]>
Commit | Line | Data |
---|---|---|
203b91b9 RS |
1 | /* More subroutines needed by GCC output code on some machines. */ |
2 | /* Compile this one with gcc. */ | |
83ffe9cd | 3 | /* Copyright (C) 1989-2023 Free Software Foundation, Inc. |
203b91b9 | 4 | |
1322177d | 5 | This file is part of GCC. |
203b91b9 | 6 | |
1322177d LB |
7 | GCC is free software; you can redistribute it and/or modify it under |
8 | the terms of the GNU General Public License as published by the Free | |
748086b7 | 9 | Software Foundation; either version 3, or (at your option) any later |
1322177d | 10 | version. |
203b91b9 | 11 | |
1322177d LB |
12 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
13 | WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
14 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
15 | for more details. | |
203b91b9 | 16 | |
748086b7 JJ |
17 | Under Section 7 of GPL version 3, you are granted additional |
18 | permissions described in the GCC Runtime Library Exception, version | |
19 | 3.1, as published by the Free Software Foundation. | |
20 | ||
21 | You should have received a copy of the GNU General Public License and | |
22 | a copy of the GCC Runtime Library Exception along with this program; | |
23 | see the files COPYING3 and COPYING.RUNTIME respectively. If not, see | |
24 | <http://www.gnu.org/licenses/>. */ | |
203b91b9 | 25 | |
0dadecf6 | 26 | #include "tconfig.h" |
2e39bdbe | 27 | #include "tsystem.h" |
4977bab6 ZW |
28 | #include "coretypes.h" |
29 | #include "tm.h" | |
852b75ed | 30 | #include "libgcc_tm.h" |
2467749d | 31 | |
53585c36 RH |
32 | #ifdef HAVE_GAS_HIDDEN |
33 | #define ATTRIBUTE_HIDDEN __attribute__ ((__visibility__ ("hidden"))) | |
34 | #else | |
35 | #define ATTRIBUTE_HIDDEN | |
36 | #endif | |
37 | ||
b2a203c8 RS |
38 | /* Work out the largest "word" size that we can deal with on this target. */ |
39 | #if MIN_UNITS_PER_WORD > 4 | |
40 | # define LIBGCC2_MAX_UNITS_PER_WORD 8 | |
41 | #elif (MIN_UNITS_PER_WORD > 2 \ | |
4471aff6 | 42 | || (MIN_UNITS_PER_WORD > 1 && __SIZEOF_LONG_LONG__ > 4)) |
b2a203c8 RS |
43 | # define LIBGCC2_MAX_UNITS_PER_WORD 4 |
44 | #else | |
45 | # define LIBGCC2_MAX_UNITS_PER_WORD MIN_UNITS_PER_WORD | |
46 | #endif | |
47 | ||
48 | /* Work out what word size we are using for this compilation. | |
49 | The value can be set on the command line. */ | |
baffad1f | 50 | #ifndef LIBGCC2_UNITS_PER_WORD |
b2a203c8 | 51 | #define LIBGCC2_UNITS_PER_WORD LIBGCC2_MAX_UNITS_PER_WORD |
baffad1f RS |
52 | #endif |
53 | ||
b2a203c8 | 54 | #if LIBGCC2_UNITS_PER_WORD <= LIBGCC2_MAX_UNITS_PER_WORD |
baffad1f | 55 | |
299b83b7 | 56 | #include "libgcc2.h" |
203b91b9 | 57 | \f |
d8088c6f BS |
58 | #ifdef DECLARE_LIBRARY_RENAMES |
59 | DECLARE_LIBRARY_RENAMES | |
60 | #endif | |
61 | ||
b68daef4 | 62 | #if defined (L_negdi2) |
3d2adde6 CC |
63 | DWtype |
64 | __negdi2 (DWtype u) | |
65 | { | |
b982024e KG |
66 | const DWunion uu = {.ll = u}; |
67 | const DWunion w = { {.low = -uu.s.low, | |
68 | .high = -uu.s.high - ((UWtype) -uu.s.low > 0) } }; | |
3d2adde6 CC |
69 | |
70 | return w.ll; | |
71 | } | |
72 | #endif | |
91ce572a CC |
73 | |
74 | #ifdef L_addvsi3 | |
66f77154 | 75 | Wtype |
0aec6014 | 76 | __addvSI3 (Wtype a, Wtype b) |
91ce572a | 77 | { |
ebc4cd54 | 78 | Wtype w; |
91ce572a | 79 | |
831f24a7 | 80 | if (__builtin_add_overflow (a, b, &w)) |
91ce572a CC |
81 | abort (); |
82 | ||
83 | return w; | |
23190837 | 84 | } |
0aec6014 EB |
85 | #ifdef COMPAT_SIMODE_TRAPPING_ARITHMETIC |
86 | SItype | |
87 | __addvsi3 (SItype a, SItype b) | |
88 | { | |
ebc4cd54 | 89 | SItype w; |
0aec6014 | 90 | |
831f24a7 | 91 | if (__builtin_add_overflow (a, b, &w)) |
0aec6014 EB |
92 | abort (); |
93 | ||
94 | return w; | |
95 | } | |
96 | #endif /* COMPAT_SIMODE_TRAPPING_ARITHMETIC */ | |
3d2adde6 | 97 | #endif |
91ce572a CC |
98 | \f |
99 | #ifdef L_addvdi3 | |
66f77154 | 100 | DWtype |
0aec6014 | 101 | __addvDI3 (DWtype a, DWtype b) |
91ce572a | 102 | { |
ebc4cd54 | 103 | DWtype w; |
91ce572a | 104 | |
831f24a7 | 105 | if (__builtin_add_overflow (a, b, &w)) |
91ce572a CC |
106 | abort (); |
107 | ||
108 | return w; | |
109 | } | |
110 | #endif | |
111 | \f | |
112 | #ifdef L_subvsi3 | |
66f77154 | 113 | Wtype |
0aec6014 | 114 | __subvSI3 (Wtype a, Wtype b) |
91ce572a | 115 | { |
ebc4cd54 | 116 | Wtype w; |
91ce572a | 117 | |
831f24a7 | 118 | if (__builtin_sub_overflow (a, b, &w)) |
91ce572a CC |
119 | abort (); |
120 | ||
121 | return w; | |
91ce572a | 122 | } |
0aec6014 EB |
123 | #ifdef COMPAT_SIMODE_TRAPPING_ARITHMETIC |
124 | SItype | |
125 | __subvsi3 (SItype a, SItype b) | |
126 | { | |
ebc4cd54 | 127 | SItype w; |
0aec6014 | 128 | |
831f24a7 | 129 | if (__builtin_sub_overflow (a, b, &w)) |
0aec6014 EB |
130 | abort (); |
131 | ||
132 | return w; | |
133 | } | |
134 | #endif /* COMPAT_SIMODE_TRAPPING_ARITHMETIC */ | |
91ce572a CC |
135 | #endif |
136 | \f | |
137 | #ifdef L_subvdi3 | |
66f77154 | 138 | DWtype |
0aec6014 | 139 | __subvDI3 (DWtype a, DWtype b) |
91ce572a | 140 | { |
ebc4cd54 | 141 | DWtype w; |
91ce572a | 142 | |
831f24a7 | 143 | if (__builtin_sub_overflow (a, b, &w)) |
91ce572a CC |
144 | abort (); |
145 | ||
146 | return w; | |
91ce572a CC |
147 | } |
148 | #endif | |
149 | \f | |
150 | #ifdef L_mulvsi3 | |
66f77154 | 151 | Wtype |
0aec6014 | 152 | __mulvSI3 (Wtype a, Wtype b) |
91ce572a | 153 | { |
ebc4cd54 | 154 | Wtype w; |
91ce572a | 155 | |
831f24a7 | 156 | if (__builtin_mul_overflow (a, b, &w)) |
91ce572a CC |
157 | abort (); |
158 | ||
159 | return w; | |
160 | } | |
0aec6014 | 161 | #ifdef COMPAT_SIMODE_TRAPPING_ARITHMETIC |
0aec6014 EB |
162 | SItype |
163 | __mulvsi3 (SItype a, SItype b) | |
164 | { | |
ebc4cd54 | 165 | SItype w; |
0aec6014 | 166 | |
831f24a7 | 167 | if (__builtin_mul_overflow (a, b, &w)) |
0aec6014 EB |
168 | abort (); |
169 | ||
170 | return w; | |
171 | } | |
172 | #endif /* COMPAT_SIMODE_TRAPPING_ARITHMETIC */ | |
91ce572a CC |
173 | #endif |
174 | \f | |
175 | #ifdef L_negvsi2 | |
66f77154 | 176 | Wtype |
0aec6014 | 177 | __negvSI2 (Wtype a) |
91ce572a | 178 | { |
ebc4cd54 | 179 | Wtype w; |
91ce572a | 180 | |
831f24a7 | 181 | if (__builtin_sub_overflow (0, a, &w)) |
91ce572a CC |
182 | abort (); |
183 | ||
ebc4cd54 | 184 | return w; |
91ce572a | 185 | } |
0aec6014 EB |
186 | #ifdef COMPAT_SIMODE_TRAPPING_ARITHMETIC |
187 | SItype | |
188 | __negvsi2 (SItype a) | |
189 | { | |
ebc4cd54 | 190 | SItype w; |
0aec6014 | 191 | |
831f24a7 | 192 | if (__builtin_sub_overflow (0, a, &w)) |
0aec6014 EB |
193 | abort (); |
194 | ||
ebc4cd54 | 195 | return w; |
0aec6014 EB |
196 | } |
197 | #endif /* COMPAT_SIMODE_TRAPPING_ARITHMETIC */ | |
91ce572a CC |
198 | #endif |
199 | \f | |
200 | #ifdef L_negvdi2 | |
66f77154 | 201 | DWtype |
0aec6014 | 202 | __negvDI2 (DWtype a) |
91ce572a | 203 | { |
ebc4cd54 | 204 | DWtype w; |
91ce572a | 205 | |
831f24a7 | 206 | if (__builtin_sub_overflow (0, a, &w)) |
91ce572a CC |
207 | abort (); |
208 | ||
e11e816e | 209 | return w; |
91ce572a CC |
210 | } |
211 | #endif | |
212 | \f | |
213 | #ifdef L_absvsi2 | |
66f77154 | 214 | Wtype |
0aec6014 | 215 | __absvSI2 (Wtype a) |
91ce572a | 216 | { |
4919ed71 SK |
217 | const Wtype v = 0 - (a < 0); |
218 | Wtype w; | |
0aec6014 | 219 | |
4919ed71 | 220 | if (__builtin_add_overflow (a, v, &w)) |
0aec6014 | 221 | abort (); |
0aec6014 | 222 | |
4919ed71 | 223 | return v ^ w; |
0aec6014 EB |
224 | } |
225 | #ifdef COMPAT_SIMODE_TRAPPING_ARITHMETIC | |
226 | SItype | |
227 | __absvsi2 (SItype a) | |
228 | { | |
4919ed71 SK |
229 | const SItype v = 0 - (a < 0); |
230 | SItype w; | |
91ce572a | 231 | |
4919ed71 | 232 | if (__builtin_add_overflow (a, v, &w)) |
e11e816e | 233 | abort (); |
91ce572a | 234 | |
4919ed71 | 235 | return v ^ w; |
91ce572a | 236 | } |
0aec6014 | 237 | #endif /* COMPAT_SIMODE_TRAPPING_ARITHMETIC */ |
91ce572a CC |
238 | #endif |
239 | \f | |
240 | #ifdef L_absvdi2 | |
66f77154 | 241 | DWtype |
0aec6014 | 242 | __absvDI2 (DWtype a) |
91ce572a | 243 | { |
4919ed71 SK |
244 | const DWtype v = 0 - (a < 0); |
245 | DWtype w; | |
91ce572a | 246 | |
4919ed71 | 247 | if (__builtin_add_overflow (a, v, &w)) |
e11e816e | 248 | abort (); |
91ce572a | 249 | |
4919ed71 | 250 | return v ^ w; |
91ce572a CC |
251 | } |
252 | #endif | |
253 | \f | |
254 | #ifdef L_mulvdi3 | |
66f77154 | 255 | DWtype |
0aec6014 | 256 | __mulvDI3 (DWtype u, DWtype v) |
91ce572a | 257 | { |
4c20b2e7 BH |
258 | /* The unchecked multiplication needs 3 Wtype x Wtype multiplications, |
259 | but the checked multiplication needs only two. */ | |
b982024e KG |
260 | const DWunion uu = {.ll = u}; |
261 | const DWunion vv = {.ll = v}; | |
91ce572a | 262 | |
4f2e0d5e | 263 | if (__builtin_expect (uu.s.high == uu.s.low >> (W_TYPE_SIZE - 1), 1)) |
4c20b2e7 BH |
264 | { |
265 | /* u fits in a single Wtype. */ | |
4f2e0d5e | 266 | if (__builtin_expect (vv.s.high == vv.s.low >> (W_TYPE_SIZE - 1), 1)) |
4c20b2e7 BH |
267 | { |
268 | /* v fits in a single Wtype as well. */ | |
269 | /* A single multiplication. No overflow risk. */ | |
270 | return (DWtype) uu.s.low * (DWtype) vv.s.low; | |
271 | } | |
272 | else | |
273 | { | |
274 | /* Two multiplications. */ | |
b982024e KG |
275 | DWunion w0 = {.ll = (UDWtype) (UWtype) uu.s.low |
276 | * (UDWtype) (UWtype) vv.s.low}; | |
277 | DWunion w1 = {.ll = (UDWtype) (UWtype) uu.s.low | |
278 | * (UDWtype) (UWtype) vv.s.high}; | |
4c20b2e7 | 279 | |
4c20b2e7 BH |
280 | if (vv.s.high < 0) |
281 | w1.s.high -= uu.s.low; | |
282 | if (uu.s.low < 0) | |
283 | w1.ll -= vv.ll; | |
284 | w1.ll += (UWtype) w0.s.high; | |
4f2e0d5e | 285 | if (__builtin_expect (w1.s.high == w1.s.low >> (W_TYPE_SIZE - 1), 1)) |
4c20b2e7 BH |
286 | { |
287 | w0.s.high = w1.s.low; | |
288 | return w0.ll; | |
289 | } | |
290 | } | |
291 | } | |
292 | else | |
293 | { | |
4f2e0d5e | 294 | if (__builtin_expect (vv.s.high == vv.s.low >> (W_TYPE_SIZE - 1), 1)) |
4c20b2e7 BH |
295 | { |
296 | /* v fits into a single Wtype. */ | |
297 | /* Two multiplications. */ | |
b982024e KG |
298 | DWunion w0 = {.ll = (UDWtype) (UWtype) uu.s.low |
299 | * (UDWtype) (UWtype) vv.s.low}; | |
300 | DWunion w1 = {.ll = (UDWtype) (UWtype) uu.s.high | |
301 | * (UDWtype) (UWtype) vv.s.low}; | |
4c20b2e7 | 302 | |
4c20b2e7 BH |
303 | if (uu.s.high < 0) |
304 | w1.s.high -= vv.s.low; | |
305 | if (vv.s.low < 0) | |
306 | w1.ll -= uu.ll; | |
307 | w1.ll += (UWtype) w0.s.high; | |
4f2e0d5e | 308 | if (__builtin_expect (w1.s.high == w1.s.low >> (W_TYPE_SIZE - 1), 1)) |
4c20b2e7 BH |
309 | { |
310 | w0.s.high = w1.s.low; | |
311 | return w0.ll; | |
312 | } | |
313 | } | |
314 | else | |
315 | { | |
316 | /* A few sign checks and a single multiplication. */ | |
317 | if (uu.s.high >= 0) | |
318 | { | |
319 | if (vv.s.high >= 0) | |
320 | { | |
321 | if (uu.s.high == 0 && vv.s.high == 0) | |
322 | { | |
b982024e KG |
323 | const DWtype w = (UDWtype) (UWtype) uu.s.low |
324 | * (UDWtype) (UWtype) vv.s.low; | |
4c20b2e7 BH |
325 | if (__builtin_expect (w >= 0, 1)) |
326 | return w; | |
327 | } | |
328 | } | |
329 | else | |
330 | { | |
331 | if (uu.s.high == 0 && vv.s.high == (Wtype) -1) | |
332 | { | |
b982024e KG |
333 | DWunion ww = {.ll = (UDWtype) (UWtype) uu.s.low |
334 | * (UDWtype) (UWtype) vv.s.low}; | |
4c20b2e7 | 335 | |
4c20b2e7 BH |
336 | ww.s.high -= uu.s.low; |
337 | if (__builtin_expect (ww.s.high < 0, 1)) | |
338 | return ww.ll; | |
339 | } | |
340 | } | |
341 | } | |
342 | else | |
343 | { | |
344 | if (vv.s.high >= 0) | |
345 | { | |
346 | if (uu.s.high == (Wtype) -1 && vv.s.high == 0) | |
347 | { | |
b982024e KG |
348 | DWunion ww = {.ll = (UDWtype) (UWtype) uu.s.low |
349 | * (UDWtype) (UWtype) vv.s.low}; | |
4c20b2e7 | 350 | |
4c20b2e7 BH |
351 | ww.s.high -= vv.s.low; |
352 | if (__builtin_expect (ww.s.high < 0, 1)) | |
353 | return ww.ll; | |
354 | } | |
355 | } | |
356 | else | |
357 | { | |
e7176f75 JJ |
358 | if ((uu.s.high & vv.s.high) == (Wtype) -1 |
359 | && (uu.s.low | vv.s.low) != 0) | |
4c20b2e7 | 360 | { |
b982024e KG |
361 | DWunion ww = {.ll = (UDWtype) (UWtype) uu.s.low |
362 | * (UDWtype) (UWtype) vv.s.low}; | |
4c20b2e7 | 363 | |
4c20b2e7 BH |
364 | ww.s.high -= uu.s.low; |
365 | ww.s.high -= vv.s.low; | |
366 | if (__builtin_expect (ww.s.high >= 0, 1)) | |
367 | return ww.ll; | |
368 | } | |
369 | } | |
370 | } | |
371 | } | |
372 | } | |
91ce572a | 373 | |
4c20b2e7 BH |
374 | /* Overflow. */ |
375 | abort (); | |
91ce572a CC |
376 | } |
377 | #endif | |
378 | \f | |
203b91b9 | 379 | |
3d042e77 | 380 | /* Unless shift functions are defined with full ANSI prototypes, |
c7ff6e7a | 381 | parameter b will be promoted to int if shift_count_type is smaller than an int. */ |
203b91b9 | 382 | #ifdef L_lshrdi3 |
996ed075 | 383 | DWtype |
c7ff6e7a | 384 | __lshrdi3 (DWtype u, shift_count_type b) |
203b91b9 | 385 | { |
203b91b9 RS |
386 | if (b == 0) |
387 | return u; | |
388 | ||
b982024e | 389 | const DWunion uu = {.ll = u}; |
fdf3e18a | 390 | const shift_count_type bm = W_TYPE_SIZE - b; |
b982024e | 391 | DWunion w; |
203b91b9 | 392 | |
203b91b9 RS |
393 | if (bm <= 0) |
394 | { | |
395 | w.s.high = 0; | |
6da9c622 | 396 | w.s.low = (UWtype) uu.s.high >> -bm; |
203b91b9 RS |
397 | } |
398 | else | |
399 | { | |
b982024e | 400 | const UWtype carries = (UWtype) uu.s.high << bm; |
6da9c622 RK |
401 | |
402 | w.s.high = (UWtype) uu.s.high >> b; | |
403 | w.s.low = ((UWtype) uu.s.low >> b) | carries; | |
203b91b9 RS |
404 | } |
405 | ||
406 | return w.ll; | |
407 | } | |
408 | #endif | |
409 | ||
410 | #ifdef L_ashldi3 | |
996ed075 | 411 | DWtype |
c7ff6e7a | 412 | __ashldi3 (DWtype u, shift_count_type b) |
203b91b9 | 413 | { |
203b91b9 RS |
414 | if (b == 0) |
415 | return u; | |
416 | ||
b982024e | 417 | const DWunion uu = {.ll = u}; |
fdf3e18a | 418 | const shift_count_type bm = W_TYPE_SIZE - b; |
b982024e | 419 | DWunion w; |
203b91b9 | 420 | |
203b91b9 RS |
421 | if (bm <= 0) |
422 | { | |
423 | w.s.low = 0; | |
6da9c622 | 424 | w.s.high = (UWtype) uu.s.low << -bm; |
203b91b9 RS |
425 | } |
426 | else | |
427 | { | |
b982024e | 428 | const UWtype carries = (UWtype) uu.s.low >> bm; |
6da9c622 RK |
429 | |
430 | w.s.low = (UWtype) uu.s.low << b; | |
431 | w.s.high = ((UWtype) uu.s.high << b) | carries; | |
203b91b9 RS |
432 | } |
433 | ||
434 | return w.ll; | |
435 | } | |
436 | #endif | |
437 | ||
438 | #ifdef L_ashrdi3 | |
996ed075 | 439 | DWtype |
c7ff6e7a | 440 | __ashrdi3 (DWtype u, shift_count_type b) |
203b91b9 | 441 | { |
203b91b9 RS |
442 | if (b == 0) |
443 | return u; | |
444 | ||
b982024e | 445 | const DWunion uu = {.ll = u}; |
fdf3e18a | 446 | const shift_count_type bm = W_TYPE_SIZE - b; |
b982024e | 447 | DWunion w; |
203b91b9 | 448 | |
203b91b9 RS |
449 | if (bm <= 0) |
450 | { | |
451 | /* w.s.high = 1..1 or 0..0 */ | |
fdf3e18a | 452 | w.s.high = uu.s.high >> (W_TYPE_SIZE - 1); |
203b91b9 RS |
453 | w.s.low = uu.s.high >> -bm; |
454 | } | |
455 | else | |
456 | { | |
b982024e | 457 | const UWtype carries = (UWtype) uu.s.high << bm; |
6da9c622 | 458 | |
203b91b9 | 459 | w.s.high = uu.s.high >> b; |
6da9c622 | 460 | w.s.low = ((UWtype) uu.s.low >> b) | carries; |
203b91b9 RS |
461 | } |
462 | ||
463 | return w.ll; | |
464 | } | |
465 | #endif | |
466 | \f | |
167fa32c | 467 | #ifdef L_bswapsi2 |
e4b6bec2 EC |
468 | SItype |
469 | __bswapsi2 (SItype u) | |
167fa32c | 470 | { |
a8ae2392 SK |
471 | return ((((u) & 0xff000000u) >> 24) |
472 | | (((u) & 0x00ff0000u) >> 8) | |
473 | | (((u) & 0x0000ff00u) << 8) | |
474 | | (((u) & 0x000000ffu) << 24)); | |
167fa32c EC |
475 | } |
476 | #endif | |
477 | #ifdef L_bswapdi2 | |
e4b6bec2 EC |
478 | DItype |
479 | __bswapdi2 (DItype u) | |
167fa32c EC |
480 | { |
481 | return ((((u) & 0xff00000000000000ull) >> 56) | |
482 | | (((u) & 0x00ff000000000000ull) >> 40) | |
483 | | (((u) & 0x0000ff0000000000ull) >> 24) | |
484 | | (((u) & 0x000000ff00000000ull) >> 8) | |
485 | | (((u) & 0x00000000ff000000ull) << 8) | |
486 | | (((u) & 0x0000000000ff0000ull) << 24) | |
487 | | (((u) & 0x000000000000ff00ull) << 40) | |
488 | | (((u) & 0x00000000000000ffull) << 56)); | |
489 | } | |
490 | #endif | |
dfff898c RH |
491 | #ifdef L_ffssi2 |
492 | #undef int | |
dfff898c RH |
493 | int |
494 | __ffsSI2 (UWtype u) | |
495 | { | |
496 | UWtype count; | |
497 | ||
498 | if (u == 0) | |
499 | return 0; | |
500 | ||
501 | count_trailing_zeros (count, u); | |
502 | return count + 1; | |
503 | } | |
504 | #endif | |
505 | \f | |
aa66bd06 | 506 | #ifdef L_ffsdi2 |
dabb3f04 | 507 | #undef int |
dabb3f04 | 508 | int |
dfff898c | 509 | __ffsDI2 (DWtype u) |
aa66bd06 | 510 | { |
b982024e | 511 | const DWunion uu = {.ll = u}; |
d6eacd48 RH |
512 | UWtype word, count, add; |
513 | ||
d6eacd48 RH |
514 | if (uu.s.low != 0) |
515 | word = uu.s.low, add = 0; | |
516 | else if (uu.s.high != 0) | |
fdf3e18a | 517 | word = uu.s.high, add = W_TYPE_SIZE; |
d6eacd48 RH |
518 | else |
519 | return 0; | |
520 | ||
521 | count_trailing_zeros (count, word); | |
522 | return count + add + 1; | |
aa66bd06 RS |
523 | } |
524 | #endif | |
525 | \f | |
203b91b9 | 526 | #ifdef L_muldi3 |
996ed075 JJ |
527 | DWtype |
528 | __muldi3 (DWtype u, DWtype v) | |
203b91b9 | 529 | { |
b982024e KG |
530 | const DWunion uu = {.ll = u}; |
531 | const DWunion vv = {.ll = v}; | |
532 | DWunion w = {.ll = __umulsidi3 (uu.s.low, vv.s.low)}; | |
203b91b9 | 533 | |
996ed075 JJ |
534 | w.s.high += ((UWtype) uu.s.low * (UWtype) vv.s.high |
535 | + (UWtype) uu.s.high * (UWtype) vv.s.low); | |
203b91b9 RS |
536 | |
537 | return w.ll; | |
538 | } | |
539 | #endif | |
540 | \f | |
59798a0c UW |
541 | #if (defined (L_udivdi3) || defined (L_divdi3) || \ |
542 | defined (L_umoddi3) || defined (L_moddi3)) | |
f8eef883 | 543 | #if defined (sdiv_qrnnd) |
59798a0c UW |
544 | #define L_udiv_w_sdiv |
545 | #endif | |
f8eef883 | 546 | #endif |
59798a0c | 547 | |
3904131a | 548 | #ifdef L_udiv_w_sdiv |
ce13d15f | 549 | #if defined (sdiv_qrnnd) |
59798a0c UW |
550 | #if (defined (L_udivdi3) || defined (L_divdi3) || \ |
551 | defined (L_umoddi3) || defined (L_moddi3)) | |
1ab9ba62 | 552 | static inline __attribute__ ((__always_inline__)) |
59798a0c | 553 | #endif |
996ed075 JJ |
554 | UWtype |
555 | __udiv_w_sdiv (UWtype *rp, UWtype a1, UWtype a0, UWtype d) | |
431b1ee0 | 556 | { |
996ed075 JJ |
557 | UWtype q, r; |
558 | UWtype c0, c1, b1; | |
431b1ee0 | 559 | |
996ed075 | 560 | if ((Wtype) d >= 0) |
431b1ee0 | 561 | { |
996ed075 | 562 | if (a1 < d - a1 - (a0 >> (W_TYPE_SIZE - 1))) |
431b1ee0 | 563 | { |
ea4b7848 | 564 | /* Dividend, divisor, and quotient are nonnegative. */ |
431b1ee0 TG |
565 | sdiv_qrnnd (q, r, a1, a0, d); |
566 | } | |
567 | else | |
568 | { | |
ea4b7848 | 569 | /* Compute c1*2^32 + c0 = a1*2^32 + a0 - 2^31*d. */ |
996ed075 | 570 | sub_ddmmss (c1, c0, a1, a0, d >> 1, d << (W_TYPE_SIZE - 1)); |
ea4b7848 | 571 | /* Divide (c1*2^32 + c0) by d. */ |
431b1ee0 | 572 | sdiv_qrnnd (q, r, c1, c0, d); |
ea4b7848 | 573 | /* Add 2^31 to quotient. */ |
996ed075 | 574 | q += (UWtype) 1 << (W_TYPE_SIZE - 1); |
431b1ee0 TG |
575 | } |
576 | } | |
577 | else | |
578 | { | |
579 | b1 = d >> 1; /* d/2, between 2^30 and 2^31 - 1 */ | |
580 | c1 = a1 >> 1; /* A/2 */ | |
996ed075 | 581 | c0 = (a1 << (W_TYPE_SIZE - 1)) + (a0 >> 1); |
431b1ee0 TG |
582 | |
583 | if (a1 < b1) /* A < 2^32*b1, so A/2 < 2^31*b1 */ | |
584 | { | |
585 | sdiv_qrnnd (q, r, c1, c0, b1); /* (A/2) / (d/2) */ | |
586 | ||
587 | r = 2*r + (a0 & 1); /* Remainder from A/(2*b1) */ | |
588 | if ((d & 1) != 0) | |
589 | { | |
590 | if (r >= q) | |
591 | r = r - q; | |
592 | else if (q - r <= d) | |
593 | { | |
594 | r = r - q + d; | |
595 | q--; | |
596 | } | |
597 | else | |
598 | { | |
599 | r = r - q + 2*d; | |
600 | q -= 2; | |
601 | } | |
602 | } | |
603 | } | |
604 | else if (c1 < b1) /* So 2^31 <= (A/2)/b1 < 2^32 */ | |
605 | { | |
606 | c1 = (b1 - 1) - c1; | |
607 | c0 = ~c0; /* logical NOT */ | |
608 | ||
609 | sdiv_qrnnd (q, r, c1, c0, b1); /* (A/2) / (d/2) */ | |
610 | ||
611 | q = ~q; /* (A/2)/b1 */ | |
612 | r = (b1 - 1) - r; | |
613 | ||
614 | r = 2*r + (a0 & 1); /* A/(2*b1) */ | |
615 | ||
616 | if ((d & 1) != 0) | |
617 | { | |
618 | if (r >= q) | |
619 | r = r - q; | |
620 | else if (q - r <= d) | |
621 | { | |
622 | r = r - q + d; | |
623 | q--; | |
624 | } | |
625 | else | |
626 | { | |
627 | r = r - q + 2*d; | |
628 | q -= 2; | |
629 | } | |
630 | } | |
631 | } | |
632 | else /* Implies c1 = b1 */ | |
633 | { /* Hence a1 = d - 1 = 2*b1 - 1 */ | |
634 | if (a0 >= -d) | |
635 | { | |
636 | q = -1; | |
637 | r = a0 + d; | |
638 | } | |
639 | else | |
640 | { | |
641 | q = -2; | |
642 | r = a0 + 2*d; | |
643 | } | |
644 | } | |
645 | } | |
646 | ||
647 | *rp = r; | |
648 | return q; | |
649 | } | |
ce13d15f RK |
650 | #else |
651 | /* If sdiv_qrnnd doesn't exist, define dummy __udiv_w_sdiv. */ | |
996ed075 JJ |
652 | UWtype |
653 | __udiv_w_sdiv (UWtype *rp __attribute__ ((__unused__)), | |
654 | UWtype a1 __attribute__ ((__unused__)), | |
655 | UWtype a0 __attribute__ ((__unused__)), | |
656 | UWtype d __attribute__ ((__unused__))) | |
081f5e7e KG |
657 | { |
658 | return 0; | |
659 | } | |
ce13d15f | 660 | #endif |
431b1ee0 TG |
661 | #endif |
662 | \f | |
536bfcd0 | 663 | #if (defined (L_udivdi3) || defined (L_divdi3) || \ |
18362447 UB |
664 | defined (L_umoddi3) || defined (L_moddi3) || \ |
665 | defined (L_divmoddi4)) | |
536bfcd0 RK |
666 | #define L_udivmoddi4 |
667 | #endif | |
668 | ||
d6eacd48 | 669 | #ifdef L_clz |
dcfae47c | 670 | const UQItype __clz_tab[256] = |
203b91b9 RS |
671 | { |
672 | 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, | |
673 | 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, | |
674 | 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, | |
675 | 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, | |
676 | 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, | |
677 | 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, | |
678 | 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, | |
dcfae47c | 679 | 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 |
203b91b9 | 680 | }; |
d6eacd48 | 681 | #endif |
2928cd7a RH |
682 | \f |
683 | #ifdef L_clzsi2 | |
dabb3f04 | 684 | #undef int |
dabb3f04 | 685 | int |
8275b011 | 686 | __clzSI2 (UWtype x) |
2928cd7a | 687 | { |
53585c36 | 688 | Wtype ret; |
2928cd7a | 689 | |
8275b011 | 690 | count_leading_zeros (ret, x); |
53585c36 RH |
691 | |
692 | return ret; | |
2928cd7a RH |
693 | } |
694 | #endif | |
695 | \f | |
696 | #ifdef L_clzdi2 | |
dabb3f04 | 697 | #undef int |
dabb3f04 | 698 | int |
8275b011 | 699 | __clzDI2 (UDWtype x) |
2928cd7a | 700 | { |
b982024e | 701 | const DWunion uu = {.ll = x}; |
53585c36 RH |
702 | UWtype word; |
703 | Wtype ret, add; | |
704 | ||
8275b011 RH |
705 | if (uu.s.high) |
706 | word = uu.s.high, add = 0; | |
53585c36 | 707 | else |
8275b011 | 708 | word = uu.s.low, add = W_TYPE_SIZE; |
2928cd7a | 709 | |
53585c36 RH |
710 | count_leading_zeros (ret, word); |
711 | return ret + add; | |
2928cd7a RH |
712 | } |
713 | #endif | |
714 | \f | |
715 | #ifdef L_ctzsi2 | |
dabb3f04 | 716 | #undef int |
dabb3f04 | 717 | int |
8275b011 | 718 | __ctzSI2 (UWtype x) |
2928cd7a | 719 | { |
53585c36 | 720 | Wtype ret; |
2928cd7a | 721 | |
53585c36 | 722 | count_trailing_zeros (ret, x); |
2928cd7a | 723 | |
53585c36 | 724 | return ret; |
2928cd7a RH |
725 | } |
726 | #endif | |
727 | \f | |
728 | #ifdef L_ctzdi2 | |
dabb3f04 | 729 | #undef int |
dabb3f04 | 730 | int |
8275b011 | 731 | __ctzDI2 (UDWtype x) |
2928cd7a | 732 | { |
b982024e | 733 | const DWunion uu = {.ll = x}; |
53585c36 RH |
734 | UWtype word; |
735 | Wtype ret, add; | |
736 | ||
8275b011 RH |
737 | if (uu.s.low) |
738 | word = uu.s.low, add = 0; | |
53585c36 | 739 | else |
8275b011 | 740 | word = uu.s.high, add = W_TYPE_SIZE; |
2928cd7a | 741 | |
53585c36 RH |
742 | count_trailing_zeros (ret, word); |
743 | return ret + add; | |
2928cd7a RH |
744 | } |
745 | #endif | |
3801c801 BS |
746 | \f |
747 | #ifdef L_clrsbsi2 | |
748 | #undef int | |
749 | int | |
750 | __clrsbSI2 (Wtype x) | |
751 | { | |
752 | Wtype ret; | |
2928cd7a | 753 | |
3801c801 BS |
754 | if (x < 0) |
755 | x = ~x; | |
756 | if (x == 0) | |
757 | return W_TYPE_SIZE - 1; | |
758 | count_leading_zeros (ret, x); | |
759 | return ret - 1; | |
760 | } | |
761 | #endif | |
762 | \f | |
763 | #ifdef L_clrsbdi2 | |
764 | #undef int | |
765 | int | |
766 | __clrsbDI2 (DWtype x) | |
767 | { | |
768 | const DWunion uu = {.ll = x}; | |
769 | UWtype word; | |
770 | Wtype ret, add; | |
771 | ||
772 | if (uu.s.high == 0) | |
773 | word = uu.s.low, add = W_TYPE_SIZE; | |
774 | else if (uu.s.high == -1) | |
775 | word = ~uu.s.low, add = W_TYPE_SIZE; | |
776 | else if (uu.s.high >= 0) | |
777 | word = uu.s.high, add = 0; | |
778 | else | |
779 | word = ~uu.s.high, add = 0; | |
780 | ||
781 | if (word == 0) | |
782 | ret = W_TYPE_SIZE; | |
783 | else | |
784 | count_leading_zeros (ret, word); | |
785 | ||
786 | return ret + add - 1; | |
787 | } | |
788 | #endif | |
789 | \f | |
2928cd7a | 790 | #ifdef L_popcount_tab |
dcfae47c | 791 | const UQItype __popcount_tab[256] = |
2928cd7a RH |
792 | { |
793 | 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, | |
794 | 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, | |
795 | 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, | |
796 | 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, | |
797 | 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, | |
798 | 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, | |
799 | 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, | |
dcfae47c | 800 | 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 |
2928cd7a RH |
801 | }; |
802 | #endif | |
803 | \f | |
4ea3d774 | 804 | #if defined(L_popcountsi2) || defined(L_popcountdi2) |
a153644f TS |
805 | #define POPCOUNTCST2(x) (((UWtype) x << __CHAR_BIT__) | x) |
806 | #define POPCOUNTCST4(x) (((UWtype) x << (2 * __CHAR_BIT__)) | x) | |
807 | #define POPCOUNTCST8(x) (((UWtype) x << (4 * __CHAR_BIT__)) | x) | |
808 | #if W_TYPE_SIZE == __CHAR_BIT__ | |
4ea3d774 | 809 | #define POPCOUNTCST(x) x |
a153644f | 810 | #elif W_TYPE_SIZE == 2 * __CHAR_BIT__ |
4ea3d774 | 811 | #define POPCOUNTCST(x) POPCOUNTCST2 (x) |
a153644f | 812 | #elif W_TYPE_SIZE == 4 * __CHAR_BIT__ |
4ea3d774 | 813 | #define POPCOUNTCST(x) POPCOUNTCST4 (POPCOUNTCST2 (x)) |
a153644f | 814 | #elif W_TYPE_SIZE == 8 * __CHAR_BIT__ |
4ea3d774 JJ |
815 | #define POPCOUNTCST(x) POPCOUNTCST8 (POPCOUNTCST4 (POPCOUNTCST2 (x))) |
816 | #endif | |
817 | #endif | |
818 | \f | |
2928cd7a | 819 | #ifdef L_popcountsi2 |
dabb3f04 | 820 | #undef int |
dabb3f04 | 821 | int |
8275b011 | 822 | __popcountSI2 (UWtype x) |
2928cd7a | 823 | { |
4ea3d774 JJ |
824 | /* Force table lookup on targets like AVR and RL78 which only |
825 | pretend they have LIBGCC2_UNITS_PER_WORD 4, but actually | |
826 | have 1, and other small word targets. */ | |
a153644f | 827 | #if __SIZEOF_INT__ > 2 && defined (POPCOUNTCST) && __CHAR_BIT__ == 8 |
4ea3d774 JJ |
828 | x = x - ((x >> 1) & POPCOUNTCST (0x55)); |
829 | x = (x & POPCOUNTCST (0x33)) + ((x >> 2) & POPCOUNTCST (0x33)); | |
830 | x = (x + (x >> 4)) & POPCOUNTCST (0x0F); | |
a153644f | 831 | return (x * POPCOUNTCST (0x01)) >> (W_TYPE_SIZE - __CHAR_BIT__); |
4ea3d774 | 832 | #else |
4000debb | 833 | int i, ret = 0; |
8275b011 RH |
834 | |
835 | for (i = 0; i < W_TYPE_SIZE; i += 8) | |
836 | ret += __popcount_tab[(x >> i) & 0xff]; | |
837 | ||
838 | return ret; | |
4ea3d774 | 839 | #endif |
2928cd7a RH |
840 | } |
841 | #endif | |
842 | \f | |
843 | #ifdef L_popcountdi2 | |
dabb3f04 | 844 | #undef int |
dabb3f04 | 845 | int |
8275b011 | 846 | __popcountDI2 (UDWtype x) |
2928cd7a | 847 | { |
4ea3d774 JJ |
848 | /* Force table lookup on targets like AVR and RL78 which only |
849 | pretend they have LIBGCC2_UNITS_PER_WORD 4, but actually | |
850 | have 1, and other small word targets. */ | |
a153644f | 851 | #if __SIZEOF_INT__ > 2 && defined (POPCOUNTCST) && __CHAR_BIT__ == 8 |
4ea3d774 JJ |
852 | const DWunion uu = {.ll = x}; |
853 | UWtype x1 = uu.s.low, x2 = uu.s.high; | |
854 | x1 = x1 - ((x1 >> 1) & POPCOUNTCST (0x55)); | |
855 | x2 = x2 - ((x2 >> 1) & POPCOUNTCST (0x55)); | |
856 | x1 = (x1 & POPCOUNTCST (0x33)) + ((x1 >> 2) & POPCOUNTCST (0x33)); | |
857 | x2 = (x2 & POPCOUNTCST (0x33)) + ((x2 >> 2) & POPCOUNTCST (0x33)); | |
858 | x1 = (x1 + (x1 >> 4)) & POPCOUNTCST (0x0F); | |
859 | x2 = (x2 + (x2 >> 4)) & POPCOUNTCST (0x0F); | |
860 | x1 += x2; | |
a153644f | 861 | return (x1 * POPCOUNTCST (0x01)) >> (W_TYPE_SIZE - __CHAR_BIT__); |
4ea3d774 | 862 | #else |
4000debb | 863 | int i, ret = 0; |
8275b011 RH |
864 | |
865 | for (i = 0; i < 2*W_TYPE_SIZE; i += 8) | |
866 | ret += __popcount_tab[(x >> i) & 0xff]; | |
867 | ||
868 | return ret; | |
4ea3d774 | 869 | #endif |
2928cd7a RH |
870 | } |
871 | #endif | |
872 | \f | |
873 | #ifdef L_paritysi2 | |
dabb3f04 | 874 | #undef int |
dabb3f04 | 875 | int |
8275b011 | 876 | __paritySI2 (UWtype x) |
2928cd7a | 877 | { |
8275b011 RH |
878 | #if W_TYPE_SIZE > 64 |
879 | # error "fill out the table" | |
880 | #endif | |
881 | #if W_TYPE_SIZE > 32 | |
882 | x ^= x >> 32; | |
883 | #endif | |
884 | #if W_TYPE_SIZE > 16 | |
885 | x ^= x >> 16; | |
886 | #endif | |
887 | x ^= x >> 8; | |
888 | x ^= x >> 4; | |
889 | x &= 0xf; | |
890 | return (0x6996 >> x) & 1; | |
2928cd7a RH |
891 | } |
892 | #endif | |
893 | \f | |
894 | #ifdef L_paritydi2 | |
dabb3f04 | 895 | #undef int |
dabb3f04 | 896 | int |
8275b011 | 897 | __parityDI2 (UDWtype x) |
2928cd7a | 898 | { |
b982024e KG |
899 | const DWunion uu = {.ll = x}; |
900 | UWtype nx = uu.s.low ^ uu.s.high; | |
8275b011 RH |
901 | |
902 | #if W_TYPE_SIZE > 64 | |
903 | # error "fill out the table" | |
904 | #endif | |
905 | #if W_TYPE_SIZE > 32 | |
906 | nx ^= nx >> 32; | |
907 | #endif | |
908 | #if W_TYPE_SIZE > 16 | |
2928cd7a | 909 | nx ^= nx >> 16; |
8275b011 | 910 | #endif |
2928cd7a | 911 | nx ^= nx >> 8; |
53585c36 | 912 | nx ^= nx >> 4; |
0c9ed856 RH |
913 | nx &= 0xf; |
914 | return (0x6996 >> nx) & 1; | |
2928cd7a RH |
915 | } |
916 | #endif | |
d6eacd48 RH |
917 | |
918 | #ifdef L_udivmoddi4 | |
30b8f78b KV |
919 | #ifdef TARGET_HAS_NO_HW_DIVIDE |
920 | ||
921 | #if (defined (L_udivdi3) || defined (L_divdi3) || \ | |
18362447 UB |
922 | defined (L_umoddi3) || defined (L_moddi3) || \ |
923 | defined (L_divmoddi4)) | |
30b8f78b KV |
924 | static inline __attribute__ ((__always_inline__)) |
925 | #endif | |
926 | UDWtype | |
927 | __udivmoddi4 (UDWtype n, UDWtype d, UDWtype *rp) | |
928 | { | |
929 | UDWtype q = 0, r = n, y = d; | |
930 | UWtype lz1, lz2, i, k; | |
931 | ||
932 | /* Implements align divisor shift dividend method. This algorithm | |
933 | aligns the divisor under the dividend and then perform number of | |
934 | test-subtract iterations which shift the dividend left. Number of | |
935 | iterations is k + 1 where k is the number of bit positions the | |
ebc4cd54 | 936 | divisor must be shifted left to align it under the dividend. |
30b8f78b KV |
937 | quotient bits can be saved in the rightmost positions of the dividend |
938 | as it shifts left on each test-subtract iteration. */ | |
939 | ||
940 | if (y <= r) | |
941 | { | |
942 | lz1 = __builtin_clzll (d); | |
943 | lz2 = __builtin_clzll (n); | |
944 | ||
945 | k = lz1 - lz2; | |
946 | y = (y << k); | |
947 | ||
ebc4cd54 | 948 | /* Dividend can exceed 2 ^ (width - 1) - 1 but still be less than the |
30b8f78b KV |
949 | aligned divisor. Normal iteration can drops the high order bit |
950 | of the dividend. Therefore, first test-subtract iteration is a | |
951 | special case, saving its quotient bit in a separate location and | |
952 | not shifting the dividend. */ | |
953 | if (r >= y) | |
954 | { | |
955 | r = r - y; | |
956 | q = (1ULL << k); | |
957 | } | |
958 | ||
959 | if (k > 0) | |
960 | { | |
961 | y = y >> 1; | |
962 | ||
963 | /* k additional iterations where k regular test subtract shift | |
964 | dividend iterations are done. */ | |
965 | i = k; | |
966 | do | |
967 | { | |
968 | if (r >= y) | |
969 | r = ((r - y) << 1) + 1; | |
970 | else | |
971 | r = (r << 1); | |
972 | i = i - 1; | |
973 | } while (i != 0); | |
974 | ||
975 | /* First quotient bit is combined with the quotient bits resulting | |
976 | from the k regular iterations. */ | |
977 | q = q + r; | |
978 | r = r >> k; | |
979 | q = q - (r << k); | |
980 | } | |
981 | } | |
982 | ||
983 | if (rp) | |
984 | *rp = r; | |
985 | return q; | |
986 | } | |
987 | #else | |
203b91b9 | 988 | |
536bfcd0 | 989 | #if (defined (L_udivdi3) || defined (L_divdi3) || \ |
18362447 UB |
990 | defined (L_umoddi3) || defined (L_moddi3) || \ |
991 | defined (L_divmoddi4)) | |
1ab9ba62 | 992 | static inline __attribute__ ((__always_inline__)) |
536bfcd0 | 993 | #endif |
996ed075 JJ |
994 | UDWtype |
995 | __udivmoddi4 (UDWtype n, UDWtype d, UDWtype *rp) | |
203b91b9 | 996 | { |
b982024e KG |
997 | const DWunion nn = {.ll = n}; |
998 | const DWunion dd = {.ll = d}; | |
996ed075 JJ |
999 | DWunion rr; |
1000 | UWtype d0, d1, n0, n1, n2; | |
1001 | UWtype q0, q1; | |
1002 | UWtype b, bm; | |
203b91b9 | 1003 | |
203b91b9 RS |
1004 | d0 = dd.s.low; |
1005 | d1 = dd.s.high; | |
1006 | n0 = nn.s.low; | |
1007 | n1 = nn.s.high; | |
1008 | ||
1009 | #if !UDIV_NEEDS_NORMALIZATION | |
1010 | if (d1 == 0) | |
1011 | { | |
1012 | if (d0 > n1) | |
1013 | { | |
1014 | /* 0q = nn / 0D */ | |
1015 | ||
1016 | udiv_qrnnd (q0, n0, n1, n0, d0); | |
1017 | q1 = 0; | |
1018 | ||
1019 | /* Remainder in n0. */ | |
1020 | } | |
1021 | else | |
1022 | { | |
1023 | /* qq = NN / 0d */ | |
1024 | ||
1025 | if (d0 == 0) | |
1026 | d0 = 1 / d0; /* Divide intentionally by zero. */ | |
1027 | ||
1028 | udiv_qrnnd (q1, n1, 0, n1, d0); | |
1029 | udiv_qrnnd (q0, n0, n1, n0, d0); | |
1030 | ||
1031 | /* Remainder in n0. */ | |
1032 | } | |
1033 | ||
1034 | if (rp != 0) | |
1035 | { | |
1036 | rr.s.low = n0; | |
1037 | rr.s.high = 0; | |
1038 | *rp = rr.ll; | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | #else /* UDIV_NEEDS_NORMALIZATION */ | |
1043 | ||
1044 | if (d1 == 0) | |
1045 | { | |
1046 | if (d0 > n1) | |
1047 | { | |
1048 | /* 0q = nn / 0D */ | |
1049 | ||
1050 | count_leading_zeros (bm, d0); | |
1051 | ||
1052 | if (bm != 0) | |
1053 | { | |
1054 | /* Normalize, i.e. make the most significant bit of the | |
1055 | denominator set. */ | |
1056 | ||
1057 | d0 = d0 << bm; | |
996ed075 | 1058 | n1 = (n1 << bm) | (n0 >> (W_TYPE_SIZE - bm)); |
203b91b9 RS |
1059 | n0 = n0 << bm; |
1060 | } | |
1061 | ||
1062 | udiv_qrnnd (q0, n0, n1, n0, d0); | |
1063 | q1 = 0; | |
1064 | ||
1065 | /* Remainder in n0 >> bm. */ | |
1066 | } | |
1067 | else | |
1068 | { | |
1069 | /* qq = NN / 0d */ | |
1070 | ||
1071 | if (d0 == 0) | |
1072 | d0 = 1 / d0; /* Divide intentionally by zero. */ | |
1073 | ||
1074 | count_leading_zeros (bm, d0); | |
1075 | ||
1076 | if (bm == 0) | |
1077 | { | |
1078 | /* From (n1 >= d0) /\ (the most significant bit of d0 is set), | |
1079 | conclude (the most significant bit of n1 is set) /\ (the | |
1080 | leading quotient digit q1 = 1). | |
1081 | ||
1082 | This special case is necessary, not an optimization. | |
996ed075 | 1083 | (Shifts counts of W_TYPE_SIZE are undefined.) */ |
203b91b9 RS |
1084 | |
1085 | n1 -= d0; | |
1086 | q1 = 1; | |
1087 | } | |
1088 | else | |
1089 | { | |
1090 | /* Normalize. */ | |
1091 | ||
996ed075 | 1092 | b = W_TYPE_SIZE - bm; |
203b91b9 RS |
1093 | |
1094 | d0 = d0 << bm; | |
1095 | n2 = n1 >> b; | |
1096 | n1 = (n1 << bm) | (n0 >> b); | |
1097 | n0 = n0 << bm; | |
1098 | ||
1099 | udiv_qrnnd (q1, n1, n2, n1, d0); | |
1100 | } | |
1101 | ||
0f41302f | 1102 | /* n1 != d0... */ |
203b91b9 RS |
1103 | |
1104 | udiv_qrnnd (q0, n0, n1, n0, d0); | |
1105 | ||
1106 | /* Remainder in n0 >> bm. */ | |
1107 | } | |
1108 | ||
1109 | if (rp != 0) | |
1110 | { | |
1111 | rr.s.low = n0 >> bm; | |
1112 | rr.s.high = 0; | |
1113 | *rp = rr.ll; | |
1114 | } | |
1115 | } | |
1116 | #endif /* UDIV_NEEDS_NORMALIZATION */ | |
1117 | ||
1118 | else | |
1119 | { | |
1120 | if (d1 > n1) | |
1121 | { | |
1122 | /* 00 = nn / DD */ | |
1123 | ||
1124 | q0 = 0; | |
1125 | q1 = 0; | |
1126 | ||
1127 | /* Remainder in n1n0. */ | |
1128 | if (rp != 0) | |
1129 | { | |
1130 | rr.s.low = n0; | |
1131 | rr.s.high = n1; | |
1132 | *rp = rr.ll; | |
1133 | } | |
1134 | } | |
1135 | else | |
1136 | { | |
1137 | /* 0q = NN / dd */ | |
1138 | ||
1139 | count_leading_zeros (bm, d1); | |
1140 | if (bm == 0) | |
1141 | { | |
1142 | /* From (n1 >= d1) /\ (the most significant bit of d1 is set), | |
1143 | conclude (the most significant bit of n1 is set) /\ (the | |
1144 | quotient digit q0 = 0 or 1). | |
1145 | ||
1146 | This special case is necessary, not an optimization. */ | |
1147 | ||
1148 | /* The condition on the next line takes advantage of that | |
1149 | n1 >= d1 (true due to program flow). */ | |
1150 | if (n1 > d1 || n0 >= d0) | |
1151 | { | |
1152 | q0 = 1; | |
1153 | sub_ddmmss (n1, n0, n1, n0, d1, d0); | |
1154 | } | |
1155 | else | |
1156 | q0 = 0; | |
1157 | ||
1158 | q1 = 0; | |
1159 | ||
1160 | if (rp != 0) | |
1161 | { | |
1162 | rr.s.low = n0; | |
1163 | rr.s.high = n1; | |
1164 | *rp = rr.ll; | |
1165 | } | |
1166 | } | |
1167 | else | |
1168 | { | |
996ed075 | 1169 | UWtype m1, m0; |
203b91b9 RS |
1170 | /* Normalize. */ |
1171 | ||
996ed075 | 1172 | b = W_TYPE_SIZE - bm; |
203b91b9 RS |
1173 | |
1174 | d1 = (d1 << bm) | (d0 >> b); | |
1175 | d0 = d0 << bm; | |
1176 | n2 = n1 >> b; | |
1177 | n1 = (n1 << bm) | (n0 >> b); | |
1178 | n0 = n0 << bm; | |
1179 | ||
1180 | udiv_qrnnd (q0, n1, n2, n1, d1); | |
1181 | umul_ppmm (m1, m0, q0, d0); | |
1182 | ||
1183 | if (m1 > n1 || (m1 == n1 && m0 > n0)) | |
1184 | { | |
1185 | q0--; | |
1186 | sub_ddmmss (m1, m0, m1, m0, d1, d0); | |
1187 | } | |
1188 | ||
1189 | q1 = 0; | |
1190 | ||
1191 | /* Remainder in (n1n0 - m1m0) >> bm. */ | |
1192 | if (rp != 0) | |
1193 | { | |
1194 | sub_ddmmss (n1, n0, n1, n0, m1, m0); | |
1195 | rr.s.low = (n1 << b) | (n0 >> bm); | |
1196 | rr.s.high = n1 >> bm; | |
1197 | *rp = rr.ll; | |
1198 | } | |
1199 | } | |
1200 | } | |
1201 | } | |
1202 | ||
b982024e | 1203 | const DWunion ww = {{.low = q0, .high = q1}}; |
203b91b9 RS |
1204 | return ww.ll; |
1205 | } | |
1206 | #endif | |
30b8f78b | 1207 | #endif |
203b91b9 RS |
1208 | |
1209 | #ifdef L_divdi3 | |
996ed075 JJ |
1210 | DWtype |
1211 | __divdi3 (DWtype u, DWtype v) | |
203b91b9 | 1212 | { |
c7ff6e7a | 1213 | Wtype c = 0; |
b982024e KG |
1214 | DWunion uu = {.ll = u}; |
1215 | DWunion vv = {.ll = v}; | |
996ed075 | 1216 | DWtype w; |
203b91b9 | 1217 | |
203b91b9 RS |
1218 | if (uu.s.high < 0) |
1219 | c = ~c, | |
b68daef4 | 1220 | uu.ll = -uu.ll; |
203b91b9 RS |
1221 | if (vv.s.high < 0) |
1222 | c = ~c, | |
b68daef4 | 1223 | vv.ll = -vv.ll; |
203b91b9 | 1224 | |
996ed075 | 1225 | w = __udivmoddi4 (uu.ll, vv.ll, (UDWtype *) 0); |
203b91b9 | 1226 | if (c) |
b68daef4 | 1227 | w = -w; |
203b91b9 RS |
1228 | |
1229 | return w; | |
1230 | } | |
1231 | #endif | |
1232 | ||
1233 | #ifdef L_moddi3 | |
996ed075 JJ |
1234 | DWtype |
1235 | __moddi3 (DWtype u, DWtype v) | |
203b91b9 | 1236 | { |
c7ff6e7a | 1237 | Wtype c = 0; |
b982024e KG |
1238 | DWunion uu = {.ll = u}; |
1239 | DWunion vv = {.ll = v}; | |
996ed075 | 1240 | DWtype w; |
203b91b9 | 1241 | |
203b91b9 RS |
1242 | if (uu.s.high < 0) |
1243 | c = ~c, | |
b68daef4 | 1244 | uu.ll = -uu.ll; |
203b91b9 | 1245 | if (vv.s.high < 0) |
b68daef4 | 1246 | vv.ll = -vv.ll; |
203b91b9 | 1247 | |
9c859be1 | 1248 | (void) __udivmoddi4 (uu.ll, vv.ll, (UDWtype*)&w); |
203b91b9 | 1249 | if (c) |
b68daef4 | 1250 | w = -w; |
203b91b9 RS |
1251 | |
1252 | return w; | |
1253 | } | |
1254 | #endif | |
1255 | ||
18362447 UB |
1256 | #ifdef L_divmoddi4 |
1257 | DWtype | |
1258 | __divmoddi4 (DWtype u, DWtype v, DWtype *rp) | |
1259 | { | |
1260 | Wtype c1 = 0, c2 = 0; | |
1261 | DWunion uu = {.ll = u}; | |
1262 | DWunion vv = {.ll = v}; | |
1263 | DWtype w; | |
1264 | DWtype r; | |
1265 | ||
1266 | if (uu.s.high < 0) | |
1267 | c1 = ~c1, c2 = ~c2, | |
1268 | uu.ll = -uu.ll; | |
1269 | if (vv.s.high < 0) | |
1270 | c1 = ~c1, | |
1271 | vv.ll = -vv.ll; | |
1272 | ||
1273 | w = __udivmoddi4 (uu.ll, vv.ll, (UDWtype*)&r); | |
1274 | if (c1) | |
1275 | w = -w; | |
1276 | if (c2) | |
1277 | r = -r; | |
1278 | ||
1279 | *rp = r; | |
1280 | return w; | |
1281 | } | |
1282 | #endif | |
1283 | ||
203b91b9 | 1284 | #ifdef L_umoddi3 |
996ed075 JJ |
1285 | UDWtype |
1286 | __umoddi3 (UDWtype u, UDWtype v) | |
203b91b9 | 1287 | { |
996ed075 | 1288 | UDWtype w; |
203b91b9 RS |
1289 | |
1290 | (void) __udivmoddi4 (u, v, &w); | |
1291 | ||
1292 | return w; | |
1293 | } | |
1294 | #endif | |
1295 | ||
1296 | #ifdef L_udivdi3 | |
996ed075 JJ |
1297 | UDWtype |
1298 | __udivdi3 (UDWtype n, UDWtype d) | |
203b91b9 | 1299 | { |
996ed075 | 1300 | return __udivmoddi4 (n, d, (UDWtype *) 0); |
203b91b9 RS |
1301 | } |
1302 | #endif | |
1303 | \f | |
2ce182e2 JJ |
1304 | #if (defined(__BITINT_MAXWIDTH__) \ |
1305 | && (defined(L_mulbitint3) || defined(L_divmodbitint4))) | |
1306 | /* _BitInt support. */ | |
1307 | ||
1308 | /* If *P is zero or sign extended (the latter only for PREC < 0) from | |
1309 | some narrower _BitInt value, reduce precision. */ | |
1310 | ||
1311 | static inline __attribute__((__always_inline__)) SItype | |
1312 | bitint_reduce_prec (const UWtype **p, SItype prec) | |
1313 | { | |
1314 | UWtype mslimb; | |
1315 | SItype i; | |
1316 | if (prec < 0) | |
1317 | { | |
1318 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1319 | i = 0; | |
1320 | #else | |
1321 | i = ((USItype) -1 - prec) / W_TYPE_SIZE; | |
1322 | #endif | |
1323 | mslimb = (*p)[i]; | |
1324 | if (mslimb & ((UWtype) 1 << (((USItype) -1 - prec) % W_TYPE_SIZE))) | |
1325 | { | |
1326 | SItype n = ((USItype) -prec) % W_TYPE_SIZE; | |
1327 | if (n) | |
1328 | { | |
1329 | mslimb |= ((UWtype) -1 << (((USItype) -1 - prec) % W_TYPE_SIZE)); | |
1330 | if (mslimb == (UWtype) -1) | |
1331 | { | |
1332 | prec += n; | |
1333 | if (prec >= -1) | |
1334 | return -2; | |
1335 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1336 | ++p; | |
1337 | #else | |
1338 | --i; | |
1339 | #endif | |
1340 | mslimb = (*p)[i]; | |
1341 | n = 0; | |
1342 | } | |
1343 | } | |
1344 | while (mslimb == (UWtype) -1) | |
1345 | { | |
1346 | prec += W_TYPE_SIZE; | |
1347 | if (prec >= -1) | |
1348 | return -2; | |
1349 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1350 | ++p; | |
1351 | #else | |
1352 | --i; | |
1353 | #endif | |
1354 | mslimb = (*p)[i]; | |
1355 | } | |
1356 | if (n == 0) | |
1357 | { | |
1358 | if ((Wtype) mslimb >= 0) | |
1359 | { | |
1360 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1361 | --p; | |
1362 | #endif | |
1363 | return prec - 1; | |
1364 | } | |
1365 | } | |
1366 | return prec; | |
1367 | } | |
1368 | else | |
1369 | prec = -prec; | |
1370 | } | |
1371 | else | |
1372 | { | |
1373 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1374 | i = 0; | |
1375 | #else | |
1376 | i = ((USItype) prec - 1) / W_TYPE_SIZE; | |
1377 | #endif | |
1378 | mslimb = (*p)[i]; | |
1379 | } | |
1380 | SItype n = ((USItype) prec) % W_TYPE_SIZE; | |
1381 | if (n) | |
1382 | { | |
1383 | mslimb &= ((UWtype) 1 << (((USItype) prec) % W_TYPE_SIZE)) - 1; | |
1384 | if (mslimb == 0) | |
1385 | { | |
1386 | prec -= n; | |
1387 | if (prec == 0) | |
1388 | return 1; | |
1389 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1390 | ++p; | |
1391 | #else | |
1392 | --i; | |
1393 | #endif | |
1394 | mslimb = (*p)[i]; | |
1395 | } | |
1396 | } | |
1397 | while (mslimb == 0) | |
1398 | { | |
1399 | prec -= W_TYPE_SIZE; | |
1400 | if (prec == 0) | |
1401 | return 1; | |
1402 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1403 | ++p; | |
1404 | #else | |
1405 | --i; | |
1406 | #endif | |
1407 | mslimb = (*p)[i]; | |
1408 | } | |
1409 | return prec; | |
1410 | } | |
1411 | ||
1412 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1413 | # define BITINT_INC -1 | |
1414 | # define BITINT_END(be, le) (be) | |
1415 | #else | |
1416 | # define BITINT_INC 1 | |
1417 | # define BITINT_END(be, le) (le) | |
1418 | #endif | |
1419 | ||
1420 | #ifdef L_mulbitint3 | |
1421 | /* D = S * L. */ | |
1422 | ||
1423 | static UWtype | |
1424 | bitint_mul_1 (UWtype *d, const UWtype *s, UWtype l, SItype n) | |
1425 | { | |
1426 | UWtype sv, hi, lo, c = 0; | |
1427 | do | |
1428 | { | |
1429 | sv = *s; | |
1430 | s += BITINT_INC; | |
1431 | umul_ppmm (hi, lo, sv, l); | |
1432 | c = __builtin_add_overflow (lo, c, &lo) + hi; | |
1433 | *d = lo; | |
1434 | d += BITINT_INC; | |
1435 | } | |
1436 | while (--n); | |
1437 | return c; | |
1438 | } | |
1439 | ||
1440 | /* D += S * L. */ | |
1441 | ||
1442 | static UWtype | |
1443 | bitint_addmul_1 (UWtype *d, const UWtype *s, UWtype l, SItype n) | |
1444 | { | |
1445 | UWtype sv, hi, lo, c = 0; | |
1446 | do | |
1447 | { | |
1448 | sv = *s; | |
1449 | s += BITINT_INC; | |
1450 | umul_ppmm (hi, lo, sv, l); | |
1451 | hi += __builtin_add_overflow (lo, *d, &lo); | |
1452 | c = __builtin_add_overflow (lo, c, &lo) + hi; | |
1453 | *d = lo; | |
1454 | d += BITINT_INC; | |
1455 | } | |
1456 | while (--n); | |
1457 | return c; | |
1458 | } | |
1459 | ||
1460 | /* If XPREC is positive, it is precision in bits | |
1461 | of an unsigned _BitInt operand (which has XPREC/W_TYPE_SIZE | |
1462 | full limbs and if Xprec%W_TYPE_SIZE one partial limb. | |
1463 | If Xprec is negative, -XPREC is precision in bits | |
1464 | of a signed _BitInt operand. RETPREC should be always | |
1465 | positive. */ | |
1466 | ||
1467 | void | |
1468 | __mulbitint3 (UWtype *ret, SItype retprec, | |
1469 | const UWtype *u, SItype uprec, | |
1470 | const UWtype *v, SItype vprec) | |
1471 | { | |
1472 | uprec = bitint_reduce_prec (&u, uprec); | |
1473 | vprec = bitint_reduce_prec (&v, vprec); | |
1474 | USItype auprec = uprec < 0 ? -uprec : uprec; | |
1475 | USItype avprec = vprec < 0 ? -vprec : vprec; | |
1476 | ||
1477 | /* Prefer non-negative U. | |
1478 | Otherwise make sure V doesn't have higher precision than U. */ | |
1479 | if ((uprec < 0 && vprec >= 0) | |
1480 | || (avprec > auprec && !(uprec >= 0 && vprec < 0))) | |
1481 | { | |
1482 | SItype p; | |
1483 | const UWtype *t; | |
1484 | p = uprec; uprec = vprec; vprec = p; | |
1485 | p = auprec; auprec = avprec; avprec = p; | |
1486 | t = u; u = v; v = t; | |
1487 | } | |
1488 | ||
1489 | USItype un = auprec / W_TYPE_SIZE; | |
1490 | USItype un2 = (auprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1491 | USItype vn = avprec / W_TYPE_SIZE; | |
1492 | USItype vn2 = (avprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1493 | USItype retn = ((USItype) retprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1494 | USItype retidx, uidx, vidx; | |
1495 | UWtype vv; | |
1496 | /* Indexes of least significant limb. */ | |
1497 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1498 | retidx = retn - 1; | |
1499 | uidx = un2 - 1; | |
1500 | vidx = vn2 - 1; | |
1501 | #else | |
1502 | retidx = 0; | |
1503 | uidx = 0; | |
1504 | vidx = 0; | |
1505 | #endif | |
1506 | if (__builtin_expect (auprec <= W_TYPE_SIZE, 0) && vprec < 0) | |
1507 | { | |
1508 | UWtype uu = u[uidx]; | |
1509 | if (__builtin_expect (auprec < W_TYPE_SIZE, 0)) | |
1510 | uu &= ((UWtype) 1 << (auprec % W_TYPE_SIZE)) - 1; | |
1511 | if (uu == 0) | |
1512 | { | |
1513 | /* 0 * negative would be otherwise mishandled below, so | |
1514 | handle it specially. */ | |
1515 | __builtin_memset (ret, 0, retn * sizeof (UWtype)); | |
1516 | return; | |
1517 | } | |
1518 | } | |
1519 | vv = v[vidx]; | |
1520 | if (__builtin_expect (avprec < W_TYPE_SIZE, 0)) | |
1521 | { | |
1522 | if (vprec > 0) | |
1523 | vv &= ((UWtype) 1 << (avprec % W_TYPE_SIZE)) - 1; | |
1524 | else | |
1525 | vv |= (UWtype) -1 << (avprec % W_TYPE_SIZE); | |
1526 | } | |
1527 | ||
1528 | USItype n = un > retn ? retn : un; | |
1529 | USItype n2 = n; | |
1530 | USItype retidx2 = retidx + n * BITINT_INC; | |
1531 | UWtype c = 0, uv = 0; | |
1532 | if (n) | |
1533 | c = bitint_mul_1 (ret + retidx, u + uidx, vv, n); | |
1534 | if (retn > un && un2 != un) | |
1535 | { | |
1536 | UWtype hi, lo; | |
1537 | uv = u[uidx + n * BITINT_INC]; | |
1538 | if (uprec > 0) | |
1539 | uv &= ((UWtype) 1 << (auprec % W_TYPE_SIZE)) - 1; | |
1540 | else | |
1541 | uv |= (UWtype) -1 << (auprec % W_TYPE_SIZE); | |
1542 | umul_ppmm (hi, lo, uv, vv); | |
1543 | c = __builtin_add_overflow (lo, c, &lo) + hi; | |
1544 | ret[retidx2] = lo; | |
1545 | retidx2 += BITINT_INC; | |
1546 | ++n2; | |
1547 | } | |
1548 | if (retn > un2) | |
1549 | { | |
1550 | if (uprec < 0) | |
1551 | { | |
1552 | while (n2 < retn) | |
1553 | { | |
1554 | if (n2 >= un2 + vn2) | |
1555 | break; | |
1556 | UWtype hi, lo; | |
1557 | umul_ppmm (hi, lo, (UWtype) -1, vv); | |
1558 | c = __builtin_add_overflow (lo, c, &lo) + hi; | |
1559 | ret[retidx2] = lo; | |
1560 | retidx2 += BITINT_INC; | |
1561 | ++n2; | |
1562 | } | |
1563 | } | |
1564 | else | |
1565 | { | |
1566 | ret[retidx2] = c; | |
1567 | retidx2 += BITINT_INC; | |
1568 | ++n2; | |
1569 | } | |
1570 | /* If RET has more limbs than U after precision reduction, | |
1571 | fill in the remaining limbs. */ | |
1572 | while (n2 < retn) | |
1573 | { | |
1574 | if (n2 < un2 + vn2 || (uprec ^ vprec) >= 0) | |
1575 | c = 0; | |
1576 | else | |
1577 | c = (UWtype) -1; | |
1578 | ret[retidx2] = c; | |
1579 | retidx2 += BITINT_INC; | |
1580 | ++n2; | |
1581 | } | |
1582 | } | |
1583 | /* N is now number of possibly non-zero limbs in RET (ignoring | |
1584 | limbs above UN2 + VN2 which if any have been finalized already). */ | |
1585 | USItype end = vprec < 0 ? un2 + vn2 : vn2; | |
1586 | if (retn > un2 + vn2) retn = un2 + vn2; | |
1587 | if (end > retn) end = retn; | |
1588 | for (USItype m = 1; m < end; ++m) | |
1589 | { | |
1590 | retidx += BITINT_INC; | |
1591 | vidx += BITINT_INC; | |
1592 | if (m < vn2) | |
1593 | { | |
1594 | vv = v[vidx]; | |
1595 | if (__builtin_expect (m == vn, 0)) | |
1596 | { | |
1597 | if (vprec > 0) | |
1598 | vv &= ((UWtype) 1 << (avprec % W_TYPE_SIZE)) - 1; | |
1599 | else | |
1600 | vv |= (UWtype) -1 << (avprec % W_TYPE_SIZE); | |
1601 | } | |
1602 | } | |
1603 | else | |
1604 | vv = (UWtype) -1; | |
1605 | if (m + n > retn) | |
1606 | n = retn - m; | |
1607 | c = 0; | |
1608 | if (n) | |
1609 | c = bitint_addmul_1 (ret + retidx, u + uidx, vv, n); | |
1610 | n2 = m + n; | |
1611 | retidx2 = retidx + n * BITINT_INC; | |
1612 | if (n2 < retn && un2 != un) | |
1613 | { | |
1614 | UWtype hi, lo; | |
1615 | umul_ppmm (hi, lo, uv, vv); | |
1616 | hi += __builtin_add_overflow (lo, ret[retidx2], &lo); | |
1617 | c = __builtin_add_overflow (lo, c, &lo) + hi; | |
1618 | ret[retidx2] = lo; | |
1619 | retidx2 += BITINT_INC; | |
1620 | ++n2; | |
1621 | } | |
1622 | if (uprec < 0) | |
1623 | while (n2 < retn) | |
1624 | { | |
1625 | UWtype hi, lo; | |
1626 | umul_ppmm (hi, lo, (UWtype) -1, vv); | |
1627 | hi += __builtin_add_overflow (lo, ret[retidx2], &lo); | |
1628 | c = __builtin_add_overflow (lo, c, &lo) + hi; | |
1629 | ret[retidx2] = lo; | |
1630 | retidx2 += BITINT_INC; | |
1631 | ++n2; | |
1632 | } | |
1633 | else if (n2 < retn) | |
1634 | { | |
1635 | ret[retidx2] = c; | |
1636 | retidx2 += BITINT_INC; | |
1637 | } | |
1638 | } | |
1639 | } | |
1640 | #endif | |
1641 | ||
1642 | #ifdef L_divmodbitint4 | |
1643 | static void | |
1644 | bitint_negate (UWtype *d, const UWtype *s, SItype n) | |
1645 | { | |
1646 | UWtype c = 1; | |
1647 | do | |
1648 | { | |
1649 | UWtype sv = *s, lo; | |
1650 | s += BITINT_INC; | |
1651 | c = __builtin_add_overflow (~sv, c, &lo); | |
1652 | *d = lo; | |
1653 | d += BITINT_INC; | |
1654 | } | |
1655 | while (--n); | |
1656 | } | |
1657 | ||
1658 | /* D -= S * L. */ | |
1659 | ||
1660 | static UWtype | |
1661 | bitint_submul_1 (UWtype *d, const UWtype *s, UWtype l, SItype n) | |
1662 | { | |
1663 | UWtype sv, hi, lo, c = 0; | |
1664 | do | |
1665 | { | |
1666 | sv = *s; | |
1667 | s += BITINT_INC; | |
1668 | umul_ppmm (hi, lo, sv, l); | |
1669 | hi += __builtin_sub_overflow (*d, lo, &lo); | |
1670 | c = __builtin_sub_overflow (lo, c, &lo) + hi; | |
1671 | *d = lo; | |
1672 | d += BITINT_INC; | |
1673 | } | |
1674 | while (--n); | |
1675 | return c; | |
1676 | } | |
1677 | ||
1678 | /* If XPREC is positive, it is precision in bits | |
1679 | of an unsigned _BitInt operand (which has XPREC/W_TYPE_SIZE | |
1680 | full limbs and if Xprec%W_TYPE_SIZE one partial limb. | |
1681 | If Xprec is negative, -XPREC is precision in bits | |
1682 | of a signed _BitInt operand. QPREC and RPREC should be | |
1683 | always non-negative. If either Q or R is NULL (at least | |
1684 | one should be non-NULL), then corresponding QPREC or RPREC | |
1685 | should be 0. */ | |
1686 | ||
1687 | void | |
1688 | __divmodbitint4 (UWtype *q, SItype qprec, | |
1689 | UWtype *r, SItype rprec, | |
1690 | const UWtype *u, SItype uprec, | |
1691 | const UWtype *v, SItype vprec) | |
1692 | { | |
1693 | uprec = bitint_reduce_prec (&u, uprec); | |
1694 | vprec = bitint_reduce_prec (&v, vprec); | |
1695 | USItype auprec = uprec < 0 ? -uprec : uprec; | |
1696 | USItype avprec = vprec < 0 ? -vprec : vprec; | |
1697 | USItype un = (auprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1698 | USItype vn = (avprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1699 | USItype qn = ((USItype) qprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1700 | USItype rn = ((USItype) rprec + W_TYPE_SIZE - 1) / W_TYPE_SIZE; | |
1701 | USItype up = auprec % W_TYPE_SIZE; | |
1702 | USItype vp = avprec % W_TYPE_SIZE; | |
1703 | if (__builtin_expect (un < vn, 0)) | |
1704 | { | |
1705 | /* If abs(v) > abs(u), then q is 0 and r is u. */ | |
1706 | if (q) | |
1707 | __builtin_memset (q, 0, qn * sizeof (UWtype)); | |
1708 | if (r == NULL) | |
1709 | return; | |
1710 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1711 | r += rn - 1; | |
1712 | u += un - 1; | |
1713 | #endif | |
1714 | if (up) | |
1715 | --un; | |
1716 | if (rn < un) | |
1717 | un = rn; | |
1718 | for (rn -= un; un; --un) | |
1719 | { | |
1720 | *r = *u; | |
1721 | r += BITINT_INC; | |
1722 | u += BITINT_INC; | |
1723 | } | |
1724 | if (!rn) | |
1725 | return; | |
1726 | if (up) | |
1727 | { | |
1728 | if (uprec > 0) | |
1729 | *r = *u & (((UWtype) 1 << up) - 1); | |
1730 | else | |
1731 | *r = *u | ((UWtype) -1 << up); | |
1732 | r += BITINT_INC; | |
1733 | if (!--rn) | |
1734 | return; | |
1735 | } | |
1736 | UWtype c = uprec < 0 ? (UWtype) -1 : (UWtype) 0; | |
1737 | for (; rn; --rn) | |
1738 | { | |
1739 | *r = c; | |
1740 | r += BITINT_INC; | |
1741 | } | |
1742 | return; | |
1743 | } | |
1744 | USItype qn2 = un - vn + 1; | |
1745 | if (qn >= qn2) | |
1746 | qn2 = 0; | |
1747 | USItype sz = un + 1 + vn + qn2; | |
1748 | UWtype *buf = __builtin_alloca (sz * sizeof (UWtype)); | |
1749 | USItype uidx, vidx; | |
1750 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1751 | uidx = un - 1; | |
1752 | vidx = vn - 1; | |
1753 | #else | |
1754 | uidx = 0; | |
1755 | vidx = 0; | |
1756 | #endif | |
1757 | if (uprec < 0) | |
1758 | bitint_negate (buf + BITINT_END (uidx + 1, 0), u + uidx, un); | |
1759 | else | |
1760 | __builtin_memcpy (buf + BITINT_END (1, 0), u, un * sizeof (UWtype)); | |
1761 | if (up) | |
1762 | buf[BITINT_END (1, un - 1)] &= (((UWtype) 1 << up) - 1); | |
1763 | if (vprec < 0) | |
1764 | bitint_negate (buf + un + 1 + vidx, v + vidx, vn); | |
1765 | else | |
1766 | __builtin_memcpy (buf + un + 1, v, vn * sizeof (UWtype)); | |
1767 | if (vp) | |
1768 | buf[un + 1 + BITINT_END (0, vn - 1)] &= (((UWtype) 1 << vp) - 1); | |
1769 | UWtype *u2 = buf; | |
1770 | UWtype *v2 = u2 + un + 1; | |
1771 | UWtype *q2 = v2 + vn; | |
1772 | if (!qn2) | |
1773 | q2 = q + BITINT_END (qn - (un - vn + 1), 0); | |
1774 | ||
1775 | /* Knuth's algorithm. See also ../gcc/wide-int.cc (divmod_internal_2). */ | |
1776 | ||
1777 | #ifndef UDIV_NEEDS_NORMALIZATION | |
1778 | /* Handle single limb divisor first. */ | |
1779 | if (vn == 1) | |
1780 | { | |
1781 | UWtype vv = v2[0]; | |
1782 | if (vv == 0) | |
1783 | vv = 1 / vv; /* Divide intentionally by zero. */ | |
1784 | UWtype k = 0; | |
1785 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1786 | for (SItype i = 0; i <= un - 1; ++i) | |
1787 | #else | |
1788 | for (SItype i = un - 1; i >= 0; --i) | |
1789 | #endif | |
1790 | udiv_qrnnd (q2[i], k, k, u2[BITINT_END (i + 1, i)], vv); | |
1791 | if (r != NULL) | |
1792 | r[BITINT_END (rn - 1, 0)] = k; | |
1793 | } | |
1794 | else | |
1795 | #endif | |
1796 | { | |
1797 | SItype s; | |
1798 | #ifdef UDIV_NEEDS_NORMALIZATION | |
1799 | if (vn == 1 && v2[0] == 0) | |
1800 | s = 0; | |
1801 | else | |
1802 | #endif | |
1803 | if (sizeof (0U) == sizeof (UWtype)) | |
1804 | s = __builtin_clz (v2[BITINT_END (0, vn - 1)]); | |
1805 | else if (sizeof (0UL) == sizeof (UWtype)) | |
1806 | s = __builtin_clzl (v2[BITINT_END (0, vn - 1)]); | |
1807 | else | |
1808 | s = __builtin_clzll (v2[BITINT_END (0, vn - 1)]); | |
1809 | if (s) | |
1810 | { | |
1811 | /* Normalize by shifting v2 left so that it has msb set. */ | |
1812 | const SItype n = sizeof (UWtype) * __CHAR_BIT__; | |
1813 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1814 | for (SItype i = 0; i < vn - 1; ++i) | |
1815 | #else | |
1816 | for (SItype i = vn - 1; i > 0; --i) | |
1817 | #endif | |
1818 | v2[i] = (v2[i] << s) | (v2[i - BITINT_INC] >> (n - s)); | |
1819 | v2[vidx] = v2[vidx] << s; | |
1820 | /* And shift u2 left by the same amount. */ | |
1821 | u2[BITINT_END (0, un)] = u2[BITINT_END (1, un - 1)] >> (n - s); | |
1822 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1823 | for (SItype i = 1; i < un; ++i) | |
1824 | #else | |
1825 | for (SItype i = un - 1; i > 0; --i) | |
1826 | #endif | |
1827 | u2[i] = (u2[i] << s) | (u2[i - BITINT_INC] >> (n - s)); | |
1828 | u2[BITINT_END (un, 0)] = u2[BITINT_END (un, 0)] << s; | |
1829 | } | |
1830 | else | |
1831 | u2[BITINT_END (0, un)] = 0; | |
1832 | #ifdef UDIV_NEEDS_NORMALIZATION | |
1833 | /* Handle single limb divisor first. */ | |
1834 | if (vn == 1) | |
1835 | { | |
1836 | UWtype vv = v2[0]; | |
1837 | if (vv == 0) | |
1838 | vv = 1 / vv; /* Divide intentionally by zero. */ | |
1839 | UWtype k = u2[BITINT_END (0, un)]; | |
1840 | #if __LIBGCC_BITINT_ORDER__ == __ORDER_BIG_ENDIAN__ | |
1841 | for (SItype i = 0; i <= un - 1; ++i) | |
1842 | #else | |
1843 | for (SItype i = un - 1; i >= 0; --i) | |
1844 | #endif | |
1845 | udiv_qrnnd (q2[i], k, k, u2[BITINT_END (i + 1, i)], vv); | |
1846 | if (r != NULL) | |
1847 | r[BITINT_END (rn - 1, 0)] = k >> s; | |
1848 | } | |
1849 | else | |
1850 | #endif | |
1851 | { | |
1852 | UWtype vv1 = v2[BITINT_END (0, vn - 1)]; | |
1853 | UWtype vv0 = v2[BITINT_END (1, vn - 2)]; | |
1854 | /* Main loop. */ | |
1855 | for (SItype j = un - vn; j >= 0; --j) | |
1856 | { | |
1857 | /* Compute estimate in qhat. */ | |
1858 | UWtype uv1 = u2[BITINT_END (un - j - vn, j + vn)]; | |
1859 | UWtype uv0 = u2[BITINT_END (un - j - vn + 1, j + vn - 1)]; | |
1860 | UWtype qhat, rhat, hi, lo, c; | |
1861 | if (uv1 >= vv1) | |
1862 | { | |
1863 | /* udiv_qrnnd doesn't support quotients which don't | |
1864 | fit into UWtype, so subtract from uv1:uv0 vv1 | |
1865 | first. */ | |
1866 | uv1 -= vv1 + __builtin_sub_overflow (uv0, vv1, &uv0); | |
1867 | udiv_qrnnd (qhat, rhat, uv1, uv0, vv1); | |
1868 | if (!__builtin_add_overflow (rhat, vv1, &rhat)) | |
1869 | goto again; | |
1870 | } | |
1871 | else | |
1872 | { | |
1873 | udiv_qrnnd (qhat, rhat, uv1, uv0, vv1); | |
1874 | again: | |
1875 | umul_ppmm (hi, lo, qhat, vv0); | |
1876 | if (hi > rhat | |
1877 | || (hi == rhat | |
1878 | && lo > u2[BITINT_END (un - j - vn + 2, | |
1879 | j + vn - 2)])) | |
1880 | { | |
1881 | --qhat; | |
1882 | if (!__builtin_add_overflow (rhat, vv1, &rhat)) | |
1883 | goto again; | |
1884 | } | |
1885 | } | |
1886 | ||
1887 | c = bitint_submul_1 (u2 + BITINT_END (un - j, j), | |
1888 | v2 + BITINT_END (vn - 1, 0), qhat, vn); | |
1889 | u2[BITINT_END (un - j - vn, j + vn)] -= c; | |
1890 | /* If we've subtracted too much, decrease qhat and | |
1891 | and add back. */ | |
1892 | if ((Wtype) u2[BITINT_END (un - j - vn, j + vn)] < 0) | |
1893 | { | |
1894 | --qhat; | |
1895 | c = 0; | |
1896 | for (USItype i = 0; i < vn; ++i) | |
1897 | { | |
1898 | UWtype s = v2[BITINT_END (vn - 1 - i, i)]; | |
1899 | UWtype d = u2[BITINT_END (un - i - j, i + j)]; | |
1900 | UWtype c1 = __builtin_add_overflow (d, s, &d); | |
1901 | UWtype c2 = __builtin_add_overflow (d, c, &d); | |
1902 | c = c1 + c2; | |
1903 | u2[BITINT_END (un - i - j, i + j)] = d; | |
1904 | } | |
1905 | u2[BITINT_END (un - j - vn, j + vn)] += c; | |
1906 | } | |
1907 | q2[BITINT_END (un - vn - j, j)] = qhat; | |
1908 | } | |
1909 | if (r != NULL) | |
1910 | { | |
1911 | if (s) | |
1912 | { | |
1913 | const SItype n = sizeof (UWtype) * __CHAR_BIT__; | |
1914 | /* Unnormalize remainder. */ | |
1915 | USItype i; | |
1916 | for (i = 0; i < vn && i < rn; ++i) | |
1917 | r[BITINT_END (rn - 1 - i, i)] | |
1918 | = ((u2[BITINT_END (un - i, i)] >> s) | |
1919 | | (u2[BITINT_END (un - i - 1, i + 1)] << (n - s))); | |
1920 | if (i < rn) | |
1921 | r[BITINT_END (rn - vn, vn - 1)] | |
1922 | = u2[BITINT_END (un - vn + 1, vn - 1)] >> s; | |
1923 | } | |
1924 | else if (rn > vn) | |
1925 | __builtin_memcpy (&r[BITINT_END (rn - vn, 0)], | |
1926 | &u2[BITINT_END (un + 1 - vn, 0)], | |
1927 | vn * sizeof (UWtype)); | |
1928 | else | |
1929 | __builtin_memcpy (&r[0], &u2[BITINT_END (un + 1 - rn, 0)], | |
1930 | rn * sizeof (UWtype)); | |
1931 | } | |
1932 | } | |
1933 | } | |
1934 | if (q != NULL) | |
1935 | { | |
1936 | if ((uprec < 0) ^ (vprec < 0)) | |
1937 | { | |
1938 | /* Negative quotient. */ | |
1939 | USItype n; | |
1940 | if (un - vn + 1 > qn) | |
1941 | n = qn; | |
1942 | else | |
1943 | n = un - vn + 1; | |
1944 | bitint_negate (q + BITINT_END (qn - 1, 0), | |
1945 | q2 + BITINT_END (un - vn, 0), n); | |
1946 | if (qn > n) | |
1947 | __builtin_memset (q + BITINT_END (0, n), -1, | |
1948 | (qn - n) * sizeof (UWtype)); | |
1949 | } | |
1950 | else | |
1951 | { | |
1952 | /* Positive quotient. */ | |
1953 | if (qn2) | |
1954 | __builtin_memcpy (q, q2 + BITINT_END (un - vn + 1 - qn, 0), | |
1955 | qn * sizeof (UWtype)); | |
1956 | else if (qn > un - vn + 1) | |
1957 | __builtin_memset (q + BITINT_END (0, un - vn + 1), 0, | |
1958 | (qn - (un - vn + 1)) * sizeof (UWtype)); | |
1959 | } | |
1960 | } | |
1961 | if (r != NULL) | |
1962 | { | |
1963 | if (uprec < 0) | |
1964 | { | |
1965 | /* Negative remainder. */ | |
1966 | bitint_negate (r + BITINT_END (rn - 1, 0), | |
1967 | r + BITINT_END (rn - 1, 0), | |
1968 | rn > vn ? vn : rn); | |
1969 | if (rn > vn) | |
1970 | __builtin_memset (r + BITINT_END (0, vn), -1, | |
1971 | (rn - vn) * sizeof (UWtype)); | |
1972 | } | |
1973 | else | |
1974 | { | |
1975 | /* Positive remainder. */ | |
1976 | if (rn > vn) | |
1977 | __builtin_memset (r + BITINT_END (0, vn), 0, | |
1978 | (rn - vn) * sizeof (UWtype)); | |
1979 | } | |
1980 | } | |
1981 | } | |
1982 | #endif | |
1983 | #endif | |
1984 | \f | |
203b91b9 | 1985 | #ifdef L_cmpdi2 |
c7ff6e7a | 1986 | cmp_return_type |
996ed075 | 1987 | __cmpdi2 (DWtype a, DWtype b) |
203b91b9 | 1988 | { |
ebc4cd54 | 1989 | return (a > b) - (a < b) + 1; |
203b91b9 RS |
1990 | } |
1991 | #endif | |
1992 | ||
1993 | #ifdef L_ucmpdi2 | |
c7ff6e7a | 1994 | cmp_return_type |
ebc4cd54 | 1995 | __ucmpdi2 (UDWtype a, UDWtype b) |
203b91b9 | 1996 | { |
ebc4cd54 | 1997 | return (a > b) - (a < b) + 1; |
203b91b9 RS |
1998 | } |
1999 | #endif | |
2000 | \f | |
4e9db8b2 | 2001 | #if defined(L_fixunstfdi) && LIBGCC2_HAS_TF_MODE |
f139f5fa | 2002 | UDWtype |
6da9c622 | 2003 | __fixunstfDI (TFtype a) |
ab495388 | 2004 | { |
ab495388 RS |
2005 | if (a < 0) |
2006 | return 0; | |
2007 | ||
2008 | /* Compute high word of result, as a flonum. */ | |
4f2e0d5e | 2009 | const TFtype b = (a / Wtype_MAXp1_F); |
996ed075 | 2010 | /* Convert that to fixed (but not to DWtype!), |
ab495388 | 2011 | and shift it into the high word. */ |
b982024e | 2012 | UDWtype v = (UWtype) b; |
4f2e0d5e | 2013 | v <<= W_TYPE_SIZE; |
ab495388 RS |
2014 | /* Remove high part from the TFtype, leaving the low part as flonum. */ |
2015 | a -= (TFtype)v; | |
996ed075 | 2016 | /* Convert that to fixed (but not to DWtype!) and add it in. |
ab495388 RS |
2017 | Sometimes A comes out negative. This is significant, since |
2018 | A has more bits than a long int does. */ | |
2019 | if (a < 0) | |
996ed075 | 2020 | v -= (UWtype) (- a); |
ab495388 | 2021 | else |
996ed075 | 2022 | v += (UWtype) a; |
ab495388 RS |
2023 | return v; |
2024 | } | |
2025 | #endif | |
2026 | ||
4e9db8b2 | 2027 | #if defined(L_fixtfdi) && LIBGCC2_HAS_TF_MODE |
996ed075 | 2028 | DWtype |
37ef1054 | 2029 | __fixtfdi (TFtype a) |
ab495388 RS |
2030 | { |
2031 | if (a < 0) | |
6da9c622 RK |
2032 | return - __fixunstfDI (-a); |
2033 | return __fixunstfDI (a); | |
ab495388 RS |
2034 | } |
2035 | #endif | |
2036 | ||
4e9db8b2 | 2037 | #if defined(L_fixunsxfdi) && LIBGCC2_HAS_XF_MODE |
f139f5fa | 2038 | UDWtype |
6da9c622 | 2039 | __fixunsxfDI (XFtype a) |
e0799b34 | 2040 | { |
e0799b34 RS |
2041 | if (a < 0) |
2042 | return 0; | |
2043 | ||
2044 | /* Compute high word of result, as a flonum. */ | |
4f2e0d5e | 2045 | const XFtype b = (a / Wtype_MAXp1_F); |
996ed075 | 2046 | /* Convert that to fixed (but not to DWtype!), |
e0799b34 | 2047 | and shift it into the high word. */ |
b982024e | 2048 | UDWtype v = (UWtype) b; |
4f2e0d5e | 2049 | v <<= W_TYPE_SIZE; |
e0799b34 RS |
2050 | /* Remove high part from the XFtype, leaving the low part as flonum. */ |
2051 | a -= (XFtype)v; | |
996ed075 | 2052 | /* Convert that to fixed (but not to DWtype!) and add it in. |
e0799b34 RS |
2053 | Sometimes A comes out negative. This is significant, since |
2054 | A has more bits than a long int does. */ | |
2055 | if (a < 0) | |
996ed075 | 2056 | v -= (UWtype) (- a); |
e0799b34 | 2057 | else |
996ed075 | 2058 | v += (UWtype) a; |
e0799b34 RS |
2059 | return v; |
2060 | } | |
2061 | #endif | |
2062 | ||
4e9db8b2 | 2063 | #if defined(L_fixxfdi) && LIBGCC2_HAS_XF_MODE |
996ed075 | 2064 | DWtype |
37ef1054 | 2065 | __fixxfdi (XFtype a) |
e0799b34 RS |
2066 | { |
2067 | if (a < 0) | |
6da9c622 RK |
2068 | return - __fixunsxfDI (-a); |
2069 | return __fixunsxfDI (a); | |
e0799b34 RS |
2070 | } |
2071 | #endif | |
2072 | ||
4e9db8b2 | 2073 | #if defined(L_fixunsdfdi) && LIBGCC2_HAS_DF_MODE |
f139f5fa | 2074 | UDWtype |
6da9c622 | 2075 | __fixunsdfDI (DFtype a) |
203b91b9 | 2076 | { |
4977bab6 ZW |
2077 | /* Get high part of result. The division here will just moves the radix |
2078 | point and will not cause any rounding. Then the conversion to integral | |
2079 | type chops result as desired. */ | |
4f2e0d5e | 2080 | const UWtype hi = a / Wtype_MAXp1_F; |
203b91b9 | 2081 | |
4977bab6 ZW |
2082 | /* Get low part of result. Convert `hi' to floating type and scale it back, |
2083 | then subtract this from the number being converted. This leaves the low | |
2084 | part. Convert that to integral type. */ | |
4f2e0d5e | 2085 | const UWtype lo = a - (DFtype) hi * Wtype_MAXp1_F; |
4977bab6 ZW |
2086 | |
2087 | /* Assemble result from the two parts. */ | |
4f2e0d5e | 2088 | return ((UDWtype) hi << W_TYPE_SIZE) | lo; |
203b91b9 RS |
2089 | } |
2090 | #endif | |
2091 | ||
4e9db8b2 | 2092 | #if defined(L_fixdfdi) && LIBGCC2_HAS_DF_MODE |
996ed075 | 2093 | DWtype |
37ef1054 | 2094 | __fixdfdi (DFtype a) |
203b91b9 RS |
2095 | { |
2096 | if (a < 0) | |
6da9c622 RK |
2097 | return - __fixunsdfDI (-a); |
2098 | return __fixunsdfDI (a); | |
203b91b9 RS |
2099 | } |
2100 | #endif | |
2101 | ||
cfa7bd9c | 2102 | #if defined(L_fixunssfdi) && LIBGCC2_HAS_SF_MODE |
f139f5fa | 2103 | UDWtype |
4f2e0d5e | 2104 | __fixunssfDI (SFtype a) |
203b91b9 | 2105 | { |
4e9db8b2 | 2106 | #if LIBGCC2_HAS_DF_MODE |
ab495388 | 2107 | /* Convert the SFtype to a DFtype, because that is surely not going |
203b91b9 | 2108 | to lose any bits. Some day someone else can write a faster version |
ab495388 | 2109 | that avoids converting to DFtype, and verify it really works right. */ |
4f2e0d5e | 2110 | const DFtype dfa = a; |
203b91b9 | 2111 | |
4977bab6 ZW |
2112 | /* Get high part of result. The division here will just moves the radix |
2113 | point and will not cause any rounding. Then the conversion to integral | |
2114 | type chops result as desired. */ | |
4f2e0d5e | 2115 | const UWtype hi = dfa / Wtype_MAXp1_F; |
203b91b9 | 2116 | |
4977bab6 ZW |
2117 | /* Get low part of result. Convert `hi' to floating type and scale it back, |
2118 | then subtract this from the number being converted. This leaves the low | |
2119 | part. Convert that to integral type. */ | |
4f2e0d5e | 2120 | const UWtype lo = dfa - (DFtype) hi * Wtype_MAXp1_F; |
4977bab6 ZW |
2121 | |
2122 | /* Assemble result from the two parts. */ | |
4f2e0d5e RH |
2123 | return ((UDWtype) hi << W_TYPE_SIZE) | lo; |
2124 | #elif FLT_MANT_DIG < W_TYPE_SIZE | |
2125 | if (a < 1) | |
2126 | return 0; | |
2127 | if (a < Wtype_MAXp1_F) | |
2128 | return (UWtype)a; | |
2129 | if (a < Wtype_MAXp1_F * Wtype_MAXp1_F) | |
2130 | { | |
2131 | /* Since we know that there are fewer significant bits in the SFmode | |
2132 | quantity than in a word, we know that we can convert out all the | |
2e681715 | 2133 | significant bits in one step, and thus avoid losing bits. */ |
4f2e0d5e RH |
2134 | |
2135 | /* ??? This following loop essentially performs frexpf. If we could | |
2136 | use the real libm function, or poke at the actual bits of the fp | |
2137 | format, it would be significantly faster. */ | |
2138 | ||
2139 | UWtype shift = 0, counter; | |
2140 | SFtype msb; | |
2141 | ||
2142 | a /= Wtype_MAXp1_F; | |
2143 | for (counter = W_TYPE_SIZE / 2; counter != 0; counter >>= 1) | |
2144 | { | |
2145 | SFtype counterf = (UWtype)1 << counter; | |
2146 | if (a >= counterf) | |
2147 | { | |
2148 | shift |= counter; | |
2149 | a /= counterf; | |
2150 | } | |
2151 | } | |
2152 | ||
2153 | /* Rescale into the range of one word, extract the bits of that | |
2154 | one word, and shift the result into position. */ | |
2155 | a *= Wtype_MAXp1_F; | |
2156 | counter = a; | |
2157 | return (DWtype)counter << shift; | |
2158 | } | |
2159 | return -1; | |
2160 | #else | |
2161 | # error | |
2162 | #endif | |
203b91b9 RS |
2163 | } |
2164 | #endif | |
2165 | ||
cfa7bd9c | 2166 | #if defined(L_fixsfdi) && LIBGCC2_HAS_SF_MODE |
996ed075 | 2167 | DWtype |
ab495388 | 2168 | __fixsfdi (SFtype a) |
203b91b9 RS |
2169 | { |
2170 | if (a < 0) | |
6da9c622 RK |
2171 | return - __fixunssfDI (-a); |
2172 | return __fixunssfDI (a); | |
203b91b9 RS |
2173 | } |
2174 | #endif | |
2175 | ||
4e9db8b2 | 2176 | #if defined(L_floatdixf) && LIBGCC2_HAS_XF_MODE |
e0799b34 | 2177 | XFtype |
996ed075 | 2178 | __floatdixf (DWtype u) |
e0799b34 | 2179 | { |
66bb34c0 | 2180 | #if W_TYPE_SIZE > __LIBGCC_XF_MANT_DIG__ |
4a73d865 JM |
2181 | # error |
2182 | #endif | |
4f2e0d5e RH |
2183 | XFtype d = (Wtype) (u >> W_TYPE_SIZE); |
2184 | d *= Wtype_MAXp1_F; | |
2185 | d += (UWtype)u; | |
e5e809f4 | 2186 | return d; |
e0799b34 RS |
2187 | } |
2188 | #endif | |
2189 | ||
d7735880 JM |
2190 | #if defined(L_floatundixf) && LIBGCC2_HAS_XF_MODE |
2191 | XFtype | |
2192 | __floatundixf (UDWtype u) | |
2193 | { | |
66bb34c0 | 2194 | #if W_TYPE_SIZE > __LIBGCC_XF_MANT_DIG__ |
4a73d865 JM |
2195 | # error |
2196 | #endif | |
d7735880 JM |
2197 | XFtype d = (UWtype) (u >> W_TYPE_SIZE); |
2198 | d *= Wtype_MAXp1_F; | |
2199 | d += (UWtype)u; | |
2200 | return d; | |
2201 | } | |
2202 | #endif | |
2203 | ||
4e9db8b2 | 2204 | #if defined(L_floatditf) && LIBGCC2_HAS_TF_MODE |
ab495388 | 2205 | TFtype |
996ed075 | 2206 | __floatditf (DWtype u) |
ab495388 | 2207 | { |
66bb34c0 | 2208 | #if W_TYPE_SIZE > __LIBGCC_TF_MANT_DIG__ |
4a73d865 JM |
2209 | # error |
2210 | #endif | |
4f2e0d5e RH |
2211 | TFtype d = (Wtype) (u >> W_TYPE_SIZE); |
2212 | d *= Wtype_MAXp1_F; | |
2213 | d += (UWtype)u; | |
e5e809f4 | 2214 | return d; |
ab495388 RS |
2215 | } |
2216 | #endif | |
2217 | ||
d7735880 JM |
2218 | #if defined(L_floatunditf) && LIBGCC2_HAS_TF_MODE |
2219 | TFtype | |
2220 | __floatunditf (UDWtype u) | |
2221 | { | |
66bb34c0 | 2222 | #if W_TYPE_SIZE > __LIBGCC_TF_MANT_DIG__ |
4a73d865 | 2223 | # error |
203b91b9 | 2224 | #endif |
4a73d865 | 2225 | TFtype d = (UWtype) (u >> W_TYPE_SIZE); |
d7735880 JM |
2226 | d *= Wtype_MAXp1_F; |
2227 | d += (UWtype)u; | |
2228 | return d; | |
2229 | } | |
2230 | #endif | |
2231 | ||
4a73d865 JM |
2232 | #if (defined(L_floatdisf) && LIBGCC2_HAS_SF_MODE) \ |
2233 | || (defined(L_floatdidf) && LIBGCC2_HAS_DF_MODE) | |
4f2e0d5e | 2234 | #define DI_SIZE (W_TYPE_SIZE * 2) |
b04c9063 AM |
2235 | #define F_MODE_OK(SIZE) \ |
2236 | (SIZE < DI_SIZE \ | |
2237 | && SIZE > (DI_SIZE - SIZE + FSSIZE) \ | |
5fb54b91 | 2238 | && !AVOID_FP_TYPE_CONVERSION(SIZE)) |
4a73d865 JM |
2239 | #if defined(L_floatdisf) |
2240 | #define FUNC __floatdisf | |
2241 | #define FSTYPE SFtype | |
66bb34c0 | 2242 | #define FSSIZE __LIBGCC_SF_MANT_DIG__ |
4a73d865 JM |
2243 | #else |
2244 | #define FUNC __floatdidf | |
2245 | #define FSTYPE DFtype | |
66bb34c0 | 2246 | #define FSSIZE __LIBGCC_DF_MANT_DIG__ |
4a73d865 | 2247 | #endif |
203b91b9 | 2248 | |
4a73d865 JM |
2249 | FSTYPE |
2250 | FUNC (DWtype u) | |
203b91b9 | 2251 | { |
4a73d865 | 2252 | #if FSSIZE >= W_TYPE_SIZE |
4f2e0d5e | 2253 | /* When the word size is small, we never get any rounding error. */ |
4a73d865 | 2254 | FSTYPE f = (Wtype) (u >> W_TYPE_SIZE); |
4f2e0d5e RH |
2255 | f *= Wtype_MAXp1_F; |
2256 | f += (UWtype)u; | |
2257 | return f; | |
66bb34c0 JM |
2258 | #elif (LIBGCC2_HAS_DF_MODE && F_MODE_OK (__LIBGCC_DF_MANT_DIG__)) \ |
2259 | || (LIBGCC2_HAS_XF_MODE && F_MODE_OK (__LIBGCC_XF_MANT_DIG__)) \ | |
2260 | || (LIBGCC2_HAS_TF_MODE && F_MODE_OK (__LIBGCC_TF_MANT_DIG__)) | |
4a73d865 | 2261 | |
66bb34c0 JM |
2262 | #if (LIBGCC2_HAS_DF_MODE && F_MODE_OK (__LIBGCC_DF_MANT_DIG__)) |
2263 | # define FSIZE __LIBGCC_DF_MANT_DIG__ | |
4a73d865 | 2264 | # define FTYPE DFtype |
66bb34c0 JM |
2265 | #elif (LIBGCC2_HAS_XF_MODE && F_MODE_OK (__LIBGCC_XF_MANT_DIG__)) |
2266 | # define FSIZE __LIBGCC_XF_MANT_DIG__ | |
4a73d865 | 2267 | # define FTYPE XFtype |
66bb34c0 JM |
2268 | #elif (LIBGCC2_HAS_TF_MODE && F_MODE_OK (__LIBGCC_TF_MANT_DIG__)) |
2269 | # define FSIZE __LIBGCC_TF_MANT_DIG__ | |
4a73d865 | 2270 | # define FTYPE TFtype |
4f2e0d5e RH |
2271 | #else |
2272 | # error | |
2273 | #endif | |
2274 | ||
4a73d865 | 2275 | #define REP_BIT ((UDWtype) 1 << (DI_SIZE - FSIZE)) |
4f2e0d5e | 2276 | |
d9e1ab8d | 2277 | /* Protect against double-rounding error. |
4f2e0d5e RH |
2278 | Represent any low-order bits, that might be truncated by a bit that |
2279 | won't be lost. The bit can go in anywhere below the rounding position | |
4a73d865 JM |
2280 | of the FSTYPE. A fixed mask and bit position handles all usual |
2281 | configurations. */ | |
2282 | if (! (- ((DWtype) 1 << FSIZE) < u | |
2283 | && u < ((DWtype) 1 << FSIZE))) | |
d9e1ab8d | 2284 | { |
4a73d865 | 2285 | if ((UDWtype) u & (REP_BIT - 1)) |
d9e1ab8d | 2286 | { |
4a73d865 JM |
2287 | u &= ~ (REP_BIT - 1); |
2288 | u |= REP_BIT; | |
d9e1ab8d RK |
2289 | } |
2290 | } | |
203b91b9 | 2291 | |
4a73d865 JM |
2292 | /* Do the calculation in a wider type so that we don't lose any of |
2293 | the precision of the high word while multiplying it. */ | |
2294 | FTYPE f = (Wtype) (u >> W_TYPE_SIZE); | |
4f2e0d5e RH |
2295 | f *= Wtype_MAXp1_F; |
2296 | f += (UWtype)u; | |
4a73d865 | 2297 | return (FSTYPE) f; |
4f2e0d5e | 2298 | #else |
4a73d865 JM |
2299 | #if FSSIZE >= W_TYPE_SIZE - 2 |
2300 | # error | |
2301 | #endif | |
2302 | /* Finally, the word size is larger than the number of bits in the | |
2303 | required FSTYPE, and we've got no suitable wider type. The only | |
2304 | way to avoid double rounding is to special case the | |
2305 | extraction. */ | |
4f2e0d5e RH |
2306 | |
2307 | /* If there are no high bits set, fall back to one conversion. */ | |
2308 | if ((Wtype)u == u) | |
4a73d865 | 2309 | return (FSTYPE)(Wtype)u; |
4f2e0d5e RH |
2310 | |
2311 | /* Otherwise, find the power of two. */ | |
2312 | Wtype hi = u >> W_TYPE_SIZE; | |
2313 | if (hi < 0) | |
1f6eac90 | 2314 | hi = -(UWtype) hi; |
4f2e0d5e RH |
2315 | |
2316 | UWtype count, shift; | |
5de3e2d8 BE |
2317 | #if !defined (COUNT_LEADING_ZEROS_0) || COUNT_LEADING_ZEROS_0 != W_TYPE_SIZE |
2318 | if (hi == 0) | |
2319 | count = W_TYPE_SIZE; | |
2320 | else | |
2321 | #endif | |
4f2e0d5e RH |
2322 | count_leading_zeros (count, hi); |
2323 | ||
2324 | /* No leading bits means u == minimum. */ | |
2325 | if (count == 0) | |
6395ba73 | 2326 | return Wtype_MAXp1_F * (FSTYPE) (hi | ((UWtype) u != 0)); |
4f2e0d5e | 2327 | |
4a73d865 | 2328 | shift = 1 + W_TYPE_SIZE - count; |
4f2e0d5e RH |
2329 | |
2330 | /* Shift down the most significant bits. */ | |
2331 | hi = u >> shift; | |
2332 | ||
2333 | /* If we lost any nonzero bits, set the lsb to ensure correct rounding. */ | |
5fb54b91 | 2334 | if ((UWtype)u << (W_TYPE_SIZE - shift)) |
4f2e0d5e RH |
2335 | hi |= 1; |
2336 | ||
2337 | /* Convert the one word of data, and rescale. */ | |
5fb54b91 RH |
2338 | FSTYPE f = hi, e; |
2339 | if (shift == W_TYPE_SIZE) | |
2340 | e = Wtype_MAXp1_F; | |
2341 | /* The following two cases could be merged if we knew that the target | |
2342 | supported a native unsigned->float conversion. More often, we only | |
2343 | have a signed conversion, and have to add extra fixup code. */ | |
2344 | else if (shift == W_TYPE_SIZE - 1) | |
2345 | e = Wtype_MAXp1_F / 2; | |
2346 | else | |
2347 | e = (Wtype)1 << shift; | |
2348 | return f * e; | |
4f2e0d5e | 2349 | #endif |
203b91b9 RS |
2350 | } |
2351 | #endif | |
2352 | ||
4a73d865 JM |
2353 | #if (defined(L_floatundisf) && LIBGCC2_HAS_SF_MODE) \ |
2354 | || (defined(L_floatundidf) && LIBGCC2_HAS_DF_MODE) | |
d7735880 | 2355 | #define DI_SIZE (W_TYPE_SIZE * 2) |
b04c9063 AM |
2356 | #define F_MODE_OK(SIZE) \ |
2357 | (SIZE < DI_SIZE \ | |
2358 | && SIZE > (DI_SIZE - SIZE + FSSIZE) \ | |
5fb54b91 | 2359 | && !AVOID_FP_TYPE_CONVERSION(SIZE)) |
4a73d865 JM |
2360 | #if defined(L_floatundisf) |
2361 | #define FUNC __floatundisf | |
2362 | #define FSTYPE SFtype | |
66bb34c0 | 2363 | #define FSSIZE __LIBGCC_SF_MANT_DIG__ |
4a73d865 JM |
2364 | #else |
2365 | #define FUNC __floatundidf | |
2366 | #define FSTYPE DFtype | |
66bb34c0 | 2367 | #define FSSIZE __LIBGCC_DF_MANT_DIG__ |
4a73d865 | 2368 | #endif |
d7735880 | 2369 | |
4a73d865 JM |
2370 | FSTYPE |
2371 | FUNC (UDWtype u) | |
d7735880 | 2372 | { |
4a73d865 | 2373 | #if FSSIZE >= W_TYPE_SIZE |
d7735880 | 2374 | /* When the word size is small, we never get any rounding error. */ |
4a73d865 | 2375 | FSTYPE f = (UWtype) (u >> W_TYPE_SIZE); |
d7735880 JM |
2376 | f *= Wtype_MAXp1_F; |
2377 | f += (UWtype)u; | |
2378 | return f; | |
66bb34c0 JM |
2379 | #elif (LIBGCC2_HAS_DF_MODE && F_MODE_OK (__LIBGCC_DF_MANT_DIG__)) \ |
2380 | || (LIBGCC2_HAS_XF_MODE && F_MODE_OK (__LIBGCC_XF_MANT_DIG__)) \ | |
2381 | || (LIBGCC2_HAS_TF_MODE && F_MODE_OK (__LIBGCC_TF_MANT_DIG__)) | |
4a73d865 | 2382 | |
66bb34c0 JM |
2383 | #if (LIBGCC2_HAS_DF_MODE && F_MODE_OK (__LIBGCC_DF_MANT_DIG__)) |
2384 | # define FSIZE __LIBGCC_DF_MANT_DIG__ | |
4a73d865 | 2385 | # define FTYPE DFtype |
66bb34c0 JM |
2386 | #elif (LIBGCC2_HAS_XF_MODE && F_MODE_OK (__LIBGCC_XF_MANT_DIG__)) |
2387 | # define FSIZE __LIBGCC_XF_MANT_DIG__ | |
4a73d865 | 2388 | # define FTYPE XFtype |
66bb34c0 JM |
2389 | #elif (LIBGCC2_HAS_TF_MODE && F_MODE_OK (__LIBGCC_TF_MANT_DIG__)) |
2390 | # define FSIZE __LIBGCC_TF_MANT_DIG__ | |
4a73d865 | 2391 | # define FTYPE TFtype |
d7735880 JM |
2392 | #else |
2393 | # error | |
2394 | #endif | |
2395 | ||
4a73d865 | 2396 | #define REP_BIT ((UDWtype) 1 << (DI_SIZE - FSIZE)) |
d7735880 JM |
2397 | |
2398 | /* Protect against double-rounding error. | |
2399 | Represent any low-order bits, that might be truncated by a bit that | |
2400 | won't be lost. The bit can go in anywhere below the rounding position | |
4a73d865 JM |
2401 | of the FSTYPE. A fixed mask and bit position handles all usual |
2402 | configurations. */ | |
2403 | if (u >= ((UDWtype) 1 << FSIZE)) | |
d7735880 | 2404 | { |
4a73d865 | 2405 | if ((UDWtype) u & (REP_BIT - 1)) |
d7735880 | 2406 | { |
4a73d865 JM |
2407 | u &= ~ (REP_BIT - 1); |
2408 | u |= REP_BIT; | |
d7735880 JM |
2409 | } |
2410 | } | |
2411 | ||
4a73d865 JM |
2412 | /* Do the calculation in a wider type so that we don't lose any of |
2413 | the precision of the high word while multiplying it. */ | |
2414 | FTYPE f = (UWtype) (u >> W_TYPE_SIZE); | |
d7735880 JM |
2415 | f *= Wtype_MAXp1_F; |
2416 | f += (UWtype)u; | |
4a73d865 | 2417 | return (FSTYPE) f; |
d7735880 | 2418 | #else |
4a73d865 JM |
2419 | #if FSSIZE == W_TYPE_SIZE - 1 |
2420 | # error | |
2421 | #endif | |
2422 | /* Finally, the word size is larger than the number of bits in the | |
2423 | required FSTYPE, and we've got no suitable wider type. The only | |
2424 | way to avoid double rounding is to special case the | |
2425 | extraction. */ | |
d7735880 JM |
2426 | |
2427 | /* If there are no high bits set, fall back to one conversion. */ | |
2428 | if ((UWtype)u == u) | |
4a73d865 | 2429 | return (FSTYPE)(UWtype)u; |
d7735880 JM |
2430 | |
2431 | /* Otherwise, find the power of two. */ | |
2432 | UWtype hi = u >> W_TYPE_SIZE; | |
2433 | ||
2434 | UWtype count, shift; | |
2435 | count_leading_zeros (count, hi); | |
2436 | ||
2437 | shift = W_TYPE_SIZE - count; | |
2438 | ||
2439 | /* Shift down the most significant bits. */ | |
2440 | hi = u >> shift; | |
2441 | ||
2442 | /* If we lost any nonzero bits, set the lsb to ensure correct rounding. */ | |
5fb54b91 | 2443 | if ((UWtype)u << (W_TYPE_SIZE - shift)) |
d7735880 JM |
2444 | hi |= 1; |
2445 | ||
2446 | /* Convert the one word of data, and rescale. */ | |
5fb54b91 RH |
2447 | FSTYPE f = hi, e; |
2448 | if (shift == W_TYPE_SIZE) | |
2449 | e = Wtype_MAXp1_F; | |
2450 | /* The following two cases could be merged if we knew that the target | |
2451 | supported a native unsigned->float conversion. More often, we only | |
2452 | have a signed conversion, and have to add extra fixup code. */ | |
2453 | else if (shift == W_TYPE_SIZE - 1) | |
2454 | e = Wtype_MAXp1_F / 2; | |
2455 | else | |
2456 | e = (Wtype)1 << shift; | |
2457 | return f * e; | |
d7735880 JM |
2458 | #endif |
2459 | } | |
2460 | #endif | |
2461 | ||
4e9db8b2 | 2462 | #if defined(L_fixunsxfsi) && LIBGCC2_HAS_XF_MODE |
996ed075 | 2463 | UWtype |
6da9c622 | 2464 | __fixunsxfSI (XFtype a) |
e0799b34 | 2465 | { |
5d0e6486 AO |
2466 | if (a >= - (DFtype) Wtype_MIN) |
2467 | return (Wtype) (a + Wtype_MIN) - Wtype_MIN; | |
996ed075 | 2468 | return (Wtype) a; |
e0799b34 RS |
2469 | } |
2470 | #endif | |
2471 | ||
4e9db8b2 | 2472 | #if defined(L_fixunsdfsi) && LIBGCC2_HAS_DF_MODE |
996ed075 | 2473 | UWtype |
6da9c622 | 2474 | __fixunsdfSI (DFtype a) |
203b91b9 | 2475 | { |
5d0e6486 AO |
2476 | if (a >= - (DFtype) Wtype_MIN) |
2477 | return (Wtype) (a + Wtype_MIN) - Wtype_MIN; | |
996ed075 | 2478 | return (Wtype) a; |
203b91b9 RS |
2479 | } |
2480 | #endif | |
2481 | ||
cfa7bd9c | 2482 | #if defined(L_fixunssfsi) && LIBGCC2_HAS_SF_MODE |
996ed075 | 2483 | UWtype |
6da9c622 | 2484 | __fixunssfSI (SFtype a) |
203b91b9 | 2485 | { |
5d0e6486 AO |
2486 | if (a >= - (SFtype) Wtype_MIN) |
2487 | return (Wtype) (a + Wtype_MIN) - Wtype_MIN; | |
996ed075 | 2488 | return (Wtype) a; |
203b91b9 | 2489 | } |
17684d46 RG |
2490 | #endif |
2491 | \f | |
2492 | /* Integer power helper used from __builtin_powi for non-constant | |
2493 | exponents. */ | |
2494 | ||
cfa7bd9c | 2495 | #if (defined(L_powisf2) && LIBGCC2_HAS_SF_MODE) \ |
4e9db8b2 SE |
2496 | || (defined(L_powidf2) && LIBGCC2_HAS_DF_MODE) \ |
2497 | || (defined(L_powixf2) && LIBGCC2_HAS_XF_MODE) \ | |
2498 | || (defined(L_powitf2) && LIBGCC2_HAS_TF_MODE) | |
17684d46 RG |
2499 | # if defined(L_powisf2) |
2500 | # define TYPE SFtype | |
2501 | # define NAME __powisf2 | |
2502 | # elif defined(L_powidf2) | |
2503 | # define TYPE DFtype | |
2504 | # define NAME __powidf2 | |
2505 | # elif defined(L_powixf2) | |
2506 | # define TYPE XFtype | |
2507 | # define NAME __powixf2 | |
2508 | # elif defined(L_powitf2) | |
2509 | # define TYPE TFtype | |
2510 | # define NAME __powitf2 | |
2511 | # endif | |
2512 | ||
0b8495ae FJ |
2513 | #undef int |
2514 | #undef unsigned | |
17684d46 | 2515 | TYPE |
0b8495ae | 2516 | NAME (TYPE x, int m) |
17684d46 | 2517 | { |
35da095d | 2518 | unsigned int n = m < 0 ? -(unsigned int) m : (unsigned int) m; |
17684d46 RG |
2519 | TYPE y = n % 2 ? x : 1; |
2520 | while (n >>= 1) | |
2521 | { | |
2522 | x = x * x; | |
2523 | if (n % 2) | |
2524 | y = y * x; | |
2525 | } | |
2526 | return m < 0 ? 1/y : y; | |
2527 | } | |
2528 | ||
203b91b9 RS |
2529 | #endif |
2530 | \f | |
0abcd6cc JG |
2531 | #if((defined(L_mulhc3) || defined(L_divhc3)) && LIBGCC2_HAS_HF_MODE) \ |
2532 | || ((defined(L_mulsc3) || defined(L_divsc3)) && LIBGCC2_HAS_SF_MODE) \ | |
4e9db8b2 SE |
2533 | || ((defined(L_muldc3) || defined(L_divdc3)) && LIBGCC2_HAS_DF_MODE) \ |
2534 | || ((defined(L_mulxc3) || defined(L_divxc3)) && LIBGCC2_HAS_XF_MODE) \ | |
2535 | || ((defined(L_multc3) || defined(L_divtc3)) && LIBGCC2_HAS_TF_MODE) | |
7e7e470f RH |
2536 | |
2537 | #undef float | |
2538 | #undef double | |
2539 | #undef long | |
2540 | ||
0abcd6cc JG |
2541 | #if defined(L_mulhc3) || defined(L_divhc3) |
2542 | # define MTYPE HFtype | |
2543 | # define CTYPE HCtype | |
54f0224d | 2544 | # define AMTYPE SFtype |
0abcd6cc JG |
2545 | # define MODE hc |
2546 | # define CEXT __LIBGCC_HF_FUNC_EXT__ | |
2547 | # define NOTRUNC (!__LIBGCC_HF_EXCESS_PRECISION__) | |
2548 | #elif defined(L_mulsc3) || defined(L_divsc3) | |
7e7e470f RH |
2549 | # define MTYPE SFtype |
2550 | # define CTYPE SCtype | |
54f0224d | 2551 | # define AMTYPE DFtype |
7e7e470f | 2552 | # define MODE sc |
dd69f047 | 2553 | # define CEXT __LIBGCC_SF_FUNC_EXT__ |
d758aeb5 | 2554 | # define NOTRUNC (!__LIBGCC_SF_EXCESS_PRECISION__) |
54f0224d PM |
2555 | # define RBIG (__LIBGCC_SF_MAX__ / 2) |
2556 | # define RMIN (__LIBGCC_SF_MIN__) | |
2557 | # define RMIN2 (__LIBGCC_SF_EPSILON__) | |
2558 | # define RMINSCAL (1 / __LIBGCC_SF_EPSILON__) | |
2559 | # define RMAX2 (RBIG * RMIN2) | |
7e7e470f RH |
2560 | #elif defined(L_muldc3) || defined(L_divdc3) |
2561 | # define MTYPE DFtype | |
2562 | # define CTYPE DCtype | |
2563 | # define MODE dc | |
dd69f047 | 2564 | # define CEXT __LIBGCC_DF_FUNC_EXT__ |
d758aeb5 | 2565 | # define NOTRUNC (!__LIBGCC_DF_EXCESS_PRECISION__) |
54f0224d PM |
2566 | # define RBIG (__LIBGCC_DF_MAX__ / 2) |
2567 | # define RMIN (__LIBGCC_DF_MIN__) | |
2568 | # define RMIN2 (__LIBGCC_DF_EPSILON__) | |
2569 | # define RMINSCAL (1 / __LIBGCC_DF_EPSILON__) | |
2570 | # define RMAX2 (RBIG * RMIN2) | |
7e7e470f RH |
2571 | #elif defined(L_mulxc3) || defined(L_divxc3) |
2572 | # define MTYPE XFtype | |
2573 | # define CTYPE XCtype | |
2574 | # define MODE xc | |
dd69f047 | 2575 | # define CEXT __LIBGCC_XF_FUNC_EXT__ |
d758aeb5 | 2576 | # define NOTRUNC (!__LIBGCC_XF_EXCESS_PRECISION__) |
54f0224d PM |
2577 | # define RBIG (__LIBGCC_XF_MAX__ / 2) |
2578 | # define RMIN (__LIBGCC_XF_MIN__) | |
2579 | # define RMIN2 (__LIBGCC_XF_EPSILON__) | |
2580 | # define RMINSCAL (1 / __LIBGCC_XF_EPSILON__) | |
2581 | # define RMAX2 (RBIG * RMIN2) | |
7e7e470f RH |
2582 | #elif defined(L_multc3) || defined(L_divtc3) |
2583 | # define MTYPE TFtype | |
2584 | # define CTYPE TCtype | |
2585 | # define MODE tc | |
dd69f047 | 2586 | # define CEXT __LIBGCC_TF_FUNC_EXT__ |
d758aeb5 | 2587 | # define NOTRUNC (!__LIBGCC_TF_EXCESS_PRECISION__) |
d9105685 PM |
2588 | # if __LIBGCC_TF_MANT_DIG__ == 106 |
2589 | # define RBIG (__LIBGCC_DF_MAX__ / 2) | |
2590 | # define RMIN (__LIBGCC_DF_MIN__) | |
2591 | # define RMIN2 (__LIBGCC_DF_EPSILON__) | |
2592 | # define RMINSCAL (1 / __LIBGCC_DF_EPSILON__) | |
2593 | # else | |
2594 | # define RBIG (__LIBGCC_TF_MAX__ / 2) | |
2595 | # define RMIN (__LIBGCC_TF_MIN__) | |
2596 | # define RMIN2 (__LIBGCC_TF_EPSILON__) | |
2597 | # define RMINSCAL (1 / __LIBGCC_TF_EPSILON__) | |
2598 | # endif | |
54f0224d | 2599 | # define RMAX2 (RBIG * RMIN2) |
7e7e470f RH |
2600 | #else |
2601 | # error | |
2602 | #endif | |
2603 | ||
2604 | #define CONCAT3(A,B,C) _CONCAT3(A,B,C) | |
2605 | #define _CONCAT3(A,B,C) A##B##C | |
2606 | ||
2607 | #define CONCAT2(A,B) _CONCAT2(A,B) | |
2608 | #define _CONCAT2(A,B) A##B | |
2609 | ||
af8096fc UB |
2610 | #define isnan(x) __builtin_isnan (x) |
2611 | #define isfinite(x) __builtin_isfinite (x) | |
2612 | #define isinf(x) __builtin_isinf (x) | |
7e7e470f | 2613 | |
ca22d882 | 2614 | #define INFINITY CONCAT2(__builtin_huge_val, CEXT) () |
7e7e470f RH |
2615 | #define I 1i |
2616 | ||
2617 | /* Helpers to make the following code slightly less gross. */ | |
2618 | #define COPYSIGN CONCAT2(__builtin_copysign, CEXT) | |
2619 | #define FABS CONCAT2(__builtin_fabs, CEXT) | |
2620 | ||
2621 | /* Verify that MTYPE matches up with CEXT. */ | |
2622 | extern void *compile_type_assert[sizeof(INFINITY) == sizeof(MTYPE) ? 1 : -1]; | |
2623 | ||
2624 | /* Ensure that we've lost any extra precision. */ | |
2625 | #if NOTRUNC | |
2626 | # define TRUNC(x) | |
2627 | #else | |
2628 | # define TRUNC(x) __asm__ ("" : "=m"(x) : "m"(x)) | |
2629 | #endif | |
2630 | ||
0abcd6cc | 2631 | #if defined(L_mulhc3) || defined(L_mulsc3) || defined(L_muldc3) \ |
7e7e470f RH |
2632 | || defined(L_mulxc3) || defined(L_multc3) |
2633 | ||
2634 | CTYPE | |
2635 | CONCAT3(__mul,MODE,3) (MTYPE a, MTYPE b, MTYPE c, MTYPE d) | |
2636 | { | |
2637 | MTYPE ac, bd, ad, bc, x, y; | |
ddef83d2 | 2638 | CTYPE res; |
7e7e470f RH |
2639 | |
2640 | ac = a * c; | |
2641 | bd = b * d; | |
2642 | ad = a * d; | |
2643 | bc = b * c; | |
2644 | ||
2645 | TRUNC (ac); | |
2646 | TRUNC (bd); | |
2647 | TRUNC (ad); | |
2648 | TRUNC (bc); | |
2649 | ||
2650 | x = ac - bd; | |
2651 | y = ad + bc; | |
2652 | ||
2653 | if (isnan (x) && isnan (y)) | |
2654 | { | |
2655 | /* Recover infinities that computed as NaN + iNaN. */ | |
2656 | _Bool recalc = 0; | |
2657 | if (isinf (a) || isinf (b)) | |
2658 | { | |
2659 | /* z is infinite. "Box" the infinity and change NaNs in | |
2660 | the other factor to 0. */ | |
2661 | a = COPYSIGN (isinf (a) ? 1 : 0, a); | |
2662 | b = COPYSIGN (isinf (b) ? 1 : 0, b); | |
2663 | if (isnan (c)) c = COPYSIGN (0, c); | |
2664 | if (isnan (d)) d = COPYSIGN (0, d); | |
2665 | recalc = 1; | |
2666 | } | |
2667 | if (isinf (c) || isinf (d)) | |
2668 | { | |
2669 | /* w is infinite. "Box" the infinity and change NaNs in | |
2670 | the other factor to 0. */ | |
2671 | c = COPYSIGN (isinf (c) ? 1 : 0, c); | |
2672 | d = COPYSIGN (isinf (d) ? 1 : 0, d); | |
2673 | if (isnan (a)) a = COPYSIGN (0, a); | |
2674 | if (isnan (b)) b = COPYSIGN (0, b); | |
2675 | recalc = 1; | |
2676 | } | |
2677 | if (!recalc | |
2678 | && (isinf (ac) || isinf (bd) | |
2679 | || isinf (ad) || isinf (bc))) | |
2680 | { | |
2681 | /* Recover infinities from overflow by changing NaNs to 0. */ | |
2682 | if (isnan (a)) a = COPYSIGN (0, a); | |
2683 | if (isnan (b)) b = COPYSIGN (0, b); | |
2684 | if (isnan (c)) c = COPYSIGN (0, c); | |
2685 | if (isnan (d)) d = COPYSIGN (0, d); | |
2686 | recalc = 1; | |
2687 | } | |
2688 | if (recalc) | |
2689 | { | |
2690 | x = INFINITY * (a * c - b * d); | |
2691 | y = INFINITY * (a * d + b * c); | |
2692 | } | |
2693 | } | |
2694 | ||
ddef83d2 RG |
2695 | __real__ res = x; |
2696 | __imag__ res = y; | |
2697 | return res; | |
7e7e470f RH |
2698 | } |
2699 | #endif /* complex multiply */ | |
2700 | ||
0abcd6cc | 2701 | #if defined(L_divhc3) || defined(L_divsc3) || defined(L_divdc3) \ |
7e7e470f RH |
2702 | || defined(L_divxc3) || defined(L_divtc3) |
2703 | ||
2704 | CTYPE | |
2705 | CONCAT3(__div,MODE,3) (MTYPE a, MTYPE b, MTYPE c, MTYPE d) | |
2706 | { | |
54f0224d PM |
2707 | #if defined(L_divhc3) \ |
2708 | || (defined(L_divsc3) && defined(__LIBGCC_HAVE_HWDBL__) ) | |
2709 | ||
2710 | /* Half precision is handled with float precision. | |
2711 | float is handled with double precision when double precision | |
2712 | hardware is available. | |
2713 | Due to the additional precision, the simple complex divide | |
2714 | method (without Smith's method) is sufficient to get accurate | |
2715 | answers and runs slightly faster than Smith's method. */ | |
2716 | ||
2717 | AMTYPE aa, bb, cc, dd; | |
2718 | AMTYPE denom; | |
2719 | MTYPE x, y; | |
2720 | CTYPE res; | |
2721 | aa = a; | |
2722 | bb = b; | |
2723 | cc = c; | |
2724 | dd = d; | |
2725 | ||
2726 | denom = (cc * cc) + (dd * dd); | |
2727 | x = ((aa * cc) + (bb * dd)) / denom; | |
2728 | y = ((bb * cc) - (aa * dd)) / denom; | |
2729 | ||
2730 | #else | |
7e7e470f | 2731 | MTYPE denom, ratio, x, y; |
ddef83d2 | 2732 | CTYPE res; |
7e7e470f | 2733 | |
54f0224d PM |
2734 | /* double, extended, long double have significant potential |
2735 | underflow/overflow errors that can be greatly reduced with | |
2736 | a limited number of tests and adjustments. float is handled | |
2737 | the same way when no HW double is available. | |
2738 | */ | |
2739 | ||
2740 | /* Scale by max(c,d) to reduce chances of denominator overflowing. */ | |
7e7e470f RH |
2741 | if (FABS (c) < FABS (d)) |
2742 | { | |
54f0224d PM |
2743 | /* Prevent underflow when denominator is near max representable. */ |
2744 | if (FABS (d) >= RBIG) | |
2745 | { | |
2746 | a = a / 2; | |
2747 | b = b / 2; | |
2748 | c = c / 2; | |
2749 | d = d / 2; | |
2750 | } | |
2751 | /* Avoid overflow/underflow issues when c and d are small. | |
2752 | Scaling up helps avoid some underflows. | |
2753 | No new overflow possible since c&d < RMIN2. */ | |
2754 | if (FABS (d) < RMIN2) | |
2755 | { | |
2756 | a = a * RMINSCAL; | |
2757 | b = b * RMINSCAL; | |
2758 | c = c * RMINSCAL; | |
2759 | d = d * RMINSCAL; | |
2760 | } | |
2761 | else | |
2762 | { | |
2763 | if (((FABS (a) < RMIN) && (FABS (b) < RMAX2) && (FABS (d) < RMAX2)) | |
2764 | || ((FABS (b) < RMIN) && (FABS (a) < RMAX2) | |
2765 | && (FABS (d) < RMAX2))) | |
2766 | { | |
2767 | a = a * RMINSCAL; | |
2768 | b = b * RMINSCAL; | |
2769 | c = c * RMINSCAL; | |
2770 | d = d * RMINSCAL; | |
2771 | } | |
2772 | } | |
7e7e470f RH |
2773 | ratio = c / d; |
2774 | denom = (c * ratio) + d; | |
54f0224d PM |
2775 | /* Choose alternate order of computation if ratio is subnormal. */ |
2776 | if (FABS (ratio) > RMIN) | |
2777 | { | |
2778 | x = ((a * ratio) + b) / denom; | |
2779 | y = ((b * ratio) - a) / denom; | |
2780 | } | |
2781 | else | |
2782 | { | |
2783 | x = ((c * (a / d)) + b) / denom; | |
2784 | y = ((c * (b / d)) - a) / denom; | |
2785 | } | |
7e7e470f RH |
2786 | } |
2787 | else | |
2788 | { | |
54f0224d PM |
2789 | /* Prevent underflow when denominator is near max representable. */ |
2790 | if (FABS (c) >= RBIG) | |
2791 | { | |
2792 | a = a / 2; | |
2793 | b = b / 2; | |
2794 | c = c / 2; | |
2795 | d = d / 2; | |
2796 | } | |
2797 | /* Avoid overflow/underflow issues when both c and d are small. | |
2798 | Scaling up helps avoid some underflows. | |
2799 | No new overflow possible since both c&d are less than RMIN2. */ | |
2800 | if (FABS (c) < RMIN2) | |
2801 | { | |
2802 | a = a * RMINSCAL; | |
2803 | b = b * RMINSCAL; | |
2804 | c = c * RMINSCAL; | |
2805 | d = d * RMINSCAL; | |
2806 | } | |
2807 | else | |
2808 | { | |
2809 | if (((FABS (a) < RMIN) && (FABS (b) < RMAX2) && (FABS (c) < RMAX2)) | |
2810 | || ((FABS (b) < RMIN) && (FABS (a) < RMAX2) | |
2811 | && (FABS (c) < RMAX2))) | |
2812 | { | |
2813 | a = a * RMINSCAL; | |
2814 | b = b * RMINSCAL; | |
2815 | c = c * RMINSCAL; | |
2816 | d = d * RMINSCAL; | |
2817 | } | |
2818 | } | |
7e7e470f RH |
2819 | ratio = d / c; |
2820 | denom = (d * ratio) + c; | |
54f0224d PM |
2821 | /* Choose alternate order of computation if ratio is subnormal. */ |
2822 | if (FABS (ratio) > RMIN) | |
2823 | { | |
2824 | x = ((b * ratio) + a) / denom; | |
2825 | y = (b - (a * ratio)) / denom; | |
2826 | } | |
2827 | else | |
2828 | { | |
2829 | x = (a + (d * (b / c))) / denom; | |
2830 | y = (b - (d * (a / c))) / denom; | |
2831 | } | |
7e7e470f | 2832 | } |
54f0224d | 2833 | #endif |
7e7e470f | 2834 | |
54f0224d PM |
2835 | /* Recover infinities and zeros that computed as NaN+iNaN; the only |
2836 | cases are nonzero/zero, infinite/finite, and finite/infinite. */ | |
7e7e470f RH |
2837 | if (isnan (x) && isnan (y)) |
2838 | { | |
698ac934 | 2839 | if (c == 0.0 && d == 0.0 && (!isnan (a) || !isnan (b))) |
7e7e470f RH |
2840 | { |
2841 | x = COPYSIGN (INFINITY, c) * a; | |
2842 | y = COPYSIGN (INFINITY, c) * b; | |
2843 | } | |
2844 | else if ((isinf (a) || isinf (b)) && isfinite (c) && isfinite (d)) | |
2845 | { | |
2846 | a = COPYSIGN (isinf (a) ? 1 : 0, a); | |
2847 | b = COPYSIGN (isinf (b) ? 1 : 0, b); | |
2848 | x = INFINITY * (a * c + b * d); | |
2849 | y = INFINITY * (b * c - a * d); | |
2850 | } | |
2851 | else if ((isinf (c) || isinf (d)) && isfinite (a) && isfinite (b)) | |
2852 | { | |
2853 | c = COPYSIGN (isinf (c) ? 1 : 0, c); | |
2854 | d = COPYSIGN (isinf (d) ? 1 : 0, d); | |
2855 | x = 0.0 * (a * c + b * d); | |
2856 | y = 0.0 * (b * c - a * d); | |
2857 | } | |
2858 | } | |
2859 | ||
ddef83d2 RG |
2860 | __real__ res = x; |
2861 | __imag__ res = y; | |
2862 | return res; | |
7e7e470f RH |
2863 | } |
2864 | #endif /* complex divide */ | |
2865 | ||
2866 | #endif /* all complex float routines */ | |
2867 | \f | |
ab495388 RS |
2868 | /* From here on down, the routines use normal data types. */ |
2869 | ||
2870 | #define SItype bogus_type | |
2871 | #define USItype bogus_type | |
2872 | #define DItype bogus_type | |
2873 | #define UDItype bogus_type | |
2874 | #define SFtype bogus_type | |
2875 | #define DFtype bogus_type | |
996ed075 JJ |
2876 | #undef Wtype |
2877 | #undef UWtype | |
2878 | #undef HWtype | |
2879 | #undef UHWtype | |
2880 | #undef DWtype | |
2881 | #undef UDWtype | |
ab495388 RS |
2882 | |
2883 | #undef char | |
2884 | #undef short | |
2885 | #undef int | |
2886 | #undef long | |
2887 | #undef unsigned | |
2888 | #undef float | |
2889 | #undef double | |
9bd23d2c RS |
2890 | \f |
2891 | #ifdef L__gcc_bcmp | |
2892 | ||
2893 | /* Like bcmp except the sign is meaningful. | |
9faa82d8 | 2894 | Result is negative if S1 is less than S2, |
9bd23d2c RS |
2895 | positive if S1 is greater, 0 if S1 and S2 are equal. */ |
2896 | ||
2897 | int | |
299b83b7 | 2898 | __gcc_bcmp (const unsigned char *s1, const unsigned char *s2, size_t size) |
9bd23d2c RS |
2899 | { |
2900 | while (size > 0) | |
2901 | { | |
b982024e | 2902 | const unsigned char c1 = *s1++, c2 = *s2++; |
9bd23d2c RS |
2903 | if (c1 != c2) |
2904 | return c1 - c2; | |
2905 | size--; | |
2906 | } | |
2907 | return 0; | |
2908 | } | |
ab495388 | 2909 | |
3fe68d0a ZW |
2910 | #endif |
2911 | \f | |
2912 | /* __eprintf used to be used by GCC's private version of <assert.h>. | |
2913 | We no longer provide that header, but this routine remains in libgcc.a | |
2914 | for binary backward compatibility. Note that it is not included in | |
2915 | the shared version of libgcc. */ | |
2916 | #ifdef L_eprintf | |
2917 | #ifndef inhibit_libc | |
2918 | ||
2919 | #undef NULL /* Avoid errors if stdio.h and our stddef.h mismatch. */ | |
2920 | #include <stdio.h> | |
2921 | ||
2922 | void | |
2923 | __eprintf (const char *string, const char *expression, | |
2924 | unsigned int line, const char *filename) | |
2925 | { | |
2926 | fprintf (stderr, string, expression, line, filename); | |
2927 | fflush (stderr); | |
2928 | abort (); | |
2929 | } | |
2930 | ||
2931 | #endif | |
203b91b9 RS |
2932 | #endif |
2933 | ||
203b91b9 | 2934 | \f |
203b91b9 RS |
2935 | #ifdef L_clear_cache |
2936 | /* Clear part of an instruction cache. */ | |
2937 | ||
203b91b9 | 2938 | void |
a90b0cdd MS |
2939 | __clear_cache (void *beg __attribute__((__unused__)), |
2940 | void *end __attribute__((__unused__))) | |
203b91b9 | 2941 | { |
23190837 | 2942 | #ifdef CLEAR_INSN_CACHE |
a90b0cdd MS |
2943 | /* Cast the void* pointers to char* as some implementations |
2944 | of the macro assume the pointers can be subtracted from | |
2945 | one another. */ | |
2946 | CLEAR_INSN_CACHE ((char *) beg, (char *) end); | |
e1178973 | 2947 | #endif /* CLEAR_INSN_CACHE */ |
203b91b9 RS |
2948 | } |
2949 | ||
2950 | #endif /* L_clear_cache */ | |
2951 | \f | |
2952 | #ifdef L_trampoline | |
2953 | ||
2954 | /* Jump to a trampoline, loading the static chain address. */ | |
2955 | ||
cd985f66 | 2956 | #if defined(WINNT) && ! defined(__CYGWIN__) |
902c7559 | 2957 | #define WIN32_LEAN_AND_MEAN |
bf806a90 | 2958 | #include <windows.h> |
0a38153f KT |
2959 | int getpagesize (void); |
2960 | int mprotect (char *,int, int); | |
e3367a77 | 2961 | |
94c1e7ac | 2962 | int |
3e7d8ef1 | 2963 | getpagesize (void) |
f5ea9817 RK |
2964 | { |
2965 | #ifdef _ALPHA_ | |
2966 | return 8192; | |
2967 | #else | |
2968 | return 4096; | |
2969 | #endif | |
2970 | } | |
2971 | ||
272e2587 RK |
2972 | int |
2973 | mprotect (char *addr, int len, int prot) | |
f5ea9817 | 2974 | { |
234952b3 | 2975 | DWORD np, op; |
f5ea9817 | 2976 | |
272e2587 RK |
2977 | if (prot == 7) |
2978 | np = 0x40; | |
2979 | else if (prot == 5) | |
2980 | np = 0x20; | |
2981 | else if (prot == 4) | |
2982 | np = 0x10; | |
2983 | else if (prot == 3) | |
2984 | np = 0x04; | |
2985 | else if (prot == 1) | |
2986 | np = 0x02; | |
2987 | else if (prot == 0) | |
2988 | np = 0x01; | |
234952b3 OS |
2989 | else |
2990 | return -1; | |
f5ea9817 RK |
2991 | |
2992 | if (VirtualProtect (addr, len, np, &op)) | |
2993 | return 0; | |
2994 | else | |
2995 | return -1; | |
f5ea9817 RK |
2996 | } |
2997 | ||
cd985f66 | 2998 | #endif /* WINNT && ! __CYGWIN__ */ |
f5ea9817 | 2999 | |
23190837 AJ |
3000 | #ifdef TRANSFER_FROM_TRAMPOLINE |
3001 | TRANSFER_FROM_TRAMPOLINE | |
203b91b9 | 3002 | #endif |
203b91b9 RS |
3003 | #endif /* L_trampoline */ |
3004 | \f | |
cae21ae8 | 3005 | #ifndef __CYGWIN__ |
203b91b9 RS |
3006 | #ifdef L__main |
3007 | ||
3008 | #include "gbl-ctors.h" | |
7abc66b1 | 3009 | |
c06cff95 RS |
3010 | /* Some systems use __main in a way incompatible with its use in gcc, in these |
3011 | cases use the macros NAME__MAIN to give a quoted symbol and SYMBOL__MAIN to | |
3012 | give the same symbol without quotes for an alternative entry point. You | |
0f41302f | 3013 | must define both, or neither. */ |
c06cff95 RS |
3014 | #ifndef NAME__MAIN |
3015 | #define NAME__MAIN "__main" | |
3016 | #define SYMBOL__MAIN __main | |
3017 | #endif | |
203b91b9 | 3018 | |
53d68b9f JM |
3019 | #if defined (__LIBGCC_INIT_SECTION_ASM_OP__) \ |
3020 | || defined (__LIBGCC_INIT_ARRAY_SECTION_ASM_OP__) | |
fe1fd353 JM |
3021 | #undef HAS_INIT_SECTION |
3022 | #define HAS_INIT_SECTION | |
3023 | #endif | |
3024 | ||
3025 | #if !defined (HAS_INIT_SECTION) || !defined (OBJECT_FORMAT_ELF) | |
31cf0144 JM |
3026 | |
3027 | /* Some ELF crosses use crtstuff.c to provide __CTOR_LIST__, but use this | |
72d1a48d EB |
3028 | code to run constructors. In that case, we need to handle EH here, too. |
3029 | But MINGW32 is special because it handles CRTSTUFF and EH on its own. */ | |
3030 | ||
3031 | #ifdef __MINGW32__ | |
3032 | #undef __LIBGCC_EH_FRAME_SECTION_NAME__ | |
3033 | #endif | |
31cf0144 | 3034 | |
53d68b9f | 3035 | #ifdef __LIBGCC_EH_FRAME_SECTION_NAME__ |
e4b776a6 | 3036 | #include "unwind-dw2-fde.h" |
31cf0144 JM |
3037 | extern unsigned char __EH_FRAME_BEGIN__[]; |
3038 | #endif | |
3039 | ||
203b91b9 RS |
3040 | /* Run all the global destructors on exit from the program. */ |
3041 | ||
3042 | void | |
3e7d8ef1 | 3043 | __do_global_dtors (void) |
203b91b9 | 3044 | { |
89cf554b RS |
3045 | #ifdef DO_GLOBAL_DTORS_BODY |
3046 | DO_GLOBAL_DTORS_BODY; | |
3047 | #else | |
b40b9d93 MS |
3048 | static func_ptr *p = __DTOR_LIST__ + 1; |
3049 | while (*p) | |
3050 | { | |
3051 | p++; | |
3052 | (*(p-1)) (); | |
3053 | } | |
89cf554b | 3054 | #endif |
53d68b9f | 3055 | #if defined (__LIBGCC_EH_FRAME_SECTION_NAME__) && !defined (HAS_INIT_SECTION) |
a4ebb0e6 GRK |
3056 | { |
3057 | static int completed = 0; | |
3058 | if (! completed) | |
3059 | { | |
3060 | completed = 1; | |
3061 | __deregister_frame_info (__EH_FRAME_BEGIN__); | |
3062 | } | |
3063 | } | |
31cf0144 | 3064 | #endif |
203b91b9 | 3065 | } |
68d69835 | 3066 | #endif |
203b91b9 | 3067 | |
fe1fd353 | 3068 | #ifndef HAS_INIT_SECTION |
203b91b9 RS |
3069 | /* Run all the global constructors on entry to the program. */ |
3070 | ||
203b91b9 | 3071 | void |
3e7d8ef1 | 3072 | __do_global_ctors (void) |
203b91b9 | 3073 | { |
53d68b9f | 3074 | #ifdef __LIBGCC_EH_FRAME_SECTION_NAME__ |
31cf0144 JM |
3075 | { |
3076 | static struct object object; | |
3077 | __register_frame_info (__EH_FRAME_BEGIN__, &object); | |
3078 | } | |
3079 | #endif | |
203b91b9 | 3080 | DO_GLOBAL_CTORS_BODY; |
a218d5ba | 3081 | atexit (__do_global_dtors); |
203b91b9 | 3082 | } |
fe1fd353 | 3083 | #endif /* no HAS_INIT_SECTION */ |
203b91b9 | 3084 | |
fe1fd353 | 3085 | #if !defined (HAS_INIT_SECTION) || defined (INVOKE__main) |
203b91b9 RS |
3086 | /* Subroutine called automatically by `main'. |
3087 | Compiling a global function named `main' | |
3088 | produces an automatic call to this function at the beginning. | |
3089 | ||
3090 | For many systems, this routine calls __do_global_ctors. | |
3091 | For systems which support a .init section we use the .init section | |
3092 | to run __do_global_ctors, so we need not do anything here. */ | |
3093 | ||
4043d9c1 | 3094 | extern void SYMBOL__MAIN (void); |
203b91b9 | 3095 | void |
4043d9c1 | 3096 | SYMBOL__MAIN (void) |
203b91b9 RS |
3097 | { |
3098 | /* Support recursive calls to `main': run initializers just once. */ | |
7e6f1890 | 3099 | static int initialized; |
203b91b9 RS |
3100 | if (! initialized) |
3101 | { | |
3102 | initialized = 1; | |
3103 | __do_global_ctors (); | |
3104 | } | |
3105 | } | |
fe1fd353 | 3106 | #endif /* no HAS_INIT_SECTION or INVOKE__main */ |
203b91b9 RS |
3107 | |
3108 | #endif /* L__main */ | |
cae21ae8 | 3109 | #endif /* __CYGWIN__ */ |
203b91b9 | 3110 | \f |
ad38743d | 3111 | #ifdef L_ctors |
203b91b9 RS |
3112 | |
3113 | #include "gbl-ctors.h" | |
3114 | ||
3115 | /* Provide default definitions for the lists of constructors and | |
657be7af JL |
3116 | destructors, so that we don't get linker errors. These symbols are |
3117 | intentionally bss symbols, so that gld and/or collect will provide | |
3118 | the right values. */ | |
203b91b9 RS |
3119 | |
3120 | /* We declare the lists here with two elements each, | |
657be7af JL |
3121 | so that they are valid empty lists if no other definition is loaded. |
3122 | ||
3123 | If we are using the old "set" extensions to have the gnu linker | |
3124 | collect ctors and dtors, then we __CTOR_LIST__ and __DTOR_LIST__ | |
3125 | must be in the bss/common section. | |
3126 | ||
3127 | Long term no port should use those extensions. But many still do. */ | |
1770511a | 3128 | #if !defined(__LIBGCC_INIT_SECTION_ASM_OP__) |
aa6ad1a6 | 3129 | #if defined (TARGET_ASM_CONSTRUCTOR) || defined (USE_COLLECT2) |
d15d0264 RS |
3130 | func_ptr __CTOR_LIST__[2] = {0, 0}; |
3131 | func_ptr __DTOR_LIST__[2] = {0, 0}; | |
657be7af JL |
3132 | #else |
3133 | func_ptr __CTOR_LIST__[2]; | |
3134 | func_ptr __DTOR_LIST__[2]; | |
3135 | #endif | |
1770511a | 3136 | #endif /* no __LIBGCC_INIT_SECTION_ASM_OP__ */ |
ad38743d | 3137 | #endif /* L_ctors */ |
baffad1f | 3138 | #endif /* LIBGCC2_UNITS_PER_WORD <= MIN_UNITS_PER_WORD */ |