2003-12-26 Guilhem Lavaux <guilhem@kaffe.org>
[official-gcc.git] / gcc / longlong.h
blob2fb68e7c373ee3728a33cb28ace3eac43828650a
1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2 Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000
3 Free Software Foundation, Inc.
5 This definition file is free software; you can redistribute it
6 and/or modify it under the terms of the GNU General Public
7 License as published by the Free Software Foundation; either
8 version 2, or (at your option) any later version.
10 This definition file is distributed in the hope that it will be
11 useful, but WITHOUT ANY WARRANTY; without even the implied
12 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
13 See the GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 59 Temple Place - Suite 330,
18 Boston, MA 02111-1307, USA. */
20 /* You have to define the following before including this file:
22 UWtype -- An unsigned type, default type for operations (typically a "word")
23 UHWtype -- An unsigned type, at least half the size of UWtype.
24 UDWtype -- An unsigned type, at least twice as large a UWtype
25 W_TYPE_SIZE -- size in bits of UWtype
27 UQItype -- Unsigned 8 bit type.
28 SItype, USItype -- Signed and unsigned 32 bit types.
29 DItype, UDItype -- Signed and unsigned 64 bit types.
31 On a 32 bit machine UWtype should typically be USItype;
32 on a 64 bit machine, UWtype should typically be UDItype.
35 #define __BITS4 (W_TYPE_SIZE / 4)
36 #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
37 #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
38 #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
40 #ifndef W_TYPE_SIZE
41 #define W_TYPE_SIZE 32
42 #define UWtype USItype
43 #define UHWtype USItype
44 #define UDWtype UDItype
45 #endif
47 /* Define auxiliary asm macros.
49 1) umul_ppmm(high_prod, low_prod, multipler, multiplicand) multiplies two
50 UWtype integers MULTIPLER and MULTIPLICAND, and generates a two UWtype
51 word product in HIGH_PROD and LOW_PROD.
53 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
54 UDWtype product. This is just a variant of umul_ppmm.
56 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
57 denominator) divides a UDWtype, composed by the UWtype integers
58 HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
59 in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less
60 than DENOMINATOR for correct operation. If, in addition, the most
61 significant bit of DENOMINATOR must be 1, then the pre-processor symbol
62 UDIV_NEEDS_NORMALIZATION is defined to 1.
64 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
65 denominator). Like udiv_qrnnd but the numbers are signed. The quotient
66 is rounded towards 0.
68 5) count_leading_zeros(count, x) counts the number of zero-bits from the
69 msb to the first nonzero bit in the UWtype X. This is the number of
70 steps X needs to be shifted left to set the msb. Undefined for X == 0,
71 unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
73 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
74 from the least significant end.
76 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
77 high_addend_2, low_addend_2) adds two UWtype integers, composed by
78 HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
79 respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow
80 (i.e. carry out) is not stored anywhere, and is lost.
82 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
83 high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
84 composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
85 LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE
86 and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
87 and is lost.
89 If any of these macros are left undefined for a particular CPU,
90 C macros are used. */
92 /* The CPUs come in alphabetical order below.
94 Please add support for more CPUs here, or improve the current support
95 for the CPUs below!
96 (E.g. WE32100, IBM360.) */
98 #if defined (__GNUC__) && !defined (NO_ASM)
100 /* We sometimes need to clobber "cc" with gcc2, but that would not be
101 understood by gcc1. Use cpp to avoid major code duplication. */
102 #if __GNUC__ < 2
103 #define __CLOBBER_CC
104 #define __AND_CLOBBER_CC
105 #else /* __GNUC__ >= 2 */
106 #define __CLOBBER_CC : "cc"
107 #define __AND_CLOBBER_CC , "cc"
108 #endif /* __GNUC__ < 2 */
110 #if defined (__alpha) && W_TYPE_SIZE == 64
111 #define umul_ppmm(ph, pl, m0, m1) \
112 do { \
113 UDItype __m0 = (m0), __m1 = (m1); \
114 (ph) = __builtin_alpha_umulh (__m0, __m1); \
115 (pl) = __m0 * __m1; \
116 } while (0)
117 #define UMUL_TIME 46
118 #ifndef LONGLONG_STANDALONE
119 #define udiv_qrnnd(q, r, n1, n0, d) \
120 do { UDItype __r; \
121 (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \
122 (r) = __r; \
123 } while (0)
124 extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
125 #define UDIV_TIME 220
126 #endif /* LONGLONG_STANDALONE */
127 #ifdef __alpha_cix__
128 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X))
129 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
130 #define COUNT_LEADING_ZEROS_0 64
131 #else
132 extern const UQItype __clz_tab[] ATTRIBUTE_HIDDEN;
133 #define count_leading_zeros(COUNT,X) \
134 do { \
135 UDItype __xr = (X), __t, __a; \
136 __t = __builtin_alpha_cmpbge (0, __xr); \
137 __a = __clz_tab[__t ^ 0xff] - 1; \
138 __t = __builtin_alpha_extbl (__xr, __a); \
139 (COUNT) = 64 - (__clz_tab[__t] + __a*8); \
140 } while (0)
141 #define count_trailing_zeros(COUNT,X) \
142 do { \
143 UDItype __xr = (X), __t, __a; \
144 __t = __builtin_alpha_cmpbge (0, __xr); \
145 __t = ~__t & -~__t; \
146 __a = ((__t & 0xCC) != 0) * 2; \
147 __a += ((__t & 0xF0) != 0) * 4; \
148 __a += ((__t & 0xAA) != 0); \
149 __t = __builtin_alpha_extbl (__xr, __a); \
150 __a <<= 3; \
151 __t &= -__t; \
152 __a += ((__t & 0xCC) != 0) * 2; \
153 __a += ((__t & 0xF0) != 0) * 4; \
154 __a += ((__t & 0xAA) != 0); \
155 (COUNT) = __a; \
156 } while (0)
157 #endif /* __alpha_cix__ */
158 #endif /* __alpha */
160 #if defined (__arc__) && W_TYPE_SIZE == 32
161 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
162 __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \
163 : "=r" ((USItype) (sh)), \
164 "=&r" ((USItype) (sl)) \
165 : "%r" ((USItype) (ah)), \
166 "rIJ" ((USItype) (bh)), \
167 "%r" ((USItype) (al)), \
168 "rIJ" ((USItype) (bl)))
169 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
170 __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \
171 : "=r" ((USItype) (sh)), \
172 "=&r" ((USItype) (sl)) \
173 : "r" ((USItype) (ah)), \
174 "rIJ" ((USItype) (bh)), \
175 "r" ((USItype) (al)), \
176 "rIJ" ((USItype) (bl)))
177 /* Call libgcc routine. */
178 #define umul_ppmm(w1, w0, u, v) \
179 do { \
180 DWunion __w; \
181 __w.ll = __umulsidi3 (u, v); \
182 w1 = __w.s.high; \
183 w0 = __w.s.low; \
184 } while (0)
185 #define __umulsidi3 __umulsidi3
186 UDItype __umulsidi3 (USItype, USItype);
187 #endif
189 #if defined (__arm__) && W_TYPE_SIZE == 32
190 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
191 __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \
192 : "=r" ((USItype) (sh)), \
193 "=&r" ((USItype) (sl)) \
194 : "%r" ((USItype) (ah)), \
195 "rI" ((USItype) (bh)), \
196 "%r" ((USItype) (al)), \
197 "rI" ((USItype) (bl)))
198 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
199 __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \
200 : "=r" ((USItype) (sh)), \
201 "=&r" ((USItype) (sl)) \
202 : "r" ((USItype) (ah)), \
203 "rI" ((USItype) (bh)), \
204 "r" ((USItype) (al)), \
205 "rI" ((USItype) (bl)))
206 #define umul_ppmm(xh, xl, a, b) \
207 {register USItype __t0, __t1, __t2; \
208 __asm__ ("%@ Inlined umul_ppmm\n" \
209 " mov %2, %5, lsr #16\n" \
210 " mov %0, %6, lsr #16\n" \
211 " bic %3, %5, %2, lsl #16\n" \
212 " bic %4, %6, %0, lsl #16\n" \
213 " mul %1, %3, %4\n" \
214 " mul %4, %2, %4\n" \
215 " mul %3, %0, %3\n" \
216 " mul %0, %2, %0\n" \
217 " adds %3, %4, %3\n" \
218 " addcs %0, %0, #65536\n" \
219 " adds %1, %1, %3, lsl #16\n" \
220 " adc %0, %0, %3, lsr #16" \
221 : "=&r" ((USItype) (xh)), \
222 "=r" ((USItype) (xl)), \
223 "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \
224 : "r" ((USItype) (a)), \
225 "r" ((USItype) (b)));}
226 #define UMUL_TIME 20
227 #define UDIV_TIME 100
228 #endif /* __arm__ */
230 #if defined (__hppa) && W_TYPE_SIZE == 32
231 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
232 __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \
233 : "=r" ((USItype) (sh)), \
234 "=&r" ((USItype) (sl)) \
235 : "%rM" ((USItype) (ah)), \
236 "rM" ((USItype) (bh)), \
237 "%rM" ((USItype) (al)), \
238 "rM" ((USItype) (bl)))
239 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
240 __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \
241 : "=r" ((USItype) (sh)), \
242 "=&r" ((USItype) (sl)) \
243 : "rM" ((USItype) (ah)), \
244 "rM" ((USItype) (bh)), \
245 "rM" ((USItype) (al)), \
246 "rM" ((USItype) (bl)))
247 #if defined (_PA_RISC1_1)
248 #define umul_ppmm(w1, w0, u, v) \
249 do { \
250 union \
252 UDItype __f; \
253 struct {USItype __w1, __w0;} __w1w0; \
254 } __t; \
255 __asm__ ("xmpyu %1,%2,%0" \
256 : "=x" (__t.__f) \
257 : "x" ((USItype) (u)), \
258 "x" ((USItype) (v))); \
259 (w1) = __t.__w1w0.__w1; \
260 (w0) = __t.__w1w0.__w0; \
261 } while (0)
262 #define UMUL_TIME 8
263 #else
264 #define UMUL_TIME 30
265 #endif
266 #define UDIV_TIME 40
267 #define count_leading_zeros(count, x) \
268 do { \
269 USItype __tmp; \
270 __asm__ ( \
271 "ldi 1,%0\n" \
272 " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \
273 " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\
274 " ldo 16(%0),%0 ; Yes. Perform add.\n" \
275 " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \
276 " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\
277 " ldo 8(%0),%0 ; Yes. Perform add.\n" \
278 " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \
279 " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\
280 " ldo 4(%0),%0 ; Yes. Perform add.\n" \
281 " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \
282 " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\
283 " ldo 2(%0),%0 ; Yes. Perform add.\n" \
284 " extru %1,30,1,%1 ; Extract bit 1.\n" \
285 " sub %0,%1,%0 ; Subtract it.\n" \
286 : "=r" (count), "=r" (__tmp) : "1" (x)); \
287 } while (0)
288 #endif
290 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
291 #define smul_ppmm(xh, xl, m0, m1) \
292 do { \
293 union {DItype __ll; \
294 struct {USItype __h, __l;} __i; \
295 } __x; \
296 __asm__ ("lr %N0,%1\n\tmr %0,%2" \
297 : "=&r" (__x.__ll) \
298 : "r" (m0), "r" (m1)); \
299 (xh) = __x.__i.__h; (xl) = __x.__i.__l; \
300 } while (0)
301 #define sdiv_qrnnd(q, r, n1, n0, d) \
302 do { \
303 union {DItype __ll; \
304 struct {USItype __h, __l;} __i; \
305 } __x; \
306 __x.__i.__h = n1; __x.__i.__l = n0; \
307 __asm__ ("dr %0,%2" \
308 : "=r" (__x.__ll) \
309 : "0" (__x.__ll), "r" (d)); \
310 (q) = __x.__i.__l; (r) = __x.__i.__h; \
311 } while (0)
312 #endif
314 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
315 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
316 __asm__ ("addl %5,%1\n\tadcl %3,%0" \
317 : "=r" ((USItype) (sh)), \
318 "=&r" ((USItype) (sl)) \
319 : "%0" ((USItype) (ah)), \
320 "g" ((USItype) (bh)), \
321 "%1" ((USItype) (al)), \
322 "g" ((USItype) (bl)))
323 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
324 __asm__ ("subl %5,%1\n\tsbbl %3,%0" \
325 : "=r" ((USItype) (sh)), \
326 "=&r" ((USItype) (sl)) \
327 : "0" ((USItype) (ah)), \
328 "g" ((USItype) (bh)), \
329 "1" ((USItype) (al)), \
330 "g" ((USItype) (bl)))
331 #define umul_ppmm(w1, w0, u, v) \
332 __asm__ ("mull %3" \
333 : "=a" ((USItype) (w0)), \
334 "=d" ((USItype) (w1)) \
335 : "%0" ((USItype) (u)), \
336 "rm" ((USItype) (v)))
337 #define udiv_qrnnd(q, r, n1, n0, dv) \
338 __asm__ ("divl %4" \
339 : "=a" ((USItype) (q)), \
340 "=d" ((USItype) (r)) \
341 : "0" ((USItype) (n0)), \
342 "1" ((USItype) (n1)), \
343 "rm" ((USItype) (dv)))
344 #define count_leading_zeros(count, x) \
345 do { \
346 USItype __cbtmp; \
347 __asm__ ("bsrl %1,%0" \
348 : "=r" (__cbtmp) : "rm" ((USItype) (x))); \
349 (count) = __cbtmp ^ 31; \
350 } while (0)
351 #define count_trailing_zeros(count, x) \
352 __asm__ ("bsfl %1,%0" : "=r" (count) : "rm" ((USItype)(x)))
353 #define UMUL_TIME 40
354 #define UDIV_TIME 40
355 #endif /* 80x86 */
357 #if defined (__i960__) && W_TYPE_SIZE == 32
358 #define umul_ppmm(w1, w0, u, v) \
359 ({union {UDItype __ll; \
360 struct {USItype __l, __h;} __i; \
361 } __xx; \
362 __asm__ ("emul %2,%1,%0" \
363 : "=d" (__xx.__ll) \
364 : "%dI" ((USItype) (u)), \
365 "dI" ((USItype) (v))); \
366 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
367 #define __umulsidi3(u, v) \
368 ({UDItype __w; \
369 __asm__ ("emul %2,%1,%0" \
370 : "=d" (__w) \
371 : "%dI" ((USItype) (u)), \
372 "dI" ((USItype) (v))); \
373 __w; })
374 #endif /* __i960__ */
376 #if defined (__M32R__) && W_TYPE_SIZE == 32
377 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
378 /* The cmp clears the condition bit. */ \
379 __asm__ ("cmp %0,%0\n\taddx %%5,%1\n\taddx %%3,%0" \
380 : "=r" ((USItype) (sh)), \
381 "=&r" ((USItype) (sl)) \
382 : "%0" ((USItype) (ah)), \
383 "r" ((USItype) (bh)), \
384 "%1" ((USItype) (al)), \
385 "r" ((USItype) (bl)) \
386 : "cbit")
387 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
388 /* The cmp clears the condition bit. */ \
389 __asm__ ("cmp %0,%0\n\tsubx %5,%1\n\tsubx %3,%0" \
390 : "=r" ((USItype) (sh)), \
391 "=&r" ((USItype) (sl)) \
392 : "0" ((USItype) (ah)), \
393 "r" ((USItype) (bh)), \
394 "1" ((USItype) (al)), \
395 "r" ((USItype) (bl)) \
396 : "cbit")
397 #endif /* __M32R__ */
399 #if defined (__mc68000__) && W_TYPE_SIZE == 32
400 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
401 __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \
402 : "=d" ((USItype) (sh)), \
403 "=&d" ((USItype) (sl)) \
404 : "%0" ((USItype) (ah)), \
405 "d" ((USItype) (bh)), \
406 "%1" ((USItype) (al)), \
407 "g" ((USItype) (bl)))
408 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
409 __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \
410 : "=d" ((USItype) (sh)), \
411 "=&d" ((USItype) (sl)) \
412 : "0" ((USItype) (ah)), \
413 "d" ((USItype) (bh)), \
414 "1" ((USItype) (al)), \
415 "g" ((USItype) (bl)))
417 /* The '020, '030, '040 and CPU32 have 32x32->64 and 64/32->32q-32r. */
418 #if defined (__mc68020__) || defined(mc68020) \
419 || defined(__mc68030__) || defined(mc68030) \
420 || defined(__mc68040__) || defined(mc68040) \
421 || defined(__mcpu32__) || defined(mcpu32)
422 #define umul_ppmm(w1, w0, u, v) \
423 __asm__ ("mulu%.l %3,%1:%0" \
424 : "=d" ((USItype) (w0)), \
425 "=d" ((USItype) (w1)) \
426 : "%0" ((USItype) (u)), \
427 "dmi" ((USItype) (v)))
428 #define UMUL_TIME 45
429 #define udiv_qrnnd(q, r, n1, n0, d) \
430 __asm__ ("divu%.l %4,%1:%0" \
431 : "=d" ((USItype) (q)), \
432 "=d" ((USItype) (r)) \
433 : "0" ((USItype) (n0)), \
434 "1" ((USItype) (n1)), \
435 "dmi" ((USItype) (d)))
436 #define UDIV_TIME 90
437 #define sdiv_qrnnd(q, r, n1, n0, d) \
438 __asm__ ("divs%.l %4,%1:%0" \
439 : "=d" ((USItype) (q)), \
440 "=d" ((USItype) (r)) \
441 : "0" ((USItype) (n0)), \
442 "1" ((USItype) (n1)), \
443 "dmi" ((USItype) (d)))
445 #else /* not mc68020 */
446 #if defined(__mcoldfire__)
447 #define umul_ppmm(xh, xl, a, b) \
448 __asm__ ("| Inlined umul_ppmm\n" \
449 " move%.l %2,%/d0\n" \
450 " move%.l %3,%/d1\n" \
451 " move%.l %/d0,%/d2\n" \
452 " swap %/d0\n" \
453 " move%.l %/d1,%/d3\n" \
454 " swap %/d1\n" \
455 " move%.w %/d2,%/d4\n" \
456 " mulu %/d3,%/d4\n" \
457 " mulu %/d1,%/d2\n" \
458 " mulu %/d0,%/d3\n" \
459 " mulu %/d0,%/d1\n" \
460 " move%.l %/d4,%/d0\n" \
461 " clr%.w %/d0\n" \
462 " swap %/d0\n" \
463 " add%.l %/d0,%/d2\n" \
464 " add%.l %/d3,%/d2\n" \
465 " jcc 1f\n" \
466 " add%.l %#65536,%/d1\n" \
467 "1: swap %/d2\n" \
468 " moveq %#0,%/d0\n" \
469 " move%.w %/d2,%/d0\n" \
470 " move%.w %/d4,%/d2\n" \
471 " move%.l %/d2,%1\n" \
472 " add%.l %/d1,%/d0\n" \
473 " move%.l %/d0,%0" \
474 : "=g" ((USItype) (xh)), \
475 "=g" ((USItype) (xl)) \
476 : "g" ((USItype) (a)), \
477 "g" ((USItype) (b)) \
478 : "d0", "d1", "d2", "d3", "d4")
479 #define UMUL_TIME 100
480 #define UDIV_TIME 400
481 #else /* not ColdFire */
482 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */
483 #define umul_ppmm(xh, xl, a, b) \
484 __asm__ ("| Inlined umul_ppmm\n" \
485 " move%.l %2,%/d0\n" \
486 " move%.l %3,%/d1\n" \
487 " move%.l %/d0,%/d2\n" \
488 " swap %/d0\n" \
489 " move%.l %/d1,%/d3\n" \
490 " swap %/d1\n" \
491 " move%.w %/d2,%/d4\n" \
492 " mulu %/d3,%/d4\n" \
493 " mulu %/d1,%/d2\n" \
494 " mulu %/d0,%/d3\n" \
495 " mulu %/d0,%/d1\n" \
496 " move%.l %/d4,%/d0\n" \
497 " eor%.w %/d0,%/d0\n" \
498 " swap %/d0\n" \
499 " add%.l %/d0,%/d2\n" \
500 " add%.l %/d3,%/d2\n" \
501 " jcc 1f\n" \
502 " add%.l %#65536,%/d1\n" \
503 "1: swap %/d2\n" \
504 " moveq %#0,%/d0\n" \
505 " move%.w %/d2,%/d0\n" \
506 " move%.w %/d4,%/d2\n" \
507 " move%.l %/d2,%1\n" \
508 " add%.l %/d1,%/d0\n" \
509 " move%.l %/d0,%0" \
510 : "=g" ((USItype) (xh)), \
511 "=g" ((USItype) (xl)) \
512 : "g" ((USItype) (a)), \
513 "g" ((USItype) (b)) \
514 : "d0", "d1", "d2", "d3", "d4")
515 #define UMUL_TIME 100
516 #define UDIV_TIME 400
517 #endif /* not ColdFire */
518 #endif /* not mc68020 */
520 /* The '020, '030, '040 and '060 have bitfield insns.
521 cpu32 disguises as a 68020, but lacks them. */
522 #if ( defined (__mc68020__) || defined(mc68020) \
523 || defined(__mc68030__) || defined(mc68030) \
524 || defined(__mc68040__) || defined(mc68040) \
525 || defined(__mc68060__) || defined(mc68060) ) \
526 && !defined(__mcpu32__)
527 #define count_leading_zeros(count, x) \
528 __asm__ ("bfffo %1{%b2:%b2},%0" \
529 : "=d" ((USItype) (count)) \
530 : "od" ((USItype) (x)), "n" (0))
531 #endif
532 #endif /* mc68000 */
534 #if defined (__m88000__) && W_TYPE_SIZE == 32
535 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
536 __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \
537 : "=r" ((USItype) (sh)), \
538 "=&r" ((USItype) (sl)) \
539 : "%rJ" ((USItype) (ah)), \
540 "rJ" ((USItype) (bh)), \
541 "%rJ" ((USItype) (al)), \
542 "rJ" ((USItype) (bl)))
543 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
544 __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \
545 : "=r" ((USItype) (sh)), \
546 "=&r" ((USItype) (sl)) \
547 : "rJ" ((USItype) (ah)), \
548 "rJ" ((USItype) (bh)), \
549 "rJ" ((USItype) (al)), \
550 "rJ" ((USItype) (bl)))
551 #define count_leading_zeros(count, x) \
552 do { \
553 USItype __cbtmp; \
554 __asm__ ("ff1 %0,%1" \
555 : "=r" (__cbtmp) \
556 : "r" ((USItype) (x))); \
557 (count) = __cbtmp ^ 31; \
558 } while (0)
559 #define COUNT_LEADING_ZEROS_0 63 /* sic */
560 #if defined (__mc88110__)
561 #define umul_ppmm(wh, wl, u, v) \
562 do { \
563 union {UDItype __ll; \
564 struct {USItype __h, __l;} __i; \
565 } __xx; \
566 __asm__ ("mulu.d %0,%1,%2" \
567 : "=r" (__xx.__ll) \
568 : "r" ((USItype) (u)), \
569 "r" ((USItype) (v))); \
570 (wh) = __xx.__i.__h; \
571 (wl) = __xx.__i.__l; \
572 } while (0)
573 #define udiv_qrnnd(q, r, n1, n0, d) \
574 ({union {UDItype __ll; \
575 struct {USItype __h, __l;} __i; \
576 } __xx; \
577 USItype __q; \
578 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
579 __asm__ ("divu.d %0,%1,%2" \
580 : "=r" (__q) \
581 : "r" (__xx.__ll), \
582 "r" ((USItype) (d))); \
583 (r) = (n0) - __q * (d); (q) = __q; })
584 #define UMUL_TIME 5
585 #define UDIV_TIME 25
586 #else
587 #define UMUL_TIME 17
588 #define UDIV_TIME 150
589 #endif /* __mc88110__ */
590 #endif /* __m88000__ */
592 #if defined (__mips__) && W_TYPE_SIZE == 32
593 #define umul_ppmm(w1, w0, u, v) \
594 __asm__ ("multu %2,%3" \
595 : "=l" ((USItype) (w0)), \
596 "=h" ((USItype) (w1)) \
597 : "d" ((USItype) (u)), \
598 "d" ((USItype) (v)))
599 #define UMUL_TIME 10
600 #define UDIV_TIME 100
601 #endif /* __mips__ */
603 #if defined (__ns32000__) && W_TYPE_SIZE == 32
604 #define umul_ppmm(w1, w0, u, v) \
605 ({union {UDItype __ll; \
606 struct {USItype __l, __h;} __i; \
607 } __xx; \
608 __asm__ ("meid %2,%0" \
609 : "=g" (__xx.__ll) \
610 : "%0" ((USItype) (u)), \
611 "g" ((USItype) (v))); \
612 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
613 #define __umulsidi3(u, v) \
614 ({UDItype __w; \
615 __asm__ ("meid %2,%0" \
616 : "=g" (__w) \
617 : "%0" ((USItype) (u)), \
618 "g" ((USItype) (v))); \
619 __w; })
620 #define udiv_qrnnd(q, r, n1, n0, d) \
621 ({union {UDItype __ll; \
622 struct {USItype __l, __h;} __i; \
623 } __xx; \
624 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
625 __asm__ ("deid %2,%0" \
626 : "=g" (__xx.__ll) \
627 : "0" (__xx.__ll), \
628 "g" ((USItype) (d))); \
629 (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
630 #define count_trailing_zeros(count,x) \
631 do { \
632 __asm__ ("ffsd %2,%0" \
633 : "=r" ((USItype) (count)) \
634 : "0" ((USItype) 0), \
635 "r" ((USItype) (x))); \
636 } while (0)
637 #endif /* __ns32000__ */
639 /* FIXME: We should test _IBMR2 here when we add assembly support for the
640 system vendor compilers.
641 FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good
642 enough, since that hits ARM and m68k too. */
643 #if (defined (_ARCH_PPC) /* AIX */ \
644 || defined (_ARCH_PWR) /* AIX */ \
645 || defined (_ARCH_COM) /* AIX */ \
646 || defined (__powerpc__) /* gcc */ \
647 || defined (__POWERPC__) /* BEOS */ \
648 || defined (__ppc__) /* Darwin */ \
649 || defined (PPC) /* GNU/Linux, SysV */ \
650 ) && W_TYPE_SIZE == 32
651 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
652 do { \
653 if (__builtin_constant_p (bh) && (bh) == 0) \
654 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
655 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
656 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
657 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
658 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
659 else \
660 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
661 : "=r" (sh), "=&r" (sl) \
662 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
663 } while (0)
664 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
665 do { \
666 if (__builtin_constant_p (ah) && (ah) == 0) \
667 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
668 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
669 else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \
670 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
671 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
672 else if (__builtin_constant_p (bh) && (bh) == 0) \
673 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
674 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
675 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
676 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
677 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
678 else \
679 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
680 : "=r" (sh), "=&r" (sl) \
681 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
682 } while (0)
683 #define count_leading_zeros(count, x) \
684 __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x))
685 #define COUNT_LEADING_ZEROS_0 32
686 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
687 || defined (__ppc__) || defined (PPC)
688 #define umul_ppmm(ph, pl, m0, m1) \
689 do { \
690 USItype __m0 = (m0), __m1 = (m1); \
691 __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
692 (pl) = __m0 * __m1; \
693 } while (0)
694 #define UMUL_TIME 15
695 #define smul_ppmm(ph, pl, m0, m1) \
696 do { \
697 SItype __m0 = (m0), __m1 = (m1); \
698 __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
699 (pl) = __m0 * __m1; \
700 } while (0)
701 #define SMUL_TIME 14
702 #define UDIV_TIME 120
703 #elif defined (_ARCH_PWR)
704 #define UMUL_TIME 8
705 #define smul_ppmm(xh, xl, m0, m1) \
706 __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1))
707 #define SMUL_TIME 4
708 #define sdiv_qrnnd(q, r, nh, nl, d) \
709 __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d))
710 #define UDIV_TIME 100
711 #endif
712 #endif /* 32-bit POWER architecture variants. */
714 /* We should test _IBMR2 here when we add assembly support for the system
715 vendor compilers. */
716 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
717 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
718 do { \
719 if (__builtin_constant_p (bh) && (bh) == 0) \
720 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
721 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
722 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
723 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
724 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
725 else \
726 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
727 : "=r" (sh), "=&r" (sl) \
728 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
729 } while (0)
730 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
731 do { \
732 if (__builtin_constant_p (ah) && (ah) == 0) \
733 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
734 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
735 else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \
736 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
737 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
738 else if (__builtin_constant_p (bh) && (bh) == 0) \
739 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
740 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
741 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
742 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
743 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
744 else \
745 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
746 : "=r" (sh), "=&r" (sl) \
747 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
748 } while (0)
749 #define count_leading_zeros(count, x) \
750 __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
751 #define COUNT_LEADING_ZEROS_0 64
752 #define umul_ppmm(ph, pl, m0, m1) \
753 do { \
754 UDItype __m0 = (m0), __m1 = (m1); \
755 __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
756 (pl) = __m0 * __m1; \
757 } while (0)
758 #define UMUL_TIME 15
759 #define smul_ppmm(ph, pl, m0, m1) \
760 do { \
761 DItype __m0 = (m0), __m1 = (m1); \
762 __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
763 (pl) = __m0 * __m1; \
764 } while (0)
765 #define SMUL_TIME 14 /* ??? */
766 #define UDIV_TIME 120 /* ??? */
767 #endif /* 64-bit PowerPC. */
769 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
770 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
771 __asm__ ("a %1,%5\n\tae %0,%3" \
772 : "=r" ((USItype) (sh)), \
773 "=&r" ((USItype) (sl)) \
774 : "%0" ((USItype) (ah)), \
775 "r" ((USItype) (bh)), \
776 "%1" ((USItype) (al)), \
777 "r" ((USItype) (bl)))
778 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
779 __asm__ ("s %1,%5\n\tse %0,%3" \
780 : "=r" ((USItype) (sh)), \
781 "=&r" ((USItype) (sl)) \
782 : "0" ((USItype) (ah)), \
783 "r" ((USItype) (bh)), \
784 "1" ((USItype) (al)), \
785 "r" ((USItype) (bl)))
786 #define umul_ppmm(ph, pl, m0, m1) \
787 do { \
788 USItype __m0 = (m0), __m1 = (m1); \
789 __asm__ ( \
790 "s r2,r2\n" \
791 " mts r10,%2\n" \
792 " m r2,%3\n" \
793 " m r2,%3\n" \
794 " m r2,%3\n" \
795 " m r2,%3\n" \
796 " m r2,%3\n" \
797 " m r2,%3\n" \
798 " m r2,%3\n" \
799 " m r2,%3\n" \
800 " m r2,%3\n" \
801 " m r2,%3\n" \
802 " m r2,%3\n" \
803 " m r2,%3\n" \
804 " m r2,%3\n" \
805 " m r2,%3\n" \
806 " m r2,%3\n" \
807 " m r2,%3\n" \
808 " cas %0,r2,r0\n" \
809 " mfs r10,%1" \
810 : "=r" ((USItype) (ph)), \
811 "=r" ((USItype) (pl)) \
812 : "%r" (__m0), \
813 "r" (__m1) \
814 : "r2"); \
815 (ph) += ((((SItype) __m0 >> 31) & __m1) \
816 + (((SItype) __m1 >> 31) & __m0)); \
817 } while (0)
818 #define UMUL_TIME 20
819 #define UDIV_TIME 200
820 #define count_leading_zeros(count, x) \
821 do { \
822 if ((x) >= 0x10000) \
823 __asm__ ("clz %0,%1" \
824 : "=r" ((USItype) (count)) \
825 : "r" ((USItype) (x) >> 16)); \
826 else \
828 __asm__ ("clz %0,%1" \
829 : "=r" ((USItype) (count)) \
830 : "r" ((USItype) (x))); \
831 (count) += 16; \
833 } while (0)
834 #endif
836 #if defined (__sh2__) && W_TYPE_SIZE == 32
837 #define umul_ppmm(w1, w0, u, v) \
838 __asm__ ( \
839 "dmulu.l %2,%3\n\tsts macl,%1\n\tsts mach,%0" \
840 : "=r" ((USItype)(w1)), \
841 "=r" ((USItype)(w0)) \
842 : "r" ((USItype)(u)), \
843 "r" ((USItype)(v)) \
844 : "macl", "mach")
845 #define UMUL_TIME 5
846 #endif
848 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
849 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
850 #define count_leading_zeros(count, x) \
851 do \
853 UDItype x_ = (USItype)(x); \
854 SItype c_; \
856 __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \
857 (count) = c_ - 31; \
859 while (0)
860 #define COUNT_LEADING_ZEROS_0 32
861 #endif
863 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
864 && W_TYPE_SIZE == 32
865 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
866 __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \
867 : "=r" ((USItype) (sh)), \
868 "=&r" ((USItype) (sl)) \
869 : "%rJ" ((USItype) (ah)), \
870 "rI" ((USItype) (bh)), \
871 "%rJ" ((USItype) (al)), \
872 "rI" ((USItype) (bl)) \
873 __CLOBBER_CC)
874 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
875 __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \
876 : "=r" ((USItype) (sh)), \
877 "=&r" ((USItype) (sl)) \
878 : "rJ" ((USItype) (ah)), \
879 "rI" ((USItype) (bh)), \
880 "rJ" ((USItype) (al)), \
881 "rI" ((USItype) (bl)) \
882 __CLOBBER_CC)
883 #if defined (__sparc_v8__)
884 #define umul_ppmm(w1, w0, u, v) \
885 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
886 : "=r" ((USItype) (w1)), \
887 "=r" ((USItype) (w0)) \
888 : "r" ((USItype) (u)), \
889 "r" ((USItype) (v)))
890 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
891 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
892 : "=&r" ((USItype) (__q)), \
893 "=&r" ((USItype) (__r)) \
894 : "r" ((USItype) (__n1)), \
895 "r" ((USItype) (__n0)), \
896 "r" ((USItype) (__d)))
897 #else
898 #if defined (__sparclite__)
899 /* This has hardware multiply but not divide. It also has two additional
900 instructions scan (ffs from high bit) and divscc. */
901 #define umul_ppmm(w1, w0, u, v) \
902 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
903 : "=r" ((USItype) (w1)), \
904 "=r" ((USItype) (w0)) \
905 : "r" ((USItype) (u)), \
906 "r" ((USItype) (v)))
907 #define udiv_qrnnd(q, r, n1, n0, d) \
908 __asm__ ("! Inlined udiv_qrnnd\n" \
909 " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
910 " tst %%g0\n" \
911 " divscc %3,%4,%%g1\n" \
912 " divscc %%g1,%4,%%g1\n" \
913 " divscc %%g1,%4,%%g1\n" \
914 " divscc %%g1,%4,%%g1\n" \
915 " divscc %%g1,%4,%%g1\n" \
916 " divscc %%g1,%4,%%g1\n" \
917 " divscc %%g1,%4,%%g1\n" \
918 " divscc %%g1,%4,%%g1\n" \
919 " divscc %%g1,%4,%%g1\n" \
920 " divscc %%g1,%4,%%g1\n" \
921 " divscc %%g1,%4,%%g1\n" \
922 " divscc %%g1,%4,%%g1\n" \
923 " divscc %%g1,%4,%%g1\n" \
924 " divscc %%g1,%4,%%g1\n" \
925 " divscc %%g1,%4,%%g1\n" \
926 " divscc %%g1,%4,%%g1\n" \
927 " divscc %%g1,%4,%%g1\n" \
928 " divscc %%g1,%4,%%g1\n" \
929 " divscc %%g1,%4,%%g1\n" \
930 " divscc %%g1,%4,%%g1\n" \
931 " divscc %%g1,%4,%%g1\n" \
932 " divscc %%g1,%4,%%g1\n" \
933 " divscc %%g1,%4,%%g1\n" \
934 " divscc %%g1,%4,%%g1\n" \
935 " divscc %%g1,%4,%%g1\n" \
936 " divscc %%g1,%4,%%g1\n" \
937 " divscc %%g1,%4,%%g1\n" \
938 " divscc %%g1,%4,%%g1\n" \
939 " divscc %%g1,%4,%%g1\n" \
940 " divscc %%g1,%4,%%g1\n" \
941 " divscc %%g1,%4,%%g1\n" \
942 " divscc %%g1,%4,%0\n" \
943 " rd %%y,%1\n" \
944 " bl,a 1f\n" \
945 " add %1,%4,%1\n" \
946 "1: ! End of inline udiv_qrnnd" \
947 : "=r" ((USItype) (q)), \
948 "=r" ((USItype) (r)) \
949 : "r" ((USItype) (n1)), \
950 "r" ((USItype) (n0)), \
951 "rI" ((USItype) (d)) \
952 : "g1" __AND_CLOBBER_CC)
953 #define UDIV_TIME 37
954 #define count_leading_zeros(count, x) \
955 do { \
956 __asm__ ("scan %1,1,%0" \
957 : "=r" ((USItype) (count)) \
958 : "r" ((USItype) (x))); \
959 } while (0)
960 /* Early sparclites return 63 for an argument of 0, but they warn that future
961 implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0
962 undefined. */
963 #else
964 /* SPARC without integer multiplication and divide instructions.
965 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
966 #define umul_ppmm(w1, w0, u, v) \
967 __asm__ ("! Inlined umul_ppmm\n" \
968 " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\
969 " sra %3,31,%%o5 ! Don't move this insn\n" \
970 " and %2,%%o5,%%o5 ! Don't move this insn\n" \
971 " andcc %%g0,0,%%g1 ! Don't move this insn\n" \
972 " mulscc %%g1,%3,%%g1\n" \
973 " mulscc %%g1,%3,%%g1\n" \
974 " mulscc %%g1,%3,%%g1\n" \
975 " mulscc %%g1,%3,%%g1\n" \
976 " mulscc %%g1,%3,%%g1\n" \
977 " mulscc %%g1,%3,%%g1\n" \
978 " mulscc %%g1,%3,%%g1\n" \
979 " mulscc %%g1,%3,%%g1\n" \
980 " mulscc %%g1,%3,%%g1\n" \
981 " mulscc %%g1,%3,%%g1\n" \
982 " mulscc %%g1,%3,%%g1\n" \
983 " mulscc %%g1,%3,%%g1\n" \
984 " mulscc %%g1,%3,%%g1\n" \
985 " mulscc %%g1,%3,%%g1\n" \
986 " mulscc %%g1,%3,%%g1\n" \
987 " mulscc %%g1,%3,%%g1\n" \
988 " mulscc %%g1,%3,%%g1\n" \
989 " mulscc %%g1,%3,%%g1\n" \
990 " mulscc %%g1,%3,%%g1\n" \
991 " mulscc %%g1,%3,%%g1\n" \
992 " mulscc %%g1,%3,%%g1\n" \
993 " mulscc %%g1,%3,%%g1\n" \
994 " mulscc %%g1,%3,%%g1\n" \
995 " mulscc %%g1,%3,%%g1\n" \
996 " mulscc %%g1,%3,%%g1\n" \
997 " mulscc %%g1,%3,%%g1\n" \
998 " mulscc %%g1,%3,%%g1\n" \
999 " mulscc %%g1,%3,%%g1\n" \
1000 " mulscc %%g1,%3,%%g1\n" \
1001 " mulscc %%g1,%3,%%g1\n" \
1002 " mulscc %%g1,%3,%%g1\n" \
1003 " mulscc %%g1,%3,%%g1\n" \
1004 " mulscc %%g1,0,%%g1\n" \
1005 " add %%g1,%%o5,%0\n" \
1006 " rd %%y,%1" \
1007 : "=r" ((USItype) (w1)), \
1008 "=r" ((USItype) (w0)) \
1009 : "%rI" ((USItype) (u)), \
1010 "r" ((USItype) (v)) \
1011 : "g1", "o5" __AND_CLOBBER_CC)
1012 #define UMUL_TIME 39 /* 39 instructions */
1013 /* It's quite necessary to add this much assembler for the sparc.
1014 The default udiv_qrnnd (in C) is more than 10 times slower! */
1015 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1016 __asm__ ("! Inlined udiv_qrnnd\n" \
1017 " mov 32,%%g1\n" \
1018 " subcc %1,%2,%%g0\n" \
1019 "1: bcs 5f\n" \
1020 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1021 " sub %1,%2,%1 ! this kills msb of n\n" \
1022 " addx %1,%1,%1 ! so this can't give carry\n" \
1023 " subcc %%g1,1,%%g1\n" \
1024 "2: bne 1b\n" \
1025 " subcc %1,%2,%%g0\n" \
1026 " bcs 3f\n" \
1027 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1028 " b 3f\n" \
1029 " sub %1,%2,%1 ! this kills msb of n\n" \
1030 "4: sub %1,%2,%1\n" \
1031 "5: addxcc %1,%1,%1\n" \
1032 " bcc 2b\n" \
1033 " subcc %%g1,1,%%g1\n" \
1034 "! Got carry from n. Subtract next step to cancel this carry.\n" \
1035 " bne 4b\n" \
1036 " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \
1037 " sub %1,%2,%1\n" \
1038 "3: xnor %0,0,%0\n" \
1039 " ! End of inline udiv_qrnnd" \
1040 : "=&r" ((USItype) (__q)), \
1041 "=&r" ((USItype) (__r)) \
1042 : "r" ((USItype) (__d)), \
1043 "1" ((USItype) (__n1)), \
1044 "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1045 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1046 #endif /* __sparclite__ */
1047 #endif /* __sparc_v8__ */
1048 #endif /* sparc32 */
1050 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1051 && W_TYPE_SIZE == 64
1052 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1053 __asm__ ("addcc %r4,%5,%1\n\t" \
1054 "add %r2,%3,%0\n\t" \
1055 "bcs,a,pn %%xcc, 1f\n\t" \
1056 "add %0, 1, %0\n" \
1057 "1:" \
1058 : "=r" ((UDItype)(sh)), \
1059 "=&r" ((UDItype)(sl)) \
1060 : "%rJ" ((UDItype)(ah)), \
1061 "rI" ((UDItype)(bh)), \
1062 "%rJ" ((UDItype)(al)), \
1063 "rI" ((UDItype)(bl)) \
1064 __CLOBBER_CC)
1066 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1067 __asm__ ("subcc %r4,%5,%1\n\t" \
1068 "sub %r2,%3,%0\n\t" \
1069 "bcs,a,pn %%xcc, 1f\n\t" \
1070 "sub %0, 1, %0\n\t" \
1071 "1:" \
1072 : "=r" ((UDItype)(sh)), \
1073 "=&r" ((UDItype)(sl)) \
1074 : "rJ" ((UDItype)(ah)), \
1075 "rI" ((UDItype)(bh)), \
1076 "rJ" ((UDItype)(al)), \
1077 "rI" ((UDItype)(bl)) \
1078 __CLOBBER_CC)
1080 #define umul_ppmm(wh, wl, u, v) \
1081 do { \
1082 UDItype tmp1, tmp2, tmp3, tmp4; \
1083 __asm__ __volatile__ ( \
1084 "srl %7,0,%3\n\t" \
1085 "mulx %3,%6,%1\n\t" \
1086 "srlx %6,32,%2\n\t" \
1087 "mulx %2,%3,%4\n\t" \
1088 "sllx %4,32,%5\n\t" \
1089 "srl %6,0,%3\n\t" \
1090 "sub %1,%5,%5\n\t" \
1091 "srlx %5,32,%5\n\t" \
1092 "addcc %4,%5,%4\n\t" \
1093 "srlx %7,32,%5\n\t" \
1094 "mulx %3,%5,%3\n\t" \
1095 "mulx %2,%5,%5\n\t" \
1096 "sethi %%hi(0x80000000),%2\n\t" \
1097 "addcc %4,%3,%4\n\t" \
1098 "srlx %4,32,%4\n\t" \
1099 "add %2,%2,%2\n\t" \
1100 "movcc %%xcc,%%g0,%2\n\t" \
1101 "addcc %5,%4,%5\n\t" \
1102 "sllx %3,32,%3\n\t" \
1103 "add %1,%3,%1\n\t" \
1104 "add %5,%2,%0" \
1105 : "=r" ((UDItype)(wh)), \
1106 "=&r" ((UDItype)(wl)), \
1107 "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \
1108 : "r" ((UDItype)(u)), \
1109 "r" ((UDItype)(v)) \
1110 __CLOBBER_CC); \
1111 } while (0)
1112 #define UMUL_TIME 96
1113 #define UDIV_TIME 230
1114 #endif /* sparc64 */
1116 #if defined (__vax__) && W_TYPE_SIZE == 32
1117 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1118 __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \
1119 : "=g" ((USItype) (sh)), \
1120 "=&g" ((USItype) (sl)) \
1121 : "%0" ((USItype) (ah)), \
1122 "g" ((USItype) (bh)), \
1123 "%1" ((USItype) (al)), \
1124 "g" ((USItype) (bl)))
1125 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1126 __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \
1127 : "=g" ((USItype) (sh)), \
1128 "=&g" ((USItype) (sl)) \
1129 : "0" ((USItype) (ah)), \
1130 "g" ((USItype) (bh)), \
1131 "1" ((USItype) (al)), \
1132 "g" ((USItype) (bl)))
1133 #define umul_ppmm(xh, xl, m0, m1) \
1134 do { \
1135 union { \
1136 UDItype __ll; \
1137 struct {USItype __l, __h;} __i; \
1138 } __xx; \
1139 USItype __m0 = (m0), __m1 = (m1); \
1140 __asm__ ("emul %1,%2,$0,%0" \
1141 : "=r" (__xx.__ll) \
1142 : "g" (__m0), \
1143 "g" (__m1)); \
1144 (xh) = __xx.__i.__h; \
1145 (xl) = __xx.__i.__l; \
1146 (xh) += ((((SItype) __m0 >> 31) & __m1) \
1147 + (((SItype) __m1 >> 31) & __m0)); \
1148 } while (0)
1149 #define sdiv_qrnnd(q, r, n1, n0, d) \
1150 do { \
1151 union {DItype __ll; \
1152 struct {SItype __l, __h;} __i; \
1153 } __xx; \
1154 __xx.__i.__h = n1; __xx.__i.__l = n0; \
1155 __asm__ ("ediv %3,%2,%0,%1" \
1156 : "=g" (q), "=g" (r) \
1157 : "g" (__xx.__ll), "g" (d)); \
1158 } while (0)
1159 #endif /* __vax__ */
1161 #if defined (__z8000__) && W_TYPE_SIZE == 16
1162 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1163 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1164 : "=r" ((unsigned int)(sh)), \
1165 "=&r" ((unsigned int)(sl)) \
1166 : "%0" ((unsigned int)(ah)), \
1167 "r" ((unsigned int)(bh)), \
1168 "%1" ((unsigned int)(al)), \
1169 "rQR" ((unsigned int)(bl)))
1170 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1171 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1172 : "=r" ((unsigned int)(sh)), \
1173 "=&r" ((unsigned int)(sl)) \
1174 : "0" ((unsigned int)(ah)), \
1175 "r" ((unsigned int)(bh)), \
1176 "1" ((unsigned int)(al)), \
1177 "rQR" ((unsigned int)(bl)))
1178 #define umul_ppmm(xh, xl, m0, m1) \
1179 do { \
1180 union {long int __ll; \
1181 struct {unsigned int __h, __l;} __i; \
1182 } __xx; \
1183 unsigned int __m0 = (m0), __m1 = (m1); \
1184 __asm__ ("mult %S0,%H3" \
1185 : "=r" (__xx.__i.__h), \
1186 "=r" (__xx.__i.__l) \
1187 : "%1" (__m0), \
1188 "rQR" (__m1)); \
1189 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1190 (xh) += ((((signed int) __m0 >> 15) & __m1) \
1191 + (((signed int) __m1 >> 15) & __m0)); \
1192 } while (0)
1193 #endif /* __z8000__ */
1195 #endif /* __GNUC__ */
1197 /* If this machine has no inline assembler, use C macros. */
1199 #if !defined (add_ssaaaa)
1200 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1201 do { \
1202 UWtype __x; \
1203 __x = (al) + (bl); \
1204 (sh) = (ah) + (bh) + (__x < (al)); \
1205 (sl) = __x; \
1206 } while (0)
1207 #endif
1209 #if !defined (sub_ddmmss)
1210 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1211 do { \
1212 UWtype __x; \
1213 __x = (al) - (bl); \
1214 (sh) = (ah) - (bh) - (__x > (al)); \
1215 (sl) = __x; \
1216 } while (0)
1217 #endif
1219 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1220 smul_ppmm. */
1221 #if !defined (umul_ppmm) && defined (smul_ppmm)
1222 #define umul_ppmm(w1, w0, u, v) \
1223 do { \
1224 UWtype __w1; \
1225 UWtype __xm0 = (u), __xm1 = (v); \
1226 smul_ppmm (__w1, w0, __xm0, __xm1); \
1227 (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \
1228 + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \
1229 } while (0)
1230 #endif
1232 /* If we still don't have umul_ppmm, define it using plain C. */
1233 #if !defined (umul_ppmm)
1234 #define umul_ppmm(w1, w0, u, v) \
1235 do { \
1236 UWtype __x0, __x1, __x2, __x3; \
1237 UHWtype __ul, __vl, __uh, __vh; \
1239 __ul = __ll_lowpart (u); \
1240 __uh = __ll_highpart (u); \
1241 __vl = __ll_lowpart (v); \
1242 __vh = __ll_highpart (v); \
1244 __x0 = (UWtype) __ul * __vl; \
1245 __x1 = (UWtype) __ul * __vh; \
1246 __x2 = (UWtype) __uh * __vl; \
1247 __x3 = (UWtype) __uh * __vh; \
1249 __x1 += __ll_highpart (__x0);/* this can't give carry */ \
1250 __x1 += __x2; /* but this indeed can */ \
1251 if (__x1 < __x2) /* did we get it? */ \
1252 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1254 (w1) = __x3 + __ll_highpart (__x1); \
1255 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
1256 } while (0)
1257 #endif
1259 #if !defined (__umulsidi3)
1260 #define __umulsidi3(u, v) \
1261 ({DWunion __w; \
1262 umul_ppmm (__w.s.high, __w.s.low, u, v); \
1263 __w.ll; })
1264 #endif
1266 /* Define this unconditionally, so it can be used for debugging. */
1267 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1268 do { \
1269 UWtype __d1, __d0, __q1, __q0; \
1270 UWtype __r1, __r0, __m; \
1271 __d1 = __ll_highpart (d); \
1272 __d0 = __ll_lowpart (d); \
1274 __r1 = (n1) % __d1; \
1275 __q1 = (n1) / __d1; \
1276 __m = (UWtype) __q1 * __d0; \
1277 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
1278 if (__r1 < __m) \
1280 __q1--, __r1 += (d); \
1281 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1282 if (__r1 < __m) \
1283 __q1--, __r1 += (d); \
1285 __r1 -= __m; \
1287 __r0 = __r1 % __d1; \
1288 __q0 = __r1 / __d1; \
1289 __m = (UWtype) __q0 * __d0; \
1290 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
1291 if (__r0 < __m) \
1293 __q0--, __r0 += (d); \
1294 if (__r0 >= (d)) \
1295 if (__r0 < __m) \
1296 __q0--, __r0 += (d); \
1298 __r0 -= __m; \
1300 (q) = (UWtype) __q1 * __ll_B | __q0; \
1301 (r) = __r0; \
1302 } while (0)
1304 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1305 __udiv_w_sdiv (defined in libgcc or elsewhere). */
1306 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1307 #define udiv_qrnnd(q, r, nh, nl, d) \
1308 do { \
1309 USItype __r; \
1310 (q) = __udiv_w_sdiv (&__r, nh, nl, d); \
1311 (r) = __r; \
1312 } while (0)
1313 #endif
1315 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1316 #if !defined (udiv_qrnnd)
1317 #define UDIV_NEEDS_NORMALIZATION 1
1318 #define udiv_qrnnd __udiv_qrnnd_c
1319 #endif
1321 #if !defined (count_leading_zeros)
1322 extern const UQItype __clz_tab[] ATTRIBUTE_HIDDEN;
1323 #define count_leading_zeros(count, x) \
1324 do { \
1325 UWtype __xr = (x); \
1326 UWtype __a; \
1328 if (W_TYPE_SIZE <= 32) \
1330 __a = __xr < ((UWtype)1<<2*__BITS4) \
1331 ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \
1332 : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1334 else \
1336 for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1337 if (((__xr >> __a) & 0xff) != 0) \
1338 break; \
1341 (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1342 } while (0)
1343 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1344 #endif
1346 #if !defined (count_trailing_zeros)
1347 /* Define count_trailing_zeros using count_leading_zeros. The latter might be
1348 defined in asm, but if it is not, the C version above is good enough. */
1349 #define count_trailing_zeros(count, x) \
1350 do { \
1351 UWtype __ctz_x = (x); \
1352 UWtype __ctz_c; \
1353 count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \
1354 (count) = W_TYPE_SIZE - 1 - __ctz_c; \
1355 } while (0)
1356 #endif
1358 #ifndef UDIV_NEEDS_NORMALIZATION
1359 #define UDIV_NEEDS_NORMALIZATION 0
1360 #endif