
---|
2695 | struct floatX x;
|
---|
2696 |
|
---|
2697 | f32ToFloatX( a, &x );
|
---|
2698 | floatXRoundToInt( &x, roundingMode, exact );
|
---|
2699 | return floatXToF32( &x );
|
---|
2700 |
|
---|
2701 | }
|
---|
2702 |
|
---|
2703 | float32_t slow_f32_add( float32_t a, float32_t b )
|
---|
2704 | {
|
---|
2705 | struct floatX x, y;
|
---|
2706 |
|
---|
2707 | f32ToFloatX( a, &x );
|
---|
2708 | f32ToFloatX( b, &y );
|
---|
2709 | floatXAdd( &x, &y );
|
---|
2710 | return floatXToF32( &x );
|
---|
2711 |
|
---|
2712 | }
|
---|
2713 |
|
---|
2714 | float32_t slow_f32_sub( float32_t a, float32_t b )
|
---|
2715 | {
|
---|
2716 | struct floatX x, y;
|
---|
2717 |
|
---|
2718 | f32ToFloatX( a, &x );
|
---|
2719 | f32ToFloatX( b, &y );
|
---|
2720 | y.sign = !y.sign;
|
---|
2721 | floatXAdd( &x, &y );
|
---|
2722 | return floatXToF32( &x );
|
---|
2723 |
|
---|
2724 |
|
---|
2725 | }
|
---|
2726 |
|
---|
2727 | float32_t slow_f32_mul( float32_t a, float32_t b )
|
---|
2728 | {
|
---|
2729 | struct floatX x, y;
|
---|
2730 |
|
---|
2731 | f32ToFloatX( a, &x );
|
---|
2732 | f32ToFloatX( b, &y );
|
---|
2733 | floatXMul( &x, &y );
|
---|
2734 | return floatXToF32( &x );
|
---|
2735 |
|
---|
2736 | }
|
---|
2737 |
|
---|
2738 | float32_t slow_f32_mulAdd( float32_t a, float32_t b, float32_t c )
|
---|
2739 | {
|
---|
2740 | struct floatX x, y;
|
---|
2741 |
|
---|
2742 | f32ToFloatX( a, &x );
|
---|
2743 | f32ToFloatX( b, &y );
|
---|
2744 | floatXMul( &x, &y );
|
---|
2745 | f32ToFloatX( c, &y );
|
---|
2746 | floatXAdd( &x, &y );
|
---|
2747 | return floatXToF32( &x );
|
---|
2748 |
|
---|
2749 | }
|
---|
2750 |
|
---|
2751 | float32_t slow_f32_div( float32_t a, float32_t b )
|
---|
2752 | {
|
---|
2753 | struct floatX x, y;
|
---|
2754 |
|
---|
2755 | f32ToFloatX( a, &x );
|
---|
2756 | f32ToFloatX( b, &y );
|
---|
2757 | floatXDiv( &x, &y );
|
---|
2758 | return floatXToF32( &x );
|
---|
2759 |
|
---|
2760 | }
|
---|
2761 |
|
---|
2762 | float32_t slow_f32_rem( float32_t a, float32_t b )
|
---|
2763 | {
|
---|
2764 | struct floatX x, y;
|
---|
2765 |
|
---|
2766 | f32ToFloatX( a, &x );
|
---|
2767 | f32ToFloatX( b, &y );
|
---|
2768 | floatXRem( &x, &y );
|
---|
2769 | return floatXToF32( &x );
|
---|
2770 |
|
---|
2771 | }
|
---|
2772 |
|
---|
2773 | float32_t slow_f32_sqrt( float32_t a )
|
---|
2774 | {
|
---|
2775 | struct floatX x;
|
---|
2776 |
|
---|
2777 | f32ToFloatX( a, &x );
|
---|
2778 | floatXSqrt( &x );
|
---|
2779 | return floatXToF32( &x );
|
---|
2780 |
|
---|
2781 | }
|
---|
2782 |
|
---|
2783 | bool slow_f32_eq( float32_t a, float32_t b )
|
---|
2784 | {
|
---|
2785 | struct floatX x, y;
|
---|
2786 |
|
---|
2787 | f32ToFloatX( a, &x );
|
---|
2788 | f32ToFloatX( b, &y );
|
---|
2789 | return floatXEq( &x, &y );
|
---|
2790 |
|
---|
2791 | }
|
---|
2792 |
|
---|
2793 | bool slow_f32_le( float32_t a, float32_t b )
|
---|
2794 | {
|
---|
2795 | struct floatX x, y;
|
---|
2796 |
|
---|
2797 | f32ToFloatX( a, &x );
|
---|
2798 | f32ToFloatX( b, &y );
|
---|
2799 | if ( x.isNaN || y.isNaN ) {
|
---|
2800 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
2801 | }
|
---|
2802 | return floatXLe( &x, &y );
|
---|
2803 |
|
---|
2804 | }
|
---|
2805 |
|
---|
2806 | bool slow_f32_lt( float32_t a, float32_t b )
|
---|
2807 | {
|
---|
2808 | struct floatX x, y;
|
---|
2809 |
|
---|
2810 | f32ToFloatX( a, &x );
|
---|
2811 | f32ToFloatX( b, &y );
|
---|
2812 | if ( x.isNaN || y.isNaN ) {
|
---|
2813 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
2814 | }
|
---|
2815 | return floatXLt( &x, &y );
|
---|
2816 |
|
---|
2817 | }
|
---|
2818 |
|
---|
2819 | bool slow_f32_eq_signaling( float32_t a, float32_t b )
|
---|
2820 | {
|
---|
2821 | struct floatX x, y;
|
---|
2822 |
|
---|
2823 | f32ToFloatX( a, &x );
|
---|
2824 | f32ToFloatX( b, &y );
|
---|
2825 | if ( x.isNaN || y.isNaN ) {
|
---|
2826 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
2827 | }
|
---|
2828 | return floatXEq( &x, &y );
|
---|
2829 |
|
---|
2830 | }
|
---|
2831 |
|
---|
2832 | bool slow_f32_le_quiet( float32_t a, float32_t b )
|
---|
2833 | {
|
---|
2834 | struct floatX x, y;
|
---|
2835 |
|
---|
2836 | f32ToFloatX( a, &x );
|
---|
2837 | f32ToFloatX( b, &y );
|
---|
2838 | return floatXLe( &x, &y );
|
---|
2839 |
|
---|
2840 | }
|
---|
2841 |
|
---|
2842 | bool slow_f32_lt_quiet( float32_t a, float32_t b )
|
---|
2843 | {
|
---|
2844 | struct floatX x, y;
|
---|
2845 |
|
---|
2846 | f32ToFloatX( a, &x );
|
---|
2847 | f32ToFloatX( b, &y );
|
---|
2848 | return floatXLt( &x, &y );
|
---|
2849 |
|
---|
2850 | }
|
---|
2851 |
|
---|
2852 | #ifdef FLOAT64
|
---|
2853 |
|
---|
2854 | uint_fast32_t
|
---|
2855 | slow_f64_to_ui32( float64_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2856 | {
|
---|
2857 | struct floatX x;
|
---|
2858 |
|
---|
2859 | f64ToFloatX( a, &x );
|
---|
2860 | return floatXToUI32( &x, roundingMode, exact );
|
---|
2861 |
|
---|
2862 | }
|
---|
2863 |
|
---|
2864 | uint_fast64_t
|
---|
2865 | slow_f64_to_ui64( float64_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2866 | {
|
---|
2867 | struct floatX x;
|
---|
2868 |
|
---|
2869 | f64ToFloatX( a, &x );
|
---|
2870 | return floatXToUI64( &x, roundingMode, exact );
|
---|
2871 |
|
---|
2872 | }
|
---|
2873 |
|
---|
2874 | int_fast32_t
|
---|
2875 | slow_f64_to_i32( float64_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2876 | {
|
---|
2877 | struct floatX x;
|
---|
2878 |
|
---|
2879 | f64ToFloatX( a, &x );
|
---|
2880 | return floatXToI32( &x, roundingMode, exact );
|
---|
2881 |
|
---|
2882 | }
|
---|
2883 |
|
---|
2884 | int_fast64_t
|
---|
2885 | slow_f64_to_i64( float64_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2886 | {
|
---|
2887 | struct floatX x;
|
---|
2888 |
|
---|
2889 | f64ToFloatX( a, &x );
|
---|
2890 | return floatXToI64( &x, roundingMode, exact );
|
---|
2891 |
|
---|
2892 | }
|
---|
2893 |
|
---|
2894 | uint_fast32_t slow_f64_to_ui32_r_minMag( float64_t a, bool exact )
|
---|
2895 | {
|
---|
2896 | struct floatX x;
|
---|
2897 |
|
---|
2898 | f64ToFloatX( a, &x );
|
---|
2899 | return floatXToUI32( &x, softfloat_round_minMag, exact );
|
---|
2900 |
|
---|
2901 | }
|
---|
2902 |
|
---|
2903 | uint_fast64_t slow_f64_to_ui64_r_minMag( float64_t a, bool exact )
|
---|
2904 | {
|
---|
2905 | struct floatX x;
|
---|
2906 |
|
---|
2907 | f64ToFloatX( a, &x );
|
---|
2908 | return floatXToUI64( &x, softfloat_round_minMag, exact );
|
---|
2909 |
|
---|
2910 | }
|
---|
2911 |
|
---|
2912 | int_fast32_t slow_f64_to_i32_r_minMag( float64_t a, bool exact )
|
---|
2913 | {
|
---|
2914 | struct floatX x;
|
---|
2915 |
|
---|
2916 | f64ToFloatX( a, &x );
|
---|
2917 | return floatXToI32( &x, softfloat_round_minMag, exact );
|
---|
2918 |
|
---|
2919 | }
|
---|
2920 |
|
---|
2921 | int_fast64_t slow_f64_to_i64_r_minMag( float64_t a, bool exact )
|
---|
2922 | {
|
---|
2923 | struct floatX x;
|
---|
2924 |
|
---|
2925 | f64ToFloatX( a, &x );
|
---|
2926 | return floatXToI64( &x, softfloat_round_minMag, exact );
|
---|
2927 |
|
---|
2928 | }
|
---|
2929 |
|
---|
2930 | #ifdef FLOAT16
|
---|
2931 |
|
---|
2932 | float16_t slow_f64_to_f16( float64_t a )
|
---|
2933 | {
|
---|
2934 | struct floatX x;
|
---|
2935 |
|
---|
2936 | f64ToFloatX( a, &x );
|
---|
2937 | return floatXToF16( &x );
|
---|
2938 |
|
---|
2939 | }
|
---|
2940 |
|
---|
2941 | #endif
|
---|
2942 |
|
---|
2943 | float32_t slow_f64_to_f32( float64_t a )
|
---|
2944 | {
|
---|
2945 | struct floatX x;
|
---|
2946 |
|
---|
2947 | f64ToFloatX( a, &x );
|
---|
2948 | return floatXToF32( &x );
|
---|
2949 |
|
---|
2950 | }
|
---|
2951 |
|
---|
2952 | #ifdef EXTFLOAT80
|
---|
2953 |
|
---|
2954 | void slow_f64_to_extF80M( float64_t a, extFloat80_t *zPtr )
|
---|
2955 | {
|
---|
2956 | struct floatX x;
|
---|
2957 |
|
---|
2958 | f64ToFloatX( a, &x );
|
---|
2959 | floatXToExtF80M( &x, zPtr );
|
---|
2960 |
|
---|
2961 | }
|
---|
2962 |
|
---|
2963 | #endif
|
---|
2964 |
|
---|
2965 | #ifdef FLOAT128
|
---|
2966 |
|
---|
2967 | void slow_f64_to_f128M( float64_t a, float128_t *zPtr )
|
---|
2968 | {
|
---|
2969 | struct floatX x;
|
---|
2970 |
|
---|
2971 | f64ToFloatX( a, &x );
|
---|
2972 | floatXToF128M( &x, zPtr );
|
---|
2973 |
|
---|
2974 | }
|
---|
2975 |
|
---|
2976 | #endif
|
---|
2977 |
|
---|
2978 | float64_t
|
---|
2979 | slow_f64_roundToInt( float64_t a, uint_fast8_t roundingMode, bool exact )
|
---|
2980 | {
|
---|
2981 | struct floatX x;
|
---|
2982 |
|
---|
2983 | f64ToFloatX( a, &x );
|
---|
2984 | floatXRoundToInt( &x, roundingMode, exact );
|
---|
2985 | return floatXToF64( &x );
|
---|
2986 |
|
---|
2987 | }
|
---|
2988 |
|
---|
2989 | float64_t slow_f64_add( float64_t a, float64_t b )
|
---|
2990 | {
|
---|
2991 | struct floatX x, y;
|
---|
2992 |
|
---|
2993 | f64ToFloatX( a, &x );
|
---|
2994 | f64ToFloatX( b, &y );
|
---|
2995 | floatXAdd( &x, &y );
|
---|
2996 | return floatXToF64( &x );
|
---|
2997 |
|
---|
2998 | }
|
---|
2999 |
|
---|
3000 | float64_t slow_f64_sub( float64_t a, float64_t b )
|
---|
3001 | {
|
---|
3002 | struct floatX x, y;
|
---|
3003 |
|
---|
3004 | f64ToFloatX( a, &x );
|
---|
3005 | f64ToFloatX( b, &y );
|
---|
3006 | y.sign = !y.sign;
|
---|
3007 | floatXAdd( &x, &y );
|
---|
3008 | return floatXToF64( &x );
|
---|
3009 |
|
---|
3010 | }
|
---|
3011 |
|
---|
3012 | float64_t slow_f64_mul( float64_t a, float64_t b )
|
---|
3013 | {
|
---|
3014 | struct floatX x, y;
|
---|
3015 |
|
---|
3016 | f64ToFloatX( a, &x );
|
---|
3017 | f64ToFloatX( b, &y );
|
---|
3018 | floatXMul( &x, &y );
|
---|
3019 | return floatXToF64( &x );
|
---|
3020 |
|
---|
3021 | }
|
---|
3022 |
|
---|
3023 | float64_t slow_f64_mulAdd( float64_t a, float64_t b, float64_t c )
|
---|
3024 | {
|
---|
3025 | struct floatX x, y;
|
---|
3026 |
|
---|
3027 | f64ToFloatX( a, &x );
|
---|
3028 | f64ToFloatX( b, &y );
|
---|
3029 | floatXMul( &x, &y );
|
---|
3030 | f64ToFloatX( c, &y );
|
---|
3031 | floatXAdd( &x, &y );
|
---|
3032 | return floatXToF64( &x );
|
---|
3033 |
|
---|
3034 | }
|
---|
3035 |
|
---|
3036 | float64_t slow_f64_div( float64_t a, float64_t b )
|
---|
3037 | {
|
---|
3038 | struct floatX x, y;
|
---|
3039 |
|
---|
3040 | f64ToFloatX( a, &x );
|
---|
3041 | f64ToFloatX( b, &y );
|
---|
3042 | floatXDiv( &x, &y );
|
---|
3043 | return floatXToF64( &x );
|
---|
3044 |
|
---|
3045 | }
|
---|
3046 |
|
---|
3047 | float64_t slow_f64_rem( float64_t a, float64_t b )
|
---|
3048 | {
|
---|
3049 | struct floatX x, y;
|
---|
3050 |
|
---|
3051 | f64ToFloatX( a, &x );
|
---|
3052 | f64ToFloatX( b, &y );
|
---|
3053 | floatXRem( &x, &y );
|
---|
3054 | return floatXToF64( &x );
|
---|
3055 |
|
---|
3056 | }
|
---|
3057 |
|
---|
3058 | float64_t slow_f64_sqrt( float64_t a )
|
---|
3059 | {
|
---|
3060 | struct floatX x;
|
---|
3061 |
|
---|
3062 | f64ToFloatX( a, &x );
|
---|
3063 | floatXSqrt( &x );
|
---|
3064 | return floatXToF64( &x );
|
---|
3065 |
|
---|
3066 | }
|
---|
3067 |
|
---|
3068 | bool slow_f64_eq( float64_t a, float64_t b )
|
---|
3069 | {
|
---|
3070 | struct floatX x, y;
|
---|
3071 |
|
---|
3072 | f64ToFloatX( a, &x );
|
---|
3073 | f64ToFloatX( b, &y );
|
---|
3074 | return floatXEq( &x, &y );
|
---|
3075 |
|
---|
3076 | }
|
---|
3077 |
|
---|
3078 | bool slow_f64_le( float64_t a, float64_t b )
|
---|
3079 | {
|
---|
3080 | struct floatX x, y;
|
---|
3081 |
|
---|
3082 | f64ToFloatX( a, &x );
|
---|
3083 | f64ToFloatX( b, &y );
|
---|
3084 | if ( x.isNaN || y.isNaN ) {
|
---|
3085 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3086 | }
|
---|
3087 | return floatXLe( &x, &y );
|
---|
3088 |
|
---|
3089 | }
|
---|
3090 |
|
---|
3091 | bool slow_f64_lt( float64_t a, float64_t b )
|
---|
3092 | {
|
---|
3093 | struct floatX x, y;
|
---|
3094 |
|
---|
3095 | f64ToFloatX( a, &x );
|
---|
3096 | f64ToFloatX( b, &y );
|
---|
3097 | if ( x.isNaN || y.isNaN ) {
|
---|
3098 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3099 | }
|
---|
3100 | return floatXLt( &x, &y );
|
---|
3101 |
|
---|
3102 | }
|
---|
3103 |
|
---|
3104 | bool slow_f64_eq_signaling( float64_t a, float64_t b )
|
---|
3105 | {
|
---|
3106 | struct floatX x, y;
|
---|
3107 |
|
---|
3108 | f64ToFloatX( a, &x );
|
---|
3109 | f64ToFloatX( b, &y );
|
---|
3110 | if ( x.isNaN || y.isNaN ) {
|
---|
3111 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3112 | }
|
---|
3113 | return floatXEq( &x, &y );
|
---|
3114 |
|
---|
3115 | }
|
---|
3116 |
|
---|
3117 | bool slow_f64_le_quiet( float64_t a, float64_t b )
|
---|
3118 | {
|
---|
3119 | struct floatX x, y;
|
---|
3120 |
|
---|
3121 | f64ToFloatX( a, &x );
|
---|
3122 | f64ToFloatX( b, &y );
|
---|
3123 | return floatXLe( &x, &y );
|
---|
3124 |
|
---|
3125 | }
|
---|
3126 |
|
---|
3127 | bool slow_f64_lt_quiet( float64_t a, float64_t b )
|
---|
3128 | {
|
---|
3129 | struct floatX x, y;
|
---|
3130 |
|
---|
3131 | f64ToFloatX( a, &x );
|
---|
3132 | f64ToFloatX( b, &y );
|
---|
3133 | return floatXLt( &x, &y );
|
---|
3134 |
|
---|
3135 | }
|
---|
3136 |
|
---|
3137 | #endif
|
---|
3138 |
|
---|
3139 | #ifdef EXTFLOAT80
|
---|
3140 |
|
---|
3141 | uint_fast32_t
|
---|
3142 | slow_extF80M_to_ui32(
|
---|
3143 | const extFloat80_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3144 | {
|
---|
3145 | struct floatX x;
|
---|
3146 |
|
---|
3147 | extF80MToFloatX( aPtr, &x );
|
---|
3148 | return floatXToUI32( &x, roundingMode, exact );
|
---|
3149 |
|
---|
3150 | }
|
---|
3151 |
|
---|
3152 | uint_fast64_t
|
---|
3153 | slow_extF80M_to_ui64(
|
---|
3154 | const extFloat80_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3155 | {
|
---|
3156 | struct floatX x;
|
---|
3157 |
|
---|
3158 | extF80MToFloatX( aPtr, &x );
|
---|
3159 | return floatXToUI64( &x, roundingMode, exact );
|
---|
3160 |
|
---|
3161 | }
|
---|
3162 |
|
---|
3163 | int_fast32_t
|
---|
3164 | slow_extF80M_to_i32(
|
---|
3165 | const extFloat80_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3166 | {
|
---|
3167 | struct floatX x;
|
---|
3168 |
|
---|
3169 | extF80MToFloatX( aPtr, &x );
|
---|
3170 | return floatXToI32( &x, roundingMode, exact );
|
---|
3171 |
|
---|
3172 | }
|
---|
3173 |
|
---|
3174 | int_fast64_t
|
---|
3175 | slow_extF80M_to_i64(
|
---|
3176 | const extFloat80_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3177 | {
|
---|
3178 | struct floatX x;
|
---|
3179 |
|
---|
3180 | extF80MToFloatX( aPtr, &x );
|
---|
3181 | return floatXToI64( &x, roundingMode, exact );
|
---|
3182 |
|
---|
3183 | }
|
---|
3184 |
|
---|
3185 | uint_fast32_t
|
---|
3186 | slow_extF80M_to_ui32_r_minMag( const extFloat80_t *aPtr, bool exact )
|
---|
3187 | {
|
---|
3188 | struct floatX x;
|
---|
3189 |
|
---|
3190 | extF80MToFloatX( aPtr, &x );
|
---|
3191 | return floatXToUI32( &x, softfloat_round_minMag, exact );
|
---|
3192 |
|
---|
3193 | }
|
---|
3194 |
|
---|
3195 | uint_fast64_t
|
---|
3196 | slow_extF80M_to_ui64_r_minMag( const extFloat80_t *aPtr, bool exact )
|
---|
3197 | {
|
---|
3198 | struct floatX x;
|
---|
3199 |
|
---|
3200 | extF80MToFloatX( aPtr, &x );
|
---|
3201 | return floatXToUI64( &x, softfloat_round_minMag, exact );
|
---|
3202 |
|
---|
3203 | }
|
---|
3204 |
|
---|
3205 | int_fast32_t
|
---|
3206 | slow_extF80M_to_i32_r_minMag( const extFloat80_t *aPtr, bool exact )
|
---|
3207 | {
|
---|
3208 | struct floatX x;
|
---|
3209 |
|
---|
3210 | extF80MToFloatX( aPtr, &x );
|
---|
3211 | return floatXToI32( &x, softfloat_round_minMag, exact );
|
---|
3212 |
|
---|
3213 | }
|
---|
3214 |
|
---|
3215 | int_fast64_t
|
---|
3216 | slow_extF80M_to_i64_r_minMag( const extFloat80_t *aPtr, bool exact )
|
---|
3217 | {
|
---|
3218 | struct floatX x;
|
---|
3219 |
|
---|
3220 | extF80MToFloatX( aPtr, &x );
|
---|
3221 | return floatXToI64( &x, softfloat_round_minMag, exact );
|
---|
3222 |
|
---|
3223 | }
|
---|
3224 |
|
---|
3225 | #ifdef FLOAT16
|
---|
3226 |
|
---|
3227 | float16_t slow_extF80M_to_f16( const extFloat80_t *aPtr )
|
---|
3228 | {
|
---|
3229 | struct floatX x;
|
---|
3230 |
|
---|
3231 | extF80MToFloatX( aPtr, &x );
|
---|
3232 | return floatXToF16( &x );
|
---|
3233 |
|
---|
3234 | }
|
---|
3235 |
|
---|
3236 | #endif
|
---|
3237 |
|
---|
3238 | float32_t slow_extF80M_to_f32( const extFloat80_t *aPtr )
|
---|
3239 | {
|
---|
3240 | struct floatX x;
|
---|
3241 |
|
---|
3242 | extF80MToFloatX( aPtr, &x );
|
---|
3243 | return floatXToF32( &x );
|
---|
3244 |
|
---|
3245 | }
|
---|
3246 |
|
---|
3247 | #ifdef FLOAT64
|
---|
3248 |
|
---|
3249 | float64_t slow_extF80M_to_f64( const extFloat80_t *aPtr )
|
---|
3250 | {
|
---|
3251 | struct floatX x;
|
---|
3252 |
|
---|
3253 | extF80MToFloatX( aPtr, &x );
|
---|
3254 | return floatXToF64( &x );
|
---|
3255 |
|
---|
3256 | }
|
---|
3257 |
|
---|
3258 | #endif
|
---|
3259 |
|
---|
3260 | #ifdef FLOAT128
|
---|
3261 |
|
---|
3262 | void slow_extF80M_to_f128M( const extFloat80_t *aPtr, float128_t *zPtr )
|
---|
3263 | {
|
---|
3264 | struct floatX x;
|
---|
3265 |
|
---|
3266 | extF80MToFloatX( aPtr, &x );
|
---|
3267 | floatXToF128M( &x, zPtr );
|
---|
3268 |
|
---|
3269 | }
|
---|
3270 |
|
---|
3271 | #endif
|
---|
3272 |
|
---|
3273 | void
|
---|
3274 | slow_extF80M_roundToInt(
|
---|
3275 | const extFloat80_t *aPtr,
|
---|
3276 | uint_fast8_t roundingMode,
|
---|
3277 | bool exact,
|
---|
3278 | extFloat80_t *zPtr
|
---|
3279 | )
|
---|
3280 | {
|
---|
3281 | struct floatX x;
|
---|
3282 |
|
---|
3283 | extF80MToFloatX( aPtr, &x );
|
---|
3284 | floatXRoundToInt( &x, roundingMode, exact );
|
---|
3285 | floatXToExtF80M( &x, zPtr );
|
---|
3286 |
|
---|
3287 | }
|
---|
3288 |
|
---|
3289 | void
|
---|
3290 | slow_extF80M_add(
|
---|
3291 | const extFloat80_t *aPtr, const extFloat80_t *bPtr, extFloat80_t *zPtr )
|
---|
3292 | {
|
---|
3293 | struct floatX x, y;
|
---|
3294 |
|
---|
3295 | extF80MToFloatX( aPtr, &x );
|
---|
3296 | extF80MToFloatX( bPtr, &y );
|
---|
3297 | floatXAdd( &x, &y );
|
---|
3298 | floatXToExtF80M( &x, zPtr );
|
---|
3299 |
|
---|
3300 | }
|
---|
3301 |
|
---|
3302 | void
|
---|
3303 | slow_extF80M_sub(
|
---|
3304 | const extFloat80_t *aPtr, const extFloat80_t *bPtr, extFloat80_t *zPtr )
|
---|
3305 | {
|
---|
3306 | struct floatX x, y;
|
---|
3307 |
|
---|
3308 | extF80MToFloatX( aPtr, &x );
|
---|
3309 | extF80MToFloatX( bPtr, &y );
|
---|
3310 | y.sign = !y.sign;
|
---|
3311 | floatXAdd( &x, &y );
|
---|
3312 | floatXToExtF80M( &x, zPtr );
|
---|
3313 |
|
---|
3314 | }
|
---|
3315 |
|
---|
3316 | void
|
---|
3317 | slow_extF80M_mul(
|
---|
3318 | const extFloat80_t *aPtr, const extFloat80_t *bPtr, extFloat80_t *zPtr )
|
---|
3319 | {
|
---|
3320 | struct floatX x, y;
|
---|
3321 |
|
---|
3322 | extF80MToFloatX( aPtr, &x );
|
---|
3323 | extF80MToFloatX( bPtr, &y );
|
---|
3324 | floatXMul( &x, &y );
|
---|
3325 | floatXToExtF80M( &x, zPtr );
|
---|
3326 |
|
---|
3327 | }
|
---|
3328 |
|
---|
3329 | void
|
---|
3330 | slow_extF80M_div(
|
---|
3331 | const extFloat80_t *aPtr, const extFloat80_t *bPtr, extFloat80_t *zPtr )
|
---|
3332 | {
|
---|
3333 | struct floatX x, y;
|
---|
3334 |
|
---|
3335 | extF80MToFloatX( aPtr, &x );
|
---|
3336 | extF80MToFloatX( bPtr, &y );
|
---|
3337 | floatXDiv( &x, &y );
|
---|
3338 | floatXToExtF80M( &x, zPtr );
|
---|
3339 |
|
---|
3340 | }
|
---|
3341 |
|
---|
3342 | void
|
---|
3343 | slow_extF80M_rem(
|
---|
3344 | const extFloat80_t *aPtr, const extFloat80_t *bPtr, extFloat80_t *zPtr )
|
---|
3345 | {
|
---|
3346 | struct floatX x, y;
|
---|
3347 |
|
---|
3348 | extF80MToFloatX( aPtr, &x );
|
---|
3349 | extF80MToFloatX( bPtr, &y );
|
---|
3350 | floatXRem( &x, &y );
|
---|
3351 | floatXToExtF80M( &x, zPtr );
|
---|
3352 |
|
---|
3353 | }
|
---|
3354 |
|
---|
3355 | void slow_extF80M_sqrt( const extFloat80_t *aPtr, extFloat80_t *zPtr )
|
---|
3356 | {
|
---|
3357 | struct floatX x;
|
---|
3358 |
|
---|
3359 | extF80MToFloatX( aPtr, &x );
|
---|
3360 | floatXSqrt( &x );
|
---|
3361 | floatXToExtF80M( &x, zPtr );
|
---|
3362 |
|
---|
3363 | }
|
---|
3364 |
|
---|
3365 | bool slow_extF80M_eq( const extFloat80_t *aPtr, const extFloat80_t *bPtr )
|
---|
3366 | {
|
---|
3367 | struct floatX x, y;
|
---|
3368 |
|
---|
3369 | extF80MToFloatX( aPtr, &x );
|
---|
3370 | extF80MToFloatX( bPtr, &y );
|
---|
3371 | return floatXEq( &x, &y );
|
---|
3372 |
|
---|
3373 | }
|
---|
3374 |
|
---|
3375 | bool slow_extF80M_le( const extFloat80_t *aPtr, const extFloat80_t *bPtr )
|
---|
3376 | {
|
---|
3377 | struct floatX x, y;
|
---|
3378 |
|
---|
3379 | extF80MToFloatX( aPtr, &x );
|
---|
3380 | extF80MToFloatX( bPtr, &y );
|
---|
3381 | if ( x.isNaN || y.isNaN ) {
|
---|
3382 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3383 | }
|
---|
3384 | return floatXLe( &x, &y );
|
---|
3385 |
|
---|
3386 | }
|
---|
3387 |
|
---|
3388 | bool slow_extF80M_lt( const extFloat80_t *aPtr, const extFloat80_t *bPtr )
|
---|
3389 | {
|
---|
3390 | struct floatX x, y;
|
---|
3391 |
|
---|
3392 | extF80MToFloatX( aPtr, &x );
|
---|
3393 | extF80MToFloatX( bPtr, &y );
|
---|
3394 | if ( x.isNaN || y.isNaN ) {
|
---|
3395 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3396 | }
|
---|
3397 | return floatXLt( &x, &y );
|
---|
3398 |
|
---|
3399 | }
|
---|
3400 |
|
---|
3401 | bool
|
---|
3402 | slow_extF80M_eq_signaling(
|
---|
3403 | const extFloat80_t *aPtr, const extFloat80_t *bPtr )
|
---|
3404 | {
|
---|
3405 | struct floatX x, y;
|
---|
3406 |
|
---|
3407 | extF80MToFloatX( aPtr, &x );
|
---|
3408 | extF80MToFloatX( bPtr, &y );
|
---|
3409 | if ( x.isNaN || y.isNaN ) {
|
---|
3410 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3411 | }
|
---|
3412 | return floatXEq( &x, &y );
|
---|
3413 |
|
---|
3414 | }
|
---|
3415 |
|
---|
3416 | bool slow_extF80M_le_quiet( const extFloat80_t *aPtr, const extFloat80_t *bPtr )
|
---|
3417 | {
|
---|
3418 | struct floatX x, y;
|
---|
3419 |
|
---|
3420 | extF80MToFloatX( aPtr, &x );
|
---|
3421 | extF80MToFloatX( bPtr, &y );
|
---|
3422 | return floatXLe( &x, &y );
|
---|
3423 |
|
---|
3424 | }
|
---|
3425 |
|
---|
3426 | bool slow_extF80M_lt_quiet( const extFloat80_t *aPtr, const extFloat80_t *bPtr )
|
---|
3427 | {
|
---|
3428 | struct floatX x, y;
|
---|
3429 |
|
---|
3430 | extF80MToFloatX( aPtr, &x );
|
---|
3431 | extF80MToFloatX( bPtr, &y );
|
---|
3432 | return floatXLt( &x, &y );
|
---|
3433 |
|
---|
3434 | }
|
---|
3435 |
|
---|
3436 | #endif
|
---|
3437 |
|
---|
3438 | #ifdef FLOAT128
|
---|
3439 |
|
---|
3440 | uint_fast32_t
|
---|
3441 | slow_f128M_to_ui32(
|
---|
3442 | const float128_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3443 | {
|
---|
3444 | struct floatX x;
|
---|
3445 |
|
---|
3446 | f128MToFloatX( aPtr, &x );
|
---|
3447 | return floatXToUI32( &x, roundingMode, exact );
|
---|
3448 |
|
---|
3449 | }
|
---|
3450 |
|
---|
3451 | uint_fast64_t
|
---|
3452 | slow_f128M_to_ui64(
|
---|
3453 | const float128_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3454 | {
|
---|
3455 | struct floatX x;
|
---|
3456 |
|
---|
3457 | f128MToFloatX( aPtr, &x );
|
---|
3458 | return floatXToUI64( &x, roundingMode, exact );
|
---|
3459 |
|
---|
3460 | }
|
---|
3461 |
|
---|
3462 | int_fast32_t
|
---|
3463 | slow_f128M_to_i32(
|
---|
3464 | const float128_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3465 | {
|
---|
3466 | struct floatX x;
|
---|
3467 |
|
---|
3468 | f128MToFloatX( aPtr, &x );
|
---|
3469 | return floatXToI32( &x, roundingMode, exact );
|
---|
3470 |
|
---|
3471 | }
|
---|
3472 |
|
---|
3473 | int_fast64_t
|
---|
3474 | slow_f128M_to_i64(
|
---|
3475 | const float128_t *aPtr, uint_fast8_t roundingMode, bool exact )
|
---|
3476 | {
|
---|
3477 | struct floatX x;
|
---|
3478 |
|
---|
3479 | f128MToFloatX( aPtr, &x );
|
---|
3480 | return floatXToI64( &x, roundingMode, exact );
|
---|
3481 |
|
---|
3482 | }
|
---|
3483 |
|
---|
3484 | uint_fast32_t slow_f128M_to_ui32_r_minMag( const float128_t *aPtr, bool exact )
|
---|
3485 | {
|
---|
3486 | struct floatX x;
|
---|
3487 |
|
---|
3488 | f128MToFloatX( aPtr, &x );
|
---|
3489 | return floatXToUI32( &x, softfloat_round_minMag, exact );
|
---|
3490 |
|
---|
3491 | }
|
---|
3492 |
|
---|
3493 | uint_fast64_t slow_f128M_to_ui64_r_minMag( const float128_t *aPtr, bool exact )
|
---|
3494 | {
|
---|
3495 | struct floatX x;
|
---|
3496 |
|
---|
3497 | f128MToFloatX( aPtr, &x );
|
---|
3498 | return floatXToUI64( &x, softfloat_round_minMag, exact );
|
---|
3499 |
|
---|
3500 | }
|
---|
3501 |
|
---|
3502 | int_fast32_t slow_f128M_to_i32_r_minMag( const float128_t *aPtr, bool exact )
|
---|
3503 | {
|
---|
3504 | struct floatX x;
|
---|
3505 |
|
---|
3506 | f128MToFloatX( aPtr, &x );
|
---|
3507 | return floatXToI32( &x, softfloat_round_minMag, exact );
|
---|
3508 |
|
---|
3509 | }
|
---|
3510 |
|
---|
3511 | int_fast64_t slow_f128M_to_i64_r_minMag( const float128_t *aPtr, bool exact )
|
---|
3512 | {
|
---|
3513 | struct floatX x;
|
---|
3514 |
|
---|
3515 | f128MToFloatX( aPtr, &x );
|
---|
3516 | return floatXToI64( &x, softfloat_round_minMag, exact );
|
---|
3517 |
|
---|
3518 | }
|
---|
3519 |
|
---|
3520 | #ifdef FLOAT16
|
---|
3521 |
|
---|
3522 | float16_t slow_f128M_to_f16( const float128_t *aPtr )
|
---|
3523 | {
|
---|
3524 | struct floatX x;
|
---|
3525 |
|
---|
3526 | f128MToFloatX( aPtr, &x );
|
---|
3527 | return floatXToF16( &x );
|
---|
3528 |
|
---|
3529 | }
|
---|
3530 |
|
---|
3531 | #endif
|
---|
3532 |
|
---|
3533 | float32_t slow_f128M_to_f32( const float128_t *aPtr )
|
---|
3534 | {
|
---|
3535 | struct floatX x;
|
---|
3536 |
|
---|
3537 | f128MToFloatX( aPtr, &x );
|
---|
3538 | return floatXToF32( &x );
|
---|
3539 |
|
---|
3540 | }
|
---|
3541 |
|
---|
3542 | #ifdef FLOAT64
|
---|
3543 |
|
---|
3544 | float64_t slow_f128M_to_f64( const float128_t *aPtr )
|
---|
3545 | {
|
---|
3546 | struct floatX x;
|
---|
3547 |
|
---|
3548 | f128MToFloatX( aPtr, &x );
|
---|
3549 | return floatXToF64( &x );
|
---|
3550 |
|
---|
3551 | }
|
---|
3552 |
|
---|
3553 | #endif
|
---|
3554 |
|
---|
3555 | #ifdef EXTFLOAT80
|
---|
3556 |
|
---|
3557 | void slow_f128M_to_extF80M( const float128_t *aPtr, extFloat80_t *zPtr )
|
---|
3558 | {
|
---|
3559 | struct floatX x;
|
---|
3560 |
|
---|
3561 | f128MToFloatX( aPtr, &x );
|
---|
3562 | floatXToExtF80M( &x, zPtr );
|
---|
3563 |
|
---|
3564 | }
|
---|
3565 |
|
---|
3566 | #endif
|
---|
3567 |
|
---|
3568 | void
|
---|
3569 | slow_f128M_roundToInt(
|
---|
3570 | const float128_t *aPtr,
|
---|
3571 | uint_fast8_t roundingMode,
|
---|
3572 | bool exact,
|
---|
3573 | float128_t *zPtr
|
---|
3574 | )
|
---|
3575 | {
|
---|
3576 | struct floatX x;
|
---|
3577 |
|
---|
3578 | f128MToFloatX( aPtr, &x );
|
---|
3579 | floatXRoundToInt( &x, roundingMode, exact );
|
---|
3580 | floatXToF128M( &x, zPtr );
|
---|
3581 |
|
---|
3582 | }
|
---|
3583 |
|
---|
3584 | void
|
---|
3585 | slow_f128M_add(
|
---|
3586 | const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
---|
3587 | {
|
---|
3588 | struct floatX x, y;
|
---|
3589 |
|
---|
3590 | f128MToFloatX( aPtr, &x );
|
---|
3591 | f128MToFloatX( bPtr, &y );
|
---|
3592 | floatXAdd( &x, &y );
|
---|
3593 | floatXToF128M( &x, zPtr );
|
---|
3594 |
|
---|
3595 | }
|
---|
3596 |
|
---|
3597 | void
|
---|
3598 | slow_f128M_sub(
|
---|
3599 | const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
---|
3600 | {
|
---|
3601 | struct floatX x, y;
|
---|
3602 |
|
---|
3603 | f128MToFloatX( aPtr, &x );
|
---|
3604 | f128MToFloatX( bPtr, &y );
|
---|
3605 | y.sign = !y.sign;
|
---|
3606 | floatXAdd( &x, &y );
|
---|
3607 | floatXToF128M( &x, zPtr );
|
---|
3608 |
|
---|
3609 | }
|
---|
3610 |
|
---|
3611 | void
|
---|
3612 | slow_f128M_mul(
|
---|
3613 | const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
---|
3614 | {
|
---|
3615 | struct floatX x, y;
|
---|
3616 |
|
---|
3617 | f128MToFloatX( aPtr, &x );
|
---|
3618 | f128MToFloatX( bPtr, &y );
|
---|
3619 | floatXMul( &x, &y );
|
---|
3620 | floatXToF128M( &x, zPtr );
|
---|
3621 |
|
---|
3622 | }
|
---|
3623 |
|
---|
3624 | void
|
---|
3625 | slow_f128M_mulAdd(
|
---|
3626 | const float128_t *aPtr,
|
---|
3627 | const float128_t *bPtr,
|
---|
3628 | const float128_t *cPtr,
|
---|
3629 | float128_t *zPtr
|
---|
3630 | )
|
---|
3631 | {
|
---|
3632 | struct floatX256 x, y;
|
---|
3633 |
|
---|
3634 | f128MToFloatX256( aPtr, &x );
|
---|
3635 | f128MToFloatX256( bPtr, &y );
|
---|
3636 | floatX256Mul( &x, &y );
|
---|
3637 | f128MToFloatX256( cPtr, &y );
|
---|
3638 | floatX256Add( &x, &y );
|
---|
3639 | floatX256ToF128M( &x, zPtr );
|
---|
3640 |
|
---|
3641 | }
|
---|
3642 |
|
---|
3643 | void
|
---|
3644 | slow_f128M_div(
|
---|
3645 | const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
---|
3646 | {
|
---|
3647 | struct floatX x, y;
|
---|
3648 |
|
---|
3649 | f128MToFloatX( aPtr, &x );
|
---|
3650 | f128MToFloatX( bPtr, &y );
|
---|
3651 | floatXDiv( &x, &y );
|
---|
3652 | floatXToF128M( &x, zPtr );
|
---|
3653 |
|
---|
3654 | }
|
---|
3655 |
|
---|
3656 | void
|
---|
3657 | slow_f128M_rem(
|
---|
3658 | const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
---|
3659 | {
|
---|
3660 | struct floatX x, y;
|
---|
3661 |
|
---|
3662 | f128MToFloatX( aPtr, &x );
|
---|
3663 | f128MToFloatX( bPtr, &y );
|
---|
3664 | floatXRem( &x, &y );
|
---|
3665 | floatXToF128M( &x, zPtr );
|
---|
3666 |
|
---|
3667 | }
|
---|
3668 |
|
---|
3669 | void slow_f128M_sqrt( const float128_t *aPtr, float128_t *zPtr )
|
---|
3670 | {
|
---|
3671 | struct floatX x;
|
---|
3672 |
|
---|
3673 | f128MToFloatX( aPtr, &x );
|
---|
3674 | floatXSqrt( &x );
|
---|
3675 | floatXToF128M( &x, zPtr );
|
---|
3676 |
|
---|
3677 | }
|
---|
3678 |
|
---|
3679 | bool slow_f128M_eq( const float128_t *aPtr, const float128_t *bPtr )
|
---|
3680 | {
|
---|
3681 | struct floatX x, y;
|
---|
3682 |
|
---|
3683 | f128MToFloatX( aPtr, &x );
|
---|
3684 | f128MToFloatX( bPtr, &y );
|
---|
3685 | return floatXEq( &x, &y );
|
---|
3686 |
|
---|
3687 | }
|
---|
3688 |
|
---|
3689 | bool slow_f128M_le( const float128_t *aPtr, const float128_t *bPtr )
|
---|
3690 | {
|
---|
3691 | struct floatX x, y;
|
---|
3692 |
|
---|
3693 | f128MToFloatX( aPtr, &x );
|
---|
3694 | f128MToFloatX( bPtr, &y );
|
---|
3695 | if ( x.isNaN || y.isNaN ) {
|
---|
3696 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3697 | }
|
---|
3698 | return floatXLe( &x, &y );
|
---|
3699 |
|
---|
3700 | }
|
---|
3701 |
|
---|
3702 | bool slow_f128M_lt( const float128_t *aPtr, const float128_t *bPtr )
|
---|
3703 | {
|
---|
3704 | struct floatX x, y;
|
---|
3705 |
|
---|
3706 | f128MToFloatX( aPtr, &x );
|
---|
3707 | f128MToFloatX( bPtr, &y );
|
---|
3708 | if ( x.isNaN || y.isNaN ) {
|
---|
3709 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3710 | }
|
---|
3711 | return floatXLt( &x, &y );
|
---|
3712 |
|
---|
3713 | }
|
---|
3714 |
|
---|
3715 | bool slow_f128M_eq_signaling( const float128_t *aPtr, const float128_t *bPtr )
|
---|
3716 | {
|
---|
3717 | struct floatX x, y;
|
---|
3718 |
|
---|
3719 | f128MToFloatX( aPtr, &x );
|
---|
3720 | f128MToFloatX( bPtr, &y );
|
---|
3721 | if ( x.isNaN || y.isNaN ) {
|
---|
3722 | slowfloat_exceptionFlags |= softfloat_flag_invalid;
|
---|
3723 | }
|
---|
3724 | return floatXEq( &x, &y );
|
---|
3725 |
|
---|
3726 | }
|
---|
3727 |
|
---|
3728 | bool slow_f128M_le_quiet( const float128_t *aPtr, const float128_t *bPtr )
|
---|
3729 | {
|
---|
3730 | struct floatX x, y;
|
---|
3731 |
|
---|
3732 | f128MToFloatX( aPtr, &x );
|
---|
3733 | f128MToFloatX( bPtr, &y );
|
---|
3734 | return floatXLe( &x, &y );
|
---|
3735 |
|
---|
3736 | }
|
---|
3737 |
|
---|
3738 | bool slow_f128M_lt_quiet( const float128_t *aPtr, const float128_t *bPtr )
|
---|
3739 | {
|
---|
3740 | struct floatX x, y;
|
---|
3741 |
|
---|
3742 | f128MToFloatX( aPtr, &x );
|
---|
3743 | f128MToFloatX( bPtr, &y );
|
---|
3744 | return floatXLt( &x, &y );
|
---|
3745 |
|
---|
3746 | }
|
---|
3747 |
|
---|
3748 | #endif
|
---|
3749 |
|
---|