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, 2011
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 (__AVR__)
255 #if W_TYPE_SIZE == 16
256 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
257 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X))
258 #define COUNT_LEADING_ZEROS_0 16
259 #endif /* W_TYPE_SIZE == 16 */
261 #if W_TYPE_SIZE == 32
262 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X))
263 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
264 #define COUNT_LEADING_ZEROS_0 32
265 #endif /* W_TYPE_SIZE == 32 */
267 #if W_TYPE_SIZE == 64
268 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzll (X))
269 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzll (X))
270 #define COUNT_LEADING_ZEROS_0 64
271 #endif /* W_TYPE_SIZE == 64 */
273 #endif /* defined (__AVR__) */
275 #if defined (__CRIS__) && __CRIS_arch_version >= 3
276 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
277 #if __CRIS_arch_version >= 8
278 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
280 #endif /* __CRIS__ */
282 #if defined (__hppa) && W_TYPE_SIZE == 32
283 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
284 __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \
285 : "=r" ((USItype) (sh)), \
286 "=&r" ((USItype) (sl)) \
287 : "%rM" ((USItype) (ah)), \
288 "rM" ((USItype) (bh)), \
289 "%rM" ((USItype) (al)), \
290 "rM" ((USItype) (bl)))
291 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
292 __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \
293 : "=r" ((USItype) (sh)), \
294 "=&r" ((USItype) (sl)) \
295 : "rM" ((USItype) (ah)), \
296 "rM" ((USItype) (bh)), \
297 "rM" ((USItype) (al)), \
298 "rM" ((USItype) (bl)))
299 #if defined (_PA_RISC1_1)
300 #define umul_ppmm(w1, w0, u, v) \
305 struct {USItype __w1, __w0;} __w1w0; \
307 __asm__ ("xmpyu %1,%2,%0" \
309 : "x" ((USItype) (u)), \
310 "x" ((USItype) (v))); \
311 (w1) = __t.__w1w0.__w1; \
312 (w0) = __t.__w1w0.__w0; \
319 #define count_leading_zeros(count, x) \
324 " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \
325 " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\
326 " ldo 16(%0),%0 ; Yes. Perform add.\n" \
327 " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \
328 " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\
329 " ldo 8(%0),%0 ; Yes. Perform add.\n" \
330 " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \
331 " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\
332 " ldo 4(%0),%0 ; Yes. Perform add.\n" \
333 " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \
334 " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\
335 " ldo 2(%0),%0 ; Yes. Perform add.\n" \
336 " extru %1,30,1,%1 ; Extract bit 1.\n" \
337 " sub %0,%1,%0 ; Subtract it.\n" \
338 : "=r" (count), "=r" (__tmp) : "1" (x)); \
342 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
343 #if !defined (__zarch__)
344 #define smul_ppmm(xh, xl, m0, m1) \
346 union {DItype __ll; \
347 struct {USItype __h, __l;} __i; \
349 __asm__ ("lr %N0,%1\n\tmr %0,%2" \
351 : "r" (m0), "r" (m1)); \
352 (xh) = __x.__i.__h; (xl) = __x.__i.__l; \
354 #define sdiv_qrnnd(q, r, n1, n0, d) \
356 union {DItype __ll; \
357 struct {USItype __h, __l;} __i; \
359 __x.__i.__h = n1; __x.__i.__l = n0; \
360 __asm__ ("dr %0,%2" \
362 : "0" (__x.__ll), "r" (d)); \
363 (q) = __x.__i.__l; (r) = __x.__i.__h; \
366 #define smul_ppmm(xh, xl, m0, m1) \
368 register SItype r0 __asm__ ("0"); \
369 register SItype r1 __asm__ ("1") = m0; \
371 __asm__ ("mr\t%%r0,%3" \
372 : "=r" (r0), "=r" (r1) \
373 : "r" (r1), "r" (m1)); \
374 (xh) = r0; (xl) = r1; \
376 #define sdiv_qrnnd(q, r, n1, n0, d) \
378 register SItype r0 __asm__ ("0") = n0; \
379 register SItype r1 __asm__ ("1") = n1; \
381 __asm__ ("dr\t%%r0,%3" \
382 : "=r" (r0), "=r" (r1) \
383 : "r" (r0), "r" (r1), "r" (d)); \
384 (q) = r0; (r) = r1; \
386 #endif /* __zarch__ */
389 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
390 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
391 __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}" \
392 : "=r" ((USItype) (sh)), \
393 "=&r" ((USItype) (sl)) \
394 : "%0" ((USItype) (ah)), \
395 "g" ((USItype) (bh)), \
396 "%1" ((USItype) (al)), \
397 "g" ((USItype) (bl)))
398 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
399 __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}" \
400 : "=r" ((USItype) (sh)), \
401 "=&r" ((USItype) (sl)) \
402 : "0" ((USItype) (ah)), \
403 "g" ((USItype) (bh)), \
404 "1" ((USItype) (al)), \
405 "g" ((USItype) (bl)))
406 #define umul_ppmm(w1, w0, u, v) \
407 __asm__ ("mul{l} %3" \
408 : "=a" ((USItype) (w0)), \
409 "=d" ((USItype) (w1)) \
410 : "%0" ((USItype) (u)), \
411 "rm" ((USItype) (v)))
412 #define udiv_qrnnd(q, r, n1, n0, dv) \
413 __asm__ ("div{l} %4" \
414 : "=a" ((USItype) (q)), \
415 "=d" ((USItype) (r)) \
416 : "0" ((USItype) (n0)), \
417 "1" ((USItype) (n1)), \
418 "rm" ((USItype) (dv)))
419 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x))
420 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x))
425 #if (defined (__x86_64__) || defined (__i386__)) && W_TYPE_SIZE == 64
426 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
427 __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}" \
428 : "=r" ((UDItype) (sh)), \
429 "=&r" ((UDItype) (sl)) \
430 : "%0" ((UDItype) (ah)), \
431 "rme" ((UDItype) (bh)), \
432 "%1" ((UDItype) (al)), \
433 "rme" ((UDItype) (bl)))
434 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
435 __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}" \
436 : "=r" ((UDItype) (sh)), \
437 "=&r" ((UDItype) (sl)) \
438 : "0" ((UDItype) (ah)), \
439 "rme" ((UDItype) (bh)), \
440 "1" ((UDItype) (al)), \
441 "rme" ((UDItype) (bl)))
442 #define umul_ppmm(w1, w0, u, v) \
443 __asm__ ("mul{q} %3" \
444 : "=a" ((UDItype) (w0)), \
445 "=d" ((UDItype) (w1)) \
446 : "%0" ((UDItype) (u)), \
447 "rm" ((UDItype) (v)))
448 #define udiv_qrnnd(q, r, n1, n0, dv) \
449 __asm__ ("div{q} %4" \
450 : "=a" ((UDItype) (q)), \
451 "=d" ((UDItype) (r)) \
452 : "0" ((UDItype) (n0)), \
453 "1" ((UDItype) (n1)), \
454 "rm" ((UDItype) (dv)))
455 #define count_leading_zeros(count, x) ((count) = __builtin_clzll (x))
456 #define count_trailing_zeros(count, x) ((count) = __builtin_ctzll (x))
461 #if defined (__i960__) && W_TYPE_SIZE == 32
462 #define umul_ppmm(w1, w0, u, v) \
463 ({union {UDItype __ll; \
464 struct {USItype __l, __h;} __i; \
466 __asm__ ("emul %2,%1,%0" \
468 : "%dI" ((USItype) (u)), \
469 "dI" ((USItype) (v))); \
470 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
471 #define __umulsidi3(u, v) \
473 __asm__ ("emul %2,%1,%0" \
475 : "%dI" ((USItype) (u)), \
476 "dI" ((USItype) (v))); \
478 #endif /* __i960__ */
480 #if defined (__ia64) && W_TYPE_SIZE == 64
481 /* This form encourages gcc (pre-release 3.4 at least) to emit predicated
482 "sub r=r,r" and "sub r=r,r,1", giving a 2 cycle latency. The generic
483 code using "al<bl" arithmetically comes out making an actual 0 or 1 in a
484 register, which takes an extra cycle. */
485 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
490 (sh) = (ah) - (bh) - 1; \
492 (sh) = (ah) - (bh); \
496 /* Do both product parts in assembly, since that gives better code with
497 all gcc versions. Some callers will just use the upper part, and in
498 that situation we waste an instruction, but not any cycles. */
499 #define umul_ppmm(ph, pl, m0, m1) \
500 __asm__ ("xma.hu %0 = %2, %3, f0\n\txma.l %1 = %2, %3, f0" \
501 : "=&f" (ph), "=f" (pl) \
502 : "f" (m0), "f" (m1))
503 #define count_leading_zeros(count, x) \
505 UWtype _x = (x), _y, _a, _c; \
506 __asm__ ("mux1 %0 = %1, @rev" : "=r" (_y) : "r" (_x)); \
507 __asm__ ("czx1.l %0 = %1" : "=r" (_a) : "r" (-_y | _y)); \
508 _c = (_a - 1) << 3; \
515 (count) = W_TYPE_SIZE - 1 - _c; \
517 /* similar to what gcc does for __builtin_ffs, but 0 based rather than 1
518 based, and we don't need a special case for x==0 here */
519 #define count_trailing_zeros(count, x) \
521 UWtype __ctz_x = (x); \
522 __asm__ ("popcnt %0 = %1" \
524 : "r" ((__ctz_x-1) & ~__ctz_x)); \
529 #if defined (__M32R__) && W_TYPE_SIZE == 32
530 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
531 /* The cmp clears the condition bit. */ \
532 __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \
533 : "=r" ((USItype) (sh)), \
534 "=&r" ((USItype) (sl)) \
535 : "0" ((USItype) (ah)), \
536 "r" ((USItype) (bh)), \
537 "1" ((USItype) (al)), \
538 "r" ((USItype) (bl)) \
540 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
541 /* The cmp clears the condition bit. */ \
542 __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \
543 : "=r" ((USItype) (sh)), \
544 "=&r" ((USItype) (sl)) \
545 : "0" ((USItype) (ah)), \
546 "r" ((USItype) (bh)), \
547 "1" ((USItype) (al)), \
548 "r" ((USItype) (bl)) \
550 #endif /* __M32R__ */
552 #if defined (__mc68000__) && W_TYPE_SIZE == 32
553 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
554 __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \
555 : "=d" ((USItype) (sh)), \
556 "=&d" ((USItype) (sl)) \
557 : "%0" ((USItype) (ah)), \
558 "d" ((USItype) (bh)), \
559 "%1" ((USItype) (al)), \
560 "g" ((USItype) (bl)))
561 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
562 __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \
563 : "=d" ((USItype) (sh)), \
564 "=&d" ((USItype) (sl)) \
565 : "0" ((USItype) (ah)), \
566 "d" ((USItype) (bh)), \
567 "1" ((USItype) (al)), \
568 "g" ((USItype) (bl)))
570 /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r. */
571 #if (defined (__mc68020__) && !defined (__mc68060__))
572 #define umul_ppmm(w1, w0, u, v) \
573 __asm__ ("mulu%.l %3,%1:%0" \
574 : "=d" ((USItype) (w0)), \
575 "=d" ((USItype) (w1)) \
576 : "%0" ((USItype) (u)), \
577 "dmi" ((USItype) (v)))
579 #define udiv_qrnnd(q, r, n1, n0, d) \
580 __asm__ ("divu%.l %4,%1:%0" \
581 : "=d" ((USItype) (q)), \
582 "=d" ((USItype) (r)) \
583 : "0" ((USItype) (n0)), \
584 "1" ((USItype) (n1)), \
585 "dmi" ((USItype) (d)))
587 #define sdiv_qrnnd(q, r, n1, n0, d) \
588 __asm__ ("divs%.l %4,%1:%0" \
589 : "=d" ((USItype) (q)), \
590 "=d" ((USItype) (r)) \
591 : "0" ((USItype) (n0)), \
592 "1" ((USItype) (n1)), \
593 "dmi" ((USItype) (d)))
595 #elif defined (__mcoldfire__) /* not mc68020 */
597 #define umul_ppmm(xh, xl, a, b) \
598 __asm__ ("| Inlined umul_ppmm\n" \
599 " move%.l %2,%/d0\n" \
600 " move%.l %3,%/d1\n" \
601 " move%.l %/d0,%/d2\n" \
603 " move%.l %/d1,%/d3\n" \
605 " move%.w %/d2,%/d4\n" \
606 " mulu %/d3,%/d4\n" \
607 " mulu %/d1,%/d2\n" \
608 " mulu %/d0,%/d3\n" \
609 " mulu %/d0,%/d1\n" \
610 " move%.l %/d4,%/d0\n" \
613 " add%.l %/d0,%/d2\n" \
614 " add%.l %/d3,%/d2\n" \
616 " add%.l %#65536,%/d1\n" \
618 " moveq %#0,%/d0\n" \
619 " move%.w %/d2,%/d0\n" \
620 " move%.w %/d4,%/d2\n" \
621 " move%.l %/d2,%1\n" \
622 " add%.l %/d1,%/d0\n" \
624 : "=g" ((USItype) (xh)), \
625 "=g" ((USItype) (xl)) \
626 : "g" ((USItype) (a)), \
627 "g" ((USItype) (b)) \
628 : "d0", "d1", "d2", "d3", "d4")
629 #define UMUL_TIME 100
630 #define UDIV_TIME 400
631 #else /* not ColdFire */
632 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */
633 #define umul_ppmm(xh, xl, a, b) \
634 __asm__ ("| Inlined umul_ppmm\n" \
635 " move%.l %2,%/d0\n" \
636 " move%.l %3,%/d1\n" \
637 " move%.l %/d0,%/d2\n" \
639 " move%.l %/d1,%/d3\n" \
641 " move%.w %/d2,%/d4\n" \
642 " mulu %/d3,%/d4\n" \
643 " mulu %/d1,%/d2\n" \
644 " mulu %/d0,%/d3\n" \
645 " mulu %/d0,%/d1\n" \
646 " move%.l %/d4,%/d0\n" \
647 " eor%.w %/d0,%/d0\n" \
649 " add%.l %/d0,%/d2\n" \
650 " add%.l %/d3,%/d2\n" \
652 " add%.l %#65536,%/d1\n" \
654 " moveq %#0,%/d0\n" \
655 " move%.w %/d2,%/d0\n" \
656 " move%.w %/d4,%/d2\n" \
657 " move%.l %/d2,%1\n" \
658 " add%.l %/d1,%/d0\n" \
660 : "=g" ((USItype) (xh)), \
661 "=g" ((USItype) (xl)) \
662 : "g" ((USItype) (a)), \
663 "g" ((USItype) (b)) \
664 : "d0", "d1", "d2", "d3", "d4")
665 #define UMUL_TIME 100
666 #define UDIV_TIME 400
668 #endif /* not mc68020 */
670 /* The '020, '030, '040 and '060 have bitfield insns.
671 cpu32 disguises as a 68020, but lacks them. */
672 #if defined (__mc68020__) && !defined (__mcpu32__)
673 #define count_leading_zeros(count, x) \
674 __asm__ ("bfffo %1{%b2:%b2},%0" \
675 : "=d" ((USItype) (count)) \
676 : "od" ((USItype) (x)), "n" (0))
677 /* Some ColdFire architectures have a ff1 instruction supported via
679 #elif defined (__mcfisaaplus__) || defined (__mcfisac__)
680 #define count_leading_zeros(count,x) ((count) = __builtin_clz (x))
681 #define COUNT_LEADING_ZEROS_0 32
685 #if defined (__m88000__) && W_TYPE_SIZE == 32
686 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
687 __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \
688 : "=r" ((USItype) (sh)), \
689 "=&r" ((USItype) (sl)) \
690 : "%rJ" ((USItype) (ah)), \
691 "rJ" ((USItype) (bh)), \
692 "%rJ" ((USItype) (al)), \
693 "rJ" ((USItype) (bl)))
694 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
695 __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \
696 : "=r" ((USItype) (sh)), \
697 "=&r" ((USItype) (sl)) \
698 : "rJ" ((USItype) (ah)), \
699 "rJ" ((USItype) (bh)), \
700 "rJ" ((USItype) (al)), \
701 "rJ" ((USItype) (bl)))
702 #define count_leading_zeros(count, x) \
705 __asm__ ("ff1 %0,%1" \
707 : "r" ((USItype) (x))); \
708 (count) = __cbtmp ^ 31; \
710 #define COUNT_LEADING_ZEROS_0 63 /* sic */
711 #if defined (__mc88110__)
712 #define umul_ppmm(wh, wl, u, v) \
714 union {UDItype __ll; \
715 struct {USItype __h, __l;} __i; \
717 __asm__ ("mulu.d %0,%1,%2" \
719 : "r" ((USItype) (u)), \
720 "r" ((USItype) (v))); \
721 (wh) = __xx.__i.__h; \
722 (wl) = __xx.__i.__l; \
724 #define udiv_qrnnd(q, r, n1, n0, d) \
725 ({union {UDItype __ll; \
726 struct {USItype __h, __l;} __i; \
729 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
730 __asm__ ("divu.d %0,%1,%2" \
733 "r" ((USItype) (d))); \
734 (r) = (n0) - __q * (d); (q) = __q; })
739 #define UDIV_TIME 150
740 #endif /* __mc88110__ */
741 #endif /* __m88000__ */
743 #if defined (__mn10300__)
744 # if defined (__AM33__)
745 # define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
746 # define umul_ppmm(w1, w0, u, v) \
747 asm("mulu %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
748 # define smul_ppmm(w1, w0, u, v) \
749 asm("mul %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
751 # define umul_ppmm(w1, w0, u, v) \
752 asm("nop; nop; mulu %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
753 # define smul_ppmm(w1, w0, u, v) \
754 asm("nop; nop; mul %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
756 # define add_ssaaaa(sh, sl, ah, al, bh, bl) \
758 DWunion __s, __a, __b; \
759 __a.s.low = (al); __a.s.high = (ah); \
760 __b.s.low = (bl); __b.s.high = (bh); \
761 __s.ll = __a.ll + __b.ll; \
762 (sl) = __s.s.low; (sh) = __s.s.high; \
764 # define sub_ddmmss(sh, sl, ah, al, bh, bl) \
766 DWunion __s, __a, __b; \
767 __a.s.low = (al); __a.s.high = (ah); \
768 __b.s.low = (bl); __b.s.high = (bh); \
769 __s.ll = __a.ll - __b.ll; \
770 (sl) = __s.s.low; (sh) = __s.s.high; \
772 # define udiv_qrnnd(q, r, nh, nl, d) \
773 asm("divu %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
774 # define sdiv_qrnnd(q, r, nh, nl, d) \
775 asm("div %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
777 # define UDIV_TIME 38
780 #if defined (__mips__) && W_TYPE_SIZE == 32
781 #define umul_ppmm(w1, w0, u, v) \
783 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \
784 (w1) = (USItype) (__x >> 32); \
785 (w0) = (USItype) (__x); \
788 #define UDIV_TIME 100
790 #if (__mips == 32 || __mips == 64) && ! __mips16
791 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
792 #define COUNT_LEADING_ZEROS_0 32
794 #endif /* __mips__ */
796 #if defined (__ns32000__) && W_TYPE_SIZE == 32
797 #define umul_ppmm(w1, w0, u, v) \
798 ({union {UDItype __ll; \
799 struct {USItype __l, __h;} __i; \
801 __asm__ ("meid %2,%0" \
803 : "%0" ((USItype) (u)), \
804 "g" ((USItype) (v))); \
805 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
806 #define __umulsidi3(u, v) \
808 __asm__ ("meid %2,%0" \
810 : "%0" ((USItype) (u)), \
811 "g" ((USItype) (v))); \
813 #define udiv_qrnnd(q, r, n1, n0, d) \
814 ({union {UDItype __ll; \
815 struct {USItype __l, __h;} __i; \
817 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
818 __asm__ ("deid %2,%0" \
821 "g" ((USItype) (d))); \
822 (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
823 #define count_trailing_zeros(count,x) \
825 __asm__ ("ffsd %2,%0" \
826 : "=r" ((USItype) (count)) \
827 : "0" ((USItype) 0), \
828 "r" ((USItype) (x))); \
830 #endif /* __ns32000__ */
832 /* FIXME: We should test _IBMR2 here when we add assembly support for the
833 system vendor compilers.
834 FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good
835 enough, since that hits ARM and m68k too. */
836 #if (defined (_ARCH_PPC) /* AIX */ \
837 || defined (_ARCH_PWR) /* AIX */ \
838 || defined (_ARCH_COM) /* AIX */ \
839 || defined (__powerpc__) /* gcc */ \
840 || defined (__POWERPC__) /* BEOS */ \
841 || defined (__ppc__) /* Darwin */ \
842 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
843 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
844 && CPU_FAMILY == PPC) \
845 ) && W_TYPE_SIZE == 32
846 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
848 if (__builtin_constant_p (bh) && (bh) == 0) \
849 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
850 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
851 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
852 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
853 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
855 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
856 : "=r" (sh), "=&r" (sl) \
857 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
859 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
861 if (__builtin_constant_p (ah) && (ah) == 0) \
862 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
863 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
864 else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \
865 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
866 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
867 else if (__builtin_constant_p (bh) && (bh) == 0) \
868 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
869 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
870 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
871 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
872 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
874 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
875 : "=r" (sh), "=&r" (sl) \
876 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
878 #define count_leading_zeros(count, x) \
879 __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x))
880 #define COUNT_LEADING_ZEROS_0 32
881 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
882 || defined (__ppc__) \
883 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
884 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
885 && CPU_FAMILY == PPC)
886 #define umul_ppmm(ph, pl, m0, m1) \
888 USItype __m0 = (m0), __m1 = (m1); \
889 __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
890 (pl) = __m0 * __m1; \
893 #define smul_ppmm(ph, pl, m0, m1) \
895 SItype __m0 = (m0), __m1 = (m1); \
896 __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
897 (pl) = __m0 * __m1; \
900 #define UDIV_TIME 120
901 #elif defined (_ARCH_PWR)
903 #define smul_ppmm(xh, xl, m0, m1) \
904 __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1))
906 #define sdiv_qrnnd(q, r, nh, nl, d) \
907 __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d))
908 #define UDIV_TIME 100
910 #endif /* 32-bit POWER architecture variants. */
912 /* We should test _IBMR2 here when we add assembly support for the system
914 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
915 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
917 if (__builtin_constant_p (bh) && (bh) == 0) \
918 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
919 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
920 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
921 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
922 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
924 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
925 : "=r" (sh), "=&r" (sl) \
926 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
928 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
930 if (__builtin_constant_p (ah) && (ah) == 0) \
931 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
932 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
933 else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \
934 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
935 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
936 else if (__builtin_constant_p (bh) && (bh) == 0) \
937 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
938 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
939 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
940 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
941 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
943 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
944 : "=r" (sh), "=&r" (sl) \
945 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
947 #define count_leading_zeros(count, x) \
948 __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
949 #define COUNT_LEADING_ZEROS_0 64
950 #define umul_ppmm(ph, pl, m0, m1) \
952 UDItype __m0 = (m0), __m1 = (m1); \
953 __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
954 (pl) = __m0 * __m1; \
957 #define smul_ppmm(ph, pl, m0, m1) \
959 DItype __m0 = (m0), __m1 = (m1); \
960 __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
961 (pl) = __m0 * __m1; \
963 #define SMUL_TIME 14 /* ??? */
964 #define UDIV_TIME 120 /* ??? */
965 #endif /* 64-bit PowerPC. */
967 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
968 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
969 __asm__ ("a %1,%5\n\tae %0,%3" \
970 : "=r" ((USItype) (sh)), \
971 "=&r" ((USItype) (sl)) \
972 : "%0" ((USItype) (ah)), \
973 "r" ((USItype) (bh)), \
974 "%1" ((USItype) (al)), \
975 "r" ((USItype) (bl)))
976 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
977 __asm__ ("s %1,%5\n\tse %0,%3" \
978 : "=r" ((USItype) (sh)), \
979 "=&r" ((USItype) (sl)) \
980 : "0" ((USItype) (ah)), \
981 "r" ((USItype) (bh)), \
982 "1" ((USItype) (al)), \
983 "r" ((USItype) (bl)))
984 #define umul_ppmm(ph, pl, m0, m1) \
986 USItype __m0 = (m0), __m1 = (m1); \
1008 : "=r" ((USItype) (ph)), \
1009 "=r" ((USItype) (pl)) \
1013 (ph) += ((((SItype) __m0 >> 31) & __m1) \
1014 + (((SItype) __m1 >> 31) & __m0)); \
1016 #define UMUL_TIME 20
1017 #define UDIV_TIME 200
1018 #define count_leading_zeros(count, x) \
1020 if ((x) >= 0x10000) \
1021 __asm__ ("clz %0,%1" \
1022 : "=r" ((USItype) (count)) \
1023 : "r" ((USItype) (x) >> 16)); \
1026 __asm__ ("clz %0,%1" \
1027 : "=r" ((USItype) (count)) \
1028 : "r" ((USItype) (x))); \
1034 #if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
1036 #define umul_ppmm(w1, w0, u, v) \
1038 "dmulu.l %2,%3\n\tsts%M1 macl,%1\n\tsts%M0 mach,%0" \
1039 : "=r<" ((USItype)(w1)), \
1040 "=r<" ((USItype)(w0)) \
1041 : "r" ((USItype)(u)), \
1042 "r" ((USItype)(v)) \
1047 /* This is the same algorithm as __udiv_qrnnd_c. */
1048 #define UDIV_NEEDS_NORMALIZATION 1
1050 #define udiv_qrnnd(q, r, n1, n0, d) \
1052 extern UWtype __udiv_qrnnd_16 (UWtype, UWtype) \
1053 __attribute__ ((visibility ("hidden"))); \
1054 /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */ \
1065 : "=r" (q), "=&z" (r) \
1066 : "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16) \
1067 : "r1", "r2", "r4", "r5", "r6", "pr", "t"); \
1070 #define UDIV_TIME 80
1072 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1073 __asm__ ("clrt;subc %5,%1; subc %4,%0" \
1074 : "=r" (sh), "=r" (sl) \
1075 : "0" (ah), "1" (al), "r" (bh), "r" (bl) : "t")
1079 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
1080 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1081 #define count_leading_zeros(count, x) \
1084 UDItype x_ = (USItype)(x); \
1087 __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \
1088 (count) = c_ - 31; \
1091 #define COUNT_LEADING_ZEROS_0 32
1094 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
1095 && W_TYPE_SIZE == 32
1096 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1097 __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \
1098 : "=r" ((USItype) (sh)), \
1099 "=&r" ((USItype) (sl)) \
1100 : "%rJ" ((USItype) (ah)), \
1101 "rI" ((USItype) (bh)), \
1102 "%rJ" ((USItype) (al)), \
1103 "rI" ((USItype) (bl)) \
1105 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1106 __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \
1107 : "=r" ((USItype) (sh)), \
1108 "=&r" ((USItype) (sl)) \
1109 : "rJ" ((USItype) (ah)), \
1110 "rI" ((USItype) (bh)), \
1111 "rJ" ((USItype) (al)), \
1112 "rI" ((USItype) (bl)) \
1114 #if defined (__sparc_v8__)
1115 #define umul_ppmm(w1, w0, u, v) \
1116 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1117 : "=r" ((USItype) (w1)), \
1118 "=r" ((USItype) (w0)) \
1119 : "r" ((USItype) (u)), \
1120 "r" ((USItype) (v)))
1121 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1122 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
1123 : "=&r" ((USItype) (__q)), \
1124 "=&r" ((USItype) (__r)) \
1125 : "r" ((USItype) (__n1)), \
1126 "r" ((USItype) (__n0)), \
1127 "r" ((USItype) (__d)))
1129 #if defined (__sparclite__)
1130 /* This has hardware multiply but not divide. It also has two additional
1131 instructions scan (ffs from high bit) and divscc. */
1132 #define umul_ppmm(w1, w0, u, v) \
1133 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1134 : "=r" ((USItype) (w1)), \
1135 "=r" ((USItype) (w0)) \
1136 : "r" ((USItype) (u)), \
1137 "r" ((USItype) (v)))
1138 #define udiv_qrnnd(q, r, n1, n0, d) \
1139 __asm__ ("! Inlined udiv_qrnnd\n" \
1140 " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
1142 " divscc %3,%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,%%g1\n" \
1153 " divscc %%g1,%4,%%g1\n" \
1154 " divscc %%g1,%4,%%g1\n" \
1155 " divscc %%g1,%4,%%g1\n" \
1156 " divscc %%g1,%4,%%g1\n" \
1157 " divscc %%g1,%4,%%g1\n" \
1158 " divscc %%g1,%4,%%g1\n" \
1159 " divscc %%g1,%4,%%g1\n" \
1160 " divscc %%g1,%4,%%g1\n" \
1161 " divscc %%g1,%4,%%g1\n" \
1162 " divscc %%g1,%4,%%g1\n" \
1163 " divscc %%g1,%4,%%g1\n" \
1164 " divscc %%g1,%4,%%g1\n" \
1165 " divscc %%g1,%4,%%g1\n" \
1166 " divscc %%g1,%4,%%g1\n" \
1167 " divscc %%g1,%4,%%g1\n" \
1168 " divscc %%g1,%4,%%g1\n" \
1169 " divscc %%g1,%4,%%g1\n" \
1170 " divscc %%g1,%4,%%g1\n" \
1171 " divscc %%g1,%4,%%g1\n" \
1172 " divscc %%g1,%4,%%g1\n" \
1173 " divscc %%g1,%4,%0\n" \
1177 "1: ! End of inline udiv_qrnnd" \
1178 : "=r" ((USItype) (q)), \
1179 "=r" ((USItype) (r)) \
1180 : "r" ((USItype) (n1)), \
1181 "r" ((USItype) (n0)), \
1182 "rI" ((USItype) (d)) \
1183 : "g1" __AND_CLOBBER_CC)
1184 #define UDIV_TIME 37
1185 #define count_leading_zeros(count, x) \
1187 __asm__ ("scan %1,1,%0" \
1188 : "=r" ((USItype) (count)) \
1189 : "r" ((USItype) (x))); \
1191 /* Early sparclites return 63 for an argument of 0, but they warn that future
1192 implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0
1195 /* SPARC without integer multiplication and divide instructions.
1196 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
1197 #define umul_ppmm(w1, w0, u, v) \
1198 __asm__ ("! Inlined umul_ppmm\n" \
1199 " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\
1200 " sra %3,31,%%o5 ! Don't move this insn\n" \
1201 " and %2,%%o5,%%o5 ! Don't move this insn\n" \
1202 " andcc %%g0,0,%%g1 ! Don't move this insn\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,%3,%%g1\n" \
1215 " mulscc %%g1,%3,%%g1\n" \
1216 " mulscc %%g1,%3,%%g1\n" \
1217 " mulscc %%g1,%3,%%g1\n" \
1218 " mulscc %%g1,%3,%%g1\n" \
1219 " mulscc %%g1,%3,%%g1\n" \
1220 " mulscc %%g1,%3,%%g1\n" \
1221 " mulscc %%g1,%3,%%g1\n" \
1222 " mulscc %%g1,%3,%%g1\n" \
1223 " mulscc %%g1,%3,%%g1\n" \
1224 " mulscc %%g1,%3,%%g1\n" \
1225 " mulscc %%g1,%3,%%g1\n" \
1226 " mulscc %%g1,%3,%%g1\n" \
1227 " mulscc %%g1,%3,%%g1\n" \
1228 " mulscc %%g1,%3,%%g1\n" \
1229 " mulscc %%g1,%3,%%g1\n" \
1230 " mulscc %%g1,%3,%%g1\n" \
1231 " mulscc %%g1,%3,%%g1\n" \
1232 " mulscc %%g1,%3,%%g1\n" \
1233 " mulscc %%g1,%3,%%g1\n" \
1234 " mulscc %%g1,%3,%%g1\n" \
1235 " mulscc %%g1,0,%%g1\n" \
1236 " add %%g1,%%o5,%0\n" \
1238 : "=r" ((USItype) (w1)), \
1239 "=r" ((USItype) (w0)) \
1240 : "%rI" ((USItype) (u)), \
1241 "r" ((USItype) (v)) \
1242 : "g1", "o5" __AND_CLOBBER_CC)
1243 #define UMUL_TIME 39 /* 39 instructions */
1244 /* It's quite necessary to add this much assembler for the sparc.
1245 The default udiv_qrnnd (in C) is more than 10 times slower! */
1246 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1247 __asm__ ("! Inlined udiv_qrnnd\n" \
1249 " subcc %1,%2,%%g0\n" \
1251 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1252 " sub %1,%2,%1 ! this kills msb of n\n" \
1253 " addx %1,%1,%1 ! so this can't give carry\n" \
1254 " subcc %%g1,1,%%g1\n" \
1256 " subcc %1,%2,%%g0\n" \
1258 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1260 " sub %1,%2,%1 ! this kills msb of n\n" \
1261 "4: sub %1,%2,%1\n" \
1262 "5: addxcc %1,%1,%1\n" \
1264 " subcc %%g1,1,%%g1\n" \
1265 "! Got carry from n. Subtract next step to cancel this carry.\n" \
1267 " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \
1269 "3: xnor %0,0,%0\n" \
1270 " ! End of inline udiv_qrnnd" \
1271 : "=&r" ((USItype) (__q)), \
1272 "=&r" ((USItype) (__r)) \
1273 : "r" ((USItype) (__d)), \
1274 "1" ((USItype) (__n1)), \
1275 "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1276 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1277 #endif /* __sparclite__ */
1278 #endif /* __sparc_v8__ */
1279 #endif /* sparc32 */
1281 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1282 && W_TYPE_SIZE == 64
1283 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1284 __asm__ ("addcc %r4,%5,%1\n\t" \
1285 "add %r2,%3,%0\n\t" \
1286 "bcs,a,pn %%xcc, 1f\n\t" \
1289 : "=r" ((UDItype)(sh)), \
1290 "=&r" ((UDItype)(sl)) \
1291 : "%rJ" ((UDItype)(ah)), \
1292 "rI" ((UDItype)(bh)), \
1293 "%rJ" ((UDItype)(al)), \
1294 "rI" ((UDItype)(bl)) \
1297 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1298 __asm__ ("subcc %r4,%5,%1\n\t" \
1299 "sub %r2,%3,%0\n\t" \
1300 "bcs,a,pn %%xcc, 1f\n\t" \
1301 "sub %0, 1, %0\n\t" \
1303 : "=r" ((UDItype)(sh)), \
1304 "=&r" ((UDItype)(sl)) \
1305 : "rJ" ((UDItype)(ah)), \
1306 "rI" ((UDItype)(bh)), \
1307 "rJ" ((UDItype)(al)), \
1308 "rI" ((UDItype)(bl)) \
1311 #define umul_ppmm(wh, wl, u, v) \
1313 UDItype tmp1, tmp2, tmp3, tmp4; \
1314 __asm__ __volatile__ ( \
1316 "mulx %3,%6,%1\n\t" \
1317 "srlx %6,32,%2\n\t" \
1318 "mulx %2,%3,%4\n\t" \
1319 "sllx %4,32,%5\n\t" \
1321 "sub %1,%5,%5\n\t" \
1322 "srlx %5,32,%5\n\t" \
1323 "addcc %4,%5,%4\n\t" \
1324 "srlx %7,32,%5\n\t" \
1325 "mulx %3,%5,%3\n\t" \
1326 "mulx %2,%5,%5\n\t" \
1327 "sethi %%hi(0x80000000),%2\n\t" \
1328 "addcc %4,%3,%4\n\t" \
1329 "srlx %4,32,%4\n\t" \
1330 "add %2,%2,%2\n\t" \
1331 "movcc %%xcc,%%g0,%2\n\t" \
1332 "addcc %5,%4,%5\n\t" \
1333 "sllx %3,32,%3\n\t" \
1334 "add %1,%3,%1\n\t" \
1336 : "=r" ((UDItype)(wh)), \
1337 "=&r" ((UDItype)(wl)), \
1338 "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \
1339 : "r" ((UDItype)(u)), \
1340 "r" ((UDItype)(v)) \
1343 #define UMUL_TIME 96
1344 #define UDIV_TIME 230
1345 #endif /* sparc64 */
1347 #if defined (__vax__) && W_TYPE_SIZE == 32
1348 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1349 __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \
1350 : "=g" ((USItype) (sh)), \
1351 "=&g" ((USItype) (sl)) \
1352 : "%0" ((USItype) (ah)), \
1353 "g" ((USItype) (bh)), \
1354 "%1" ((USItype) (al)), \
1355 "g" ((USItype) (bl)))
1356 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1357 __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \
1358 : "=g" ((USItype) (sh)), \
1359 "=&g" ((USItype) (sl)) \
1360 : "0" ((USItype) (ah)), \
1361 "g" ((USItype) (bh)), \
1362 "1" ((USItype) (al)), \
1363 "g" ((USItype) (bl)))
1364 #define umul_ppmm(xh, xl, m0, m1) \
1368 struct {USItype __l, __h;} __i; \
1370 USItype __m0 = (m0), __m1 = (m1); \
1371 __asm__ ("emul %1,%2,$0,%0" \
1372 : "=r" (__xx.__ll) \
1375 (xh) = __xx.__i.__h; \
1376 (xl) = __xx.__i.__l; \
1377 (xh) += ((((SItype) __m0 >> 31) & __m1) \
1378 + (((SItype) __m1 >> 31) & __m0)); \
1380 #define sdiv_qrnnd(q, r, n1, n0, d) \
1382 union {DItype __ll; \
1383 struct {SItype __l, __h;} __i; \
1385 __xx.__i.__h = n1; __xx.__i.__l = n0; \
1386 __asm__ ("ediv %3,%2,%0,%1" \
1387 : "=g" (q), "=g" (r) \
1388 : "g" (__xx.__ll), "g" (d)); \
1390 #endif /* __vax__ */
1393 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1397 __asm__ ("addu .l1 %1, %2, %0" \
1398 : "=a" (__ll) : "a" (al), "a" (bl)); \
1399 (sl) = (USItype)__ll; \
1400 (sh) = ((USItype)(__ll >> 32)) + (ah) + (bh); \
1404 #ifdef _TMS320C6400_PLUS
1405 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1406 #define umul_ppmm(w1, w0, u, v) \
1408 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \
1409 (w1) = (USItype) (__x >> 32); \
1410 (w0) = (USItype) (__x); \
1412 #endif /* _TMS320C6400_PLUS */
1414 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x))
1416 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x))
1419 #define UDIV_TIME 40
1420 #endif /* _TMS320C6X */
1422 #if defined (__xtensa__) && W_TYPE_SIZE == 32
1423 /* This code is not Xtensa-configuration-specific, so rely on the compiler
1424 to expand builtin functions depending on what configuration features
1425 are available. This avoids library calls when the operation can be
1426 performed in-line. */
1427 #define umul_ppmm(w1, w0, u, v) \
1430 __w.ll = __builtin_umulsidi3 (u, v); \
1434 #define __umulsidi3(u, v) __builtin_umulsidi3 (u, v)
1435 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
1436 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
1437 #endif /* __xtensa__ */
1439 #if defined xstormy16
1440 extern UHItype __stormy16_count_leading_zeros (UHItype);
1441 #define count_leading_zeros(count, x) \
1446 /* We assume that W_TYPE_SIZE is a multiple of 16... */ \
1447 for ((count) = 0, size = W_TYPE_SIZE; size; size -= 16) \
1451 c = __clzhi2 ((x) >> (size - 16)); \
1458 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1461 #if defined (__z8000__) && W_TYPE_SIZE == 16
1462 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1463 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1464 : "=r" ((unsigned int)(sh)), \
1465 "=&r" ((unsigned int)(sl)) \
1466 : "%0" ((unsigned int)(ah)), \
1467 "r" ((unsigned int)(bh)), \
1468 "%1" ((unsigned int)(al)), \
1469 "rQR" ((unsigned int)(bl)))
1470 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1471 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1472 : "=r" ((unsigned int)(sh)), \
1473 "=&r" ((unsigned int)(sl)) \
1474 : "0" ((unsigned int)(ah)), \
1475 "r" ((unsigned int)(bh)), \
1476 "1" ((unsigned int)(al)), \
1477 "rQR" ((unsigned int)(bl)))
1478 #define umul_ppmm(xh, xl, m0, m1) \
1480 union {long int __ll; \
1481 struct {unsigned int __h, __l;} __i; \
1483 unsigned int __m0 = (m0), __m1 = (m1); \
1484 __asm__ ("mult %S0,%H3" \
1485 : "=r" (__xx.__i.__h), \
1486 "=r" (__xx.__i.__l) \
1489 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1490 (xh) += ((((signed int) __m0 >> 15) & __m1) \
1491 + (((signed int) __m1 >> 15) & __m0)); \
1493 #endif /* __z8000__ */
1495 #endif /* __GNUC__ */
1497 /* If this machine has no inline assembler, use C macros. */
1499 #if !defined (add_ssaaaa)
1500 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1503 __x = (al) + (bl); \
1504 (sh) = (ah) + (bh) + (__x < (al)); \
1509 #if !defined (sub_ddmmss)
1510 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1513 __x = (al) - (bl); \
1514 (sh) = (ah) - (bh) - (__x > (al)); \
1519 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1521 #if !defined (umul_ppmm) && defined (smul_ppmm)
1522 #define umul_ppmm(w1, w0, u, v) \
1525 UWtype __xm0 = (u), __xm1 = (v); \
1526 smul_ppmm (__w1, w0, __xm0, __xm1); \
1527 (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \
1528 + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \
1532 /* If we still don't have umul_ppmm, define it using plain C. */
1533 #if !defined (umul_ppmm)
1534 #define umul_ppmm(w1, w0, u, v) \
1536 UWtype __x0, __x1, __x2, __x3; \
1537 UHWtype __ul, __vl, __uh, __vh; \
1539 __ul = __ll_lowpart (u); \
1540 __uh = __ll_highpart (u); \
1541 __vl = __ll_lowpart (v); \
1542 __vh = __ll_highpart (v); \
1544 __x0 = (UWtype) __ul * __vl; \
1545 __x1 = (UWtype) __ul * __vh; \
1546 __x2 = (UWtype) __uh * __vl; \
1547 __x3 = (UWtype) __uh * __vh; \
1549 __x1 += __ll_highpart (__x0);/* this can't give carry */ \
1550 __x1 += __x2; /* but this indeed can */ \
1551 if (__x1 < __x2) /* did we get it? */ \
1552 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1554 (w1) = __x3 + __ll_highpart (__x1); \
1555 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
1559 #if !defined (__umulsidi3)
1560 #define __umulsidi3(u, v) \
1562 umul_ppmm (__w.s.high, __w.s.low, u, v); \
1566 /* Define this unconditionally, so it can be used for debugging. */
1567 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1569 UWtype __d1, __d0, __q1, __q0; \
1570 UWtype __r1, __r0, __m; \
1571 __d1 = __ll_highpart (d); \
1572 __d0 = __ll_lowpart (d); \
1574 __r1 = (n1) % __d1; \
1575 __q1 = (n1) / __d1; \
1576 __m = (UWtype) __q1 * __d0; \
1577 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
1580 __q1--, __r1 += (d); \
1581 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1583 __q1--, __r1 += (d); \
1587 __r0 = __r1 % __d1; \
1588 __q0 = __r1 / __d1; \
1589 __m = (UWtype) __q0 * __d0; \
1590 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
1593 __q0--, __r0 += (d); \
1596 __q0--, __r0 += (d); \
1600 (q) = (UWtype) __q1 * __ll_B | __q0; \
1604 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1605 __udiv_w_sdiv (defined in libgcc or elsewhere). */
1606 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1607 #define udiv_qrnnd(q, r, nh, nl, d) \
1610 (q) = __udiv_w_sdiv (&__r, nh, nl, d); \
1615 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1616 #if !defined (udiv_qrnnd)
1617 #define UDIV_NEEDS_NORMALIZATION 1
1618 #define udiv_qrnnd __udiv_qrnnd_c
1621 #if !defined (count_leading_zeros)
1622 #define count_leading_zeros(count, x) \
1624 UWtype __xr = (x); \
1627 if (W_TYPE_SIZE <= 32) \
1629 __a = __xr < ((UWtype)1<<2*__BITS4) \
1630 ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \
1631 : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1635 for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1636 if (((__xr >> __a) & 0xff) != 0) \
1640 (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1642 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1645 #if !defined (count_trailing_zeros)
1646 /* Define count_trailing_zeros using count_leading_zeros. The latter might be
1647 defined in asm, but if it is not, the C version above is good enough. */
1648 #define count_trailing_zeros(count, x) \
1650 UWtype __ctz_x = (x); \
1652 count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \
1653 (count) = W_TYPE_SIZE - 1 - __ctz_c; \
1657 #ifndef UDIV_NEEDS_NORMALIZATION
1658 #define UDIV_NEEDS_NORMALIZATION 0
OSDN Git Service
