?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
2008-06-04 Xinliang David Li <davidxl@google.com>
[official-gcc.git] / gcc / longlong.h
blob6c24564bd3e846db05c16ca367df25e093e3fa63
1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2 Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2004,
3 2005, 2007 Free Software Foundation, Inc.
4
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 3, or (at your option) any later version.
9
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.
14
15 You should have received a copy of the GNU General Public License
16 along with this program; see the file COPYING3. If not see
17 <http://www.gnu.org/licenses/>. */
18
19 /* You have to define the following before including this file:
20
21 UWtype -- An unsigned type, default type for operations (typically a "word")
22 UHWtype -- An unsigned type, at least half the size of UWtype.
23 UDWtype -- An unsigned type, at least twice as large a UWtype
24 W_TYPE_SIZE -- size in bits of UWtype
25
26 UQItype -- Unsigned 8 bit type.
27 SItype, USItype -- Signed and unsigned 32 bit types.
28 DItype, UDItype -- Signed and unsigned 64 bit types.
29
30 On a 32 bit machine UWtype should typically be USItype;
31 on a 64 bit machine, UWtype should typically be UDItype. */
32
33 #define __BITS4 (W_TYPE_SIZE / 4)
34 #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
35 #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
36 #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
37
38 #ifndef W_TYPE_SIZE
39 #define W_TYPE_SIZE 32
40 #define UWtype USItype
41 #define UHWtype USItype
42 #define UDWtype UDItype
43 #endif
44
45 extern const UQItype __clz_tab[256];
46
47 /* Define auxiliary asm macros.
48
49 1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
50 UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
51 word product in HIGH_PROD and LOW_PROD.
52
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.
55
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.
63
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.
67
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.
72
73 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
74 from the least significant end.
75
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.
81
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.
88
89 If any of these macros are left undefined for a particular CPU,
90 C macros are used. */
91
92 /* The CPUs come in alphabetical order below.
93
94 Please add support for more CPUs here, or improve the current support
95 for the CPUs below!
96 (E.g. WE32100, IBM360.) */
97
98 #if defined (__GNUC__) && !defined (NO_ASM)
99
100 /* We sometimes need to clobber "cc" with gcc2, but that would not be
101 understood by gcc1. Use cpp to avoid major code duplication. */
102 #if __GNUC__ < 2
103 #define __CLOBBER_CC
104 #define __AND_CLOBBER_CC
105 #else /* __GNUC__ >= 2 */
106 #define __CLOBBER_CC : "cc"
107 #define __AND_CLOBBER_CC , "cc"
108 #endif /* __GNUC__ < 2 */
109
110 #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 #define count_leading_zeros(COUNT,X) \
133 do { \
134 UDItype __xr = (X), __t, __a; \
135 __t = __builtin_alpha_cmpbge (0, __xr); \
136 __a = __clz_tab[__t ^ 0xff] - 1; \
137 __t = __builtin_alpha_extbl (__xr, __a); \
138 (COUNT) = 64 - (__clz_tab[__t] + __a*8); \
139 } while (0)
140 #define count_trailing_zeros(COUNT,X) \
141 do { \
142 UDItype __xr = (X), __t, __a; \
143 __t = __builtin_alpha_cmpbge (0, __xr); \
144 __t = ~__t & -~__t; \
145 __a = ((__t & 0xCC) != 0) * 2; \
146 __a += ((__t & 0xF0) != 0) * 4; \
147 __a += ((__t & 0xAA) != 0); \
148 __t = __builtin_alpha_extbl (__xr, __a); \
149 __a <<= 3; \
150 __t &= -__t; \
151 __a += ((__t & 0xCC) != 0) * 2; \
152 __a += ((__t & 0xF0) != 0) * 4; \
153 __a += ((__t & 0xAA) != 0); \
154 (COUNT) = __a; \
155 } while (0)
156 #endif /* __alpha_cix__ */
157 #endif /* __alpha */
158
159 #if defined (__arc__) && W_TYPE_SIZE == 32
160 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
161 __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \
162 : "=r" ((USItype) (sh)), \
163 "=&r" ((USItype) (sl)) \
164 : "%r" ((USItype) (ah)), \
165 "rIJ" ((USItype) (bh)), \
166 "%r" ((USItype) (al)), \
167 "rIJ" ((USItype) (bl)))
168 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
169 __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \
170 : "=r" ((USItype) (sh)), \
171 "=&r" ((USItype) (sl)) \
172 : "r" ((USItype) (ah)), \
173 "rIJ" ((USItype) (bh)), \
174 "r" ((USItype) (al)), \
175 "rIJ" ((USItype) (bl)))
176 /* Call libgcc routine. */
177 #define umul_ppmm(w1, w0, u, v) \
178 do { \
179 DWunion __w; \
180 __w.ll = __umulsidi3 (u, v); \
181 w1 = __w.s.high; \
182 w0 = __w.s.low; \
183 } while (0)
184 #define __umulsidi3 __umulsidi3
185 UDItype __umulsidi3 (USItype, USItype);
186 #endif
187
188 #if defined (__arm__) && !defined (__thumb__) && W_TYPE_SIZE == 32
189 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
190 __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \
191 : "=r" ((USItype) (sh)), \
192 "=&r" ((USItype) (sl)) \
193 : "%r" ((USItype) (ah)), \
194 "rI" ((USItype) (bh)), \
195 "%r" ((USItype) (al)), \
196 "rI" ((USItype) (bl)) __CLOBBER_CC)
197 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
198 __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \
199 : "=r" ((USItype) (sh)), \
200 "=&r" ((USItype) (sl)) \
201 : "r" ((USItype) (ah)), \
202 "rI" ((USItype) (bh)), \
203 "r" ((USItype) (al)), \
204 "rI" ((USItype) (bl)) __CLOBBER_CC)
205 #define umul_ppmm(xh, xl, a, b) \
206 {register USItype __t0, __t1, __t2; \
207 __asm__ ("%@ Inlined umul_ppmm\n" \
208 " mov %2, %5, lsr #16\n" \
209 " mov %0, %6, lsr #16\n" \
210 " bic %3, %5, %2, lsl #16\n" \
211 " bic %4, %6, %0, lsl #16\n" \
212 " mul %1, %3, %4\n" \
213 " mul %4, %2, %4\n" \
214 " mul %3, %0, %3\n" \
215 " mul %0, %2, %0\n" \
216 " adds %3, %4, %3\n" \
217 " addcs %0, %0, #65536\n" \
218 " adds %1, %1, %3, lsl #16\n" \
219 " adc %0, %0, %3, lsr #16" \
220 : "=&r" ((USItype) (xh)), \
221 "=r" ((USItype) (xl)), \
222 "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \
223 : "r" ((USItype) (a)), \
224 "r" ((USItype) (b)) __CLOBBER_CC );}
225 #define UMUL_TIME 20
226 #define UDIV_TIME 100
227 #endif /* __arm__ */
228
229 #if defined (__CRIS__) && __CRIS_arch_version >= 3
230 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
231 #if __CRIS_arch_version >= 8
232 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
233 #endif
234 #endif /* __CRIS__ */
235
236 #if defined (__hppa) && W_TYPE_SIZE == 32
237 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
238 __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \
239 : "=r" ((USItype) (sh)), \
240 "=&r" ((USItype) (sl)) \
241 : "%rM" ((USItype) (ah)), \
242 "rM" ((USItype) (bh)), \
243 "%rM" ((USItype) (al)), \
244 "rM" ((USItype) (bl)))
245 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
246 __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \
247 : "=r" ((USItype) (sh)), \
248 "=&r" ((USItype) (sl)) \
249 : "rM" ((USItype) (ah)), \
250 "rM" ((USItype) (bh)), \
251 "rM" ((USItype) (al)), \
252 "rM" ((USItype) (bl)))
253 #if defined (_PA_RISC1_1)
254 #define umul_ppmm(w1, w0, u, v) \
255 do { \
256 union \
257 { \
258 UDItype __f; \
259 struct {USItype __w1, __w0;} __w1w0; \
260 } __t; \
261 __asm__ ("xmpyu %1,%2,%0" \
262 : "=x" (__t.__f) \
263 : "x" ((USItype) (u)), \
264 "x" ((USItype) (v))); \
265 (w1) = __t.__w1w0.__w1; \
266 (w0) = __t.__w1w0.__w0; \
267 } while (0)
268 #define UMUL_TIME 8
269 #else
270 #define UMUL_TIME 30
271 #endif
272 #define UDIV_TIME 40
273 #define count_leading_zeros(count, x) \
274 do { \
275 USItype __tmp; \
276 __asm__ ( \
277 "ldi 1,%0\n" \
278 " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \
279 " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\
280 " ldo 16(%0),%0 ; Yes. Perform add.\n" \
281 " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \
282 " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\
283 " ldo 8(%0),%0 ; Yes. Perform add.\n" \
284 " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \
285 " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\
286 " ldo 4(%0),%0 ; Yes. Perform add.\n" \
287 " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \
288 " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\
289 " ldo 2(%0),%0 ; Yes. Perform add.\n" \
290 " extru %1,30,1,%1 ; Extract bit 1.\n" \
291 " sub %0,%1,%0 ; Subtract it.\n" \
292 : "=r" (count), "=r" (__tmp) : "1" (x)); \
293 } while (0)
294 #endif
295
296 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
297 #define smul_ppmm(xh, xl, m0, m1) \
298 do { \
299 union {DItype __ll; \
300 struct {USItype __h, __l;} __i; \
301 } __x; \
302 __asm__ ("lr %N0,%1\n\tmr %0,%2" \
303 : "=&r" (__x.__ll) \
304 : "r" (m0), "r" (m1)); \
305 (xh) = __x.__i.__h; (xl) = __x.__i.__l; \
306 } while (0)
307 #define sdiv_qrnnd(q, r, n1, n0, d) \
308 do { \
309 union {DItype __ll; \
310 struct {USItype __h, __l;} __i; \
311 } __x; \
312 __x.__i.__h = n1; __x.__i.__l = n0; \
313 __asm__ ("dr %0,%2" \
314 : "=r" (__x.__ll) \
315 : "0" (__x.__ll), "r" (d)); \
316 (q) = __x.__i.__l; (r) = __x.__i.__h; \
317 } while (0)
318 #endif
319
320 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
321 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
322 __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}" \
323 : "=r" ((USItype) (sh)), \
324 "=&r" ((USItype) (sl)) \
325 : "%0" ((USItype) (ah)), \
326 "g" ((USItype) (bh)), \
327 "%1" ((USItype) (al)), \
328 "g" ((USItype) (bl)))
329 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
330 __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}" \
331 : "=r" ((USItype) (sh)), \
332 "=&r" ((USItype) (sl)) \
333 : "0" ((USItype) (ah)), \
334 "g" ((USItype) (bh)), \
335 "1" ((USItype) (al)), \
336 "g" ((USItype) (bl)))
337 #define umul_ppmm(w1, w0, u, v) \
338 __asm__ ("mul{l} %3" \
339 : "=a" ((USItype) (w0)), \
340 "=d" ((USItype) (w1)) \
341 : "%0" ((USItype) (u)), \
342 "rm" ((USItype) (v)))
343 #define udiv_qrnnd(q, r, n1, n0, dv) \
344 __asm__ ("div{l} %4" \
345 : "=a" ((USItype) (q)), \
346 "=d" ((USItype) (r)) \
347 : "0" ((USItype) (n0)), \
348 "1" ((USItype) (n1)), \
349 "rm" ((USItype) (dv)))
350 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x))
351 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x))
352 #define UMUL_TIME 40
353 #define UDIV_TIME 40
354 #endif /* 80x86 */
355
356 #if (defined (__x86_64__) || defined (__i386__)) && W_TYPE_SIZE == 64
357 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
358 __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}" \
359 : "=r" ((UDItype) (sh)), \
360 "=&r" ((UDItype) (sl)) \
361 : "%0" ((UDItype) (ah)), \
362 "rme" ((UDItype) (bh)), \
363 "%1" ((UDItype) (al)), \
364 "rme" ((UDItype) (bl)))
365 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
366 __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}" \
367 : "=r" ((UDItype) (sh)), \
368 "=&r" ((UDItype) (sl)) \
369 : "0" ((UDItype) (ah)), \
370 "rme" ((UDItype) (bh)), \
371 "1" ((UDItype) (al)), \
372 "rme" ((UDItype) (bl)))
373 #define umul_ppmm(w1, w0, u, v) \
374 __asm__ ("mul{q} %3" \
375 : "=a" ((UDItype) (w0)), \
376 "=d" ((UDItype) (w1)) \
377 : "%0" ((UDItype) (u)), \
378 "rm" ((UDItype) (v)))
379 #define udiv_qrnnd(q, r, n1, n0, dv) \
380 __asm__ ("div{q} %4" \
381 : "=a" ((UDItype) (q)), \
382 "=d" ((UDItype) (r)) \
383 : "0" ((UDItype) (n0)), \
384 "1" ((UDItype) (n1)), \
385 "rm" ((UDItype) (dv)))
386 #define count_leading_zeros(count, x) ((count) = __builtin_clzl (x))
387 #define count_trailing_zeros(count, x) ((count) = __builtin_ctzl (x))
388 #define UMUL_TIME 40
389 #define UDIV_TIME 40
390 #endif /* x86_64 */
391
392 #if defined (__i960__) && W_TYPE_SIZE == 32
393 #define umul_ppmm(w1, w0, u, v) \
394 ({union {UDItype __ll; \
395 struct {USItype __l, __h;} __i; \
396 } __xx; \
397 __asm__ ("emul %2,%1,%0" \
398 : "=d" (__xx.__ll) \
399 : "%dI" ((USItype) (u)), \
400 "dI" ((USItype) (v))); \
401 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
402 #define __umulsidi3(u, v) \
403 ({UDItype __w; \
404 __asm__ ("emul %2,%1,%0" \
405 : "=d" (__w) \
406 : "%dI" ((USItype) (u)), \
407 "dI" ((USItype) (v))); \
408 __w; })
409 #endif /* __i960__ */
410
411 #if defined (__M32R__) && W_TYPE_SIZE == 32
412 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
413 /* The cmp clears the condition bit. */ \
414 __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \
415 : "=r" ((USItype) (sh)), \
416 "=&r" ((USItype) (sl)) \
417 : "0" ((USItype) (ah)), \
418 "r" ((USItype) (bh)), \
419 "1" ((USItype) (al)), \
420 "r" ((USItype) (bl)) \
421 : "cbit")
422 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
423 /* The cmp clears the condition bit. */ \
424 __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \
425 : "=r" ((USItype) (sh)), \
426 "=&r" ((USItype) (sl)) \
427 : "0" ((USItype) (ah)), \
428 "r" ((USItype) (bh)), \
429 "1" ((USItype) (al)), \
430 "r" ((USItype) (bl)) \
431 : "cbit")
432 #endif /* __M32R__ */
433
434 #if defined (__mc68000__) && W_TYPE_SIZE == 32
435 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
436 __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \
437 : "=d" ((USItype) (sh)), \
438 "=&d" ((USItype) (sl)) \
439 : "%0" ((USItype) (ah)), \
440 "d" ((USItype) (bh)), \
441 "%1" ((USItype) (al)), \
442 "g" ((USItype) (bl)))
443 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
444 __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \
445 : "=d" ((USItype) (sh)), \
446 "=&d" ((USItype) (sl)) \
447 : "0" ((USItype) (ah)), \
448 "d" ((USItype) (bh)), \
449 "1" ((USItype) (al)), \
450 "g" ((USItype) (bl)))
451
452 /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r. */
453 #if (defined (__mc68020__) && !defined (__mc68060__))
454 #define umul_ppmm(w1, w0, u, v) \
455 __asm__ ("mulu%.l %3,%1:%0" \
456 : "=d" ((USItype) (w0)), \
457 "=d" ((USItype) (w1)) \
458 : "%0" ((USItype) (u)), \
459 "dmi" ((USItype) (v)))
460 #define UMUL_TIME 45
461 #define udiv_qrnnd(q, r, n1, n0, d) \
462 __asm__ ("divu%.l %4,%1:%0" \
463 : "=d" ((USItype) (q)), \
464 "=d" ((USItype) (r)) \
465 : "0" ((USItype) (n0)), \
466 "1" ((USItype) (n1)), \
467 "dmi" ((USItype) (d)))
468 #define UDIV_TIME 90
469 #define sdiv_qrnnd(q, r, n1, n0, d) \
470 __asm__ ("divs%.l %4,%1:%0" \
471 : "=d" ((USItype) (q)), \
472 "=d" ((USItype) (r)) \
473 : "0" ((USItype) (n0)), \
474 "1" ((USItype) (n1)), \
475 "dmi" ((USItype) (d)))
476
477 #elif defined (__mcoldfire__) /* not mc68020 */
478
479 #define umul_ppmm(xh, xl, a, b) \
480 __asm__ ("| Inlined umul_ppmm\n" \
481 " move%.l %2,%/d0\n" \
482 " move%.l %3,%/d1\n" \
483 " move%.l %/d0,%/d2\n" \
484 " swap %/d0\n" \
485 " move%.l %/d1,%/d3\n" \
486 " swap %/d1\n" \
487 " move%.w %/d2,%/d4\n" \
488 " mulu %/d3,%/d4\n" \
489 " mulu %/d1,%/d2\n" \
490 " mulu %/d0,%/d3\n" \
491 " mulu %/d0,%/d1\n" \
492 " move%.l %/d4,%/d0\n" \
493 " clr%.w %/d0\n" \
494 " swap %/d0\n" \
495 " add%.l %/d0,%/d2\n" \
496 " add%.l %/d3,%/d2\n" \
497 " jcc 1f\n" \
498 " add%.l %#65536,%/d1\n" \
499 "1: swap %/d2\n" \
500 " moveq %#0,%/d0\n" \
501 " move%.w %/d2,%/d0\n" \
502 " move%.w %/d4,%/d2\n" \
503 " move%.l %/d2,%1\n" \
504 " add%.l %/d1,%/d0\n" \
505 " move%.l %/d0,%0" \
506 : "=g" ((USItype) (xh)), \
507 "=g" ((USItype) (xl)) \
508 : "g" ((USItype) (a)), \
509 "g" ((USItype) (b)) \
510 : "d0", "d1", "d2", "d3", "d4")
511 #define UMUL_TIME 100
512 #define UDIV_TIME 400
513 #else /* not ColdFire */
514 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */
515 #define umul_ppmm(xh, xl, a, b) \
516 __asm__ ("| Inlined umul_ppmm\n" \
517 " move%.l %2,%/d0\n" \
518 " move%.l %3,%/d1\n" \
519 " move%.l %/d0,%/d2\n" \
520 " swap %/d0\n" \
521 " move%.l %/d1,%/d3\n" \
522 " swap %/d1\n" \
523 " move%.w %/d2,%/d4\n" \
524 " mulu %/d3,%/d4\n" \
525 " mulu %/d1,%/d2\n" \
526 " mulu %/d0,%/d3\n" \
527 " mulu %/d0,%/d1\n" \
528 " move%.l %/d4,%/d0\n" \
529 " eor%.w %/d0,%/d0\n" \
530 " swap %/d0\n" \
531 " add%.l %/d0,%/d2\n" \
532 " add%.l %/d3,%/d2\n" \
533 " jcc 1f\n" \
534 " add%.l %#65536,%/d1\n" \
535 "1: swap %/d2\n" \
536 " moveq %#0,%/d0\n" \
537 " move%.w %/d2,%/d0\n" \
538 " move%.w %/d4,%/d2\n" \
539 " move%.l %/d2,%1\n" \
540 " add%.l %/d1,%/d0\n" \
541 " move%.l %/d0,%0" \
542 : "=g" ((USItype) (xh)), \
543 "=g" ((USItype) (xl)) \
544 : "g" ((USItype) (a)), \
545 "g" ((USItype) (b)) \
546 : "d0", "d1", "d2", "d3", "d4")
547 #define UMUL_TIME 100
548 #define UDIV_TIME 400
549
550 #endif /* not mc68020 */
551
552 /* The '020, '030, '040 and '060 have bitfield insns.
553 cpu32 disguises as a 68020, but lacks them. */
554 #if defined (__mc68020__) && !defined (__mcpu32__)
555 #define count_leading_zeros(count, x) \
556 __asm__ ("bfffo %1{%b2:%b2},%0" \
557 : "=d" ((USItype) (count)) \
558 : "od" ((USItype) (x)), "n" (0))
559 /* Some ColdFire architectures have a ff1 instruction supported via
560 __builtin_clz. */
561 #elif defined (__mcfisaaplus__) || defined (__mcfisac__)
562 #define count_leading_zeros(count,x) ((count) = __builtin_clz (x))
563 #define COUNT_LEADING_ZEROS_0 32
564 #endif
565 #endif /* mc68000 */
566
567 #if defined (__m88000__) && W_TYPE_SIZE == 32
568 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
569 __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \
570 : "=r" ((USItype) (sh)), \
571 "=&r" ((USItype) (sl)) \
572 : "%rJ" ((USItype) (ah)), \
573 "rJ" ((USItype) (bh)), \
574 "%rJ" ((USItype) (al)), \
575 "rJ" ((USItype) (bl)))
576 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
577 __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \
578 : "=r" ((USItype) (sh)), \
579 "=&r" ((USItype) (sl)) \
580 : "rJ" ((USItype) (ah)), \
581 "rJ" ((USItype) (bh)), \
582 "rJ" ((USItype) (al)), \
583 "rJ" ((USItype) (bl)))
584 #define count_leading_zeros(count, x) \
585 do { \
586 USItype __cbtmp; \
587 __asm__ ("ff1 %0,%1" \
588 : "=r" (__cbtmp) \
589 : "r" ((USItype) (x))); \
590 (count) = __cbtmp ^ 31; \
591 } while (0)
592 #define COUNT_LEADING_ZEROS_0 63 /* sic */
593 #if defined (__mc88110__)
594 #define umul_ppmm(wh, wl, u, v) \
595 do { \
596 union {UDItype __ll; \
597 struct {USItype __h, __l;} __i; \
598 } __xx; \
599 __asm__ ("mulu.d %0,%1,%2" \
600 : "=r" (__xx.__ll) \
601 : "r" ((USItype) (u)), \
602 "r" ((USItype) (v))); \
603 (wh) = __xx.__i.__h; \
604 (wl) = __xx.__i.__l; \
605 } while (0)
606 #define udiv_qrnnd(q, r, n1, n0, d) \
607 ({union {UDItype __ll; \
608 struct {USItype __h, __l;} __i; \
609 } __xx; \
610 USItype __q; \
611 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
612 __asm__ ("divu.d %0,%1,%2" \
613 : "=r" (__q) \
614 : "r" (__xx.__ll), \
615 "r" ((USItype) (d))); \
616 (r) = (n0) - __q * (d); (q) = __q; })
617 #define UMUL_TIME 5
618 #define UDIV_TIME 25
619 #else
620 #define UMUL_TIME 17
621 #define UDIV_TIME 150
622 #endif /* __mc88110__ */
623 #endif /* __m88000__ */
624
625 #if defined (__mips__) && W_TYPE_SIZE == 32
626 #define umul_ppmm(w1, w0, u, v) \
627 __asm__ ("multu %2,%3" \
628 : "=l" ((USItype) (w0)), \
629 "=h" ((USItype) (w1)) \
630 : "d" ((USItype) (u)), \
631 "d" ((USItype) (v)))
632 #define UMUL_TIME 10
633 #define UDIV_TIME 100
634
635 #if (__mips == 32 || __mips == 64) && ! __mips16
636 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
637 #define COUNT_LEADING_ZEROS_0 32
638 #endif
639 #endif /* __mips__ */
640
641 #if defined (__ns32000__) && W_TYPE_SIZE == 32
642 #define umul_ppmm(w1, w0, u, v) \
643 ({union {UDItype __ll; \
644 struct {USItype __l, __h;} __i; \
645 } __xx; \
646 __asm__ ("meid %2,%0" \
647 : "=g" (__xx.__ll) \
648 : "%0" ((USItype) (u)), \
649 "g" ((USItype) (v))); \
650 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
651 #define __umulsidi3(u, v) \
652 ({UDItype __w; \
653 __asm__ ("meid %2,%0" \
654 : "=g" (__w) \
655 : "%0" ((USItype) (u)), \
656 "g" ((USItype) (v))); \
657 __w; })
658 #define udiv_qrnnd(q, r, n1, n0, d) \
659 ({union {UDItype __ll; \
660 struct {USItype __l, __h;} __i; \
661 } __xx; \
662 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
663 __asm__ ("deid %2,%0" \
664 : "=g" (__xx.__ll) \
665 : "0" (__xx.__ll), \
666 "g" ((USItype) (d))); \
667 (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
668 #define count_trailing_zeros(count,x) \
669 do { \
670 __asm__ ("ffsd %2,%0" \
671 : "=r" ((USItype) (count)) \
672 : "0" ((USItype) 0), \
673 "r" ((USItype) (x))); \
674 } while (0)
675 #endif /* __ns32000__ */
676
677 /* FIXME: We should test _IBMR2 here when we add assembly support for the
678 system vendor compilers.
679 FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good
680 enough, since that hits ARM and m68k too. */
681 #if (defined (_ARCH_PPC) /* AIX */ \
682 || defined (_ARCH_PWR) /* AIX */ \
683 || defined (_ARCH_COM) /* AIX */ \
684 || defined (__powerpc__) /* gcc */ \
685 || defined (__POWERPC__) /* BEOS */ \
686 || defined (__ppc__) /* Darwin */ \
687 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
688 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
689 && CPU_FAMILY == PPC) \
690 ) && W_TYPE_SIZE == 32
691 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
692 do { \
693 if (__builtin_constant_p (bh) && (bh) == 0) \
694 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
695 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
696 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
697 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
698 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
699 else \
700 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
701 : "=r" (sh), "=&r" (sl) \
702 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
703 } while (0)
704 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
705 do { \
706 if (__builtin_constant_p (ah) && (ah) == 0) \
707 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
708 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
709 else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \
710 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
711 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
712 else if (__builtin_constant_p (bh) && (bh) == 0) \
713 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
714 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
715 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
716 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
717 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
718 else \
719 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
720 : "=r" (sh), "=&r" (sl) \
721 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
722 } while (0)
723 #define count_leading_zeros(count, x) \
724 __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x))
725 #define COUNT_LEADING_ZEROS_0 32
726 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
727 || defined (__ppc__) \
728 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
729 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
730 && CPU_FAMILY == PPC)
731 #define umul_ppmm(ph, pl, m0, m1) \
732 do { \
733 USItype __m0 = (m0), __m1 = (m1); \
734 __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
735 (pl) = __m0 * __m1; \
736 } while (0)
737 #define UMUL_TIME 15
738 #define smul_ppmm(ph, pl, m0, m1) \
739 do { \
740 SItype __m0 = (m0), __m1 = (m1); \
741 __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
742 (pl) = __m0 * __m1; \
743 } while (0)
744 #define SMUL_TIME 14
745 #define UDIV_TIME 120
746 #elif defined (_ARCH_PWR)
747 #define UMUL_TIME 8
748 #define smul_ppmm(xh, xl, m0, m1) \
749 __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1))
750 #define SMUL_TIME 4
751 #define sdiv_qrnnd(q, r, nh, nl, d) \
752 __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d))
753 #define UDIV_TIME 100
754 #endif
755 #endif /* 32-bit POWER architecture variants. */
756
757 /* We should test _IBMR2 here when we add assembly support for the system
758 vendor compilers. */
759 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
760 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
761 do { \
762 if (__builtin_constant_p (bh) && (bh) == 0) \
763 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
764 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
765 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
766 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
767 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
768 else \
769 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
770 : "=r" (sh), "=&r" (sl) \
771 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
772 } while (0)
773 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
774 do { \
775 if (__builtin_constant_p (ah) && (ah) == 0) \
776 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
777 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
778 else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \
779 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
780 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
781 else if (__builtin_constant_p (bh) && (bh) == 0) \
782 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
783 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
784 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
785 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
786 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
787 else \
788 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
789 : "=r" (sh), "=&r" (sl) \
790 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
791 } while (0)
792 #define count_leading_zeros(count, x) \
793 __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
794 #define COUNT_LEADING_ZEROS_0 64
795 #define umul_ppmm(ph, pl, m0, m1) \
796 do { \
797 UDItype __m0 = (m0), __m1 = (m1); \
798 __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
799 (pl) = __m0 * __m1; \
800 } while (0)
801 #define UMUL_TIME 15
802 #define smul_ppmm(ph, pl, m0, m1) \
803 do { \
804 DItype __m0 = (m0), __m1 = (m1); \
805 __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
806 (pl) = __m0 * __m1; \
807 } while (0)
808 #define SMUL_TIME 14 /* ??? */
809 #define UDIV_TIME 120 /* ??? */
810 #endif /* 64-bit PowerPC. */
811
812 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
813 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
814 __asm__ ("a %1,%5\n\tae %0,%3" \
815 : "=r" ((USItype) (sh)), \
816 "=&r" ((USItype) (sl)) \
817 : "%0" ((USItype) (ah)), \
818 "r" ((USItype) (bh)), \
819 "%1" ((USItype) (al)), \
820 "r" ((USItype) (bl)))
821 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
822 __asm__ ("s %1,%5\n\tse %0,%3" \
823 : "=r" ((USItype) (sh)), \
824 "=&r" ((USItype) (sl)) \
825 : "0" ((USItype) (ah)), \
826 "r" ((USItype) (bh)), \
827 "1" ((USItype) (al)), \
828 "r" ((USItype) (bl)))
829 #define umul_ppmm(ph, pl, m0, m1) \
830 do { \
831 USItype __m0 = (m0), __m1 = (m1); \
832 __asm__ ( \
833 "s r2,r2\n" \
834 " mts r10,%2\n" \
835 " m r2,%3\n" \
836 " m r2,%3\n" \
837 " m r2,%3\n" \
838 " m r2,%3\n" \
839 " m r2,%3\n" \
840 " m r2,%3\n" \
841 " m r2,%3\n" \
842 " m r2,%3\n" \
843 " m r2,%3\n" \
844 " m r2,%3\n" \
845 " m r2,%3\n" \
846 " m r2,%3\n" \
847 " m r2,%3\n" \
848 " m r2,%3\n" \
849 " m r2,%3\n" \
850 " m r2,%3\n" \
851 " cas %0,r2,r0\n" \
852 " mfs r10,%1" \
853 : "=r" ((USItype) (ph)), \
854 "=r" ((USItype) (pl)) \
855 : "%r" (__m0), \
856 "r" (__m1) \
857 : "r2"); \
858 (ph) += ((((SItype) __m0 >> 31) & __m1) \
859 + (((SItype) __m1 >> 31) & __m0)); \
860 } while (0)
861 #define UMUL_TIME 20
862 #define UDIV_TIME 200
863 #define count_leading_zeros(count, x) \
864 do { \
865 if ((x) >= 0x10000) \
866 __asm__ ("clz %0,%1" \
867 : "=r" ((USItype) (count)) \
868 : "r" ((USItype) (x) >> 16)); \
869 else \
870 { \
871 __asm__ ("clz %0,%1" \
872 : "=r" ((USItype) (count)) \
873 : "r" ((USItype) (x))); \
874 (count) += 16; \
875 } \
876 } while (0)
877 #endif
878
879 #if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
880 #ifndef __sh1__
881 #define umul_ppmm(w1, w0, u, v) \
882 __asm__ ( \
883 "dmulu.l %2,%3\n\tsts%M1 macl,%1\n\tsts%M0 mach,%0" \
884 : "=r<" ((USItype)(w1)), \
885 "=r<" ((USItype)(w0)) \
886 : "r" ((USItype)(u)), \
887 "r" ((USItype)(v)) \
888 : "macl", "mach")
889 #define UMUL_TIME 5
890 #endif
891
892 /* This is the same algorithm as __udiv_qrnnd_c. */
893 #define UDIV_NEEDS_NORMALIZATION 1
894
895 #define udiv_qrnnd(q, r, n1, n0, d) \
896 do { \
897 extern UWtype __udiv_qrnnd_16 (UWtype, UWtype) \
898 __attribute__ ((visibility ("hidden"))); \
899 /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */ \
900 __asm__ ( \
901 "mov%M4 %4,r5\n" \
902 " swap.w %3,r4\n" \
903 " swap.w r5,r6\n" \
904 " jsr @%5\n" \
905 " shll16 r6\n" \
906 " swap.w r4,r4\n" \
907 " jsr @%5\n" \
908 " swap.w r1,%0\n" \
909 " or r1,%0" \
910 : "=r" (q), "=&z" (r) \
911 : "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16) \
912 : "r1", "r2", "r4", "r5", "r6", "pr"); \
913 } while (0)
914
915 #define UDIV_TIME 80
916
917 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
918 __asm__ ("clrt;subc %5,%1; subc %4,%0" \
919 : "=r" (sh), "=r" (sl) \
920 : "0" (ah), "1" (al), "r" (bh), "r" (bl))
921
922 #endif /* __sh__ */
923
924 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
925 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
926 #define count_leading_zeros(count, x) \
927 do \
928 { \
929 UDItype x_ = (USItype)(x); \
930 SItype c_; \
931 \
932 __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \
933 (count) = c_ - 31; \
934 } \
935 while (0)
936 #define COUNT_LEADING_ZEROS_0 32
937 #endif
938
939 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
940 && W_TYPE_SIZE == 32
941 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
942 __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \
943 : "=r" ((USItype) (sh)), \
944 "=&r" ((USItype) (sl)) \
945 : "%rJ" ((USItype) (ah)), \
946 "rI" ((USItype) (bh)), \
947 "%rJ" ((USItype) (al)), \
948 "rI" ((USItype) (bl)) \
949 __CLOBBER_CC)
950 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
951 __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \
952 : "=r" ((USItype) (sh)), \
953 "=&r" ((USItype) (sl)) \
954 : "rJ" ((USItype) (ah)), \
955 "rI" ((USItype) (bh)), \
956 "rJ" ((USItype) (al)), \
957 "rI" ((USItype) (bl)) \
958 __CLOBBER_CC)
959 #if defined (__sparc_v8__)
960 #define umul_ppmm(w1, w0, u, v) \
961 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
962 : "=r" ((USItype) (w1)), \
963 "=r" ((USItype) (w0)) \
964 : "r" ((USItype) (u)), \
965 "r" ((USItype) (v)))
966 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
967 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
968 : "=&r" ((USItype) (__q)), \
969 "=&r" ((USItype) (__r)) \
970 : "r" ((USItype) (__n1)), \
971 "r" ((USItype) (__n0)), \
972 "r" ((USItype) (__d)))
973 #else
974 #if defined (__sparclite__)
975 /* This has hardware multiply but not divide. It also has two additional
976 instructions scan (ffs from high bit) and divscc. */
977 #define umul_ppmm(w1, w0, u, v) \
978 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
979 : "=r" ((USItype) (w1)), \
980 "=r" ((USItype) (w0)) \
981 : "r" ((USItype) (u)), \
982 "r" ((USItype) (v)))
983 #define udiv_qrnnd(q, r, n1, n0, d) \
984 __asm__ ("! Inlined udiv_qrnnd\n" \
985 " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
986 " tst %%g0\n" \
987 " divscc %3,%4,%%g1\n" \
988 " divscc %%g1,%4,%%g1\n" \
989 " divscc %%g1,%4,%%g1\n" \
990 " divscc %%g1,%4,%%g1\n" \
991 " divscc %%g1,%4,%%g1\n" \
992 " divscc %%g1,%4,%%g1\n" \
993 " divscc %%g1,%4,%%g1\n" \
994 " divscc %%g1,%4,%%g1\n" \
995 " divscc %%g1,%4,%%g1\n" \
996 " divscc %%g1,%4,%%g1\n" \
997 " divscc %%g1,%4,%%g1\n" \
998 " divscc %%g1,%4,%%g1\n" \
999 " divscc %%g1,%4,%%g1\n" \
1000 " divscc %%g1,%4,%%g1\n" \
1001 " divscc %%g1,%4,%%g1\n" \
1002 " divscc %%g1,%4,%%g1\n" \
1003 " divscc %%g1,%4,%%g1\n" \
1004 " divscc %%g1,%4,%%g1\n" \
1005 " divscc %%g1,%4,%%g1\n" \
1006 " divscc %%g1,%4,%%g1\n" \
1007 " divscc %%g1,%4,%%g1\n" \
1008 " divscc %%g1,%4,%%g1\n" \
1009 " divscc %%g1,%4,%%g1\n" \
1010 " divscc %%g1,%4,%%g1\n" \
1011 " divscc %%g1,%4,%%g1\n" \
1012 " divscc %%g1,%4,%%g1\n" \
1013 " divscc %%g1,%4,%%g1\n" \
1014 " divscc %%g1,%4,%%g1\n" \
1015 " divscc %%g1,%4,%%g1\n" \
1016 " divscc %%g1,%4,%%g1\n" \
1017 " divscc %%g1,%4,%%g1\n" \
1018 " divscc %%g1,%4,%0\n" \
1019 " rd %%y,%1\n" \
1020 " bl,a 1f\n" \
1021 " add %1,%4,%1\n" \
1022 "1: ! End of inline udiv_qrnnd" \
1023 : "=r" ((USItype) (q)), \
1024 "=r" ((USItype) (r)) \
1025 : "r" ((USItype) (n1)), \
1026 "r" ((USItype) (n0)), \
1027 "rI" ((USItype) (d)) \
1028 : "g1" __AND_CLOBBER_CC)
1029 #define UDIV_TIME 37
1030 #define count_leading_zeros(count, x) \
1031 do { \
1032 __asm__ ("scan %1,1,%0" \
1033 : "=r" ((USItype) (count)) \
1034 : "r" ((USItype) (x))); \
1035 } while (0)
1036 /* Early sparclites return 63 for an argument of 0, but they warn that future
1037 implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0
1038 undefined. */
1039 #else
1040 /* SPARC without integer multiplication and divide instructions.
1041 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
1042 #define umul_ppmm(w1, w0, u, v) \
1043 __asm__ ("! Inlined umul_ppmm\n" \
1044 " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\
1045 " sra %3,31,%%o5 ! Don't move this insn\n" \
1046 " and %2,%%o5,%%o5 ! Don't move this insn\n" \
1047 " andcc %%g0,0,%%g1 ! Don't move this insn\n" \
1048 " mulscc %%g1,%3,%%g1\n" \
1049 " mulscc %%g1,%3,%%g1\n" \
1050 " mulscc %%g1,%3,%%g1\n" \
1051 " mulscc %%g1,%3,%%g1\n" \
1052 " mulscc %%g1,%3,%%g1\n" \
1053 " mulscc %%g1,%3,%%g1\n" \
1054 " mulscc %%g1,%3,%%g1\n" \
1055 " mulscc %%g1,%3,%%g1\n" \
1056 " mulscc %%g1,%3,%%g1\n" \
1057 " mulscc %%g1,%3,%%g1\n" \
1058 " mulscc %%g1,%3,%%g1\n" \
1059 " mulscc %%g1,%3,%%g1\n" \
1060 " mulscc %%g1,%3,%%g1\n" \
1061 " mulscc %%g1,%3,%%g1\n" \
1062 " mulscc %%g1,%3,%%g1\n" \
1063 " mulscc %%g1,%3,%%g1\n" \
1064 " mulscc %%g1,%3,%%g1\n" \
1065 " mulscc %%g1,%3,%%g1\n" \
1066 " mulscc %%g1,%3,%%g1\n" \
1067 " mulscc %%g1,%3,%%g1\n" \
1068 " mulscc %%g1,%3,%%g1\n" \
1069 " mulscc %%g1,%3,%%g1\n" \
1070 " mulscc %%g1,%3,%%g1\n" \
1071 " mulscc %%g1,%3,%%g1\n" \
1072 " mulscc %%g1,%3,%%g1\n" \
1073 " mulscc %%g1,%3,%%g1\n" \
1074 " mulscc %%g1,%3,%%g1\n" \
1075 " mulscc %%g1,%3,%%g1\n" \
1076 " mulscc %%g1,%3,%%g1\n" \
1077 " mulscc %%g1,%3,%%g1\n" \
1078 " mulscc %%g1,%3,%%g1\n" \
1079 " mulscc %%g1,%3,%%g1\n" \
1080 " mulscc %%g1,0,%%g1\n" \
1081 " add %%g1,%%o5,%0\n" \
1082 " rd %%y,%1" \
1083 : "=r" ((USItype) (w1)), \
1084 "=r" ((USItype) (w0)) \
1085 : "%rI" ((USItype) (u)), \
1086 "r" ((USItype) (v)) \
1087 : "g1", "o5" __AND_CLOBBER_CC)
1088 #define UMUL_TIME 39 /* 39 instructions */
1089 /* It's quite necessary to add this much assembler for the sparc.
1090 The default udiv_qrnnd (in C) is more than 10 times slower! */
1091 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1092 __asm__ ("! Inlined udiv_qrnnd\n" \
1093 " mov 32,%%g1\n" \
1094 " subcc %1,%2,%%g0\n" \
1095 "1: bcs 5f\n" \
1096 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1097 " sub %1,%2,%1 ! this kills msb of n\n" \
1098 " addx %1,%1,%1 ! so this can't give carry\n" \
1099 " subcc %%g1,1,%%g1\n" \
1100 "2: bne 1b\n" \
1101 " subcc %1,%2,%%g0\n" \
1102 " bcs 3f\n" \
1103 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1104 " b 3f\n" \
1105 " sub %1,%2,%1 ! this kills msb of n\n" \
1106 "4: sub %1,%2,%1\n" \
1107 "5: addxcc %1,%1,%1\n" \
1108 " bcc 2b\n" \
1109 " subcc %%g1,1,%%g1\n" \
1110 "! Got carry from n. Subtract next step to cancel this carry.\n" \
1111 " bne 4b\n" \
1112 " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \
1113 " sub %1,%2,%1\n" \
1114 "3: xnor %0,0,%0\n" \
1115 " ! End of inline udiv_qrnnd" \
1116 : "=&r" ((USItype) (__q)), \
1117 "=&r" ((USItype) (__r)) \
1118 : "r" ((USItype) (__d)), \
1119 "1" ((USItype) (__n1)), \
1120 "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1121 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1122 #endif /* __sparclite__ */
1123 #endif /* __sparc_v8__ */
1124 #endif /* sparc32 */
1125
1126 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1127 && W_TYPE_SIZE == 64
1128 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1129 __asm__ ("addcc %r4,%5,%1\n\t" \
1130 "add %r2,%3,%0\n\t" \
1131 "bcs,a,pn %%xcc, 1f\n\t" \
1132 "add %0, 1, %0\n" \
1133 "1:" \
1134 : "=r" ((UDItype)(sh)), \
1135 "=&r" ((UDItype)(sl)) \
1136 : "%rJ" ((UDItype)(ah)), \
1137 "rI" ((UDItype)(bh)), \
1138 "%rJ" ((UDItype)(al)), \
1139 "rI" ((UDItype)(bl)) \
1140 __CLOBBER_CC)
1141
1142 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1143 __asm__ ("subcc %r4,%5,%1\n\t" \
1144 "sub %r2,%3,%0\n\t" \
1145 "bcs,a,pn %%xcc, 1f\n\t" \
1146 "sub %0, 1, %0\n\t" \
1147 "1:" \
1148 : "=r" ((UDItype)(sh)), \
1149 "=&r" ((UDItype)(sl)) \
1150 : "rJ" ((UDItype)(ah)), \
1151 "rI" ((UDItype)(bh)), \
1152 "rJ" ((UDItype)(al)), \
1153 "rI" ((UDItype)(bl)) \
1154 __CLOBBER_CC)
1155
1156 #define umul_ppmm(wh, wl, u, v) \
1157 do { \
1158 UDItype tmp1, tmp2, tmp3, tmp4; \
1159 __asm__ __volatile__ ( \
1160 "srl %7,0,%3\n\t" \
1161 "mulx %3,%6,%1\n\t" \
1162 "srlx %6,32,%2\n\t" \
1163 "mulx %2,%3,%4\n\t" \
1164 "sllx %4,32,%5\n\t" \
1165 "srl %6,0,%3\n\t" \
1166 "sub %1,%5,%5\n\t" \
1167 "srlx %5,32,%5\n\t" \
1168 "addcc %4,%5,%4\n\t" \
1169 "srlx %7,32,%5\n\t" \
1170 "mulx %3,%5,%3\n\t" \
1171 "mulx %2,%5,%5\n\t" \
1172 "sethi %%hi(0x80000000),%2\n\t" \
1173 "addcc %4,%3,%4\n\t" \
1174 "srlx %4,32,%4\n\t" \
1175 "add %2,%2,%2\n\t" \
1176 "movcc %%xcc,%%g0,%2\n\t" \
1177 "addcc %5,%4,%5\n\t" \
1178 "sllx %3,32,%3\n\t" \
1179 "add %1,%3,%1\n\t" \
1180 "add %5,%2,%0" \
1181 : "=r" ((UDItype)(wh)), \
1182 "=&r" ((UDItype)(wl)), \
1183 "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \
1184 : "r" ((UDItype)(u)), \
1185 "r" ((UDItype)(v)) \
1186 __CLOBBER_CC); \
1187 } while (0)
1188 #define UMUL_TIME 96
1189 #define UDIV_TIME 230
1190 #endif /* sparc64 */
1191
1192 #if defined (__vax__) && W_TYPE_SIZE == 32
1193 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1194 __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \
1195 : "=g" ((USItype) (sh)), \
1196 "=&g" ((USItype) (sl)) \
1197 : "%0" ((USItype) (ah)), \
1198 "g" ((USItype) (bh)), \
1199 "%1" ((USItype) (al)), \
1200 "g" ((USItype) (bl)))
1201 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1202 __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \
1203 : "=g" ((USItype) (sh)), \
1204 "=&g" ((USItype) (sl)) \
1205 : "0" ((USItype) (ah)), \
1206 "g" ((USItype) (bh)), \
1207 "1" ((USItype) (al)), \
1208 "g" ((USItype) (bl)))
1209 #define umul_ppmm(xh, xl, m0, m1) \
1210 do { \
1211 union { \
1212 UDItype __ll; \
1213 struct {USItype __l, __h;} __i; \
1214 } __xx; \
1215 USItype __m0 = (m0), __m1 = (m1); \
1216 __asm__ ("emul %1,%2,$0,%0" \
1217 : "=r" (__xx.__ll) \
1218 : "g" (__m0), \
1219 "g" (__m1)); \
1220 (xh) = __xx.__i.__h; \
1221 (xl) = __xx.__i.__l; \
1222 (xh) += ((((SItype) __m0 >> 31) & __m1) \
1223 + (((SItype) __m1 >> 31) & __m0)); \
1224 } while (0)
1225 #define sdiv_qrnnd(q, r, n1, n0, d) \
1226 do { \
1227 union {DItype __ll; \
1228 struct {SItype __l, __h;} __i; \
1229 } __xx; \
1230 __xx.__i.__h = n1; __xx.__i.__l = n0; \
1231 __asm__ ("ediv %3,%2,%0,%1" \
1232 : "=g" (q), "=g" (r) \
1233 : "g" (__xx.__ll), "g" (d)); \
1234 } while (0)
1235 #endif /* __vax__ */
1236
1237 #if defined (__xtensa__) && W_TYPE_SIZE == 32
1238 /* This code is not Xtensa-configuration-specific, so rely on the compiler
1239 to expand builtin functions depending on what configuration features
1240 are available. This avoids library calls when the operation can be
1241 performed in-line. */
1242 #define umul_ppmm(w1, w0, u, v) \
1243 do { \
1244 DWunion __w; \
1245 __w.ll = __builtin_umulsidi3 (u, v); \
1246 w1 = __w.s.high; \
1247 w0 = __w.s.low; \
1248 } while (0)
1249 #define __umulsidi3(u, v) __builtin_umulsidi3 (u, v)
1250 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
1251 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
1252 #endif /* __xtensa__ */
1253
1254 #if defined (__z8000__) && W_TYPE_SIZE == 16
1255 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1256 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1257 : "=r" ((unsigned int)(sh)), \
1258 "=&r" ((unsigned int)(sl)) \
1259 : "%0" ((unsigned int)(ah)), \
1260 "r" ((unsigned int)(bh)), \
1261 "%1" ((unsigned int)(al)), \
1262 "rQR" ((unsigned int)(bl)))
1263 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1264 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1265 : "=r" ((unsigned int)(sh)), \
1266 "=&r" ((unsigned int)(sl)) \
1267 : "0" ((unsigned int)(ah)), \
1268 "r" ((unsigned int)(bh)), \
1269 "1" ((unsigned int)(al)), \
1270 "rQR" ((unsigned int)(bl)))
1271 #define umul_ppmm(xh, xl, m0, m1) \
1272 do { \
1273 union {long int __ll; \
1274 struct {unsigned int __h, __l;} __i; \
1275 } __xx; \
1276 unsigned int __m0 = (m0), __m1 = (m1); \
1277 __asm__ ("mult %S0,%H3" \
1278 : "=r" (__xx.__i.__h), \
1279 "=r" (__xx.__i.__l) \
1280 : "%1" (__m0), \
1281 "rQR" (__m1)); \
1282 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1283 (xh) += ((((signed int) __m0 >> 15) & __m1) \
1284 + (((signed int) __m1 >> 15) & __m0)); \
1285 } while (0)
1286 #endif /* __z8000__ */
1287
1288 #endif /* __GNUC__ */
1289
1290 /* If this machine has no inline assembler, use C macros. */
1291
1292 #if !defined (add_ssaaaa)
1293 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1294 do { \
1295 UWtype __x; \
1296 __x = (al) + (bl); \
1297 (sh) = (ah) + (bh) + (__x < (al)); \
1298 (sl) = __x; \
1299 } while (0)
1300 #endif
1301
1302 #if !defined (sub_ddmmss)
1303 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1304 do { \
1305 UWtype __x; \
1306 __x = (al) - (bl); \
1307 (sh) = (ah) - (bh) - (__x > (al)); \
1308 (sl) = __x; \
1309 } while (0)
1310 #endif
1311
1312 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1313 smul_ppmm. */
1314 #if !defined (umul_ppmm) && defined (smul_ppmm)
1315 #define umul_ppmm(w1, w0, u, v) \
1316 do { \
1317 UWtype __w1; \
1318 UWtype __xm0 = (u), __xm1 = (v); \
1319 smul_ppmm (__w1, w0, __xm0, __xm1); \
1320 (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \
1321 + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \
1322 } while (0)
1323 #endif
1324
1325 /* If we still don't have umul_ppmm, define it using plain C. */
1326 #if !defined (umul_ppmm)
1327 #define umul_ppmm(w1, w0, u, v) \
1328 do { \
1329 UWtype __x0, __x1, __x2, __x3; \
1330 UHWtype __ul, __vl, __uh, __vh; \
1331 \
1332 __ul = __ll_lowpart (u); \
1333 __uh = __ll_highpart (u); \
1334 __vl = __ll_lowpart (v); \
1335 __vh = __ll_highpart (v); \
1336 \
1337 __x0 = (UWtype) __ul * __vl; \
1338 __x1 = (UWtype) __ul * __vh; \
1339 __x2 = (UWtype) __uh * __vl; \
1340 __x3 = (UWtype) __uh * __vh; \
1341 \
1342 __x1 += __ll_highpart (__x0);/* this can't give carry */ \
1343 __x1 += __x2; /* but this indeed can */ \
1344 if (__x1 < __x2) /* did we get it? */ \
1345 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1346 \
1347 (w1) = __x3 + __ll_highpart (__x1); \
1348 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
1349 } while (0)
1350 #endif
1351
1352 #if !defined (__umulsidi3)
1353 #define __umulsidi3(u, v) \
1354 ({DWunion __w; \
1355 umul_ppmm (__w.s.high, __w.s.low, u, v); \
1356 __w.ll; })
1357 #endif
1358
1359 /* Define this unconditionally, so it can be used for debugging. */
1360 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1361 do { \
1362 UWtype __d1, __d0, __q1, __q0; \
1363 UWtype __r1, __r0, __m; \
1364 __d1 = __ll_highpart (d); \
1365 __d0 = __ll_lowpart (d); \
1366 \
1367 __r1 = (n1) % __d1; \
1368 __q1 = (n1) / __d1; \
1369 __m = (UWtype) __q1 * __d0; \
1370 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
1371 if (__r1 < __m) \
1372 { \
1373 __q1--, __r1 += (d); \
1374 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1375 if (__r1 < __m) \
1376 __q1--, __r1 += (d); \
1377 } \
1378 __r1 -= __m; \
1379 \
1380 __r0 = __r1 % __d1; \
1381 __q0 = __r1 / __d1; \
1382 __m = (UWtype) __q0 * __d0; \
1383 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
1384 if (__r0 < __m) \
1385 { \
1386 __q0--, __r0 += (d); \
1387 if (__r0 >= (d)) \
1388 if (__r0 < __m) \
1389 __q0--, __r0 += (d); \
1390 } \
1391 __r0 -= __m; \
1392 \
1393 (q) = (UWtype) __q1 * __ll_B | __q0; \
1394 (r) = __r0; \
1395 } while (0)
1396
1397 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1398 __udiv_w_sdiv (defined in libgcc or elsewhere). */
1399 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1400 #define udiv_qrnnd(q, r, nh, nl, d) \
1401 do { \
1402 USItype __r; \
1403 (q) = __udiv_w_sdiv (&__r, nh, nl, d); \
1404 (r) = __r; \
1405 } while (0)
1406 #endif
1407
1408 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1409 #if !defined (udiv_qrnnd)
1410 #define UDIV_NEEDS_NORMALIZATION 1
1411 #define udiv_qrnnd __udiv_qrnnd_c
1412 #endif
1413
1414 #if !defined (count_leading_zeros)
1415 #define count_leading_zeros(count, x) \
1416 do { \
1417 UWtype __xr = (x); \
1418 UWtype __a; \
1419 \
1420 if (W_TYPE_SIZE <= 32) \
1421 { \
1422 __a = __xr < ((UWtype)1<<2*__BITS4) \
1423 ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \
1424 : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1425 } \
1426 else \
1427 { \
1428 for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1429 if (((__xr >> __a) & 0xff) != 0) \
1430 break; \
1431 } \
1432 \
1433 (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1434 } while (0)
1435 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1436 #endif
1437
1438 #if !defined (count_trailing_zeros)
1439 /* Define count_trailing_zeros using count_leading_zeros. The latter might be
1440 defined in asm, but if it is not, the C version above is good enough. */
1441 #define count_trailing_zeros(count, x) \
1442 do { \
1443 UWtype __ctz_x = (x); \
1444 UWtype __ctz_c; \
1445 count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \
1446 (count) = W_TYPE_SIZE - 1 - __ctz_c; \
1447 } while (0)
1448 #endif
1449
1450 #ifndef UDIV_NEEDS_NORMALIZATION
1451 #define UDIV_NEEDS_NORMALIZATION 0
1452 #endif
Mirror of the offical gcc git repository hosted at gcc.gnu.org