1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2 Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
3 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
4 Free Software Foundation, Inc.
6 This file is part of the GNU C Library.
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
13 In addition to the permissions in the GNU Lesser General Public
14 License, the Free Software Foundation gives you unlimited
15 permission to link the compiled version of this file into
16 combinations with other programs, and to distribute those
17 combinations without any restriction coming from the use of this
18 file. (The Lesser General Public License restrictions do apply in
19 other respects; for example, they cover modification of the file,
20 and distribution when not linked into a combine executable.)
22 The GNU C Library is distributed in the hope that it will be useful,
23 but WITHOUT ANY WARRANTY; without even the implied warranty of
24 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 Lesser General Public License for more details.
27 You should have received a copy of the GNU Lesser General Public
28 License along with the GNU C Library; if not, write to the Free
29 Software Foundation, 51 Franklin Street, Fifth Floor, Boston,
30 MA 02110-1301, USA. */
32 /* You have to define the following before including this file:
34 UWtype -- An unsigned type, default type for operations (typically a "word")
35 UHWtype -- An unsigned type, at least half the size of UWtype.
36 UDWtype -- An unsigned type, at least twice as large a UWtype
37 W_TYPE_SIZE -- size in bits of UWtype
39 UQItype -- Unsigned 8 bit type.
40 SItype, USItype -- Signed and unsigned 32 bit types.
41 DItype, UDItype -- Signed and unsigned 64 bit types.
43 On a 32 bit machine UWtype should typically be USItype;
44 on a 64 bit machine, UWtype should typically be UDItype. */
46 #define __BITS4 (W_TYPE_SIZE / 4)
47 #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
48 #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
49 #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
52 #define W_TYPE_SIZE 32
53 #define UWtype USItype
54 #define UHWtype USItype
55 #define UDWtype UDItype
58 /* Used in glibc only. */
59 #ifndef attribute_hidden
60 #define attribute_hidden
63 extern const UQItype __clz_tab[256] attribute_hidden;
65 /* Define auxiliary asm macros.
67 1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
68 UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
69 word product in HIGH_PROD and LOW_PROD.
71 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
72 UDWtype product. This is just a variant of umul_ppmm.
74 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
75 denominator) divides a UDWtype, composed by the UWtype integers
76 HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
77 in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less
78 than DENOMINATOR for correct operation. If, in addition, the most
79 significant bit of DENOMINATOR must be 1, then the pre-processor symbol
80 UDIV_NEEDS_NORMALIZATION is defined to 1.
82 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
83 denominator). Like udiv_qrnnd but the numbers are signed. The quotient
86 5) count_leading_zeros(count, x) counts the number of zero-bits from the
87 msb to the first nonzero bit in the UWtype X. This is the number of
88 steps X needs to be shifted left to set the msb. Undefined for X == 0,
89 unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
91 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
92 from the least significant end.
94 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
95 high_addend_2, low_addend_2) adds two UWtype integers, composed by
96 HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
97 respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow
98 (i.e. carry out) is not stored anywhere, and is lost.
100 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
101 high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
102 composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
103 LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE
104 and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
107 If any of these macros are left undefined for a particular CPU,
108 C macros are used. */
110 /* The CPUs come in alphabetical order below.
112 Please add support for more CPUs here, or improve the current support
114 (E.g. WE32100, IBM360.) */
116 #if defined (__GNUC__) && !defined (NO_ASM)
118 /* We sometimes need to clobber "cc" with gcc2, but that would not be
119 understood by gcc1. Use cpp to avoid major code duplication. */
122 #define __AND_CLOBBER_CC
123 #else /* __GNUC__ >= 2 */
124 #define __CLOBBER_CC : "cc"
125 #define __AND_CLOBBER_CC , "cc"
126 #endif /* __GNUC__ < 2 */
128 #if defined (__alpha) && W_TYPE_SIZE == 64
129 #define umul_ppmm(ph, pl, m0, m1) \
131 UDItype __m0 = (m0), __m1 = (m1); \
132 (ph) = __builtin_alpha_umulh (__m0, __m1); \
133 (pl) = __m0 * __m1; \
136 #ifndef LONGLONG_STANDALONE
137 #define udiv_qrnnd(q, r, n1, n0, d) \
139 (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \
142 extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
143 #define UDIV_TIME 220
144 #endif /* LONGLONG_STANDALONE */
146 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X))
147 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
148 #define COUNT_LEADING_ZEROS_0 64
150 #define count_leading_zeros(COUNT,X) \
152 UDItype __xr = (X), __t, __a; \
153 __t = __builtin_alpha_cmpbge (0, __xr); \
154 __a = __clz_tab[__t ^ 0xff] - 1; \
155 __t = __builtin_alpha_extbl (__xr, __a); \
156 (COUNT) = 64 - (__clz_tab[__t] + __a*8); \
158 #define count_trailing_zeros(COUNT,X) \
160 UDItype __xr = (X), __t, __a; \
161 __t = __builtin_alpha_cmpbge (0, __xr); \
162 __t = ~__t & -~__t; \
163 __a = ((__t & 0xCC) != 0) * 2; \
164 __a += ((__t & 0xF0) != 0) * 4; \
165 __a += ((__t & 0xAA) != 0); \
166 __t = __builtin_alpha_extbl (__xr, __a); \
169 __a += ((__t & 0xCC) != 0) * 2; \
170 __a += ((__t & 0xF0) != 0) * 4; \
171 __a += ((__t & 0xAA) != 0); \
174 #endif /* __alpha_cix__ */
177 #if defined (__arc__) && W_TYPE_SIZE == 32
178 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
179 __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \
180 : "=r" ((USItype) (sh)), \
181 "=&r" ((USItype) (sl)) \
182 : "%r" ((USItype) (ah)), \
183 "rIJ" ((USItype) (bh)), \
184 "%r" ((USItype) (al)), \
185 "rIJ" ((USItype) (bl)))
186 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
187 __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \
188 : "=r" ((USItype) (sh)), \
189 "=&r" ((USItype) (sl)) \
190 : "r" ((USItype) (ah)), \
191 "rIJ" ((USItype) (bh)), \
192 "r" ((USItype) (al)), \
193 "rIJ" ((USItype) (bl)))
194 /* Call libgcc routine. */
195 #define umul_ppmm(w1, w0, u, v) \
198 __w.ll = __umulsidi3 (u, v); \
202 #define __umulsidi3 __umulsidi3
203 UDItype __umulsidi3 (USItype, USItype);
206 #if defined (__arm__) && !defined (__thumb__) && W_TYPE_SIZE == 32
207 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
208 __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \
209 : "=r" ((USItype) (sh)), \
210 "=&r" ((USItype) (sl)) \
211 : "%r" ((USItype) (ah)), \
212 "rI" ((USItype) (bh)), \
213 "%r" ((USItype) (al)), \
214 "rI" ((USItype) (bl)) __CLOBBER_CC)
215 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
216 __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \
217 : "=r" ((USItype) (sh)), \
218 "=&r" ((USItype) (sl)) \
219 : "r" ((USItype) (ah)), \
220 "rI" ((USItype) (bh)), \
221 "r" ((USItype) (al)), \
222 "rI" ((USItype) (bl)) __CLOBBER_CC)
223 #define umul_ppmm(xh, xl, a, b) \
224 {register USItype __t0, __t1, __t2; \
225 __asm__ ("%@ Inlined umul_ppmm\n" \
226 " mov %2, %5, lsr #16\n" \
227 " mov %0, %6, lsr #16\n" \
228 " bic %3, %5, %2, lsl #16\n" \
229 " bic %4, %6, %0, lsl #16\n" \
230 " mul %1, %3, %4\n" \
231 " mul %4, %2, %4\n" \
232 " mul %3, %0, %3\n" \
233 " mul %0, %2, %0\n" \
234 " adds %3, %4, %3\n" \
235 " addcs %0, %0, #65536\n" \
236 " adds %1, %1, %3, lsl #16\n" \
237 " adc %0, %0, %3, lsr #16" \
238 : "=&r" ((USItype) (xh)), \
239 "=r" ((USItype) (xl)), \
240 "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \
241 : "r" ((USItype) (a)), \
242 "r" ((USItype) (b)) __CLOBBER_CC );}
244 #define UDIV_TIME 100
248 /* Let gcc decide how best to implement count_leading_zeros. */
249 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
250 #define COUNT_LEADING_ZEROS_0 32
253 #if defined (__CRIS__) && __CRIS_arch_version >= 3
254 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
255 #if __CRIS_arch_version >= 8
256 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
258 #endif /* __CRIS__ */
260 #if defined (__hppa) && W_TYPE_SIZE == 32
261 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
262 __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \
263 : "=r" ((USItype) (sh)), \
264 "=&r" ((USItype) (sl)) \
265 : "%rM" ((USItype) (ah)), \
266 "rM" ((USItype) (bh)), \
267 "%rM" ((USItype) (al)), \
268 "rM" ((USItype) (bl)))
269 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
270 __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \
271 : "=r" ((USItype) (sh)), \
272 "=&r" ((USItype) (sl)) \
273 : "rM" ((USItype) (ah)), \
274 "rM" ((USItype) (bh)), \
275 "rM" ((USItype) (al)), \
276 "rM" ((USItype) (bl)))
277 #if defined (_PA_RISC1_1)
278 #define umul_ppmm(w1, w0, u, v) \
283 struct {USItype __w1, __w0;} __w1w0; \
285 __asm__ ("xmpyu %1,%2,%0" \
287 : "x" ((USItype) (u)), \
288 "x" ((USItype) (v))); \
289 (w1) = __t.__w1w0.__w1; \
290 (w0) = __t.__w1w0.__w0; \
297 #define count_leading_zeros(count, x) \
302 " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \
303 " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\
304 " ldo 16(%0),%0 ; Yes. Perform add.\n" \
305 " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \
306 " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\
307 " ldo 8(%0),%0 ; Yes. Perform add.\n" \
308 " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \
309 " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\
310 " ldo 4(%0),%0 ; Yes. Perform add.\n" \
311 " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \
312 " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\
313 " ldo 2(%0),%0 ; Yes. Perform add.\n" \
314 " extru %1,30,1,%1 ; Extract bit 1.\n" \
315 " sub %0,%1,%0 ; Subtract it.\n" \
316 : "=r" (count), "=r" (__tmp) : "1" (x)); \
320 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
321 #if !defined (__zarch__)
322 #define smul_ppmm(xh, xl, m0, m1) \
324 union {DItype __ll; \
325 struct {USItype __h, __l;} __i; \
327 __asm__ ("lr %N0,%1\n\tmr %0,%2" \
329 : "r" (m0), "r" (m1)); \
330 (xh) = __x.__i.__h; (xl) = __x.__i.__l; \
332 #define sdiv_qrnnd(q, r, n1, n0, d) \
334 union {DItype __ll; \
335 struct {USItype __h, __l;} __i; \
337 __x.__i.__h = n1; __x.__i.__l = n0; \
338 __asm__ ("dr %0,%2" \
340 : "0" (__x.__ll), "r" (d)); \
341 (q) = __x.__i.__l; (r) = __x.__i.__h; \
344 #define smul_ppmm(xh, xl, m0, m1) \
346 register SItype __r0 __asm__ ("0"); \
347 register SItype __r1 __asm__ ("1") = (m0); \
349 __asm__ ("mr\t%%r0,%3" \
350 : "=r" (__r0), "=r" (__r1) \
351 : "r" (__r1), "r" (m1)); \
352 (xh) = __r0; (xl) = __r1; \
355 #define sdiv_qrnnd(q, r, n1, n0, d) \
357 register SItype __r0 __asm__ ("0") = (n1); \
358 register SItype __r1 __asm__ ("1") = (n0); \
360 __asm__ ("dr\t%%r0,%4" \
361 : "=r" (__r0), "=r" (__r1) \
362 : "r" (__r0), "r" (__r1), "r" (d)); \
363 (q) = __r1; (r) = __r0; \
365 #endif /* __zarch__ */
368 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
369 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
370 __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}" \
371 : "=r" ((USItype) (sh)), \
372 "=&r" ((USItype) (sl)) \
373 : "%0" ((USItype) (ah)), \
374 "g" ((USItype) (bh)), \
375 "%1" ((USItype) (al)), \
376 "g" ((USItype) (bl)))
377 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
378 __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}" \
379 : "=r" ((USItype) (sh)), \
380 "=&r" ((USItype) (sl)) \
381 : "0" ((USItype) (ah)), \
382 "g" ((USItype) (bh)), \
383 "1" ((USItype) (al)), \
384 "g" ((USItype) (bl)))
385 #define umul_ppmm(w1, w0, u, v) \
386 __asm__ ("mul{l} %3" \
387 : "=a" ((USItype) (w0)), \
388 "=d" ((USItype) (w1)) \
389 : "%0" ((USItype) (u)), \
390 "rm" ((USItype) (v)))
391 #define udiv_qrnnd(q, r, n1, n0, dv) \
392 __asm__ ("div{l} %4" \
393 : "=a" ((USItype) (q)), \
394 "=d" ((USItype) (r)) \
395 : "0" ((USItype) (n0)), \
396 "1" ((USItype) (n1)), \
397 "rm" ((USItype) (dv)))
398 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x))
399 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x))
404 #if (defined (__x86_64__) || defined (__i386__)) && W_TYPE_SIZE == 64
405 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
406 __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}" \
407 : "=r" ((UDItype) (sh)), \
408 "=&r" ((UDItype) (sl)) \
409 : "%0" ((UDItype) (ah)), \
410 "rme" ((UDItype) (bh)), \
411 "%1" ((UDItype) (al)), \
412 "rme" ((UDItype) (bl)))
413 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
414 __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}" \
415 : "=r" ((UDItype) (sh)), \
416 "=&r" ((UDItype) (sl)) \
417 : "0" ((UDItype) (ah)), \
418 "rme" ((UDItype) (bh)), \
419 "1" ((UDItype) (al)), \
420 "rme" ((UDItype) (bl)))
421 #define umul_ppmm(w1, w0, u, v) \
422 __asm__ ("mul{q} %3" \
423 : "=a" ((UDItype) (w0)), \
424 "=d" ((UDItype) (w1)) \
425 : "%0" ((UDItype) (u)), \
426 "rm" ((UDItype) (v)))
427 #define udiv_qrnnd(q, r, n1, n0, dv) \
428 __asm__ ("div{q} %4" \
429 : "=a" ((UDItype) (q)), \
430 "=d" ((UDItype) (r)) \
431 : "0" ((UDItype) (n0)), \
432 "1" ((UDItype) (n1)), \
433 "rm" ((UDItype) (dv)))
434 #define count_leading_zeros(count, x) ((count) = __builtin_clzl (x))
435 #define count_trailing_zeros(count, x) ((count) = __builtin_ctzl (x))
440 #if defined (__i960__) && W_TYPE_SIZE == 32
441 #define umul_ppmm(w1, w0, u, v) \
442 ({union {UDItype __ll; \
443 struct {USItype __l, __h;} __i; \
445 __asm__ ("emul %2,%1,%0" \
447 : "%dI" ((USItype) (u)), \
448 "dI" ((USItype) (v))); \
449 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
450 #define __umulsidi3(u, v) \
452 __asm__ ("emul %2,%1,%0" \
454 : "%dI" ((USItype) (u)), \
455 "dI" ((USItype) (v))); \
457 #endif /* __i960__ */
459 #if defined (__ia64) && W_TYPE_SIZE == 64
460 /* This form encourages gcc (pre-release 3.4 at least) to emit predicated
461 "sub r=r,r" and "sub r=r,r,1", giving a 2 cycle latency. The generic
462 code using "al<bl" arithmetically comes out making an actual 0 or 1 in a
463 register, which takes an extra cycle. */
464 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
469 (sh) = (ah) - (bh) - 1; \
471 (sh) = (ah) - (bh); \
475 /* Do both product parts in assembly, since that gives better code with
476 all gcc versions. Some callers will just use the upper part, and in
477 that situation we waste an instruction, but not any cycles. */
478 #define umul_ppmm(ph, pl, m0, m1) \
479 __asm__ ("xma.hu %0 = %2, %3, f0\n\txma.l %1 = %2, %3, f0" \
480 : "=&f" (ph), "=f" (pl) \
481 : "f" (m0), "f" (m1))
482 #define count_leading_zeros(count, x) \
484 UWtype _x = (x), _y, _a, _c; \
485 __asm__ ("mux1 %0 = %1, @rev" : "=r" (_y) : "r" (_x)); \
486 __asm__ ("czx1.l %0 = %1" : "=r" (_a) : "r" (-_y | _y)); \
487 _c = (_a - 1) << 3; \
494 (count) = W_TYPE_SIZE - 1 - _c; \
496 /* similar to what gcc does for __builtin_ffs, but 0 based rather than 1
497 based, and we don't need a special case for x==0 here */
498 #define count_trailing_zeros(count, x) \
500 UWtype __ctz_x = (x); \
501 __asm__ ("popcnt %0 = %1" \
503 : "r" ((__ctz_x-1) & ~__ctz_x)); \
508 #if defined (__M32R__) && W_TYPE_SIZE == 32
509 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
510 /* The cmp clears the condition bit. */ \
511 __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \
512 : "=r" ((USItype) (sh)), \
513 "=&r" ((USItype) (sl)) \
514 : "0" ((USItype) (ah)), \
515 "r" ((USItype) (bh)), \
516 "1" ((USItype) (al)), \
517 "r" ((USItype) (bl)) \
519 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
520 /* The cmp clears the condition bit. */ \
521 __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \
522 : "=r" ((USItype) (sh)), \
523 "=&r" ((USItype) (sl)) \
524 : "0" ((USItype) (ah)), \
525 "r" ((USItype) (bh)), \
526 "1" ((USItype) (al)), \
527 "r" ((USItype) (bl)) \
529 #endif /* __M32R__ */
531 #if defined (__mc68000__) && W_TYPE_SIZE == 32
532 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
533 __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \
534 : "=d" ((USItype) (sh)), \
535 "=&d" ((USItype) (sl)) \
536 : "%0" ((USItype) (ah)), \
537 "d" ((USItype) (bh)), \
538 "%1" ((USItype) (al)), \
539 "g" ((USItype) (bl)))
540 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
541 __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \
542 : "=d" ((USItype) (sh)), \
543 "=&d" ((USItype) (sl)) \
544 : "0" ((USItype) (ah)), \
545 "d" ((USItype) (bh)), \
546 "1" ((USItype) (al)), \
547 "g" ((USItype) (bl)))
549 /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r. */
550 #if (defined (__mc68020__) && !defined (__mc68060__))
551 #define umul_ppmm(w1, w0, u, v) \
552 __asm__ ("mulu%.l %3,%1:%0" \
553 : "=d" ((USItype) (w0)), \
554 "=d" ((USItype) (w1)) \
555 : "%0" ((USItype) (u)), \
556 "dmi" ((USItype) (v)))
558 #define udiv_qrnnd(q, r, n1, n0, d) \
559 __asm__ ("divu%.l %4,%1:%0" \
560 : "=d" ((USItype) (q)), \
561 "=d" ((USItype) (r)) \
562 : "0" ((USItype) (n0)), \
563 "1" ((USItype) (n1)), \
564 "dmi" ((USItype) (d)))
566 #define sdiv_qrnnd(q, r, n1, n0, d) \
567 __asm__ ("divs%.l %4,%1:%0" \
568 : "=d" ((USItype) (q)), \
569 "=d" ((USItype) (r)) \
570 : "0" ((USItype) (n0)), \
571 "1" ((USItype) (n1)), \
572 "dmi" ((USItype) (d)))
574 #elif defined (__mcoldfire__) /* not mc68020 */
576 #define umul_ppmm(xh, xl, a, b) \
577 __asm__ ("| Inlined umul_ppmm\n" \
578 " move%.l %2,%/d0\n" \
579 " move%.l %3,%/d1\n" \
580 " move%.l %/d0,%/d2\n" \
582 " move%.l %/d1,%/d3\n" \
584 " move%.w %/d2,%/d4\n" \
585 " mulu %/d3,%/d4\n" \
586 " mulu %/d1,%/d2\n" \
587 " mulu %/d0,%/d3\n" \
588 " mulu %/d0,%/d1\n" \
589 " move%.l %/d4,%/d0\n" \
592 " add%.l %/d0,%/d2\n" \
593 " add%.l %/d3,%/d2\n" \
595 " add%.l %#65536,%/d1\n" \
597 " moveq %#0,%/d0\n" \
598 " move%.w %/d2,%/d0\n" \
599 " move%.w %/d4,%/d2\n" \
600 " move%.l %/d2,%1\n" \
601 " add%.l %/d1,%/d0\n" \
603 : "=g" ((USItype) (xh)), \
604 "=g" ((USItype) (xl)) \
605 : "g" ((USItype) (a)), \
606 "g" ((USItype) (b)) \
607 : "d0", "d1", "d2", "d3", "d4")
608 #define UMUL_TIME 100
609 #define UDIV_TIME 400
610 #else /* not ColdFire */
611 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */
612 #define umul_ppmm(xh, xl, a, b) \
613 __asm__ ("| Inlined umul_ppmm\n" \
614 " move%.l %2,%/d0\n" \
615 " move%.l %3,%/d1\n" \
616 " move%.l %/d0,%/d2\n" \
618 " move%.l %/d1,%/d3\n" \
620 " move%.w %/d2,%/d4\n" \
621 " mulu %/d3,%/d4\n" \
622 " mulu %/d1,%/d2\n" \
623 " mulu %/d0,%/d3\n" \
624 " mulu %/d0,%/d1\n" \
625 " move%.l %/d4,%/d0\n" \
626 " eor%.w %/d0,%/d0\n" \
628 " add%.l %/d0,%/d2\n" \
629 " add%.l %/d3,%/d2\n" \
631 " add%.l %#65536,%/d1\n" \
633 " moveq %#0,%/d0\n" \
634 " move%.w %/d2,%/d0\n" \
635 " move%.w %/d4,%/d2\n" \
636 " move%.l %/d2,%1\n" \
637 " add%.l %/d1,%/d0\n" \
639 : "=g" ((USItype) (xh)), \
640 "=g" ((USItype) (xl)) \
641 : "g" ((USItype) (a)), \
642 "g" ((USItype) (b)) \
643 : "d0", "d1", "d2", "d3", "d4")
644 #define UMUL_TIME 100
645 #define UDIV_TIME 400
647 #endif /* not mc68020 */
649 /* The '020, '030, '040 and '060 have bitfield insns.
650 cpu32 disguises as a 68020, but lacks them. */
651 #if defined (__mc68020__) && !defined (__mcpu32__)
652 #define count_leading_zeros(count, x) \
653 __asm__ ("bfffo %1{%b2:%b2},%0" \
654 : "=d" ((USItype) (count)) \
655 : "od" ((USItype) (x)), "n" (0))
656 /* Some ColdFire architectures have a ff1 instruction supported via
658 #elif defined (__mcfisaaplus__) || defined (__mcfisac__)
659 #define count_leading_zeros(count,x) ((count) = __builtin_clz (x))
660 #define COUNT_LEADING_ZEROS_0 32
664 #if defined (__m88000__) && W_TYPE_SIZE == 32
665 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
666 __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \
667 : "=r" ((USItype) (sh)), \
668 "=&r" ((USItype) (sl)) \
669 : "%rJ" ((USItype) (ah)), \
670 "rJ" ((USItype) (bh)), \
671 "%rJ" ((USItype) (al)), \
672 "rJ" ((USItype) (bl)))
673 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
674 __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \
675 : "=r" ((USItype) (sh)), \
676 "=&r" ((USItype) (sl)) \
677 : "rJ" ((USItype) (ah)), \
678 "rJ" ((USItype) (bh)), \
679 "rJ" ((USItype) (al)), \
680 "rJ" ((USItype) (bl)))
681 #define count_leading_zeros(count, x) \
684 __asm__ ("ff1 %0,%1" \
686 : "r" ((USItype) (x))); \
687 (count) = __cbtmp ^ 31; \
689 #define COUNT_LEADING_ZEROS_0 63 /* sic */
690 #if defined (__mc88110__)
691 #define umul_ppmm(wh, wl, u, v) \
693 union {UDItype __ll; \
694 struct {USItype __h, __l;} __i; \
696 __asm__ ("mulu.d %0,%1,%2" \
698 : "r" ((USItype) (u)), \
699 "r" ((USItype) (v))); \
700 (wh) = __xx.__i.__h; \
701 (wl) = __xx.__i.__l; \
703 #define udiv_qrnnd(q, r, n1, n0, d) \
704 ({union {UDItype __ll; \
705 struct {USItype __h, __l;} __i; \
708 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
709 __asm__ ("divu.d %0,%1,%2" \
712 "r" ((USItype) (d))); \
713 (r) = (n0) - __q * (d); (q) = __q; })
718 #define UDIV_TIME 150
719 #endif /* __mc88110__ */
720 #endif /* __m88000__ */
722 #if defined (__mn10300__)
723 # if defined (__AM33__)
724 # define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
725 # define umul_ppmm(w1, w0, u, v) \
726 asm("mulu %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
727 # define smul_ppmm(w1, w0, u, v) \
728 asm("mul %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
730 # define umul_ppmm(w1, w0, u, v) \
731 asm("nop; nop; mulu %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
732 # define smul_ppmm(w1, w0, u, v) \
733 asm("nop; nop; mul %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
735 # define add_ssaaaa(sh, sl, ah, al, bh, bl) \
737 DWunion __s, __a, __b; \
738 __a.s.low = (al); __a.s.high = (ah); \
739 __b.s.low = (bl); __b.s.high = (bh); \
740 __s.ll = __a.ll + __b.ll; \
741 (sl) = __s.s.low; (sh) = __s.s.high; \
743 # define sub_ddmmss(sh, sl, ah, al, bh, bl) \
745 DWunion __s, __a, __b; \
746 __a.s.low = (al); __a.s.high = (ah); \
747 __b.s.low = (bl); __b.s.high = (bh); \
748 __s.ll = __a.ll - __b.ll; \
749 (sl) = __s.s.low; (sh) = __s.s.high; \
751 # define udiv_qrnnd(q, r, nh, nl, d) \
752 asm("divu %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
753 # define sdiv_qrnnd(q, r, nh, nl, d) \
754 asm("div %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
756 # define UDIV_TIME 38
759 #if defined (__mips__) && W_TYPE_SIZE == 32
760 #define umul_ppmm(w1, w0, u, v) \
762 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \
763 (w1) = (USItype) (__x >> 32); \
764 (w0) = (USItype) (__x); \
767 #define UDIV_TIME 100
769 #if (__mips == 32 || __mips == 64) && ! __mips16
770 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
771 #define COUNT_LEADING_ZEROS_0 32
773 #endif /* __mips__ */
775 #if defined (__ns32000__) && W_TYPE_SIZE == 32
776 #define umul_ppmm(w1, w0, u, v) \
777 ({union {UDItype __ll; \
778 struct {USItype __l, __h;} __i; \
780 __asm__ ("meid %2,%0" \
782 : "%0" ((USItype) (u)), \
783 "g" ((USItype) (v))); \
784 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
785 #define __umulsidi3(u, v) \
787 __asm__ ("meid %2,%0" \
789 : "%0" ((USItype) (u)), \
790 "g" ((USItype) (v))); \
792 #define udiv_qrnnd(q, r, n1, n0, d) \
793 ({union {UDItype __ll; \
794 struct {USItype __l, __h;} __i; \
796 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
797 __asm__ ("deid %2,%0" \
800 "g" ((USItype) (d))); \
801 (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
802 #define count_trailing_zeros(count,x) \
804 __asm__ ("ffsd %2,%0" \
805 : "=r" ((USItype) (count)) \
806 : "0" ((USItype) 0), \
807 "r" ((USItype) (x))); \
809 #endif /* __ns32000__ */
811 /* FIXME: We should test _IBMR2 here when we add assembly support for the
812 system vendor compilers.
813 FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good
814 enough, since that hits ARM and m68k too. */
815 #if (defined (_ARCH_PPC) /* AIX */ \
816 || defined (_ARCH_PWR) /* AIX */ \
817 || defined (_ARCH_COM) /* AIX */ \
818 || defined (__powerpc__) /* gcc */ \
819 || defined (__POWERPC__) /* BEOS */ \
820 || defined (__ppc__) /* Darwin */ \
821 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
822 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
823 && CPU_FAMILY == PPC) \
824 ) && W_TYPE_SIZE == 32
825 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
827 if (__builtin_constant_p (bh) && (bh) == 0) \
828 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
829 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
830 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
831 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
832 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
834 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
835 : "=r" (sh), "=&r" (sl) \
836 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
838 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
840 if (__builtin_constant_p (ah) && (ah) == 0) \
841 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
842 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
843 else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \
844 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
845 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
846 else if (__builtin_constant_p (bh) && (bh) == 0) \
847 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
848 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
849 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
850 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
851 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
853 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
854 : "=r" (sh), "=&r" (sl) \
855 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
857 #define count_leading_zeros(count, x) \
858 __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x))
859 #define COUNT_LEADING_ZEROS_0 32
860 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
861 || defined (__ppc__) \
862 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
863 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
864 && CPU_FAMILY == PPC)
865 #define umul_ppmm(ph, pl, m0, m1) \
867 USItype __m0 = (m0), __m1 = (m1); \
868 __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
869 (pl) = __m0 * __m1; \
872 #define smul_ppmm(ph, pl, m0, m1) \
874 SItype __m0 = (m0), __m1 = (m1); \
875 __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
876 (pl) = __m0 * __m1; \
879 #define UDIV_TIME 120
880 #elif defined (_ARCH_PWR)
882 #define smul_ppmm(xh, xl, m0, m1) \
883 __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1))
885 #define sdiv_qrnnd(q, r, nh, nl, d) \
886 __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d))
887 #define UDIV_TIME 100
889 #endif /* 32-bit POWER architecture variants. */
891 /* We should test _IBMR2 here when we add assembly support for the system
893 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
894 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
896 if (__builtin_constant_p (bh) && (bh) == 0) \
897 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
898 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
899 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
900 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
901 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
903 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
904 : "=r" (sh), "=&r" (sl) \
905 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
907 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
909 if (__builtin_constant_p (ah) && (ah) == 0) \
910 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
911 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
912 else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \
913 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
914 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
915 else if (__builtin_constant_p (bh) && (bh) == 0) \
916 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
917 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
918 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
919 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
920 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
922 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
923 : "=r" (sh), "=&r" (sl) \
924 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
926 #define count_leading_zeros(count, x) \
927 __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
928 #define COUNT_LEADING_ZEROS_0 64
929 #define umul_ppmm(ph, pl, m0, m1) \
931 UDItype __m0 = (m0), __m1 = (m1); \
932 __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
933 (pl) = __m0 * __m1; \
936 #define smul_ppmm(ph, pl, m0, m1) \
938 DItype __m0 = (m0), __m1 = (m1); \
939 __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
940 (pl) = __m0 * __m1; \
942 #define SMUL_TIME 14 /* ??? */
943 #define UDIV_TIME 120 /* ??? */
944 #endif /* 64-bit PowerPC. */
946 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
947 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
948 __asm__ ("a %1,%5\n\tae %0,%3" \
949 : "=r" ((USItype) (sh)), \
950 "=&r" ((USItype) (sl)) \
951 : "%0" ((USItype) (ah)), \
952 "r" ((USItype) (bh)), \
953 "%1" ((USItype) (al)), \
954 "r" ((USItype) (bl)))
955 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
956 __asm__ ("s %1,%5\n\tse %0,%3" \
957 : "=r" ((USItype) (sh)), \
958 "=&r" ((USItype) (sl)) \
959 : "0" ((USItype) (ah)), \
960 "r" ((USItype) (bh)), \
961 "1" ((USItype) (al)), \
962 "r" ((USItype) (bl)))
963 #define umul_ppmm(ph, pl, m0, m1) \
965 USItype __m0 = (m0), __m1 = (m1); \
987 : "=r" ((USItype) (ph)), \
988 "=r" ((USItype) (pl)) \
992 (ph) += ((((SItype) __m0 >> 31) & __m1) \
993 + (((SItype) __m1 >> 31) & __m0)); \
996 #define UDIV_TIME 200
997 #define count_leading_zeros(count, x) \
999 if ((x) >= 0x10000) \
1000 __asm__ ("clz %0,%1" \
1001 : "=r" ((USItype) (count)) \
1002 : "r" ((USItype) (x) >> 16)); \
1005 __asm__ ("clz %0,%1" \
1006 : "=r" ((USItype) (count)) \
1007 : "r" ((USItype) (x))); \
1013 #if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
1015 #define umul_ppmm(w1, w0, u, v) \
1017 "dmulu.l %2,%3\n\tsts%M1 macl,%1\n\tsts%M0 mach,%0" \
1018 : "=r<" ((USItype)(w1)), \
1019 "=r<" ((USItype)(w0)) \
1020 : "r" ((USItype)(u)), \
1021 "r" ((USItype)(v)) \
1026 /* This is the same algorithm as __udiv_qrnnd_c. */
1027 #define UDIV_NEEDS_NORMALIZATION 1
1029 #define udiv_qrnnd(q, r, n1, n0, d) \
1031 extern UWtype __udiv_qrnnd_16 (UWtype, UWtype) \
1032 __attribute__ ((visibility ("hidden"))); \
1033 /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */ \
1044 : "=r" (q), "=&z" (r) \
1045 : "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16) \
1046 : "r1", "r2", "r4", "r5", "r6", "pr", "t"); \
1049 #define UDIV_TIME 80
1051 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1052 __asm__ ("clrt;subc %5,%1; subc %4,%0" \
1053 : "=r" (sh), "=r" (sl) \
1054 : "0" (ah), "1" (al), "r" (bh), "r" (bl) : "t")
1058 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
1059 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1060 #define count_leading_zeros(count, x) \
1063 UDItype x_ = (USItype)(x); \
1066 __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \
1067 (count) = c_ - 31; \
1070 #define COUNT_LEADING_ZEROS_0 32
1073 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
1074 && W_TYPE_SIZE == 32
1075 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1076 __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \
1077 : "=r" ((USItype) (sh)), \
1078 "=&r" ((USItype) (sl)) \
1079 : "%rJ" ((USItype) (ah)), \
1080 "rI" ((USItype) (bh)), \
1081 "%rJ" ((USItype) (al)), \
1082 "rI" ((USItype) (bl)) \
1084 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1085 __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \
1086 : "=r" ((USItype) (sh)), \
1087 "=&r" ((USItype) (sl)) \
1088 : "rJ" ((USItype) (ah)), \
1089 "rI" ((USItype) (bh)), \
1090 "rJ" ((USItype) (al)), \
1091 "rI" ((USItype) (bl)) \
1093 #if defined (__sparc_v8__)
1094 #define umul_ppmm(w1, w0, u, v) \
1095 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1096 : "=r" ((USItype) (w1)), \
1097 "=r" ((USItype) (w0)) \
1098 : "r" ((USItype) (u)), \
1099 "r" ((USItype) (v)))
1100 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1101 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
1102 : "=&r" ((USItype) (__q)), \
1103 "=&r" ((USItype) (__r)) \
1104 : "r" ((USItype) (__n1)), \
1105 "r" ((USItype) (__n0)), \
1106 "r" ((USItype) (__d)))
1108 #if defined (__sparclite__)
1109 /* This has hardware multiply but not divide. It also has two additional
1110 instructions scan (ffs from high bit) and divscc. */
1111 #define umul_ppmm(w1, w0, u, v) \
1112 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1113 : "=r" ((USItype) (w1)), \
1114 "=r" ((USItype) (w0)) \
1115 : "r" ((USItype) (u)), \
1116 "r" ((USItype) (v)))
1117 #define udiv_qrnnd(q, r, n1, n0, d) \
1118 __asm__ ("! Inlined udiv_qrnnd\n" \
1119 " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
1121 " divscc %3,%4,%%g1\n" \
1122 " divscc %%g1,%4,%%g1\n" \
1123 " divscc %%g1,%4,%%g1\n" \
1124 " divscc %%g1,%4,%%g1\n" \
1125 " divscc %%g1,%4,%%g1\n" \
1126 " divscc %%g1,%4,%%g1\n" \
1127 " divscc %%g1,%4,%%g1\n" \
1128 " divscc %%g1,%4,%%g1\n" \
1129 " divscc %%g1,%4,%%g1\n" \
1130 " divscc %%g1,%4,%%g1\n" \
1131 " divscc %%g1,%4,%%g1\n" \
1132 " divscc %%g1,%4,%%g1\n" \
1133 " divscc %%g1,%4,%%g1\n" \
1134 " divscc %%g1,%4,%%g1\n" \
1135 " divscc %%g1,%4,%%g1\n" \
1136 " divscc %%g1,%4,%%g1\n" \
1137 " divscc %%g1,%4,%%g1\n" \
1138 " divscc %%g1,%4,%%g1\n" \
1139 " divscc %%g1,%4,%%g1\n" \
1140 " divscc %%g1,%4,%%g1\n" \
1141 " divscc %%g1,%4,%%g1\n" \
1142 " divscc %%g1,%4,%%g1\n" \
1143 " divscc %%g1,%4,%%g1\n" \
1144 " divscc %%g1,%4,%%g1\n" \
1145 " divscc %%g1,%4,%%g1\n" \
1146 " divscc %%g1,%4,%%g1\n" \
1147 " divscc %%g1,%4,%%g1\n" \
1148 " divscc %%g1,%4,%%g1\n" \
1149 " divscc %%g1,%4,%%g1\n" \
1150 " divscc %%g1,%4,%%g1\n" \
1151 " divscc %%g1,%4,%%g1\n" \
1152 " divscc %%g1,%4,%0\n" \
1156 "1: ! End of inline udiv_qrnnd" \
1157 : "=r" ((USItype) (q)), \
1158 "=r" ((USItype) (r)) \
1159 : "r" ((USItype) (n1)), \
1160 "r" ((USItype) (n0)), \
1161 "rI" ((USItype) (d)) \
1162 : "g1" __AND_CLOBBER_CC)
1163 #define UDIV_TIME 37
1164 #define count_leading_zeros(count, x) \
1166 __asm__ ("scan %1,1,%0" \
1167 : "=r" ((USItype) (count)) \
1168 : "r" ((USItype) (x))); \
1170 /* Early sparclites return 63 for an argument of 0, but they warn that future
1171 implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0
1174 /* SPARC without integer multiplication and divide instructions.
1175 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
1176 #define umul_ppmm(w1, w0, u, v) \
1177 __asm__ ("! Inlined umul_ppmm\n" \
1178 " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\
1179 " sra %3,31,%%o5 ! Don't move this insn\n" \
1180 " and %2,%%o5,%%o5 ! Don't move this insn\n" \
1181 " andcc %%g0,0,%%g1 ! Don't move this insn\n" \
1182 " mulscc %%g1,%3,%%g1\n" \
1183 " mulscc %%g1,%3,%%g1\n" \
1184 " mulscc %%g1,%3,%%g1\n" \
1185 " mulscc %%g1,%3,%%g1\n" \
1186 " mulscc %%g1,%3,%%g1\n" \
1187 " mulscc %%g1,%3,%%g1\n" \
1188 " mulscc %%g1,%3,%%g1\n" \
1189 " mulscc %%g1,%3,%%g1\n" \
1190 " mulscc %%g1,%3,%%g1\n" \
1191 " mulscc %%g1,%3,%%g1\n" \
1192 " mulscc %%g1,%3,%%g1\n" \
1193 " mulscc %%g1,%3,%%g1\n" \
1194 " mulscc %%g1,%3,%%g1\n" \
1195 " mulscc %%g1,%3,%%g1\n" \
1196 " mulscc %%g1,%3,%%g1\n" \
1197 " mulscc %%g1,%3,%%g1\n" \
1198 " mulscc %%g1,%3,%%g1\n" \
1199 " mulscc %%g1,%3,%%g1\n" \
1200 " mulscc %%g1,%3,%%g1\n" \
1201 " mulscc %%g1,%3,%%g1\n" \
1202 " mulscc %%g1,%3,%%g1\n" \
1203 " mulscc %%g1,%3,%%g1\n" \
1204 " mulscc %%g1,%3,%%g1\n" \
1205 " mulscc %%g1,%3,%%g1\n" \
1206 " mulscc %%g1,%3,%%g1\n" \
1207 " mulscc %%g1,%3,%%g1\n" \
1208 " mulscc %%g1,%3,%%g1\n" \
1209 " mulscc %%g1,%3,%%g1\n" \
1210 " mulscc %%g1,%3,%%g1\n" \
1211 " mulscc %%g1,%3,%%g1\n" \
1212 " mulscc %%g1,%3,%%g1\n" \
1213 " mulscc %%g1,%3,%%g1\n" \
1214 " mulscc %%g1,0,%%g1\n" \
1215 " add %%g1,%%o5,%0\n" \
1217 : "=r" ((USItype) (w1)), \
1218 "=r" ((USItype) (w0)) \
1219 : "%rI" ((USItype) (u)), \
1220 "r" ((USItype) (v)) \
1221 : "g1", "o5" __AND_CLOBBER_CC)
1222 #define UMUL_TIME 39 /* 39 instructions */
1223 /* It's quite necessary to add this much assembler for the sparc.
1224 The default udiv_qrnnd (in C) is more than 10 times slower! */
1225 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1226 __asm__ ("! Inlined udiv_qrnnd\n" \
1228 " subcc %1,%2,%%g0\n" \
1230 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1231 " sub %1,%2,%1 ! this kills msb of n\n" \
1232 " addx %1,%1,%1 ! so this can't give carry\n" \
1233 " subcc %%g1,1,%%g1\n" \
1235 " subcc %1,%2,%%g0\n" \
1237 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1239 " sub %1,%2,%1 ! this kills msb of n\n" \
1240 "4: sub %1,%2,%1\n" \
1241 "5: addxcc %1,%1,%1\n" \
1243 " subcc %%g1,1,%%g1\n" \
1244 "! Got carry from n. Subtract next step to cancel this carry.\n" \
1246 " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \
1248 "3: xnor %0,0,%0\n" \
1249 " ! End of inline udiv_qrnnd" \
1250 : "=&r" ((USItype) (__q)), \
1251 "=&r" ((USItype) (__r)) \
1252 : "r" ((USItype) (__d)), \
1253 "1" ((USItype) (__n1)), \
1254 "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1255 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1256 #endif /* __sparclite__ */
1257 #endif /* __sparc_v8__ */
1258 #endif /* sparc32 */
1260 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1261 && W_TYPE_SIZE == 64
1262 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1263 __asm__ ("addcc %r4,%5,%1\n\t" \
1264 "add %r2,%3,%0\n\t" \
1265 "bcs,a,pn %%xcc, 1f\n\t" \
1268 : "=r" ((UDItype)(sh)), \
1269 "=&r" ((UDItype)(sl)) \
1270 : "%rJ" ((UDItype)(ah)), \
1271 "rI" ((UDItype)(bh)), \
1272 "%rJ" ((UDItype)(al)), \
1273 "rI" ((UDItype)(bl)) \
1276 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1277 __asm__ ("subcc %r4,%5,%1\n\t" \
1278 "sub %r2,%3,%0\n\t" \
1279 "bcs,a,pn %%xcc, 1f\n\t" \
1280 "sub %0, 1, %0\n\t" \
1282 : "=r" ((UDItype)(sh)), \
1283 "=&r" ((UDItype)(sl)) \
1284 : "rJ" ((UDItype)(ah)), \
1285 "rI" ((UDItype)(bh)), \
1286 "rJ" ((UDItype)(al)), \
1287 "rI" ((UDItype)(bl)) \
1290 #define umul_ppmm(wh, wl, u, v) \
1292 UDItype tmp1, tmp2, tmp3, tmp4; \
1293 __asm__ __volatile__ ( \
1295 "mulx %3,%6,%1\n\t" \
1296 "srlx %6,32,%2\n\t" \
1297 "mulx %2,%3,%4\n\t" \
1298 "sllx %4,32,%5\n\t" \
1300 "sub %1,%5,%5\n\t" \
1301 "srlx %5,32,%5\n\t" \
1302 "addcc %4,%5,%4\n\t" \
1303 "srlx %7,32,%5\n\t" \
1304 "mulx %3,%5,%3\n\t" \
1305 "mulx %2,%5,%5\n\t" \
1306 "sethi %%hi(0x80000000),%2\n\t" \
1307 "addcc %4,%3,%4\n\t" \
1308 "srlx %4,32,%4\n\t" \
1309 "add %2,%2,%2\n\t" \
1310 "movcc %%xcc,%%g0,%2\n\t" \
1311 "addcc %5,%4,%5\n\t" \
1312 "sllx %3,32,%3\n\t" \
1313 "add %1,%3,%1\n\t" \
1315 : "=r" ((UDItype)(wh)), \
1316 "=&r" ((UDItype)(wl)), \
1317 "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \
1318 : "r" ((UDItype)(u)), \
1319 "r" ((UDItype)(v)) \
1322 #define UMUL_TIME 96
1323 #define UDIV_TIME 230
1324 #endif /* sparc64 */
1326 #if defined (__vax__) && W_TYPE_SIZE == 32
1327 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1328 __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \
1329 : "=g" ((USItype) (sh)), \
1330 "=&g" ((USItype) (sl)) \
1331 : "%0" ((USItype) (ah)), \
1332 "g" ((USItype) (bh)), \
1333 "%1" ((USItype) (al)), \
1334 "g" ((USItype) (bl)))
1335 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1336 __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \
1337 : "=g" ((USItype) (sh)), \
1338 "=&g" ((USItype) (sl)) \
1339 : "0" ((USItype) (ah)), \
1340 "g" ((USItype) (bh)), \
1341 "1" ((USItype) (al)), \
1342 "g" ((USItype) (bl)))
1343 #define umul_ppmm(xh, xl, m0, m1) \
1347 struct {USItype __l, __h;} __i; \
1349 USItype __m0 = (m0), __m1 = (m1); \
1350 __asm__ ("emul %1,%2,$0,%0" \
1351 : "=r" (__xx.__ll) \
1354 (xh) = __xx.__i.__h; \
1355 (xl) = __xx.__i.__l; \
1356 (xh) += ((((SItype) __m0 >> 31) & __m1) \
1357 + (((SItype) __m1 >> 31) & __m0)); \
1359 #define sdiv_qrnnd(q, r, n1, n0, d) \
1361 union {DItype __ll; \
1362 struct {SItype __l, __h;} __i; \
1364 __xx.__i.__h = n1; __xx.__i.__l = n0; \
1365 __asm__ ("ediv %3,%2,%0,%1" \
1366 : "=g" (q), "=g" (r) \
1367 : "g" (__xx.__ll), "g" (d)); \
1369 #endif /* __vax__ */
1371 #if defined (__xtensa__) && W_TYPE_SIZE == 32
1372 /* This code is not Xtensa-configuration-specific, so rely on the compiler
1373 to expand builtin functions depending on what configuration features
1374 are available. This avoids library calls when the operation can be
1375 performed in-line. */
1376 #define umul_ppmm(w1, w0, u, v) \
1379 __w.ll = __builtin_umulsidi3 (u, v); \
1383 #define __umulsidi3(u, v) __builtin_umulsidi3 (u, v)
1384 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
1385 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
1386 #endif /* __xtensa__ */
1388 #if defined xstormy16
1389 extern UHItype __stormy16_count_leading_zeros (UHItype);
1390 #define count_leading_zeros(count, x) \
1395 /* We assume that W_TYPE_SIZE is a multiple of 16... */ \
1396 for ((count) = 0, size = W_TYPE_SIZE; size; size -= 16) \
1400 c = __clzhi2 ((x) >> (size - 16)); \
1407 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1410 #if defined (__z8000__) && W_TYPE_SIZE == 16
1411 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1412 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1413 : "=r" ((unsigned int)(sh)), \
1414 "=&r" ((unsigned int)(sl)) \
1415 : "%0" ((unsigned int)(ah)), \
1416 "r" ((unsigned int)(bh)), \
1417 "%1" ((unsigned int)(al)), \
1418 "rQR" ((unsigned int)(bl)))
1419 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1420 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1421 : "=r" ((unsigned int)(sh)), \
1422 "=&r" ((unsigned int)(sl)) \
1423 : "0" ((unsigned int)(ah)), \
1424 "r" ((unsigned int)(bh)), \
1425 "1" ((unsigned int)(al)), \
1426 "rQR" ((unsigned int)(bl)))
1427 #define umul_ppmm(xh, xl, m0, m1) \
1429 union {long int __ll; \
1430 struct {unsigned int __h, __l;} __i; \
1432 unsigned int __m0 = (m0), __m1 = (m1); \
1433 __asm__ ("mult %S0,%H3" \
1434 : "=r" (__xx.__i.__h), \
1435 "=r" (__xx.__i.__l) \
1438 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1439 (xh) += ((((signed int) __m0 >> 15) & __m1) \
1440 + (((signed int) __m1 >> 15) & __m0)); \
1442 #endif /* __z8000__ */
1444 #endif /* __GNUC__ */
1446 /* If this machine has no inline assembler, use C macros. */
1448 #if !defined (add_ssaaaa)
1449 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1452 __x = (al) + (bl); \
1453 (sh) = (ah) + (bh) + (__x < (al)); \
1458 #if !defined (sub_ddmmss)
1459 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1462 __x = (al) - (bl); \
1463 (sh) = (ah) - (bh) - (__x > (al)); \
1468 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1470 #if !defined (umul_ppmm) && defined (smul_ppmm)
1471 #define umul_ppmm(w1, w0, u, v) \
1474 UWtype __xm0 = (u), __xm1 = (v); \
1475 smul_ppmm (__w1, w0, __xm0, __xm1); \
1476 (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \
1477 + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \
1481 /* If we still don't have umul_ppmm, define it using plain C. */
1482 #if !defined (umul_ppmm)
1483 #define umul_ppmm(w1, w0, u, v) \
1485 UWtype __x0, __x1, __x2, __x3; \
1486 UHWtype __ul, __vl, __uh, __vh; \
1488 __ul = __ll_lowpart (u); \
1489 __uh = __ll_highpart (u); \
1490 __vl = __ll_lowpart (v); \
1491 __vh = __ll_highpart (v); \
1493 __x0 = (UWtype) __ul * __vl; \
1494 __x1 = (UWtype) __ul * __vh; \
1495 __x2 = (UWtype) __uh * __vl; \
1496 __x3 = (UWtype) __uh * __vh; \
1498 __x1 += __ll_highpart (__x0);/* this can't give carry */ \
1499 __x1 += __x2; /* but this indeed can */ \
1500 if (__x1 < __x2) /* did we get it? */ \
1501 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1503 (w1) = __x3 + __ll_highpart (__x1); \
1504 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
1508 #if !defined (__umulsidi3)
1509 #define __umulsidi3(u, v) \
1511 umul_ppmm (__w.s.high, __w.s.low, u, v); \
1515 /* Define this unconditionally, so it can be used for debugging. */
1516 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1518 UWtype __d1, __d0, __q1, __q0; \
1519 UWtype __r1, __r0, __m; \
1520 __d1 = __ll_highpart (d); \
1521 __d0 = __ll_lowpart (d); \
1523 __r1 = (n1) % __d1; \
1524 __q1 = (n1) / __d1; \
1525 __m = (UWtype) __q1 * __d0; \
1526 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
1529 __q1--, __r1 += (d); \
1530 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1532 __q1--, __r1 += (d); \
1536 __r0 = __r1 % __d1; \
1537 __q0 = __r1 / __d1; \
1538 __m = (UWtype) __q0 * __d0; \
1539 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
1542 __q0--, __r0 += (d); \
1545 __q0--, __r0 += (d); \
1549 (q) = (UWtype) __q1 * __ll_B | __q0; \
1553 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1554 __udiv_w_sdiv (defined in libgcc or elsewhere). */
1555 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1556 #define udiv_qrnnd(q, r, nh, nl, d) \
1559 (q) = __udiv_w_sdiv (&__r, nh, nl, d); \
1564 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1565 #if !defined (udiv_qrnnd)
1566 #define UDIV_NEEDS_NORMALIZATION 1
1567 #define udiv_qrnnd __udiv_qrnnd_c
1570 #if !defined (count_leading_zeros)
1571 #define count_leading_zeros(count, x) \
1573 UWtype __xr = (x); \
1576 if (W_TYPE_SIZE <= 32) \
1578 __a = __xr < ((UWtype)1<<2*__BITS4) \
1579 ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \
1580 : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1584 for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1585 if (((__xr >> __a) & 0xff) != 0) \
1589 (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1591 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1594 #if !defined (count_trailing_zeros)
1595 /* Define count_trailing_zeros using count_leading_zeros. The latter might be
1596 defined in asm, but if it is not, the C version above is good enough. */
1597 #define count_trailing_zeros(count, x) \
1599 UWtype __ctz_x = (x); \
1601 count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \
1602 (count) = W_TYPE_SIZE - 1 - __ctz_c; \
1606 #ifndef UDIV_NEEDS_NORMALIZATION
1607 #define UDIV_NEEDS_NORMALIZATION 0
OSDN Git Service
