]>
Commit | Line | Data |
---|---|---|
1 | the niosii-gnutools-src-5.1.tgz release from altera is missing a lot of | |
2 | core gcc files, this patch adds them back in | |
3 | ||
4 | --- ./gcc/gcc/config/nios2/crti.asm | |
5 | +++ ./gcc/gcc/config/nios2/crti.asm | |
6 | @@ -0,0 +1,88 @@ | |
7 | +/* | |
8 | + Copyright (C) 2003 | |
9 | + by Jonah Graham (jgraham@altera.com) | |
10 | + | |
11 | +This file is free software; you can redistribute it and/or modify it | |
12 | +under the terms of the GNU General Public License as published by the | |
13 | +Free Software Foundation; either version 2, or (at your option) any | |
14 | +later version. | |
15 | + | |
16 | +In addition to the permissions in the GNU General Public License, the | |
17 | +Free Software Foundation gives you unlimited permission to link the | |
18 | +compiled version of this file with other programs, and to distribute | |
19 | +those programs without any restriction coming from the use of this | |
20 | +file. (The General Public License restrictions do apply in other | |
21 | +respects; for example, they cover modification of the file, and | |
22 | +distribution when not linked into another program.) | |
23 | + | |
24 | +This file is distributed in the hope that it will be useful, but | |
25 | +WITHOUT ANY WARRANTY; without even the implied warranty of | |
26 | +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
27 | +General Public License for more details. | |
28 | + | |
29 | +You should have received a copy of the GNU General Public License | |
30 | +along with this program; see the file COPYING. If not, write to | |
31 | +the Free Software Foundation, 59 Temple Place - Suite 330, | |
32 | +Boston, MA 02111-1307, USA. | |
33 | + | |
34 | + As a special exception, if you link this library with files | |
35 | + compiled with GCC to produce an executable, this does not cause | |
36 | + the resulting executable to be covered by the GNU General Public License. | |
37 | + This exception does not however invalidate any other reasons why | |
38 | + the executable file might be covered by the GNU General Public License. | |
39 | + | |
40 | + | |
41 | +This file just make a stack frame for the contents of the .fini and | |
42 | +.init sections. Users may put any desired instructions in those | |
43 | +sections. | |
44 | + | |
45 | + | |
46 | +While technically any code can be put in the init and fini sections | |
47 | +most stuff will not work other than stuff which obeys the call frame | |
48 | +and ABI. All the call-preserved registers are saved, the call clobbered | |
49 | +registers should have been saved by the code calling init and fini. | |
50 | + | |
51 | +See crtstuff.c for an example of code that inserts itself in the | |
52 | +init and fini sections. | |
53 | + | |
54 | +See crt0.s for the code that calls init and fini. | |
55 | +*/ | |
56 | + | |
57 | + .file "crti.asm" | |
58 | + | |
59 | + .section ".init" | |
60 | + .align 2 | |
61 | + .global _init | |
62 | +_init: | |
63 | + addi sp, sp, -48 | |
64 | + stw ra, 44(sp) | |
65 | + stw r23, 40(sp) | |
66 | + stw r22, 36(sp) | |
67 | + stw r21, 32(sp) | |
68 | + stw r20, 28(sp) | |
69 | + stw r19, 24(sp) | |
70 | + stw r18, 20(sp) | |
71 | + stw r17, 16(sp) | |
72 | + stw r16, 12(sp) | |
73 | + stw fp, 8(sp) | |
74 | + mov fp, sp | |
75 | + | |
76 | + | |
77 | + .section ".fini" | |
78 | + .align 2 | |
79 | + .global _fini | |
80 | +_fini: | |
81 | + addi sp, sp, -48 | |
82 | + stw ra, 44(sp) | |
83 | + stw r23, 40(sp) | |
84 | + stw r22, 36(sp) | |
85 | + stw r21, 32(sp) | |
86 | + stw r20, 28(sp) | |
87 | + stw r19, 24(sp) | |
88 | + stw r18, 20(sp) | |
89 | + stw r17, 16(sp) | |
90 | + stw r16, 12(sp) | |
91 | + stw fp, 8(sp) | |
92 | + mov fp, sp | |
93 | + | |
94 | + | |
95 | --- ./gcc/gcc/config/nios2/crtn.asm | |
96 | +++ ./gcc/gcc/config/nios2/crtn.asm | |
97 | @@ -0,0 +1,70 @@ | |
98 | +/* | |
99 | + Copyright (C) 2003 | |
100 | + by Jonah Graham (jgraham@altera.com) | |
101 | + | |
102 | +This file is free software; you can redistribute it and/or modify it | |
103 | +under the terms of the GNU General Public License as published by the | |
104 | +Free Software Foundation; either version 2, or (at your option) any | |
105 | +later version. | |
106 | + | |
107 | +In addition to the permissions in the GNU General Public License, the | |
108 | +Free Software Foundation gives you unlimited permission to link the | |
109 | +compiled version of this file with other programs, and to distribute | |
110 | +those programs without any restriction coming from the use of this | |
111 | +file. (The General Public License restrictions do apply in other | |
112 | +respects; for example, they cover modification of the file, and | |
113 | +distribution when not linked into another program.) | |
114 | + | |
115 | +This file is distributed in the hope that it will be useful, but | |
116 | +WITHOUT ANY WARRANTY; without even the implied warranty of | |
117 | +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
118 | +General Public License for more details. | |
119 | + | |
120 | +You should have received a copy of the GNU General Public License | |
121 | +along with this program; see the file COPYING. If not, write to | |
122 | +the Free Software Foundation, 59 Temple Place - Suite 330, | |
123 | +Boston, MA 02111-1307, USA. | |
124 | + | |
125 | + As a special exception, if you link this library with files | |
126 | + compiled with GCC to produce an executable, this does not cause | |
127 | + the resulting executable to be covered by the GNU General Public License. | |
128 | + This exception does not however invalidate any other reasons why | |
129 | + the executable file might be covered by the GNU General Public License. | |
130 | + | |
131 | + | |
132 | +This file just makes sure that the .fini and .init sections do in | |
133 | +fact return. Users may put any desired instructions in those sections. | |
134 | +This file is the last thing linked into any executable. | |
135 | +*/ | |
136 | + .file "crtn.asm" | |
137 | + | |
138 | + | |
139 | + | |
140 | + .section ".init" | |
141 | + ldw ra, 44(sp) | |
142 | + ldw r23, 40(sp) | |
143 | + ldw r22, 36(sp) | |
144 | + ldw r21, 32(sp) | |
145 | + ldw r20, 28(sp) | |
146 | + ldw r19, 24(sp) | |
147 | + ldw r18, 20(sp) | |
148 | + ldw r17, 16(sp) | |
149 | + ldw r16, 12(sp) | |
150 | + ldw fp, 8(sp) | |
151 | + addi sp, sp, -48 | |
152 | + ret | |
153 | + | |
154 | + .section ".fini" | |
155 | + ldw ra, 44(sp) | |
156 | + ldw r23, 40(sp) | |
157 | + ldw r22, 36(sp) | |
158 | + ldw r21, 32(sp) | |
159 | + ldw r20, 28(sp) | |
160 | + ldw r19, 24(sp) | |
161 | + ldw r18, 20(sp) | |
162 | + ldw r17, 16(sp) | |
163 | + ldw r16, 12(sp) | |
164 | + ldw fp, 8(sp) | |
165 | + addi sp, sp, -48 | |
166 | + ret | |
167 | + | |
168 | --- ./gcc/gcc/config/nios2/lib2-divmod-hi.c | |
169 | +++ ./gcc/gcc/config/nios2/lib2-divmod-hi.c | |
170 | @@ -0,0 +1,123 @@ | |
171 | + | |
172 | +/* We include auto-host.h here to get HAVE_GAS_HIDDEN. This is | |
173 | + supposedly valid even though this is a "target" file. */ | |
174 | +#include "auto-host.h" | |
175 | + | |
176 | + | |
177 | +#include "tconfig.h" | |
178 | +#include "tsystem.h" | |
179 | +#include "coretypes.h" | |
180 | +#include "tm.h" | |
181 | + | |
182 | + | |
183 | +/* Don't use `fancy_abort' here even if config.h says to use it. */ | |
184 | +#ifdef abort | |
185 | +#undef abort | |
186 | +#endif | |
187 | + | |
188 | + | |
189 | +#ifdef HAVE_GAS_HIDDEN | |
190 | +#define ATTRIBUTE_HIDDEN __attribute__ ((__visibility__ ("hidden"))) | |
191 | +#else | |
192 | +#define ATTRIBUTE_HIDDEN | |
193 | +#endif | |
194 | + | |
195 | +#include "libgcc2.h" | |
196 | + | |
197 | +extern HItype __modhi3 (HItype, HItype); | |
198 | +extern HItype __divhi3 (HItype, HItype); | |
199 | +extern HItype __umodhi3 (HItype, HItype); | |
200 | +extern HItype __udivhi3 (HItype, HItype); | |
201 | + | |
202 | +static UHItype udivmodhi4(UHItype, UHItype, word_type); | |
203 | + | |
204 | +static UHItype | |
205 | +udivmodhi4(UHItype num, UHItype den, word_type modwanted) | |
206 | +{ | |
207 | + UHItype bit = 1; | |
208 | + UHItype res = 0; | |
209 | + | |
210 | + while (den < num && bit && !(den & (1L<<15))) | |
211 | + { | |
212 | + den <<=1; | |
213 | + bit <<=1; | |
214 | + } | |
215 | + while (bit) | |
216 | + { | |
217 | + if (num >= den) | |
218 | + { | |
219 | + num -= den; | |
220 | + res |= bit; | |
221 | + } | |
222 | + bit >>=1; | |
223 | + den >>=1; | |
224 | + } | |
225 | + if (modwanted) return num; | |
226 | + return res; | |
227 | +} | |
228 | + | |
229 | + | |
230 | +HItype | |
231 | +__divhi3 (HItype a, HItype b) | |
232 | +{ | |
233 | + word_type neg = 0; | |
234 | + HItype res; | |
235 | + | |
236 | + if (a < 0) | |
237 | + { | |
238 | + a = -a; | |
239 | + neg = !neg; | |
240 | + } | |
241 | + | |
242 | + if (b < 0) | |
243 | + { | |
244 | + b = -b; | |
245 | + neg = !neg; | |
246 | + } | |
247 | + | |
248 | + res = udivmodhi4 (a, b, 0); | |
249 | + | |
250 | + if (neg) | |
251 | + res = -res; | |
252 | + | |
253 | + return res; | |
254 | +} | |
255 | + | |
256 | + | |
257 | +HItype | |
258 | +__modhi3 (HItype a, HItype b) | |
259 | +{ | |
260 | + word_type neg = 0; | |
261 | + HItype res; | |
262 | + | |
263 | + if (a < 0) | |
264 | + { | |
265 | + a = -a; | |
266 | + neg = 1; | |
267 | + } | |
268 | + | |
269 | + if (b < 0) | |
270 | + b = -b; | |
271 | + | |
272 | + res = udivmodhi4 (a, b, 1); | |
273 | + | |
274 | + if (neg) | |
275 | + res = -res; | |
276 | + | |
277 | + return res; | |
278 | +} | |
279 | + | |
280 | + | |
281 | +HItype | |
282 | +__udivhi3 (HItype a, HItype b) | |
283 | +{ | |
284 | + return udivmodhi4 (a, b, 0); | |
285 | +} | |
286 | + | |
287 | + | |
288 | +HItype | |
289 | +__umodhi3 (HItype a, HItype b) | |
290 | +{ | |
291 | + return udivmodhi4 (a, b, 1); | |
292 | +} | |
293 | + | |
294 | --- ./gcc/gcc/config/nios2/lib2-divmod.c | |
295 | +++ ./gcc/gcc/config/nios2/lib2-divmod.c | |
296 | @@ -0,0 +1,126 @@ | |
297 | + | |
298 | +/* We include auto-host.h here to get HAVE_GAS_HIDDEN. This is | |
299 | + supposedly valid even though this is a "target" file. */ | |
300 | +#include "auto-host.h" | |
301 | + | |
302 | + | |
303 | +#include "tconfig.h" | |
304 | +#include "tsystem.h" | |
305 | +#include "coretypes.h" | |
306 | +#include "tm.h" | |
307 | + | |
308 | + | |
309 | +/* Don't use `fancy_abort' here even if config.h says to use it. */ | |
310 | +#ifdef abort | |
311 | +#undef abort | |
312 | +#endif | |
313 | + | |
314 | + | |
315 | +#ifdef HAVE_GAS_HIDDEN | |
316 | +#define ATTRIBUTE_HIDDEN __attribute__ ((__visibility__ ("hidden"))) | |
317 | +#else | |
318 | +#define ATTRIBUTE_HIDDEN | |
319 | +#endif | |
320 | + | |
321 | +#include "libgcc2.h" | |
322 | + | |
323 | +extern SItype __modsi3 (SItype, SItype); | |
324 | +extern SItype __divsi3 (SItype, SItype); | |
325 | +extern SItype __umodsi3 (SItype, SItype); | |
326 | +extern SItype __udivsi3 (SItype, SItype); | |
327 | + | |
328 | +static USItype udivmodsi4(USItype, USItype, word_type); | |
329 | + | |
330 | +/* 16-bit SI divide and modulo as used in NIOS */ | |
331 | + | |
332 | + | |
333 | +static USItype | |
334 | +udivmodsi4(USItype num, USItype den, word_type modwanted) | |
335 | +{ | |
336 | + USItype bit = 1; | |
337 | + USItype res = 0; | |
338 | + | |
339 | + while (den < num && bit && !(den & (1L<<31))) | |
340 | + { | |
341 | + den <<=1; | |
342 | + bit <<=1; | |
343 | + } | |
344 | + while (bit) | |
345 | + { | |
346 | + if (num >= den) | |
347 | + { | |
348 | + num -= den; | |
349 | + res |= bit; | |
350 | + } | |
351 | + bit >>=1; | |
352 | + den >>=1; | |
353 | + } | |
354 | + if (modwanted) return num; | |
355 | + return res; | |
356 | +} | |
357 | + | |
358 | + | |
359 | +SItype | |
360 | +__divsi3 (SItype a, SItype b) | |
361 | +{ | |
362 | + word_type neg = 0; | |
363 | + SItype res; | |
364 | + | |
365 | + if (a < 0) | |
366 | + { | |
367 | + a = -a; | |
368 | + neg = !neg; | |
369 | + } | |
370 | + | |
371 | + if (b < 0) | |
372 | + { | |
373 | + b = -b; | |
374 | + neg = !neg; | |
375 | + } | |
376 | + | |
377 | + res = udivmodsi4 (a, b, 0); | |
378 | + | |
379 | + if (neg) | |
380 | + res = -res; | |
381 | + | |
382 | + return res; | |
383 | +} | |
384 | + | |
385 | + | |
386 | +SItype | |
387 | +__modsi3 (SItype a, SItype b) | |
388 | +{ | |
389 | + word_type neg = 0; | |
390 | + SItype res; | |
391 | + | |
392 | + if (a < 0) | |
393 | + { | |
394 | + a = -a; | |
395 | + neg = 1; | |
396 | + } | |
397 | + | |
398 | + if (b < 0) | |
399 | + b = -b; | |
400 | + | |
401 | + res = udivmodsi4 (a, b, 1); | |
402 | + | |
403 | + if (neg) | |
404 | + res = -res; | |
405 | + | |
406 | + return res; | |
407 | +} | |
408 | + | |
409 | + | |
410 | +SItype | |
411 | +__udivsi3 (SItype a, SItype b) | |
412 | +{ | |
413 | + return udivmodsi4 (a, b, 0); | |
414 | +} | |
415 | + | |
416 | + | |
417 | +SItype | |
418 | +__umodsi3 (SItype a, SItype b) | |
419 | +{ | |
420 | + return udivmodsi4 (a, b, 1); | |
421 | +} | |
422 | + | |
423 | --- ./gcc/gcc/config/nios2/lib2-divtable.c | |
424 | +++ ./gcc/gcc/config/nios2/lib2-divtable.c | |
425 | @@ -0,0 +1,46 @@ | |
426 | + | |
427 | +/* We include auto-host.h here to get HAVE_GAS_HIDDEN. This is | |
428 | + supposedly valid even though this is a "target" file. */ | |
429 | +#include "auto-host.h" | |
430 | + | |
431 | + | |
432 | +#include "tconfig.h" | |
433 | +#include "tsystem.h" | |
434 | +#include "coretypes.h" | |
435 | +#include "tm.h" | |
436 | + | |
437 | + | |
438 | +/* Don't use `fancy_abort' here even if config.h says to use it. */ | |
439 | +#ifdef abort | |
440 | +#undef abort | |
441 | +#endif | |
442 | + | |
443 | + | |
444 | +#ifdef HAVE_GAS_HIDDEN | |
445 | +#define ATTRIBUTE_HIDDEN __attribute__ ((__visibility__ ("hidden"))) | |
446 | +#else | |
447 | +#define ATTRIBUTE_HIDDEN | |
448 | +#endif | |
449 | + | |
450 | +#include "libgcc2.h" | |
451 | + | |
452 | +UQItype __divsi3_table[] = | |
453 | +{ | |
454 | + 0, 0/1, 0/2, 0/3, 0/4, 0/5, 0/6, 0/7, 0/8, 0/9, 0/10, 0/11, 0/12, 0/13, 0/14, 0/15, | |
455 | + 0, 1/1, 1/2, 1/3, 1/4, 1/5, 1/6, 1/7, 1/8, 1/9, 1/10, 1/11, 1/12, 1/13, 1/14, 1/15, | |
456 | + 0, 2/1, 2/2, 2/3, 2/4, 2/5, 2/6, 2/7, 2/8, 2/9, 2/10, 2/11, 2/12, 2/13, 2/14, 2/15, | |
457 | + 0, 3/1, 3/2, 3/3, 3/4, 3/5, 3/6, 3/7, 3/8, 3/9, 3/10, 3/11, 3/12, 3/13, 3/14, 3/15, | |
458 | + 0, 4/1, 4/2, 4/3, 4/4, 4/5, 4/6, 4/7, 4/8, 4/9, 4/10, 4/11, 4/12, 4/13, 4/14, 4/15, | |
459 | + 0, 5/1, 5/2, 5/3, 5/4, 5/5, 5/6, 5/7, 5/8, 5/9, 5/10, 5/11, 5/12, 5/13, 5/14, 5/15, | |
460 | + 0, 6/1, 6/2, 6/3, 6/4, 6/5, 6/6, 6/7, 6/8, 6/9, 6/10, 6/11, 6/12, 6/13, 6/14, 6/15, | |
461 | + 0, 7/1, 7/2, 7/3, 7/4, 7/5, 7/6, 7/7, 7/8, 7/9, 7/10, 7/11, 7/12, 7/13, 7/14, 7/15, | |
462 | + 0, 8/1, 8/2, 8/3, 8/4, 8/5, 8/6, 8/7, 8/8, 8/9, 8/10, 8/11, 8/12, 8/13, 8/14, 8/15, | |
463 | + 0, 9/1, 9/2, 9/3, 9/4, 9/5, 9/6, 9/7, 9/8, 9/9, 9/10, 9/11, 9/12, 9/13, 9/14, 9/15, | |
464 | + 0, 10/1, 10/2, 10/3, 10/4, 10/5, 10/6, 10/7, 10/8, 10/9, 10/10, 10/11, 10/12, 10/13, 10/14, 10/15, | |
465 | + 0, 11/1, 11/2, 11/3, 11/4, 11/5, 11/6, 11/7, 11/8, 11/9, 11/10, 11/11, 11/12, 11/13, 11/14, 11/15, | |
466 | + 0, 12/1, 12/2, 12/3, 12/4, 12/5, 12/6, 12/7, 12/8, 12/9, 12/10, 12/11, 12/12, 12/13, 12/14, 12/15, | |
467 | + 0, 13/1, 13/2, 13/3, 13/4, 13/5, 13/6, 13/7, 13/8, 13/9, 13/10, 13/11, 13/12, 13/13, 13/14, 13/15, | |
468 | + 0, 14/1, 14/2, 14/3, 14/4, 14/5, 14/6, 14/7, 14/8, 14/9, 14/10, 14/11, 14/12, 14/13, 14/14, 14/15, | |
469 | + 0, 15/1, 15/2, 15/3, 15/4, 15/5, 15/6, 15/7, 15/8, 15/9, 15/10, 15/11, 15/12, 15/13, 15/14, 15/15, | |
470 | +}; | |
471 | + | |
472 | --- ./gcc/gcc/config/nios2/lib2-mul.c | |
473 | +++ ./gcc/gcc/config/nios2/lib2-mul.c | |
474 | @@ -0,0 +1,103 @@ | |
475 | +/* while we are debugging (ie compile outside of gcc build) | |
476 | + disable gcc specific headers */ | |
477 | +#ifndef DEBUG_MULSI3 | |
478 | + | |
479 | + | |
480 | +/* We include auto-host.h here to get HAVE_GAS_HIDDEN. This is | |
481 | + supposedly valid even though this is a "target" file. */ | |
482 | +#include "auto-host.h" | |
483 | + | |
484 | + | |
485 | +#include "tconfig.h" | |
486 | +#include "tsystem.h" | |
487 | +#include "coretypes.h" | |
488 | +#include "tm.h" | |
489 | + | |
490 | + | |
491 | +/* Don't use `fancy_abort' here even if config.h says to use it. */ | |
492 | +#ifdef abort | |
493 | +#undef abort | |
494 | +#endif | |
495 | + | |
496 | + | |
497 | +#ifdef HAVE_GAS_HIDDEN | |
498 | +#define ATTRIBUTE_HIDDEN __attribute__ ((__visibility__ ("hidden"))) | |
499 | +#else | |
500 | +#define ATTRIBUTE_HIDDEN | |
501 | +#endif | |
502 | + | |
503 | +#include "libgcc2.h" | |
504 | + | |
505 | +#else | |
506 | +#define SItype int | |
507 | +#define USItype unsigned int | |
508 | +#endif | |
509 | + | |
510 | + | |
511 | +extern SItype __mulsi3 (SItype, SItype); | |
512 | + | |
513 | +SItype | |
514 | +__mulsi3 (SItype a, SItype b) | |
515 | +{ | |
516 | + SItype res = 0; | |
517 | + USItype cnt = a; | |
518 | + | |
519 | + while (cnt) | |
520 | + { | |
521 | + if (cnt & 1) | |
522 | + { | |
523 | + res += b; | |
524 | + } | |
525 | + b <<= 1; | |
526 | + cnt >>= 1; | |
527 | + } | |
528 | + | |
529 | + return res; | |
530 | +} | |
531 | +/* | |
532 | +TODO: Choose best alternative implementation. | |
533 | + | |
534 | +SItype | |
535 | +__divsi3 (SItype a, SItype b) | |
536 | +{ | |
537 | + SItype res = 0; | |
538 | + USItype cnt = 0; | |
539 | + | |
540 | + while (cnt < 32) | |
541 | + { | |
542 | + if (a & (1L << cnt)) | |
543 | + { | |
544 | + res += b; | |
545 | + } | |
546 | + b <<= 1; | |
547 | + cnt++; | |
548 | + } | |
549 | + | |
550 | + return res; | |
551 | +} | |
552 | +*/ | |
553 | + | |
554 | + | |
555 | +#ifdef DEBUG_MULSI3 | |
556 | + | |
557 | +int | |
558 | +main () | |
559 | +{ | |
560 | + int i, j; | |
561 | + int error = 0; | |
562 | + | |
563 | + for (i = -1000; i < 1000; i++) | |
564 | + for (j = -1000; j < 1000; j++) | |
565 | + { | |
566 | + int expect = i * j; | |
567 | + int actual = A__divsi3 (i, j); | |
568 | + if (expect != actual) | |
569 | + { | |
570 | + printf ("error: %d * %d = %d not %d\n", i, j, expect, actual); | |
571 | + error = 1; | |
572 | + } | |
573 | + } | |
574 | + | |
575 | + return error; | |
576 | +} | |
577 | +#endif | |
578 | --- ./gcc/gcc/config/nios2/nios2-dp-bit.c | |
579 | +++ ./gcc/gcc/config/nios2/nios2-dp-bit.c | |
580 | @@ -0,0 +1,1652 @@ | |
581 | + | |
582 | +/* This is a software floating point library which can be used | |
583 | + for targets without hardware floating point. | |
584 | + Copyright (C) 1994, 1995, 1996, 1997, 1998, 2000, 2001, 2002, 2003, 2004 | |
585 | + Free Software Foundation, Inc. | |
586 | + | |
587 | +This file is free software; you can redistribute it and/or modify it | |
588 | +under the terms of the GNU General Public License as published by the | |
589 | +Free Software Foundation; either version 2, or (at your option) any | |
590 | +later version. | |
591 | + | |
592 | +In addition to the permissions in the GNU General Public License, the | |
593 | +Free Software Foundation gives you unlimited permission to link the | |
594 | +compiled version of this file with other programs, and to distribute | |
595 | +those programs without any restriction coming from the use of this | |
596 | +file. (The General Public License restrictions do apply in other | |
597 | +respects; for example, they cover modification of the file, and | |
598 | +distribution when not linked into another program.) | |
599 | + | |
600 | +This file is distributed in the hope that it will be useful, but | |
601 | +WITHOUT ANY WARRANTY; without even the implied warranty of | |
602 | +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
603 | +General Public License for more details. | |
604 | + | |
605 | +You should have received a copy of the GNU General Public License | |
606 | +along with this program; see the file COPYING. If not, write to | |
607 | +the Free Software Foundation, 59 Temple Place - Suite 330, | |
608 | +Boston, MA 02111-1307, USA. */ | |
609 | + | |
610 | +/* As a special exception, if you link this library with other files, | |
611 | + some of which are compiled with GCC, to produce an executable, | |
612 | + this library does not by itself cause the resulting executable | |
613 | + to be covered by the GNU General Public License. | |
614 | + This exception does not however invalidate any other reasons why | |
615 | + the executable file might be covered by the GNU General Public License. */ | |
616 | + | |
617 | +/* This implements IEEE 754 format arithmetic, but does not provide a | |
618 | + mechanism for setting the rounding mode, or for generating or handling | |
619 | + exceptions. | |
620 | + | |
621 | + The original code by Steve Chamberlain, hacked by Mark Eichin and Jim | |
622 | + Wilson, all of Cygnus Support. */ | |
623 | + | |
624 | +/* The intended way to use this file is to make two copies, add `#define FLOAT' | |
625 | + to one copy, then compile both copies and add them to libgcc.a. */ | |
626 | + | |
627 | +#include "tconfig.h" | |
628 | +#include "coretypes.h" | |
629 | +#include "tm.h" | |
630 | +#include "config/fp-bit.h" | |
631 | + | |
632 | +/* The following macros can be defined to change the behavior of this file: | |
633 | + FLOAT: Implement a `float', aka SFmode, fp library. If this is not | |
634 | + defined, then this file implements a `double', aka DFmode, fp library. | |
635 | + FLOAT_ONLY: Used with FLOAT, to implement a `float' only library, i.e. | |
636 | + don't include float->double conversion which requires the double library. | |
637 | + This is useful only for machines which can't support doubles, e.g. some | |
638 | + 8-bit processors. | |
639 | + CMPtype: Specify the type that floating point compares should return. | |
640 | + This defaults to SItype, aka int. | |
641 | + US_SOFTWARE_GOFAST: This makes all entry points use the same names as the | |
642 | + US Software goFast library. | |
643 | + _DEBUG_BITFLOAT: This makes debugging the code a little easier, by adding | |
644 | + two integers to the FLO_union_type. | |
645 | + NO_DENORMALS: Disable handling of denormals. | |
646 | + NO_NANS: Disable nan and infinity handling | |
647 | + SMALL_MACHINE: Useful when operations on QIs and HIs are faster | |
648 | + than on an SI */ | |
649 | + | |
650 | +/* We don't currently support extended floats (long doubles) on machines | |
651 | + without hardware to deal with them. | |
652 | + | |
653 | + These stubs are just to keep the linker from complaining about unresolved | |
654 | + references which can be pulled in from libio & libstdc++, even if the | |
655 | + user isn't using long doubles. However, they may generate an unresolved | |
656 | + external to abort if abort is not used by the function, and the stubs | |
657 | + are referenced from within libc, since libgcc goes before and after the | |
658 | + system library. */ | |
659 | + | |
660 | +#ifdef DECLARE_LIBRARY_RENAMES | |
661 | + DECLARE_LIBRARY_RENAMES | |
662 | +#endif | |
663 | + | |
664 | +#ifdef EXTENDED_FLOAT_STUBS | |
665 | +extern void abort (void); | |
666 | +void __extendsfxf2 (void) { abort(); } | |
667 | +void __extenddfxf2 (void) { abort(); } | |
668 | +void __truncxfdf2 (void) { abort(); } | |
669 | +void __truncxfsf2 (void) { abort(); } | |
670 | +void __fixxfsi (void) { abort(); } | |
671 | +void __floatsixf (void) { abort(); } | |
672 | +void __addxf3 (void) { abort(); } | |
673 | +void __subxf3 (void) { abort(); } | |
674 | +void __mulxf3 (void) { abort(); } | |
675 | +void __divxf3 (void) { abort(); } | |
676 | +void __negxf2 (void) { abort(); } | |
677 | +void __eqxf2 (void) { abort(); } | |
678 | +void __nexf2 (void) { abort(); } | |
679 | +void __gtxf2 (void) { abort(); } | |
680 | +void __gexf2 (void) { abort(); } | |
681 | +void __lexf2 (void) { abort(); } | |
682 | +void __ltxf2 (void) { abort(); } | |
683 | + | |
684 | +void __extendsftf2 (void) { abort(); } | |
685 | +void __extenddftf2 (void) { abort(); } | |
686 | +void __trunctfdf2 (void) { abort(); } | |
687 | +void __trunctfsf2 (void) { abort(); } | |
688 | +void __fixtfsi (void) { abort(); } | |
689 | +void __floatsitf (void) { abort(); } | |
690 | +void __addtf3 (void) { abort(); } | |
691 | +void __subtf3 (void) { abort(); } | |
692 | +void __multf3 (void) { abort(); } | |
693 | +void __divtf3 (void) { abort(); } | |
694 | +void __negtf2 (void) { abort(); } | |
695 | +void __eqtf2 (void) { abort(); } | |
696 | +void __netf2 (void) { abort(); } | |
697 | +void __gttf2 (void) { abort(); } | |
698 | +void __getf2 (void) { abort(); } | |
699 | +void __letf2 (void) { abort(); } | |
700 | +void __lttf2 (void) { abort(); } | |
701 | +#else /* !EXTENDED_FLOAT_STUBS, rest of file */ | |
702 | + | |
703 | +/* IEEE "special" number predicates */ | |
704 | + | |
705 | +#ifdef NO_NANS | |
706 | + | |
707 | +#define nan() 0 | |
708 | +#define isnan(x) 0 | |
709 | +#define isinf(x) 0 | |
710 | +#else | |
711 | + | |
712 | +#if defined L_thenan_sf | |
713 | +const fp_number_type __thenan_sf = { CLASS_SNAN, 0, 0, {(fractype) 0} }; | |
714 | +#elif defined L_thenan_df | |
715 | +const fp_number_type __thenan_df = { CLASS_SNAN, 0, 0, {(fractype) 0} }; | |
716 | +#elif defined L_thenan_tf | |
717 | +const fp_number_type __thenan_tf = { CLASS_SNAN, 0, 0, {(fractype) 0} }; | |
718 | +#elif defined TFLOAT | |
719 | +extern const fp_number_type __thenan_tf; | |
720 | +#elif defined FLOAT | |
721 | +extern const fp_number_type __thenan_sf; | |
722 | +#else | |
723 | +extern const fp_number_type __thenan_df; | |
724 | +#endif | |
725 | + | |
726 | +INLINE | |
727 | +static fp_number_type * | |
728 | +nan (void) | |
729 | +{ | |
730 | + /* Discard the const qualifier... */ | |
731 | +#ifdef TFLOAT | |
732 | + return (fp_number_type *) (& __thenan_tf); | |
733 | +#elif defined FLOAT | |
734 | + return (fp_number_type *) (& __thenan_sf); | |
735 | +#else | |
736 | + return (fp_number_type *) (& __thenan_df); | |
737 | +#endif | |
738 | +} | |
739 | + | |
740 | +INLINE | |
741 | +static int | |
742 | +isnan ( fp_number_type * x) | |
743 | +{ | |
744 | + return x->class == CLASS_SNAN || x->class == CLASS_QNAN; | |
745 | +} | |
746 | + | |
747 | +INLINE | |
748 | +static int | |
749 | +isinf ( fp_number_type * x) | |
750 | +{ | |
751 | + return x->class == CLASS_INFINITY; | |
752 | +} | |
753 | + | |
754 | +#endif /* NO_NANS */ | |
755 | + | |
756 | +INLINE | |
757 | +static int | |
758 | +iszero ( fp_number_type * x) | |
759 | +{ | |
760 | + return x->class == CLASS_ZERO; | |
761 | +} | |
762 | + | |
763 | +INLINE | |
764 | +static void | |
765 | +flip_sign ( fp_number_type * x) | |
766 | +{ | |
767 | + x->sign = !x->sign; | |
768 | +} | |
769 | + | |
770 | +extern FLO_type pack_d ( fp_number_type * ); | |
771 | + | |
772 | +#if defined(L_pack_df) || defined(L_pack_sf) || defined(L_pack_tf) | |
773 | +FLO_type | |
774 | +pack_d ( fp_number_type * src) | |
775 | +{ | |
776 | + FLO_union_type dst; | |
777 | + fractype fraction = src->fraction.ll; /* wasn't unsigned before? */ | |
778 | + int sign = src->sign; | |
779 | + int exp = 0; | |
780 | + | |
781 | + if (LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) && (isnan (src) || isinf (src))) | |
782 | + { | |
783 | + /* We can't represent these values accurately. By using the | |
784 | + largest possible magnitude, we guarantee that the conversion | |
785 | + of infinity is at least as big as any finite number. */ | |
786 | + exp = EXPMAX; | |
787 | + fraction = ((fractype) 1 << FRACBITS) - 1; | |
788 | + } | |
789 | + else if (isnan (src)) | |
790 | + { | |
791 | + exp = EXPMAX; | |
792 | + if (src->class == CLASS_QNAN || 1) | |
793 | + { | |
794 | +#ifdef QUIET_NAN_NEGATED | |
795 | + fraction |= QUIET_NAN - 1; | |
796 | +#else | |
797 | + fraction |= QUIET_NAN; | |
798 | +#endif | |
799 | + } | |
800 | + } | |
801 | + else if (isinf (src)) | |
802 | + { | |
803 | + exp = EXPMAX; | |
804 | + fraction = 0; | |
805 | + } | |
806 | + else if (iszero (src)) | |
807 | + { | |
808 | + exp = 0; | |
809 | + fraction = 0; | |
810 | + } | |
811 | + else if (fraction == 0) | |
812 | + { | |
813 | + exp = 0; | |
814 | + } | |
815 | + else | |
816 | + { | |
817 | + if (src->normal_exp < NORMAL_EXPMIN) | |
818 | + { | |
819 | +#ifdef NO_DENORMALS | |
820 | + /* Go straight to a zero representation if denormals are not | |
821 | + supported. The denormal handling would be harmless but | |
822 | + isn't unnecessary. */ | |
823 | + exp = 0; | |
824 | + fraction = 0; | |
825 | +#else /* NO_DENORMALS */ | |
826 | + /* This number's exponent is too low to fit into the bits | |
827 | + available in the number, so we'll store 0 in the exponent and | |
828 | + shift the fraction to the right to make up for it. */ | |
829 | + | |
830 | + int shift = NORMAL_EXPMIN - src->normal_exp; | |
831 | + | |
832 | + exp = 0; | |
833 | + | |
834 | + if (shift > FRAC_NBITS - NGARDS) | |
835 | + { | |
836 | + /* No point shifting, since it's more that 64 out. */ | |
837 | + fraction = 0; | |
838 | + } | |
839 | + else | |
840 | + { | |
841 | + int lowbit = (fraction & (((fractype)1 << shift) - 1)) ? 1 : 0; | |
842 | + fraction = (fraction >> shift) | lowbit; | |
843 | + } | |
844 | + if ((fraction & GARDMASK) == GARDMSB) | |
845 | + { | |
846 | + if ((fraction & (1 << NGARDS))) | |
847 | + fraction += GARDROUND + 1; | |
848 | + } | |
849 | + else | |
850 | + { | |
851 | + /* Add to the guards to round up. */ | |
852 | + fraction += GARDROUND; | |
853 | + } | |
854 | + /* Perhaps the rounding means we now need to change the | |
855 | + exponent, because the fraction is no longer denormal. */ | |
856 | + if (fraction >= IMPLICIT_1) | |
857 | + { | |
858 | + exp += 1; | |
859 | + } | |
860 | + fraction >>= NGARDS; | |
861 | +#endif /* NO_DENORMALS */ | |
862 | + } | |
863 | + else if (!LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) | |
864 | + && src->normal_exp > EXPBIAS) | |
865 | + { | |
866 | + exp = EXPMAX; | |
867 | + fraction = 0; | |
868 | + } | |
869 | + else | |
870 | + { | |
871 | + exp = src->normal_exp + EXPBIAS; | |
872 | + if (!ROUND_TOWARDS_ZERO) | |
873 | + { | |
874 | + /* IF the gard bits are the all zero, but the first, then we're | |
875 | + half way between two numbers, choose the one which makes the | |
876 | + lsb of the answer 0. */ | |
877 | + if ((fraction & GARDMASK) == GARDMSB) | |
878 | + { | |
879 | + if (fraction & (1 << NGARDS)) | |
880 | + fraction += GARDROUND + 1; | |
881 | + } | |
882 | + else | |
883 | + { | |
884 | + /* Add a one to the guards to round up */ | |
885 | + fraction += GARDROUND; | |
886 | + } | |
887 | + if (fraction >= IMPLICIT_2) | |
888 | + { | |
889 | + fraction >>= 1; | |
890 | + exp += 1; | |
891 | + } | |
892 | + } | |
893 | + fraction >>= NGARDS; | |
894 | + | |
895 | + if (LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) && exp > EXPMAX) | |
896 | + { | |
897 | + /* Saturate on overflow. */ | |
898 | + exp = EXPMAX; | |
899 | + fraction = ((fractype) 1 << FRACBITS) - 1; | |
900 | + } | |
901 | + } | |
902 | + } | |
903 | + | |
904 | + /* We previously used bitfields to store the number, but this doesn't | |
905 | + handle little/big endian systems conveniently, so use shifts and | |
906 | + masks */ | |
907 | +#ifdef FLOAT_BIT_ORDER_MISMATCH | |
908 | + dst.bits.fraction = fraction; | |
909 | + dst.bits.exp = exp; | |
910 | + dst.bits.sign = sign; | |
911 | +#else | |
912 | +# if defined TFLOAT && defined HALFFRACBITS | |
913 | + { | |
914 | + halffractype high, low, unity; | |
915 | + int lowsign, lowexp; | |
916 | + | |
917 | + unity = (halffractype) 1 << HALFFRACBITS; | |
918 | + | |
919 | + /* Set HIGH to the high double's significand, masking out the implicit 1. | |
920 | + Set LOW to the low double's full significand. */ | |
921 | + high = (fraction >> (FRACBITS - HALFFRACBITS)) & (unity - 1); | |
922 | + low = fraction & (unity * 2 - 1); | |
923 | + | |
924 | + /* Get the initial sign and exponent of the low double. */ | |
925 | + lowexp = exp - HALFFRACBITS - 1; | |
926 | + lowsign = sign; | |
927 | + | |
928 | + /* HIGH should be rounded like a normal double, making |LOW| <= | |
929 | + 0.5 ULP of HIGH. Assume round-to-nearest. */ | |
930 | + if (exp < EXPMAX) | |
931 | + if (low > unity || (low == unity && (high & 1) == 1)) | |
932 | + { | |
933 | + /* Round HIGH up and adjust LOW to match. */ | |
934 | + high++; | |
935 | + if (high == unity) | |
936 | + { | |
937 | + /* May make it infinite, but that's OK. */ | |
938 | + high = 0; | |
939 | + exp++; | |
940 | + } | |
941 | + low = unity * 2 - low; | |
942 | + lowsign ^= 1; | |
943 | + } | |
944 | + | |
945 | + high |= (halffractype) exp << HALFFRACBITS; | |
946 | + high |= (halffractype) sign << (HALFFRACBITS + EXPBITS); | |
947 | + | |
948 | + if (exp == EXPMAX || exp == 0 || low == 0) | |
949 | + low = 0; | |
950 | + else | |
951 | + { | |
952 | + while (lowexp > 0 && low < unity) | |
953 | + { | |
954 | + low <<= 1; | |
955 | + lowexp--; | |
956 | + } | |
957 | + | |
958 | + if (lowexp <= 0) | |
959 | + { | |
960 | + halffractype roundmsb, round; | |
961 | + int shift; | |
962 | + | |
963 | + shift = 1 - lowexp; | |
964 | + roundmsb = (1 << (shift - 1)); | |
965 | + round = low & ((roundmsb << 1) - 1); | |
966 | + | |
967 | + low >>= shift; | |
968 | + lowexp = 0; | |
969 | + | |
970 | + if (round > roundmsb || (round == roundmsb && (low & 1) == 1)) | |
971 | + { | |
972 | + low++; | |
973 | + if (low == unity) | |
974 | + /* LOW rounds up to the smallest normal number. */ | |
975 | + lowexp++; | |
976 | + } | |
977 | + } | |
978 | + | |
979 | + low &= unity - 1; | |
980 | + low |= (halffractype) lowexp << HALFFRACBITS; | |
981 | + low |= (halffractype) lowsign << (HALFFRACBITS + EXPBITS); | |
982 | + } | |
983 | + dst.value_raw = ((fractype) high << HALFSHIFT) | low; | |
984 | + } | |
985 | +# else | |
986 | + dst.value_raw = fraction & ((((fractype)1) << FRACBITS) - (fractype)1); | |
987 | + dst.value_raw |= ((fractype) (exp & ((1 << EXPBITS) - 1))) << FRACBITS; | |
988 | + dst.value_raw |= ((fractype) (sign & 1)) << (FRACBITS | EXPBITS); | |
989 | +# endif | |
990 | +#endif | |
991 | + | |
992 | +#if defined(FLOAT_WORD_ORDER_MISMATCH) && !defined(FLOAT) | |
993 | +#ifdef TFLOAT | |
994 | + { | |
995 | + qrtrfractype tmp1 = dst.words[0]; | |
996 | + qrtrfractype tmp2 = dst.words[1]; | |
997 | + dst.words[0] = dst.words[3]; | |
998 | + dst.words[1] = dst.words[2]; | |
999 | + dst.words[2] = tmp2; | |
1000 | + dst.words[3] = tmp1; | |
1001 | + } | |
1002 | +#else | |
1003 | + { | |
1004 | + halffractype tmp = dst.words[0]; | |
1005 | + dst.words[0] = dst.words[1]; | |
1006 | + dst.words[1] = tmp; | |
1007 | + } | |
1008 | +#endif | |
1009 | +#endif | |
1010 | + | |
1011 | + return dst.value; | |
1012 | +} | |
1013 | +#endif | |
1014 | + | |
1015 | +#if defined(L_unpack_df) || defined(L_unpack_sf) || defined(L_unpack_tf) | |
1016 | +void | |
1017 | +unpack_d (FLO_union_type * src, fp_number_type * dst) | |
1018 | +{ | |
1019 | + /* We previously used bitfields to store the number, but this doesn't | |
1020 | + handle little/big endian systems conveniently, so use shifts and | |
1021 | + masks */ | |
1022 | + fractype fraction; | |
1023 | + int exp; | |
1024 | + int sign; | |
1025 | + | |
1026 | +#if defined(FLOAT_WORD_ORDER_MISMATCH) && !defined(FLOAT) | |
1027 | + FLO_union_type swapped; | |
1028 | + | |
1029 | +#ifdef TFLOAT | |
1030 | + swapped.words[0] = src->words[3]; | |
1031 | + swapped.words[1] = src->words[2]; | |
1032 | + swapped.words[2] = src->words[1]; | |
1033 | + swapped.words[3] = src->words[0]; | |
1034 | +#else | |
1035 | + swapped.words[0] = src->words[1]; | |
1036 | + swapped.words[1] = src->words[0]; | |
1037 | +#endif | |
1038 | + src = &swapped; | |
1039 | +#endif | |
1040 | + | |
1041 | +#ifdef FLOAT_BIT_ORDER_MISMATCH | |
1042 | + fraction = src->bits.fraction; | |
1043 | + exp = src->bits.exp; | |
1044 | + sign = src->bits.sign; | |
1045 | +#else | |
1046 | +# if defined TFLOAT && defined HALFFRACBITS | |
1047 | + { | |
1048 | + halffractype high, low; | |
1049 | + | |
1050 | + high = src->value_raw >> HALFSHIFT; | |
1051 | + low = src->value_raw & (((fractype)1 << HALFSHIFT) - 1); | |
1052 | + | |
1053 | + fraction = high & ((((fractype)1) << HALFFRACBITS) - 1); | |
1054 | + fraction <<= FRACBITS - HALFFRACBITS; | |
1055 | + exp = ((int)(high >> HALFFRACBITS)) & ((1 << EXPBITS) - 1); | |
1056 | + sign = ((int)(high >> (((HALFFRACBITS + EXPBITS))))) & 1; | |
1057 | + | |
1058 | + if (exp != EXPMAX && exp != 0 && low != 0) | |
1059 | + { | |
1060 | + int lowexp = ((int)(low >> HALFFRACBITS)) & ((1 << EXPBITS) - 1); | |
1061 | + int lowsign = ((int)(low >> (((HALFFRACBITS + EXPBITS))))) & 1; | |
1062 | + int shift; | |
1063 | + fractype xlow; | |
1064 | + | |
1065 | + xlow = low & ((((fractype)1) << HALFFRACBITS) - 1); | |
1066 | + if (lowexp) | |
1067 | + xlow |= (((halffractype)1) << HALFFRACBITS); | |
1068 | + else | |
1069 | + lowexp = 1; | |
1070 | + shift = (FRACBITS - HALFFRACBITS) - (exp - lowexp); | |
1071 | + if (shift > 0) | |
1072 | + xlow <<= shift; | |
1073 | + else if (shift < 0) | |
1074 | + xlow >>= -shift; | |
1075 | + if (sign == lowsign) | |
1076 | + fraction += xlow; | |
1077 | + else if (fraction >= xlow) | |
1078 | + fraction -= xlow; | |
1079 | + else | |
1080 | + { | |
1081 | + /* The high part is a power of two but the full number is lower. | |
1082 | + This code will leave the implicit 1 in FRACTION, but we'd | |
1083 | + have added that below anyway. */ | |
1084 | + fraction = (((fractype) 1 << FRACBITS) - xlow) << 1; | |
1085 | + exp--; | |
1086 | + } | |
1087 | + } | |
1088 | + } | |
1089 | +# else | |
1090 | + fraction = src->value_raw & ((((fractype)1) << FRACBITS) - 1); | |
1091 | + exp = ((int)(src->value_raw >> FRACBITS)) & ((1 << EXPBITS) - 1); | |
1092 | + sign = ((int)(src->value_raw >> (FRACBITS + EXPBITS))) & 1; | |
1093 | +# endif | |
1094 | +#endif | |
1095 | + | |
1096 | + dst->sign = sign; | |
1097 | + if (exp == 0) | |
1098 | + { | |
1099 | + /* Hmm. Looks like 0 */ | |
1100 | + if (fraction == 0 | |
1101 | +#ifdef NO_DENORMALS | |
1102 | + || 1 | |
1103 | +#endif | |
1104 | + ) | |
1105 | + { | |
1106 | + /* tastes like zero */ | |
1107 | + dst->class = CLASS_ZERO; | |
1108 | + } | |
1109 | + else | |
1110 | + { | |
1111 | + /* Zero exponent with nonzero fraction - it's denormalized, | |
1112 | + so there isn't a leading implicit one - we'll shift it so | |
1113 | + it gets one. */ | |
1114 | + dst->normal_exp = exp - EXPBIAS + 1; | |
1115 | + fraction <<= NGARDS; | |
1116 | + | |
1117 | + dst->class = CLASS_NUMBER; | |
1118 | +#if 1 | |
1119 | + while (fraction < IMPLICIT_1) | |
1120 | + { | |
1121 | + fraction <<= 1; | |
1122 | + dst->normal_exp--; | |
1123 | + } | |
1124 | +#endif | |
1125 | + dst->fraction.ll = fraction; | |
1126 | + } | |
1127 | + } | |
1128 | + else if (!LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) && exp == EXPMAX) | |
1129 | + { | |
1130 | + /* Huge exponent*/ | |
1131 | + if (fraction == 0) | |
1132 | + { | |
1133 | + /* Attached to a zero fraction - means infinity */ | |
1134 | + dst->class = CLASS_INFINITY; | |
1135 | + } | |
1136 | + else | |
1137 | + { | |
1138 | + /* Nonzero fraction, means nan */ | |
1139 | +#ifdef QUIET_NAN_NEGATED | |
1140 | + if ((fraction & QUIET_NAN) == 0) | |
1141 | +#else | |
1142 | + if (fraction & QUIET_NAN) | |
1143 | +#endif | |
1144 | + { | |
1145 | + dst->class = CLASS_QNAN; | |
1146 | + } | |
1147 | + else | |
1148 | + { | |
1149 | + dst->class = CLASS_SNAN; | |
1150 | + } | |
1151 | + /* Keep the fraction part as the nan number */ | |
1152 | + dst->fraction.ll = fraction; | |
1153 | + } | |
1154 | + } | |
1155 | + else | |
1156 | + { | |
1157 | + /* Nothing strange about this number */ | |
1158 | + dst->normal_exp = exp - EXPBIAS; | |
1159 | + dst->class = CLASS_NUMBER; | |
1160 | + dst->fraction.ll = (fraction << NGARDS) | IMPLICIT_1; | |
1161 | + } | |
1162 | +} | |
1163 | +#endif /* L_unpack_df || L_unpack_sf */ | |
1164 | + | |
1165 | +#if defined(L_addsub_sf) || defined(L_addsub_df) || defined(L_addsub_tf) | |
1166 | +static fp_number_type * | |
1167 | +_fpadd_parts (fp_number_type * a, | |
1168 | + fp_number_type * b, | |
1169 | + fp_number_type * tmp) | |
1170 | +{ | |
1171 | + intfrac tfraction; | |
1172 | + | |
1173 | + /* Put commonly used fields in local variables. */ | |
1174 | + int a_normal_exp; | |
1175 | + int b_normal_exp; | |
1176 | + fractype a_fraction; | |
1177 | + fractype b_fraction; | |
1178 | + | |
1179 | + if (isnan (a)) | |
1180 | + { | |
1181 | + return a; | |
1182 | + } | |
1183 | + if (isnan (b)) | |
1184 | + { | |
1185 | + return b; | |
1186 | + } | |
1187 | + if (isinf (a)) | |
1188 | + { | |
1189 | + /* Adding infinities with opposite signs yields a NaN. */ | |
1190 | + if (isinf (b) && a->sign != b->sign) | |
1191 | + return nan (); | |
1192 | + return a; | |
1193 | + } | |
1194 | + if (isinf (b)) | |
1195 | + { | |
1196 | + return b; | |
1197 | + } | |
1198 | + if (iszero (b)) | |
1199 | + { | |
1200 | + if (iszero (a)) | |
1201 | + { | |
1202 | + *tmp = *a; | |
1203 | + tmp->sign = a->sign & b->sign; | |
1204 | + return tmp; | |
1205 | + } | |
1206 | + return a; | |
1207 | + } | |
1208 | + if (iszero (a)) | |
1209 | + { | |
1210 | + return b; | |
1211 | + } | |
1212 | + | |
1213 | + /* Got two numbers. shift the smaller and increment the exponent till | |
1214 | + they're the same */ | |
1215 | + { | |
1216 | + int diff; | |
1217 | + | |
1218 | + a_normal_exp = a->normal_exp; | |
1219 | + b_normal_exp = b->normal_exp; | |
1220 | + a_fraction = a->fraction.ll; | |
1221 | + b_fraction = b->fraction.ll; | |
1222 | + | |
1223 | + diff = a_normal_exp - b_normal_exp; | |
1224 | + | |
1225 | + if (diff < 0) | |
1226 | + diff = -diff; | |
1227 | + if (diff < FRAC_NBITS) | |
1228 | + { | |
1229 | + /* ??? This does shifts one bit at a time. Optimize. */ | |
1230 | + while (a_normal_exp > b_normal_exp) | |
1231 | + { | |
1232 | + b_normal_exp++; | |
1233 | + LSHIFT (b_fraction); | |
1234 | + } | |
1235 | + while (b_normal_exp > a_normal_exp) | |
1236 | + { | |
1237 | + a_normal_exp++; | |
1238 | + LSHIFT (a_fraction); | |
1239 | + } | |
1240 | + } | |
1241 | + else | |
1242 | + { | |
1243 | + /* Somethings's up.. choose the biggest */ | |
1244 | + if (a_normal_exp > b_normal_exp) | |
1245 | + { | |
1246 | + b_normal_exp = a_normal_exp; | |
1247 | + b_fraction = 0; | |
1248 | + } | |
1249 | + else | |
1250 | + { | |
1251 | + a_normal_exp = b_normal_exp; | |
1252 | + a_fraction = 0; | |
1253 | + } | |
1254 | + } | |
1255 | + } | |
1256 | + | |
1257 | + if (a->sign != b->sign) | |
1258 | + { | |
1259 | + if (a->sign) | |
1260 | + { | |
1261 | + tfraction = -a_fraction + b_fraction; | |
1262 | + } | |
1263 | + else | |
1264 | + { | |
1265 | + tfraction = a_fraction - b_fraction; | |
1266 | + } | |
1267 | + if (tfraction >= 0) | |
1268 | + { | |
1269 | + tmp->sign = 0; | |
1270 | + tmp->normal_exp = a_normal_exp; | |
1271 | + tmp->fraction.ll = tfraction; | |
1272 | + } | |
1273 | + else | |
1274 | + { | |
1275 | + tmp->sign = 1; | |
1276 | + tmp->normal_exp = a_normal_exp; | |
1277 | + tmp->fraction.ll = -tfraction; | |
1278 | + } | |
1279 | + /* and renormalize it */ | |
1280 | + | |
1281 | + while (tmp->fraction.ll < IMPLICIT_1 && tmp->fraction.ll) | |
1282 | + { | |
1283 | + tmp->fraction.ll <<= 1; | |
1284 | + tmp->normal_exp--; | |
1285 | + } | |
1286 | + } | |
1287 | + else | |
1288 | + { | |
1289 | + tmp->sign = a->sign; | |
1290 | + tmp->normal_exp = a_normal_exp; | |
1291 | + tmp->fraction.ll = a_fraction + b_fraction; | |
1292 | + } | |
1293 | + tmp->class = CLASS_NUMBER; | |
1294 | + /* Now the fraction is added, we have to shift down to renormalize the | |
1295 | + number */ | |
1296 | + | |
1297 | + if (tmp->fraction.ll >= IMPLICIT_2) | |
1298 | + { | |
1299 | + LSHIFT (tmp->fraction.ll); | |
1300 | + tmp->normal_exp++; | |
1301 | + } | |
1302 | + return tmp; | |
1303 | + | |
1304 | +} | |
1305 | + | |
1306 | +FLO_type | |
1307 | +add (FLO_type arg_a, FLO_type arg_b) | |
1308 | +{ | |
1309 | + fp_number_type a; | |
1310 | + fp_number_type b; | |
1311 | + fp_number_type tmp; | |
1312 | + fp_number_type *res; | |
1313 | + FLO_union_type au, bu; | |
1314 | + | |
1315 | + au.value = arg_a; | |
1316 | + bu.value = arg_b; | |
1317 | + | |
1318 | + unpack_d (&au, &a); | |
1319 | + unpack_d (&bu, &b); | |
1320 | + | |
1321 | + res = _fpadd_parts (&a, &b, &tmp); | |
1322 | + | |
1323 | + return pack_d (res); | |
1324 | +} | |
1325 | + | |
1326 | +FLO_type | |
1327 | +sub (FLO_type arg_a, FLO_type arg_b) | |
1328 | +{ | |
1329 | + fp_number_type a; | |
1330 | + fp_number_type b; | |
1331 | + fp_number_type tmp; | |
1332 | + fp_number_type *res; | |
1333 | + FLO_union_type au, bu; | |
1334 | + | |
1335 | + au.value = arg_a; | |
1336 | + bu.value = arg_b; | |
1337 | + | |
1338 | + unpack_d (&au, &a); | |
1339 | + unpack_d (&bu, &b); | |
1340 | + | |
1341 | + b.sign ^= 1; | |
1342 | + | |
1343 | + res = _fpadd_parts (&a, &b, &tmp); | |
1344 | + | |
1345 | + return pack_d (res); | |
1346 | +} | |
1347 | +#endif /* L_addsub_sf || L_addsub_df */ | |
1348 | + | |
1349 | +#if defined(L_mul_sf) || defined(L_mul_df) || defined(L_mul_tf) | |
1350 | +static inline __attribute__ ((__always_inline__)) fp_number_type * | |
1351 | +_fpmul_parts ( fp_number_type * a, | |
1352 | + fp_number_type * b, | |
1353 | + fp_number_type * tmp) | |
1354 | +{ | |
1355 | + fractype low = 0; | |
1356 | + fractype high = 0; | |
1357 | + | |
1358 | + if (isnan (a)) | |
1359 | + { | |
1360 | + a->sign = a->sign != b->sign; | |
1361 | + return a; | |
1362 | + } | |
1363 | + if (isnan (b)) | |
1364 | + { | |
1365 | + b->sign = a->sign != b->sign; | |
1366 | + return b; | |
1367 | + } | |
1368 | + if (isinf (a)) | |
1369 | + { | |
1370 | + if (iszero (b)) | |
1371 | + return nan (); | |
1372 | + a->sign = a->sign != b->sign; | |
1373 | + return a; | |
1374 | + } | |
1375 | + if (isinf (b)) | |
1376 | + { | |
1377 | + if (iszero (a)) | |
1378 | + { | |
1379 | + return nan (); | |
1380 | + } | |
1381 | + b->sign = a->sign != b->sign; | |
1382 | + return b; | |
1383 | + } | |
1384 | + if (iszero (a)) | |
1385 | + { | |
1386 | + a->sign = a->sign != b->sign; | |
1387 | + return a; | |
1388 | + } | |
1389 | + if (iszero (b)) | |
1390 | + { | |
1391 | + b->sign = a->sign != b->sign; | |
1392 | + return b; | |
1393 | + } | |
1394 | + | |
1395 | + /* Calculate the mantissa by multiplying both numbers to get a | |
1396 | + twice-as-wide number. */ | |
1397 | + { | |
1398 | +#if defined(NO_DI_MODE) || defined(TFLOAT) | |
1399 | + { | |
1400 | + fractype x = a->fraction.ll; | |
1401 | + fractype ylow = b->fraction.ll; | |
1402 | + fractype yhigh = 0; | |
1403 | + int bit; | |
1404 | + | |
1405 | + /* ??? This does multiplies one bit at a time. Optimize. */ | |
1406 | + for (bit = 0; bit < FRAC_NBITS; bit++) | |
1407 | + { | |
1408 | + int carry; | |
1409 | + | |
1410 | + if (x & 1) | |
1411 | + { | |
1412 | + carry = (low += ylow) < ylow; | |
1413 | + high += yhigh + carry; | |
1414 | + } | |
1415 | + yhigh <<= 1; | |
1416 | + if (ylow & FRACHIGH) | |
1417 | + { | |
1418 | + yhigh |= 1; | |
1419 | + } | |
1420 | + ylow <<= 1; | |
1421 | + x >>= 1; | |
1422 | + } | |
1423 | + } | |
1424 | +#elif defined(FLOAT) | |
1425 | + /* Multiplying two USIs to get a UDI, we're safe. */ | |
1426 | + { | |
1427 | + UDItype answer = (UDItype)a->fraction.ll * (UDItype)b->fraction.ll; | |
1428 | + | |
1429 | + high = answer >> BITS_PER_SI; | |
1430 | + low = answer; | |
1431 | + } | |
1432 | +#else | |
1433 | + /* fractype is DImode, but we need the result to be twice as wide. | |
1434 | + Assuming a widening multiply from DImode to TImode is not | |
1435 | + available, build one by hand. */ | |
1436 | + { | |
1437 | + USItype nl = a->fraction.ll; | |
1438 | + USItype nh = a->fraction.ll >> BITS_PER_SI; | |
1439 | + USItype ml = b->fraction.ll; | |
1440 | + USItype mh = b->fraction.ll >> BITS_PER_SI; | |
1441 | + UDItype pp_ll = (UDItype) ml * nl; | |
1442 | + UDItype pp_hl = (UDItype) mh * nl; | |
1443 | + UDItype pp_lh = (UDItype) ml * nh; | |
1444 | + UDItype pp_hh = (UDItype) mh * nh; | |
1445 | + UDItype res2 = 0; | |
1446 | + UDItype res0 = 0; | |
1447 | + UDItype ps_hh__ = pp_hl + pp_lh; | |
1448 | + if (ps_hh__ < pp_hl) | |
1449 | + res2 += (UDItype)1 << BITS_PER_SI; | |
1450 | + pp_hl = (UDItype)(USItype)ps_hh__ << BITS_PER_SI; | |
1451 | + res0 = pp_ll + pp_hl; | |
1452 | + if (res0 < pp_ll) | |
1453 | + res2++; | |
1454 | + res2 += (ps_hh__ >> BITS_PER_SI) + pp_hh; | |
1455 | + high = res2; | |
1456 | + low = res0; | |
1457 | + } | |
1458 | +#endif | |
1459 | + } | |
1460 | + | |
1461 | + tmp->normal_exp = a->normal_exp + b->normal_exp | |
1462 | + + FRAC_NBITS - (FRACBITS + NGARDS); | |
1463 | + tmp->sign = a->sign != b->sign; | |
1464 | + while (high >= IMPLICIT_2) | |
1465 | + { | |
1466 | + tmp->normal_exp++; | |
1467 | + if (high & 1) | |
1468 | + { | |
1469 | + low >>= 1; | |
1470 | + low |= FRACHIGH; | |
1471 | + } | |
1472 | + high >>= 1; | |
1473 | + } | |
1474 | + while (high < IMPLICIT_1) | |
1475 | + { | |
1476 | + tmp->normal_exp--; | |
1477 | + | |
1478 | + high <<= 1; | |
1479 | + if (low & FRACHIGH) | |
1480 | + high |= 1; | |
1481 | + low <<= 1; | |
1482 | + } | |
1483 | + /* rounding is tricky. if we only round if it won't make us round later. */ | |
1484 | +#if 0 | |
1485 | + if (low & FRACHIGH2) | |
1486 | + { | |
1487 | + if (((high & GARDMASK) != GARDMSB) | |
1488 | + && (((high + 1) & GARDMASK) == GARDMSB)) | |
1489 | + { | |
1490 | + /* don't round, it gets done again later. */ | |
1491 | + } | |
1492 | + else | |
1493 | + { | |
1494 | + high++; | |
1495 | + } | |
1496 | + } | |
1497 | +#endif | |
1498 | + if (!ROUND_TOWARDS_ZERO && (high & GARDMASK) == GARDMSB) | |
1499 | + { | |
1500 | + if (high & (1 << NGARDS)) | |
1501 | + { | |
1502 | + /* half way, so round to even */ | |
1503 | + high += GARDROUND + 1; | |
1504 | + } | |
1505 | + else if (low) | |
1506 | + { | |
1507 | + /* but we really weren't half way */ | |
1508 | + high += GARDROUND + 1; | |
1509 | + } | |
1510 | + } | |
1511 | + tmp->fraction.ll = high; | |
1512 | + tmp->class = CLASS_NUMBER; | |
1513 | + return tmp; | |
1514 | +} | |
1515 | + | |
1516 | +FLO_type | |
1517 | +multiply (FLO_type arg_a, FLO_type arg_b) | |
1518 | +{ | |
1519 | + fp_number_type a; | |
1520 | + fp_number_type b; | |
1521 | + fp_number_type tmp; | |
1522 | + fp_number_type *res; | |
1523 | + FLO_union_type au, bu; | |
1524 | + | |
1525 | + au.value = arg_a; | |
1526 | + bu.value = arg_b; | |
1527 | + | |
1528 | + unpack_d (&au, &a); | |
1529 | + unpack_d (&bu, &b); | |
1530 | + | |
1531 | + res = _fpmul_parts (&a, &b, &tmp); | |
1532 | + | |
1533 | + return pack_d (res); | |
1534 | +} | |
1535 | +#endif /* L_mul_sf || L_mul_df */ | |
1536 | + | |
1537 | +#if defined(L_div_sf) || defined(L_div_df) || defined(L_div_tf) | |
1538 | +static inline __attribute__ ((__always_inline__)) fp_number_type * | |
1539 | +_fpdiv_parts (fp_number_type * a, | |
1540 | + fp_number_type * b) | |
1541 | +{ | |
1542 | + fractype bit; | |
1543 | + fractype numerator; | |
1544 | + fractype denominator; | |
1545 | + fractype quotient; | |
1546 | + | |
1547 | + if (isnan (a)) | |
1548 | + { | |
1549 | + return a; | |
1550 | + } | |
1551 | + if (isnan (b)) | |
1552 | + { | |
1553 | + return b; | |
1554 | + } | |
1555 | + | |
1556 | + a->sign = a->sign ^ b->sign; | |
1557 | + | |
1558 | + if (isinf (a) || iszero (a)) | |
1559 | + { | |
1560 | + if (a->class == b->class) | |
1561 | + return nan (); | |
1562 | + return a; | |
1563 | + } | |
1564 | + | |
1565 | + if (isinf (b)) | |
1566 | + { | |
1567 | + a->fraction.ll = 0; | |
1568 | + a->normal_exp = 0; | |
1569 | + return a; | |
1570 | + } | |
1571 | + if (iszero (b)) | |
1572 | + { | |
1573 | + a->class = CLASS_INFINITY; | |
1574 | + return a; | |
1575 | + } | |
1576 | + | |
1577 | + /* Calculate the mantissa by multiplying both 64bit numbers to get a | |
1578 | + 128 bit number */ | |
1579 | + { | |
1580 | + /* quotient = | |
1581 | + ( numerator / denominator) * 2^(numerator exponent - denominator exponent) | |
1582 | + */ | |
1583 | + | |
1584 | + a->normal_exp = a->normal_exp - b->normal_exp; | |
1585 | + numerator = a->fraction.ll; | |
1586 | + denominator = b->fraction.ll; | |
1587 | + | |
1588 | + if (numerator < denominator) | |
1589 | + { | |
1590 | + /* Fraction will be less than 1.0 */ | |
1591 | + numerator *= 2; | |
1592 | + a->normal_exp--; | |
1593 | + } | |
1594 | + bit = IMPLICIT_1; | |
1595 | + quotient = 0; | |
1596 | + /* ??? Does divide one bit at a time. Optimize. */ | |
1597 | + while (bit) | |
1598 | + { | |
1599 | + if (numerator >= denominator) | |
1600 | + { | |
1601 | + quotient |= bit; | |
1602 | + numerator -= denominator; | |
1603 | + } | |
1604 | + bit >>= 1; | |
1605 | + numerator *= 2; | |
1606 | + } | |
1607 | + | |
1608 | + if (!ROUND_TOWARDS_ZERO && (quotient & GARDMASK) == GARDMSB) | |
1609 | + { | |
1610 | + if (quotient & (1 << NGARDS)) | |
1611 | + { | |
1612 | + /* half way, so round to even */ | |
1613 | + quotient += GARDROUND + 1; | |
1614 | + } | |
1615 | + else if (numerator) | |
1616 | + { | |
1617 | + /* but we really weren't half way, more bits exist */ | |
1618 | + quotient += GARDROUND + 1; | |
1619 | + } | |
1620 | + } | |
1621 | + | |
1622 | + a->fraction.ll = quotient; | |
1623 | + return (a); | |
1624 | + } | |
1625 | +} | |
1626 | + | |
1627 | +FLO_type | |
1628 | +divide (FLO_type arg_a, FLO_type arg_b) | |
1629 | +{ | |
1630 | + fp_number_type a; | |
1631 | + fp_number_type b; | |
1632 | + fp_number_type *res; | |
1633 | + FLO_union_type au, bu; | |
1634 | + | |
1635 | + au.value = arg_a; | |
1636 | + bu.value = arg_b; | |
1637 | + | |
1638 | + unpack_d (&au, &a); | |
1639 | + unpack_d (&bu, &b); | |
1640 | + | |
1641 | + res = _fpdiv_parts (&a, &b); | |
1642 | + | |
1643 | + return pack_d (res); | |
1644 | +} | |
1645 | +#endif /* L_div_sf || L_div_df */ | |
1646 | + | |
1647 | +#if defined(L_fpcmp_parts_sf) || defined(L_fpcmp_parts_df) \ | |
1648 | + || defined(L_fpcmp_parts_tf) | |
1649 | +/* according to the demo, fpcmp returns a comparison with 0... thus | |
1650 | + a<b -> -1 | |
1651 | + a==b -> 0 | |
1652 | + a>b -> +1 | |
1653 | + */ | |
1654 | + | |
1655 | +int | |
1656 | +__fpcmp_parts (fp_number_type * a, fp_number_type * b) | |
1657 | +{ | |
1658 | +#if 0 | |
1659 | + /* either nan -> unordered. Must be checked outside of this routine. */ | |
1660 | + if (isnan (a) && isnan (b)) | |
1661 | + { | |
1662 | + return 1; /* still unordered! */ | |
1663 | + } | |
1664 | +#endif | |
1665 | + | |
1666 | + if (isnan (a) || isnan (b)) | |
1667 | + { | |
1668 | + return 1; /* how to indicate unordered compare? */ | |
1669 | + } | |
1670 | + if (isinf (a) && isinf (b)) | |
1671 | + { | |
1672 | + /* +inf > -inf, but +inf != +inf */ | |
1673 | + /* b \a| +inf(0)| -inf(1) | |
1674 | + ______\+--------+-------- | |
1675 | + +inf(0)| a==b(0)| a<b(-1) | |
1676 | + -------+--------+-------- | |
1677 | + -inf(1)| a>b(1) | a==b(0) | |
1678 | + -------+--------+-------- | |
1679 | + So since unordered must be nonzero, just line up the columns... | |
1680 | + */ | |
1681 | + return b->sign - a->sign; | |
1682 | + } | |
1683 | + /* but not both... */ | |
1684 | + if (isinf (a)) | |
1685 | + { | |
1686 | + return a->sign ? -1 : 1; | |
1687 | + } | |
1688 | + if (isinf (b)) | |
1689 | + { | |
1690 | + return b->sign ? 1 : -1; | |
1691 | + } | |
1692 | + if (iszero (a) && iszero (b)) | |
1693 | + { | |
1694 | + return 0; | |
1695 | + } | |
1696 | + if (iszero (a)) | |
1697 | + { | |
1698 | + return b->sign ? 1 : -1; | |
1699 | + } | |
1700 | + if (iszero (b)) | |
1701 | + { | |
1702 | + return a->sign ? -1 : 1; | |
1703 | + } | |
1704 | + /* now both are "normal". */ | |
1705 | + if (a->sign != b->sign) | |
1706 | + { | |
1707 | + /* opposite signs */ | |
1708 | + return a->sign ? -1 : 1; | |
1709 | + } | |
1710 | + /* same sign; exponents? */ | |
1711 | + if (a->normal_exp > b->normal_exp) | |
1712 | + { | |
1713 | + return a->sign ? -1 : 1; | |
1714 | + } | |
1715 | + if (a->normal_exp < b->normal_exp) | |
1716 | + { | |
1717 | + return a->sign ? 1 : -1; | |
1718 | + } | |
1719 | + /* same exponents; check size. */ | |
1720 | + if (a->fraction.ll > b->fraction.ll) | |
1721 | + { | |
1722 | + return a->sign ? -1 : 1; | |
1723 | + } | |
1724 | + if (a->fraction.ll < b->fraction.ll) | |
1725 | + { | |
1726 | + return a->sign ? 1 : -1; | |
1727 | + } | |
1728 | + /* after all that, they're equal. */ | |
1729 | + return 0; | |
1730 | +} | |
1731 | +#endif | |
1732 | + | |
1733 | +#if defined(L_compare_sf) || defined(L_compare_df) || defined(L_compoare_tf) | |
1734 | +CMPtype | |
1735 | +compare (FLO_type arg_a, FLO_type arg_b) | |
1736 | +{ | |
1737 | + fp_number_type a; | |
1738 | + fp_number_type b; | |
1739 | + FLO_union_type au, bu; | |
1740 | + | |
1741 | + au.value = arg_a; | |
1742 | + bu.value = arg_b; | |
1743 | + | |
1744 | + unpack_d (&au, &a); | |
1745 | + unpack_d (&bu, &b); | |
1746 | + | |
1747 | + return __fpcmp_parts (&a, &b); | |
1748 | +} | |
1749 | +#endif /* L_compare_sf || L_compare_df */ | |
1750 | + | |
1751 | +#ifndef US_SOFTWARE_GOFAST | |
1752 | + | |
1753 | +/* These should be optimized for their specific tasks someday. */ | |
1754 | + | |
1755 | +#if defined(L_eq_sf) || defined(L_eq_df) || defined(L_eq_tf) | |
1756 | +CMPtype | |
1757 | +_eq_f2 (FLO_type arg_a, FLO_type arg_b) | |
1758 | +{ | |
1759 | + fp_number_type a; | |
1760 | + fp_number_type b; | |
1761 | + FLO_union_type au, bu; | |
1762 | + | |
1763 | + au.value = arg_a; | |
1764 | + bu.value = arg_b; | |
1765 | + | |
1766 | + unpack_d (&au, &a); | |
1767 | + unpack_d (&bu, &b); | |
1768 | + | |
1769 | + if (isnan (&a) || isnan (&b)) | |
1770 | + return 1; /* false, truth == 0 */ | |
1771 | + | |
1772 | + return __fpcmp_parts (&a, &b) ; | |
1773 | +} | |
1774 | +#endif /* L_eq_sf || L_eq_df */ | |
1775 | + | |
1776 | +#if defined(L_ne_sf) || defined(L_ne_df) || defined(L_ne_tf) | |
1777 | +CMPtype | |
1778 | +_ne_f2 (FLO_type arg_a, FLO_type arg_b) | |
1779 | +{ | |
1780 | + fp_number_type a; | |
1781 | + fp_number_type b; | |
1782 | + FLO_union_type au, bu; | |
1783 | + | |
1784 | + au.value = arg_a; | |
1785 | + bu.value = arg_b; | |
1786 | + | |
1787 | + unpack_d (&au, &a); | |
1788 | + unpack_d (&bu, &b); | |
1789 | + | |
1790 | + if (isnan (&a) || isnan (&b)) | |
1791 | + return 1; /* true, truth != 0 */ | |
1792 | + | |
1793 | + return __fpcmp_parts (&a, &b) ; | |
1794 | +} | |
1795 | +#endif /* L_ne_sf || L_ne_df */ | |
1796 | + | |
1797 | +#if defined(L_gt_sf) || defined(L_gt_df) || defined(L_gt_tf) | |
1798 | +CMPtype | |
1799 | +_gt_f2 (FLO_type arg_a, FLO_type arg_b) | |
1800 | +{ | |
1801 | + fp_number_type a; | |
1802 | + fp_number_type b; | |
1803 | + FLO_union_type au, bu; | |
1804 | + | |
1805 | + au.value = arg_a; | |
1806 | + bu.value = arg_b; | |
1807 | + | |
1808 | + unpack_d (&au, &a); | |
1809 | + unpack_d (&bu, &b); | |
1810 | + | |
1811 | + if (isnan (&a) || isnan (&b)) | |
1812 | + return -1; /* false, truth > 0 */ | |
1813 | + | |
1814 | + return __fpcmp_parts (&a, &b); | |
1815 | +} | |
1816 | +#endif /* L_gt_sf || L_gt_df */ | |
1817 | + | |
1818 | +#if defined(L_ge_sf) || defined(L_ge_df) || defined(L_ge_tf) | |
1819 | +CMPtype | |
1820 | +_ge_f2 (FLO_type arg_a, FLO_type arg_b) | |
1821 | +{ | |
1822 | + fp_number_type a; | |
1823 | + fp_number_type b; | |
1824 | + FLO_union_type au, bu; | |
1825 | + | |
1826 | + au.value = arg_a; | |
1827 | + bu.value = arg_b; | |
1828 | + | |
1829 | + unpack_d (&au, &a); | |
1830 | + unpack_d (&bu, &b); | |
1831 | + | |
1832 | + if (isnan (&a) || isnan (&b)) | |
1833 | + return -1; /* false, truth >= 0 */ | |
1834 | + return __fpcmp_parts (&a, &b) ; | |
1835 | +} | |
1836 | +#endif /* L_ge_sf || L_ge_df */ | |
1837 | + | |
1838 | +#if defined(L_lt_sf) || defined(L_lt_df) || defined(L_lt_tf) | |
1839 | +CMPtype | |
1840 | +_lt_f2 (FLO_type arg_a, FLO_type arg_b) | |
1841 | +{ | |
1842 | + fp_number_type a; | |
1843 | + fp_number_type b; | |
1844 | + FLO_union_type au, bu; | |
1845 | + | |
1846 | + au.value = arg_a; | |
1847 | + bu.value = arg_b; | |
1848 | + | |
1849 | + unpack_d (&au, &a); | |
1850 | + unpack_d (&bu, &b); | |
1851 | + | |
1852 | + if (isnan (&a) || isnan (&b)) | |
1853 | + return 1; /* false, truth < 0 */ | |
1854 | + | |
1855 | + return __fpcmp_parts (&a, &b); | |
1856 | +} | |
1857 | +#endif /* L_lt_sf || L_lt_df */ | |
1858 | + | |
1859 | +#if defined(L_le_sf) || defined(L_le_df) || defined(L_le_tf) | |
1860 | +CMPtype | |
1861 | +_le_f2 (FLO_type arg_a, FLO_type arg_b) | |
1862 | +{ | |
1863 | + fp_number_type a; | |
1864 | + fp_number_type b; | |
1865 | + FLO_union_type au, bu; | |
1866 | + | |
1867 | + au.value = arg_a; | |
1868 | + bu.value = arg_b; | |
1869 | + | |
1870 | + unpack_d (&au, &a); | |
1871 | + unpack_d (&bu, &b); | |
1872 | + | |
1873 | + if (isnan (&a) || isnan (&b)) | |
1874 | + return 1; /* false, truth <= 0 */ | |
1875 | + | |
1876 | + return __fpcmp_parts (&a, &b) ; | |
1877 | +} | |
1878 | +#endif /* L_le_sf || L_le_df */ | |
1879 | + | |
1880 | +#endif /* ! US_SOFTWARE_GOFAST */ | |
1881 | + | |
1882 | +#if defined(L_unord_sf) || defined(L_unord_df) || defined(L_unord_tf) | |
1883 | +CMPtype | |
1884 | +_unord_f2 (FLO_type arg_a, FLO_type arg_b) | |
1885 | +{ | |
1886 | + fp_number_type a; | |
1887 | + fp_number_type b; | |
1888 | + FLO_union_type au, bu; | |
1889 | + | |
1890 | + au.value = arg_a; | |
1891 | + bu.value = arg_b; | |
1892 | + | |
1893 | + unpack_d (&au, &a); | |
1894 | + unpack_d (&bu, &b); | |
1895 | + | |
1896 | + return (isnan (&a) || isnan (&b)); | |
1897 | +} | |
1898 | +#endif /* L_unord_sf || L_unord_df */ | |
1899 | + | |
1900 | +#if defined(L_si_to_sf) || defined(L_si_to_df) || defined(L_si_to_tf) | |
1901 | +FLO_type | |
1902 | +si_to_float (SItype arg_a) | |
1903 | +{ | |
1904 | + fp_number_type in; | |
1905 | + | |
1906 | + in.class = CLASS_NUMBER; | |
1907 | + in.sign = arg_a < 0; | |
1908 | + if (!arg_a) | |
1909 | + { | |
1910 | + in.class = CLASS_ZERO; | |
1911 | + } | |
1912 | + else | |
1913 | + { | |
1914 | + in.normal_exp = FRACBITS + NGARDS; | |
1915 | + if (in.sign) | |
1916 | + { | |
1917 | + /* Special case for minint, since there is no +ve integer | |
1918 | + representation for it */ | |
1919 | + if (arg_a == (- MAX_SI_INT - 1)) | |
1920 | + { | |
1921 | + return (FLO_type)(- MAX_SI_INT - 1); | |
1922 | + } | |
1923 | + in.fraction.ll = (-arg_a); | |
1924 | + } | |
1925 | + else | |
1926 | + in.fraction.ll = arg_a; | |
1927 | + | |
1928 | + while (in.fraction.ll < ((fractype)1 << (FRACBITS + NGARDS))) | |
1929 | + { | |
1930 | + in.fraction.ll <<= 1; | |
1931 | + in.normal_exp -= 1; | |
1932 | + } | |
1933 | + } | |
1934 | + return pack_d (&in); | |
1935 | +} | |
1936 | +#endif /* L_si_to_sf || L_si_to_df */ | |
1937 | + | |
1938 | +#if defined(L_usi_to_sf) || defined(L_usi_to_df) || defined(L_usi_to_tf) | |
1939 | +FLO_type | |
1940 | +usi_to_float (USItype arg_a) | |
1941 | +{ | |
1942 | + fp_number_type in; | |
1943 | + | |
1944 | + in.sign = 0; | |
1945 | + if (!arg_a) | |
1946 | + { | |
1947 | + in.class = CLASS_ZERO; | |
1948 | + } | |
1949 | + else | |
1950 | + { | |
1951 | + in.class = CLASS_NUMBER; | |
1952 | + in.normal_exp = FRACBITS + NGARDS; | |
1953 | + in.fraction.ll = arg_a; | |
1954 | + | |
1955 | + while (in.fraction.ll > ((fractype)1 << (FRACBITS + NGARDS))) | |
1956 | + { | |
1957 | + in.fraction.ll >>= 1; | |
1958 | + in.normal_exp += 1; | |
1959 | + } | |
1960 | + while (in.fraction.ll < ((fractype)1 << (FRACBITS + NGARDS))) | |
1961 | + { | |
1962 | + in.fraction.ll <<= 1; | |
1963 | + in.normal_exp -= 1; | |
1964 | + } | |
1965 | + } | |
1966 | + return pack_d (&in); | |
1967 | +} | |
1968 | +#endif | |
1969 | + | |
1970 | +#if defined(L_sf_to_si) || defined(L_df_to_si) || defined(L_tf_to_si) | |
1971 | +SItype | |
1972 | +float_to_si (FLO_type arg_a) | |
1973 | +{ | |
1974 | + fp_number_type a; | |
1975 | + SItype tmp; | |
1976 | + FLO_union_type au; | |
1977 | + | |
1978 | + au.value = arg_a; | |
1979 | + unpack_d (&au, &a); | |
1980 | + | |
1981 | + if (iszero (&a)) | |
1982 | + return 0; | |
1983 | + if (isnan (&a)) | |
1984 | + return 0; | |
1985 | + /* get reasonable MAX_SI_INT... */ | |
1986 | + if (isinf (&a)) | |
1987 | + return a.sign ? (-MAX_SI_INT)-1 : MAX_SI_INT; | |
1988 | + /* it is a number, but a small one */ | |
1989 | + if (a.normal_exp < 0) | |
1990 | + return 0; | |
1991 | + if (a.normal_exp > BITS_PER_SI - 2) | |
1992 | + return a.sign ? (-MAX_SI_INT)-1 : MAX_SI_INT; | |
1993 | + tmp = a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp); | |
1994 | + return a.sign ? (-tmp) : (tmp); | |
1995 | +} | |
1996 | +#endif /* L_sf_to_si || L_df_to_si */ | |
1997 | + | |
1998 | +#if defined(L_sf_to_usi) || defined(L_df_to_usi) || defined(L_tf_to_usi) | |
1999 | +#if defined US_SOFTWARE_GOFAST || defined(L_tf_to_usi) | |
2000 | +/* While libgcc2.c defines its own __fixunssfsi and __fixunsdfsi routines, | |
2001 | + we also define them for GOFAST because the ones in libgcc2.c have the | |
2002 | + wrong names and I'd rather define these here and keep GOFAST CYG-LOC's | |
2003 | + out of libgcc2.c. We can't define these here if not GOFAST because then | |
2004 | + there'd be duplicate copies. */ | |
2005 | + | |
2006 | +USItype | |
2007 | +float_to_usi (FLO_type arg_a) | |
2008 | +{ | |
2009 | + fp_number_type a; | |
2010 | + FLO_union_type au; | |
2011 | + | |
2012 | + au.value = arg_a; | |
2013 | + unpack_d (&au, &a); | |
2014 | + | |
2015 | + if (iszero (&a)) | |
2016 | + return 0; | |
2017 | + if (isnan (&a)) | |
2018 | + return 0; | |
2019 | + /* it is a negative number */ | |
2020 | + if (a.sign) | |
2021 | + return 0; | |
2022 | + /* get reasonable MAX_USI_INT... */ | |
2023 | + if (isinf (&a)) | |
2024 | + return MAX_USI_INT; | |
2025 | + /* it is a number, but a small one */ | |
2026 | + if (a.normal_exp < 0) | |
2027 | + return 0; | |
2028 | + if (a.normal_exp > BITS_PER_SI - 1) | |
2029 | + return MAX_USI_INT; | |
2030 | + else if (a.normal_exp > (FRACBITS + NGARDS)) | |
2031 | + return a.fraction.ll << (a.normal_exp - (FRACBITS + NGARDS)); | |
2032 | + else | |
2033 | + return a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp); | |
2034 | +} | |
2035 | +#endif /* US_SOFTWARE_GOFAST */ | |
2036 | +#endif /* L_sf_to_usi || L_df_to_usi */ | |
2037 | + | |
2038 | +#if defined(L_negate_sf) || defined(L_negate_df) || defined(L_negate_tf) | |
2039 | +FLO_type | |
2040 | +negate (FLO_type arg_a) | |
2041 | +{ | |
2042 | + fp_number_type a; | |
2043 | + FLO_union_type au; | |
2044 | + | |
2045 | + au.value = arg_a; | |
2046 | + unpack_d (&au, &a); | |
2047 | + | |
2048 | + flip_sign (&a); | |
2049 | + return pack_d (&a); | |
2050 | +} | |
2051 | +#endif /* L_negate_sf || L_negate_df */ | |
2052 | + | |
2053 | +#ifdef FLOAT | |
2054 | + | |
2055 | +#if defined(L_make_sf) | |
2056 | +SFtype | |
2057 | +__make_fp(fp_class_type class, | |
2058 | + unsigned int sign, | |
2059 | + int exp, | |
2060 | + USItype frac) | |
2061 | +{ | |
2062 | + fp_number_type in; | |
2063 | + | |
2064 | + in.class = class; | |
2065 | + in.sign = sign; | |
2066 | + in.normal_exp = exp; | |
2067 | + in.fraction.ll = frac; | |
2068 | + return pack_d (&in); | |
2069 | +} | |
2070 | +#endif /* L_make_sf */ | |
2071 | + | |
2072 | +#ifndef FLOAT_ONLY | |
2073 | + | |
2074 | +/* This enables one to build an fp library that supports float but not double. | |
2075 | + Otherwise, we would get an undefined reference to __make_dp. | |
2076 | + This is needed for some 8-bit ports that can't handle well values that | |
2077 | + are 8-bytes in size, so we just don't support double for them at all. */ | |
2078 | + | |
2079 | +#if defined(L_sf_to_df) | |
2080 | +DFtype | |
2081 | +sf_to_df (SFtype arg_a) | |
2082 | +{ | |
2083 | + fp_number_type in; | |
2084 | + FLO_union_type au; | |
2085 | + | |
2086 | + au.value = arg_a; | |
2087 | + unpack_d (&au, &in); | |
2088 | + | |
2089 | + return __make_dp (in.class, in.sign, in.normal_exp, | |
2090 | + ((UDItype) in.fraction.ll) << F_D_BITOFF); | |
2091 | +} | |
2092 | +#endif /* L_sf_to_df */ | |
2093 | + | |
2094 | +#if defined(L_sf_to_tf) && defined(TMODES) | |
2095 | +TFtype | |
2096 | +sf_to_tf (SFtype arg_a) | |
2097 | +{ | |
2098 | + fp_number_type in; | |
2099 | + FLO_union_type au; | |
2100 | + | |
2101 | + au.value = arg_a; | |
2102 | + unpack_d (&au, &in); | |
2103 | + | |
2104 | + return __make_tp (in.class, in.sign, in.normal_exp, | |
2105 | + ((UTItype) in.fraction.ll) << F_T_BITOFF); | |
2106 | +} | |
2107 | +#endif /* L_sf_to_df */ | |
2108 | + | |
2109 | +#endif /* ! FLOAT_ONLY */ | |
2110 | +#endif /* FLOAT */ | |
2111 | + | |
2112 | +#ifndef FLOAT | |
2113 | + | |
2114 | +extern SFtype __make_fp (fp_class_type, unsigned int, int, USItype); | |
2115 | + | |
2116 | +#if defined(L_make_df) | |
2117 | +DFtype | |
2118 | +__make_dp (fp_class_type class, unsigned int sign, int exp, UDItype frac) | |
2119 | +{ | |
2120 | + fp_number_type in; | |
2121 | + | |
2122 | + in.class = class; | |
2123 | + in.sign = sign; | |
2124 | + in.normal_exp = exp; | |
2125 | + in.fraction.ll = frac; | |
2126 | + return pack_d (&in); | |
2127 | +} | |
2128 | +#endif /* L_make_df */ | |
2129 | + | |
2130 | +#if defined(L_df_to_sf) | |
2131 | +SFtype | |
2132 | +df_to_sf (DFtype arg_a) | |
2133 | +{ | |
2134 | + fp_number_type in; | |
2135 | + USItype sffrac; | |
2136 | + FLO_union_type au; | |
2137 | + | |
2138 | + au.value = arg_a; | |
2139 | + unpack_d (&au, &in); | |
2140 | + | |
2141 | + sffrac = in.fraction.ll >> F_D_BITOFF; | |
2142 | + | |
2143 | + /* We set the lowest guard bit in SFFRAC if we discarded any non | |
2144 | + zero bits. */ | |
2145 | + if ((in.fraction.ll & (((USItype) 1 << F_D_BITOFF) - 1)) != 0) | |
2146 | + sffrac |= 1; | |
2147 | + | |
2148 | + return __make_fp (in.class, in.sign, in.normal_exp, sffrac); | |
2149 | +} | |
2150 | +#endif /* L_df_to_sf */ | |
2151 | + | |
2152 | +#if defined(L_df_to_tf) && defined(TMODES) \ | |
2153 | + && !defined(FLOAT) && !defined(TFLOAT) | |
2154 | +TFtype | |
2155 | +df_to_tf (DFtype arg_a) | |
2156 | +{ | |
2157 | + fp_number_type in; | |
2158 | + FLO_union_type au; | |
2159 | + | |
2160 | + au.value = arg_a; | |
2161 | + unpack_d (&au, &in); | |
2162 | + | |
2163 | + return __make_tp (in.class, in.sign, in.normal_exp, | |
2164 | + ((UTItype) in.fraction.ll) << D_T_BITOFF); | |
2165 | +} | |
2166 | +#endif /* L_sf_to_df */ | |
2167 | + | |
2168 | +#ifdef TFLOAT | |
2169 | +#if defined(L_make_tf) | |
2170 | +TFtype | |
2171 | +__make_tp(fp_class_type class, | |
2172 | + unsigned int sign, | |
2173 | + int exp, | |
2174 | + UTItype frac) | |
2175 | +{ | |
2176 | + fp_number_type in; | |
2177 | + | |
2178 | + in.class = class; | |
2179 | + in.sign = sign; | |
2180 | + in.normal_exp = exp; | |
2181 | + in.fraction.ll = frac; | |
2182 | + return pack_d (&in); | |
2183 | +} | |
2184 | +#endif /* L_make_tf */ | |
2185 | + | |
2186 | +#if defined(L_tf_to_df) | |
2187 | +DFtype | |
2188 | +tf_to_df (TFtype arg_a) | |
2189 | +{ | |
2190 | + fp_number_type in; | |
2191 | + UDItype sffrac; | |
2192 | + FLO_union_type au; | |
2193 | + | |
2194 | + au.value = arg_a; | |
2195 | + unpack_d (&au, &in); | |
2196 | + | |
2197 | + sffrac = in.fraction.ll >> D_T_BITOFF; | |
2198 | + | |
2199 | + /* We set the lowest guard bit in SFFRAC if we discarded any non | |
2200 | + zero bits. */ | |
2201 | + if ((in.fraction.ll & (((UTItype) 1 << D_T_BITOFF) - 1)) != 0) | |
2202 | + sffrac |= 1; | |
2203 | + | |
2204 | + return __make_dp (in.class, in.sign, in.normal_exp, sffrac); | |
2205 | +} | |
2206 | +#endif /* L_tf_to_df */ | |
2207 | + | |
2208 | +#if defined(L_tf_to_sf) | |
2209 | +SFtype | |
2210 | +tf_to_sf (TFtype arg_a) | |
2211 | +{ | |
2212 | + fp_number_type in; | |
2213 | + USItype sffrac; | |
2214 | + FLO_union_type au; | |
2215 | + | |
2216 | + au.value = arg_a; | |
2217 | + unpack_d (&au, &in); | |
2218 | + | |
2219 | + sffrac = in.fraction.ll >> F_T_BITOFF; | |
2220 | + | |
2221 | + /* We set the lowest guard bit in SFFRAC if we discarded any non | |
2222 | + zero bits. */ | |
2223 | + if ((in.fraction.ll & (((UTItype) 1 << F_T_BITOFF) - 1)) != 0) | |
2224 | + sffrac |= 1; | |
2225 | + | |
2226 | + return __make_fp (in.class, in.sign, in.normal_exp, sffrac); | |
2227 | +} | |
2228 | +#endif /* L_tf_to_sf */ | |
2229 | +#endif /* TFLOAT */ | |
2230 | + | |
2231 | +#endif /* ! FLOAT */ | |
2232 | +#endif /* !EXTENDED_FLOAT_STUBS */ | |
2233 | --- ./gcc/gcc/config/nios2/nios2-fp-bit.c | |
2234 | +++ ./gcc/gcc/config/nios2/nios2-fp-bit.c | |
2235 | @@ -0,0 +1,1652 @@ | |
2236 | +#define FLOAT | |
2237 | +/* This is a software floating point library which can be used | |
2238 | + for targets without hardware floating point. | |
2239 | + Copyright (C) 1994, 1995, 1996, 1997, 1998, 2000, 2001, 2002, 2003, 2004 | |
2240 | + Free Software Foundation, Inc. | |
2241 | + | |
2242 | +This file is free software; you can redistribute it and/or modify it | |
2243 | +under the terms of the GNU General Public License as published by the | |
2244 | +Free Software Foundation; either version 2, or (at your option) any | |
2245 | +later version. | |
2246 | + | |
2247 | +In addition to the permissions in the GNU General Public License, the | |
2248 | +Free Software Foundation gives you unlimited permission to link the | |
2249 | +compiled version of this file with other programs, and to distribute | |
2250 | +those programs without any restriction coming from the use of this | |
2251 | +file. (The General Public License restrictions do apply in other | |
2252 | +respects; for example, they cover modification of the file, and | |
2253 | +distribution when not linked into another program.) | |
2254 | + | |
2255 | +This file is distributed in the hope that it will be useful, but | |
2256 | +WITHOUT ANY WARRANTY; without even the implied warranty of | |
2257 | +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
2258 | +General Public License for more details. | |
2259 | + | |
2260 | +You should have received a copy of the GNU General Public License | |
2261 | +along with this program; see the file COPYING. If not, write to | |
2262 | +the Free Software Foundation, 59 Temple Place - Suite 330, | |
2263 | +Boston, MA 02111-1307, USA. */ | |
2264 | + | |
2265 | +/* As a special exception, if you link this library with other files, | |
2266 | + some of which are compiled with GCC, to produce an executable, | |
2267 | + this library does not by itself cause the resulting executable | |
2268 | + to be covered by the GNU General Public License. | |
2269 | + This exception does not however invalidate any other reasons why | |
2270 | + the executable file might be covered by the GNU General Public License. */ | |
2271 | + | |
2272 | +/* This implements IEEE 754 format arithmetic, but does not provide a | |
2273 | + mechanism for setting the rounding mode, or for generating or handling | |
2274 | + exceptions. | |
2275 | + | |
2276 | + The original code by Steve Chamberlain, hacked by Mark Eichin and Jim | |
2277 | + Wilson, all of Cygnus Support. */ | |
2278 | + | |
2279 | +/* The intended way to use this file is to make two copies, add `#define FLOAT' | |
2280 | + to one copy, then compile both copies and add them to libgcc.a. */ | |
2281 | + | |
2282 | +#include "tconfig.h" | |
2283 | +#include "coretypes.h" | |
2284 | +#include "tm.h" | |
2285 | +#include "config/fp-bit.h" | |
2286 | + | |
2287 | +/* The following macros can be defined to change the behavior of this file: | |
2288 | + FLOAT: Implement a `float', aka SFmode, fp library. If this is not | |
2289 | + defined, then this file implements a `double', aka DFmode, fp library. | |
2290 | + FLOAT_ONLY: Used with FLOAT, to implement a `float' only library, i.e. | |
2291 | + don't include float->double conversion which requires the double library. | |
2292 | + This is useful only for machines which can't support doubles, e.g. some | |
2293 | + 8-bit processors. | |
2294 | + CMPtype: Specify the type that floating point compares should return. | |
2295 | + This defaults to SItype, aka int. | |
2296 | + US_SOFTWARE_GOFAST: This makes all entry points use the same names as the | |
2297 | + US Software goFast library. | |
2298 | + _DEBUG_BITFLOAT: This makes debugging the code a little easier, by adding | |
2299 | + two integers to the FLO_union_type. | |
2300 | + NO_DENORMALS: Disable handling of denormals. | |
2301 | + NO_NANS: Disable nan and infinity handling | |
2302 | + SMALL_MACHINE: Useful when operations on QIs and HIs are faster | |
2303 | + than on an SI */ | |
2304 | + | |
2305 | +/* We don't currently support extended floats (long doubles) on machines | |
2306 | + without hardware to deal with them. | |
2307 | + | |
2308 | + These stubs are just to keep the linker from complaining about unresolved | |
2309 | + references which can be pulled in from libio & libstdc++, even if the | |
2310 | + user isn't using long doubles. However, they may generate an unresolved | |
2311 | + external to abort if abort is not used by the function, and the stubs | |
2312 | + are referenced from within libc, since libgcc goes before and after the | |
2313 | + system library. */ | |
2314 | + | |
2315 | +#ifdef DECLARE_LIBRARY_RENAMES | |
2316 | + DECLARE_LIBRARY_RENAMES | |
2317 | +#endif | |
2318 | + | |
2319 | +#ifdef EXTENDED_FLOAT_STUBS | |
2320 | +extern void abort (void); | |
2321 | +void __extendsfxf2 (void) { abort(); } | |
2322 | +void __extenddfxf2 (void) { abort(); } | |
2323 | +void __truncxfdf2 (void) { abort(); } | |
2324 | +void __truncxfsf2 (void) { abort(); } | |
2325 | +void __fixxfsi (void) { abort(); } | |
2326 | +void __floatsixf (void) { abort(); } | |
2327 | +void __addxf3 (void) { abort(); } | |
2328 | +void __subxf3 (void) { abort(); } | |
2329 | +void __mulxf3 (void) { abort(); } | |
2330 | +void __divxf3 (void) { abort(); } | |
2331 | +void __negxf2 (void) { abort(); } | |
2332 | +void __eqxf2 (void) { abort(); } | |
2333 | +void __nexf2 (void) { abort(); } | |
2334 | +void __gtxf2 (void) { abort(); } | |
2335 | +void __gexf2 (void) { abort(); } | |
2336 | +void __lexf2 (void) { abort(); } | |
2337 | +void __ltxf2 (void) { abort(); } | |
2338 | + | |
2339 | +void __extendsftf2 (void) { abort(); } | |
2340 | +void __extenddftf2 (void) { abort(); } | |
2341 | +void __trunctfdf2 (void) { abort(); } | |
2342 | +void __trunctfsf2 (void) { abort(); } | |
2343 | +void __fixtfsi (void) { abort(); } | |
2344 | +void __floatsitf (void) { abort(); } | |
2345 | +void __addtf3 (void) { abort(); } | |
2346 | +void __subtf3 (void) { abort(); } | |
2347 | +void __multf3 (void) { abort(); } | |
2348 | +void __divtf3 (void) { abort(); } | |
2349 | +void __negtf2 (void) { abort(); } | |
2350 | +void __eqtf2 (void) { abort(); } | |
2351 | +void __netf2 (void) { abort(); } | |
2352 | +void __gttf2 (void) { abort(); } | |
2353 | +void __getf2 (void) { abort(); } | |
2354 | +void __letf2 (void) { abort(); } | |
2355 | +void __lttf2 (void) { abort(); } | |
2356 | +#else /* !EXTENDED_FLOAT_STUBS, rest of file */ | |
2357 | + | |
2358 | +/* IEEE "special" number predicates */ | |
2359 | + | |
2360 | +#ifdef NO_NANS | |
2361 | + | |
2362 | +#define nan() 0 | |
2363 | +#define isnan(x) 0 | |
2364 | +#define isinf(x) 0 | |
2365 | +#else | |
2366 | + | |
2367 | +#if defined L_thenan_sf | |
2368 | +const fp_number_type __thenan_sf = { CLASS_SNAN, 0, 0, {(fractype) 0} }; | |
2369 | +#elif defined L_thenan_df | |
2370 | +const fp_number_type __thenan_df = { CLASS_SNAN, 0, 0, {(fractype) 0} }; | |
2371 | +#elif defined L_thenan_tf | |
2372 | +const fp_number_type __thenan_tf = { CLASS_SNAN, 0, 0, {(fractype) 0} }; | |
2373 | +#elif defined TFLOAT | |
2374 | +extern const fp_number_type __thenan_tf; | |
2375 | +#elif defined FLOAT | |
2376 | +extern const fp_number_type __thenan_sf; | |
2377 | +#else | |
2378 | +extern const fp_number_type __thenan_df; | |
2379 | +#endif | |
2380 | + | |
2381 | +INLINE | |
2382 | +static fp_number_type * | |
2383 | +nan (void) | |
2384 | +{ | |
2385 | + /* Discard the const qualifier... */ | |
2386 | +#ifdef TFLOAT | |
2387 | + return (fp_number_type *) (& __thenan_tf); | |
2388 | +#elif defined FLOAT | |
2389 | + return (fp_number_type *) (& __thenan_sf); | |
2390 | +#else | |
2391 | + return (fp_number_type *) (& __thenan_df); | |
2392 | +#endif | |
2393 | +} | |
2394 | + | |
2395 | +INLINE | |
2396 | +static int | |
2397 | +isnan ( fp_number_type * x) | |
2398 | +{ | |
2399 | + return x->class == CLASS_SNAN || x->class == CLASS_QNAN; | |
2400 | +} | |
2401 | + | |
2402 | +INLINE | |
2403 | +static int | |
2404 | +isinf ( fp_number_type * x) | |
2405 | +{ | |
2406 | + return x->class == CLASS_INFINITY; | |
2407 | +} | |
2408 | + | |
2409 | +#endif /* NO_NANS */ | |
2410 | + | |
2411 | +INLINE | |
2412 | +static int | |
2413 | +iszero ( fp_number_type * x) | |
2414 | +{ | |
2415 | + return x->class == CLASS_ZERO; | |
2416 | +} | |
2417 | + | |
2418 | +INLINE | |
2419 | +static void | |
2420 | +flip_sign ( fp_number_type * x) | |
2421 | +{ | |
2422 | + x->sign = !x->sign; | |
2423 | +} | |
2424 | + | |
2425 | +extern FLO_type pack_d ( fp_number_type * ); | |
2426 | + | |
2427 | +#if defined(L_pack_df) || defined(L_pack_sf) || defined(L_pack_tf) | |
2428 | +FLO_type | |
2429 | +pack_d ( fp_number_type * src) | |
2430 | +{ | |
2431 | + FLO_union_type dst; | |
2432 | + fractype fraction = src->fraction.ll; /* wasn't unsigned before? */ | |
2433 | + int sign = src->sign; | |
2434 | + int exp = 0; | |
2435 | + | |
2436 | + if (LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) && (isnan (src) || isinf (src))) | |
2437 | + { | |
2438 | + /* We can't represent these values accurately. By using the | |
2439 | + largest possible magnitude, we guarantee that the conversion | |
2440 | + of infinity is at least as big as any finite number. */ | |
2441 | + exp = EXPMAX; | |
2442 | + fraction = ((fractype) 1 << FRACBITS) - 1; | |
2443 | + } | |
2444 | + else if (isnan (src)) | |
2445 | + { | |
2446 | + exp = EXPMAX; | |
2447 | + if (src->class == CLASS_QNAN || 1) | |
2448 | + { | |
2449 | +#ifdef QUIET_NAN_NEGATED | |
2450 | + fraction |= QUIET_NAN - 1; | |
2451 | +#else | |
2452 | + fraction |= QUIET_NAN; | |
2453 | +#endif | |
2454 | + } | |
2455 | + } | |
2456 | + else if (isinf (src)) | |
2457 | + { | |
2458 | + exp = EXPMAX; | |
2459 | + fraction = 0; | |
2460 | + } | |
2461 | + else if (iszero (src)) | |
2462 | + { | |
2463 | + exp = 0; | |
2464 | + fraction = 0; | |
2465 | + } | |
2466 | + else if (fraction == 0) | |
2467 | + { | |
2468 | + exp = 0; | |
2469 | + } | |
2470 | + else | |
2471 | + { | |
2472 | + if (src->normal_exp < NORMAL_EXPMIN) | |
2473 | + { | |
2474 | +#ifdef NO_DENORMALS | |
2475 | + /* Go straight to a zero representation if denormals are not | |
2476 | + supported. The denormal handling would be harmless but | |
2477 | + isn't unnecessary. */ | |
2478 | + exp = 0; | |
2479 | + fraction = 0; | |
2480 | +#else /* NO_DENORMALS */ | |
2481 | + /* This number's exponent is too low to fit into the bits | |
2482 | + available in the number, so we'll store 0 in the exponent and | |
2483 | + shift the fraction to the right to make up for it. */ | |
2484 | + | |
2485 | + int shift = NORMAL_EXPMIN - src->normal_exp; | |
2486 | + | |
2487 | + exp = 0; | |
2488 | + | |
2489 | + if (shift > FRAC_NBITS - NGARDS) | |
2490 | + { | |
2491 | + /* No point shifting, since it's more that 64 out. */ | |
2492 | + fraction = 0; | |
2493 | + } | |
2494 | + else | |
2495 | + { | |
2496 | + int lowbit = (fraction & (((fractype)1 << shift) - 1)) ? 1 : 0; | |
2497 | + fraction = (fraction >> shift) | lowbit; | |
2498 | + } | |
2499 | + if ((fraction & GARDMASK) == GARDMSB) | |
2500 | + { | |
2501 | + if ((fraction & (1 << NGARDS))) | |
2502 | + fraction += GARDROUND + 1; | |
2503 | + } | |
2504 | + else | |
2505 | + { | |
2506 | + /* Add to the guards to round up. */ | |
2507 | + fraction += GARDROUND; | |
2508 | + } | |
2509 | + /* Perhaps the rounding means we now need to change the | |
2510 | + exponent, because the fraction is no longer denormal. */ | |
2511 | + if (fraction >= IMPLICIT_1) | |
2512 | + { | |
2513 | + exp += 1; | |
2514 | + } | |
2515 | + fraction >>= NGARDS; | |
2516 | +#endif /* NO_DENORMALS */ | |
2517 | + } | |
2518 | + else if (!LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) | |
2519 | + && src->normal_exp > EXPBIAS) | |
2520 | + { | |
2521 | + exp = EXPMAX; | |
2522 | + fraction = 0; | |
2523 | + } | |
2524 | + else | |
2525 | + { | |
2526 | + exp = src->normal_exp + EXPBIAS; | |
2527 | + if (!ROUND_TOWARDS_ZERO) | |
2528 | + { | |
2529 | + /* IF the gard bits are the all zero, but the first, then we're | |
2530 | + half way between two numbers, choose the one which makes the | |
2531 | + lsb of the answer 0. */ | |
2532 | + if ((fraction & GARDMASK) == GARDMSB) | |
2533 | + { | |
2534 | + if (fraction & (1 << NGARDS)) | |
2535 | + fraction += GARDROUND + 1; | |
2536 | + } | |
2537 | + else | |
2538 | + { | |
2539 | + /* Add a one to the guards to round up */ | |
2540 | + fraction += GARDROUND; | |
2541 | + } | |
2542 | + if (fraction >= IMPLICIT_2) | |
2543 | + { | |
2544 | + fraction >>= 1; | |
2545 | + exp += 1; | |
2546 | + } | |
2547 | + } | |
2548 | + fraction >>= NGARDS; | |
2549 | + | |
2550 | + if (LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) && exp > EXPMAX) | |
2551 | + { | |
2552 | + /* Saturate on overflow. */ | |
2553 | + exp = EXPMAX; | |
2554 | + fraction = ((fractype) 1 << FRACBITS) - 1; | |
2555 | + } | |
2556 | + } | |
2557 | + } | |
2558 | + | |
2559 | + /* We previously used bitfields to store the number, but this doesn't | |
2560 | + handle little/big endian systems conveniently, so use shifts and | |
2561 | + masks */ | |
2562 | +#ifdef FLOAT_BIT_ORDER_MISMATCH | |
2563 | + dst.bits.fraction = fraction; | |
2564 | + dst.bits.exp = exp; | |
2565 | + dst.bits.sign = sign; | |
2566 | +#else | |
2567 | +# if defined TFLOAT && defined HALFFRACBITS | |
2568 | + { | |
2569 | + halffractype high, low, unity; | |
2570 | + int lowsign, lowexp; | |
2571 | + | |
2572 | + unity = (halffractype) 1 << HALFFRACBITS; | |
2573 | + | |
2574 | + /* Set HIGH to the high double's significand, masking out the implicit 1. | |
2575 | + Set LOW to the low double's full significand. */ | |
2576 | + high = (fraction >> (FRACBITS - HALFFRACBITS)) & (unity - 1); | |
2577 | + low = fraction & (unity * 2 - 1); | |
2578 | + | |
2579 | + /* Get the initial sign and exponent of the low double. */ | |
2580 | + lowexp = exp - HALFFRACBITS - 1; | |
2581 | + lowsign = sign; | |
2582 | + | |
2583 | + /* HIGH should be rounded like a normal double, making |LOW| <= | |
2584 | + 0.5 ULP of HIGH. Assume round-to-nearest. */ | |
2585 | + if (exp < EXPMAX) | |
2586 | + if (low > unity || (low == unity && (high & 1) == 1)) | |
2587 | + { | |
2588 | + /* Round HIGH up and adjust LOW to match. */ | |
2589 | + high++; | |
2590 | + if (high == unity) | |
2591 | + { | |
2592 | + /* May make it infinite, but that's OK. */ | |
2593 | + high = 0; | |
2594 | + exp++; | |
2595 | + } | |
2596 | + low = unity * 2 - low; | |
2597 | + lowsign ^= 1; | |
2598 | + } | |
2599 | + | |
2600 | + high |= (halffractype) exp << HALFFRACBITS; | |
2601 | + high |= (halffractype) sign << (HALFFRACBITS + EXPBITS); | |
2602 | + | |
2603 | + if (exp == EXPMAX || exp == 0 || low == 0) | |
2604 | + low = 0; | |
2605 | + else | |
2606 | + { | |
2607 | + while (lowexp > 0 && low < unity) | |
2608 | + { | |
2609 | + low <<= 1; | |
2610 | + lowexp--; | |
2611 | + } | |
2612 | + | |
2613 | + if (lowexp <= 0) | |
2614 | + { | |
2615 | + halffractype roundmsb, round; | |
2616 | + int shift; | |
2617 | + | |
2618 | + shift = 1 - lowexp; | |
2619 | + roundmsb = (1 << (shift - 1)); | |
2620 | + round = low & ((roundmsb << 1) - 1); | |
2621 | + | |
2622 | + low >>= shift; | |
2623 | + lowexp = 0; | |
2624 | + | |
2625 | + if (round > roundmsb || (round == roundmsb && (low & 1) == 1)) | |
2626 | + { | |
2627 | + low++; | |
2628 | + if (low == unity) | |
2629 | + /* LOW rounds up to the smallest normal number. */ | |
2630 | + lowexp++; | |
2631 | + } | |
2632 | + } | |
2633 | + | |
2634 | + low &= unity - 1; | |
2635 | + low |= (halffractype) lowexp << HALFFRACBITS; | |
2636 | + low |= (halffractype) lowsign << (HALFFRACBITS + EXPBITS); | |
2637 | + } | |
2638 | + dst.value_raw = ((fractype) high << HALFSHIFT) | low; | |
2639 | + } | |
2640 | +# else | |
2641 | + dst.value_raw = fraction & ((((fractype)1) << FRACBITS) - (fractype)1); | |
2642 | + dst.value_raw |= ((fractype) (exp & ((1 << EXPBITS) - 1))) << FRACBITS; | |
2643 | + dst.value_raw |= ((fractype) (sign & 1)) << (FRACBITS | EXPBITS); | |
2644 | +# endif | |
2645 | +#endif | |
2646 | + | |
2647 | +#if defined(FLOAT_WORD_ORDER_MISMATCH) && !defined(FLOAT) | |
2648 | +#ifdef TFLOAT | |
2649 | + { | |
2650 | + qrtrfractype tmp1 = dst.words[0]; | |
2651 | + qrtrfractype tmp2 = dst.words[1]; | |
2652 | + dst.words[0] = dst.words[3]; | |
2653 | + dst.words[1] = dst.words[2]; | |
2654 | + dst.words[2] = tmp2; | |
2655 | + dst.words[3] = tmp1; | |
2656 | + } | |
2657 | +#else | |
2658 | + { | |
2659 | + halffractype tmp = dst.words[0]; | |
2660 | + dst.words[0] = dst.words[1]; | |
2661 | + dst.words[1] = tmp; | |
2662 | + } | |
2663 | +#endif | |
2664 | +#endif | |
2665 | + | |
2666 | + return dst.value; | |
2667 | +} | |
2668 | +#endif | |
2669 | + | |
2670 | +#if defined(L_unpack_df) || defined(L_unpack_sf) || defined(L_unpack_tf) | |
2671 | +void | |
2672 | +unpack_d (FLO_union_type * src, fp_number_type * dst) | |
2673 | +{ | |
2674 | + /* We previously used bitfields to store the number, but this doesn't | |
2675 | + handle little/big endian systems conveniently, so use shifts and | |
2676 | + masks */ | |
2677 | + fractype fraction; | |
2678 | + int exp; | |
2679 | + int sign; | |
2680 | + | |
2681 | +#if defined(FLOAT_WORD_ORDER_MISMATCH) && !defined(FLOAT) | |
2682 | + FLO_union_type swapped; | |
2683 | + | |
2684 | +#ifdef TFLOAT | |
2685 | + swapped.words[0] = src->words[3]; | |
2686 | + swapped.words[1] = src->words[2]; | |
2687 | + swapped.words[2] = src->words[1]; | |
2688 | + swapped.words[3] = src->words[0]; | |
2689 | +#else | |
2690 | + swapped.words[0] = src->words[1]; | |
2691 | + swapped.words[1] = src->words[0]; | |
2692 | +#endif | |
2693 | + src = &swapped; | |
2694 | +#endif | |
2695 | + | |
2696 | +#ifdef FLOAT_BIT_ORDER_MISMATCH | |
2697 | + fraction = src->bits.fraction; | |
2698 | + exp = src->bits.exp; | |
2699 | + sign = src->bits.sign; | |
2700 | +#else | |
2701 | +# if defined TFLOAT && defined HALFFRACBITS | |
2702 | + { | |
2703 | + halffractype high, low; | |
2704 | + | |
2705 | + high = src->value_raw >> HALFSHIFT; | |
2706 | + low = src->value_raw & (((fractype)1 << HALFSHIFT) - 1); | |
2707 | + | |
2708 | + fraction = high & ((((fractype)1) << HALFFRACBITS) - 1); | |
2709 | + fraction <<= FRACBITS - HALFFRACBITS; | |
2710 | + exp = ((int)(high >> HALFFRACBITS)) & ((1 << EXPBITS) - 1); | |
2711 | + sign = ((int)(high >> (((HALFFRACBITS + EXPBITS))))) & 1; | |
2712 | + | |
2713 | + if (exp != EXPMAX && exp != 0 && low != 0) | |
2714 | + { | |
2715 | + int lowexp = ((int)(low >> HALFFRACBITS)) & ((1 << EXPBITS) - 1); | |
2716 | + int lowsign = ((int)(low >> (((HALFFRACBITS + EXPBITS))))) & 1; | |
2717 | + int shift; | |
2718 | + fractype xlow; | |
2719 | + | |
2720 | + xlow = low & ((((fractype)1) << HALFFRACBITS) - 1); | |
2721 | + if (lowexp) | |
2722 | + xlow |= (((halffractype)1) << HALFFRACBITS); | |
2723 | + else | |
2724 | + lowexp = 1; | |
2725 | + shift = (FRACBITS - HALFFRACBITS) - (exp - lowexp); | |
2726 | + if (shift > 0) | |
2727 | + xlow <<= shift; | |
2728 | + else if (shift < 0) | |
2729 | + xlow >>= -shift; | |
2730 | + if (sign == lowsign) | |
2731 | + fraction += xlow; | |
2732 | + else if (fraction >= xlow) | |
2733 | + fraction -= xlow; | |
2734 | + else | |
2735 | + { | |
2736 | + /* The high part is a power of two but the full number is lower. | |
2737 | + This code will leave the implicit 1 in FRACTION, but we'd | |
2738 | + have added that below anyway. */ | |
2739 | + fraction = (((fractype) 1 << FRACBITS) - xlow) << 1; | |
2740 | + exp--; | |
2741 | + } | |
2742 | + } | |
2743 | + } | |
2744 | +# else | |
2745 | + fraction = src->value_raw & ((((fractype)1) << FRACBITS) - 1); | |
2746 | + exp = ((int)(src->value_raw >> FRACBITS)) & ((1 << EXPBITS) - 1); | |
2747 | + sign = ((int)(src->value_raw >> (FRACBITS + EXPBITS))) & 1; | |
2748 | +# endif | |
2749 | +#endif | |
2750 | + | |
2751 | + dst->sign = sign; | |
2752 | + if (exp == 0) | |
2753 | + { | |
2754 | + /* Hmm. Looks like 0 */ | |
2755 | + if (fraction == 0 | |
2756 | +#ifdef NO_DENORMALS | |
2757 | + || 1 | |
2758 | +#endif | |
2759 | + ) | |
2760 | + { | |
2761 | + /* tastes like zero */ | |
2762 | + dst->class = CLASS_ZERO; | |
2763 | + } | |
2764 | + else | |
2765 | + { | |
2766 | + /* Zero exponent with nonzero fraction - it's denormalized, | |
2767 | + so there isn't a leading implicit one - we'll shift it so | |
2768 | + it gets one. */ | |
2769 | + dst->normal_exp = exp - EXPBIAS + 1; | |
2770 | + fraction <<= NGARDS; | |
2771 | + | |
2772 | + dst->class = CLASS_NUMBER; | |
2773 | +#if 1 | |
2774 | + while (fraction < IMPLICIT_1) | |
2775 | + { | |
2776 | + fraction <<= 1; | |
2777 | + dst->normal_exp--; | |
2778 | + } | |
2779 | +#endif | |
2780 | + dst->fraction.ll = fraction; | |
2781 | + } | |
2782 | + } | |
2783 | + else if (!LARGEST_EXPONENT_IS_NORMAL (FRAC_NBITS) && exp == EXPMAX) | |
2784 | + { | |
2785 | + /* Huge exponent*/ | |
2786 | + if (fraction == 0) | |
2787 | + { | |
2788 | + /* Attached to a zero fraction - means infinity */ | |
2789 | + dst->class = CLASS_INFINITY; | |
2790 | + } | |
2791 | + else | |
2792 | + { | |
2793 | + /* Nonzero fraction, means nan */ | |
2794 | +#ifdef QUIET_NAN_NEGATED | |
2795 | + if ((fraction & QUIET_NAN) == 0) | |
2796 | +#else | |
2797 | + if (fraction & QUIET_NAN) | |
2798 | +#endif | |
2799 | + { | |
2800 | + dst->class = CLASS_QNAN; | |
2801 | + } | |
2802 | + else | |
2803 | + { | |
2804 | + dst->class = CLASS_SNAN; | |
2805 | + } | |
2806 | + /* Keep the fraction part as the nan number */ | |
2807 | + dst->fraction.ll = fraction; | |
2808 | + } | |
2809 | + } | |
2810 | + else | |
2811 | + { | |
2812 | + /* Nothing strange about this number */ | |
2813 | + dst->normal_exp = exp - EXPBIAS; | |
2814 | + dst->class = CLASS_NUMBER; | |
2815 | + dst->fraction.ll = (fraction << NGARDS) | IMPLICIT_1; | |
2816 | + } | |
2817 | +} | |
2818 | +#endif /* L_unpack_df || L_unpack_sf */ | |
2819 | + | |
2820 | +#if defined(L_addsub_sf) || defined(L_addsub_df) || defined(L_addsub_tf) | |
2821 | +static fp_number_type * | |
2822 | +_fpadd_parts (fp_number_type * a, | |
2823 | + fp_number_type * b, | |
2824 | + fp_number_type * tmp) | |
2825 | +{ | |
2826 | + intfrac tfraction; | |
2827 | + | |
2828 | + /* Put commonly used fields in local variables. */ | |
2829 | + int a_normal_exp; | |
2830 | + int b_normal_exp; | |
2831 | + fractype a_fraction; | |
2832 | + fractype b_fraction; | |
2833 | + | |
2834 | + if (isnan (a)) | |
2835 | + { | |
2836 | + return a; | |
2837 | + } | |
2838 | + if (isnan (b)) | |
2839 | + { | |
2840 | + return b; | |
2841 | + } | |
2842 | + if (isinf (a)) | |
2843 | + { | |
2844 | + /* Adding infinities with opposite signs yields a NaN. */ | |
2845 | + if (isinf (b) && a->sign != b->sign) | |
2846 | + return nan (); | |
2847 | + return a; | |
2848 | + } | |
2849 | + if (isinf (b)) | |
2850 | + { | |
2851 | + return b; | |
2852 | + } | |
2853 | + if (iszero (b)) | |
2854 | + { | |
2855 | + if (iszero (a)) | |
2856 | + { | |
2857 | + *tmp = *a; | |
2858 | + tmp->sign = a->sign & b->sign; | |
2859 | + return tmp; | |
2860 | + } | |
2861 | + return a; | |
2862 | + } | |
2863 | + if (iszero (a)) | |
2864 | + { | |
2865 | + return b; | |
2866 | + } | |
2867 | + | |
2868 | + /* Got two numbers. shift the smaller and increment the exponent till | |
2869 | + they're the same */ | |
2870 | + { | |
2871 | + int diff; | |
2872 | + | |
2873 | + a_normal_exp = a->normal_exp; | |
2874 | + b_normal_exp = b->normal_exp; | |
2875 | + a_fraction = a->fraction.ll; | |
2876 | + b_fraction = b->fraction.ll; | |
2877 | + | |
2878 | + diff = a_normal_exp - b_normal_exp; | |
2879 | + | |
2880 | + if (diff < 0) | |
2881 | + diff = -diff; | |
2882 | + if (diff < FRAC_NBITS) | |
2883 | + { | |
2884 | + /* ??? This does shifts one bit at a time. Optimize. */ | |
2885 | + while (a_normal_exp > b_normal_exp) | |
2886 | + { | |
2887 | + b_normal_exp++; | |
2888 | + LSHIFT (b_fraction); | |
2889 | + } | |
2890 | + while (b_normal_exp > a_normal_exp) | |
2891 | + { | |
2892 | + a_normal_exp++; | |
2893 | + LSHIFT (a_fraction); | |
2894 | + } | |
2895 | + } | |
2896 | + else | |
2897 | + { | |
2898 | + /* Somethings's up.. choose the biggest */ | |
2899 | + if (a_normal_exp > b_normal_exp) | |
2900 | + { | |
2901 | + b_normal_exp = a_normal_exp; | |
2902 | + b_fraction = 0; | |
2903 | + } | |
2904 | + else | |
2905 | + { | |
2906 | + a_normal_exp = b_normal_exp; | |
2907 | + a_fraction = 0; | |
2908 | + } | |
2909 | + } | |
2910 | + } | |
2911 | + | |
2912 | + if (a->sign != b->sign) | |
2913 | + { | |
2914 | + if (a->sign) | |
2915 | + { | |
2916 | + tfraction = -a_fraction + b_fraction; | |
2917 | + } | |
2918 | + else | |
2919 | + { | |
2920 | + tfraction = a_fraction - b_fraction; | |
2921 | + } | |
2922 | + if (tfraction >= 0) | |
2923 | + { | |
2924 | + tmp->sign = 0; | |
2925 | + tmp->normal_exp = a_normal_exp; | |
2926 | + tmp->fraction.ll = tfraction; | |
2927 | + } | |
2928 | + else | |
2929 | + { | |
2930 | + tmp->sign = 1; | |
2931 | + tmp->normal_exp = a_normal_exp; | |
2932 | + tmp->fraction.ll = -tfraction; | |
2933 | + } | |
2934 | + /* and renormalize it */ | |
2935 | + | |
2936 | + while (tmp->fraction.ll < IMPLICIT_1 && tmp->fraction.ll) | |
2937 | + { | |
2938 | + tmp->fraction.ll <<= 1; | |
2939 | + tmp->normal_exp--; | |
2940 | + } | |
2941 | + } | |
2942 | + else | |
2943 | + { | |
2944 | + tmp->sign = a->sign; | |
2945 | + tmp->normal_exp = a_normal_exp; | |
2946 | + tmp->fraction.ll = a_fraction + b_fraction; | |
2947 | + } | |
2948 | + tmp->class = CLASS_NUMBER; | |
2949 | + /* Now the fraction is added, we have to shift down to renormalize the | |
2950 | + number */ | |
2951 | + | |
2952 | + if (tmp->fraction.ll >= IMPLICIT_2) | |
2953 | + { | |
2954 | + LSHIFT (tmp->fraction.ll); | |
2955 | + tmp->normal_exp++; | |
2956 | + } | |
2957 | + return tmp; | |
2958 | + | |
2959 | +} | |
2960 | + | |
2961 | +FLO_type | |
2962 | +add (FLO_type arg_a, FLO_type arg_b) | |
2963 | +{ | |
2964 | + fp_number_type a; | |
2965 | + fp_number_type b; | |
2966 | + fp_number_type tmp; | |
2967 | + fp_number_type *res; | |
2968 | + FLO_union_type au, bu; | |
2969 | + | |
2970 | + au.value = arg_a; | |
2971 | + bu.value = arg_b; | |
2972 | + | |
2973 | + unpack_d (&au, &a); | |
2974 | + unpack_d (&bu, &b); | |
2975 | + | |
2976 | + res = _fpadd_parts (&a, &b, &tmp); | |
2977 | + | |
2978 | + return pack_d (res); | |
2979 | +} | |
2980 | + | |
2981 | +FLO_type | |
2982 | +sub (FLO_type arg_a, FLO_type arg_b) | |
2983 | +{ | |
2984 | + fp_number_type a; | |
2985 | + fp_number_type b; | |
2986 | + fp_number_type tmp; | |
2987 | + fp_number_type *res; | |
2988 | + FLO_union_type au, bu; | |
2989 | + | |
2990 | + au.value = arg_a; | |
2991 | + bu.value = arg_b; | |
2992 | + | |
2993 | + unpack_d (&au, &a); | |
2994 | + unpack_d (&bu, &b); | |
2995 | + | |
2996 | + b.sign ^= 1; | |
2997 | + | |
2998 | + res = _fpadd_parts (&a, &b, &tmp); | |
2999 | + | |
3000 | + return pack_d (res); | |
3001 | +} | |
3002 | +#endif /* L_addsub_sf || L_addsub_df */ | |
3003 | + | |
3004 | +#if defined(L_mul_sf) || defined(L_mul_df) || defined(L_mul_tf) | |
3005 | +static inline __attribute__ ((__always_inline__)) fp_number_type * | |
3006 | +_fpmul_parts ( fp_number_type * a, | |
3007 | + fp_number_type * b, | |
3008 | + fp_number_type * tmp) | |
3009 | +{ | |
3010 | + fractype low = 0; | |
3011 | + fractype high = 0; | |
3012 | + | |
3013 | + if (isnan (a)) | |
3014 | + { | |
3015 | + a->sign = a->sign != b->sign; | |
3016 | + return a; | |
3017 | + } | |
3018 | + if (isnan (b)) | |
3019 | + { | |
3020 | + b->sign = a->sign != b->sign; | |
3021 | + return b; | |
3022 | + } | |
3023 | + if (isinf (a)) | |
3024 | + { | |
3025 | + if (iszero (b)) | |
3026 | + return nan (); | |
3027 | + a->sign = a->sign != b->sign; | |
3028 | + return a; | |
3029 | + } | |
3030 | + if (isinf (b)) | |
3031 | + { | |
3032 | + if (iszero (a)) | |
3033 | + { | |
3034 | + return nan (); | |
3035 | + } | |
3036 | + b->sign = a->sign != b->sign; | |
3037 | + return b; | |
3038 | + } | |
3039 | + if (iszero (a)) | |
3040 | + { | |
3041 | + a->sign = a->sign != b->sign; | |
3042 | + return a; | |
3043 | + } | |
3044 | + if (iszero (b)) | |
3045 | + { | |
3046 | + b->sign = a->sign != b->sign; | |
3047 | + return b; | |
3048 | + } | |
3049 | + | |
3050 | + /* Calculate the mantissa by multiplying both numbers to get a | |
3051 | + twice-as-wide number. */ | |
3052 | + { | |
3053 | +#if defined(NO_DI_MODE) || defined(TFLOAT) | |
3054 | + { | |
3055 | + fractype x = a->fraction.ll; | |
3056 | + fractype ylow = b->fraction.ll; | |
3057 | + fractype yhigh = 0; | |
3058 | + int bit; | |
3059 | + | |
3060 | + /* ??? This does multiplies one bit at a time. Optimize. */ | |
3061 | + for (bit = 0; bit < FRAC_NBITS; bit++) | |
3062 | + { | |
3063 | + int carry; | |
3064 | + | |
3065 | + if (x & 1) | |
3066 | + { | |
3067 | + carry = (low += ylow) < ylow; | |
3068 | + high += yhigh + carry; | |
3069 | + } | |
3070 | + yhigh <<= 1; | |
3071 | + if (ylow & FRACHIGH) | |
3072 | + { | |
3073 | + yhigh |= 1; | |
3074 | + } | |
3075 | + ylow <<= 1; | |
3076 | + x >>= 1; | |
3077 | + } | |
3078 | + } | |
3079 | +#elif defined(FLOAT) | |
3080 | + /* Multiplying two USIs to get a UDI, we're safe. */ | |
3081 | + { | |
3082 | + UDItype answer = (UDItype)a->fraction.ll * (UDItype)b->fraction.ll; | |
3083 | + | |
3084 | + high = answer >> BITS_PER_SI; | |
3085 | + low = answer; | |
3086 | + } | |
3087 | +#else | |
3088 | + /* fractype is DImode, but we need the result to be twice as wide. | |
3089 | + Assuming a widening multiply from DImode to TImode is not | |
3090 | + available, build one by hand. */ | |
3091 | + { | |
3092 | + USItype nl = a->fraction.ll; | |
3093 | + USItype nh = a->fraction.ll >> BITS_PER_SI; | |
3094 | + USItype ml = b->fraction.ll; | |
3095 | + USItype mh = b->fraction.ll >> BITS_PER_SI; | |
3096 | + UDItype pp_ll = (UDItype) ml * nl; | |
3097 | + UDItype pp_hl = (UDItype) mh * nl; | |
3098 | + UDItype pp_lh = (UDItype) ml * nh; | |
3099 | + UDItype pp_hh = (UDItype) mh * nh; | |
3100 | + UDItype res2 = 0; | |
3101 | + UDItype res0 = 0; | |
3102 | + UDItype ps_hh__ = pp_hl + pp_lh; | |
3103 | + if (ps_hh__ < pp_hl) | |
3104 | + res2 += (UDItype)1 << BITS_PER_SI; | |
3105 | + pp_hl = (UDItype)(USItype)ps_hh__ << BITS_PER_SI; | |
3106 | + res0 = pp_ll + pp_hl; | |
3107 | + if (res0 < pp_ll) | |
3108 | + res2++; | |
3109 | + res2 += (ps_hh__ >> BITS_PER_SI) + pp_hh; | |
3110 | + high = res2; | |
3111 | + low = res0; | |
3112 | + } | |
3113 | +#endif | |
3114 | + } | |
3115 | + | |
3116 | + tmp->normal_exp = a->normal_exp + b->normal_exp | |
3117 | + + FRAC_NBITS - (FRACBITS + NGARDS); | |
3118 | + tmp->sign = a->sign != b->sign; | |
3119 | + while (high >= IMPLICIT_2) | |
3120 | + { | |
3121 | + tmp->normal_exp++; | |
3122 | + if (high & 1) | |
3123 | + { | |
3124 | + low >>= 1; | |
3125 | + low |= FRACHIGH; | |
3126 | + } | |
3127 | + high >>= 1; | |
3128 | + } | |
3129 | + while (high < IMPLICIT_1) | |
3130 | + { | |
3131 | + tmp->normal_exp--; | |
3132 | + | |
3133 | + high <<= 1; | |
3134 | + if (low & FRACHIGH) | |
3135 | + high |= 1; | |
3136 | + low <<= 1; | |
3137 | + } | |
3138 | + /* rounding is tricky. if we only round if it won't make us round later. */ | |
3139 | +#if 0 | |
3140 | + if (low & FRACHIGH2) | |
3141 | + { | |
3142 | + if (((high & GARDMASK) != GARDMSB) | |
3143 | + && (((high + 1) & GARDMASK) == GARDMSB)) | |
3144 | + { | |
3145 | + /* don't round, it gets done again later. */ | |
3146 | + } | |
3147 | + else | |
3148 | + { | |
3149 | + high++; | |
3150 | + } | |
3151 | + } | |
3152 | +#endif | |
3153 | + if (!ROUND_TOWARDS_ZERO && (high & GARDMASK) == GARDMSB) | |
3154 | + { | |
3155 | + if (high & (1 << NGARDS)) | |
3156 | + { | |
3157 | + /* half way, so round to even */ | |
3158 | + high += GARDROUND + 1; | |
3159 | + } | |
3160 | + else if (low) | |
3161 | + { | |
3162 | + /* but we really weren't half way */ | |
3163 | + high += GARDROUND + 1; | |
3164 | + } | |
3165 | + } | |
3166 | + tmp->fraction.ll = high; | |
3167 | + tmp->class = CLASS_NUMBER; | |
3168 | + return tmp; | |
3169 | +} | |
3170 | + | |
3171 | +FLO_type | |
3172 | +multiply (FLO_type arg_a, FLO_type arg_b) | |
3173 | +{ | |
3174 | + fp_number_type a; | |
3175 | + fp_number_type b; | |
3176 | + fp_number_type tmp; | |
3177 | + fp_number_type *res; | |
3178 | + FLO_union_type au, bu; | |
3179 | + | |
3180 | + au.value = arg_a; | |
3181 | + bu.value = arg_b; | |
3182 | + | |
3183 | + unpack_d (&au, &a); | |
3184 | + unpack_d (&bu, &b); | |
3185 | + | |
3186 | + res = _fpmul_parts (&a, &b, &tmp); | |
3187 | + | |
3188 | + return pack_d (res); | |
3189 | +} | |
3190 | +#endif /* L_mul_sf || L_mul_df */ | |
3191 | + | |
3192 | +#if defined(L_div_sf) || defined(L_div_df) || defined(L_div_tf) | |
3193 | +static inline __attribute__ ((__always_inline__)) fp_number_type * | |
3194 | +_fpdiv_parts (fp_number_type * a, | |
3195 | + fp_number_type * b) | |
3196 | +{ | |
3197 | + fractype bit; | |
3198 | + fractype numerator; | |
3199 | + fractype denominator; | |
3200 | + fractype quotient; | |
3201 | + | |
3202 | + if (isnan (a)) | |
3203 | + { | |
3204 | + return a; | |
3205 | + } | |
3206 | + if (isnan (b)) | |
3207 | + { | |
3208 | + return b; | |
3209 | + } | |
3210 | + | |
3211 | + a->sign = a->sign ^ b->sign; | |
3212 | + | |
3213 | + if (isinf (a) || iszero (a)) | |
3214 | + { | |
3215 | + if (a->class == b->class) | |
3216 | + return nan (); | |
3217 | + return a; | |
3218 | + } | |
3219 | + | |
3220 | + if (isinf (b)) | |
3221 | + { | |
3222 | + a->fraction.ll = 0; | |
3223 | + a->normal_exp = 0; | |
3224 | + return a; | |
3225 | + } | |
3226 | + if (iszero (b)) | |
3227 | + { | |
3228 | + a->class = CLASS_INFINITY; | |
3229 | + return a; | |
3230 | + } | |
3231 | + | |
3232 | + /* Calculate the mantissa by multiplying both 64bit numbers to get a | |
3233 | + 128 bit number */ | |
3234 | + { | |
3235 | + /* quotient = | |
3236 | + ( numerator / denominator) * 2^(numerator exponent - denominator exponent) | |
3237 | + */ | |
3238 | + | |
3239 | + a->normal_exp = a->normal_exp - b->normal_exp; | |
3240 | + numerator = a->fraction.ll; | |
3241 | + denominator = b->fraction.ll; | |
3242 | + | |
3243 | + if (numerator < denominator) | |
3244 | + { | |
3245 | + /* Fraction will be less than 1.0 */ | |
3246 | + numerator *= 2; | |
3247 | + a->normal_exp--; | |
3248 | + } | |
3249 | + bit = IMPLICIT_1; | |
3250 | + quotient = 0; | |
3251 | + /* ??? Does divide one bit at a time. Optimize. */ | |
3252 | + while (bit) | |
3253 | + { | |
3254 | + if (numerator >= denominator) | |
3255 | + { | |
3256 | + quotient |= bit; | |
3257 | + numerator -= denominator; | |
3258 | + } | |
3259 | + bit >>= 1; | |
3260 | + numerator *= 2; | |
3261 | + } | |
3262 | + | |
3263 | + if (!ROUND_TOWARDS_ZERO && (quotient & GARDMASK) == GARDMSB) | |
3264 | + { | |
3265 | + if (quotient & (1 << NGARDS)) | |
3266 | + { | |
3267 | + /* half way, so round to even */ | |
3268 | + quotient += GARDROUND + 1; | |
3269 | + } | |
3270 | + else if (numerator) | |
3271 | + { | |
3272 | + /* but we really weren't half way, more bits exist */ | |
3273 | + quotient += GARDROUND + 1; | |
3274 | + } | |
3275 | + } | |
3276 | + | |
3277 | + a->fraction.ll = quotient; | |
3278 | + return (a); | |
3279 | + } | |
3280 | +} | |
3281 | + | |
3282 | +FLO_type | |
3283 | +divide (FLO_type arg_a, FLO_type arg_b) | |
3284 | +{ | |
3285 | + fp_number_type a; | |
3286 | + fp_number_type b; | |
3287 | + fp_number_type *res; | |
3288 | + FLO_union_type au, bu; | |
3289 | + | |
3290 | + au.value = arg_a; | |
3291 | + bu.value = arg_b; | |
3292 | + | |
3293 | + unpack_d (&au, &a); | |
3294 | + unpack_d (&bu, &b); | |
3295 | + | |
3296 | + res = _fpdiv_parts (&a, &b); | |
3297 | + | |
3298 | + return pack_d (res); | |
3299 | +} | |
3300 | +#endif /* L_div_sf || L_div_df */ | |
3301 | + | |
3302 | +#if defined(L_fpcmp_parts_sf) || defined(L_fpcmp_parts_df) \ | |
3303 | + || defined(L_fpcmp_parts_tf) | |
3304 | +/* according to the demo, fpcmp returns a comparison with 0... thus | |
3305 | + a<b -> -1 | |
3306 | + a==b -> 0 | |
3307 | + a>b -> +1 | |
3308 | + */ | |
3309 | + | |
3310 | +int | |
3311 | +__fpcmp_parts (fp_number_type * a, fp_number_type * b) | |
3312 | +{ | |
3313 | +#if 0 | |
3314 | + /* either nan -> unordered. Must be checked outside of this routine. */ | |
3315 | + if (isnan (a) && isnan (b)) | |
3316 | + { | |
3317 | + return 1; /* still unordered! */ | |
3318 | + } | |
3319 | +#endif | |
3320 | + | |
3321 | + if (isnan (a) || isnan (b)) | |
3322 | + { | |
3323 | + return 1; /* how to indicate unordered compare? */ | |
3324 | + } | |
3325 | + if (isinf (a) && isinf (b)) | |
3326 | + { | |
3327 | + /* +inf > -inf, but +inf != +inf */ | |
3328 | + /* b \a| +inf(0)| -inf(1) | |
3329 | + ______\+--------+-------- | |
3330 | + +inf(0)| a==b(0)| a<b(-1) | |
3331 | + -------+--------+-------- | |
3332 | + -inf(1)| a>b(1) | a==b(0) | |
3333 | + -------+--------+-------- | |
3334 | + So since unordered must be nonzero, just line up the columns... | |
3335 | + */ | |
3336 | + return b->sign - a->sign; | |
3337 | + } | |
3338 | + /* but not both... */ | |
3339 | + if (isinf (a)) | |
3340 | + { | |
3341 | + return a->sign ? -1 : 1; | |
3342 | + } | |
3343 | + if (isinf (b)) | |
3344 | + { | |
3345 | + return b->sign ? 1 : -1; | |
3346 | + } | |
3347 | + if (iszero (a) && iszero (b)) | |
3348 | + { | |
3349 | + return 0; | |
3350 | + } | |
3351 | + if (iszero (a)) | |
3352 | + { | |
3353 | + return b->sign ? 1 : -1; | |
3354 | + } | |
3355 | + if (iszero (b)) | |
3356 | + { | |
3357 | + return a->sign ? -1 : 1; | |
3358 | + } | |
3359 | + /* now both are "normal". */ | |
3360 | + if (a->sign != b->sign) | |
3361 | + { | |
3362 | + /* opposite signs */ | |
3363 | + return a->sign ? -1 : 1; | |
3364 | + } | |
3365 | + /* same sign; exponents? */ | |
3366 | + if (a->normal_exp > b->normal_exp) | |
3367 | + { | |
3368 | + return a->sign ? -1 : 1; | |
3369 | + } | |
3370 | + if (a->normal_exp < b->normal_exp) | |
3371 | + { | |
3372 | + return a->sign ? 1 : -1; | |
3373 | + } | |
3374 | + /* same exponents; check size. */ | |
3375 | + if (a->fraction.ll > b->fraction.ll) | |
3376 | + { | |
3377 | + return a->sign ? -1 : 1; | |
3378 | + } | |
3379 | + if (a->fraction.ll < b->fraction.ll) | |
3380 | + { | |
3381 | + return a->sign ? 1 : -1; | |
3382 | + } | |
3383 | + /* after all that, they're equal. */ | |
3384 | + return 0; | |
3385 | +} | |
3386 | +#endif | |
3387 | + | |
3388 | +#if defined(L_compare_sf) || defined(L_compare_df) || defined(L_compoare_tf) | |
3389 | +CMPtype | |
3390 | +compare (FLO_type arg_a, FLO_type arg_b) | |
3391 | +{ | |
3392 | + fp_number_type a; | |
3393 | + fp_number_type b; | |
3394 | + FLO_union_type au, bu; | |
3395 | + | |
3396 | + au.value = arg_a; | |
3397 | + bu.value = arg_b; | |
3398 | + | |
3399 | + unpack_d (&au, &a); | |
3400 | + unpack_d (&bu, &b); | |
3401 | + | |
3402 | + return __fpcmp_parts (&a, &b); | |
3403 | +} | |
3404 | +#endif /* L_compare_sf || L_compare_df */ | |
3405 | + | |
3406 | +#ifndef US_SOFTWARE_GOFAST | |
3407 | + | |
3408 | +/* These should be optimized for their specific tasks someday. */ | |
3409 | + | |
3410 | +#if defined(L_eq_sf) || defined(L_eq_df) || defined(L_eq_tf) | |
3411 | +CMPtype | |
3412 | +_eq_f2 (FLO_type arg_a, FLO_type arg_b) | |
3413 | +{ | |
3414 | + fp_number_type a; | |
3415 | + fp_number_type b; | |
3416 | + FLO_union_type au, bu; | |
3417 | + | |
3418 | + au.value = arg_a; | |
3419 | + bu.value = arg_b; | |
3420 | + | |
3421 | + unpack_d (&au, &a); | |
3422 | + unpack_d (&bu, &b); | |
3423 | + | |
3424 | + if (isnan (&a) || isnan (&b)) | |
3425 | + return 1; /* false, truth == 0 */ | |
3426 | + | |
3427 | + return __fpcmp_parts (&a, &b) ; | |
3428 | +} | |
3429 | +#endif /* L_eq_sf || L_eq_df */ | |
3430 | + | |
3431 | +#if defined(L_ne_sf) || defined(L_ne_df) || defined(L_ne_tf) | |
3432 | +CMPtype | |
3433 | +_ne_f2 (FLO_type arg_a, FLO_type arg_b) | |
3434 | +{ | |
3435 | + fp_number_type a; | |
3436 | + fp_number_type b; | |
3437 | + FLO_union_type au, bu; | |
3438 | + | |
3439 | + au.value = arg_a; | |
3440 | + bu.value = arg_b; | |
3441 | + | |
3442 | + unpack_d (&au, &a); | |
3443 | + unpack_d (&bu, &b); | |
3444 | + | |
3445 | + if (isnan (&a) || isnan (&b)) | |
3446 | + return 1; /* true, truth != 0 */ | |
3447 | + | |
3448 | + return __fpcmp_parts (&a, &b) ; | |
3449 | +} | |
3450 | +#endif /* L_ne_sf || L_ne_df */ | |
3451 | + | |
3452 | +#if defined(L_gt_sf) || defined(L_gt_df) || defined(L_gt_tf) | |
3453 | +CMPtype | |
3454 | +_gt_f2 (FLO_type arg_a, FLO_type arg_b) | |
3455 | +{ | |
3456 | + fp_number_type a; | |
3457 | + fp_number_type b; | |
3458 | + FLO_union_type au, bu; | |
3459 | + | |
3460 | + au.value = arg_a; | |
3461 | + bu.value = arg_b; | |
3462 | + | |
3463 | + unpack_d (&au, &a); | |
3464 | + unpack_d (&bu, &b); | |
3465 | + | |
3466 | + if (isnan (&a) || isnan (&b)) | |
3467 | + return -1; /* false, truth > 0 */ | |
3468 | + | |
3469 | + return __fpcmp_parts (&a, &b); | |
3470 | +} | |
3471 | +#endif /* L_gt_sf || L_gt_df */ | |
3472 | + | |
3473 | +#if defined(L_ge_sf) || defined(L_ge_df) || defined(L_ge_tf) | |
3474 | +CMPtype | |
3475 | +_ge_f2 (FLO_type arg_a, FLO_type arg_b) | |
3476 | +{ | |
3477 | + fp_number_type a; | |
3478 | + fp_number_type b; | |
3479 | + FLO_union_type au, bu; | |
3480 | + | |
3481 | + au.value = arg_a; | |
3482 | + bu.value = arg_b; | |
3483 | + | |
3484 | + unpack_d (&au, &a); | |
3485 | + unpack_d (&bu, &b); | |
3486 | + | |
3487 | + if (isnan (&a) || isnan (&b)) | |
3488 | + return -1; /* false, truth >= 0 */ | |
3489 | + return __fpcmp_parts (&a, &b) ; | |
3490 | +} | |
3491 | +#endif /* L_ge_sf || L_ge_df */ | |
3492 | + | |
3493 | +#if defined(L_lt_sf) || defined(L_lt_df) || defined(L_lt_tf) | |
3494 | +CMPtype | |
3495 | +_lt_f2 (FLO_type arg_a, FLO_type arg_b) | |
3496 | +{ | |
3497 | + fp_number_type a; | |
3498 | + fp_number_type b; | |
3499 | + FLO_union_type au, bu; | |
3500 | + | |
3501 | + au.value = arg_a; | |
3502 | + bu.value = arg_b; | |
3503 | + | |
3504 | + unpack_d (&au, &a); | |
3505 | + unpack_d (&bu, &b); | |
3506 | + | |
3507 | + if (isnan (&a) || isnan (&b)) | |
3508 | + return 1; /* false, truth < 0 */ | |
3509 | + | |
3510 | + return __fpcmp_parts (&a, &b); | |
3511 | +} | |
3512 | +#endif /* L_lt_sf || L_lt_df */ | |
3513 | + | |
3514 | +#if defined(L_le_sf) || defined(L_le_df) || defined(L_le_tf) | |
3515 | +CMPtype | |
3516 | +_le_f2 (FLO_type arg_a, FLO_type arg_b) | |
3517 | +{ | |
3518 | + fp_number_type a; | |
3519 | + fp_number_type b; | |
3520 | + FLO_union_type au, bu; | |
3521 | + | |
3522 | + au.value = arg_a; | |
3523 | + bu.value = arg_b; | |
3524 | + | |
3525 | + unpack_d (&au, &a); | |
3526 | + unpack_d (&bu, &b); | |
3527 | + | |
3528 | + if (isnan (&a) || isnan (&b)) | |
3529 | + return 1; /* false, truth <= 0 */ | |
3530 | + | |
3531 | + return __fpcmp_parts (&a, &b) ; | |
3532 | +} | |
3533 | +#endif /* L_le_sf || L_le_df */ | |
3534 | + | |
3535 | +#endif /* ! US_SOFTWARE_GOFAST */ | |
3536 | + | |
3537 | +#if defined(L_unord_sf) || defined(L_unord_df) || defined(L_unord_tf) | |
3538 | +CMPtype | |
3539 | +_unord_f2 (FLO_type arg_a, FLO_type arg_b) | |
3540 | +{ | |
3541 | + fp_number_type a; | |
3542 | + fp_number_type b; | |
3543 | + FLO_union_type au, bu; | |
3544 | + | |
3545 | + au.value = arg_a; | |
3546 | + bu.value = arg_b; | |
3547 | + | |
3548 | + unpack_d (&au, &a); | |
3549 | + unpack_d (&bu, &b); | |
3550 | + | |
3551 | + return (isnan (&a) || isnan (&b)); | |
3552 | +} | |
3553 | +#endif /* L_unord_sf || L_unord_df */ | |
3554 | + | |
3555 | +#if defined(L_si_to_sf) || defined(L_si_to_df) || defined(L_si_to_tf) | |
3556 | +FLO_type | |
3557 | +si_to_float (SItype arg_a) | |
3558 | +{ | |
3559 | + fp_number_type in; | |
3560 | + | |
3561 | + in.class = CLASS_NUMBER; | |
3562 | + in.sign = arg_a < 0; | |
3563 | + if (!arg_a) | |
3564 | + { | |
3565 | + in.class = CLASS_ZERO; | |
3566 | + } | |
3567 | + else | |
3568 | + { | |
3569 | + in.normal_exp = FRACBITS + NGARDS; | |
3570 | + if (in.sign) | |
3571 | + { | |
3572 | + /* Special case for minint, since there is no +ve integer | |
3573 | + representation for it */ | |
3574 | + if (arg_a == (- MAX_SI_INT - 1)) | |
3575 | + { | |
3576 | + return (FLO_type)(- MAX_SI_INT - 1); | |
3577 | + } | |
3578 | + in.fraction.ll = (-arg_a); | |
3579 | + } | |
3580 | + else | |
3581 | + in.fraction.ll = arg_a; | |
3582 | + | |
3583 | + while (in.fraction.ll < ((fractype)1 << (FRACBITS + NGARDS))) | |
3584 | + { | |
3585 | + in.fraction.ll <<= 1; | |
3586 | + in.normal_exp -= 1; | |
3587 | + } | |
3588 | + } | |
3589 | + return pack_d (&in); | |
3590 | +} | |
3591 | +#endif /* L_si_to_sf || L_si_to_df */ | |
3592 | + | |
3593 | +#if defined(L_usi_to_sf) || defined(L_usi_to_df) || defined(L_usi_to_tf) | |
3594 | +FLO_type | |
3595 | +usi_to_float (USItype arg_a) | |
3596 | +{ | |
3597 | + fp_number_type in; | |
3598 | + | |
3599 | + in.sign = 0; | |
3600 | + if (!arg_a) | |
3601 | + { | |
3602 | + in.class = CLASS_ZERO; | |
3603 | + } | |
3604 | + else | |
3605 | + { | |
3606 | + in.class = CLASS_NUMBER; | |
3607 | + in.normal_exp = FRACBITS + NGARDS; | |
3608 | + in.fraction.ll = arg_a; | |
3609 | + | |
3610 | + while (in.fraction.ll > ((fractype)1 << (FRACBITS + NGARDS))) | |
3611 | + { | |
3612 | + in.fraction.ll >>= 1; | |
3613 | + in.normal_exp += 1; | |
3614 | + } | |
3615 | + while (in.fraction.ll < ((fractype)1 << (FRACBITS + NGARDS))) | |
3616 | + { | |
3617 | + in.fraction.ll <<= 1; | |
3618 | + in.normal_exp -= 1; | |
3619 | + } | |
3620 | + } | |
3621 | + return pack_d (&in); | |
3622 | +} | |
3623 | +#endif | |
3624 | + | |
3625 | +#if defined(L_sf_to_si) || defined(L_df_to_si) || defined(L_tf_to_si) | |
3626 | +SItype | |
3627 | +float_to_si (FLO_type arg_a) | |
3628 | +{ | |
3629 | + fp_number_type a; | |
3630 | + SItype tmp; | |
3631 | + FLO_union_type au; | |
3632 | + | |
3633 | + au.value = arg_a; | |
3634 | + unpack_d (&au, &a); | |
3635 | + | |
3636 | + if (iszero (&a)) | |
3637 | + return 0; | |
3638 | + if (isnan (&a)) | |
3639 | + return 0; | |
3640 | + /* get reasonable MAX_SI_INT... */ | |
3641 | + if (isinf (&a)) | |
3642 | + return a.sign ? (-MAX_SI_INT)-1 : MAX_SI_INT; | |
3643 | + /* it is a number, but a small one */ | |
3644 | + if (a.normal_exp < 0) | |
3645 | + return 0; | |
3646 | + if (a.normal_exp > BITS_PER_SI - 2) | |
3647 | + return a.sign ? (-MAX_SI_INT)-1 : MAX_SI_INT; | |
3648 | + tmp = a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp); | |
3649 | + return a.sign ? (-tmp) : (tmp); | |
3650 | +} | |
3651 | +#endif /* L_sf_to_si || L_df_to_si */ | |
3652 | + | |
3653 | +#if defined(L_sf_to_usi) || defined(L_df_to_usi) || defined(L_tf_to_usi) | |
3654 | +#if defined US_SOFTWARE_GOFAST || defined(L_tf_to_usi) | |
3655 | +/* While libgcc2.c defines its own __fixunssfsi and __fixunsdfsi routines, | |
3656 | + we also define them for GOFAST because the ones in libgcc2.c have the | |
3657 | + wrong names and I'd rather define these here and keep GOFAST CYG-LOC's | |
3658 | + out of libgcc2.c. We can't define these here if not GOFAST because then | |
3659 | + there'd be duplicate copies. */ | |
3660 | + | |
3661 | +USItype | |
3662 | +float_to_usi (FLO_type arg_a) | |
3663 | +{ | |
3664 | + fp_number_type a; | |
3665 | + FLO_union_type au; | |
3666 | + | |
3667 | + au.value = arg_a; | |
3668 | + unpack_d (&au, &a); | |
3669 | + | |
3670 | + if (iszero (&a)) | |
3671 | + return 0; | |
3672 | + if (isnan (&a)) | |
3673 | + return 0; | |
3674 | + /* it is a negative number */ | |
3675 | + if (a.sign) | |
3676 | + return 0; | |
3677 | + /* get reasonable MAX_USI_INT... */ | |
3678 | + if (isinf (&a)) | |
3679 | + return MAX_USI_INT; | |
3680 | + /* it is a number, but a small one */ | |
3681 | + if (a.normal_exp < 0) | |
3682 | + return 0; | |
3683 | + if (a.normal_exp > BITS_PER_SI - 1) | |
3684 | + return MAX_USI_INT; | |
3685 | + else if (a.normal_exp > (FRACBITS + NGARDS)) | |
3686 | + return a.fraction.ll << (a.normal_exp - (FRACBITS + NGARDS)); | |
3687 | + else | |
3688 | + return a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp); | |
3689 | +} | |
3690 | +#endif /* US_SOFTWARE_GOFAST */ | |
3691 | +#endif /* L_sf_to_usi || L_df_to_usi */ | |
3692 | + | |
3693 | +#if defined(L_negate_sf) || defined(L_negate_df) || defined(L_negate_tf) | |
3694 | +FLO_type | |
3695 | +negate (FLO_type arg_a) | |
3696 | +{ | |
3697 | + fp_number_type a; | |
3698 | + FLO_union_type au; | |
3699 | + | |
3700 | + au.value = arg_a; | |
3701 | + unpack_d (&au, &a); | |
3702 | + | |
3703 | + flip_sign (&a); | |
3704 | + return pack_d (&a); | |
3705 | +} | |
3706 | +#endif /* L_negate_sf || L_negate_df */ | |
3707 | + | |
3708 | +#ifdef FLOAT | |
3709 | + | |
3710 | +#if defined(L_make_sf) | |
3711 | +SFtype | |
3712 | +__make_fp(fp_class_type class, | |
3713 | + unsigned int sign, | |
3714 | + int exp, | |
3715 | + USItype frac) | |
3716 | +{ | |
3717 | + fp_number_type in; | |
3718 | + | |
3719 | + in.class = class; | |
3720 | + in.sign = sign; | |
3721 | + in.normal_exp = exp; | |
3722 | + in.fraction.ll = frac; | |
3723 | + return pack_d (&in); | |
3724 | +} | |
3725 | +#endif /* L_make_sf */ | |
3726 | + | |
3727 | +#ifndef FLOAT_ONLY | |
3728 | + | |
3729 | +/* This enables one to build an fp library that supports float but not double. | |
3730 | + Otherwise, we would get an undefined reference to __make_dp. | |
3731 | + This is needed for some 8-bit ports that can't handle well values that | |
3732 | + are 8-bytes in size, so we just don't support double for them at all. */ | |
3733 | + | |
3734 | +#if defined(L_sf_to_df) | |
3735 | +DFtype | |
3736 | +sf_to_df (SFtype arg_a) | |
3737 | +{ | |
3738 | + fp_number_type in; | |
3739 | + FLO_union_type au; | |
3740 | + | |
3741 | + au.value = arg_a; | |
3742 | + unpack_d (&au, &in); | |
3743 | + | |
3744 | + return __make_dp (in.class, in.sign, in.normal_exp, | |
3745 | + ((UDItype) in.fraction.ll) << F_D_BITOFF); | |
3746 | +} | |
3747 | +#endif /* L_sf_to_df */ | |
3748 | + | |
3749 | +#if defined(L_sf_to_tf) && defined(TMODES) | |
3750 | +TFtype | |
3751 | +sf_to_tf (SFtype arg_a) | |
3752 | +{ | |
3753 | + fp_number_type in; | |
3754 | + FLO_union_type au; | |
3755 | + | |
3756 | + au.value = arg_a; | |
3757 | + unpack_d (&au, &in); | |
3758 | + | |
3759 | + return __make_tp (in.class, in.sign, in.normal_exp, | |
3760 | + ((UTItype) in.fraction.ll) << F_T_BITOFF); | |
3761 | +} | |
3762 | +#endif /* L_sf_to_df */ | |
3763 | + | |
3764 | +#endif /* ! FLOAT_ONLY */ | |
3765 | +#endif /* FLOAT */ | |
3766 | + | |
3767 | +#ifndef FLOAT | |
3768 | + | |
3769 | +extern SFtype __make_fp (fp_class_type, unsigned int, int, USItype); | |
3770 | + | |
3771 | +#if defined(L_make_df) | |
3772 | +DFtype | |
3773 | +__make_dp (fp_class_type class, unsigned int sign, int exp, UDItype frac) | |
3774 | +{ | |
3775 | + fp_number_type in; | |
3776 | + | |
3777 | + in.class = class; | |
3778 | + in.sign = sign; | |
3779 | + in.normal_exp = exp; | |
3780 | + in.fraction.ll = frac; | |
3781 | + return pack_d (&in); | |
3782 | +} | |
3783 | +#endif /* L_make_df */ | |
3784 | + | |
3785 | +#if defined(L_df_to_sf) | |
3786 | +SFtype | |
3787 | +df_to_sf (DFtype arg_a) | |
3788 | +{ | |
3789 | + fp_number_type in; | |
3790 | + USItype sffrac; | |
3791 | + FLO_union_type au; | |
3792 | + | |
3793 | + au.value = arg_a; | |
3794 | + unpack_d (&au, &in); | |
3795 | + | |
3796 | + sffrac = in.fraction.ll >> F_D_BITOFF; | |
3797 | + | |
3798 | + /* We set the lowest guard bit in SFFRAC if we discarded any non | |
3799 | + zero bits. */ | |
3800 | + if ((in.fraction.ll & (((USItype) 1 << F_D_BITOFF) - 1)) != 0) | |
3801 | + sffrac |= 1; | |
3802 | + | |
3803 | + return __make_fp (in.class, in.sign, in.normal_exp, sffrac); | |
3804 | +} | |
3805 | +#endif /* L_df_to_sf */ | |
3806 | + | |
3807 | +#if defined(L_df_to_tf) && defined(TMODES) \ | |
3808 | + && !defined(FLOAT) && !defined(TFLOAT) | |
3809 | +TFtype | |
3810 | +df_to_tf (DFtype arg_a) | |
3811 | +{ | |
3812 | + fp_number_type in; | |
3813 | + FLO_union_type au; | |
3814 | + | |
3815 | + au.value = arg_a; | |
3816 | + unpack_d (&au, &in); | |
3817 | + | |
3818 | + return __make_tp (in.class, in.sign, in.normal_exp, | |
3819 | + ((UTItype) in.fraction.ll) << D_T_BITOFF); | |
3820 | +} | |
3821 | +#endif /* L_sf_to_df */ | |
3822 | + | |
3823 | +#ifdef TFLOAT | |
3824 | +#if defined(L_make_tf) | |
3825 | +TFtype | |
3826 | +__make_tp(fp_class_type class, | |
3827 | + unsigned int sign, | |
3828 | + int exp, | |
3829 | + UTItype frac) | |
3830 | +{ | |
3831 | + fp_number_type in; | |
3832 | + | |
3833 | + in.class = class; | |
3834 | + in.sign = sign; | |
3835 | + in.normal_exp = exp; | |
3836 | + in.fraction.ll = frac; | |
3837 | + return pack_d (&in); | |
3838 | +} | |
3839 | +#endif /* L_make_tf */ | |
3840 | + | |
3841 | +#if defined(L_tf_to_df) | |
3842 | +DFtype | |
3843 | +tf_to_df (TFtype arg_a) | |
3844 | +{ | |
3845 | + fp_number_type in; | |
3846 | + UDItype sffrac; | |
3847 | + FLO_union_type au; | |
3848 | + | |
3849 | + au.value = arg_a; | |
3850 | + unpack_d (&au, &in); | |
3851 | + | |
3852 | + sffrac = in.fraction.ll >> D_T_BITOFF; | |
3853 | + | |
3854 | + /* We set the lowest guard bit in SFFRAC if we discarded any non | |
3855 | + zero bits. */ | |
3856 | + if ((in.fraction.ll & (((UTItype) 1 << D_T_BITOFF) - 1)) != 0) | |
3857 | + sffrac |= 1; | |
3858 | + | |
3859 | + return __make_dp (in.class, in.sign, in.normal_exp, sffrac); | |
3860 | +} | |
3861 | +#endif /* L_tf_to_df */ | |
3862 | + | |
3863 | +#if defined(L_tf_to_sf) | |
3864 | +SFtype | |
3865 | +tf_to_sf (TFtype arg_a) | |
3866 | +{ | |
3867 | + fp_number_type in; | |
3868 | + USItype sffrac; | |
3869 | + FLO_union_type au; | |
3870 | + | |
3871 | + au.value = arg_a; | |
3872 | + unpack_d (&au, &in); | |
3873 | + | |
3874 | + sffrac = in.fraction.ll >> F_T_BITOFF; | |
3875 | + | |
3876 | + /* We set the lowest guard bit in SFFRAC if we discarded any non | |
3877 | + zero bits. */ | |
3878 | + if ((in.fraction.ll & (((UTItype) 1 << F_T_BITOFF) - 1)) != 0) | |
3879 | + sffrac |= 1; | |
3880 | + | |
3881 | + return __make_fp (in.class, in.sign, in.normal_exp, sffrac); | |
3882 | +} | |
3883 | +#endif /* L_tf_to_sf */ | |
3884 | +#endif /* TFLOAT */ | |
3885 | + | |
3886 | +#endif /* ! FLOAT */ | |
3887 | +#endif /* !EXTENDED_FLOAT_STUBS */ | |
3888 | --- ./gcc/gcc/libgcc2.c | |
3889 | +++ ./gcc/gcc/libgcc2.c | |
3890 | @@ -0,0 +1,1669 @@ | |
3891 | +/* More subroutines needed by GCC output code on some machines. */ | |
3892 | +/* Compile this one with gcc. */ | |
3893 | +/* Copyright (C) 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, | |
3894 | + 2000, 2001, 2002, 2003 Free Software Foundation, Inc. | |
3895 | + | |
3896 | +This file is part of GCC. | |
3897 | + | |
3898 | +GCC is free software; you can redistribute it and/or modify it under | |
3899 | +the terms of the GNU General Public License as published by the Free | |
3900 | +Software Foundation; either version 2, or (at your option) any later | |
3901 | +version. | |
3902 | + | |
3903 | +In addition to the permissions in the GNU General Public License, the | |
3904 | +Free Software Foundation gives you unlimited permission to link the | |
3905 | +compiled version of this file into combinations with other programs, | |
3906 | +and to distribute those combinations without any restriction coming | |
3907 | +from the use of this file. (The General Public License restrictions | |
3908 | +do apply in other respects; for example, they cover modification of | |
3909 | +the file, and distribution when not linked into a combine | |
3910 | +executable.) | |
3911 | + | |
3912 | +GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
3913 | +WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
3914 | +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
3915 | +for more details. | |
3916 | + | |
3917 | +You should have received a copy of the GNU General Public License | |
3918 | +along with GCC; see the file COPYING. If not, write to the Free | |
3919 | +Software Foundation, 59 Temple Place - Suite 330, Boston, MA | |
3920 | +02111-1307, USA. */ | |
3921 | + | |
3922 | + | |
3923 | +/* We include auto-host.h here to get HAVE_GAS_HIDDEN. This is | |
3924 | + supposedly valid even though this is a "target" file. */ | |
3925 | +#include "auto-host.h" | |
3926 | + | |
3927 | +/* It is incorrect to include config.h here, because this file is being | |
3928 | + compiled for the target, and hence definitions concerning only the host | |
3929 | + do not apply. */ | |
3930 | +#include "tconfig.h" | |
3931 | +#include "tsystem.h" | |
3932 | +#include "coretypes.h" | |
3933 | +#include "tm.h" | |
3934 | + | |
3935 | +/* Don't use `fancy_abort' here even if config.h says to use it. */ | |
3936 | +#ifdef abort | |
3937 | +#undef abort | |
3938 | +#endif | |
3939 | + | |
3940 | +#ifdef HAVE_GAS_HIDDEN | |
3941 | +#define ATTRIBUTE_HIDDEN __attribute__ ((__visibility__ ("hidden"))) | |
3942 | +#else | |
3943 | +#define ATTRIBUTE_HIDDEN | |
3944 | +#endif | |
3945 | + | |
3946 | +#include "libgcc2.h" | |
3947 | +\f | |
3948 | +#ifdef DECLARE_LIBRARY_RENAMES | |
3949 | + DECLARE_LIBRARY_RENAMES | |
3950 | +#endif | |
3951 | + | |
3952 | +#if defined (L_negdi2) | |
3953 | +DWtype | |
3954 | +__negdi2 (DWtype u) | |
3955 | +{ | |
3956 | + const DWunion uu = {.ll = u}; | |
3957 | + const DWunion w = { {.low = -uu.s.low, | |
3958 | + .high = -uu.s.high - ((UWtype) -uu.s.low > 0) } }; | |
3959 | + | |
3960 | + return w.ll; | |
3961 | +} | |
3962 | +#endif | |
3963 | + | |
3964 | +#ifdef L_addvsi3 | |
3965 | +Wtype | |
3966 | +__addvsi3 (Wtype a, Wtype b) | |
3967 | +{ | |
3968 | + const Wtype w = a + b; | |
3969 | + | |
3970 | + if (b >= 0 ? w < a : w > a) | |
3971 | + abort (); | |
3972 | + | |
3973 | + return w; | |
3974 | +} | |
3975 | +#endif | |
3976 | +\f | |
3977 | +#ifdef L_addvdi3 | |
3978 | +DWtype | |
3979 | +__addvdi3 (DWtype a, DWtype b) | |
3980 | +{ | |
3981 | + const DWtype w = a + b; | |
3982 | + | |
3983 | + if (b >= 0 ? w < a : w > a) | |
3984 | + abort (); | |
3985 | + | |
3986 | + return w; | |
3987 | +} | |
3988 | +#endif | |
3989 | +\f | |
3990 | +#ifdef L_subvsi3 | |
3991 | +Wtype | |
3992 | +__subvsi3 (Wtype a, Wtype b) | |
3993 | +{ | |
3994 | + const DWtype w = a - b; | |
3995 | + | |
3996 | + if (b >= 0 ? w > a : w < a) | |
3997 | + abort (); | |
3998 | + | |
3999 | + return w; | |
4000 | +} | |
4001 | +#endif | |
4002 | +\f | |
4003 | +#ifdef L_subvdi3 | |
4004 | +DWtype | |
4005 | +__subvdi3 (DWtype a, DWtype b) | |
4006 | +{ | |
4007 | + const DWtype w = a - b; | |
4008 | + | |
4009 | + if (b >= 0 ? w > a : w < a) | |
4010 | + abort (); | |
4011 | + | |
4012 | + return w; | |
4013 | +} | |
4014 | +#endif | |
4015 | +\f | |
4016 | +#ifdef L_mulvsi3 | |
4017 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
4018 | +Wtype | |
4019 | +__mulvsi3 (Wtype a, Wtype b) | |
4020 | +{ | |
4021 | + const DWtype w = (DWtype) a * (DWtype) b; | |
4022 | + | |
4023 | + if (((a >= 0) == (b >= 0)) | |
4024 | + ? (UDWtype) w > (UDWtype) (((DWtype) 1 << (WORD_SIZE - 1)) - 1) | |
4025 | + : (UDWtype) w < (UDWtype) ((DWtype) -1 << (WORD_SIZE - 1))) | |
4026 | + abort (); | |
4027 | + | |
4028 | + return w; | |
4029 | +} | |
4030 | +#endif | |
4031 | +\f | |
4032 | +#ifdef L_negvsi2 | |
4033 | +Wtype | |
4034 | +__negvsi2 (Wtype a) | |
4035 | +{ | |
4036 | + const Wtype w = -a; | |
4037 | + | |
4038 | + if (a >= 0 ? w > 0 : w < 0) | |
4039 | + abort (); | |
4040 | + | |
4041 | + return w; | |
4042 | +} | |
4043 | +#endif | |
4044 | +\f | |
4045 | +#ifdef L_negvdi2 | |
4046 | +DWtype | |
4047 | +__negvdi2 (DWtype a) | |
4048 | +{ | |
4049 | + const DWtype w = -a; | |
4050 | + | |
4051 | + if (a >= 0 ? w > 0 : w < 0) | |
4052 | + abort (); | |
4053 | + | |
4054 | + return w; | |
4055 | +} | |
4056 | +#endif | |
4057 | +\f | |
4058 | +#ifdef L_absvsi2 | |
4059 | +Wtype | |
4060 | +__absvsi2 (Wtype a) | |
4061 | +{ | |
4062 | + Wtype w = a; | |
4063 | + | |
4064 | + if (a < 0) | |
4065 | +#ifdef L_negvsi2 | |
4066 | + w = __negvsi2 (a); | |
4067 | +#else | |
4068 | + w = -a; | |
4069 | + | |
4070 | + if (w < 0) | |
4071 | + abort (); | |
4072 | +#endif | |
4073 | + | |
4074 | + return w; | |
4075 | +} | |
4076 | +#endif | |
4077 | +\f | |
4078 | +#ifdef L_absvdi2 | |
4079 | +DWtype | |
4080 | +__absvdi2 (DWtype a) | |
4081 | +{ | |
4082 | + DWtype w = a; | |
4083 | + | |
4084 | + if (a < 0) | |
4085 | +#ifdef L_negvdi2 | |
4086 | + w = __negvdi2 (a); | |
4087 | +#else | |
4088 | + w = -a; | |
4089 | + | |
4090 | + if (w < 0) | |
4091 | + abort (); | |
4092 | +#endif | |
4093 | + | |
4094 | + return w; | |
4095 | +} | |
4096 | +#endif | |
4097 | +\f | |
4098 | +#ifdef L_mulvdi3 | |
4099 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
4100 | +DWtype | |
4101 | +__mulvdi3 (DWtype u, DWtype v) | |
4102 | +{ | |
4103 | + /* The unchecked multiplication needs 3 Wtype x Wtype multiplications, | |
4104 | + but the checked multiplication needs only two. */ | |
4105 | + const DWunion uu = {.ll = u}; | |
4106 | + const DWunion vv = {.ll = v}; | |
4107 | + | |
4108 | + if (__builtin_expect (uu.s.high == uu.s.low >> (WORD_SIZE - 1), 1)) | |
4109 | + { | |
4110 | + /* u fits in a single Wtype. */ | |
4111 | + if (__builtin_expect (vv.s.high == vv.s.low >> (WORD_SIZE - 1), 1)) | |
4112 | + { | |
4113 | + /* v fits in a single Wtype as well. */ | |
4114 | + /* A single multiplication. No overflow risk. */ | |
4115 | + return (DWtype) uu.s.low * (DWtype) vv.s.low; | |
4116 | + } | |
4117 | + else | |
4118 | + { | |
4119 | + /* Two multiplications. */ | |
4120 | + DWunion w0 = {.ll = (UDWtype) (UWtype) uu.s.low | |
4121 | + * (UDWtype) (UWtype) vv.s.low}; | |
4122 | + DWunion w1 = {.ll = (UDWtype) (UWtype) uu.s.low | |
4123 | + * (UDWtype) (UWtype) vv.s.high}; | |
4124 | + | |
4125 | + if (vv.s.high < 0) | |
4126 | + w1.s.high -= uu.s.low; | |
4127 | + if (uu.s.low < 0) | |
4128 | + w1.ll -= vv.ll; | |
4129 | + w1.ll += (UWtype) w0.s.high; | |
4130 | + if (__builtin_expect (w1.s.high == w1.s.low >> (WORD_SIZE - 1), 1)) | |
4131 | + { | |
4132 | + w0.s.high = w1.s.low; | |
4133 | + return w0.ll; | |
4134 | + } | |
4135 | + } | |
4136 | + } | |
4137 | + else | |
4138 | + { | |
4139 | + if (__builtin_expect (vv.s.high == vv.s.low >> (WORD_SIZE - 1), 1)) | |
4140 | + { | |
4141 | + /* v fits into a single Wtype. */ | |
4142 | + /* Two multiplications. */ | |
4143 | + DWunion w0 = {.ll = (UDWtype) (UWtype) uu.s.low | |
4144 | + * (UDWtype) (UWtype) vv.s.low}; | |
4145 | + DWunion w1 = {.ll = (UDWtype) (UWtype) uu.s.high | |
4146 | + * (UDWtype) (UWtype) vv.s.low}; | |
4147 | + | |
4148 | + if (uu.s.high < 0) | |
4149 | + w1.s.high -= vv.s.low; | |
4150 | + if (vv.s.low < 0) | |
4151 | + w1.ll -= uu.ll; | |
4152 | + w1.ll += (UWtype) w0.s.high; | |
4153 | + if (__builtin_expect (w1.s.high == w1.s.low >> (WORD_SIZE - 1), 1)) | |
4154 | + { | |
4155 | + w0.s.high = w1.s.low; | |
4156 | + return w0.ll; | |
4157 | + } | |
4158 | + } | |
4159 | + else | |
4160 | + { | |
4161 | + /* A few sign checks and a single multiplication. */ | |
4162 | + if (uu.s.high >= 0) | |
4163 | + { | |
4164 | + if (vv.s.high >= 0) | |
4165 | + { | |
4166 | + if (uu.s.high == 0 && vv.s.high == 0) | |
4167 | + { | |
4168 | + const DWtype w = (UDWtype) (UWtype) uu.s.low | |
4169 | + * (UDWtype) (UWtype) vv.s.low; | |
4170 | + if (__builtin_expect (w >= 0, 1)) | |
4171 | + return w; | |
4172 | + } | |
4173 | + } | |
4174 | + else | |
4175 | + { | |
4176 | + if (uu.s.high == 0 && vv.s.high == (Wtype) -1) | |
4177 | + { | |
4178 | + DWunion ww = {.ll = (UDWtype) (UWtype) uu.s.low | |
4179 | + * (UDWtype) (UWtype) vv.s.low}; | |
4180 | + | |
4181 | + ww.s.high -= uu.s.low; | |
4182 | + if (__builtin_expect (ww.s.high < 0, 1)) | |
4183 | + return ww.ll; | |
4184 | + } | |
4185 | + } | |
4186 | + } | |
4187 | + else | |
4188 | + { | |
4189 | + if (vv.s.high >= 0) | |
4190 | + { | |
4191 | + if (uu.s.high == (Wtype) -1 && vv.s.high == 0) | |
4192 | + { | |
4193 | + DWunion ww = {.ll = (UDWtype) (UWtype) uu.s.low | |
4194 | + * (UDWtype) (UWtype) vv.s.low}; | |
4195 | + | |
4196 | + ww.s.high -= vv.s.low; | |
4197 | + if (__builtin_expect (ww.s.high < 0, 1)) | |
4198 | + return ww.ll; | |
4199 | + } | |
4200 | + } | |
4201 | + else | |
4202 | + { | |
4203 | + if (uu.s.high == (Wtype) -1 && vv.s.high == (Wtype) - 1) | |
4204 | + { | |
4205 | + DWunion ww = {.ll = (UDWtype) (UWtype) uu.s.low | |
4206 | + * (UDWtype) (UWtype) vv.s.low}; | |
4207 | + | |
4208 | + ww.s.high -= uu.s.low; | |
4209 | + ww.s.high -= vv.s.low; | |
4210 | + if (__builtin_expect (ww.s.high >= 0, 1)) | |
4211 | + return ww.ll; | |
4212 | + } | |
4213 | + } | |
4214 | + } | |
4215 | + } | |
4216 | + } | |
4217 | + | |
4218 | + /* Overflow. */ | |
4219 | + abort (); | |
4220 | +} | |
4221 | +#endif | |
4222 | +\f | |
4223 | + | |
4224 | +/* Unless shift functions are defined with full ANSI prototypes, | |
4225 | + parameter b will be promoted to int if word_type is smaller than an int. */ | |
4226 | +#ifdef L_lshrdi3 | |
4227 | +DWtype | |
4228 | +__lshrdi3 (DWtype u, word_type b) | |
4229 | +{ | |
4230 | + if (b == 0) | |
4231 | + return u; | |
4232 | + | |
4233 | + const DWunion uu = {.ll = u}; | |
4234 | + const word_type bm = (sizeof (Wtype) * BITS_PER_UNIT) - b; | |
4235 | + DWunion w; | |
4236 | + | |
4237 | + if (bm <= 0) | |
4238 | + { | |
4239 | + w.s.high = 0; | |
4240 | + w.s.low = (UWtype) uu.s.high >> -bm; | |
4241 | + } | |
4242 | + else | |
4243 | + { | |
4244 | + const UWtype carries = (UWtype) uu.s.high << bm; | |
4245 | + | |
4246 | + w.s.high = (UWtype) uu.s.high >> b; | |
4247 | + w.s.low = ((UWtype) uu.s.low >> b) | carries; | |
4248 | + } | |
4249 | + | |
4250 | + return w.ll; | |
4251 | +} | |
4252 | +#endif | |
4253 | + | |
4254 | +#ifdef L_ashldi3 | |
4255 | +DWtype | |
4256 | +__ashldi3 (DWtype u, word_type b) | |
4257 | +{ | |
4258 | + if (b == 0) | |
4259 | + return u; | |
4260 | + | |
4261 | + const DWunion uu = {.ll = u}; | |
4262 | + const word_type bm = (sizeof (Wtype) * BITS_PER_UNIT) - b; | |
4263 | + DWunion w; | |
4264 | + | |
4265 | + if (bm <= 0) | |
4266 | + { | |
4267 | + w.s.low = 0; | |
4268 | + w.s.high = (UWtype) uu.s.low << -bm; | |
4269 | + } | |
4270 | + else | |
4271 | + { | |
4272 | + const UWtype carries = (UWtype) uu.s.low >> bm; | |
4273 | + | |
4274 | + w.s.low = (UWtype) uu.s.low << b; | |
4275 | + w.s.high = ((UWtype) uu.s.high << b) | carries; | |
4276 | + } | |
4277 | + | |
4278 | + return w.ll; | |
4279 | +} | |
4280 | +#endif | |
4281 | + | |
4282 | +#ifdef L_ashrdi3 | |
4283 | +DWtype | |
4284 | +__ashrdi3 (DWtype u, word_type b) | |
4285 | +{ | |
4286 | + if (b == 0) | |
4287 | + return u; | |
4288 | + | |
4289 | + const DWunion uu = {.ll = u}; | |
4290 | + const word_type bm = (sizeof (Wtype) * BITS_PER_UNIT) - b; | |
4291 | + DWunion w; | |
4292 | + | |
4293 | + if (bm <= 0) | |
4294 | + { | |
4295 | + /* w.s.high = 1..1 or 0..0 */ | |
4296 | + w.s.high = uu.s.high >> (sizeof (Wtype) * BITS_PER_UNIT - 1); | |
4297 | + w.s.low = uu.s.high >> -bm; | |
4298 | + } | |
4299 | + else | |
4300 | + { | |
4301 | + const UWtype carries = (UWtype) uu.s.high << bm; | |
4302 | + | |
4303 | + w.s.high = uu.s.high >> b; | |
4304 | + w.s.low = ((UWtype) uu.s.low >> b) | carries; | |
4305 | + } | |
4306 | + | |
4307 | + return w.ll; | |
4308 | +} | |
4309 | +#endif | |
4310 | +\f | |
4311 | +#ifdef L_ffssi2 | |
4312 | +#undef int | |
4313 | +extern int __ffsSI2 (UWtype u); | |
4314 | +int | |
4315 | +__ffsSI2 (UWtype u) | |
4316 | +{ | |
4317 | + UWtype count; | |
4318 | + | |
4319 | + if (u == 0) | |
4320 | + return 0; | |
4321 | + | |
4322 | + count_trailing_zeros (count, u); | |
4323 | + return count + 1; | |
4324 | +} | |
4325 | +#endif | |
4326 | +\f | |
4327 | +#ifdef L_ffsdi2 | |
4328 | +#undef int | |
4329 | +extern int __ffsDI2 (DWtype u); | |
4330 | +int | |
4331 | +__ffsDI2 (DWtype u) | |
4332 | +{ | |
4333 | + const DWunion uu = {.ll = u}; | |
4334 | + UWtype word, count, add; | |
4335 | + | |
4336 | + if (uu.s.low != 0) | |
4337 | + word = uu.s.low, add = 0; | |
4338 | + else if (uu.s.high != 0) | |
4339 | + word = uu.s.high, add = BITS_PER_UNIT * sizeof (Wtype); | |
4340 | + else | |
4341 | + return 0; | |
4342 | + | |
4343 | + count_trailing_zeros (count, word); | |
4344 | + return count + add + 1; | |
4345 | +} | |
4346 | +#endif | |
4347 | +\f | |
4348 | +#ifdef L_muldi3 | |
4349 | +DWtype | |
4350 | +__muldi3 (DWtype u, DWtype v) | |
4351 | +{ | |
4352 | + const DWunion uu = {.ll = u}; | |
4353 | + const DWunion vv = {.ll = v}; | |
4354 | + DWunion w = {.ll = __umulsidi3 (uu.s.low, vv.s.low)}; | |
4355 | + | |
4356 | + w.s.high += ((UWtype) uu.s.low * (UWtype) vv.s.high | |
4357 | + + (UWtype) uu.s.high * (UWtype) vv.s.low); | |
4358 | + | |
4359 | + return w.ll; | |
4360 | +} | |
4361 | +#endif | |
4362 | +\f | |
4363 | +#if (defined (L_udivdi3) || defined (L_divdi3) || \ | |
4364 | + defined (L_umoddi3) || defined (L_moddi3)) | |
4365 | +#if defined (sdiv_qrnnd) | |
4366 | +#define L_udiv_w_sdiv | |
4367 | +#endif | |
4368 | +#endif | |
4369 | + | |
4370 | +#ifdef L_udiv_w_sdiv | |
4371 | +#if defined (sdiv_qrnnd) | |
4372 | +#if (defined (L_udivdi3) || defined (L_divdi3) || \ | |
4373 | + defined (L_umoddi3) || defined (L_moddi3)) | |
4374 | +static inline __attribute__ ((__always_inline__)) | |
4375 | +#endif | |
4376 | +UWtype | |
4377 | +__udiv_w_sdiv (UWtype *rp, UWtype a1, UWtype a0, UWtype d) | |
4378 | +{ | |
4379 | + UWtype q, r; | |
4380 | + UWtype c0, c1, b1; | |
4381 | + | |
4382 | + if ((Wtype) d >= 0) | |
4383 | + { | |
4384 | + if (a1 < d - a1 - (a0 >> (W_TYPE_SIZE - 1))) | |
4385 | + { | |
4386 | + /* dividend, divisor, and quotient are nonnegative */ | |
4387 | + sdiv_qrnnd (q, r, a1, a0, d); | |
4388 | + } | |
4389 | + else | |
4390 | + { | |
4391 | + /* Compute c1*2^32 + c0 = a1*2^32 + a0 - 2^31*d */ | |
4392 | + sub_ddmmss (c1, c0, a1, a0, d >> 1, d << (W_TYPE_SIZE - 1)); | |
4393 | + /* Divide (c1*2^32 + c0) by d */ | |
4394 | + sdiv_qrnnd (q, r, c1, c0, d); | |
4395 | + /* Add 2^31 to quotient */ | |
4396 | + q += (UWtype) 1 << (W_TYPE_SIZE - 1); | |
4397 | + } | |
4398 | + } | |
4399 | + else | |
4400 | + { | |
4401 | + b1 = d >> 1; /* d/2, between 2^30 and 2^31 - 1 */ | |
4402 | + c1 = a1 >> 1; /* A/2 */ | |
4403 | + c0 = (a1 << (W_TYPE_SIZE - 1)) + (a0 >> 1); | |
4404 | + | |
4405 | + if (a1 < b1) /* A < 2^32*b1, so A/2 < 2^31*b1 */ | |
4406 | + { | |
4407 | + sdiv_qrnnd (q, r, c1, c0, b1); /* (A/2) / (d/2) */ | |
4408 | + | |
4409 | + r = 2*r + (a0 & 1); /* Remainder from A/(2*b1) */ | |
4410 | + if ((d & 1) != 0) | |
4411 | + { | |
4412 | + if (r >= q) | |
4413 | + r = r - q; | |
4414 | + else if (q - r <= d) | |
4415 | + { | |
4416 | + r = r - q + d; | |
4417 | + q--; | |
4418 | + } | |
4419 | + else | |
4420 | + { | |
4421 | + r = r - q + 2*d; | |
4422 | + q -= 2; | |
4423 | + } | |
4424 | + } | |
4425 | + } | |
4426 | + else if (c1 < b1) /* So 2^31 <= (A/2)/b1 < 2^32 */ | |
4427 | + { | |
4428 | + c1 = (b1 - 1) - c1; | |
4429 | + c0 = ~c0; /* logical NOT */ | |
4430 | + | |
4431 | + sdiv_qrnnd (q, r, c1, c0, b1); /* (A/2) / (d/2) */ | |
4432 | + | |
4433 | + q = ~q; /* (A/2)/b1 */ | |
4434 | + r = (b1 - 1) - r; | |
4435 | + | |
4436 | + r = 2*r + (a0 & 1); /* A/(2*b1) */ | |
4437 | + | |
4438 | + if ((d & 1) != 0) | |
4439 | + { | |
4440 | + if (r >= q) | |
4441 | + r = r - q; | |
4442 | + else if (q - r <= d) | |
4443 | + { | |
4444 | + r = r - q + d; | |
4445 | + q--; | |
4446 | + } | |
4447 | + else | |
4448 | + { | |
4449 | + r = r - q + 2*d; | |
4450 | + q -= 2; | |
4451 | + } | |
4452 | + } | |
4453 | + } | |
4454 | + else /* Implies c1 = b1 */ | |
4455 | + { /* Hence a1 = d - 1 = 2*b1 - 1 */ | |
4456 | + if (a0 >= -d) | |
4457 | + { | |
4458 | + q = -1; | |
4459 | + r = a0 + d; | |
4460 | + } | |
4461 | + else | |
4462 | + { | |
4463 | + q = -2; | |
4464 | + r = a0 + 2*d; | |
4465 | + } | |
4466 | + } | |
4467 | + } | |
4468 | + | |
4469 | + *rp = r; | |
4470 | + return q; | |
4471 | +} | |
4472 | +#else | |
4473 | +/* If sdiv_qrnnd doesn't exist, define dummy __udiv_w_sdiv. */ | |
4474 | +UWtype | |
4475 | +__udiv_w_sdiv (UWtype *rp __attribute__ ((__unused__)), | |
4476 | + UWtype a1 __attribute__ ((__unused__)), | |
4477 | + UWtype a0 __attribute__ ((__unused__)), | |
4478 | + UWtype d __attribute__ ((__unused__))) | |
4479 | +{ | |
4480 | + return 0; | |
4481 | +} | |
4482 | +#endif | |
4483 | +#endif | |
4484 | +\f | |
4485 | +#if (defined (L_udivdi3) || defined (L_divdi3) || \ | |
4486 | + defined (L_umoddi3) || defined (L_moddi3)) | |
4487 | +#define L_udivmoddi4 | |
4488 | +#endif | |
4489 | + | |
4490 | +#ifdef L_clz | |
4491 | +const UQItype __clz_tab[] = | |
4492 | +{ | |
4493 | + 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5, | |
4494 | + 6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6, | |
4495 | + 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, | |
4496 | + 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, | |
4497 | + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, | |
4498 | + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, | |
4499 | + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, | |
4500 | + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, | |
4501 | +}; | |
4502 | +#endif | |
4503 | +\f | |
4504 | +#ifdef L_clzsi2 | |
4505 | +#undef int | |
4506 | +extern int __clzSI2 (UWtype x); | |
4507 | +int | |
4508 | +__clzSI2 (UWtype x) | |
4509 | +{ | |
4510 | + Wtype ret; | |
4511 | + | |
4512 | + count_leading_zeros (ret, x); | |
4513 | + | |
4514 | + return ret; | |
4515 | +} | |
4516 | +#endif | |
4517 | +\f | |
4518 | +#ifdef L_clzdi2 | |
4519 | +#undef int | |
4520 | +extern int __clzDI2 (UDWtype x); | |
4521 | +int | |
4522 | +__clzDI2 (UDWtype x) | |
4523 | +{ | |
4524 | + const DWunion uu = {.ll = x}; | |
4525 | + UWtype word; | |
4526 | + Wtype ret, add; | |
4527 | + | |
4528 | + if (uu.s.high) | |
4529 | + word = uu.s.high, add = 0; | |
4530 | + else | |
4531 | + word = uu.s.low, add = W_TYPE_SIZE; | |
4532 | + | |
4533 | + count_leading_zeros (ret, word); | |
4534 | + return ret + add; | |
4535 | +} | |
4536 | +#endif | |
4537 | +\f | |
4538 | +#ifdef L_ctzsi2 | |
4539 | +#undef int | |
4540 | +extern int __ctzSI2 (UWtype x); | |
4541 | +int | |
4542 | +__ctzSI2 (UWtype x) | |
4543 | +{ | |
4544 | + Wtype ret; | |
4545 | + | |
4546 | + count_trailing_zeros (ret, x); | |
4547 | + | |
4548 | + return ret; | |
4549 | +} | |
4550 | +#endif | |
4551 | +\f | |
4552 | +#ifdef L_ctzdi2 | |
4553 | +#undef int | |
4554 | +extern int __ctzDI2 (UDWtype x); | |
4555 | +int | |
4556 | +__ctzDI2 (UDWtype x) | |
4557 | +{ | |
4558 | + const DWunion uu = {.ll = x}; | |
4559 | + UWtype word; | |
4560 | + Wtype ret, add; | |
4561 | + | |
4562 | + if (uu.s.low) | |
4563 | + word = uu.s.low, add = 0; | |
4564 | + else | |
4565 | + word = uu.s.high, add = W_TYPE_SIZE; | |
4566 | + | |
4567 | + count_trailing_zeros (ret, word); | |
4568 | + return ret + add; | |
4569 | +} | |
4570 | +#endif | |
4571 | + | |
4572 | +#if (defined (L_popcountsi2) || defined (L_popcountdi2) \ | |
4573 | + || defined (L_popcount_tab)) | |
4574 | +extern const UQItype __popcount_tab[] ATTRIBUTE_HIDDEN; | |
4575 | +#endif | |
4576 | + | |
4577 | +#ifdef L_popcount_tab | |
4578 | +const UQItype __popcount_tab[] = | |
4579 | +{ | |
4580 | + 0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5, | |
4581 | + 1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6, | |
4582 | + 1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6, | |
4583 | + 2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7, | |
4584 | + 1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6, | |
4585 | + 2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7, | |
4586 | + 2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7, | |
4587 | + 3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7,4,5,5,6,5,6,6,7,5,6,6,7,6,7,7,8, | |
4588 | +}; | |
4589 | +#endif | |
4590 | +\f | |
4591 | +#ifdef L_popcountsi2 | |
4592 | +#undef int | |
4593 | +extern int __popcountSI2 (UWtype x); | |
4594 | +int | |
4595 | +__popcountSI2 (UWtype x) | |
4596 | +{ | |
4597 | + UWtype i, ret = 0; | |
4598 | + | |
4599 | + for (i = 0; i < W_TYPE_SIZE; i += 8) | |
4600 | + ret += __popcount_tab[(x >> i) & 0xff]; | |
4601 | + | |
4602 | + return ret; | |
4603 | +} | |
4604 | +#endif | |
4605 | +\f | |
4606 | +#ifdef L_popcountdi2 | |
4607 | +#undef int | |
4608 | +extern int __popcountDI2 (UDWtype x); | |
4609 | +int | |
4610 | +__popcountDI2 (UDWtype x) | |
4611 | +{ | |
4612 | + UWtype i, ret = 0; | |
4613 | + | |
4614 | + for (i = 0; i < 2*W_TYPE_SIZE; i += 8) | |
4615 | + ret += __popcount_tab[(x >> i) & 0xff]; | |
4616 | + | |
4617 | + return ret; | |
4618 | +} | |
4619 | +#endif | |
4620 | +\f | |
4621 | +#ifdef L_paritysi2 | |
4622 | +#undef int | |
4623 | +extern int __paritySI2 (UWtype x); | |
4624 | +int | |
4625 | +__paritySI2 (UWtype x) | |
4626 | +{ | |
4627 | +#if W_TYPE_SIZE > 64 | |
4628 | +# error "fill out the table" | |
4629 | +#endif | |
4630 | +#if W_TYPE_SIZE > 32 | |
4631 | + x ^= x >> 32; | |
4632 | +#endif | |
4633 | +#if W_TYPE_SIZE > 16 | |
4634 | + x ^= x >> 16; | |
4635 | +#endif | |
4636 | + x ^= x >> 8; | |
4637 | + x ^= x >> 4; | |
4638 | + x &= 0xf; | |
4639 | + return (0x6996 >> x) & 1; | |
4640 | +} | |
4641 | +#endif | |
4642 | +\f | |
4643 | +#ifdef L_paritydi2 | |
4644 | +#undef int | |
4645 | +extern int __parityDI2 (UDWtype x); | |
4646 | +int | |
4647 | +__parityDI2 (UDWtype x) | |
4648 | +{ | |
4649 | + const DWunion uu = {.ll = x}; | |
4650 | + UWtype nx = uu.s.low ^ uu.s.high; | |
4651 | + | |
4652 | +#if W_TYPE_SIZE > 64 | |
4653 | +# error "fill out the table" | |
4654 | +#endif | |
4655 | +#if W_TYPE_SIZE > 32 | |
4656 | + nx ^= nx >> 32; | |
4657 | +#endif | |
4658 | +#if W_TYPE_SIZE > 16 | |
4659 | + nx ^= nx >> 16; | |
4660 | +#endif | |
4661 | + nx ^= nx >> 8; | |
4662 | + nx ^= nx >> 4; | |
4663 | + nx &= 0xf; | |
4664 | + return (0x6996 >> nx) & 1; | |
4665 | +} | |
4666 | +#endif | |
4667 | + | |
4668 | +#ifdef L_udivmoddi4 | |
4669 | + | |
4670 | +#if (defined (L_udivdi3) || defined (L_divdi3) || \ | |
4671 | + defined (L_umoddi3) || defined (L_moddi3)) | |
4672 | +static inline __attribute__ ((__always_inline__)) | |
4673 | +#endif | |
4674 | +UDWtype | |
4675 | +__udivmoddi4 (UDWtype n, UDWtype d, UDWtype *rp) | |
4676 | +{ | |
4677 | + const DWunion nn = {.ll = n}; | |
4678 | + const DWunion dd = {.ll = d}; | |
4679 | + DWunion rr; | |
4680 | + UWtype d0, d1, n0, n1, n2; | |
4681 | + UWtype q0, q1; | |
4682 | + UWtype b, bm; | |
4683 | + | |
4684 | + d0 = dd.s.low; | |
4685 | + d1 = dd.s.high; | |
4686 | + n0 = nn.s.low; | |
4687 | + n1 = nn.s.high; | |
4688 | + | |
4689 | +#if !UDIV_NEEDS_NORMALIZATION | |
4690 | + if (d1 == 0) | |
4691 | + { | |
4692 | + if (d0 > n1) | |
4693 | + { | |
4694 | + /* 0q = nn / 0D */ | |
4695 | + | |
4696 | + udiv_qrnnd (q0, n0, n1, n0, d0); | |
4697 | + q1 = 0; | |
4698 | + | |
4699 | + /* Remainder in n0. */ | |
4700 | + } | |
4701 | + else | |
4702 | + { | |
4703 | + /* qq = NN / 0d */ | |
4704 | + | |
4705 | + if (d0 == 0) | |
4706 | + d0 = 1 / d0; /* Divide intentionally by zero. */ | |
4707 | + | |
4708 | + udiv_qrnnd (q1, n1, 0, n1, d0); | |
4709 | + udiv_qrnnd (q0, n0, n1, n0, d0); | |
4710 | + | |
4711 | + /* Remainder in n0. */ | |
4712 | + } | |
4713 | + | |
4714 | + if (rp != 0) | |
4715 | + { | |
4716 | + rr.s.low = n0; | |
4717 | + rr.s.high = 0; | |
4718 | + *rp = rr.ll; | |
4719 | + } | |
4720 | + } | |
4721 | + | |
4722 | +#else /* UDIV_NEEDS_NORMALIZATION */ | |
4723 | + | |
4724 | + if (d1 == 0) | |
4725 | + { | |
4726 | + if (d0 > n1) | |
4727 | + { | |
4728 | + /* 0q = nn / 0D */ | |
4729 | + | |
4730 | + count_leading_zeros (bm, d0); | |
4731 | + | |
4732 | + if (bm != 0) | |
4733 | + { | |
4734 | + /* Normalize, i.e. make the most significant bit of the | |
4735 | + denominator set. */ | |
4736 | + | |
4737 | + d0 = d0 << bm; | |
4738 | + n1 = (n1 << bm) | (n0 >> (W_TYPE_SIZE - bm)); | |
4739 | + n0 = n0 << bm; | |
4740 | + } | |
4741 | + | |
4742 | + udiv_qrnnd (q0, n0, n1, n0, d0); | |
4743 | + q1 = 0; | |
4744 | + | |
4745 | + /* Remainder in n0 >> bm. */ | |
4746 | + } | |
4747 | + else | |
4748 | + { | |
4749 | + /* qq = NN / 0d */ | |
4750 | + | |
4751 | + if (d0 == 0) | |
4752 | + d0 = 1 / d0; /* Divide intentionally by zero. */ | |
4753 | + | |
4754 | + count_leading_zeros (bm, d0); | |
4755 | + | |
4756 | + if (bm == 0) | |
4757 | + { | |
4758 | + /* From (n1 >= d0) /\ (the most significant bit of d0 is set), | |
4759 | + conclude (the most significant bit of n1 is set) /\ (the | |
4760 | + leading quotient digit q1 = 1). | |
4761 | + | |
4762 | + This special case is necessary, not an optimization. | |
4763 | + (Shifts counts of W_TYPE_SIZE are undefined.) */ | |
4764 | + | |
4765 | + n1 -= d0; | |
4766 | + q1 = 1; | |
4767 | + } | |
4768 | + else | |
4769 | + { | |
4770 | + /* Normalize. */ | |
4771 | + | |
4772 | + b = W_TYPE_SIZE - bm; | |
4773 | + | |
4774 | + d0 = d0 << bm; | |
4775 | + n2 = n1 >> b; | |
4776 | + n1 = (n1 << bm) | (n0 >> b); | |
4777 | + n0 = n0 << bm; | |
4778 | + | |
4779 | + udiv_qrnnd (q1, n1, n2, n1, d0); | |
4780 | + } | |
4781 | + | |
4782 | + /* n1 != d0... */ | |
4783 | + | |
4784 | + udiv_qrnnd (q0, n0, n1, n0, d0); | |
4785 | + | |
4786 | + /* Remainder in n0 >> bm. */ | |
4787 | + } | |
4788 | + | |
4789 | + if (rp != 0) | |
4790 | + { | |
4791 | + rr.s.low = n0 >> bm; | |
4792 | + rr.s.high = 0; | |
4793 | + *rp = rr.ll; | |
4794 | + } | |
4795 | + } | |
4796 | +#endif /* UDIV_NEEDS_NORMALIZATION */ | |
4797 | + | |
4798 | + else | |
4799 | + { | |
4800 | + if (d1 > n1) | |
4801 | + { | |
4802 | + /* 00 = nn / DD */ | |
4803 | + | |
4804 | + q0 = 0; | |
4805 | + q1 = 0; | |
4806 | + | |
4807 | + /* Remainder in n1n0. */ | |
4808 | + if (rp != 0) | |
4809 | + { | |
4810 | + rr.s.low = n0; | |
4811 | + rr.s.high = n1; | |
4812 | + *rp = rr.ll; | |
4813 | + } | |
4814 | + } | |
4815 | + else | |
4816 | + { | |
4817 | + /* 0q = NN / dd */ | |
4818 | + | |
4819 | + count_leading_zeros (bm, d1); | |
4820 | + if (bm == 0) | |
4821 | + { | |
4822 | + /* From (n1 >= d1) /\ (the most significant bit of d1 is set), | |
4823 | + conclude (the most significant bit of n1 is set) /\ (the | |
4824 | + quotient digit q0 = 0 or 1). | |
4825 | + | |
4826 | + This special case is necessary, not an optimization. */ | |
4827 | + | |
4828 | + /* The condition on the next line takes advantage of that | |
4829 | + n1 >= d1 (true due to program flow). */ | |
4830 | + if (n1 > d1 || n0 >= d0) | |
4831 | + { | |
4832 | + q0 = 1; | |
4833 | + sub_ddmmss (n1, n0, n1, n0, d1, d0); | |
4834 | + } | |
4835 | + else | |
4836 | + q0 = 0; | |
4837 | + | |
4838 | + q1 = 0; | |
4839 | + | |
4840 | + if (rp != 0) | |
4841 | + { | |
4842 | + rr.s.low = n0; | |
4843 | + rr.s.high = n1; | |
4844 | + *rp = rr.ll; | |
4845 | + } | |
4846 | + } | |
4847 | + else | |
4848 | + { | |
4849 | + UWtype m1, m0; | |
4850 | + /* Normalize. */ | |
4851 | + | |
4852 | + b = W_TYPE_SIZE - bm; | |
4853 | + | |
4854 | + d1 = (d1 << bm) | (d0 >> b); | |
4855 | + d0 = d0 << bm; | |
4856 | + n2 = n1 >> b; | |
4857 | + n1 = (n1 << bm) | (n0 >> b); | |
4858 | + n0 = n0 << bm; | |
4859 | + | |
4860 | + udiv_qrnnd (q0, n1, n2, n1, d1); | |
4861 | + umul_ppmm (m1, m0, q0, d0); | |
4862 | + | |
4863 | + if (m1 > n1 || (m1 == n1 && m0 > n0)) | |
4864 | + { | |
4865 | + q0--; | |
4866 | + sub_ddmmss (m1, m0, m1, m0, d1, d0); | |
4867 | + } | |
4868 | + | |
4869 | + q1 = 0; | |
4870 | + | |
4871 | + /* Remainder in (n1n0 - m1m0) >> bm. */ | |
4872 | + if (rp != 0) | |
4873 | + { | |
4874 | + sub_ddmmss (n1, n0, n1, n0, m1, m0); | |
4875 | + rr.s.low = (n1 << b) | (n0 >> bm); | |
4876 | + rr.s.high = n1 >> bm; | |
4877 | + *rp = rr.ll; | |
4878 | + } | |
4879 | + } | |
4880 | + } | |
4881 | + } | |
4882 | + | |
4883 | + const DWunion ww = {{.low = q0, .high = q1}}; | |
4884 | + return ww.ll; | |
4885 | +} | |
4886 | +#endif | |
4887 | + | |
4888 | +#ifdef L_divdi3 | |
4889 | +DWtype | |
4890 | +__divdi3 (DWtype u, DWtype v) | |
4891 | +{ | |
4892 | + word_type c = 0; | |
4893 | + DWunion uu = {.ll = u}; | |
4894 | + DWunion vv = {.ll = v}; | |
4895 | + DWtype w; | |
4896 | + | |
4897 | + if (uu.s.high < 0) | |
4898 | + c = ~c, | |
4899 | + uu.ll = -uu.ll; | |
4900 | + if (vv.s.high < 0) | |
4901 | + c = ~c, | |
4902 | + vv.ll = -vv.ll; | |
4903 | + | |
4904 | + w = __udivmoddi4 (uu.ll, vv.ll, (UDWtype *) 0); | |
4905 | + if (c) | |
4906 | + w = -w; | |
4907 | + | |
4908 | + return w; | |
4909 | +} | |
4910 | +#endif | |
4911 | + | |
4912 | +#ifdef L_moddi3 | |
4913 | +DWtype | |
4914 | +__moddi3 (DWtype u, DWtype v) | |
4915 | +{ | |
4916 | + word_type c = 0; | |
4917 | + DWunion uu = {.ll = u}; | |
4918 | + DWunion vv = {.ll = v}; | |
4919 | + DWtype w; | |
4920 | + | |
4921 | + if (uu.s.high < 0) | |
4922 | + c = ~c, | |
4923 | + uu.ll = -uu.ll; | |
4924 | + if (vv.s.high < 0) | |
4925 | + vv.ll = -vv.ll; | |
4926 | + | |
4927 | + (void) __udivmoddi4 (uu.ll, vv.ll, &w); | |
4928 | + if (c) | |
4929 | + w = -w; | |
4930 | + | |
4931 | + return w; | |
4932 | +} | |
4933 | +#endif | |
4934 | + | |
4935 | +#ifdef L_umoddi3 | |
4936 | +UDWtype | |
4937 | +__umoddi3 (UDWtype u, UDWtype v) | |
4938 | +{ | |
4939 | + UDWtype w; | |
4940 | + | |
4941 | + (void) __udivmoddi4 (u, v, &w); | |
4942 | + | |
4943 | + return w; | |
4944 | +} | |
4945 | +#endif | |
4946 | + | |
4947 | +#ifdef L_udivdi3 | |
4948 | +UDWtype | |
4949 | +__udivdi3 (UDWtype n, UDWtype d) | |
4950 | +{ | |
4951 | + return __udivmoddi4 (n, d, (UDWtype *) 0); | |
4952 | +} | |
4953 | +#endif | |
4954 | +\f | |
4955 | +#ifdef L_cmpdi2 | |
4956 | +word_type | |
4957 | +__cmpdi2 (DWtype a, DWtype b) | |
4958 | +{ | |
4959 | + const DWunion au = {.ll = a}; | |
4960 | + const DWunion bu = {.ll = b}; | |
4961 | + | |
4962 | + if (au.s.high < bu.s.high) | |
4963 | + return 0; | |
4964 | + else if (au.s.high > bu.s.high) | |
4965 | + return 2; | |
4966 | + if ((UWtype) au.s.low < (UWtype) bu.s.low) | |
4967 | + return 0; | |
4968 | + else if ((UWtype) au.s.low > (UWtype) bu.s.low) | |
4969 | + return 2; | |
4970 | + return 1; | |
4971 | +} | |
4972 | +#endif | |
4973 | + | |
4974 | +#ifdef L_ucmpdi2 | |
4975 | +word_type | |
4976 | +__ucmpdi2 (DWtype a, DWtype b) | |
4977 | +{ | |
4978 | + const DWunion au = {.ll = a}; | |
4979 | + const DWunion bu = {.ll = b}; | |
4980 | + | |
4981 | + if ((UWtype) au.s.high < (UWtype) bu.s.high) | |
4982 | + return 0; | |
4983 | + else if ((UWtype) au.s.high > (UWtype) bu.s.high) | |
4984 | + return 2; | |
4985 | + if ((UWtype) au.s.low < (UWtype) bu.s.low) | |
4986 | + return 0; | |
4987 | + else if ((UWtype) au.s.low > (UWtype) bu.s.low) | |
4988 | + return 2; | |
4989 | + return 1; | |
4990 | +} | |
4991 | +#endif | |
4992 | +\f | |
4993 | +#if defined(L_fixunstfdi) && (LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 128) | |
4994 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
4995 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
4996 | + | |
4997 | +DWtype | |
4998 | +__fixunstfDI (TFtype a) | |
4999 | +{ | |
5000 | + if (a < 0) | |
5001 | + return 0; | |
5002 | + | |
5003 | + /* Compute high word of result, as a flonum. */ | |
5004 | + const TFtype b = (a / HIGH_WORD_COEFF); | |
5005 | + /* Convert that to fixed (but not to DWtype!), | |
5006 | + and shift it into the high word. */ | |
5007 | + UDWtype v = (UWtype) b; | |
5008 | + v <<= WORD_SIZE; | |
5009 | + /* Remove high part from the TFtype, leaving the low part as flonum. */ | |
5010 | + a -= (TFtype)v; | |
5011 | + /* Convert that to fixed (but not to DWtype!) and add it in. | |
5012 | + Sometimes A comes out negative. This is significant, since | |
5013 | + A has more bits than a long int does. */ | |
5014 | + if (a < 0) | |
5015 | + v -= (UWtype) (- a); | |
5016 | + else | |
5017 | + v += (UWtype) a; | |
5018 | + return v; | |
5019 | +} | |
5020 | +#endif | |
5021 | + | |
5022 | +#if defined(L_fixtfdi) && (LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 128) | |
5023 | +DWtype | |
5024 | +__fixtfdi (TFtype a) | |
5025 | +{ | |
5026 | + if (a < 0) | |
5027 | + return - __fixunstfDI (-a); | |
5028 | + return __fixunstfDI (a); | |
5029 | +} | |
5030 | +#endif | |
5031 | + | |
5032 | +#if defined(L_fixunsxfdi) && (LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 96) | |
5033 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5034 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5035 | + | |
5036 | +DWtype | |
5037 | +__fixunsxfDI (XFtype a) | |
5038 | +{ | |
5039 | + if (a < 0) | |
5040 | + return 0; | |
5041 | + | |
5042 | + /* Compute high word of result, as a flonum. */ | |
5043 | + const XFtype b = (a / HIGH_WORD_COEFF); | |
5044 | + /* Convert that to fixed (but not to DWtype!), | |
5045 | + and shift it into the high word. */ | |
5046 | + UDWtype v = (UWtype) b; | |
5047 | + v <<= WORD_SIZE; | |
5048 | + /* Remove high part from the XFtype, leaving the low part as flonum. */ | |
5049 | + a -= (XFtype)v; | |
5050 | + /* Convert that to fixed (but not to DWtype!) and add it in. | |
5051 | + Sometimes A comes out negative. This is significant, since | |
5052 | + A has more bits than a long int does. */ | |
5053 | + if (a < 0) | |
5054 | + v -= (UWtype) (- a); | |
5055 | + else | |
5056 | + v += (UWtype) a; | |
5057 | + return v; | |
5058 | +} | |
5059 | +#endif | |
5060 | + | |
5061 | +#if defined(L_fixxfdi) && (LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 96) | |
5062 | +DWtype | |
5063 | +__fixxfdi (XFtype a) | |
5064 | +{ | |
5065 | + if (a < 0) | |
5066 | + return - __fixunsxfDI (-a); | |
5067 | + return __fixunsxfDI (a); | |
5068 | +} | |
5069 | +#endif | |
5070 | + | |
5071 | +#ifdef L_fixunsdfdi | |
5072 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5073 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5074 | + | |
5075 | +DWtype | |
5076 | +__fixunsdfDI (DFtype a) | |
5077 | +{ | |
5078 | + /* Get high part of result. The division here will just moves the radix | |
5079 | + point and will not cause any rounding. Then the conversion to integral | |
5080 | + type chops result as desired. */ | |
5081 | + const UWtype hi = a / HIGH_WORD_COEFF; | |
5082 | + | |
5083 | + /* Get low part of result. Convert `hi' to floating type and scale it back, | |
5084 | + then subtract this from the number being converted. This leaves the low | |
5085 | + part. Convert that to integral type. */ | |
5086 | + const UWtype lo = (a - ((DFtype) hi) * HIGH_WORD_COEFF); | |
5087 | + | |
5088 | + /* Assemble result from the two parts. */ | |
5089 | + return ((UDWtype) hi << WORD_SIZE) | lo; | |
5090 | +} | |
5091 | +#endif | |
5092 | + | |
5093 | +#ifdef L_fixdfdi | |
5094 | +DWtype | |
5095 | +__fixdfdi (DFtype a) | |
5096 | +{ | |
5097 | + if (a < 0) | |
5098 | + return - __fixunsdfDI (-a); | |
5099 | + return __fixunsdfDI (a); | |
5100 | +} | |
5101 | +#endif | |
5102 | + | |
5103 | +#ifdef L_fixunssfdi | |
5104 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5105 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5106 | + | |
5107 | +DWtype | |
5108 | +__fixunssfDI (SFtype original_a) | |
5109 | +{ | |
5110 | + /* Convert the SFtype to a DFtype, because that is surely not going | |
5111 | + to lose any bits. Some day someone else can write a faster version | |
5112 | + that avoids converting to DFtype, and verify it really works right. */ | |
5113 | + const DFtype a = original_a; | |
5114 | + | |
5115 | + /* Get high part of result. The division here will just moves the radix | |
5116 | + point and will not cause any rounding. Then the conversion to integral | |
5117 | + type chops result as desired. */ | |
5118 | + const UWtype hi = a / HIGH_WORD_COEFF; | |
5119 | + | |
5120 | + /* Get low part of result. Convert `hi' to floating type and scale it back, | |
5121 | + then subtract this from the number being converted. This leaves the low | |
5122 | + part. Convert that to integral type. */ | |
5123 | + const UWtype lo = (a - ((DFtype) hi) * HIGH_WORD_COEFF); | |
5124 | + | |
5125 | + /* Assemble result from the two parts. */ | |
5126 | + return ((UDWtype) hi << WORD_SIZE) | lo; | |
5127 | +} | |
5128 | +#endif | |
5129 | + | |
5130 | +#ifdef L_fixsfdi | |
5131 | +DWtype | |
5132 | +__fixsfdi (SFtype a) | |
5133 | +{ | |
5134 | + if (a < 0) | |
5135 | + return - __fixunssfDI (-a); | |
5136 | + return __fixunssfDI (a); | |
5137 | +} | |
5138 | +#endif | |
5139 | + | |
5140 | +#if defined(L_floatdixf) && (LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 96) | |
5141 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5142 | +#define HIGH_HALFWORD_COEFF (((UDWtype) 1) << (WORD_SIZE / 2)) | |
5143 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5144 | + | |
5145 | +XFtype | |
5146 | +__floatdixf (DWtype u) | |
5147 | +{ | |
5148 | + XFtype d = (Wtype) (u >> WORD_SIZE); | |
5149 | + d *= HIGH_HALFWORD_COEFF; | |
5150 | + d *= HIGH_HALFWORD_COEFF; | |
5151 | + d += (UWtype) (u & (HIGH_WORD_COEFF - 1)); | |
5152 | + | |
5153 | + return d; | |
5154 | +} | |
5155 | +#endif | |
5156 | + | |
5157 | +#if defined(L_floatditf) && (LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 128) | |
5158 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5159 | +#define HIGH_HALFWORD_COEFF (((UDWtype) 1) << (WORD_SIZE / 2)) | |
5160 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5161 | + | |
5162 | +TFtype | |
5163 | +__floatditf (DWtype u) | |
5164 | +{ | |
5165 | + TFtype d = (Wtype) (u >> WORD_SIZE); | |
5166 | + d *= HIGH_HALFWORD_COEFF; | |
5167 | + d *= HIGH_HALFWORD_COEFF; | |
5168 | + d += (UWtype) (u & (HIGH_WORD_COEFF - 1)); | |
5169 | + | |
5170 | + return d; | |
5171 | +} | |
5172 | +#endif | |
5173 | + | |
5174 | +#ifdef L_floatdidf | |
5175 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5176 | +#define HIGH_HALFWORD_COEFF (((UDWtype) 1) << (WORD_SIZE / 2)) | |
5177 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5178 | + | |
5179 | +DFtype | |
5180 | +__floatdidf (DWtype u) | |
5181 | +{ | |
5182 | + DFtype d = (Wtype) (u >> WORD_SIZE); | |
5183 | + d *= HIGH_HALFWORD_COEFF; | |
5184 | + d *= HIGH_HALFWORD_COEFF; | |
5185 | + d += (UWtype) (u & (HIGH_WORD_COEFF - 1)); | |
5186 | + | |
5187 | + return d; | |
5188 | +} | |
5189 | +#endif | |
5190 | + | |
5191 | +#ifdef L_floatdisf | |
5192 | +#define WORD_SIZE (sizeof (Wtype) * BITS_PER_UNIT) | |
5193 | +#define HIGH_HALFWORD_COEFF (((UDWtype) 1) << (WORD_SIZE / 2)) | |
5194 | +#define HIGH_WORD_COEFF (((UDWtype) 1) << WORD_SIZE) | |
5195 | + | |
5196 | +#define DI_SIZE (sizeof (DWtype) * BITS_PER_UNIT) | |
5197 | +#define DF_SIZE DBL_MANT_DIG | |
5198 | +#define SF_SIZE FLT_MANT_DIG | |
5199 | + | |
5200 | +SFtype | |
5201 | +__floatdisf (DWtype u) | |
5202 | +{ | |
5203 | + /* Protect against double-rounding error. | |
5204 | + Represent any low-order bits, that might be truncated in DFmode, | |
5205 | + by a bit that won't be lost. The bit can go in anywhere below the | |
5206 | + rounding position of the SFmode. A fixed mask and bit position | |
5207 | + handles all usual configurations. It doesn't handle the case | |
5208 | + of 128-bit DImode, however. */ | |
5209 | + if (DF_SIZE < DI_SIZE | |
5210 | + && DF_SIZE > (DI_SIZE - DF_SIZE + SF_SIZE)) | |
5211 | + { | |
5212 | +#define REP_BIT ((UDWtype) 1 << (DI_SIZE - DF_SIZE)) | |
5213 | + if (! (- ((DWtype) 1 << DF_SIZE) < u | |
5214 | + && u < ((DWtype) 1 << DF_SIZE))) | |
5215 | + { | |
5216 | + if ((UDWtype) u & (REP_BIT - 1)) | |
5217 | + { | |
5218 | + u &= ~ (REP_BIT - 1); | |
5219 | + u |= REP_BIT; | |
5220 | + } | |
5221 | + } | |
5222 | + } | |
5223 | + /* Do the calculation in DFmode | |
5224 | + so that we don't lose any of the precision of the high word | |
5225 | + while multiplying it. */ | |
5226 | + DFtype f = (Wtype) (u >> WORD_SIZE); | |
5227 | + f *= HIGH_HALFWORD_COEFF; | |
5228 | + f *= HIGH_HALFWORD_COEFF; | |
5229 | + f += (UWtype) (u & (HIGH_WORD_COEFF - 1)); | |
5230 | + | |
5231 | + return (SFtype) f; | |
5232 | +} | |
5233 | +#endif | |
5234 | + | |
5235 | +#if defined(L_fixunsxfsi) && LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 96 | |
5236 | +/* Reenable the normal types, in case limits.h needs them. */ | |
5237 | +#undef char | |
5238 | +#undef short | |
5239 | +#undef int | |
5240 | +#undef long | |
5241 | +#undef unsigned | |
5242 | +#undef float | |
5243 | +#undef double | |
5244 | +#undef MIN | |
5245 | +#undef MAX | |
5246 | +#include <limits.h> | |
5247 | + | |
5248 | +UWtype | |
5249 | +__fixunsxfSI (XFtype a) | |
5250 | +{ | |
5251 | + if (a >= - (DFtype) Wtype_MIN) | |
5252 | + return (Wtype) (a + Wtype_MIN) - Wtype_MIN; | |
5253 | + return (Wtype) a; | |
5254 | +} | |
5255 | +#endif | |
5256 | + | |
5257 | +#ifdef L_fixunsdfsi | |
5258 | +/* Reenable the normal types, in case limits.h needs them. */ | |
5259 | +#undef char | |
5260 | +#undef short | |
5261 | +#undef int | |
5262 | +#undef long | |
5263 | +#undef unsigned | |
5264 | +#undef float | |
5265 | +#undef double | |
5266 | +#undef MIN | |
5267 | +#undef MAX | |
5268 | +#include <limits.h> | |
5269 | + | |
5270 | +UWtype | |
5271 | +__fixunsdfSI (DFtype a) | |
5272 | +{ | |
5273 | + if (a >= - (DFtype) Wtype_MIN) | |
5274 | + return (Wtype) (a + Wtype_MIN) - Wtype_MIN; | |
5275 | + return (Wtype) a; | |
5276 | +} | |
5277 | +#endif | |
5278 | + | |
5279 | +#ifdef L_fixunssfsi | |
5280 | +/* Reenable the normal types, in case limits.h needs them. */ | |
5281 | +#undef char | |
5282 | +#undef short | |
5283 | +#undef int | |
5284 | +#undef long | |
5285 | +#undef unsigned | |
5286 | +#undef float | |
5287 | +#undef double | |
5288 | +#undef MIN | |
5289 | +#undef MAX | |
5290 | +#include <limits.h> | |
5291 | + | |
5292 | +UWtype | |
5293 | +__fixunssfSI (SFtype a) | |
5294 | +{ | |
5295 | + if (a >= - (SFtype) Wtype_MIN) | |
5296 | + return (Wtype) (a + Wtype_MIN) - Wtype_MIN; | |
5297 | + return (Wtype) a; | |
5298 | +} | |
5299 | +#endif | |
5300 | +\f | |
5301 | +/* From here on down, the routines use normal data types. */ | |
5302 | + | |
5303 | +#define SItype bogus_type | |
5304 | +#define USItype bogus_type | |
5305 | +#define DItype bogus_type | |
5306 | +#define UDItype bogus_type | |
5307 | +#define SFtype bogus_type | |
5308 | +#define DFtype bogus_type | |
5309 | +#undef Wtype | |
5310 | +#undef UWtype | |
5311 | +#undef HWtype | |
5312 | +#undef UHWtype | |
5313 | +#undef DWtype | |
5314 | +#undef UDWtype | |
5315 | + | |
5316 | +#undef char | |
5317 | +#undef short | |
5318 | +#undef int | |
5319 | +#undef long | |
5320 | +#undef unsigned | |
5321 | +#undef float | |
5322 | +#undef double | |
5323 | +\f | |
5324 | +#ifdef L__gcc_bcmp | |
5325 | + | |
5326 | +/* Like bcmp except the sign is meaningful. | |
5327 | + Result is negative if S1 is less than S2, | |
5328 | + positive if S1 is greater, 0 if S1 and S2 are equal. */ | |
5329 | + | |
5330 | +int | |
5331 | +__gcc_bcmp (const unsigned char *s1, const unsigned char *s2, size_t size) | |
5332 | +{ | |
5333 | + while (size > 0) | |
5334 | + { | |
5335 | + const unsigned char c1 = *s1++, c2 = *s2++; | |
5336 | + if (c1 != c2) | |
5337 | + return c1 - c2; | |
5338 | + size--; | |
5339 | + } | |
5340 | + return 0; | |
5341 | +} | |
5342 | + | |
5343 | +#endif | |
5344 | +\f | |
5345 | +/* __eprintf used to be used by GCC's private version of <assert.h>. | |
5346 | + We no longer provide that header, but this routine remains in libgcc.a | |
5347 | + for binary backward compatibility. Note that it is not included in | |
5348 | + the shared version of libgcc. */ | |
5349 | +#ifdef L_eprintf | |
5350 | +#ifndef inhibit_libc | |
5351 | + | |
5352 | +#undef NULL /* Avoid errors if stdio.h and our stddef.h mismatch. */ | |
5353 | +#include <stdio.h> | |
5354 | + | |
5355 | +void | |
5356 | +__eprintf (const char *string, const char *expression, | |
5357 | + unsigned int line, const char *filename) | |
5358 | +{ | |
5359 | + fprintf (stderr, string, expression, line, filename); | |
5360 | + fflush (stderr); | |
5361 | + abort (); | |
5362 | +} | |
5363 | + | |
5364 | +#endif | |
5365 | +#endif | |
5366 | + | |
5367 | +\f | |
5368 | +#ifdef L_clear_cache | |
5369 | +/* Clear part of an instruction cache. */ | |
5370 | + | |
5371 | +void | |
5372 | +__clear_cache (char *beg __attribute__((__unused__)), | |
5373 | + char *end __attribute__((__unused__))) | |
5374 | +{ | |
5375 | +#ifdef CLEAR_INSN_CACHE | |
5376 | + CLEAR_INSN_CACHE (beg, end); | |
5377 | +#endif /* CLEAR_INSN_CACHE */ | |
5378 | +} | |
5379 | + | |
5380 | +#endif /* L_clear_cache */ | |
5381 | +\f | |
5382 | +#ifdef L_trampoline | |
5383 | + | |
5384 | +/* Jump to a trampoline, loading the static chain address. */ | |
5385 | + | |
5386 | +#if defined(WINNT) && ! defined(__CYGWIN__) && ! defined (_UWIN) | |
5387 | + | |
5388 | +long | |
5389 | +getpagesize (void) | |
5390 | +{ | |
5391 | +#ifdef _ALPHA_ | |
5392 | + return 8192; | |
5393 | +#else | |
5394 | + return 4096; | |
5395 | +#endif | |
5396 | +} | |
5397 | + | |
5398 | +#ifdef __i386__ | |
5399 | +extern int VirtualProtect (char *, int, int, int *) __attribute__((stdcall)); | |
5400 | +#endif | |
5401 | + | |
5402 | +int | |
5403 | +mprotect (char *addr, int len, int prot) | |
5404 | +{ | |
5405 | + int np, op; | |
5406 | + | |
5407 | + if (prot == 7) | |
5408 | + np = 0x40; | |
5409 | + else if (prot == 5) | |
5410 | + np = 0x20; | |
5411 | + else if (prot == 4) | |
5412 | + np = 0x10; | |
5413 | + else if (prot == 3) | |
5414 | + np = 0x04; | |
5415 | + else if (prot == 1) | |
5416 | + np = 0x02; | |
5417 | + else if (prot == 0) | |
5418 | + np = 0x01; | |
5419 | + | |
5420 | + if (VirtualProtect (addr, len, np, &op)) | |
5421 | + return 0; | |
5422 | + else | |
5423 | + return -1; | |
5424 | +} | |
5425 | + | |
5426 | +#endif /* WINNT && ! __CYGWIN__ && ! _UWIN */ | |
5427 | + | |
5428 | +#ifdef TRANSFER_FROM_TRAMPOLINE | |
5429 | +TRANSFER_FROM_TRAMPOLINE | |
5430 | +#endif | |
5431 | +#endif /* L_trampoline */ | |
5432 | +\f | |
5433 | +#ifndef __CYGWIN__ | |
5434 | +#ifdef L__main | |
5435 | + | |
5436 | +#include "gbl-ctors.h" | |
5437 | +/* Some systems use __main in a way incompatible with its use in gcc, in these | |
5438 | + cases use the macros NAME__MAIN to give a quoted symbol and SYMBOL__MAIN to | |
5439 | + give the same symbol without quotes for an alternative entry point. You | |
5440 | + must define both, or neither. */ | |
5441 | +#ifndef NAME__MAIN | |
5442 | +#define NAME__MAIN "__main" | |
5443 | +#define SYMBOL__MAIN __main | |
5444 | +#endif | |
5445 | + | |
5446 | +#ifdef INIT_SECTION_ASM_OP | |
5447 | +#undef HAS_INIT_SECTION | |
5448 | +#define HAS_INIT_SECTION | |
5449 | +#endif | |
5450 | + | |
5451 | +#if !defined (HAS_INIT_SECTION) || !defined (OBJECT_FORMAT_ELF) | |
5452 | + | |
5453 | +/* Some ELF crosses use crtstuff.c to provide __CTOR_LIST__, but use this | |
5454 | + code to run constructors. In that case, we need to handle EH here, too. */ | |
5455 | + | |
5456 | +#ifdef EH_FRAME_SECTION_NAME | |
5457 | +#include "unwind-dw2-fde.h" | |
5458 | +extern unsigned char __EH_FRAME_BEGIN__[]; | |
5459 | +#endif | |
5460 | + | |
5461 | +/* Run all the global destructors on exit from the program. */ | |
5462 | + | |
5463 | +void | |
5464 | +__do_global_dtors (void) | |
5465 | +{ | |
5466 | +#ifdef DO_GLOBAL_DTORS_BODY | |
5467 | + DO_GLOBAL_DTORS_BODY; | |
5468 | +#else | |
5469 | + static func_ptr *p = __DTOR_LIST__ + 1; | |
5470 | + while (*p) | |
5471 | + { | |
5472 | + p++; | |
5473 | + (*(p-1)) (); | |
5474 | + } | |
5475 | +#endif | |
5476 | +#if defined (EH_FRAME_SECTION_NAME) && !defined (HAS_INIT_SECTION) | |
5477 | + { | |
5478 | + static int completed = 0; | |
5479 | + if (! completed) | |
5480 | + { | |
5481 | + completed = 1; | |
5482 | + __deregister_frame_info (__EH_FRAME_BEGIN__); | |
5483 | + } | |
5484 | + } | |
5485 | +#endif | |
5486 | +} | |
5487 | +#endif | |
5488 | + | |
5489 | +#ifndef HAS_INIT_SECTION | |
5490 | +/* Run all the global constructors on entry to the program. */ | |
5491 | + | |
5492 | +void | |
5493 | +__do_global_ctors (void) | |
5494 | +{ | |
5495 | +#ifdef EH_FRAME_SECTION_NAME | |
5496 | + { | |
5497 | + static struct object object; | |
5498 | + __register_frame_info (__EH_FRAME_BEGIN__, &object); | |
5499 | + } | |
5500 | +#endif | |
5501 | + DO_GLOBAL_CTORS_BODY; | |
5502 | + atexit (__do_global_dtors); | |
5503 | +} | |
5504 | +#endif /* no HAS_INIT_SECTION */ | |
5505 | + | |
5506 | +#if !defined (HAS_INIT_SECTION) || defined (INVOKE__main) | |
5507 | +/* Subroutine called automatically by `main'. | |
5508 | + Compiling a global function named `main' | |
5509 | + produces an automatic call to this function at the beginning. | |
5510 | + | |
5511 | + For many systems, this routine calls __do_global_ctors. | |
5512 | + For systems which support a .init section we use the .init section | |
5513 | + to run __do_global_ctors, so we need not do anything here. */ | |
5514 | + | |
5515 | +extern void SYMBOL__MAIN (void); | |
5516 | +void | |
5517 | +SYMBOL__MAIN (void) | |
5518 | +{ | |
5519 | + /* Support recursive calls to `main': run initializers just once. */ | |
5520 | + static int initialized; | |
5521 | + if (! initialized) | |
5522 | + { | |
5523 | + initialized = 1; | |
5524 | + __do_global_ctors (); | |
5525 | + } | |
5526 | +} | |
5527 | +#endif /* no HAS_INIT_SECTION or INVOKE__main */ | |
5528 | + | |
5529 | +#endif /* L__main */ | |
5530 | +#endif /* __CYGWIN__ */ | |
5531 | +\f | |
5532 | +#ifdef L_ctors | |
5533 | + | |
5534 | +#include "gbl-ctors.h" | |
5535 | + | |
5536 | +/* Provide default definitions for the lists of constructors and | |
5537 | + destructors, so that we don't get linker errors. These symbols are | |
5538 | + intentionally bss symbols, so that gld and/or collect will provide | |
5539 | + the right values. */ | |
5540 | + | |
5541 | +/* We declare the lists here with two elements each, | |
5542 | + so that they are valid empty lists if no other definition is loaded. | |
5543 | + | |
5544 | + If we are using the old "set" extensions to have the gnu linker | |
5545 | + collect ctors and dtors, then we __CTOR_LIST__ and __DTOR_LIST__ | |
5546 | + must be in the bss/common section. | |
5547 | + | |
5548 | + Long term no port should use those extensions. But many still do. */ | |
5549 | +#if !defined(INIT_SECTION_ASM_OP) && !defined(CTOR_LISTS_DEFINED_EXTERNALLY) | |
5550 | +#if defined (TARGET_ASM_CONSTRUCTOR) || defined (USE_COLLECT2) | |
5551 | +func_ptr __CTOR_LIST__[2] = {0, 0}; | |
5552 | +func_ptr __DTOR_LIST__[2] = {0, 0}; | |
5553 | +#else | |
5554 | +func_ptr __CTOR_LIST__[2]; | |
5555 | +func_ptr __DTOR_LIST__[2]; | |
5556 | +#endif | |
5557 | +#endif /* no INIT_SECTION_ASM_OP and not CTOR_LISTS_DEFINED_EXTERNALLY */ | |
5558 | +#endif /* L_ctors */ | |
5559 | + | |
5560 | --- ./gcc/gcc/libgcc2.h | |
5561 | +++ ./gcc/gcc/libgcc2.h | |
5562 | @@ -0,0 +1,310 @@ | |
5563 | +/* Header file for libgcc2.c. */ | |
5564 | +/* Copyright (C) 2000, 2001 | |
5565 | + Free Software Foundation, Inc. | |
5566 | + | |
5567 | +This file is part of GCC. | |
5568 | + | |
5569 | +GCC is free software; you can redistribute it and/or modify it under | |
5570 | +the terms of the GNU General Public License as published by the Free | |
5571 | +Software Foundation; either version 2, or (at your option) any later | |
5572 | +version. | |
5573 | + | |
5574 | +GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
5575 | +WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
5576 | +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
5577 | +for more details. | |
5578 | + | |
5579 | +You should have received a copy of the GNU General Public License | |
5580 | +along with GCC; see the file COPYING. If not, write to the Free | |
5581 | +Software Foundation, 59 Temple Place - Suite 330, Boston, MA | |
5582 | +02111-1307, USA. */ | |
5583 | + | |
5584 | +/* As a special exception, if you link this library with other files, | |
5585 | + some of which are compiled with GCC, to produce an executable, | |
5586 | + this library does not by itself cause the resulting executable | |
5587 | + to be covered by the GNU General Public License. | |
5588 | + This exception does not however invalidate any other reasons why | |
5589 | + the executable file might be covered by the GNU General Public License. */ | |
5590 | + | |
5591 | + | |
5592 | +#ifndef GCC_LIBGCC2_H | |
5593 | +#define GCC_LIBGCC2_H | |
5594 | + | |
5595 | +extern int __gcc_bcmp (const unsigned char *, const unsigned char *, size_t); | |
5596 | +extern void __clear_cache (char *, char *); | |
5597 | +extern void __eprintf (const char *, const char *, unsigned int, const char *) | |
5598 | + __attribute__ ((__noreturn__)); | |
5599 | + | |
5600 | +struct exception_descriptor; | |
5601 | +extern short int __get_eh_table_language (struct exception_descriptor *); | |
5602 | +extern short int __get_eh_table_version (struct exception_descriptor *); | |
5603 | + | |
5604 | +/* Permit the tm.h file to select the endianness to use just for this | |
5605 | + file. This is used when the endianness is determined when the | |
5606 | + compiler is run. */ | |
5607 | + | |
5608 | +#ifndef LIBGCC2_WORDS_BIG_ENDIAN | |
5609 | +#define LIBGCC2_WORDS_BIG_ENDIAN WORDS_BIG_ENDIAN | |
5610 | +#endif | |
5611 | + | |
5612 | +#ifndef LIBGCC2_LONG_DOUBLE_TYPE_SIZE | |
5613 | +#define LIBGCC2_LONG_DOUBLE_TYPE_SIZE LONG_DOUBLE_TYPE_SIZE | |
5614 | +#endif | |
5615 | + | |
5616 | +#ifndef MIN_UNITS_PER_WORD | |
5617 | +#define MIN_UNITS_PER_WORD UNITS_PER_WORD | |
5618 | +#endif | |
5619 | + | |
5620 | +/* In the first part of this file, we are interfacing to calls generated | |
5621 | + by the compiler itself. These calls pass values into these routines | |
5622 | + which have very specific modes (rather than very specific types), and | |
5623 | + these compiler-generated calls also expect any return values to have | |
5624 | + very specific modes (rather than very specific types). Thus, we need | |
5625 | + to avoid using regular C language type names in this part of the file | |
5626 | + because the sizes for those types can be configured to be anything. | |
5627 | + Instead we use the following special type names. */ | |
5628 | + | |
5629 | +typedef int QItype __attribute__ ((mode (QI))); | |
5630 | +typedef unsigned int UQItype __attribute__ ((mode (QI))); | |
5631 | +typedef int HItype __attribute__ ((mode (HI))); | |
5632 | +typedef unsigned int UHItype __attribute__ ((mode (HI))); | |
5633 | +#if MIN_UNITS_PER_WORD > 1 | |
5634 | +/* These typedefs are usually forbidden on dsp's with UNITS_PER_WORD 1. */ | |
5635 | +typedef int SItype __attribute__ ((mode (SI))); | |
5636 | +typedef unsigned int USItype __attribute__ ((mode (SI))); | |
5637 | +#if LONG_LONG_TYPE_SIZE > 32 | |
5638 | +/* These typedefs are usually forbidden on archs with UNITS_PER_WORD 2. */ | |
5639 | +typedef int DItype __attribute__ ((mode (DI))); | |
5640 | +typedef unsigned int UDItype __attribute__ ((mode (DI))); | |
5641 | +#if MIN_UNITS_PER_WORD > 4 | |
5642 | +/* These typedefs are usually forbidden on archs with UNITS_PER_WORD 4. */ | |
5643 | +typedef int TItype __attribute__ ((mode (TI))); | |
5644 | +typedef unsigned int UTItype __attribute__ ((mode (TI))); | |
5645 | +#endif | |
5646 | +#endif | |
5647 | +#endif | |
5648 | + | |
5649 | +#if BITS_PER_UNIT == 8 | |
5650 | + | |
5651 | +typedef float SFtype __attribute__ ((mode (SF))); | |
5652 | +typedef float DFtype __attribute__ ((mode (DF))); | |
5653 | + | |
5654 | +#if LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 96 | |
5655 | +typedef float XFtype __attribute__ ((mode (XF))); | |
5656 | +#endif | |
5657 | +#if LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 128 | |
5658 | +typedef float TFtype __attribute__ ((mode (TF))); | |
5659 | +#endif | |
5660 | + | |
5661 | +#else /* BITS_PER_UNIT != 8 */ | |
5662 | + | |
5663 | +/* On dsp's there are usually qf/hf/tqf modes used instead of the above. | |
5664 | + For now we don't support them in libgcc2.c. */ | |
5665 | + | |
5666 | +#undef L_fixdfdi | |
5667 | +#undef L_fixsfdi | |
5668 | +#undef L_fixtfdi | |
5669 | +#undef L_fixunsdfdi | |
5670 | +#undef L_fixunsdfsi | |
5671 | +#undef L_fixunssfdi | |
5672 | +#undef L_fixunssfsi | |
5673 | +#undef L_fixunstfdi | |
5674 | +#undef L_fixunsxfdi | |
5675 | +#undef L_fixunsxfsi | |
5676 | +#undef L_fixxfdi | |
5677 | +#undef L_floatdidf | |
5678 | +#undef L_floatdisf | |
5679 | +#undef L_floatditf | |
5680 | +#undef L_floatdixf | |
5681 | + | |
5682 | +#endif /* BITS_PER_UNIT != 8 */ | |
5683 | + | |
5684 | +typedef int word_type __attribute__ ((mode (__word__))); | |
5685 | + | |
5686 | +/* Make sure that we don't accidentally use any normal C language built-in | |
5687 | + type names in the first part of this file. Instead we want to use *only* | |
5688 | + the type names defined above. The following macro definitions insure | |
5689 | + that if we *do* accidentally use some normal C language built-in type name, | |
5690 | + we will get a syntax error. */ | |
5691 | + | |
5692 | +#define char bogus_type | |
5693 | +#define short bogus_type | |
5694 | +#define int bogus_type | |
5695 | +#define long bogus_type | |
5696 | +#define unsigned bogus_type | |
5697 | +#define float bogus_type | |
5698 | +#define double bogus_type | |
5699 | + | |
5700 | +#if MIN_UNITS_PER_WORD > 4 | |
5701 | +#define W_TYPE_SIZE (8 * BITS_PER_UNIT) | |
5702 | +#define Wtype DItype | |
5703 | +#define UWtype UDItype | |
5704 | +#define HWtype DItype | |
5705 | +#define UHWtype UDItype | |
5706 | +#define DWtype TItype | |
5707 | +#define UDWtype UTItype | |
5708 | +#define __NW(a,b) __ ## a ## di ## b | |
5709 | +#define __NDW(a,b) __ ## a ## ti ## b | |
5710 | +#elif MIN_UNITS_PER_WORD > 2 \ | |
5711 | + || (MIN_UNITS_PER_WORD > 1 && LONG_LONG_TYPE_SIZE > 32) | |
5712 | +#define W_TYPE_SIZE (4 * BITS_PER_UNIT) | |
5713 | +#define Wtype SItype | |
5714 | +#define UWtype USItype | |
5715 | +#define HWtype SItype | |
5716 | +#define UHWtype USItype | |
5717 | +#define DWtype DItype | |
5718 | +#define UDWtype UDItype | |
5719 | +#define __NW(a,b) __ ## a ## si ## b | |
5720 | +#define __NDW(a,b) __ ## a ## di ## b | |
5721 | +#elif MIN_UNITS_PER_WORD > 1 | |
5722 | +#define W_TYPE_SIZE (2 * BITS_PER_UNIT) | |
5723 | +#define Wtype HItype | |
5724 | +#define UWtype UHItype | |
5725 | +#define HWtype HItype | |
5726 | +#define UHWtype UHItype | |
5727 | +#define DWtype SItype | |
5728 | +#define UDWtype USItype | |
5729 | +#define __NW(a,b) __ ## a ## hi ## b | |
5730 | +#define __NDW(a,b) __ ## a ## si ## b | |
5731 | +#else | |
5732 | +#define W_TYPE_SIZE BITS_PER_UNIT | |
5733 | +#define Wtype QItype | |
5734 | +#define UWtype UQItype | |
5735 | +#define HWtype QItype | |
5736 | +#define UHWtype UQItype | |
5737 | +#define DWtype HItype | |
5738 | +#define UDWtype UHItype | |
5739 | +#define __NW(a,b) __ ## a ## qi ## b | |
5740 | +#define __NDW(a,b) __ ## a ## hi ## b | |
5741 | +#endif | |
5742 | + | |
5743 | +#define Wtype_MAX ((Wtype)(((UWtype)1 << (W_TYPE_SIZE - 1)) - 1)) | |
5744 | +#define Wtype_MIN (- Wtype_MAX - 1) | |
5745 | + | |
5746 | +#define __muldi3 __NDW(mul,3) | |
5747 | +#define __divdi3 __NDW(div,3) | |
5748 | +#define __udivdi3 __NDW(udiv,3) | |
5749 | +#define __moddi3 __NDW(mod,3) | |
5750 | +#define __umoddi3 __NDW(umod,3) | |
5751 | +#define __negdi2 __NDW(neg,2) | |
5752 | +#define __lshrdi3 __NDW(lshr,3) | |
5753 | +#define __ashldi3 __NDW(ashl,3) | |
5754 | +#define __ashrdi3 __NDW(ashr,3) | |
5755 | +#define __cmpdi2 __NDW(cmp,2) | |
5756 | +#define __ucmpdi2 __NDW(ucmp,2) | |
5757 | +#define __udivmoddi4 __NDW(udivmod,4) | |
5758 | +#define __fixunstfDI __NDW(fixunstf,) | |
5759 | +#define __fixtfdi __NDW(fixtf,) | |
5760 | +#define __fixunsxfDI __NDW(fixunsxf,) | |
5761 | +#define __fixxfdi __NDW(fixxf,) | |
5762 | +#define __fixunsdfDI __NDW(fixunsdf,) | |
5763 | +#define __fixdfdi __NDW(fixdf,) | |
5764 | +#define __fixunssfDI __NDW(fixunssf,) | |
5765 | +#define __fixsfdi __NDW(fixsf,) | |
5766 | +#define __floatdixf __NDW(float,xf) | |
5767 | +#define __floatditf __NDW(float,tf) | |
5768 | +#define __floatdidf __NDW(float,df) | |
5769 | +#define __floatdisf __NDW(float,sf) | |
5770 | +#define __fixunsxfSI __NW(fixunsxf,) | |
5771 | +#define __fixunstfSI __NW(fixunstf,) | |
5772 | +#define __fixunsdfSI __NW(fixunsdf,) | |
5773 | +#define __fixunssfSI __NW(fixunssf,) | |
5774 | + | |
5775 | +#define __ffsSI2 __NW(ffs,2) | |
5776 | +#define __clzSI2 __NW(clz,2) | |
5777 | +#define __ctzSI2 __NW(ctz,2) | |
5778 | +#define __popcountSI2 __NW(popcount,2) | |
5779 | +#define __paritySI2 __NW(parity,2) | |
5780 | +#define __ffsDI2 __NDW(ffs,2) | |
5781 | +#define __clzDI2 __NDW(clz,2) | |
5782 | +#define __ctzDI2 __NDW(ctz,2) | |
5783 | +#define __popcountDI2 __NDW(popcount,2) | |
5784 | +#define __parityDI2 __NDW(parity,2) | |
5785 | + | |
5786 | +extern DWtype __muldi3 (DWtype, DWtype); | |
5787 | +extern DWtype __divdi3 (DWtype, DWtype); | |
5788 | +extern UDWtype __udivdi3 (UDWtype, UDWtype); | |
5789 | +extern UDWtype __umoddi3 (UDWtype, UDWtype); | |
5790 | +extern DWtype __moddi3 (DWtype, DWtype); | |
5791 | + | |
5792 | +/* __udivmoddi4 is static inline when building other libgcc2 portions. */ | |
5793 | +#if (!defined (L_udivdi3) && !defined (L_divdi3) && \ | |
5794 | + !defined (L_umoddi3) && !defined (L_moddi3)) | |
5795 | +extern UDWtype __udivmoddi4 (UDWtype, UDWtype, UDWtype *); | |
5796 | +#endif | |
5797 | + | |
5798 | +/* __negdi2 is static inline when building other libgcc2 portions. */ | |
5799 | +#if !defined(L_divdi3) && !defined(L_moddi3) | |
5800 | +extern DWtype __negdi2 (DWtype); | |
5801 | +#endif | |
5802 | + | |
5803 | +extern DWtype __lshrdi3 (DWtype, word_type); | |
5804 | +extern DWtype __ashldi3 (DWtype, word_type); | |
5805 | +extern DWtype __ashrdi3 (DWtype, word_type); | |
5806 | + | |
5807 | +/* __udiv_w_sdiv is static inline when building other libgcc2 portions. */ | |
5808 | +#if (!defined(L_udivdi3) && !defined(L_divdi3) && \ | |
5809 | + !defined(L_umoddi3) && !defined(L_moddi3)) | |
5810 | +extern UWtype __udiv_w_sdiv (UWtype *, UWtype, UWtype, UWtype); | |
5811 | +#endif | |
5812 | + | |
5813 | +extern word_type __cmpdi2 (DWtype, DWtype); | |
5814 | +extern word_type __ucmpdi2 (DWtype, DWtype); | |
5815 | + | |
5816 | +extern Wtype __absvsi2 (Wtype); | |
5817 | +extern DWtype __absvdi2 (DWtype); | |
5818 | +extern Wtype __addvsi3 (Wtype, Wtype); | |
5819 | +extern DWtype __addvdi3 (DWtype, DWtype); | |
5820 | +extern Wtype __subvsi3 (Wtype, Wtype); | |
5821 | +extern DWtype __subvdi3 (DWtype, DWtype); | |
5822 | +extern Wtype __mulvsi3 (Wtype, Wtype); | |
5823 | +extern DWtype __mulvdi3 (DWtype, DWtype); | |
5824 | +extern Wtype __negvsi2 (Wtype); | |
5825 | +extern DWtype __negvdi2 (DWtype); | |
5826 | + | |
5827 | +#if BITS_PER_UNIT == 8 | |
5828 | +extern DWtype __fixdfdi (DFtype); | |
5829 | +extern DWtype __fixsfdi (SFtype); | |
5830 | +extern DFtype __floatdidf (DWtype); | |
5831 | +extern SFtype __floatdisf (DWtype); | |
5832 | +extern UWtype __fixunsdfSI (DFtype); | |
5833 | +extern UWtype __fixunssfSI (SFtype); | |
5834 | +extern DWtype __fixunsdfDI (DFtype); | |
5835 | +extern DWtype __fixunssfDI (SFtype); | |
5836 | + | |
5837 | +#if LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 96 | |
5838 | +extern DWtype __fixxfdi (XFtype); | |
5839 | +extern DWtype __fixunsxfDI (XFtype); | |
5840 | +extern XFtype __floatdixf (DWtype); | |
5841 | +extern UWtype __fixunsxfSI (XFtype); | |
5842 | +#endif | |
5843 | + | |
5844 | +#if LIBGCC2_LONG_DOUBLE_TYPE_SIZE == 128 | |
5845 | +extern DWtype __fixunstfDI (TFtype); | |
5846 | +extern DWtype __fixtfdi (TFtype); | |
5847 | +extern TFtype __floatditf (DWtype); | |
5848 | +#endif | |
5849 | +#endif /* BITS_PER_UNIT == 8 */ | |
5850 | + | |
5851 | +/* DWstructs are pairs of Wtype values in the order determined by | |
5852 | + LIBGCC2_WORDS_BIG_ENDIAN. */ | |
5853 | + | |
5854 | +#if LIBGCC2_WORDS_BIG_ENDIAN | |
5855 | + struct DWstruct {Wtype high, low;}; | |
5856 | +#else | |
5857 | + struct DWstruct {Wtype low, high;}; | |
5858 | +#endif | |
5859 | + | |
5860 | +/* We need this union to unpack/pack DImode values, since we don't have | |
5861 | + any arithmetic yet. Incoming DImode parameters are stored into the | |
5862 | + `ll' field, and the unpacked result is read from the struct `s'. */ | |
5863 | + | |
5864 | +typedef union | |
5865 | +{ | |
5866 | + struct DWstruct s; | |
5867 | + DWtype ll; | |
5868 | +} DWunion; | |
5869 | + | |
5870 | +#include "longlong.h" | |
5871 | + | |
5872 | +#endif /* ! GCC_LIBGCC2_H */ | |
5873 | --- ./gcc/gcc/longlong.h | |
5874 | +++ ./gcc/gcc/longlong.h | |
5875 | @@ -0,0 +1,1360 @@ | |
5876 | +/* longlong.h -- definitions for mixed size 32/64 bit arithmetic. | |
5877 | + Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000 | |
5878 | + Free Software Foundation, Inc. | |
5879 | + | |
5880 | + This definition file is free software; you can redistribute it | |
5881 | + and/or modify it under the terms of the GNU General Public | |
5882 | + License as published by the Free Software Foundation; either | |
5883 | + version 2, or (at your option) any later version. | |
5884 | + | |
5885 | + This definition file is distributed in the hope that it will be | |
5886 | + useful, but WITHOUT ANY WARRANTY; without even the implied | |
5887 | + warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. | |
5888 | + See the GNU General Public License for more details. | |
5889 | + | |
5890 | + You should have received a copy of the GNU General Public License | |
5891 | + along with this program; if not, write to the Free Software | |
5892 | + Foundation, Inc., 59 Temple Place - Suite 330, | |
5893 | + Boston, MA 02111-1307, USA. */ | |
5894 | + | |
5895 | +/* You have to define the following before including this file: | |
5896 | + | |
5897 | + UWtype -- An unsigned type, default type for operations (typically a "word") | |
5898 | + UHWtype -- An unsigned type, at least half the size of UWtype. | |
5899 | + UDWtype -- An unsigned type, at least twice as large a UWtype | |
5900 | + W_TYPE_SIZE -- size in bits of UWtype | |
5901 | + | |
5902 | + UQItype -- Unsigned 8 bit type. | |
5903 | + SItype, USItype -- Signed and unsigned 32 bit types. | |
5904 | + DItype, UDItype -- Signed and unsigned 64 bit types. | |
5905 | + | |
5906 | + On a 32 bit machine UWtype should typically be USItype; | |
5907 | + on a 64 bit machine, UWtype should typically be UDItype. | |
5908 | +*/ | |
5909 | + | |
5910 | +#define __BITS4 (W_TYPE_SIZE / 4) | |
5911 | +#define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2)) | |
5912 | +#define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1)) | |
5913 | +#define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2)) | |
5914 | + | |
5915 | +#ifndef W_TYPE_SIZE | |
5916 | +#define W_TYPE_SIZE 32 | |
5917 | +#define UWtype USItype | |
5918 | +#define UHWtype USItype | |
5919 | +#define UDWtype UDItype | |
5920 | +#endif | |
5921 | + | |
5922 | +/* Define auxiliary asm macros. | |
5923 | + | |
5924 | + 1) umul_ppmm(high_prod, low_prod, multipler, multiplicand) multiplies two | |
5925 | + UWtype integers MULTIPLER and MULTIPLICAND, and generates a two UWtype | |
5926 | + word product in HIGH_PROD and LOW_PROD. | |
5927 | + | |
5928 | + 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a | |
5929 | + UDWtype product. This is just a variant of umul_ppmm. | |
5930 | + | |
5931 | + 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator, | |
5932 | + denominator) divides a UDWtype, composed by the UWtype integers | |
5933 | + HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient | |
5934 | + in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less | |
5935 | + than DENOMINATOR for correct operation. If, in addition, the most | |
5936 | + significant bit of DENOMINATOR must be 1, then the pre-processor symbol | |
5937 | + UDIV_NEEDS_NORMALIZATION is defined to 1. | |
5938 | + | |
5939 | + 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator, | |
5940 | + denominator). Like udiv_qrnnd but the numbers are signed. The quotient | |
5941 | + is rounded towards 0. | |
5942 | + | |
5943 | + 5) count_leading_zeros(count, x) counts the number of zero-bits from the | |
5944 | + msb to the first nonzero bit in the UWtype X. This is the number of | |
5945 | + steps X needs to be shifted left to set the msb. Undefined for X == 0, | |
5946 | + unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value. | |
5947 | + | |
5948 | + 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts | |
5949 | + from the least significant end. | |
5950 | + | |
5951 | + 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1, | |
5952 | + high_addend_2, low_addend_2) adds two UWtype integers, composed by | |
5953 | + HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2 | |
5954 | + respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow | |
5955 | + (i.e. carry out) is not stored anywhere, and is lost. | |
5956 | + | |
5957 | + 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend, | |
5958 | + high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers, | |
5959 | + composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and | |
5960 | + LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE | |
5961 | + and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere, | |
5962 | + and is lost. | |
5963 | + | |
5964 | + If any of these macros are left undefined for a particular CPU, | |
5965 | + C macros are used. */ | |
5966 | + | |
5967 | +/* The CPUs come in alphabetical order below. | |
5968 | + | |
5969 | + Please add support for more CPUs here, or improve the current support | |
5970 | + for the CPUs below! | |
5971 | + (E.g. WE32100, IBM360.) */ | |
5972 | + | |
5973 | +#if defined (__GNUC__) && !defined (NO_ASM) | |
5974 | + | |
5975 | +/* We sometimes need to clobber "cc" with gcc2, but that would not be | |
5976 | + understood by gcc1. Use cpp to avoid major code duplication. */ | |
5977 | +#if __GNUC__ < 2 | |
5978 | +#define __CLOBBER_CC | |
5979 | +#define __AND_CLOBBER_CC | |
5980 | +#else /* __GNUC__ >= 2 */ | |
5981 | +#define __CLOBBER_CC : "cc" | |
5982 | +#define __AND_CLOBBER_CC , "cc" | |
5983 | +#endif /* __GNUC__ < 2 */ | |
5984 | + | |
5985 | +#if defined (__alpha) && W_TYPE_SIZE == 64 | |
5986 | +#define umul_ppmm(ph, pl, m0, m1) \ | |
5987 | + do { \ | |
5988 | + UDItype __m0 = (m0), __m1 = (m1); \ | |
5989 | + (ph) = __builtin_alpha_umulh (__m0, __m1); \ | |
5990 | + (pl) = __m0 * __m1; \ | |
5991 | + } while (0) | |
5992 | +#define UMUL_TIME 46 | |
5993 | +#ifndef LONGLONG_STANDALONE | |
5994 | +#define udiv_qrnnd(q, r, n1, n0, d) \ | |
5995 | + do { UDItype __r; \ | |
5996 | + (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \ | |
5997 | + (r) = __r; \ | |
5998 | + } while (0) | |
5999 | +extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype); | |
6000 | +#define UDIV_TIME 220 | |
6001 | +#endif /* LONGLONG_STANDALONE */ | |
6002 | +#ifdef __alpha_cix__ | |
6003 | +#define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X)) | |
6004 | +#define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X)) | |
6005 | +#define COUNT_LEADING_ZEROS_0 64 | |
6006 | +#else | |
6007 | +extern const UQItype __clz_tab[] ATTRIBUTE_HIDDEN; | |
6008 | +#define count_leading_zeros(COUNT,X) \ | |
6009 | + do { \ | |
6010 | + UDItype __xr = (X), __t, __a; \ | |
6011 | + __t = __builtin_alpha_cmpbge (0, __xr); \ | |
6012 | + __a = __clz_tab[__t ^ 0xff] - 1; \ | |
6013 | + __t = __builtin_alpha_extbl (__xr, __a); \ | |
6014 | + (COUNT) = 64 - (__clz_tab[__t] + __a*8); \ | |
6015 | + } while (0) | |
6016 | +#define count_trailing_zeros(COUNT,X) \ | |
6017 | + do { \ | |
6018 | + UDItype __xr = (X), __t, __a; \ | |
6019 | + __t = __builtin_alpha_cmpbge (0, __xr); \ | |
6020 | + __t = ~__t & -~__t; \ | |
6021 | + __a = ((__t & 0xCC) != 0) * 2; \ | |
6022 | + __a += ((__t & 0xF0) != 0) * 4; \ | |
6023 | + __a += ((__t & 0xAA) != 0); \ | |
6024 | + __t = __builtin_alpha_extbl (__xr, __a); \ | |
6025 | + __a <<= 3; \ | |
6026 | + __t &= -__t; \ | |
6027 | + __a += ((__t & 0xCC) != 0) * 2; \ | |
6028 | + __a += ((__t & 0xF0) != 0) * 4; \ | |
6029 | + __a += ((__t & 0xAA) != 0); \ | |
6030 | + (COUNT) = __a; \ | |
6031 | + } while (0) | |
6032 | +#endif /* __alpha_cix__ */ | |
6033 | +#endif /* __alpha */ | |
6034 | + | |
6035 | +#if defined (__arc__) && W_TYPE_SIZE == 32 | |
6036 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6037 | + __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \ | |
6038 | + : "=r" ((USItype) (sh)), \ | |
6039 | + "=&r" ((USItype) (sl)) \ | |
6040 | + : "%r" ((USItype) (ah)), \ | |
6041 | + "rIJ" ((USItype) (bh)), \ | |
6042 | + "%r" ((USItype) (al)), \ | |
6043 | + "rIJ" ((USItype) (bl))) | |
6044 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6045 | + __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \ | |
6046 | + : "=r" ((USItype) (sh)), \ | |
6047 | + "=&r" ((USItype) (sl)) \ | |
6048 | + : "r" ((USItype) (ah)), \ | |
6049 | + "rIJ" ((USItype) (bh)), \ | |
6050 | + "r" ((USItype) (al)), \ | |
6051 | + "rIJ" ((USItype) (bl))) | |
6052 | +/* Call libgcc routine. */ | |
6053 | +#define umul_ppmm(w1, w0, u, v) \ | |
6054 | +do { \ | |
6055 | + DWunion __w; \ | |
6056 | + __w.ll = __umulsidi3 (u, v); \ | |
6057 | + w1 = __w.s.high; \ | |
6058 | + w0 = __w.s.low; \ | |
6059 | +} while (0) | |
6060 | +#define __umulsidi3 __umulsidi3 | |
6061 | +UDItype __umulsidi3 (USItype, USItype); | |
6062 | +#endif | |
6063 | + | |
6064 | +#if defined (__arm__) && W_TYPE_SIZE == 32 | |
6065 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6066 | + __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \ | |
6067 | + : "=r" ((USItype) (sh)), \ | |
6068 | + "=&r" ((USItype) (sl)) \ | |
6069 | + : "%r" ((USItype) (ah)), \ | |
6070 | + "rI" ((USItype) (bh)), \ | |
6071 | + "%r" ((USItype) (al)), \ | |
6072 | + "rI" ((USItype) (bl))) | |
6073 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6074 | + __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \ | |
6075 | + : "=r" ((USItype) (sh)), \ | |
6076 | + "=&r" ((USItype) (sl)) \ | |
6077 | + : "r" ((USItype) (ah)), \ | |
6078 | + "rI" ((USItype) (bh)), \ | |
6079 | + "r" ((USItype) (al)), \ | |
6080 | + "rI" ((USItype) (bl))) | |
6081 | +#define umul_ppmm(xh, xl, a, b) \ | |
6082 | +{register USItype __t0, __t1, __t2; \ | |
6083 | + __asm__ ("%@ Inlined umul_ppmm\n" \ | |
6084 | + " mov %2, %5, lsr #16\n" \ | |
6085 | + " mov %0, %6, lsr #16\n" \ | |
6086 | + " bic %3, %5, %2, lsl #16\n" \ | |
6087 | + " bic %4, %6, %0, lsl #16\n" \ | |
6088 | + " mul %1, %3, %4\n" \ | |
6089 | + " mul %4, %2, %4\n" \ | |
6090 | + " mul %3, %0, %3\n" \ | |
6091 | + " mul %0, %2, %0\n" \ | |
6092 | + " adds %3, %4, %3\n" \ | |
6093 | + " addcs %0, %0, #65536\n" \ | |
6094 | + " adds %1, %1, %3, lsl #16\n" \ | |
6095 | + " adc %0, %0, %3, lsr #16" \ | |
6096 | + : "=&r" ((USItype) (xh)), \ | |
6097 | + "=r" ((USItype) (xl)), \ | |
6098 | + "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \ | |
6099 | + : "r" ((USItype) (a)), \ | |
6100 | + "r" ((USItype) (b)));} | |
6101 | +#define UMUL_TIME 20 | |
6102 | +#define UDIV_TIME 100 | |
6103 | +#endif /* __arm__ */ | |
6104 | + | |
6105 | +#if defined (__hppa) && W_TYPE_SIZE == 32 | |
6106 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6107 | + __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \ | |
6108 | + : "=r" ((USItype) (sh)), \ | |
6109 | + "=&r" ((USItype) (sl)) \ | |
6110 | + : "%rM" ((USItype) (ah)), \ | |
6111 | + "rM" ((USItype) (bh)), \ | |
6112 | + "%rM" ((USItype) (al)), \ | |
6113 | + "rM" ((USItype) (bl))) | |
6114 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6115 | + __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \ | |
6116 | + : "=r" ((USItype) (sh)), \ | |
6117 | + "=&r" ((USItype) (sl)) \ | |
6118 | + : "rM" ((USItype) (ah)), \ | |
6119 | + "rM" ((USItype) (bh)), \ | |
6120 | + "rM" ((USItype) (al)), \ | |
6121 | + "rM" ((USItype) (bl))) | |
6122 | +#if defined (_PA_RISC1_1) | |
6123 | +#define umul_ppmm(w1, w0, u, v) \ | |
6124 | + do { \ | |
6125 | + union \ | |
6126 | + { \ | |
6127 | + UDItype __f; \ | |
6128 | + struct {USItype __w1, __w0;} __w1w0; \ | |
6129 | + } __t; \ | |
6130 | + __asm__ ("xmpyu %1,%2,%0" \ | |
6131 | + : "=x" (__t.__f) \ | |
6132 | + : "x" ((USItype) (u)), \ | |
6133 | + "x" ((USItype) (v))); \ | |
6134 | + (w1) = __t.__w1w0.__w1; \ | |
6135 | + (w0) = __t.__w1w0.__w0; \ | |
6136 | + } while (0) | |
6137 | +#define UMUL_TIME 8 | |
6138 | +#else | |
6139 | +#define UMUL_TIME 30 | |
6140 | +#endif | |
6141 | +#define UDIV_TIME 40 | |
6142 | +#define count_leading_zeros(count, x) \ | |
6143 | + do { \ | |
6144 | + USItype __tmp; \ | |
6145 | + __asm__ ( \ | |
6146 | + "ldi 1,%0\n" \ | |
6147 | +" extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \ | |
6148 | +" extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\ | |
6149 | +" ldo 16(%0),%0 ; Yes. Perform add.\n" \ | |
6150 | +" extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \ | |
6151 | +" extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\ | |
6152 | +" ldo 8(%0),%0 ; Yes. Perform add.\n" \ | |
6153 | +" extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \ | |
6154 | +" extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\ | |
6155 | +" ldo 4(%0),%0 ; Yes. Perform add.\n" \ | |
6156 | +" extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \ | |
6157 | +" extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\ | |
6158 | +" ldo 2(%0),%0 ; Yes. Perform add.\n" \ | |
6159 | +" extru %1,30,1,%1 ; Extract bit 1.\n" \ | |
6160 | +" sub %0,%1,%0 ; Subtract it.\n" \ | |
6161 | + : "=r" (count), "=r" (__tmp) : "1" (x)); \ | |
6162 | + } while (0) | |
6163 | +#endif | |
6164 | + | |
6165 | +#if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32 | |
6166 | +#define smul_ppmm(xh, xl, m0, m1) \ | |
6167 | + do { \ | |
6168 | + union {DItype __ll; \ | |
6169 | + struct {USItype __h, __l;} __i; \ | |
6170 | + } __x; \ | |
6171 | + __asm__ ("lr %N0,%1\n\tmr %0,%2" \ | |
6172 | + : "=&r" (__x.__ll) \ | |
6173 | + : "r" (m0), "r" (m1)); \ | |
6174 | + (xh) = __x.__i.__h; (xl) = __x.__i.__l; \ | |
6175 | + } while (0) | |
6176 | +#define sdiv_qrnnd(q, r, n1, n0, d) \ | |
6177 | + do { \ | |
6178 | + union {DItype __ll; \ | |
6179 | + struct {USItype __h, __l;} __i; \ | |
6180 | + } __x; \ | |
6181 | + __x.__i.__h = n1; __x.__i.__l = n0; \ | |
6182 | + __asm__ ("dr %0,%2" \ | |
6183 | + : "=r" (__x.__ll) \ | |
6184 | + : "0" (__x.__ll), "r" (d)); \ | |
6185 | + (q) = __x.__i.__l; (r) = __x.__i.__h; \ | |
6186 | + } while (0) | |
6187 | +#endif | |
6188 | + | |
6189 | +#if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32 | |
6190 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6191 | + __asm__ ("addl %5,%1\n\tadcl %3,%0" \ | |
6192 | + : "=r" ((USItype) (sh)), \ | |
6193 | + "=&r" ((USItype) (sl)) \ | |
6194 | + : "%0" ((USItype) (ah)), \ | |
6195 | + "g" ((USItype) (bh)), \ | |
6196 | + "%1" ((USItype) (al)), \ | |
6197 | + "g" ((USItype) (bl))) | |
6198 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6199 | + __asm__ ("subl %5,%1\n\tsbbl %3,%0" \ | |
6200 | + : "=r" ((USItype) (sh)), \ | |
6201 | + "=&r" ((USItype) (sl)) \ | |
6202 | + : "0" ((USItype) (ah)), \ | |
6203 | + "g" ((USItype) (bh)), \ | |
6204 | + "1" ((USItype) (al)), \ | |
6205 | + "g" ((USItype) (bl))) | |
6206 | +#define umul_ppmm(w1, w0, u, v) \ | |
6207 | + __asm__ ("mull %3" \ | |
6208 | + : "=a" ((USItype) (w0)), \ | |
6209 | + "=d" ((USItype) (w1)) \ | |
6210 | + : "%0" ((USItype) (u)), \ | |
6211 | + "rm" ((USItype) (v))) | |
6212 | +#define udiv_qrnnd(q, r, n1, n0, dv) \ | |
6213 | + __asm__ ("divl %4" \ | |
6214 | + : "=a" ((USItype) (q)), \ | |
6215 | + "=d" ((USItype) (r)) \ | |
6216 | + : "0" ((USItype) (n0)), \ | |
6217 | + "1" ((USItype) (n1)), \ | |
6218 | + "rm" ((USItype) (dv))) | |
6219 | +#define count_leading_zeros(count, x) \ | |
6220 | + do { \ | |
6221 | + USItype __cbtmp; \ | |
6222 | + __asm__ ("bsrl %1,%0" \ | |
6223 | + : "=r" (__cbtmp) : "rm" ((USItype) (x))); \ | |
6224 | + (count) = __cbtmp ^ 31; \ | |
6225 | + } while (0) | |
6226 | +#define count_trailing_zeros(count, x) \ | |
6227 | + __asm__ ("bsfl %1,%0" : "=r" (count) : "rm" ((USItype)(x))) | |
6228 | +#define UMUL_TIME 40 | |
6229 | +#define UDIV_TIME 40 | |
6230 | +#endif /* 80x86 */ | |
6231 | + | |
6232 | +#if defined (__i960__) && W_TYPE_SIZE == 32 | |
6233 | +#define umul_ppmm(w1, w0, u, v) \ | |
6234 | + ({union {UDItype __ll; \ | |
6235 | + struct {USItype __l, __h;} __i; \ | |
6236 | + } __xx; \ | |
6237 | + __asm__ ("emul %2,%1,%0" \ | |
6238 | + : "=d" (__xx.__ll) \ | |
6239 | + : "%dI" ((USItype) (u)), \ | |
6240 | + "dI" ((USItype) (v))); \ | |
6241 | + (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;}) | |
6242 | +#define __umulsidi3(u, v) \ | |
6243 | + ({UDItype __w; \ | |
6244 | + __asm__ ("emul %2,%1,%0" \ | |
6245 | + : "=d" (__w) \ | |
6246 | + : "%dI" ((USItype) (u)), \ | |
6247 | + "dI" ((USItype) (v))); \ | |
6248 | + __w; }) | |
6249 | +#endif /* __i960__ */ | |
6250 | + | |
6251 | +#if defined (__M32R__) && W_TYPE_SIZE == 32 | |
6252 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6253 | + /* The cmp clears the condition bit. */ \ | |
6254 | + __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \ | |
6255 | + : "=r" ((USItype) (sh)), \ | |
6256 | + "=&r" ((USItype) (sl)) \ | |
6257 | + : "0" ((USItype) (ah)), \ | |
6258 | + "r" ((USItype) (bh)), \ | |
6259 | + "1" ((USItype) (al)), \ | |
6260 | + "r" ((USItype) (bl)) \ | |
6261 | + : "cbit") | |
6262 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6263 | + /* The cmp clears the condition bit. */ \ | |
6264 | + __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \ | |
6265 | + : "=r" ((USItype) (sh)), \ | |
6266 | + "=&r" ((USItype) (sl)) \ | |
6267 | + : "0" ((USItype) (ah)), \ | |
6268 | + "r" ((USItype) (bh)), \ | |
6269 | + "1" ((USItype) (al)), \ | |
6270 | + "r" ((USItype) (bl)) \ | |
6271 | + : "cbit") | |
6272 | +#endif /* __M32R__ */ | |
6273 | + | |
6274 | +#if defined (__mc68000__) && W_TYPE_SIZE == 32 | |
6275 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6276 | + __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \ | |
6277 | + : "=d" ((USItype) (sh)), \ | |
6278 | + "=&d" ((USItype) (sl)) \ | |
6279 | + : "%0" ((USItype) (ah)), \ | |
6280 | + "d" ((USItype) (bh)), \ | |
6281 | + "%1" ((USItype) (al)), \ | |
6282 | + "g" ((USItype) (bl))) | |
6283 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6284 | + __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \ | |
6285 | + : "=d" ((USItype) (sh)), \ | |
6286 | + "=&d" ((USItype) (sl)) \ | |
6287 | + : "0" ((USItype) (ah)), \ | |
6288 | + "d" ((USItype) (bh)), \ | |
6289 | + "1" ((USItype) (al)), \ | |
6290 | + "g" ((USItype) (bl))) | |
6291 | + | |
6292 | +/* The '020, '030, '040 and CPU32 have 32x32->64 and 64/32->32q-32r. */ | |
6293 | +#if defined (__mc68020__) || defined(mc68020) \ | |
6294 | + || defined(__mc68030__) || defined(mc68030) \ | |
6295 | + || defined(__mc68040__) || defined(mc68040) \ | |
6296 | + || defined(__mcpu32__) || defined(mcpu32) | |
6297 | +#define umul_ppmm(w1, w0, u, v) \ | |
6298 | + __asm__ ("mulu%.l %3,%1:%0" \ | |
6299 | + : "=d" ((USItype) (w0)), \ | |
6300 | + "=d" ((USItype) (w1)) \ | |
6301 | + : "%0" ((USItype) (u)), \ | |
6302 | + "dmi" ((USItype) (v))) | |
6303 | +#define UMUL_TIME 45 | |
6304 | +#define udiv_qrnnd(q, r, n1, n0, d) \ | |
6305 | + __asm__ ("divu%.l %4,%1:%0" \ | |
6306 | + : "=d" ((USItype) (q)), \ | |
6307 | + "=d" ((USItype) (r)) \ | |
6308 | + : "0" ((USItype) (n0)), \ | |
6309 | + "1" ((USItype) (n1)), \ | |
6310 | + "dmi" ((USItype) (d))) | |
6311 | +#define UDIV_TIME 90 | |
6312 | +#define sdiv_qrnnd(q, r, n1, n0, d) \ | |
6313 | + __asm__ ("divs%.l %4,%1:%0" \ | |
6314 | + : "=d" ((USItype) (q)), \ | |
6315 | + "=d" ((USItype) (r)) \ | |
6316 | + : "0" ((USItype) (n0)), \ | |
6317 | + "1" ((USItype) (n1)), \ | |
6318 | + "dmi" ((USItype) (d))) | |
6319 | + | |
6320 | +#else /* not mc68020 */ | |
6321 | +#if defined(__mcoldfire__) | |
6322 | +#define umul_ppmm(xh, xl, a, b) \ | |
6323 | + __asm__ ("| Inlined umul_ppmm\n" \ | |
6324 | + " move%.l %2,%/d0\n" \ | |
6325 | + " move%.l %3,%/d1\n" \ | |
6326 | + " move%.l %/d0,%/d2\n" \ | |
6327 | + " swap %/d0\n" \ | |
6328 | + " move%.l %/d1,%/d3\n" \ | |
6329 | + " swap %/d1\n" \ | |
6330 | + " move%.w %/d2,%/d4\n" \ | |
6331 | + " mulu %/d3,%/d4\n" \ | |
6332 | + " mulu %/d1,%/d2\n" \ | |
6333 | + " mulu %/d0,%/d3\n" \ | |
6334 | + " mulu %/d0,%/d1\n" \ | |
6335 | + " move%.l %/d4,%/d0\n" \ | |
6336 | + " clr%.w %/d0\n" \ | |
6337 | + " swap %/d0\n" \ | |
6338 | + " add%.l %/d0,%/d2\n" \ | |
6339 | + " add%.l %/d3,%/d2\n" \ | |
6340 | + " jcc 1f\n" \ | |
6341 | + " add%.l %#65536,%/d1\n" \ | |
6342 | + "1: swap %/d2\n" \ | |
6343 | + " moveq %#0,%/d0\n" \ | |
6344 | + " move%.w %/d2,%/d0\n" \ | |
6345 | + " move%.w %/d4,%/d2\n" \ | |
6346 | + " move%.l %/d2,%1\n" \ | |
6347 | + " add%.l %/d1,%/d0\n" \ | |
6348 | + " move%.l %/d0,%0" \ | |
6349 | + : "=g" ((USItype) (xh)), \ | |
6350 | + "=g" ((USItype) (xl)) \ | |
6351 | + : "g" ((USItype) (a)), \ | |
6352 | + "g" ((USItype) (b)) \ | |
6353 | + : "d0", "d1", "d2", "d3", "d4") | |
6354 | +#define UMUL_TIME 100 | |
6355 | +#define UDIV_TIME 400 | |
6356 | +#else /* not ColdFire */ | |
6357 | +/* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */ | |
6358 | +#define umul_ppmm(xh, xl, a, b) \ | |
6359 | + __asm__ ("| Inlined umul_ppmm\n" \ | |
6360 | + " move%.l %2,%/d0\n" \ | |
6361 | + " move%.l %3,%/d1\n" \ | |
6362 | + " move%.l %/d0,%/d2\n" \ | |
6363 | + " swap %/d0\n" \ | |
6364 | + " move%.l %/d1,%/d3\n" \ | |
6365 | + " swap %/d1\n" \ | |
6366 | + " move%.w %/d2,%/d4\n" \ | |
6367 | + " mulu %/d3,%/d4\n" \ | |
6368 | + " mulu %/d1,%/d2\n" \ | |
6369 | + " mulu %/d0,%/d3\n" \ | |
6370 | + " mulu %/d0,%/d1\n" \ | |
6371 | + " move%.l %/d4,%/d0\n" \ | |
6372 | + " eor%.w %/d0,%/d0\n" \ | |
6373 | + " swap %/d0\n" \ | |
6374 | + " add%.l %/d0,%/d2\n" \ | |
6375 | + " add%.l %/d3,%/d2\n" \ | |
6376 | + " jcc 1f\n" \ | |
6377 | + " add%.l %#65536,%/d1\n" \ | |
6378 | + "1: swap %/d2\n" \ | |
6379 | + " moveq %#0,%/d0\n" \ | |
6380 | + " move%.w %/d2,%/d0\n" \ | |
6381 | + " move%.w %/d4,%/d2\n" \ | |
6382 | + " move%.l %/d2,%1\n" \ | |
6383 | + " add%.l %/d1,%/d0\n" \ | |
6384 | + " move%.l %/d0,%0" \ | |
6385 | + : "=g" ((USItype) (xh)), \ | |
6386 | + "=g" ((USItype) (xl)) \ | |
6387 | + : "g" ((USItype) (a)), \ | |
6388 | + "g" ((USItype) (b)) \ | |
6389 | + : "d0", "d1", "d2", "d3", "d4") | |
6390 | +#define UMUL_TIME 100 | |
6391 | +#define UDIV_TIME 400 | |
6392 | +#endif /* not ColdFire */ | |
6393 | +#endif /* not mc68020 */ | |
6394 | + | |
6395 | +/* The '020, '030, '040 and '060 have bitfield insns. | |
6396 | + cpu32 disguises as a 68020, but lacks them. */ | |
6397 | +#if ( defined (__mc68020__) || defined(mc68020) \ | |
6398 | + || defined(__mc68030__) || defined(mc68030) \ | |
6399 | + || defined(__mc68040__) || defined(mc68040) \ | |
6400 | + || defined(__mc68060__) || defined(mc68060) ) \ | |
6401 | + && !defined(__mcpu32__) | |
6402 | +#define count_leading_zeros(count, x) \ | |
6403 | + __asm__ ("bfffo %1{%b2:%b2},%0" \ | |
6404 | + : "=d" ((USItype) (count)) \ | |
6405 | + : "od" ((USItype) (x)), "n" (0)) | |
6406 | +#endif | |
6407 | +#endif /* mc68000 */ | |
6408 | + | |
6409 | +#if defined (__m88000__) && W_TYPE_SIZE == 32 | |
6410 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6411 | + __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \ | |
6412 | + : "=r" ((USItype) (sh)), \ | |
6413 | + "=&r" ((USItype) (sl)) \ | |
6414 | + : "%rJ" ((USItype) (ah)), \ | |
6415 | + "rJ" ((USItype) (bh)), \ | |
6416 | + "%rJ" ((USItype) (al)), \ | |
6417 | + "rJ" ((USItype) (bl))) | |
6418 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6419 | + __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \ | |
6420 | + : "=r" ((USItype) (sh)), \ | |
6421 | + "=&r" ((USItype) (sl)) \ | |
6422 | + : "rJ" ((USItype) (ah)), \ | |
6423 | + "rJ" ((USItype) (bh)), \ | |
6424 | + "rJ" ((USItype) (al)), \ | |
6425 | + "rJ" ((USItype) (bl))) | |
6426 | +#define count_leading_zeros(count, x) \ | |
6427 | + do { \ | |
6428 | + USItype __cbtmp; \ | |
6429 | + __asm__ ("ff1 %0,%1" \ | |
6430 | + : "=r" (__cbtmp) \ | |
6431 | + : "r" ((USItype) (x))); \ | |
6432 | + (count) = __cbtmp ^ 31; \ | |
6433 | + } while (0) | |
6434 | +#define COUNT_LEADING_ZEROS_0 63 /* sic */ | |
6435 | +#if defined (__mc88110__) | |
6436 | +#define umul_ppmm(wh, wl, u, v) \ | |
6437 | + do { \ | |
6438 | + union {UDItype __ll; \ | |
6439 | + struct {USItype __h, __l;} __i; \ | |
6440 | + } __xx; \ | |
6441 | + __asm__ ("mulu.d %0,%1,%2" \ | |
6442 | + : "=r" (__xx.__ll) \ | |
6443 | + : "r" ((USItype) (u)), \ | |
6444 | + "r" ((USItype) (v))); \ | |
6445 | + (wh) = __xx.__i.__h; \ | |
6446 | + (wl) = __xx.__i.__l; \ | |
6447 | + } while (0) | |
6448 | +#define udiv_qrnnd(q, r, n1, n0, d) \ | |
6449 | + ({union {UDItype __ll; \ | |
6450 | + struct {USItype __h, __l;} __i; \ | |
6451 | + } __xx; \ | |
6452 | + USItype __q; \ | |
6453 | + __xx.__i.__h = (n1); __xx.__i.__l = (n0); \ | |
6454 | + __asm__ ("divu.d %0,%1,%2" \ | |
6455 | + : "=r" (__q) \ | |
6456 | + : "r" (__xx.__ll), \ | |
6457 | + "r" ((USItype) (d))); \ | |
6458 | + (r) = (n0) - __q * (d); (q) = __q; }) | |
6459 | +#define UMUL_TIME 5 | |
6460 | +#define UDIV_TIME 25 | |
6461 | +#else | |
6462 | +#define UMUL_TIME 17 | |
6463 | +#define UDIV_TIME 150 | |
6464 | +#endif /* __mc88110__ */ | |
6465 | +#endif /* __m88000__ */ | |
6466 | + | |
6467 | +#if defined (__mips__) && W_TYPE_SIZE == 32 | |
6468 | +#define umul_ppmm(w1, w0, u, v) \ | |
6469 | + __asm__ ("multu %2,%3" \ | |
6470 | + : "=l" ((USItype) (w0)), \ | |
6471 | + "=h" ((USItype) (w1)) \ | |
6472 | + : "d" ((USItype) (u)), \ | |
6473 | + "d" ((USItype) (v))) | |
6474 | +#define UMUL_TIME 10 | |
6475 | +#define UDIV_TIME 100 | |
6476 | +#endif /* __mips__ */ | |
6477 | + | |
6478 | +#if defined (__ns32000__) && W_TYPE_SIZE == 32 | |
6479 | +#define umul_ppmm(w1, w0, u, v) \ | |
6480 | + ({union {UDItype __ll; \ | |
6481 | + struct {USItype __l, __h;} __i; \ | |
6482 | + } __xx; \ | |
6483 | + __asm__ ("meid %2,%0" \ | |
6484 | + : "=g" (__xx.__ll) \ | |
6485 | + : "%0" ((USItype) (u)), \ | |
6486 | + "g" ((USItype) (v))); \ | |
6487 | + (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;}) | |
6488 | +#define __umulsidi3(u, v) \ | |
6489 | + ({UDItype __w; \ | |
6490 | + __asm__ ("meid %2,%0" \ | |
6491 | + : "=g" (__w) \ | |
6492 | + : "%0" ((USItype) (u)), \ | |
6493 | + "g" ((USItype) (v))); \ | |
6494 | + __w; }) | |
6495 | +#define udiv_qrnnd(q, r, n1, n0, d) \ | |
6496 | + ({union {UDItype __ll; \ | |
6497 | + struct {USItype __l, __h;} __i; \ | |
6498 | + } __xx; \ | |
6499 | + __xx.__i.__h = (n1); __xx.__i.__l = (n0); \ | |
6500 | + __asm__ ("deid %2,%0" \ | |
6501 | + : "=g" (__xx.__ll) \ | |
6502 | + : "0" (__xx.__ll), \ | |
6503 | + "g" ((USItype) (d))); \ | |
6504 | + (r) = __xx.__i.__l; (q) = __xx.__i.__h; }) | |
6505 | +#define count_trailing_zeros(count,x) \ | |
6506 | + do { \ | |
6507 | + __asm__ ("ffsd %2,%0" \ | |
6508 | + : "=r" ((USItype) (count)) \ | |
6509 | + : "0" ((USItype) 0), \ | |
6510 | + "r" ((USItype) (x))); \ | |
6511 | + } while (0) | |
6512 | +#endif /* __ns32000__ */ | |
6513 | + | |
6514 | +/* FIXME: We should test _IBMR2 here when we add assembly support for the | |
6515 | + system vendor compilers. | |
6516 | + FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good | |
6517 | + enough, since that hits ARM and m68k too. */ | |
6518 | +#if (defined (_ARCH_PPC) /* AIX */ \ | |
6519 | + || defined (_ARCH_PWR) /* AIX */ \ | |
6520 | + || defined (_ARCH_COM) /* AIX */ \ | |
6521 | + || defined (__powerpc__) /* gcc */ \ | |
6522 | + || defined (__POWERPC__) /* BEOS */ \ | |
6523 | + || defined (__ppc__) /* Darwin */ \ | |
6524 | + || defined (PPC) /* GNU/Linux, SysV */ \ | |
6525 | + ) && W_TYPE_SIZE == 32 | |
6526 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6527 | + do { \ | |
6528 | + if (__builtin_constant_p (bh) && (bh) == 0) \ | |
6529 | + __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \ | |
6530 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ | |
6531 | + else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \ | |
6532 | + __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \ | |
6533 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ | |
6534 | + else \ | |
6535 | + __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \ | |
6536 | + : "=r" (sh), "=&r" (sl) \ | |
6537 | + : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \ | |
6538 | + } while (0) | |
6539 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6540 | + do { \ | |
6541 | + if (__builtin_constant_p (ah) && (ah) == 0) \ | |
6542 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \ | |
6543 | + : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ | |
6544 | + else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \ | |
6545 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \ | |
6546 | + : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ | |
6547 | + else if (__builtin_constant_p (bh) && (bh) == 0) \ | |
6548 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \ | |
6549 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ | |
6550 | + else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \ | |
6551 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \ | |
6552 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ | |
6553 | + else \ | |
6554 | + __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \ | |
6555 | + : "=r" (sh), "=&r" (sl) \ | |
6556 | + : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \ | |
6557 | + } while (0) | |
6558 | +#define count_leading_zeros(count, x) \ | |
6559 | + __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x)) | |
6560 | +#define COUNT_LEADING_ZEROS_0 32 | |
6561 | +#if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \ | |
6562 | + || defined (__ppc__) || defined (PPC) | |
6563 | +#define umul_ppmm(ph, pl, m0, m1) \ | |
6564 | + do { \ | |
6565 | + USItype __m0 = (m0), __m1 = (m1); \ | |
6566 | + __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ | |
6567 | + (pl) = __m0 * __m1; \ | |
6568 | + } while (0) | |
6569 | +#define UMUL_TIME 15 | |
6570 | +#define smul_ppmm(ph, pl, m0, m1) \ | |
6571 | + do { \ | |
6572 | + SItype __m0 = (m0), __m1 = (m1); \ | |
6573 | + __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ | |
6574 | + (pl) = __m0 * __m1; \ | |
6575 | + } while (0) | |
6576 | +#define SMUL_TIME 14 | |
6577 | +#define UDIV_TIME 120 | |
6578 | +#elif defined (_ARCH_PWR) | |
6579 | +#define UMUL_TIME 8 | |
6580 | +#define smul_ppmm(xh, xl, m0, m1) \ | |
6581 | + __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1)) | |
6582 | +#define SMUL_TIME 4 | |
6583 | +#define sdiv_qrnnd(q, r, nh, nl, d) \ | |
6584 | + __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d)) | |
6585 | +#define UDIV_TIME 100 | |
6586 | +#endif | |
6587 | +#endif /* 32-bit POWER architecture variants. */ | |
6588 | + | |
6589 | +/* We should test _IBMR2 here when we add assembly support for the system | |
6590 | + vendor compilers. */ | |
6591 | +#if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64 | |
6592 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6593 | + do { \ | |
6594 | + if (__builtin_constant_p (bh) && (bh) == 0) \ | |
6595 | + __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \ | |
6596 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ | |
6597 | + else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \ | |
6598 | + __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \ | |
6599 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ | |
6600 | + else \ | |
6601 | + __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \ | |
6602 | + : "=r" (sh), "=&r" (sl) \ | |
6603 | + : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \ | |
6604 | + } while (0) | |
6605 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6606 | + do { \ | |
6607 | + if (__builtin_constant_p (ah) && (ah) == 0) \ | |
6608 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \ | |
6609 | + : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ | |
6610 | + else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \ | |
6611 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \ | |
6612 | + : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ | |
6613 | + else if (__builtin_constant_p (bh) && (bh) == 0) \ | |
6614 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \ | |
6615 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ | |
6616 | + else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \ | |
6617 | + __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \ | |
6618 | + : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ | |
6619 | + else \ | |
6620 | + __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \ | |
6621 | + : "=r" (sh), "=&r" (sl) \ | |
6622 | + : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \ | |
6623 | + } while (0) | |
6624 | +#define count_leading_zeros(count, x) \ | |
6625 | + __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x)) | |
6626 | +#define COUNT_LEADING_ZEROS_0 64 | |
6627 | +#define umul_ppmm(ph, pl, m0, m1) \ | |
6628 | + do { \ | |
6629 | + UDItype __m0 = (m0), __m1 = (m1); \ | |
6630 | + __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ | |
6631 | + (pl) = __m0 * __m1; \ | |
6632 | + } while (0) | |
6633 | +#define UMUL_TIME 15 | |
6634 | +#define smul_ppmm(ph, pl, m0, m1) \ | |
6635 | + do { \ | |
6636 | + DItype __m0 = (m0), __m1 = (m1); \ | |
6637 | + __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ | |
6638 | + (pl) = __m0 * __m1; \ | |
6639 | + } while (0) | |
6640 | +#define SMUL_TIME 14 /* ??? */ | |
6641 | +#define UDIV_TIME 120 /* ??? */ | |
6642 | +#endif /* 64-bit PowerPC. */ | |
6643 | + | |
6644 | +#if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32 | |
6645 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6646 | + __asm__ ("a %1,%5\n\tae %0,%3" \ | |
6647 | + : "=r" ((USItype) (sh)), \ | |
6648 | + "=&r" ((USItype) (sl)) \ | |
6649 | + : "%0" ((USItype) (ah)), \ | |
6650 | + "r" ((USItype) (bh)), \ | |
6651 | + "%1" ((USItype) (al)), \ | |
6652 | + "r" ((USItype) (bl))) | |
6653 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6654 | + __asm__ ("s %1,%5\n\tse %0,%3" \ | |
6655 | + : "=r" ((USItype) (sh)), \ | |
6656 | + "=&r" ((USItype) (sl)) \ | |
6657 | + : "0" ((USItype) (ah)), \ | |
6658 | + "r" ((USItype) (bh)), \ | |
6659 | + "1" ((USItype) (al)), \ | |
6660 | + "r" ((USItype) (bl))) | |
6661 | +#define umul_ppmm(ph, pl, m0, m1) \ | |
6662 | + do { \ | |
6663 | + USItype __m0 = (m0), __m1 = (m1); \ | |
6664 | + __asm__ ( \ | |
6665 | + "s r2,r2\n" \ | |
6666 | +" mts r10,%2\n" \ | |
6667 | +" m r2,%3\n" \ | |
6668 | +" m r2,%3\n" \ | |
6669 | +" m r2,%3\n" \ | |
6670 | +" m r2,%3\n" \ | |
6671 | +" m r2,%3\n" \ | |
6672 | +" m r2,%3\n" \ | |
6673 | +" m r2,%3\n" \ | |
6674 | +" m r2,%3\n" \ | |
6675 | +" m r2,%3\n" \ | |
6676 | +" m r2,%3\n" \ | |
6677 | +" m r2,%3\n" \ | |
6678 | +" m r2,%3\n" \ | |
6679 | +" m r2,%3\n" \ | |
6680 | +" m r2,%3\n" \ | |
6681 | +" m r2,%3\n" \ | |
6682 | +" m r2,%3\n" \ | |
6683 | +" cas %0,r2,r0\n" \ | |
6684 | +" mfs r10,%1" \ | |
6685 | + : "=r" ((USItype) (ph)), \ | |
6686 | + "=r" ((USItype) (pl)) \ | |
6687 | + : "%r" (__m0), \ | |
6688 | + "r" (__m1) \ | |
6689 | + : "r2"); \ | |
6690 | + (ph) += ((((SItype) __m0 >> 31) & __m1) \ | |
6691 | + + (((SItype) __m1 >> 31) & __m0)); \ | |
6692 | + } while (0) | |
6693 | +#define UMUL_TIME 20 | |
6694 | +#define UDIV_TIME 200 | |
6695 | +#define count_leading_zeros(count, x) \ | |
6696 | + do { \ | |
6697 | + if ((x) >= 0x10000) \ | |
6698 | + __asm__ ("clz %0,%1" \ | |
6699 | + : "=r" ((USItype) (count)) \ | |
6700 | + : "r" ((USItype) (x) >> 16)); \ | |
6701 | + else \ | |
6702 | + { \ | |
6703 | + __asm__ ("clz %0,%1" \ | |
6704 | + : "=r" ((USItype) (count)) \ | |
6705 | + : "r" ((USItype) (x))); \ | |
6706 | + (count) += 16; \ | |
6707 | + } \ | |
6708 | + } while (0) | |
6709 | +#endif | |
6710 | + | |
6711 | +#if defined (__sh2__) && W_TYPE_SIZE == 32 | |
6712 | +#define umul_ppmm(w1, w0, u, v) \ | |
6713 | + __asm__ ( \ | |
6714 | + "dmulu.l %2,%3\n\tsts macl,%1\n\tsts mach,%0" \ | |
6715 | + : "=r" ((USItype)(w1)), \ | |
6716 | + "=r" ((USItype)(w0)) \ | |
6717 | + : "r" ((USItype)(u)), \ | |
6718 | + "r" ((USItype)(v)) \ | |
6719 | + : "macl", "mach") | |
6720 | +#define UMUL_TIME 5 | |
6721 | +#endif | |
6722 | + | |
6723 | +#if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32 | |
6724 | +#define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v) | |
6725 | +#define count_leading_zeros(count, x) \ | |
6726 | + do \ | |
6727 | + { \ | |
6728 | + UDItype x_ = (USItype)(x); \ | |
6729 | + SItype c_; \ | |
6730 | + \ | |
6731 | + __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \ | |
6732 | + (count) = c_ - 31; \ | |
6733 | + } \ | |
6734 | + while (0) | |
6735 | +#define COUNT_LEADING_ZEROS_0 32 | |
6736 | +#endif | |
6737 | + | |
6738 | +#if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \ | |
6739 | + && W_TYPE_SIZE == 32 | |
6740 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6741 | + __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \ | |
6742 | + : "=r" ((USItype) (sh)), \ | |
6743 | + "=&r" ((USItype) (sl)) \ | |
6744 | + : "%rJ" ((USItype) (ah)), \ | |
6745 | + "rI" ((USItype) (bh)), \ | |
6746 | + "%rJ" ((USItype) (al)), \ | |
6747 | + "rI" ((USItype) (bl)) \ | |
6748 | + __CLOBBER_CC) | |
6749 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6750 | + __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \ | |
6751 | + : "=r" ((USItype) (sh)), \ | |
6752 | + "=&r" ((USItype) (sl)) \ | |
6753 | + : "rJ" ((USItype) (ah)), \ | |
6754 | + "rI" ((USItype) (bh)), \ | |
6755 | + "rJ" ((USItype) (al)), \ | |
6756 | + "rI" ((USItype) (bl)) \ | |
6757 | + __CLOBBER_CC) | |
6758 | +#if defined (__sparc_v8__) | |
6759 | +#define umul_ppmm(w1, w0, u, v) \ | |
6760 | + __asm__ ("umul %2,%3,%1;rd %%y,%0" \ | |
6761 | + : "=r" ((USItype) (w1)), \ | |
6762 | + "=r" ((USItype) (w0)) \ | |
6763 | + : "r" ((USItype) (u)), \ | |
6764 | + "r" ((USItype) (v))) | |
6765 | +#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \ | |
6766 | + __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\ | |
6767 | + : "=&r" ((USItype) (__q)), \ | |
6768 | + "=&r" ((USItype) (__r)) \ | |
6769 | + : "r" ((USItype) (__n1)), \ | |
6770 | + "r" ((USItype) (__n0)), \ | |
6771 | + "r" ((USItype) (__d))) | |
6772 | +#else | |
6773 | +#if defined (__sparclite__) | |
6774 | +/* This has hardware multiply but not divide. It also has two additional | |
6775 | + instructions scan (ffs from high bit) and divscc. */ | |
6776 | +#define umul_ppmm(w1, w0, u, v) \ | |
6777 | + __asm__ ("umul %2,%3,%1;rd %%y,%0" \ | |
6778 | + : "=r" ((USItype) (w1)), \ | |
6779 | + "=r" ((USItype) (w0)) \ | |
6780 | + : "r" ((USItype) (u)), \ | |
6781 | + "r" ((USItype) (v))) | |
6782 | +#define udiv_qrnnd(q, r, n1, n0, d) \ | |
6783 | + __asm__ ("! Inlined udiv_qrnnd\n" \ | |
6784 | +" wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \ | |
6785 | +" tst %%g0\n" \ | |
6786 | +" divscc %3,%4,%%g1\n" \ | |
6787 | +" divscc %%g1,%4,%%g1\n" \ | |
6788 | +" divscc %%g1,%4,%%g1\n" \ | |
6789 | +" divscc %%g1,%4,%%g1\n" \ | |
6790 | +" divscc %%g1,%4,%%g1\n" \ | |
6791 | +" divscc %%g1,%4,%%g1\n" \ | |
6792 | +" divscc %%g1,%4,%%g1\n" \ | |
6793 | +" divscc %%g1,%4,%%g1\n" \ | |
6794 | +" divscc %%g1,%4,%%g1\n" \ | |
6795 | +" divscc %%g1,%4,%%g1\n" \ | |
6796 | +" divscc %%g1,%4,%%g1\n" \ | |
6797 | +" divscc %%g1,%4,%%g1\n" \ | |
6798 | +" divscc %%g1,%4,%%g1\n" \ | |
6799 | +" divscc %%g1,%4,%%g1\n" \ | |
6800 | +" divscc %%g1,%4,%%g1\n" \ | |
6801 | +" divscc %%g1,%4,%%g1\n" \ | |
6802 | +" divscc %%g1,%4,%%g1\n" \ | |
6803 | +" divscc %%g1,%4,%%g1\n" \ | |
6804 | +" divscc %%g1,%4,%%g1\n" \ | |
6805 | +" divscc %%g1,%4,%%g1\n" \ | |
6806 | +" divscc %%g1,%4,%%g1\n" \ | |
6807 | +" divscc %%g1,%4,%%g1\n" \ | |
6808 | +" divscc %%g1,%4,%%g1\n" \ | |
6809 | +" divscc %%g1,%4,%%g1\n" \ | |
6810 | +" divscc %%g1,%4,%%g1\n" \ | |
6811 | +" divscc %%g1,%4,%%g1\n" \ | |
6812 | +" divscc %%g1,%4,%%g1\n" \ | |
6813 | +" divscc %%g1,%4,%%g1\n" \ | |
6814 | +" divscc %%g1,%4,%%g1\n" \ | |
6815 | +" divscc %%g1,%4,%%g1\n" \ | |
6816 | +" divscc %%g1,%4,%%g1\n" \ | |
6817 | +" divscc %%g1,%4,%0\n" \ | |
6818 | +" rd %%y,%1\n" \ | |
6819 | +" bl,a 1f\n" \ | |
6820 | +" add %1,%4,%1\n" \ | |
6821 | +"1: ! End of inline udiv_qrnnd" \ | |
6822 | + : "=r" ((USItype) (q)), \ | |
6823 | + "=r" ((USItype) (r)) \ | |
6824 | + : "r" ((USItype) (n1)), \ | |
6825 | + "r" ((USItype) (n0)), \ | |
6826 | + "rI" ((USItype) (d)) \ | |
6827 | + : "g1" __AND_CLOBBER_CC) | |
6828 | +#define UDIV_TIME 37 | |
6829 | +#define count_leading_zeros(count, x) \ | |
6830 | + do { \ | |
6831 | + __asm__ ("scan %1,1,%0" \ | |
6832 | + : "=r" ((USItype) (count)) \ | |
6833 | + : "r" ((USItype) (x))); \ | |
6834 | + } while (0) | |
6835 | +/* Early sparclites return 63 for an argument of 0, but they warn that future | |
6836 | + implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0 | |
6837 | + undefined. */ | |
6838 | +#else | |
6839 | +/* SPARC without integer multiplication and divide instructions. | |
6840 | + (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */ | |
6841 | +#define umul_ppmm(w1, w0, u, v) \ | |
6842 | + __asm__ ("! Inlined umul_ppmm\n" \ | |
6843 | +" wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\ | |
6844 | +" sra %3,31,%%o5 ! Don't move this insn\n" \ | |
6845 | +" and %2,%%o5,%%o5 ! Don't move this insn\n" \ | |
6846 | +" andcc %%g0,0,%%g1 ! Don't move this insn\n" \ | |
6847 | +" mulscc %%g1,%3,%%g1\n" \ | |
6848 | +" mulscc %%g1,%3,%%g1\n" \ | |
6849 | +" mulscc %%g1,%3,%%g1\n" \ | |
6850 | +" mulscc %%g1,%3,%%g1\n" \ | |
6851 | +" mulscc %%g1,%3,%%g1\n" \ | |
6852 | +" mulscc %%g1,%3,%%g1\n" \ | |
6853 | +" mulscc %%g1,%3,%%g1\n" \ | |
6854 | +" mulscc %%g1,%3,%%g1\n" \ | |
6855 | +" mulscc %%g1,%3,%%g1\n" \ | |
6856 | +" mulscc %%g1,%3,%%g1\n" \ | |
6857 | +" mulscc %%g1,%3,%%g1\n" \ | |
6858 | +" mulscc %%g1,%3,%%g1\n" \ | |
6859 | +" mulscc %%g1,%3,%%g1\n" \ | |
6860 | +" mulscc %%g1,%3,%%g1\n" \ | |
6861 | +" mulscc %%g1,%3,%%g1\n" \ | |
6862 | +" mulscc %%g1,%3,%%g1\n" \ | |
6863 | +" mulscc %%g1,%3,%%g1\n" \ | |
6864 | +" mulscc %%g1,%3,%%g1\n" \ | |
6865 | +" mulscc %%g1,%3,%%g1\n" \ | |
6866 | +" mulscc %%g1,%3,%%g1\n" \ | |
6867 | +" mulscc %%g1,%3,%%g1\n" \ | |
6868 | +" mulscc %%g1,%3,%%g1\n" \ | |
6869 | +" mulscc %%g1,%3,%%g1\n" \ | |
6870 | +" mulscc %%g1,%3,%%g1\n" \ | |
6871 | +" mulscc %%g1,%3,%%g1\n" \ | |
6872 | +" mulscc %%g1,%3,%%g1\n" \ | |
6873 | +" mulscc %%g1,%3,%%g1\n" \ | |
6874 | +" mulscc %%g1,%3,%%g1\n" \ | |
6875 | +" mulscc %%g1,%3,%%g1\n" \ | |
6876 | +" mulscc %%g1,%3,%%g1\n" \ | |
6877 | +" mulscc %%g1,%3,%%g1\n" \ | |
6878 | +" mulscc %%g1,%3,%%g1\n" \ | |
6879 | +" mulscc %%g1,0,%%g1\n" \ | |
6880 | +" add %%g1,%%o5,%0\n" \ | |
6881 | +" rd %%y,%1" \ | |
6882 | + : "=r" ((USItype) (w1)), \ | |
6883 | + "=r" ((USItype) (w0)) \ | |
6884 | + : "%rI" ((USItype) (u)), \ | |
6885 | + "r" ((USItype) (v)) \ | |
6886 | + : "g1", "o5" __AND_CLOBBER_CC) | |
6887 | +#define UMUL_TIME 39 /* 39 instructions */ | |
6888 | +/* It's quite necessary to add this much assembler for the sparc. | |
6889 | + The default udiv_qrnnd (in C) is more than 10 times slower! */ | |
6890 | +#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \ | |
6891 | + __asm__ ("! Inlined udiv_qrnnd\n" \ | |
6892 | +" mov 32,%%g1\n" \ | |
6893 | +" subcc %1,%2,%%g0\n" \ | |
6894 | +"1: bcs 5f\n" \ | |
6895 | +" addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \ | |
6896 | +" sub %1,%2,%1 ! this kills msb of n\n" \ | |
6897 | +" addx %1,%1,%1 ! so this can't give carry\n" \ | |
6898 | +" subcc %%g1,1,%%g1\n" \ | |
6899 | +"2: bne 1b\n" \ | |
6900 | +" subcc %1,%2,%%g0\n" \ | |
6901 | +" bcs 3f\n" \ | |
6902 | +" addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \ | |
6903 | +" b 3f\n" \ | |
6904 | +" sub %1,%2,%1 ! this kills msb of n\n" \ | |
6905 | +"4: sub %1,%2,%1\n" \ | |
6906 | +"5: addxcc %1,%1,%1\n" \ | |
6907 | +" bcc 2b\n" \ | |
6908 | +" subcc %%g1,1,%%g1\n" \ | |
6909 | +"! Got carry from n. Subtract next step to cancel this carry.\n" \ | |
6910 | +" bne 4b\n" \ | |
6911 | +" addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \ | |
6912 | +" sub %1,%2,%1\n" \ | |
6913 | +"3: xnor %0,0,%0\n" \ | |
6914 | +" ! End of inline udiv_qrnnd" \ | |
6915 | + : "=&r" ((USItype) (__q)), \ | |
6916 | + "=&r" ((USItype) (__r)) \ | |
6917 | + : "r" ((USItype) (__d)), \ | |
6918 | + "1" ((USItype) (__n1)), \ | |
6919 | + "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC) | |
6920 | +#define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */ | |
6921 | +#endif /* __sparclite__ */ | |
6922 | +#endif /* __sparc_v8__ */ | |
6923 | +#endif /* sparc32 */ | |
6924 | + | |
6925 | +#if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \ | |
6926 | + && W_TYPE_SIZE == 64 | |
6927 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6928 | + __asm__ ("addcc %r4,%5,%1\n\t" \ | |
6929 | + "add %r2,%3,%0\n\t" \ | |
6930 | + "bcs,a,pn %%xcc, 1f\n\t" \ | |
6931 | + "add %0, 1, %0\n" \ | |
6932 | + "1:" \ | |
6933 | + : "=r" ((UDItype)(sh)), \ | |
6934 | + "=&r" ((UDItype)(sl)) \ | |
6935 | + : "%rJ" ((UDItype)(ah)), \ | |
6936 | + "rI" ((UDItype)(bh)), \ | |
6937 | + "%rJ" ((UDItype)(al)), \ | |
6938 | + "rI" ((UDItype)(bl)) \ | |
6939 | + __CLOBBER_CC) | |
6940 | + | |
6941 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
6942 | + __asm__ ("subcc %r4,%5,%1\n\t" \ | |
6943 | + "sub %r2,%3,%0\n\t" \ | |
6944 | + "bcs,a,pn %%xcc, 1f\n\t" \ | |
6945 | + "sub %0, 1, %0\n\t" \ | |
6946 | + "1:" \ | |
6947 | + : "=r" ((UDItype)(sh)), \ | |
6948 | + "=&r" ((UDItype)(sl)) \ | |
6949 | + : "rJ" ((UDItype)(ah)), \ | |
6950 | + "rI" ((UDItype)(bh)), \ | |
6951 | + "rJ" ((UDItype)(al)), \ | |
6952 | + "rI" ((UDItype)(bl)) \ | |
6953 | + __CLOBBER_CC) | |
6954 | + | |
6955 | +#define umul_ppmm(wh, wl, u, v) \ | |
6956 | + do { \ | |
6957 | + UDItype tmp1, tmp2, tmp3, tmp4; \ | |
6958 | + __asm__ __volatile__ ( \ | |
6959 | + "srl %7,0,%3\n\t" \ | |
6960 | + "mulx %3,%6,%1\n\t" \ | |
6961 | + "srlx %6,32,%2\n\t" \ | |
6962 | + "mulx %2,%3,%4\n\t" \ | |
6963 | + "sllx %4,32,%5\n\t" \ | |
6964 | + "srl %6,0,%3\n\t" \ | |
6965 | + "sub %1,%5,%5\n\t" \ | |
6966 | + "srlx %5,32,%5\n\t" \ | |
6967 | + "addcc %4,%5,%4\n\t" \ | |
6968 | + "srlx %7,32,%5\n\t" \ | |
6969 | + "mulx %3,%5,%3\n\t" \ | |
6970 | + "mulx %2,%5,%5\n\t" \ | |
6971 | + "sethi %%hi(0x80000000),%2\n\t" \ | |
6972 | + "addcc %4,%3,%4\n\t" \ | |
6973 | + "srlx %4,32,%4\n\t" \ | |
6974 | + "add %2,%2,%2\n\t" \ | |
6975 | + "movcc %%xcc,%%g0,%2\n\t" \ | |
6976 | + "addcc %5,%4,%5\n\t" \ | |
6977 | + "sllx %3,32,%3\n\t" \ | |
6978 | + "add %1,%3,%1\n\t" \ | |
6979 | + "add %5,%2,%0" \ | |
6980 | + : "=r" ((UDItype)(wh)), \ | |
6981 | + "=&r" ((UDItype)(wl)), \ | |
6982 | + "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \ | |
6983 | + : "r" ((UDItype)(u)), \ | |
6984 | + "r" ((UDItype)(v)) \ | |
6985 | + __CLOBBER_CC); \ | |
6986 | + } while (0) | |
6987 | +#define UMUL_TIME 96 | |
6988 | +#define UDIV_TIME 230 | |
6989 | +#endif /* sparc64 */ | |
6990 | + | |
6991 | +#if defined (__vax__) && W_TYPE_SIZE == 32 | |
6992 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
6993 | + __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \ | |
6994 | + : "=g" ((USItype) (sh)), \ | |
6995 | + "=&g" ((USItype) (sl)) \ | |
6996 | + : "%0" ((USItype) (ah)), \ | |
6997 | + "g" ((USItype) (bh)), \ | |
6998 | + "%1" ((USItype) (al)), \ | |
6999 | + "g" ((USItype) (bl))) | |
7000 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
7001 | + __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \ | |
7002 | + : "=g" ((USItype) (sh)), \ | |
7003 | + "=&g" ((USItype) (sl)) \ | |
7004 | + : "0" ((USItype) (ah)), \ | |
7005 | + "g" ((USItype) (bh)), \ | |
7006 | + "1" ((USItype) (al)), \ | |
7007 | + "g" ((USItype) (bl))) | |
7008 | +#define umul_ppmm(xh, xl, m0, m1) \ | |
7009 | + do { \ | |
7010 | + union { \ | |
7011 | + UDItype __ll; \ | |
7012 | + struct {USItype __l, __h;} __i; \ | |
7013 | + } __xx; \ | |
7014 | + USItype __m0 = (m0), __m1 = (m1); \ | |
7015 | + __asm__ ("emul %1,%2,$0,%0" \ | |
7016 | + : "=r" (__xx.__ll) \ | |
7017 | + : "g" (__m0), \ | |
7018 | + "g" (__m1)); \ | |
7019 | + (xh) = __xx.__i.__h; \ | |
7020 | + (xl) = __xx.__i.__l; \ | |
7021 | + (xh) += ((((SItype) __m0 >> 31) & __m1) \ | |
7022 | + + (((SItype) __m1 >> 31) & __m0)); \ | |
7023 | + } while (0) | |
7024 | +#define sdiv_qrnnd(q, r, n1, n0, d) \ | |
7025 | + do { \ | |
7026 | + union {DItype __ll; \ | |
7027 | + struct {SItype __l, __h;} __i; \ | |
7028 | + } __xx; \ | |
7029 | + __xx.__i.__h = n1; __xx.__i.__l = n0; \ | |
7030 | + __asm__ ("ediv %3,%2,%0,%1" \ | |
7031 | + : "=g" (q), "=g" (r) \ | |
7032 | + : "g" (__xx.__ll), "g" (d)); \ | |
7033 | + } while (0) | |
7034 | +#endif /* __vax__ */ | |
7035 | + | |
7036 | +#if defined (__z8000__) && W_TYPE_SIZE == 16 | |
7037 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
7038 | + __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \ | |
7039 | + : "=r" ((unsigned int)(sh)), \ | |
7040 | + "=&r" ((unsigned int)(sl)) \ | |
7041 | + : "%0" ((unsigned int)(ah)), \ | |
7042 | + "r" ((unsigned int)(bh)), \ | |
7043 | + "%1" ((unsigned int)(al)), \ | |
7044 | + "rQR" ((unsigned int)(bl))) | |
7045 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
7046 | + __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \ | |
7047 | + : "=r" ((unsigned int)(sh)), \ | |
7048 | + "=&r" ((unsigned int)(sl)) \ | |
7049 | + : "0" ((unsigned int)(ah)), \ | |
7050 | + "r" ((unsigned int)(bh)), \ | |
7051 | + "1" ((unsigned int)(al)), \ | |
7052 | + "rQR" ((unsigned int)(bl))) | |
7053 | +#define umul_ppmm(xh, xl, m0, m1) \ | |
7054 | + do { \ | |
7055 | + union {long int __ll; \ | |
7056 | + struct {unsigned int __h, __l;} __i; \ | |
7057 | + } __xx; \ | |
7058 | + unsigned int __m0 = (m0), __m1 = (m1); \ | |
7059 | + __asm__ ("mult %S0,%H3" \ | |
7060 | + : "=r" (__xx.__i.__h), \ | |
7061 | + "=r" (__xx.__i.__l) \ | |
7062 | + : "%1" (__m0), \ | |
7063 | + "rQR" (__m1)); \ | |
7064 | + (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \ | |
7065 | + (xh) += ((((signed int) __m0 >> 15) & __m1) \ | |
7066 | + + (((signed int) __m1 >> 15) & __m0)); \ | |
7067 | + } while (0) | |
7068 | +#endif /* __z8000__ */ | |
7069 | + | |
7070 | +#endif /* __GNUC__ */ | |
7071 | + | |
7072 | +/* If this machine has no inline assembler, use C macros. */ | |
7073 | + | |
7074 | +#if !defined (add_ssaaaa) | |
7075 | +#define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | |
7076 | + do { \ | |
7077 | + UWtype __x; \ | |
7078 | + __x = (al) + (bl); \ | |
7079 | + (sh) = (ah) + (bh) + (__x < (al)); \ | |
7080 | + (sl) = __x; \ | |
7081 | + } while (0) | |
7082 | +#endif | |
7083 | + | |
7084 | +#if !defined (sub_ddmmss) | |
7085 | +#define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | |
7086 | + do { \ | |
7087 | + UWtype __x; \ | |
7088 | + __x = (al) - (bl); \ | |
7089 | + (sh) = (ah) - (bh) - (__x > (al)); \ | |
7090 | + (sl) = __x; \ | |
7091 | + } while (0) | |
7092 | +#endif | |
7093 | + | |
7094 | +/* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of | |
7095 | + smul_ppmm. */ | |
7096 | +#if !defined (umul_ppmm) && defined (smul_ppmm) | |
7097 | +#define umul_ppmm(w1, w0, u, v) \ | |
7098 | + do { \ | |
7099 | + UWtype __w1; \ | |
7100 | + UWtype __xm0 = (u), __xm1 = (v); \ | |
7101 | + smul_ppmm (__w1, w0, __xm0, __xm1); \ | |
7102 | + (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \ | |
7103 | + + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \ | |
7104 | + } while (0) | |
7105 | +#endif | |
7106 | + | |
7107 | +/* If we still don't have umul_ppmm, define it using plain C. */ | |
7108 | +#if !defined (umul_ppmm) | |
7109 | +#define umul_ppmm(w1, w0, u, v) \ | |
7110 | + do { \ | |
7111 | + UWtype __x0, __x1, __x2, __x3; \ | |
7112 | + UHWtype __ul, __vl, __uh, __vh; \ | |
7113 | + \ | |
7114 | + __ul = __ll_lowpart (u); \ | |
7115 | + __uh = __ll_highpart (u); \ | |
7116 | + __vl = __ll_lowpart (v); \ | |
7117 | + __vh = __ll_highpart (v); \ | |
7118 | + \ | |
7119 | + __x0 = (UWtype) __ul * __vl; \ | |
7120 | + __x1 = (UWtype) __ul * __vh; \ | |
7121 | + __x2 = (UWtype) __uh * __vl; \ | |
7122 | + __x3 = (UWtype) __uh * __vh; \ | |
7123 | + \ | |
7124 | + __x1 += __ll_highpart (__x0);/* this can't give carry */ \ | |
7125 | + __x1 += __x2; /* but this indeed can */ \ | |
7126 | + if (__x1 < __x2) /* did we get it? */ \ | |
7127 | + __x3 += __ll_B; /* yes, add it in the proper pos. */ \ | |
7128 | + \ | |
7129 | + (w1) = __x3 + __ll_highpart (__x1); \ | |
7130 | + (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \ | |
7131 | + } while (0) | |
7132 | +#endif | |
7133 | + | |
7134 | +#if !defined (__umulsidi3) | |
7135 | +#define __umulsidi3(u, v) \ | |
7136 | + ({DWunion __w; \ | |
7137 | + umul_ppmm (__w.s.high, __w.s.low, u, v); \ | |
7138 | + __w.ll; }) | |
7139 | +#endif | |
7140 | + | |
7141 | +/* Define this unconditionally, so it can be used for debugging. */ | |
7142 | +#define __udiv_qrnnd_c(q, r, n1, n0, d) \ | |
7143 | + do { \ | |
7144 | + UWtype __d1, __d0, __q1, __q0; \ | |
7145 | + UWtype __r1, __r0, __m; \ | |
7146 | + __d1 = __ll_highpart (d); \ | |
7147 | + __d0 = __ll_lowpart (d); \ | |
7148 | + \ | |
7149 | + __r1 = (n1) % __d1; \ | |
7150 | + __q1 = (n1) / __d1; \ | |
7151 | + __m = (UWtype) __q1 * __d0; \ | |
7152 | + __r1 = __r1 * __ll_B | __ll_highpart (n0); \ | |
7153 | + if (__r1 < __m) \ | |
7154 | + { \ | |
7155 | + __q1--, __r1 += (d); \ | |
7156 | + if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\ | |
7157 | + if (__r1 < __m) \ | |
7158 | + __q1--, __r1 += (d); \ | |
7159 | + } \ | |
7160 | + __r1 -= __m; \ | |
7161 | + \ | |
7162 | + __r0 = __r1 % __d1; \ | |
7163 | + __q0 = __r1 / __d1; \ | |
7164 | + __m = (UWtype) __q0 * __d0; \ | |
7165 | + __r0 = __r0 * __ll_B | __ll_lowpart (n0); \ | |
7166 | + if (__r0 < __m) \ | |
7167 | + { \ | |
7168 | + __q0--, __r0 += (d); \ | |
7169 | + if (__r0 >= (d)) \ | |
7170 | + if (__r0 < __m) \ | |
7171 | + __q0--, __r0 += (d); \ | |
7172 | + } \ | |
7173 | + __r0 -= __m; \ | |
7174 | + \ | |
7175 | + (q) = (UWtype) __q1 * __ll_B | __q0; \ | |
7176 | + (r) = __r0; \ | |
7177 | + } while (0) | |
7178 | + | |
7179 | +/* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through | |
7180 | + __udiv_w_sdiv (defined in libgcc or elsewhere). */ | |
7181 | +#if !defined (udiv_qrnnd) && defined (sdiv_qrnnd) | |
7182 | +#define udiv_qrnnd(q, r, nh, nl, d) \ | |
7183 | + do { \ | |
7184 | + USItype __r; \ | |
7185 | + (q) = __udiv_w_sdiv (&__r, nh, nl, d); \ | |
7186 | + (r) = __r; \ | |
7187 | + } while (0) | |
7188 | +#endif | |
7189 | + | |
7190 | +/* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */ | |
7191 | +#if !defined (udiv_qrnnd) | |
7192 | +#define UDIV_NEEDS_NORMALIZATION 1 | |
7193 | +#define udiv_qrnnd __udiv_qrnnd_c | |
7194 | +#endif | |
7195 | + | |
7196 | +#if !defined (count_leading_zeros) | |
7197 | +extern const UQItype __clz_tab[] ATTRIBUTE_HIDDEN; | |
7198 | +#define count_leading_zeros(count, x) \ | |
7199 | + do { \ | |
7200 | + UWtype __xr = (x); \ | |
7201 | + UWtype __a; \ | |
7202 | + \ | |
7203 | + if (W_TYPE_SIZE <= 32) \ | |
7204 | + { \ | |
7205 | + __a = __xr < ((UWtype)1<<2*__BITS4) \ | |
7206 | + ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \ | |
7207 | + : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \ | |
7208 | + } \ | |
7209 | + else \ | |
7210 | + { \ | |
7211 | + for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \ | |
7212 | + if (((__xr >> __a) & 0xff) != 0) \ | |
7213 | + break; \ | |
7214 | + } \ | |
7215 | + \ | |
7216 | + (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \ | |
7217 | + } while (0) | |
7218 | +#define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE | |
7219 | +#endif | |
7220 | + | |
7221 | +#if !defined (count_trailing_zeros) | |
7222 | +/* Define count_trailing_zeros using count_leading_zeros. The latter might be | |
7223 | + defined in asm, but if it is not, the C version above is good enough. */ | |
7224 | +#define count_trailing_zeros(count, x) \ | |
7225 | + do { \ | |
7226 | + UWtype __ctz_x = (x); \ | |
7227 | + UWtype __ctz_c; \ | |
7228 | + count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \ | |
7229 | + (count) = W_TYPE_SIZE - 1 - __ctz_c; \ | |
7230 | + } while (0) | |
7231 | +#endif | |
7232 | + | |
7233 | +#ifndef UDIV_NEEDS_NORMALIZATION | |
7234 | +#define UDIV_NEEDS_NORMALIZATION 0 | |
7235 | +#endif |