1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2 Copyright (C) 1991, 1992, 1993 Free Software Foundation, Inc.
4 This definition file is free software; you can redistribute it
5 and/or modify it under the terms of the GNU General Public
6 License as published by the Free Software Foundation; either
7 version 2, or (at your option) any later version.
9 This definition file is distributed in the hope that it will be
10 useful, but WITHOUT ANY WARRANTY; without even the implied
11 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 See the GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
18 #ifndef LONG_TYPE_SIZE
19 #ifdef BITS_PER_LONGINT
20 #define LONG_TYPE_SIZE BITS_PER_LONGINT
22 #define LONG_TYPE_SIZE 32
26 #define __BITS4 (LONG_TYPE_SIZE / 4)
27 #define __ll_B (1L << (LONG_TYPE_SIZE / 2))
28 #define __ll_lowpart(t) ((unsigned long int) (t) % __ll_B)
29 #define __ll_highpart(t) ((unsigned long int) (t) / __ll_B)
31 /* Define auxiliary asm macros.
33 1) umul_ppmm(high_prod, low_prod, multipler, multiplicand)
34 multiplies two unsigned long integers MULTIPLER and MULTIPLICAND,
35 and generates a two unsigned word product in HIGH_PROD and
38 2) __umulsidi3(a,b) multiplies two unsigned long integers A and B,
39 and returns a long long product. This is just a variant of umul_ppmm.
41 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
42 denominator) divides a two-word unsigned integer, composed by the
43 integers HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and
44 places the quotient in QUOTIENT and the remainder in REMAINDER.
45 HIGH_NUMERATOR must be less than DENOMINATOR for correct operation.
46 If, in addition, the most significant bit of DENOMINATOR must be 1,
47 then the pre-processor symbol UDIV_NEEDS_NORMALIZATION is defined to 1.
49 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
50 denominator). Like udiv_qrnnd but the numbers are signed. The
51 quotient is rounded towards 0.
53 5) count_leading_zeros(count, x) counts the number of zero-bits from
54 the msb to the first non-zero bit. This is the number of steps X
55 needs to be shifted left to set the msb. Undefined for X == 0.
57 6) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
58 high_addend_2, low_addend_2) adds two two-word unsigned integers,
59 composed by HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and
60 LOW_ADDEND_2 respectively. The result is placed in HIGH_SUM and
61 LOW_SUM. Overflow (i.e. carry out) is not stored anywhere, and is
64 7) sub_ddmmss(high_difference, low_difference, high_minuend,
65 low_minuend, high_subtrahend, low_subtrahend) subtracts two
66 two-word unsigned integers, composed by HIGH_MINUEND_1 and
67 LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and LOW_SUBTRAHEND_2
68 respectively. The result is placed in HIGH_DIFFERENCE and
69 LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
72 If any of these macros are left undefined for a particular CPU,
75 /* The CPUs come in alphabetical order below.
77 Please add support for more CPUs here, or improve the current support
78 for the CPUs below! */
80 #if defined (__GNUC__) && !defined (NO_ASM)
82 /* We sometimes need to clobber "cc" with gcc2, but that would not be
83 understood by gcc1. Use cpp to avoid major code duplication. */
86 #define __AND_CLOBBER_CC
87 #else /* __GNUC__ >= 2 */
88 #define __CLOBBER_CC : "cc"
89 #define __AND_CLOBBER_CC , "cc"
90 #endif /* __GNUC__ < 2 */
92 #if defined (__a29k__) || defined (___AM29K__)
93 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
94 __asm__ ("add %1,%4,%5
96 : "=r" ((unsigned long int)(sh)), \
97 "=&r" ((unsigned long int)(sl)) \
98 : "%r" ((unsigned long int)(ah)), \
99 "rI" ((unsigned long int)(bh)), \
100 "%r" ((unsigned long int)(al)), \
101 "rI" ((unsigned long int)(bl)))
102 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
103 __asm__ ("sub %1,%4,%5
105 : "=r" ((unsigned long int)(sh)), \
106 "=&r" ((unsigned long int)(sl)) \
107 : "r" ((unsigned long int)(ah)), \
108 "rI" ((unsigned long int)(bh)), \
109 "r" ((unsigned long int)(al)), \
110 "rI" ((unsigned long int)(bl)))
111 #define umul_ppmm(xh, xl, m0, m1) \
113 unsigned long int __m0 = (m0), __m1 = (m1); \
114 __asm__ ("multiplu %0,%1,%2" \
115 : "=r" ((unsigned long int)(xl)) \
118 __asm__ ("multmu %0,%1,%2" \
119 : "=r" ((unsigned long int)(xh)) \
123 #define udiv_qrnnd(q, r, n1, n0, d) \
124 __asm__ ("dividu %0,%3,%4" \
125 : "=r" ((unsigned long int)(q)), \
126 "=q" ((unsigned long int)(r)) \
127 : "1" ((unsigned long int)(n1)), \
128 "r" ((unsigned long int)(n0)), \
129 "r" ((unsigned long int)(d)))
130 #define count_leading_zeros(count, x) \
131 __asm__ ("clz %0,%1" \
132 : "=r" ((unsigned long int)(count)) \
133 : "r" ((unsigned long int)(x)))
134 #endif /* __a29k__ */
136 #if defined (__alpha__)
137 #define umul_ppmm(ph, pl, m0, m1) \
139 unsigned long int __m0 = (m0), __m1 = (m1); \
140 __asm__ ("umulh %r1,%2,%0" \
141 : "=r" ((unsigned long int) ph) \
144 (pl) = (unsigned long int) (__m0) * (unsigned long int) (__m1); \
147 #define UDIV_TIME 500
150 #if defined (__arm__)
151 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
152 __asm__ ("adds %1,%4,%5
154 : "=r" ((unsigned long int)(sh)), \
155 "=&r" ((unsigned long int)(sl)) \
156 : "%r" ((unsigned long int)(ah)), \
157 "rI" ((unsigned long int)(bh)), \
158 "%r" ((unsigned long int)(al)), \
159 "rI" ((unsigned long int)(bl)))
160 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
161 __asm__ ("subs %1,%4,%5
163 : "=r" ((unsigned long int)(sh)), \
164 "=&r" ((unsigned long int)(sl)) \
165 : "r" ((unsigned long int)(ah)), \
166 "rI" ((unsigned long int)(bh)), \
167 "r" ((unsigned long int)(al)), \
168 "rI" ((unsigned long int)(bl)))
171 #if defined (__gmicro__)
172 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
173 __asm__ ("add.w %5,%1
175 : "=g" ((unsigned long int)(sh)), \
176 "=&g" ((unsigned long int)(sl)) \
177 : "%0" ((unsigned long int)(ah)), \
178 "g" ((unsigned long int)(bh)), \
179 "%1" ((unsigned long int)(al)), \
180 "g" ((unsigned long int)(bl)))
181 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
182 __asm__ ("sub.w %5,%1
184 : "=g" ((unsigned long int)(sh)), \
185 "=&g" ((unsigned long int)(sl)) \
186 : "0" ((unsigned long int)(ah)), \
187 "g" ((unsigned long int)(bh)), \
188 "1" ((unsigned long int)(al)), \
189 "g" ((unsigned long int)(bl)))
190 #define umul_ppmm(ph, pl, m0, m1) \
191 __asm__ ("mulx %3,%0,%1" \
192 : "=g" ((unsigned long int)(ph)), \
193 "=r" ((unsigned long int)(pl)) \
194 : "%0" ((unsigned long int)(m0)), \
195 "g" ((unsigned long int)(m1)))
196 #define udiv_qrnnd(q, r, nh, nl, d) \
197 __asm__ ("divx %4,%0,%1" \
198 : "=g" ((unsigned long int)(q)), \
199 "=r" ((unsigned long int)(r)) \
200 : "1" ((unsigned long int)(nh)), \
201 "0" ((unsigned long int)(nl)), \
202 "g" ((unsigned long int)(d)))
203 #define count_leading_zeros(count, x) \
204 __asm__ ("bsch/1 %1,%0" \
206 : "g" ((unsigned long int)(x)), \
211 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
212 __asm__ ("add %4,%5,%1
214 : "=r" ((unsigned long int)(sh)), \
215 "=&r" ((unsigned long int)(sl)) \
216 : "%rM" ((unsigned long int)(ah)), \
217 "rM" ((unsigned long int)(bh)), \
218 "%rM" ((unsigned long int)(al)), \
219 "rM" ((unsigned long int)(bl)))
220 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
221 __asm__ ("sub %4,%5,%1
223 : "=r" ((unsigned long int)(sh)), \
224 "=&r" ((unsigned long int)(sl)) \
225 : "rM" ((unsigned long int)(ah)), \
226 "rM" ((unsigned long int)(bh)), \
227 "rM" ((unsigned long int)(al)), \
228 "rM" ((unsigned long int)(bl)))
229 #if defined (_PA_RISC1_1)
230 #define umul_ppmm(wh, wl, u, v) \
232 union {long long int __ll; \
233 struct {unsigned long int __h, __l;} __i; \
235 __asm__ ("xmpyu %1,%2,%0" \
237 : "x" ((unsigned long int)(u)), \
238 "x" ((unsigned long int)(v))); \
239 (wh) = __xx.__i.__h; \
240 (wl) = __xx.__i.__l; \
248 #define count_leading_zeros(count, x) \
250 unsigned long int __tmp; \
253 extru,= %1,15,16,%%r0 ; Bits 31..16 zero?
254 extru,tr %1,15,16,%1 ; No. Shift down, skip add.
255 ldo 16(%0),%0 ; Yes. Perform add.
256 extru,= %1,23,8,%%r0 ; Bits 15..8 zero?
257 extru,tr %1,23,8,%1 ; No. Shift down, skip add.
258 ldo 8(%0),%0 ; Yes. Perform add.
259 extru,= %1,27,4,%%r0 ; Bits 7..4 zero?
260 extru,tr %1,27,4,%1 ; No. Shift down, skip add.
261 ldo 4(%0),%0 ; Yes. Perform add.
262 extru,= %1,29,2,%%r0 ; Bits 3..2 zero?
263 extru,tr %1,29,2,%1 ; No. Shift down, skip add.
264 ldo 2(%0),%0 ; Yes. Perform add.
265 extru %1,30,1,%1 ; Extract bit 1.
266 sub %0,%1,%0 ; Subtract it.
267 " : "=r" (count), "=r" (__tmp) : "1" (x)); \
271 #if defined (__i386__) || defined (__i486__)
272 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
275 : "=r" ((unsigned long int)(sh)), \
276 "=&r" ((unsigned long int)(sl)) \
277 : "%0" ((unsigned long int)(ah)), \
278 "g" ((unsigned long int)(bh)), \
279 "%1" ((unsigned long int)(al)), \
280 "g" ((unsigned long int)(bl)))
281 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
284 : "=r" ((unsigned long int)(sh)), \
285 "=&r" ((unsigned long int)(sl)) \
286 : "0" ((unsigned long int)(ah)), \
287 "g" ((unsigned long int)(bh)), \
288 "1" ((unsigned long int)(al)), \
289 "g" ((unsigned long int)(bl)))
290 #define umul_ppmm(w1, w0, u, v) \
292 : "=a" ((unsigned long int)(w0)), \
293 "=d" ((unsigned long int)(w1)) \
294 : "%0" ((unsigned long int)(u)), \
295 "rm" ((unsigned long int)(v)))
296 #define udiv_qrnnd(q, r, n1, n0, d) \
298 : "=a" ((unsigned long int)(q)), \
299 "=d" ((unsigned long int)(r)) \
300 : "0" ((unsigned long int)(n0)), \
301 "1" ((unsigned long int)(n1)), \
302 "rm" ((unsigned long int)(d)))
303 #define count_leading_zeros(count, x) \
305 unsigned long int __cbtmp; \
306 __asm__ ("bsrl %1,%0" \
307 : "=r" (__cbtmp) : "rm" ((unsigned long int)(x))); \
308 (count) = __cbtmp ^ 31; \
314 #if defined (__i860__)
316 /* Make sure these patterns really improve the code before
317 switching them on. */
318 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
323 struct {unsigned long int l, h;} i; \
325 __a.i.l = (al); __a.i.h = (ah); \
326 __b.i.l = (bl); __b.i.h = (bh); \
327 __asm__ ("fiadd.dd %1,%2,%0" \
329 : "%f" (__a.ll), "f" (__b.ll)); \
330 (sh) = __s.i.h; (sl) = __s.i.l; \
332 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
337 struct {unsigned long int l, h;} i; \
339 __a.i.l = (al); __a.i.h = (ah); \
340 __b.i.l = (bl); __b.i.h = (bh); \
341 __asm__ ("fisub.dd %1,%2,%0" \
343 : "%f" (__a.ll), "f" (__b.ll)); \
344 (sh) = __s.i.h; (sl) = __s.i.l; \
347 #endif /* __i860__ */
349 #if defined (__i960__)
350 #define umul_ppmm(w1, w0, u, v) \
351 ({union {long long int __ll; \
352 struct {unsigned long int __l, __h;} __i; \
354 __asm__ ("emul %2,%1,%0" \
356 : "%dI" ((unsigned long int)(u)), \
357 "dI" ((unsigned long int)(v))); \
358 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
359 #define __umulsidi3(u, v) \
360 ({long long int __w; \
361 __asm__ ("emul %2,%1,%0" \
363 : "%dI" ((unsigned long int)(u)), \
364 "dI" ((unsigned long int)(v))); \
366 #endif /* __i960__ */
368 #if defined (___IBMR2__) /* IBM RS6000 */
369 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
370 __asm__ ("a%I5 %1,%4,%5
372 : "=r" ((unsigned long int)(sh)), \
373 "=&r" ((unsigned long int)(sl)) \
374 : "%r" ((unsigned long int)(ah)), \
375 "r" ((unsigned long int)(bh)), \
376 "%r" ((unsigned long int)(al)), \
377 "rI" ((unsigned long int)(bl)))
378 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
379 __asm__ ("sf%I4 %1,%5,%4
381 : "=r" ((unsigned long int)(sh)), \
382 "=&r" ((unsigned long int)(sl)) \
383 : "r" ((unsigned long int)(ah)), \
384 "r" ((unsigned long int)(bh)), \
385 "rI" ((unsigned long int)(al)), \
386 "r" ((unsigned long int)(bl)))
387 #define umul_ppmm(xh, xl, m0, m1) \
389 unsigned long int __m0 = (m0), __m1 = (m1); \
390 __asm__ ("mul %0,%2,%3" \
391 : "=r" ((unsigned long int)(xh)), \
392 "=q" ((unsigned long int)(xl)) \
395 (xh) += ((((signed long int) __m0 >> 31) & __m1) \
396 + (((signed long int) __m1 >> 31) & __m0)); \
398 #define smul_ppmm(xh, xl, m0, m1) \
399 __asm__ ("mul %0,%2,%3" \
400 : "=r" ((unsigned long int)(xh)), \
401 "=q" ((unsigned long int)(xl)) \
405 #define sdiv_qrnnd(q, r, nh, nl, d) \
406 __asm__ ("div %0,%2,%4" \
407 : "=r" (q), "=q" (r) \
408 : "r" (nh), "1" (nl), "r" (d))
409 #define UDIV_TIME 100
410 #define count_leading_zeros(count, x) \
411 __asm__ ("cntlz %0,%1" \
412 : "=r" ((unsigned long int)(count)) \
413 : "r" ((unsigned long int)(x)))
414 #endif /* ___IBMR2__ */
416 #if defined (__mc68000__)
417 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
418 __asm__ ("add%.l %5,%1
420 : "=d" ((unsigned long int)(sh)), \
421 "=&d" ((unsigned long int)(sl)) \
422 : "%0" ((unsigned long int)(ah)), \
423 "d" ((unsigned long int)(bh)), \
424 "%1" ((unsigned long int)(al)), \
425 "g" ((unsigned long int)(bl)))
426 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
427 __asm__ ("sub%.l %5,%1
429 : "=d" ((unsigned long int)(sh)), \
430 "=&d" ((unsigned long int)(sl)) \
431 : "0" ((unsigned long int)(ah)), \
432 "d" ((unsigned long int)(bh)), \
433 "1" ((unsigned long int)(al)), \
434 "g" ((unsigned long int)(bl)))
435 #if defined (__mc68020__) || defined (__NeXT__) || defined(mc68020)
436 #define umul_ppmm(w1, w0, u, v) \
437 __asm__ ("mulu%.l %3,%1:%0" \
438 : "=d" ((unsigned long int)(w0)), \
439 "=d" ((unsigned long int)(w1)) \
440 : "%0" ((unsigned long int)(u)), \
441 "dmi" ((unsigned long int)(v)))
443 #define udiv_qrnnd(q, r, n1, n0, d) \
444 __asm__ ("divu%.l %4,%1:%0" \
445 : "=d" ((unsigned long int)(q)), \
446 "=d" ((unsigned long int)(r)) \
447 : "0" ((unsigned long int)(n0)), \
448 "1" ((unsigned long int)(n1)), \
449 "dmi" ((unsigned long int)(d)))
451 #define sdiv_qrnnd(q, r, n1, n0, d) \
452 __asm__ ("divs%.l %4,%1:%0" \
453 : "=d" ((unsigned long int)(q)), \
454 "=d" ((unsigned long int)(r)) \
455 : "0" ((unsigned long int)(n0)), \
456 "1" ((unsigned long int)(n1)), \
457 "dmi" ((unsigned long int)(d)))
458 #define count_leading_zeros(count, x) \
459 __asm__ ("bfffo %1{%b2:%b2},%0" \
460 : "=d" ((unsigned long int)(count)) \
461 : "od" ((unsigned long int)(x)), "n" (0))
462 #else /* not mc68020 */
463 /* This ought to be improved by relying on reload to move inputs and
464 outputs to their positions. */
465 #define umul_ppmm(xh, xl, a, b) \
466 __asm__ ("| Inlined umul_ppmm
492 : "=g" ((unsigned long int)(xh)), \
493 "=g" ((unsigned long int)(xl)) \
494 : "g" ((unsigned long int)(a)), \
495 "g" ((unsigned long int)(b)) \
496 : "d0", "d1", "d2", "d3", "d4")
497 #define UMUL_TIME 100
498 #define UDIV_TIME 400
499 #endif /* not mc68020 */
502 #if defined (__m88000__)
503 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
504 __asm__ ("addu.co %1,%r4,%r5
505 addu.ci %0,%r2,%r3" \
506 : "=r" ((unsigned long int)(sh)), \
507 "=&r" ((unsigned long int)(sl)) \
508 : "%rJ" ((unsigned long int)(ah)), \
509 "rJ" ((unsigned long int)(bh)), \
510 "%rJ" ((unsigned long int)(al)), \
511 "rJ" ((unsigned long int)(bl)))
512 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
513 __asm__ ("subu.co %1,%r4,%r5
514 subu.ci %0,%r2,%r3" \
515 : "=r" ((unsigned long int)(sh)), \
516 "=&r" ((unsigned long int)(sl)) \
517 : "rJ" ((unsigned long int)(ah)), \
518 "rJ" ((unsigned long int)(bh)), \
519 "rJ" ((unsigned long int)(al)), \
520 "rJ" ((unsigned long int)(bl)))
522 #define UDIV_TIME 150
523 #define count_leading_zeros(count, x) \
525 unsigned long int __cbtmp; \
526 __asm__ ("ff1 %0,%1" \
528 : "r" ((unsigned long int)(x))); \
529 (count) = __cbtmp ^ 31; \
531 #if defined (__mc88110__)
532 #define umul_ppmm(wh, wl, u, v) \
534 union {long long int __ll; \
535 struct {unsigned long int __h, __l;} __i; \
537 __asm__ ("mulu.d %0,%1,%2" \
539 : "r" ((unsigned long int)(u)), \
540 "r" ((unsigned long int)(v))); \
541 (wh) = __xx.__i.__h; \
542 (wl) = __xx.__i.__l; \
545 #define udiv_qrnnd(q, r, n1, n0, d) \
546 __asm__ ("or r10,%2,0
559 #endif /* __m88000__ */
561 #if defined (__mips__)
562 /* The LO and HI registers are fixed in gcc/mips.h, for some reason. */
563 #if 0 && __GNUC__ >= 2
564 #define umul_ppmm(w1, w0, u, v) \
565 __asm__ ("multu %2,%3" \
566 : "=l" ((unsigned long int)(w0)), \
567 "=h" ((unsigned long int)(w1)) \
568 : "d" ((unsigned long int)(u)), \
569 "d" ((unsigned long int)(v)))
571 #define umul_ppmm(w1, w0, u, v) \
572 __asm__ ("multu %2,%3
575 : "=d" ((unsigned long int)(w0)), \
576 "=d" ((unsigned long int)(w1)) \
577 : "d" ((unsigned long int)(u)), \
578 "d" ((unsigned long int)(v)))
581 #define UDIV_TIME 100
582 #endif /* __mips__ */
584 #if defined (__ns32000__)
585 #define umul_ppmm(w1, w0, u, v) \
586 ({union {long long int __ll; \
587 struct {unsigned long int __l, __h;} __i; \
589 __asm__ ("meid %2,%0" \
591 : "%0" ((unsigned long int)(u)), \
592 "g" ((unsigned long int)(v))); \
593 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
594 #define __umulsidi3(u, v) \
595 ({long long int __w; \
596 __asm__ ("meid %2,%0" \
598 : "%0" ((unsigned long int)(u)), \
599 "g" ((unsigned long int)(v))); \
601 #define udiv_qrnnd(q, r, n1, n0, d) \
607 : "=g" ((unsigned long int)(q)), \
608 "=g" ((unsigned long int)(r)) \
609 : "g" ((unsigned long int)(n0)), \
610 "g" ((unsigned long int)(n1)), \
611 "g" ((unsigned long int)(d)) \
613 #endif /* __ns32000__ */
615 #if defined (__pyr__)
616 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
619 : "=r" ((unsigned long int)(sh)), \
620 "=&r" ((unsigned long int)(sl)) \
621 : "%0" ((unsigned long int)(ah)), \
622 "g" ((unsigned long int)(bh)), \
623 "%1" ((unsigned long int)(al)), \
624 "g" ((unsigned long int)(bl)))
625 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
628 : "=r" ((unsigned long int)(sh)), \
629 "=&r" ((unsigned long int)(sl)) \
630 : "0" ((unsigned long int)(ah)), \
631 "g" ((unsigned long int)(bh)), \
632 "1" ((unsigned long int)(al)), \
633 "g" ((unsigned long int)(bl)))
634 /* This insn doesn't work on ancient pyramids. */
635 #define umul_ppmm(w1, w0, u, v) \
636 ({union {long long int __ll; \
637 struct {unsigned long int __h, __l;} __i; \
640 __asm__ ("uemul %3,%0" \
641 : "=r" (__xx.__i.__h), \
642 "=r" (__xx.__i.__l) \
643 : "1" (__xx.__i.__l), \
645 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
648 #if defined (__ibm032__) /* RT/ROMP */
649 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
652 : "=r" ((unsigned long int)(sh)), \
653 "=&r" ((unsigned long int)(sl)) \
654 : "%0" ((unsigned long int)(ah)), \
655 "r" ((unsigned long int)(bh)), \
656 "%1" ((unsigned long int)(al)), \
657 "r" ((unsigned long int)(bl)))
658 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
661 : "=r" ((unsigned long int)(sh)), \
662 "=&r" ((unsigned long int)(sl)) \
663 : "0" ((unsigned long int)(ah)), \
664 "r" ((unsigned long int)(bh)), \
665 "1" ((unsigned long int)(al)), \
666 "r" ((unsigned long int)(bl)))
667 #define umul_ppmm(ph, pl, m0, m1) \
669 unsigned long int __m0 = (m0), __m1 = (m1); \
691 : "=r" ((unsigned long int)(ph)), \
692 "=r" ((unsigned long int)(pl)) \
696 (ph) += ((((signed long int) __m0 >> 31) & __m1) \
697 + (((signed long int) __m1 >> 31) & __m0)); \
700 #define UDIV_TIME 200
701 #define count_leading_zeros(count, x) \
703 if ((x) >= 0x10000) \
704 __asm__ ("clz %0,%1" \
705 : "=r" ((unsigned long int)(count)) \
706 : "r" ((unsigned long int)(x) >> 16)); \
709 __asm__ ("clz %0,%1" \
710 : "=r" ((unsigned long int)(count)) \
711 : "r" ((unsigned long int)(x))); \
717 #if defined (__sparc__)
718 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
719 __asm__ ("addcc %4,%5,%1
721 : "=r" ((unsigned long int)(sh)), \
722 "=&r" ((unsigned long int)(sl)) \
723 : "%r" ((unsigned long int)(ah)), \
724 "rI" ((unsigned long int)(bh)), \
725 "%r" ((unsigned long int)(al)), \
726 "rI" ((unsigned long int)(bl)) \
728 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
729 __asm__ ("subcc %4,%5,%1
731 : "=r" ((unsigned long int)(sh)), \
732 "=&r" ((unsigned long int)(sl)) \
733 : "r" ((unsigned long int)(ah)), \
734 "rI" ((unsigned long int)(bh)), \
735 "r" ((unsigned long int)(al)), \
736 "rI" ((unsigned long int)(bl)) \
738 #if defined (__sparc_v8__)
739 #define umul_ppmm(w1, w0, u, v) \
740 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
741 : "=r" ((unsigned long int)(w1)), \
742 "=r" ((unsigned long int)(w0)) \
743 : "r" ((unsigned long int)(u)), \
744 "r" ((unsigned long int)(v)))
745 /* We might want to leave this undefined for `SuperSPARC (tm)' since
746 its implementation is crippled and often traps. */
747 #define udiv_qrnnd(q, r, n1, n0, d) \
748 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
749 : "=&r" ((unsigned long int)(q)), \
750 "=&r" ((unsigned long int)(r)) \
751 : "r" ((unsigned long int)(n1)), \
752 "r" ((unsigned long int)(n0)), \
753 "r" ((unsigned long int)(d)))
757 /* SPARC without integer multiplication and divide instructions.
758 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
759 #define umul_ppmm(w1, w0, u, v) \
760 __asm__ ("! Inlined umul_ppmm
761 wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr
762 sra %3,31,%%g2 ! Don't move this insn
763 and %2,%%g2,%%g2 ! Don't move this insn
764 andcc %%g0,0,%%g1 ! Don't move this insn
800 : "=r" ((unsigned long int)(w1)), \
801 "=r" ((unsigned long int)(w0)) \
802 : "%rI" ((unsigned long int)(u)), \
803 "r" ((unsigned long int)(v)) \
804 : "%g1", "%g2" __AND_CLOBBER_CC)
805 #define UMUL_TIME 39 /* 39 instructions */
806 /* It's quite necessary to add this much assembler for the sparc.
807 The default udiv_qrnnd (in C) is more than 10 times slower! */
808 #define udiv_qrnnd(q, r, n1, n0, d) \
809 __asm__ ("! Inlined udiv_qrnnd
813 addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb
814 sub %1,%2,%1 ! this kills msb of n
815 addx %1,%1,%1 ! so this can't give carry
820 addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb
822 sub %1,%2,%1 ! this kills msb of n
827 ! Got carry from n. Subtract next step to cancel this carry.
829 addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb
832 ! End of inline udiv_qrnnd" \
833 : "=&r" ((unsigned long int)(q)), \
834 "=&r" ((unsigned long int)(r)) \
835 : "r" ((unsigned long int)(d)), \
836 "1" ((unsigned long int)(n1)), \
837 "0" ((unsigned long int)(n0)) : "%g1" __AND_CLOBBER_CC)
838 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
839 #endif /* __sparc8__ */
840 #endif /* __sparc__ */
842 #if defined (__vax__)
843 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
844 __asm__ ("addl2 %5,%1
846 : "=g" ((unsigned long int)(sh)), \
847 "=&g" ((unsigned long int)(sl)) \
848 : "%0" ((unsigned long int)(ah)), \
849 "g" ((unsigned long int)(bh)), \
850 "%1" ((unsigned long int)(al)), \
851 "g" ((unsigned long int)(bl)))
852 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
853 __asm__ ("subl2 %5,%1
855 : "=g" ((unsigned long int)(sh)), \
856 "=&g" ((unsigned long int)(sl)) \
857 : "0" ((unsigned long int)(ah)), \
858 "g" ((unsigned long int)(bh)), \
859 "1" ((unsigned long int)(al)), \
860 "g" ((unsigned long int)(bl)))
861 #define umul_ppmm(xh, xl, m0, m1) \
863 union {long long int __ll; \
864 struct {unsigned long int __l, __h;} __i; \
866 unsigned long int __m0 = (m0), __m1 = (m1); \
867 __asm__ ("emul %1,%2,$0,%0" \
871 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
872 (xh) += ((((signed long int) __m0 >> 31) & __m1) \
873 + (((signed long int) __m1 >> 31) & __m0)); \
875 #define sdiv_qrnnd(q, r, n1, n0, d) \
877 union {long long int __ll; \
878 struct {unsigned long int __l, __h;} __i; \
880 __xx.__i.__h = n1; __xx.__i.__l = n0; \
881 __asm__ ("ediv %3,%2,%0,%1" \
882 : "=g" (q), "=g" (r) \
883 : "g" (__n1n0.ll), "g" (d)); \
887 #endif /* __GNUC__ */
890 #if !defined (umul_ppmm) && defined (__umulsidi3)
891 #define umul_ppmm(ph, pl, m0, m1) \
893 unsigned long long int __ll = __umulsidi3 (m0, m1); \
894 ph = (unsigned long int) (__ll >> LONG_TYPE_SIZE); \
895 pl = (unsigned long int) __ll; \
899 #if !defined (__umulsidi3)
900 #define __umulsidi3(u, v) \
902 umul_ppmm (__hi, __lo, u, v); \
903 ((unsigned long long) __hi << LONG_TYPE_SIZE) | __lo; })
906 /* If this machine has no inline assembler, use C macros. */
908 #if !defined (add_ssaaaa)
909 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
911 unsigned long int __x; \
913 (sh) = (ah) + (bh) + (__x < (al)); \
918 #if !defined (sub_ddmmss)
919 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
921 unsigned long int __x; \
923 (sh) = (ah) - (bh) - (__x > (al)); \
928 #if !defined (umul_ppmm)
929 #define umul_ppmm(w1, w0, u, v) \
931 unsigned long int __x0, __x1, __x2, __x3; \
932 unsigned int __ul, __vl, __uh, __vh; \
934 __ul = __ll_lowpart (u); \
935 __uh = __ll_highpart (u); \
936 __vl = __ll_lowpart (v); \
937 __vh = __ll_highpart (v); \
939 __x0 = (unsigned long int) __ul * __vl; \
940 __x1 = (unsigned long int) __ul * __vh; \
941 __x2 = (unsigned long int) __uh * __vl; \
942 __x3 = (unsigned long int) __uh * __vh; \
944 __x1 += __ll_highpart (__x0);/* this can't give carry */ \
945 __x1 += __x2; /* but this indeed can */ \
946 if (__x1 < __x2) /* did we get it? */ \
947 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
949 (w1) = __x3 + __ll_highpart (__x1); \
950 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
954 /* Define this unconditionally, so it can be used for debugging. */
955 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
957 unsigned long int __d1, __d0, __q1, __q0, __r1, __r0, __m; \
958 __d1 = __ll_highpart (d); \
959 __d0 = __ll_lowpart (d); \
961 __r1 = (n1) % __d1; \
962 __q1 = (n1) / __d1; \
963 __m = (unsigned long int) __q1 * __d0; \
964 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
967 __q1--, __r1 += (d); \
968 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
970 __q1--, __r1 += (d); \
974 __r0 = __r1 % __d1; \
975 __q0 = __r1 / __d1; \
976 __m = (unsigned long int) __q0 * __d0; \
977 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
980 __q0--, __r0 += (d); \
983 __q0--, __r0 += (d); \
987 (q) = (unsigned long int) __q1 * __ll_B | __q0; \
991 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
992 #if !defined (udiv_qrnnd)
993 #define UDIV_NEEDS_NORMALIZATION 1
994 #define udiv_qrnnd __udiv_qrnnd_c
997 #if !defined (count_leading_zeros)
1002 unsigned char __clz_tab[];
1003 #define count_leading_zeros(count, x) \
1005 unsigned long int __xr = (x); \
1008 if (LONG_TYPE_SIZE <= 32) \
1010 __a = __xr < (1<<2*__BITS4) \
1011 ? (__xr < (1<<__BITS4) ? 0 : __BITS4) \
1012 : (__xr < (1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1016 for (__a = LONG_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1017 if (((__xr >> __a) & 0xff) != 0) \
1021 (count) = LONG_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1025 #ifndef UDIV_NEEDS_NORMALIZATION
1026 #define UDIV_NEEDS_NORMALIZATION 0