1/* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2 * Note: I added some stuff for use with gnupg
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994, 1996, 1998,
5 * 2000, 2001, 2002, 2003 Free Software Foundation, Inc.
6 *
7 * This file is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU Library General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or (at your
10 * option) any later version.
11 *
12 * This file is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
14 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
15 * License for more details.
16 *
17 * You should have received a copy of the GNU Library General Public License
18 * along with this file; see the file COPYING.LIB. If not, write to
19 * the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
20 * MA 02111-1307, USA. */
21
22#include <linux/count_zeros.h>
23
24/* You have to define the following before including this file:
25 *
26 * UWtype -- An unsigned type, default type for operations (typically a "word")
27 * UHWtype -- An unsigned type, at least half the size of UWtype.
28 * UDWtype -- An unsigned type, at least twice as large a UWtype
29 * W_TYPE_SIZE -- size in bits of UWtype
30 *
31 * SItype, USItype -- Signed and unsigned 32 bit types.
32 * DItype, UDItype -- Signed and unsigned 64 bit types.
33 *
34 * On a 32 bit machine UWtype should typically be USItype;
35 * on a 64 bit machine, UWtype should typically be UDItype.
36*/
37
38#define __BITS4 (W_TYPE_SIZE / 4)
39#define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
40#define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
41#define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
42
43/* This is used to make sure no undesirable sharing between different libraries
44 that use this file takes place. */
45#ifndef __MPN
46#define __MPN(x) __##x
47#endif
48
49/* Define auxiliary asm macros.
50 *
51 * 1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
52 * UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
53 * word product in HIGH_PROD and LOW_PROD.
54 *
55 * 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
56 * UDWtype product. This is just a variant of umul_ppmm.
57
58 * 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
59 * denominator) divides a UDWtype, composed by the UWtype integers
60 * HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
61 * in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less
62 * than DENOMINATOR for correct operation. If, in addition, the most
63 * significant bit of DENOMINATOR must be 1, then the pre-processor symbol
64 * UDIV_NEEDS_NORMALIZATION is defined to 1.
65 * 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
66 * denominator). Like udiv_qrnnd but the numbers are signed. The quotient
67 * is rounded towards 0.
68 *
69 * 5) count_leading_zeros(count, x) counts the number of zero-bits from the
70 * msb to the first non-zero bit in the UWtype X. This is the number of
71 * steps X needs to be shifted left to set the msb. Undefined for X == 0,
72 * unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
73 *
74 * 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
75 * from the least significant end.
76 *
77 * 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
78 * high_addend_2, low_addend_2) adds two UWtype integers, composed by
79 * HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
80 * respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow
81 * (i.e. carry out) is not stored anywhere, and is lost.
82 *
83 * 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
84 * high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
85 * composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
86 * LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE
87 * and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
88 * and is lost.
89 *
90 * If any of these macros are left undefined for a particular CPU,
91 * C macros are used. */
92
93/* The CPUs come in alphabetical order below.
94 *
95 * Please add support for more CPUs here, or improve the current support
96 * for the CPUs below! */
97
98#if defined(__GNUC__) && !defined(NO_ASM)
99
100/* We sometimes need to clobber "cc" with gcc2, but that would not be
101 understood by gcc1. Use cpp to avoid major code duplication. */
102#if __GNUC__ < 2
103#define __CLOBBER_CC
104#define __AND_CLOBBER_CC
105#else /* __GNUC__ >= 2 */
106#define __CLOBBER_CC : "cc"
107#define __AND_CLOBBER_CC , "cc"
108#endif /* __GNUC__ < 2 */
109
110/***************************************
111 ************** A29K *****************
112 ***************************************/
113#if (defined(__a29k__) || defined(_AM29K)) && W_TYPE_SIZE == 32
114#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
115 __asm__ ("add %1,%4,%5\n" \
116 "addc %0,%2,%3" \
117 : "=r" ((USItype)(sh)), \
118 "=&r" ((USItype)(sl)) \
119 : "%r" ((USItype)(ah)), \
120 "rI" ((USItype)(bh)), \
121 "%r" ((USItype)(al)), \
122 "rI" ((USItype)(bl)))
123#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
124 __asm__ ("sub %1,%4,%5\n" \
125 "subc %0,%2,%3" \
126 : "=r" ((USItype)(sh)), \
127 "=&r" ((USItype)(sl)) \
128 : "r" ((USItype)(ah)), \
129 "rI" ((USItype)(bh)), \
130 "r" ((USItype)(al)), \
131 "rI" ((USItype)(bl)))
132#define umul_ppmm(xh, xl, m0, m1) \
133do { \
134 USItype __m0 = (m0), __m1 = (m1); \
135 __asm__ ("multiplu %0,%1,%2" \
136 : "=r" ((USItype)(xl)) \
137 : "r" (__m0), \
138 "r" (__m1)); \
139 __asm__ ("multmu %0,%1,%2" \
140 : "=r" ((USItype)(xh)) \
141 : "r" (__m0), \
142 "r" (__m1)); \
143} while (0)
144#define udiv_qrnnd(q, r, n1, n0, d) \
145 __asm__ ("dividu %0,%3,%4" \
146 : "=r" ((USItype)(q)), \
147 "=q" ((USItype)(r)) \
148 : "1" ((USItype)(n1)), \
149 "r" ((USItype)(n0)), \
150 "r" ((USItype)(d)))
151#endif /* __a29k__ */
152
153#if defined(__alpha) && W_TYPE_SIZE == 64
154#define umul_ppmm(ph, pl, m0, m1) \
155do { \
156 UDItype __m0 = (m0), __m1 = (m1); \
157 (ph) = __builtin_alpha_umulh(__m0, __m1); \
158 (pl) = __m0 * __m1; \
159} while (0)
160#define UMUL_TIME 46
161#ifndef LONGLONG_STANDALONE
162#define udiv_qrnnd(q, r, n1, n0, d) \
163do { UDItype __r; \
164 (q) = __udiv_qrnnd(&__r, (n1), (n0), (d)); \
165 (r) = __r; \
166} while (0)
167extern UDItype __udiv_qrnnd(UDItype *, UDItype, UDItype, UDItype);
168#define UDIV_TIME 220
169#endif /* LONGLONG_STANDALONE */
170#endif /* __alpha */
171
172/***************************************
173 ************** ARM ******************
174 ***************************************/
175#if defined(__arm__) && W_TYPE_SIZE == 32
176#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
177 __asm__ ("adds %1, %4, %5\n" \
178 "adc %0, %2, %3" \
179 : "=r" (sh), \
180 "=&r" (sl) \
181 : "%r" ((USItype)(ah)), \
182 "rI" ((USItype)(bh)), \
183 "%r" ((USItype)(al)), \
184 "rI" ((USItype)(bl)))
185#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
186 __asm__ ("subs %1, %4, %5\n" \
187 "sbc %0, %2, %3" \
188 : "=r" (sh), \
189 "=&r" (sl) \
190 : "r" ((USItype)(ah)), \
191 "rI" ((USItype)(bh)), \
192 "r" ((USItype)(al)), \
193 "rI" ((USItype)(bl)))
194#if defined __ARM_ARCH_2__ || defined __ARM_ARCH_3__
195#define umul_ppmm(xh, xl, a, b) \
196 __asm__ ("@ Inlined umul_ppmm\n" \
197 "mov %|r0, %2, lsr #16 @ AAAA\n" \
198 "mov %|r2, %3, lsr #16 @ BBBB\n" \
199 "bic %|r1, %2, %|r0, lsl #16 @ aaaa\n" \
200 "bic %0, %3, %|r2, lsl #16 @ bbbb\n" \
201 "mul %1, %|r1, %|r2 @ aaaa * BBBB\n" \
202 "mul %|r2, %|r0, %|r2 @ AAAA * BBBB\n" \
203 "mul %|r1, %0, %|r1 @ aaaa * bbbb\n" \
204 "mul %0, %|r0, %0 @ AAAA * bbbb\n" \
205 "adds %|r0, %1, %0 @ central sum\n" \
206 "addcs %|r2, %|r2, #65536\n" \
207 "adds %1, %|r1, %|r0, lsl #16\n" \
208 "adc %0, %|r2, %|r0, lsr #16" \
209 : "=&r" (xh), \
210 "=r" (xl) \
211 : "r" ((USItype)(a)), \
212 "r" ((USItype)(b)) \
213 : "r0", "r1", "r2")
214#else
215#define umul_ppmm(xh, xl, a, b) \
216 __asm__ ("@ Inlined umul_ppmm\n" \
217 "umull %1, %0, %2, %3" \
218 : "=&r" (xh), \
219 "=&r" (xl) \
220 : "r" ((USItype)(a)), \
221 "r" ((USItype)(b)) \
222 : "r0", "r1")
223#endif
224#define UMUL_TIME 20
225#define UDIV_TIME 100
226#endif /* __arm__ */
227
228/***************************************
229 ************** CLIPPER **************
230 ***************************************/
231#if defined(__clipper__) && W_TYPE_SIZE == 32
232#define umul_ppmm(w1, w0, u, v) \
233 ({union {UDItype __ll; \
234 struct {USItype __l, __h; } __i; \
235 } __xx; \
236 __asm__ ("mulwux %2,%0" \
237 : "=r" (__xx.__ll) \
238 : "%0" ((USItype)(u)), \
239 "r" ((USItype)(v))); \
240 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l; })
241#define smul_ppmm(w1, w0, u, v) \
242 ({union {DItype __ll; \
243 struct {SItype __l, __h; } __i; \
244 } __xx; \
245 __asm__ ("mulwx %2,%0" \
246 : "=r" (__xx.__ll) \
247 : "%0" ((SItype)(u)), \
248 "r" ((SItype)(v))); \
249 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l; })
250#define __umulsidi3(u, v) \
251 ({UDItype __w; \
252 __asm__ ("mulwux %2,%0" \
253 : "=r" (__w) \
254 : "%0" ((USItype)(u)), \
255 "r" ((USItype)(v))); \
256 __w; })
257#endif /* __clipper__ */
258
259/***************************************
260 ************** GMICRO ***************
261 ***************************************/
262#if defined(__gmicro__) && W_TYPE_SIZE == 32
263#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
264 __asm__ ("add.w %5,%1\n" \
265 "addx %3,%0" \
266 : "=g" ((USItype)(sh)), \
267 "=&g" ((USItype)(sl)) \
268 : "%0" ((USItype)(ah)), \
269 "g" ((USItype)(bh)), \
270 "%1" ((USItype)(al)), \
271 "g" ((USItype)(bl)))
272#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
273 __asm__ ("sub.w %5,%1\n" \
274 "subx %3,%0" \
275 : "=g" ((USItype)(sh)), \
276 "=&g" ((USItype)(sl)) \
277 : "0" ((USItype)(ah)), \
278 "g" ((USItype)(bh)), \
279 "1" ((USItype)(al)), \
280 "g" ((USItype)(bl)))
281#define umul_ppmm(ph, pl, m0, m1) \
282 __asm__ ("mulx %3,%0,%1" \
283 : "=g" ((USItype)(ph)), \
284 "=r" ((USItype)(pl)) \
285 : "%0" ((USItype)(m0)), \
286 "g" ((USItype)(m1)))
287#define udiv_qrnnd(q, r, nh, nl, d) \
288 __asm__ ("divx %4,%0,%1" \
289 : "=g" ((USItype)(q)), \
290 "=r" ((USItype)(r)) \
291 : "1" ((USItype)(nh)), \
292 "0" ((USItype)(nl)), \
293 "g" ((USItype)(d)))
294#endif
295
296/***************************************
297 ************** HPPA *****************
298 ***************************************/
299#if defined(__hppa) && W_TYPE_SIZE == 32
300#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
301 __asm__ ("add %4,%5,%1\n" \
302 "addc %2,%3,%0" \
303 : "=r" ((USItype)(sh)), \
304 "=&r" ((USItype)(sl)) \
305 : "%rM" ((USItype)(ah)), \
306 "rM" ((USItype)(bh)), \
307 "%rM" ((USItype)(al)), \
308 "rM" ((USItype)(bl)))
309#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
310 __asm__ ("sub %4,%5,%1\n" \
311 "subb %2,%3,%0" \
312 : "=r" ((USItype)(sh)), \
313 "=&r" ((USItype)(sl)) \
314 : "rM" ((USItype)(ah)), \
315 "rM" ((USItype)(bh)), \
316 "rM" ((USItype)(al)), \
317 "rM" ((USItype)(bl)))
318#if 0 && defined(_PA_RISC1_1)
319/* xmpyu uses floating point register which is not allowed in Linux kernel. */
320#define umul_ppmm(wh, wl, u, v) \
321do { \
322 union {UDItype __ll; \
323 struct {USItype __h, __l; } __i; \
324 } __xx; \
325 __asm__ ("xmpyu %1,%2,%0" \
326 : "=*f" (__xx.__ll) \
327 : "*f" ((USItype)(u)), \
328 "*f" ((USItype)(v))); \
329 (wh) = __xx.__i.__h; \
330 (wl) = __xx.__i.__l; \
331} while (0)
332#define UMUL_TIME 8
333#define UDIV_TIME 60
334#else
335#define UMUL_TIME 40
336#define UDIV_TIME 80
337#endif
338#if 0 /* #ifndef LONGLONG_STANDALONE */
339#define udiv_qrnnd(q, r, n1, n0, d) \
340do { USItype __r; \
341 (q) = __udiv_qrnnd(&__r, (n1), (n0), (d)); \
342 (r) = __r; \
343} while (0)
344extern USItype __udiv_qrnnd();
345#endif /* LONGLONG_STANDALONE */
346#endif /* hppa */
347
348/***************************************
349 ************** I370 *****************
350 ***************************************/
351#if (defined(__i370__) || defined(__mvs__)) && W_TYPE_SIZE == 32
352#define umul_ppmm(xh, xl, m0, m1) \
353do { \
354 union {UDItype __ll; \
355 struct {USItype __h, __l; } __i; \
356 } __xx; \
357 USItype __m0 = (m0), __m1 = (m1); \
358 __asm__ ("mr %0,%3" \
359 : "=r" (__xx.__i.__h), \
360 "=r" (__xx.__i.__l) \
361 : "%1" (__m0), \
362 "r" (__m1)); \
363 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
364 (xh) += ((((SItype) __m0 >> 31) & __m1) \
365 + (((SItype) __m1 >> 31) & __m0)); \
366} while (0)
367#define smul_ppmm(xh, xl, m0, m1) \
368do { \
369 union {DItype __ll; \
370 struct {USItype __h, __l; } __i; \
371 } __xx; \
372 __asm__ ("mr %0,%3" \
373 : "=r" (__xx.__i.__h), \
374 "=r" (__xx.__i.__l) \
375 : "%1" (m0), \
376 "r" (m1)); \
377 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
378} while (0)
379#define sdiv_qrnnd(q, r, n1, n0, d) \
380do { \
381 union {DItype __ll; \
382 struct {USItype __h, __l; } __i; \
383 } __xx; \
384 __xx.__i.__h = n1; __xx.__i.__l = n0; \
385 __asm__ ("dr %0,%2" \
386 : "=r" (__xx.__ll) \
387 : "0" (__xx.__ll), "r" (d)); \
388 (q) = __xx.__i.__l; (r) = __xx.__i.__h; \
389} while (0)
390#endif
391
392/***************************************
393 ************** I386 *****************
394 ***************************************/
395#undef __i386__
396#if (defined(__i386__) || defined(__i486__)) && W_TYPE_SIZE == 32
397#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
398 __asm__ ("addl %5,%1\n" \
399 "adcl %3,%0" \
400 : "=r" (sh), \
401 "=&r" (sl) \
402 : "%0" ((USItype)(ah)), \
403 "g" ((USItype)(bh)), \
404 "%1" ((USItype)(al)), \
405 "g" ((USItype)(bl)))
406#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
407 __asm__ ("subl %5,%1\n" \
408 "sbbl %3,%0" \
409 : "=r" (sh), \
410 "=&r" (sl) \
411 : "0" ((USItype)(ah)), \
412 "g" ((USItype)(bh)), \
413 "1" ((USItype)(al)), \
414 "g" ((USItype)(bl)))
415#define umul_ppmm(w1, w0, u, v) \
416 __asm__ ("mull %3" \
417 : "=a" (w0), \
418 "=d" (w1) \
419 : "%0" ((USItype)(u)), \
420 "rm" ((USItype)(v)))
421#define udiv_qrnnd(q, r, n1, n0, d) \
422 __asm__ ("divl %4" \
423 : "=a" (q), \
424 "=d" (r) \
425 : "0" ((USItype)(n0)), \
426 "1" ((USItype)(n1)), \
427 "rm" ((USItype)(d)))
428#ifndef UMUL_TIME
429#define UMUL_TIME 40
430#endif
431#ifndef UDIV_TIME
432#define UDIV_TIME 40
433#endif
434#endif /* 80x86 */
435
436/***************************************
437 ************** I860 *****************
438 ***************************************/
439#if defined(__i860__) && W_TYPE_SIZE == 32
440#define rshift_rhlc(r, h, l, c) \
441 __asm__ ("shr %3,r0,r0\n" \
442 "shrd %1,%2,%0" \
443 "=r" (r) : "r" (h), "r" (l), "rn" (c))
444#endif /* i860 */
445
446/***************************************
447 ************** I960 *****************
448 ***************************************/
449#if defined(__i960__) && W_TYPE_SIZE == 32
450#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
451 __asm__ ("cmpo 1,0\n" \
452 "addc %5,%4,%1\n" \
453 "addc %3,%2,%0" \
454 : "=r" ((USItype)(sh)), \
455 "=&r" ((USItype)(sl)) \
456 : "%dI" ((USItype)(ah)), \
457 "dI" ((USItype)(bh)), \
458 "%dI" ((USItype)(al)), \
459 "dI" ((USItype)(bl)))
460#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
461 __asm__ ("cmpo 0,0\n" \
462 "subc %5,%4,%1\n" \
463 "subc %3,%2,%0" \
464 : "=r" ((USItype)(sh)), \
465 "=&r" ((USItype)(sl)) \
466 : "dI" ((USItype)(ah)), \
467 "dI" ((USItype)(bh)), \
468 "dI" ((USItype)(al)), \
469 "dI" ((USItype)(bl)))
470#define umul_ppmm(w1, w0, u, v) \
471 ({union {UDItype __ll; \
472 struct {USItype __l, __h; } __i; \
473 } __xx; \
474 __asm__ ("emul %2,%1,%0" \
475 : "=d" (__xx.__ll) \
476 : "%dI" ((USItype)(u)), \
477 "dI" ((USItype)(v))); \
478 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l; })
479#define __umulsidi3(u, v) \
480 ({UDItype __w; \
481 __asm__ ("emul %2,%1,%0" \
482 : "=d" (__w) \
483 : "%dI" ((USItype)(u)), \
484 "dI" ((USItype)(v))); \
485 __w; })
486#define udiv_qrnnd(q, r, nh, nl, d) \
487do { \
488 union {UDItype __ll; \
489 struct {USItype __l, __h; } __i; \
490 } __nn; \
491 __nn.__i.__h = (nh); __nn.__i.__l = (nl); \
492 __asm__ ("ediv %d,%n,%0" \
493 : "=d" (__rq.__ll) \
494 : "dI" (__nn.__ll), \
495 "dI" ((USItype)(d))); \
496 (r) = __rq.__i.__l; (q) = __rq.__i.__h; \
497} while (0)
498#if defined(__i960mx) /* what is the proper symbol to test??? */
499#define rshift_rhlc(r, h, l, c) \
500do { \
501 union {UDItype __ll; \
502 struct {USItype __l, __h; } __i; \
503 } __nn; \
504 __nn.__i.__h = (h); __nn.__i.__l = (l); \
505 __asm__ ("shre %2,%1,%0" \
506 : "=d" (r) : "dI" (__nn.__ll), "dI" (c)); \
507}
508#endif /* i960mx */
509#endif /* i960 */
510
511/***************************************
512 ************** 68000 ****************
513 ***************************************/
514#if (defined(__mc68000__) || defined(__mc68020__) || defined(__NeXT__) || defined(mc68020)) && W_TYPE_SIZE == 32
515#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
516 __asm__ ("add%.l %5,%1\n" \
517 "addx%.l %3,%0" \
518 : "=d" ((USItype)(sh)), \
519 "=&d" ((USItype)(sl)) \
520 : "%0" ((USItype)(ah)), \
521 "d" ((USItype)(bh)), \
522 "%1" ((USItype)(al)), \
523 "g" ((USItype)(bl)))
524#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
525 __asm__ ("sub%.l %5,%1\n" \
526 "subx%.l %3,%0" \
527 : "=d" ((USItype)(sh)), \
528 "=&d" ((USItype)(sl)) \
529 : "0" ((USItype)(ah)), \
530 "d" ((USItype)(bh)), \
531 "1" ((USItype)(al)), \
532 "g" ((USItype)(bl)))
533#if (defined(__mc68020__) || defined(__NeXT__) || defined(mc68020))
534#define umul_ppmm(w1, w0, u, v) \
535 __asm__ ("mulu%.l %3,%1:%0" \
536 : "=d" ((USItype)(w0)), \
537 "=d" ((USItype)(w1)) \
538 : "%0" ((USItype)(u)), \
539 "dmi" ((USItype)(v)))
540#define UMUL_TIME 45
541#define udiv_qrnnd(q, r, n1, n0, d) \
542 __asm__ ("divu%.l %4,%1:%0" \
543 : "=d" ((USItype)(q)), \
544 "=d" ((USItype)(r)) \
545 : "0" ((USItype)(n0)), \
546 "1" ((USItype)(n1)), \
547 "dmi" ((USItype)(d)))
548#define UDIV_TIME 90
549#define sdiv_qrnnd(q, r, n1, n0, d) \
550 __asm__ ("divs%.l %4,%1:%0" \
551 : "=d" ((USItype)(q)), \
552 "=d" ((USItype)(r)) \
553 : "0" ((USItype)(n0)), \
554 "1" ((USItype)(n1)), \
555 "dmi" ((USItype)(d)))
556#else /* not mc68020 */
557#define umul_ppmm(xh, xl, a, b) \
558do { USItype __umul_tmp1, __umul_tmp2; \
559 __asm__ ("| Inlined umul_ppmm\n" \
560 "move%.l %5,%3\n" \
561 "move%.l %2,%0\n" \
562 "move%.w %3,%1\n" \
563 "swap %3\n" \
564 "swap %0\n" \
565 "mulu %2,%1\n" \
566 "mulu %3,%0\n" \
567 "mulu %2,%3\n" \
568 "swap %2\n" \
569 "mulu %5,%2\n" \
570 "add%.l %3,%2\n" \
571 "jcc 1f\n" \
572 "add%.l %#0x10000,%0\n" \
573 "1: move%.l %2,%3\n" \
574 "clr%.w %2\n" \
575 "swap %2\n" \
576 "swap %3\n" \
577 "clr%.w %3\n" \
578 "add%.l %3,%1\n" \
579 "addx%.l %2,%0\n" \
580 "| End inlined umul_ppmm" \
581 : "=&d" ((USItype)(xh)), "=&d" ((USItype)(xl)), \
582 "=d" (__umul_tmp1), "=&d" (__umul_tmp2) \
583 : "%2" ((USItype)(a)), "d" ((USItype)(b))); \
584} while (0)
585#define UMUL_TIME 100
586#define UDIV_TIME 400
587#endif /* not mc68020 */
588#endif /* mc68000 */
589
590/***************************************
591 ************** 88000 ****************
592 ***************************************/
593#if defined(__m88000__) && W_TYPE_SIZE == 32
594#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
595 __asm__ ("addu.co %1,%r4,%r5\n" \
596 "addu.ci %0,%r2,%r3" \
597 : "=r" ((USItype)(sh)), \
598 "=&r" ((USItype)(sl)) \
599 : "%rJ" ((USItype)(ah)), \
600 "rJ" ((USItype)(bh)), \
601 "%rJ" ((USItype)(al)), \
602 "rJ" ((USItype)(bl)))
603#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
604 __asm__ ("subu.co %1,%r4,%r5\n" \
605 "subu.ci %0,%r2,%r3" \
606 : "=r" ((USItype)(sh)), \
607 "=&r" ((USItype)(sl)) \
608 : "rJ" ((USItype)(ah)), \
609 "rJ" ((USItype)(bh)), \
610 "rJ" ((USItype)(al)), \
611 "rJ" ((USItype)(bl)))
612#if defined(__m88110__)
613#define umul_ppmm(wh, wl, u, v) \
614do { \
615 union {UDItype __ll; \
616 struct {USItype __h, __l; } __i; \
617 } __x; \
618 __asm__ ("mulu.d %0,%1,%2" : "=r" (__x.__ll) : "r" (u), "r" (v)); \
619 (wh) = __x.__i.__h; \
620 (wl) = __x.__i.__l; \
621} while (0)
622#define udiv_qrnnd(q, r, n1, n0, d) \
623 ({union {UDItype __ll; \
624 struct {USItype __h, __l; } __i; \
625 } __x, __q; \
626 __x.__i.__h = (n1); __x.__i.__l = (n0); \
627 __asm__ ("divu.d %0,%1,%2" \
628 : "=r" (__q.__ll) : "r" (__x.__ll), "r" (d)); \
629 (r) = (n0) - __q.__l * (d); (q) = __q.__l; })
630#define UMUL_TIME 5
631#define UDIV_TIME 25
632#else
633#define UMUL_TIME 17
634#define UDIV_TIME 150
635#endif /* __m88110__ */
636#endif /* __m88000__ */
637
638/***************************************
639 ************** MIPS *****************
640 ***************************************/
641#if defined(__mips__) && W_TYPE_SIZE == 32
642#define umul_ppmm(w1, w0, u, v) \
643do { \
644 UDItype __ll = (UDItype)(u) * (v); \
645 w1 = __ll >> 32; \
646 w0 = __ll; \
647} while (0)
648#define UMUL_TIME 10
649#define UDIV_TIME 100
650#endif /* __mips__ */
651
652/***************************************
653 ************** MIPS/64 **************
654 ***************************************/
655#if (defined(__mips) && __mips >= 3) && W_TYPE_SIZE == 64
656#if defined(__mips_isa_rev) && __mips_isa_rev >= 6 && defined(CONFIG_CC_IS_GCC)
657/*
658 * GCC ends up emitting a __multi3 intrinsic call for MIPS64r6 with the plain C
659 * code below, so we special case MIPS64r6 until the compiler can do better.
660 */
661#define umul_ppmm(w1, w0, u, v) \
662do { \
663 __asm__ ("dmulu %0,%1,%2" \
664 : "=d" ((UDItype)(w0)) \
665 : "d" ((UDItype)(u)), \
666 "d" ((UDItype)(v))); \
667 __asm__ ("dmuhu %0,%1,%2" \
668 : "=d" ((UDItype)(w1)) \
669 : "d" ((UDItype)(u)), \
670 "d" ((UDItype)(v))); \
671} while (0)
672#else
673#define umul_ppmm(w1, w0, u, v) \
674do { \
675 typedef unsigned int __ll_UTItype __attribute__((mode(TI))); \
676 __ll_UTItype __ll = (__ll_UTItype)(u) * (v); \
677 w1 = __ll >> 64; \
678 w0 = __ll; \
679} while (0)
680#endif
681#define UMUL_TIME 20
682#define UDIV_TIME 140
683#endif /* __mips__ */
684
685/***************************************
686 ************** 32000 ****************
687 ***************************************/
688#if defined(__ns32000__) && W_TYPE_SIZE == 32
689#define umul_ppmm(w1, w0, u, v) \
690 ({union {UDItype __ll; \
691 struct {USItype __l, __h; } __i; \
692 } __xx; \
693 __asm__ ("meid %2,%0" \
694 : "=g" (__xx.__ll) \
695 : "%0" ((USItype)(u)), \
696 "g" ((USItype)(v))); \
697 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l; })
698#define __umulsidi3(u, v) \
699 ({UDItype __w; \
700 __asm__ ("meid %2,%0" \
701 : "=g" (__w) \
702 : "%0" ((USItype)(u)), \
703 "g" ((USItype)(v))); \
704 __w; })
705#define udiv_qrnnd(q, r, n1, n0, d) \
706 ({union {UDItype __ll; \
707 struct {USItype __l, __h; } __i; \
708 } __xx; \
709 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
710 __asm__ ("deid %2,%0" \
711 : "=g" (__xx.__ll) \
712 : "0" (__xx.__ll), \
713 "g" ((USItype)(d))); \
714 (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
715#endif /* __ns32000__ */
716
717/***************************************
718 ************** PPC ******************
719 ***************************************/
720#if (defined(_ARCH_PPC) || defined(_IBMR2)) && W_TYPE_SIZE == 32
721#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
722do { \
723 if (__builtin_constant_p(bh) && (bh) == 0) \
724 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
725 : "=r" (sh), \
726 "=&r" (sl) \
727 : "%r" ((USItype)(ah)), \
728 "%r" ((USItype)(al)), \
729 "rI" ((USItype)(bl))); \
730 else if (__builtin_constant_p(bh) && (bh) == ~(USItype) 0) \
731 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
732 : "=r" (sh), \
733 "=&r" (sl) \
734 : "%r" ((USItype)(ah)), \
735 "%r" ((USItype)(al)), \
736 "rI" ((USItype)(bl))); \
737 else \
738 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
739 : "=r" (sh), \
740 "=&r" (sl) \
741 : "%r" ((USItype)(ah)), \
742 "r" ((USItype)(bh)), \
743 "%r" ((USItype)(al)), \
744 "rI" ((USItype)(bl))); \
745} while (0)
746#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
747do { \
748 if (__builtin_constant_p(ah) && (ah) == 0) \
749 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
750 : "=r" (sh), \
751 "=&r" (sl) \
752 : "r" ((USItype)(bh)), \
753 "rI" ((USItype)(al)), \
754 "r" ((USItype)(bl))); \
755 else if (__builtin_constant_p(ah) && (ah) == ~(USItype) 0) \
756 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
757 : "=r" (sh), \
758 "=&r" (sl) \
759 : "r" ((USItype)(bh)), \
760 "rI" ((USItype)(al)), \
761 "r" ((USItype)(bl))); \
762 else if (__builtin_constant_p(bh) && (bh) == 0) \
763 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
764 : "=r" (sh), \
765 "=&r" (sl) \
766 : "r" ((USItype)(ah)), \
767 "rI" ((USItype)(al)), \
768 "r" ((USItype)(bl))); \
769 else if (__builtin_constant_p(bh) && (bh) == ~(USItype) 0) \
770 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
771 : "=r" (sh), \
772 "=&r" (sl) \
773 : "r" ((USItype)(ah)), \
774 "rI" ((USItype)(al)), \
775 "r" ((USItype)(bl))); \
776 else \
777 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
778 : "=r" (sh), \
779 "=&r" (sl) \
780 : "r" ((USItype)(ah)), \
781 "r" ((USItype)(bh)), \
782 "rI" ((USItype)(al)), \
783 "r" ((USItype)(bl))); \
784} while (0)
785#if defined(_ARCH_PPC)
786#define umul_ppmm(ph, pl, m0, m1) \
787do { \
788 USItype __m0 = (m0), __m1 = (m1); \
789 __asm__ ("mulhwu %0,%1,%2" \
790 : "=r" (ph) \
791 : "%r" (__m0), \
792 "r" (__m1)); \
793 (pl) = __m0 * __m1; \
794} while (0)
795#define UMUL_TIME 15
796#define smul_ppmm(ph, pl, m0, m1) \
797do { \
798 SItype __m0 = (m0), __m1 = (m1); \
799 __asm__ ("mulhw %0,%1,%2" \
800 : "=r" ((SItype) ph) \
801 : "%r" (__m0), \
802 "r" (__m1)); \
803 (pl) = __m0 * __m1; \
804} while (0)
805#define SMUL_TIME 14
806#define UDIV_TIME 120
807#else
808#define umul_ppmm(xh, xl, m0, m1) \
809do { \
810 USItype __m0 = (m0), __m1 = (m1); \
811 __asm__ ("mul %0,%2,%3" \
812 : "=r" ((USItype)(xh)), \
813 "=q" ((USItype)(xl)) \
814 : "r" (__m0), \
815 "r" (__m1)); \
816 (xh) += ((((SItype) __m0 >> 31) & __m1) \
817 + (((SItype) __m1 >> 31) & __m0)); \
818} while (0)
819#define UMUL_TIME 8
820#define smul_ppmm(xh, xl, m0, m1) \
821 __asm__ ("mul %0,%2,%3" \
822 : "=r" ((SItype)(xh)), \
823 "=q" ((SItype)(xl)) \
824 : "r" (m0), \
825 "r" (m1))
826#define SMUL_TIME 4
827#define sdiv_qrnnd(q, r, nh, nl, d) \
828 __asm__ ("div %0,%2,%4" \
829 : "=r" ((SItype)(q)), "=q" ((SItype)(r)) \
830 : "r" ((SItype)(nh)), "1" ((SItype)(nl)), "r" ((SItype)(d)))
831#define UDIV_TIME 100
832#endif
833#endif /* Power architecture variants. */
834
835/***************************************
836 ************** PYR ******************
837 ***************************************/
838#if defined(__pyr__) && W_TYPE_SIZE == 32
839#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
840 __asm__ ("addw %5,%1\n" \
841 "addwc %3,%0" \
842 : "=r" ((USItype)(sh)), \
843 "=&r" ((USItype)(sl)) \
844 : "%0" ((USItype)(ah)), \
845 "g" ((USItype)(bh)), \
846 "%1" ((USItype)(al)), \
847 "g" ((USItype)(bl)))
848#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
849 __asm__ ("subw %5,%1\n" \
850 "subwb %3,%0" \
851 : "=r" ((USItype)(sh)), \
852 "=&r" ((USItype)(sl)) \
853 : "0" ((USItype)(ah)), \
854 "g" ((USItype)(bh)), \
855 "1" ((USItype)(al)), \
856 "g" ((USItype)(bl)))
857 /* This insn works on Pyramids with AP, XP, or MI CPUs, but not with SP. */
858#define umul_ppmm(w1, w0, u, v) \
859 ({union {UDItype __ll; \
860 struct {USItype __h, __l; } __i; \
861 } __xx; \
862 __asm__ ("movw %1,%R0\n" \
863 "uemul %2,%0" \
864 : "=&r" (__xx.__ll) \
865 : "g" ((USItype) (u)), \
866 "g" ((USItype)(v))); \
867 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l; })
868#endif /* __pyr__ */
869
870/***************************************
871 ************** RT/ROMP **************
872 ***************************************/
873#if defined(__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
874#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
875 __asm__ ("a %1,%5\n" \
876 "ae %0,%3" \
877 : "=r" ((USItype)(sh)), \
878 "=&r" ((USItype)(sl)) \
879 : "%0" ((USItype)(ah)), \
880 "r" ((USItype)(bh)), \
881 "%1" ((USItype)(al)), \
882 "r" ((USItype)(bl)))
883#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
884 __asm__ ("s %1,%5\n" \
885 "se %0,%3" \
886 : "=r" ((USItype)(sh)), \
887 "=&r" ((USItype)(sl)) \
888 : "0" ((USItype)(ah)), \
889 "r" ((USItype)(bh)), \
890 "1" ((USItype)(al)), \
891 "r" ((USItype)(bl)))
892#define umul_ppmm(ph, pl, m0, m1) \
893do { \
894 USItype __m0 = (m0), __m1 = (m1); \
895 __asm__ ( \
896 "s r2,r2\n" \
897 "mts r10,%2\n" \
898 "m r2,%3\n" \
899 "m r2,%3\n" \
900 "m r2,%3\n" \
901 "m r2,%3\n" \
902 "m r2,%3\n" \
903 "m r2,%3\n" \
904 "m r2,%3\n" \
905 "m r2,%3\n" \
906 "m r2,%3\n" \
907 "m r2,%3\n" \
908 "m r2,%3\n" \
909 "m r2,%3\n" \
910 "m r2,%3\n" \
911 "m r2,%3\n" \
912 "m r2,%3\n" \
913 "m r2,%3\n" \
914 "cas %0,r2,r0\n" \
915 "mfs r10,%1" \
916 : "=r" ((USItype)(ph)), \
917 "=r" ((USItype)(pl)) \
918 : "%r" (__m0), \
919 "r" (__m1) \
920 : "r2"); \
921 (ph) += ((((SItype) __m0 >> 31) & __m1) \
922 + (((SItype) __m1 >> 31) & __m0)); \
923} while (0)
924#define UMUL_TIME 20
925#define UDIV_TIME 200
926#endif /* RT/ROMP */
927
928/***************************************
929 ************** SH2 ******************
930 ***************************************/
931#if (defined(__sh2__) || defined(__sh3__) || defined(__SH4__)) \
932 && W_TYPE_SIZE == 32
933#define umul_ppmm(w1, w0, u, v) \
934 __asm__ ( \
935 "dmulu.l %2,%3\n" \
936 "sts macl,%1\n" \
937 "sts mach,%0" \
938 : "=r" ((USItype)(w1)), \
939 "=r" ((USItype)(w0)) \
940 : "r" ((USItype)(u)), \
941 "r" ((USItype)(v)) \
942 : "macl", "mach")
943#define UMUL_TIME 5
944#endif
945
946/***************************************
947 ************** SPARC ****************
948 ***************************************/
949#if defined(__sparc__) && W_TYPE_SIZE == 32
950#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
951 __asm__ ("addcc %r4,%5,%1\n" \
952 "addx %r2,%3,%0" \
953 : "=r" ((USItype)(sh)), \
954 "=&r" ((USItype)(sl)) \
955 : "%rJ" ((USItype)(ah)), \
956 "rI" ((USItype)(bh)), \
957 "%rJ" ((USItype)(al)), \
958 "rI" ((USItype)(bl)) \
959 __CLOBBER_CC)
960#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
961 __asm__ ("subcc %r4,%5,%1\n" \
962 "subx %r2,%3,%0" \
963 : "=r" ((USItype)(sh)), \
964 "=&r" ((USItype)(sl)) \
965 : "rJ" ((USItype)(ah)), \
966 "rI" ((USItype)(bh)), \
967 "rJ" ((USItype)(al)), \
968 "rI" ((USItype)(bl)) \
969 __CLOBBER_CC)
970#if defined(__sparc_v8__)
971/* Don't match immediate range because, 1) it is not often useful,
972 2) the 'I' flag thinks of the range as a 13 bit signed interval,
973 while we want to match a 13 bit interval, sign extended to 32 bits,
974 but INTERPRETED AS UNSIGNED. */
975#define umul_ppmm(w1, w0, u, v) \
976 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
977 : "=r" ((USItype)(w1)), \
978 "=r" ((USItype)(w0)) \
979 : "r" ((USItype)(u)), \
980 "r" ((USItype)(v)))
981#define UMUL_TIME 5
982#ifndef SUPERSPARC /* SuperSPARC's udiv only handles 53 bit dividends */
983#define udiv_qrnnd(q, r, n1, n0, d) \
984do { \
985 USItype __q; \
986 __asm__ ("mov %1,%%y;nop;nop;nop;udiv %2,%3,%0" \
987 : "=r" ((USItype)(__q)) \
988 : "r" ((USItype)(n1)), \
989 "r" ((USItype)(n0)), \
990 "r" ((USItype)(d))); \
991 (r) = (n0) - __q * (d); \
992 (q) = __q; \
993} while (0)
994#define UDIV_TIME 25
995#endif /* SUPERSPARC */
996#else /* ! __sparc_v8__ */
997#if defined(__sparclite__)
998/* This has hardware multiply but not divide. It also has two additional
999 instructions scan (ffs from high bit) and divscc. */
1000#define umul_ppmm(w1, w0, u, v) \
1001 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1002 : "=r" ((USItype)(w1)), \
1003 "=r" ((USItype)(w0)) \
1004 : "r" ((USItype)(u)), \
1005 "r" ((USItype)(v)))
1006#define UMUL_TIME 5
1007#define udiv_qrnnd(q, r, n1, n0, d) \
1008 __asm__ ("! Inlined udiv_qrnnd\n" \
1009 "wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
1010 "tst %%g0\n" \
1011 "divscc %3,%4,%%g1\n" \
1012 "divscc %%g1,%4,%%g1\n" \
1013 "divscc %%g1,%4,%%g1\n" \
1014 "divscc %%g1,%4,%%g1\n" \
1015 "divscc %%g1,%4,%%g1\n" \
1016 "divscc %%g1,%4,%%g1\n" \
1017 "divscc %%g1,%4,%%g1\n" \
1018 "divscc %%g1,%4,%%g1\n" \
1019 "divscc %%g1,%4,%%g1\n" \
1020 "divscc %%g1,%4,%%g1\n" \
1021 "divscc %%g1,%4,%%g1\n" \
1022 "divscc %%g1,%4,%%g1\n" \
1023 "divscc %%g1,%4,%%g1\n" \
1024 "divscc %%g1,%4,%%g1\n" \
1025 "divscc %%g1,%4,%%g1\n" \
1026 "divscc %%g1,%4,%%g1\n" \
1027 "divscc %%g1,%4,%%g1\n" \
1028 "divscc %%g1,%4,%%g1\n" \
1029 "divscc %%g1,%4,%%g1\n" \
1030 "divscc %%g1,%4,%%g1\n" \
1031 "divscc %%g1,%4,%%g1\n" \
1032 "divscc %%g1,%4,%%g1\n" \
1033 "divscc %%g1,%4,%%g1\n" \
1034 "divscc %%g1,%4,%%g1\n" \
1035 "divscc %%g1,%4,%%g1\n" \
1036 "divscc %%g1,%4,%%g1\n" \
1037 "divscc %%g1,%4,%%g1\n" \
1038 "divscc %%g1,%4,%%g1\n" \
1039 "divscc %%g1,%4,%%g1\n" \
1040 "divscc %%g1,%4,%%g1\n" \
1041 "divscc %%g1,%4,%%g1\n" \
1042 "divscc %%g1,%4,%0\n" \
1043 "rd %%y,%1\n" \
1044 "bl,a 1f\n" \
1045 "add %1,%4,%1\n" \
1046 "1: ! End of inline udiv_qrnnd" \
1047 : "=r" ((USItype)(q)), \
1048 "=r" ((USItype)(r)) \
1049 : "r" ((USItype)(n1)), \
1050 "r" ((USItype)(n0)), \
1051 "rI" ((USItype)(d)) \
1052 : "%g1" __AND_CLOBBER_CC)
1053#define UDIV_TIME 37
1054#endif /* __sparclite__ */
1055#endif /* __sparc_v8__ */
1056 /* Default to sparc v7 versions of umul_ppmm and udiv_qrnnd. */
1057#ifndef umul_ppmm
1058#define umul_ppmm(w1, w0, u, v) \
1059 __asm__ ("! Inlined umul_ppmm\n" \
1060 "wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n" \
1061 "sra %3,31,%%g2 ! Don't move this insn\n" \
1062 "and %2,%%g2,%%g2 ! Don't move this insn\n" \
1063 "andcc %%g0,0,%%g1 ! Don't move this insn\n" \
1064 "mulscc %%g1,%3,%%g1\n" \
1065 "mulscc %%g1,%3,%%g1\n" \
1066 "mulscc %%g1,%3,%%g1\n" \
1067 "mulscc %%g1,%3,%%g1\n" \
1068 "mulscc %%g1,%3,%%g1\n" \
1069 "mulscc %%g1,%3,%%g1\n" \
1070 "mulscc %%g1,%3,%%g1\n" \
1071 "mulscc %%g1,%3,%%g1\n" \
1072 "mulscc %%g1,%3,%%g1\n" \
1073 "mulscc %%g1,%3,%%g1\n" \
1074 "mulscc %%g1,%3,%%g1\n" \
1075 "mulscc %%g1,%3,%%g1\n" \
1076 "mulscc %%g1,%3,%%g1\n" \
1077 "mulscc %%g1,%3,%%g1\n" \
1078 "mulscc %%g1,%3,%%g1\n" \
1079 "mulscc %%g1,%3,%%g1\n" \
1080 "mulscc %%g1,%3,%%g1\n" \
1081 "mulscc %%g1,%3,%%g1\n" \
1082 "mulscc %%g1,%3,%%g1\n" \
1083 "mulscc %%g1,%3,%%g1\n" \
1084 "mulscc %%g1,%3,%%g1\n" \
1085 "mulscc %%g1,%3,%%g1\n" \
1086 "mulscc %%g1,%3,%%g1\n" \
1087 "mulscc %%g1,%3,%%g1\n" \
1088 "mulscc %%g1,%3,%%g1\n" \
1089 "mulscc %%g1,%3,%%g1\n" \
1090 "mulscc %%g1,%3,%%g1\n" \
1091 "mulscc %%g1,%3,%%g1\n" \
1092 "mulscc %%g1,%3,%%g1\n" \
1093 "mulscc %%g1,%3,%%g1\n" \
1094 "mulscc %%g1,%3,%%g1\n" \
1095 "mulscc %%g1,%3,%%g1\n" \
1096 "mulscc %%g1,0,%%g1\n" \
1097 "add %%g1,%%g2,%0\n" \
1098 "rd %%y,%1" \
1099 : "=r" ((USItype)(w1)), \
1100 "=r" ((USItype)(w0)) \
1101 : "%rI" ((USItype)(u)), \
1102 "r" ((USItype)(v)) \
1103 : "%g1", "%g2" __AND_CLOBBER_CC)
1104#define UMUL_TIME 39 /* 39 instructions */
1105/* It's quite necessary to add this much assembler for the sparc.
1106 The default udiv_qrnnd (in C) is more than 10 times slower! */
1107#define udiv_qrnnd(q, r, n1, n0, d) \
1108 __asm__ ("! Inlined udiv_qrnnd\n\t" \
1109 "mov 32,%%g1\n\t" \
1110 "subcc %1,%2,%%g0\n\t" \
1111 "1: bcs 5f\n\t" \
1112 "addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n\t" \
1113 "sub %1,%2,%1 ! this kills msb of n\n\t" \
1114 "addx %1,%1,%1 ! so this can't give carry\n\t" \
1115 "subcc %%g1,1,%%g1\n\t" \
1116 "2: bne 1b\n\t" \
1117 "subcc %1,%2,%%g0\n\t" \
1118 "bcs 3f\n\t" \
1119 "addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n\t" \
1120 "b 3f\n\t" \
1121 "sub %1,%2,%1 ! this kills msb of n\n\t" \
1122 "4: sub %1,%2,%1\n\t" \
1123 "5: addxcc %1,%1,%1\n\t" \
1124 "bcc 2b\n\t" \
1125 "subcc %%g1,1,%%g1\n\t" \
1126 "! Got carry from n. Subtract next step to cancel this carry.\n\t" \
1127 "bne 4b\n\t" \
1128 "addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n\t" \
1129 "sub %1,%2,%1\n\t" \
1130 "3: xnor %0,0,%0\n\t" \
1131 "! End of inline udiv_qrnnd\n" \
1132 : "=&r" ((USItype)(q)), \
1133 "=&r" ((USItype)(r)) \
1134 : "r" ((USItype)(d)), \
1135 "1" ((USItype)(n1)), \
1136 "0" ((USItype)(n0)) : "%g1", "cc")
1137#define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1138#endif
1139#endif /* __sparc__ */
1140
1141/***************************************
1142 ************** VAX ******************
1143 ***************************************/
1144#if defined(__vax__) && W_TYPE_SIZE == 32
1145#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1146 __asm__ ("addl2 %5,%1\n" \
1147 "adwc %3,%0" \
1148 : "=g" ((USItype)(sh)), \
1149 "=&g" ((USItype)(sl)) \
1150 : "%0" ((USItype)(ah)), \
1151 "g" ((USItype)(bh)), \
1152 "%1" ((USItype)(al)), \
1153 "g" ((USItype)(bl)))
1154#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1155 __asm__ ("subl2 %5,%1\n" \
1156 "sbwc %3,%0" \
1157 : "=g" ((USItype)(sh)), \
1158 "=&g" ((USItype)(sl)) \
1159 : "0" ((USItype)(ah)), \
1160 "g" ((USItype)(bh)), \
1161 "1" ((USItype)(al)), \
1162 "g" ((USItype)(bl)))
1163#define umul_ppmm(xh, xl, m0, m1) \
1164do { \
1165 union {UDItype __ll; \
1166 struct {USItype __l, __h; } __i; \
1167 } __xx; \
1168 USItype __m0 = (m0), __m1 = (m1); \
1169 __asm__ ("emul %1,%2,$0,%0" \
1170 : "=g" (__xx.__ll) \
1171 : "g" (__m0), \
1172 "g" (__m1)); \
1173 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1174 (xh) += ((((SItype) __m0 >> 31) & __m1) \
1175 + (((SItype) __m1 >> 31) & __m0)); \
1176} while (0)
1177#define sdiv_qrnnd(q, r, n1, n0, d) \
1178do { \
1179 union {DItype __ll; \
1180 struct {SItype __l, __h; } __i; \
1181 } __xx; \
1182 __xx.__i.__h = n1; __xx.__i.__l = n0; \
1183 __asm__ ("ediv %3,%2,%0,%1" \
1184 : "=g" (q), "=g" (r) \
1185 : "g" (__xx.__ll), "g" (d)); \
1186} while (0)
1187#endif /* __vax__ */
1188
1189/***************************************
1190 ************** Z8000 ****************
1191 ***************************************/
1192#if defined(__z8000__) && W_TYPE_SIZE == 16
1193#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1194 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1195 : "=r" ((unsigned int)(sh)), \
1196 "=&r" ((unsigned int)(sl)) \
1197 : "%0" ((unsigned int)(ah)), \
1198 "r" ((unsigned int)(bh)), \
1199 "%1" ((unsigned int)(al)), \
1200 "rQR" ((unsigned int)(bl)))
1201#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1202 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1203 : "=r" ((unsigned int)(sh)), \
1204 "=&r" ((unsigned int)(sl)) \
1205 : "0" ((unsigned int)(ah)), \
1206 "r" ((unsigned int)(bh)), \
1207 "1" ((unsigned int)(al)), \
1208 "rQR" ((unsigned int)(bl)))
1209#define umul_ppmm(xh, xl, m0, m1) \
1210do { \
1211 union {long int __ll; \
1212 struct {unsigned int __h, __l; } __i; \
1213 } __xx; \
1214 unsigned int __m0 = (m0), __m1 = (m1); \
1215 __asm__ ("mult %S0,%H3" \
1216 : "=r" (__xx.__i.__h), \
1217 "=r" (__xx.__i.__l) \
1218 : "%1" (__m0), \
1219 "rQR" (__m1)); \
1220 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1221 (xh) += ((((signed int) __m0 >> 15) & __m1) \
1222 + (((signed int) __m1 >> 15) & __m0)); \
1223} while (0)
1224#endif /* __z8000__ */
1225
1226#endif /* __GNUC__ */
1227
1228/***************************************
1229 *********** Generic Versions ********
1230 ***************************************/
1231#if !defined(umul_ppmm) && defined(__umulsidi3)
1232#define umul_ppmm(ph, pl, m0, m1) \
1233{ \
1234 UDWtype __ll = __umulsidi3(m0, m1); \
1235 ph = (UWtype) (__ll >> W_TYPE_SIZE); \
1236 pl = (UWtype) __ll; \
1237}
1238#endif
1239
1240#if !defined(__umulsidi3)
1241#define __umulsidi3(u, v) \
1242 ({UWtype __hi, __lo; \
1243 umul_ppmm(__hi, __lo, u, v); \
1244 ((UDWtype) __hi << W_TYPE_SIZE) | __lo; })
1245#endif
1246
1247 /* If this machine has no inline assembler, use C macros. */
1248
1249#if !defined(add_ssaaaa)
1250#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1251do { \
1252 UWtype __x; \
1253 __x = (al) + (bl); \
1254 (sh) = (ah) + (bh) + (__x < (al)); \
1255 (sl) = __x; \
1256} while (0)
1257#endif
1258
1259#if !defined(sub_ddmmss)
1260#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1261do { \
1262 UWtype __x; \
1263 __x = (al) - (bl); \
1264 (sh) = (ah) - (bh) - (__x > (al)); \
1265 (sl) = __x; \
1266} while (0)
1267#endif
1268
1269#if !defined(umul_ppmm)
1270#define umul_ppmm(w1, w0, u, v) \
1271do { \
1272 UWtype __x0, __x1, __x2, __x3; \
1273 UHWtype __ul, __vl, __uh, __vh; \
1274 UWtype __u = (u), __v = (v); \
1275 \
1276 __ul = __ll_lowpart(__u); \
1277 __uh = __ll_highpart(__u); \
1278 __vl = __ll_lowpart(__v); \
1279 __vh = __ll_highpart(__v); \
1280 \
1281 __x0 = (UWtype) __ul * __vl; \
1282 __x1 = (UWtype) __ul * __vh; \
1283 __x2 = (UWtype) __uh * __vl; \
1284 __x3 = (UWtype) __uh * __vh; \
1285 \
1286 __x1 += __ll_highpart(__x0);/* this can't give carry */ \
1287 __x1 += __x2; /* but this indeed can */ \
1288 if (__x1 < __x2) /* did we get it? */ \
1289 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1290 \
1291 (w1) = __x3 + __ll_highpart(__x1); \
1292 (w0) = (__ll_lowpart(__x1) << W_TYPE_SIZE/2) + __ll_lowpart(__x0); \
1293} while (0)
1294#endif
1295
1296#if !defined(umul_ppmm)
1297#define smul_ppmm(w1, w0, u, v) \
1298do { \
1299 UWtype __w1; \
1300 UWtype __m0 = (u), __m1 = (v); \
1301 umul_ppmm(__w1, w0, __m0, __m1); \
1302 (w1) = __w1 - (-(__m0 >> (W_TYPE_SIZE - 1)) & __m1) \
1303 - (-(__m1 >> (W_TYPE_SIZE - 1)) & __m0); \
1304} while (0)
1305#endif
1306
1307 /* Define this unconditionally, so it can be used for debugging. */
1308#define __udiv_qrnnd_c(q, r, n1, n0, d) \
1309do { \
1310 UWtype __d1, __d0, __q1, __q0, __r1, __r0, __m; \
1311 __d1 = __ll_highpart(d); \
1312 __d0 = __ll_lowpart(d); \
1313 \
1314 __r1 = (n1) % __d1; \
1315 __q1 = (n1) / __d1; \
1316 __m = (UWtype) __q1 * __d0; \
1317 __r1 = __r1 * __ll_B | __ll_highpart(n0); \
1318 if (__r1 < __m) { \
1319 __q1--, __r1 += (d); \
1320 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */ \
1321 if (__r1 < __m) \
1322 __q1--, __r1 += (d); \
1323 } \
1324 __r1 -= __m; \
1325 \
1326 __r0 = __r1 % __d1; \
1327 __q0 = __r1 / __d1; \
1328 __m = (UWtype) __q0 * __d0; \
1329 __r0 = __r0 * __ll_B | __ll_lowpart(n0); \
1330 if (__r0 < __m) { \
1331 __q0--, __r0 += (d); \
1332 if (__r0 >= (d)) \
1333 if (__r0 < __m) \
1334 __q0--, __r0 += (d); \
1335 } \
1336 __r0 -= __m; \
1337 \
1338 (q) = (UWtype) __q1 * __ll_B | __q0; \
1339 (r) = __r0; \
1340} while (0)
1341
1342/* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1343 __udiv_w_sdiv (defined in libgcc or elsewhere). */
1344#if !defined(udiv_qrnnd) && defined(sdiv_qrnnd)
1345#define udiv_qrnnd(q, r, nh, nl, d) \
1346do { \
1347 UWtype __r; \
1348 (q) = __MPN(udiv_w_sdiv) (&__r, nh, nl, d); \
1349 (r) = __r; \
1350} while (0)
1351#endif
1352
1353 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1354#if !defined(udiv_qrnnd)
1355#define UDIV_NEEDS_NORMALIZATION 1
1356#define udiv_qrnnd __udiv_qrnnd_c
1357#endif
1358
1359#ifndef UDIV_NEEDS_NORMALIZATION
1360#define UDIV_NEEDS_NORMALIZATION 0
1361#endif
1362

source code of linux/lib/crypto/mpi/longlong.h