?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge trunk version 194962 into gupc branch.
[official-gcc.git] / libgcc / longlong.h
blob30e638ca1310b519e20d104115f65304b33a3f61
1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2 Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
3 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011
4 Free Software Foundation, Inc.
5
6 This file is part of the GNU C Library.
7
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
12
13 In addition to the permissions in the GNU Lesser General Public
14 License, the Free Software Foundation gives you unlimited
15 permission to link the compiled version of this file into
16 combinations with other programs, and to distribute those
17 combinations without any restriction coming from the use of this
18 file. (The Lesser General Public License restrictions do apply in
19 other respects; for example, they cover modification of the file,
20 and distribution when not linked into a combine executable.)
21
22 The GNU C Library is distributed in the hope that it will be useful,
23 but WITHOUT ANY WARRANTY; without even the implied warranty of
24 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 Lesser General Public License for more details.
26
27 You should have received a copy of the GNU Lesser General Public
28 License along with the GNU C Library; if not, see
29 <http://www.gnu.org/licenses/>. */
30
31 /* You have to define the following before including this file:
32
33 UWtype -- An unsigned type, default type for operations (typically a "word")
34 UHWtype -- An unsigned type, at least half the size of UWtype.
35 UDWtype -- An unsigned type, at least twice as large a UWtype
36 W_TYPE_SIZE -- size in bits of UWtype
37
38 UQItype -- Unsigned 8 bit type.
39 SItype, USItype -- Signed and unsigned 32 bit types.
40 DItype, UDItype -- Signed and unsigned 64 bit types.
41
42 On a 32 bit machine UWtype should typically be USItype;
43 on a 64 bit machine, UWtype should typically be UDItype. */
44
45 #define __BITS4 (W_TYPE_SIZE / 4)
46 #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
47 #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
48 #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
49
50 #ifndef W_TYPE_SIZE
51 #define W_TYPE_SIZE 32
52 #define UWtype USItype
53 #define UHWtype USItype
54 #define UDWtype UDItype
55 #endif
56
57 /* Used in glibc only. */
58 #ifndef attribute_hidden
59 #define attribute_hidden
60 #endif
61
62 extern const UQItype __clz_tab[256] attribute_hidden;
63
64 /* Define auxiliary asm macros.
65
66 1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
67 UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
68 word product in HIGH_PROD and LOW_PROD.
69
70 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
71 UDWtype product. This is just a variant of umul_ppmm.
72
73 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
74 denominator) divides a UDWtype, composed by the UWtype integers
75 HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
76 in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less
77 than DENOMINATOR for correct operation. If, in addition, the most
78 significant bit of DENOMINATOR must be 1, then the pre-processor symbol
79 UDIV_NEEDS_NORMALIZATION is defined to 1.
80
81 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
82 denominator). Like udiv_qrnnd but the numbers are signed. The quotient
83 is rounded towards 0.
84
85 5) count_leading_zeros(count, x) counts the number of zero-bits from the
86 msb to the first nonzero bit in the UWtype X. This is the number of
87 steps X needs to be shifted left to set the msb. Undefined for X == 0,
88 unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
89
90 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
91 from the least significant end.
92
93 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
94 high_addend_2, low_addend_2) adds two UWtype integers, composed by
95 HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
96 respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow
97 (i.e. carry out) is not stored anywhere, and is lost.
98
99 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
100 high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
101 composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
102 LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE
103 and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
104 and is lost.
105
106 If any of these macros are left undefined for a particular CPU,
107 C macros are used. */
108
109 /* The CPUs come in alphabetical order below.
110
111 Please add support for more CPUs here, or improve the current support
112 for the CPUs below!
113 (E.g. WE32100, IBM360.) */
114
115 #if defined (__GNUC__) && !defined (NO_ASM)
116
117 /* We sometimes need to clobber "cc" with gcc2, but that would not be
118 understood by gcc1. Use cpp to avoid major code duplication. */
119 #if __GNUC__ < 2
120 #define __CLOBBER_CC
121 #define __AND_CLOBBER_CC
122 #else /* __GNUC__ >= 2 */
123 #define __CLOBBER_CC : "cc"
124 #define __AND_CLOBBER_CC , "cc"
125 #endif /* __GNUC__ < 2 */
126
127 #if defined (__alpha) && W_TYPE_SIZE == 64
128 #define umul_ppmm(ph, pl, m0, m1) \
129 do { \
130 UDItype __m0 = (m0), __m1 = (m1); \
131 (ph) = __builtin_alpha_umulh (__m0, __m1); \
132 (pl) = __m0 * __m1; \
133 } while (0)
134 #define UMUL_TIME 46
135 #ifndef LONGLONG_STANDALONE
136 #define udiv_qrnnd(q, r, n1, n0, d) \
137 do { UDItype __r; \
138 (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \
139 (r) = __r; \
140 } while (0)
141 extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
142 #define UDIV_TIME 220
143 #endif /* LONGLONG_STANDALONE */
144 #ifdef __alpha_cix__
145 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X))
146 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
147 #define COUNT_LEADING_ZEROS_0 64
148 #else
149 #define count_leading_zeros(COUNT,X) \
150 do { \
151 UDItype __xr = (X), __t, __a; \
152 __t = __builtin_alpha_cmpbge (0, __xr); \
153 __a = __clz_tab[__t ^ 0xff] - 1; \
154 __t = __builtin_alpha_extbl (__xr, __a); \
155 (COUNT) = 64 - (__clz_tab[__t] + __a*8); \
156 } while (0)
157 #define count_trailing_zeros(COUNT,X) \
158 do { \
159 UDItype __xr = (X), __t, __a; \
160 __t = __builtin_alpha_cmpbge (0, __xr); \
161 __t = ~__t & -~__t; \
162 __a = ((__t & 0xCC) != 0) * 2; \
163 __a += ((__t & 0xF0) != 0) * 4; \
164 __a += ((__t & 0xAA) != 0); \
165 __t = __builtin_alpha_extbl (__xr, __a); \
166 __a <<= 3; \
167 __t &= -__t; \
168 __a += ((__t & 0xCC) != 0) * 2; \
169 __a += ((__t & 0xF0) != 0) * 4; \
170 __a += ((__t & 0xAA) != 0); \
171 (COUNT) = __a; \
172 } while (0)
173 #endif /* __alpha_cix__ */
174 #endif /* __alpha */
175
176 #if defined (__arc__) && W_TYPE_SIZE == 32
177 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
178 __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \
179 : "=r" ((USItype) (sh)), \
180 "=&r" ((USItype) (sl)) \
181 : "%r" ((USItype) (ah)), \
182 "rIJ" ((USItype) (bh)), \
183 "%r" ((USItype) (al)), \
184 "rIJ" ((USItype) (bl)))
185 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
186 __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \
187 : "=r" ((USItype) (sh)), \
188 "=&r" ((USItype) (sl)) \
189 : "r" ((USItype) (ah)), \
190 "rIJ" ((USItype) (bh)), \
191 "r" ((USItype) (al)), \
192 "rIJ" ((USItype) (bl)))
193 /* Call libgcc routine. */
194 #define umul_ppmm(w1, w0, u, v) \
195 do { \
196 DWunion __w; \
197 __w.ll = __umulsidi3 (u, v); \
198 w1 = __w.s.high; \
199 w0 = __w.s.low; \
200 } while (0)
201 #define __umulsidi3 __umulsidi3
202 UDItype __umulsidi3 (USItype, USItype);
203 #endif
204
205 #if defined (__arm__) && (defined (__thumb2__) || !defined (__thumb__)) \
206 && W_TYPE_SIZE == 32
207 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
208 __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \
209 : "=r" ((USItype) (sh)), \
210 "=&r" ((USItype) (sl)) \
211 : "%r" ((USItype) (ah)), \
212 "rI" ((USItype) (bh)), \
213 "%r" ((USItype) (al)), \
214 "rI" ((USItype) (bl)) __CLOBBER_CC)
215 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
216 __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \
217 : "=r" ((USItype) (sh)), \
218 "=&r" ((USItype) (sl)) \
219 : "r" ((USItype) (ah)), \
220 "rI" ((USItype) (bh)), \
221 "r" ((USItype) (al)), \
222 "rI" ((USItype) (bl)) __CLOBBER_CC)
223 # if defined(__ARM_ARCH_2__) || defined(__ARM_ARCH_2A__) \
224 || defined(__ARM_ARCH_3__)
225 # define umul_ppmm(xh, xl, a, b) \
226 do { \
227 register USItype __t0, __t1, __t2; \
228 __asm__ ("%@ Inlined umul_ppmm\n" \
229 " mov %2, %5, lsr #16\n" \
230 " mov %0, %6, lsr #16\n" \
231 " bic %3, %5, %2, lsl #16\n" \
232 " bic %4, %6, %0, lsl #16\n" \
233 " mul %1, %3, %4\n" \
234 " mul %4, %2, %4\n" \
235 " mul %3, %0, %3\n" \
236 " mul %0, %2, %0\n" \
237 " adds %3, %4, %3\n" \
238 " addcs %0, %0, #65536\n" \
239 " adds %1, %1, %3, lsl #16\n" \
240 " adc %0, %0, %3, lsr #16" \
241 : "=&r" ((USItype) (xh)), \
242 "=r" ((USItype) (xl)), \
243 "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \
244 : "r" ((USItype) (a)), \
245 "r" ((USItype) (b)) __CLOBBER_CC ); \
246 } while (0)
247 # define UMUL_TIME 20
248 # else
249 # define umul_ppmm(xh, xl, a, b) \
250 do { \
251 /* Generate umull, under compiler control. */ \
252 register UDItype __t0 = (UDItype)(USItype)(a) * (USItype)(b); \
253 (xl) = (USItype)__t0; \
254 (xh) = (USItype)(__t0 >> 32); \
255 } while (0)
256 # define UMUL_TIME 3
257 # endif
258 # define UDIV_TIME 100
259 #endif /* __arm__ */
260
261 #if defined(__arm__)
262 /* Let gcc decide how best to implement count_leading_zeros. */
263 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
264 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X))
265 #define COUNT_LEADING_ZEROS_0 32
266 #endif
267
268 #if defined (__AVR__)
269
270 #if W_TYPE_SIZE == 16
271 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
272 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X))
273 #define COUNT_LEADING_ZEROS_0 16
274 #endif /* W_TYPE_SIZE == 16 */
275
276 #if W_TYPE_SIZE == 32
277 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X))
278 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
279 #define COUNT_LEADING_ZEROS_0 32
280 #endif /* W_TYPE_SIZE == 32 */
281
282 #if W_TYPE_SIZE == 64
283 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzll (X))
284 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzll (X))
285 #define COUNT_LEADING_ZEROS_0 64
286 #endif /* W_TYPE_SIZE == 64 */
287
288 #endif /* defined (__AVR__) */
289
290 #if defined (__CRIS__) && __CRIS_arch_version >= 3
291 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
292 #if __CRIS_arch_version >= 8
293 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
294 #endif
295 #endif /* __CRIS__ */
296
297 #if defined (__hppa) && W_TYPE_SIZE == 32
298 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
299 __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \
300 : "=r" ((USItype) (sh)), \
301 "=&r" ((USItype) (sl)) \
302 : "%rM" ((USItype) (ah)), \
303 "rM" ((USItype) (bh)), \
304 "%rM" ((USItype) (al)), \
305 "rM" ((USItype) (bl)))
306 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
307 __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \
308 : "=r" ((USItype) (sh)), \
309 "=&r" ((USItype) (sl)) \
310 : "rM" ((USItype) (ah)), \
311 "rM" ((USItype) (bh)), \
312 "rM" ((USItype) (al)), \
313 "rM" ((USItype) (bl)))
314 #if defined (_PA_RISC1_1)
315 #define umul_ppmm(w1, w0, u, v) \
316 do { \
317 union \
318 { \
319 UDItype __f; \
320 struct {USItype __w1, __w0;} __w1w0; \
321 } __t; \
322 __asm__ ("xmpyu %1,%2,%0" \
323 : "=x" (__t.__f) \
324 : "x" ((USItype) (u)), \
325 "x" ((USItype) (v))); \
326 (w1) = __t.__w1w0.__w1; \
327 (w0) = __t.__w1w0.__w0; \
328 } while (0)
329 #define UMUL_TIME 8
330 #else
331 #define UMUL_TIME 30
332 #endif
333 #define UDIV_TIME 40
334 #define count_leading_zeros(count, x) \
335 do { \
336 USItype __tmp; \
337 __asm__ ( \
338 "ldi 1,%0\n" \
339 " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \
340 " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\
341 " ldo 16(%0),%0 ; Yes. Perform add.\n" \
342 " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \
343 " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\
344 " ldo 8(%0),%0 ; Yes. Perform add.\n" \
345 " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \
346 " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\
347 " ldo 4(%0),%0 ; Yes. Perform add.\n" \
348 " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \
349 " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\
350 " ldo 2(%0),%0 ; Yes. Perform add.\n" \
351 " extru %1,30,1,%1 ; Extract bit 1.\n" \
352 " sub %0,%1,%0 ; Subtract it.\n" \
353 : "=r" (count), "=r" (__tmp) : "1" (x)); \
354 } while (0)
355 #endif
356
357 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
358 #if !defined (__zarch__)
359 #define smul_ppmm(xh, xl, m0, m1) \
360 do { \
361 union {DItype __ll; \
362 struct {USItype __h, __l;} __i; \
363 } __x; \
364 __asm__ ("lr %N0,%1\n\tmr %0,%2" \
365 : "=&r" (__x.__ll) \
366 : "r" (m0), "r" (m1)); \
367 (xh) = __x.__i.__h; (xl) = __x.__i.__l; \
368 } while (0)
369 #define sdiv_qrnnd(q, r, n1, n0, d) \
370 do { \
371 union {DItype __ll; \
372 struct {USItype __h, __l;} __i; \
373 } __x; \
374 __x.__i.__h = n1; __x.__i.__l = n0; \
375 __asm__ ("dr %0,%2" \
376 : "=r" (__x.__ll) \
377 : "0" (__x.__ll), "r" (d)); \
378 (q) = __x.__i.__l; (r) = __x.__i.__h; \
379 } while (0)
380 #else
381 #define smul_ppmm(xh, xl, m0, m1) \
382 do { \
383 register SItype __r0 __asm__ ("0"); \
384 register SItype __r1 __asm__ ("1") = (m0); \
385 \
386 __asm__ ("mr\t%%r0,%3" \
387 : "=r" (__r0), "=r" (__r1) \
388 : "r" (__r1), "r" (m1)); \
389 (xh) = __r0; (xl) = __r1; \
390 } while (0)
391
392 #define sdiv_qrnnd(q, r, n1, n0, d) \
393 do { \
394 register SItype __r0 __asm__ ("0") = (n1); \
395 register SItype __r1 __asm__ ("1") = (n0); \
396 \
397 __asm__ ("dr\t%%r0,%4" \
398 : "=r" (__r0), "=r" (__r1) \
399 : "r" (__r0), "r" (__r1), "r" (d)); \
400 (q) = __r1; (r) = __r0; \
401 } while (0)
402 #endif /* __zarch__ */
403 #endif
404
405 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
406 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
407 __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}" \
408 : "=r" ((USItype) (sh)), \
409 "=&r" ((USItype) (sl)) \
410 : "%0" ((USItype) (ah)), \
411 "g" ((USItype) (bh)), \
412 "%1" ((USItype) (al)), \
413 "g" ((USItype) (bl)))
414 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
415 __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}" \
416 : "=r" ((USItype) (sh)), \
417 "=&r" ((USItype) (sl)) \
418 : "0" ((USItype) (ah)), \
419 "g" ((USItype) (bh)), \
420 "1" ((USItype) (al)), \
421 "g" ((USItype) (bl)))
422 #define umul_ppmm(w1, w0, u, v) \
423 __asm__ ("mul{l} %3" \
424 : "=a" ((USItype) (w0)), \
425 "=d" ((USItype) (w1)) \
426 : "%0" ((USItype) (u)), \
427 "rm" ((USItype) (v)))
428 #define udiv_qrnnd(q, r, n1, n0, dv) \
429 __asm__ ("div{l} %4" \
430 : "=a" ((USItype) (q)), \
431 "=d" ((USItype) (r)) \
432 : "0" ((USItype) (n0)), \
433 "1" ((USItype) (n1)), \
434 "rm" ((USItype) (dv)))
435 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x))
436 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x))
437 #define UMUL_TIME 40
438 #define UDIV_TIME 40
439 #endif /* 80x86 */
440
441 #if (defined (__x86_64__) || defined (__i386__)) && W_TYPE_SIZE == 64
442 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
443 __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}" \
444 : "=r" ((UDItype) (sh)), \
445 "=&r" ((UDItype) (sl)) \
446 : "%0" ((UDItype) (ah)), \
447 "rme" ((UDItype) (bh)), \
448 "%1" ((UDItype) (al)), \
449 "rme" ((UDItype) (bl)))
450 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
451 __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}" \
452 : "=r" ((UDItype) (sh)), \
453 "=&r" ((UDItype) (sl)) \
454 : "0" ((UDItype) (ah)), \
455 "rme" ((UDItype) (bh)), \
456 "1" ((UDItype) (al)), \
457 "rme" ((UDItype) (bl)))
458 #define umul_ppmm(w1, w0, u, v) \
459 __asm__ ("mul{q} %3" \
460 : "=a" ((UDItype) (w0)), \
461 "=d" ((UDItype) (w1)) \
462 : "%0" ((UDItype) (u)), \
463 "rm" ((UDItype) (v)))
464 #define udiv_qrnnd(q, r, n1, n0, dv) \
465 __asm__ ("div{q} %4" \
466 : "=a" ((UDItype) (q)), \
467 "=d" ((UDItype) (r)) \
468 : "0" ((UDItype) (n0)), \
469 "1" ((UDItype) (n1)), \
470 "rm" ((UDItype) (dv)))
471 #define count_leading_zeros(count, x) ((count) = __builtin_clzll (x))
472 #define count_trailing_zeros(count, x) ((count) = __builtin_ctzll (x))
473 #define UMUL_TIME 40
474 #define UDIV_TIME 40
475 #endif /* x86_64 */
476
477 #if defined (__i960__) && W_TYPE_SIZE == 32
478 #define umul_ppmm(w1, w0, u, v) \
479 ({union {UDItype __ll; \
480 struct {USItype __l, __h;} __i; \
481 } __xx; \
482 __asm__ ("emul %2,%1,%0" \
483 : "=d" (__xx.__ll) \
484 : "%dI" ((USItype) (u)), \
485 "dI" ((USItype) (v))); \
486 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
487 #define __umulsidi3(u, v) \
488 ({UDItype __w; \
489 __asm__ ("emul %2,%1,%0" \
490 : "=d" (__w) \
491 : "%dI" ((USItype) (u)), \
492 "dI" ((USItype) (v))); \
493 __w; })
494 #endif /* __i960__ */
495
496 #if defined (__ia64) && W_TYPE_SIZE == 64
497 /* This form encourages gcc (pre-release 3.4 at least) to emit predicated
498 "sub r=r,r" and "sub r=r,r,1", giving a 2 cycle latency. The generic
499 code using "al<bl" arithmetically comes out making an actual 0 or 1 in a
500 register, which takes an extra cycle. */
501 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
502 do { \
503 UWtype __x; \
504 __x = (al) - (bl); \
505 if ((al) < (bl)) \
506 (sh) = (ah) - (bh) - 1; \
507 else \
508 (sh) = (ah) - (bh); \
509 (sl) = __x; \
510 } while (0)
511
512 /* Do both product parts in assembly, since that gives better code with
513 all gcc versions. Some callers will just use the upper part, and in
514 that situation we waste an instruction, but not any cycles. */
515 #define umul_ppmm(ph, pl, m0, m1) \
516 __asm__ ("xma.hu %0 = %2, %3, f0\n\txma.l %1 = %2, %3, f0" \
517 : "=&f" (ph), "=f" (pl) \
518 : "f" (m0), "f" (m1))
519 #define count_leading_zeros(count, x) \
520 do { \
521 UWtype _x = (x), _y, _a, _c; \
522 __asm__ ("mux1 %0 = %1, @rev" : "=r" (_y) : "r" (_x)); \
523 __asm__ ("czx1.l %0 = %1" : "=r" (_a) : "r" (-_y | _y)); \
524 _c = (_a - 1) << 3; \
525 _x >>= _c; \
526 if (_x >= 1 << 4) \
527 _x >>= 4, _c += 4; \
528 if (_x >= 1 << 2) \
529 _x >>= 2, _c += 2; \
530 _c += _x >> 1; \
531 (count) = W_TYPE_SIZE - 1 - _c; \
532 } while (0)
533 /* similar to what gcc does for __builtin_ffs, but 0 based rather than 1
534 based, and we don't need a special case for x==0 here */
535 #define count_trailing_zeros(count, x) \
536 do { \
537 UWtype __ctz_x = (x); \
538 __asm__ ("popcnt %0 = %1" \
539 : "=r" (count) \
540 : "r" ((__ctz_x-1) & ~__ctz_x)); \
541 } while (0)
542 #define UMUL_TIME 14
543 #endif
544
545 #if defined (__M32R__) && W_TYPE_SIZE == 32
546 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
547 /* The cmp clears the condition bit. */ \
548 __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \
549 : "=r" ((USItype) (sh)), \
550 "=&r" ((USItype) (sl)) \
551 : "0" ((USItype) (ah)), \
552 "r" ((USItype) (bh)), \
553 "1" ((USItype) (al)), \
554 "r" ((USItype) (bl)) \
555 : "cbit")
556 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
557 /* The cmp clears the condition bit. */ \
558 __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \
559 : "=r" ((USItype) (sh)), \
560 "=&r" ((USItype) (sl)) \
561 : "0" ((USItype) (ah)), \
562 "r" ((USItype) (bh)), \
563 "1" ((USItype) (al)), \
564 "r" ((USItype) (bl)) \
565 : "cbit")
566 #endif /* __M32R__ */
567
568 #if defined (__mc68000__) && W_TYPE_SIZE == 32
569 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
570 __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \
571 : "=d" ((USItype) (sh)), \
572 "=&d" ((USItype) (sl)) \
573 : "%0" ((USItype) (ah)), \
574 "d" ((USItype) (bh)), \
575 "%1" ((USItype) (al)), \
576 "g" ((USItype) (bl)))
577 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
578 __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \
579 : "=d" ((USItype) (sh)), \
580 "=&d" ((USItype) (sl)) \
581 : "0" ((USItype) (ah)), \
582 "d" ((USItype) (bh)), \
583 "1" ((USItype) (al)), \
584 "g" ((USItype) (bl)))
585
586 /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r. */
587 #if (defined (__mc68020__) && !defined (__mc68060__))
588 #define umul_ppmm(w1, w0, u, v) \
589 __asm__ ("mulu%.l %3,%1:%0" \
590 : "=d" ((USItype) (w0)), \
591 "=d" ((USItype) (w1)) \
592 : "%0" ((USItype) (u)), \
593 "dmi" ((USItype) (v)))
594 #define UMUL_TIME 45
595 #define udiv_qrnnd(q, r, n1, n0, d) \
596 __asm__ ("divu%.l %4,%1:%0" \
597 : "=d" ((USItype) (q)), \
598 "=d" ((USItype) (r)) \
599 : "0" ((USItype) (n0)), \
600 "1" ((USItype) (n1)), \
601 "dmi" ((USItype) (d)))
602 #define UDIV_TIME 90
603 #define sdiv_qrnnd(q, r, n1, n0, d) \
604 __asm__ ("divs%.l %4,%1:%0" \
605 : "=d" ((USItype) (q)), \
606 "=d" ((USItype) (r)) \
607 : "0" ((USItype) (n0)), \
608 "1" ((USItype) (n1)), \
609 "dmi" ((USItype) (d)))
610
611 #elif defined (__mcoldfire__) /* not mc68020 */
612
613 #define umul_ppmm(xh, xl, a, b) \
614 __asm__ ("| Inlined umul_ppmm\n" \
615 " move%.l %2,%/d0\n" \
616 " move%.l %3,%/d1\n" \
617 " move%.l %/d0,%/d2\n" \
618 " swap %/d0\n" \
619 " move%.l %/d1,%/d3\n" \
620 " swap %/d1\n" \
621 " move%.w %/d2,%/d4\n" \
622 " mulu %/d3,%/d4\n" \
623 " mulu %/d1,%/d2\n" \
624 " mulu %/d0,%/d3\n" \
625 " mulu %/d0,%/d1\n" \
626 " move%.l %/d4,%/d0\n" \
627 " clr%.w %/d0\n" \
628 " swap %/d0\n" \
629 " add%.l %/d0,%/d2\n" \
630 " add%.l %/d3,%/d2\n" \
631 " jcc 1f\n" \
632 " add%.l %#65536,%/d1\n" \
633 "1: swap %/d2\n" \
634 " moveq %#0,%/d0\n" \
635 " move%.w %/d2,%/d0\n" \
636 " move%.w %/d4,%/d2\n" \
637 " move%.l %/d2,%1\n" \
638 " add%.l %/d1,%/d0\n" \
639 " move%.l %/d0,%0" \
640 : "=g" ((USItype) (xh)), \
641 "=g" ((USItype) (xl)) \
642 : "g" ((USItype) (a)), \
643 "g" ((USItype) (b)) \
644 : "d0", "d1", "d2", "d3", "d4")
645 #define UMUL_TIME 100
646 #define UDIV_TIME 400
647 #else /* not ColdFire */
648 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */
649 #define umul_ppmm(xh, xl, a, b) \
650 __asm__ ("| Inlined umul_ppmm\n" \
651 " move%.l %2,%/d0\n" \
652 " move%.l %3,%/d1\n" \
653 " move%.l %/d0,%/d2\n" \
654 " swap %/d0\n" \
655 " move%.l %/d1,%/d3\n" \
656 " swap %/d1\n" \
657 " move%.w %/d2,%/d4\n" \
658 " mulu %/d3,%/d4\n" \
659 " mulu %/d1,%/d2\n" \
660 " mulu %/d0,%/d3\n" \
661 " mulu %/d0,%/d1\n" \
662 " move%.l %/d4,%/d0\n" \
663 " eor%.w %/d0,%/d0\n" \
664 " swap %/d0\n" \
665 " add%.l %/d0,%/d2\n" \
666 " add%.l %/d3,%/d2\n" \
667 " jcc 1f\n" \
668 " add%.l %#65536,%/d1\n" \
669 "1: swap %/d2\n" \
670 " moveq %#0,%/d0\n" \
671 " move%.w %/d2,%/d0\n" \
672 " move%.w %/d4,%/d2\n" \
673 " move%.l %/d2,%1\n" \
674 " add%.l %/d1,%/d0\n" \
675 " move%.l %/d0,%0" \
676 : "=g" ((USItype) (xh)), \
677 "=g" ((USItype) (xl)) \
678 : "g" ((USItype) (a)), \
679 "g" ((USItype) (b)) \
680 : "d0", "d1", "d2", "d3", "d4")
681 #define UMUL_TIME 100
682 #define UDIV_TIME 400
683
684 #endif /* not mc68020 */
685
686 /* The '020, '030, '040 and '060 have bitfield insns.
687 cpu32 disguises as a 68020, but lacks them. */
688 #if defined (__mc68020__) && !defined (__mcpu32__)
689 #define count_leading_zeros(count, x) \
690 __asm__ ("bfffo %1{%b2:%b2},%0" \
691 : "=d" ((USItype) (count)) \
692 : "od" ((USItype) (x)), "n" (0))
693 /* Some ColdFire architectures have a ff1 instruction supported via
694 __builtin_clz. */
695 #elif defined (__mcfisaaplus__) || defined (__mcfisac__)
696 #define count_leading_zeros(count,x) ((count) = __builtin_clz (x))
697 #define COUNT_LEADING_ZEROS_0 32
698 #endif
699 #endif /* mc68000 */
700
701 #if defined (__m88000__) && W_TYPE_SIZE == 32
702 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
703 __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \
704 : "=r" ((USItype) (sh)), \
705 "=&r" ((USItype) (sl)) \
706 : "%rJ" ((USItype) (ah)), \
707 "rJ" ((USItype) (bh)), \
708 "%rJ" ((USItype) (al)), \
709 "rJ" ((USItype) (bl)))
710 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
711 __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \
712 : "=r" ((USItype) (sh)), \
713 "=&r" ((USItype) (sl)) \
714 : "rJ" ((USItype) (ah)), \
715 "rJ" ((USItype) (bh)), \
716 "rJ" ((USItype) (al)), \
717 "rJ" ((USItype) (bl)))
718 #define count_leading_zeros(count, x) \
719 do { \
720 USItype __cbtmp; \
721 __asm__ ("ff1 %0,%1" \
722 : "=r" (__cbtmp) \
723 : "r" ((USItype) (x))); \
724 (count) = __cbtmp ^ 31; \
725 } while (0)
726 #define COUNT_LEADING_ZEROS_0 63 /* sic */
727 #if defined (__mc88110__)
728 #define umul_ppmm(wh, wl, u, v) \
729 do { \
730 union {UDItype __ll; \
731 struct {USItype __h, __l;} __i; \
732 } __xx; \
733 __asm__ ("mulu.d %0,%1,%2" \
734 : "=r" (__xx.__ll) \
735 : "r" ((USItype) (u)), \
736 "r" ((USItype) (v))); \
737 (wh) = __xx.__i.__h; \
738 (wl) = __xx.__i.__l; \
739 } while (0)
740 #define udiv_qrnnd(q, r, n1, n0, d) \
741 ({union {UDItype __ll; \
742 struct {USItype __h, __l;} __i; \
743 } __xx; \
744 USItype __q; \
745 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
746 __asm__ ("divu.d %0,%1,%2" \
747 : "=r" (__q) \
748 : "r" (__xx.__ll), \
749 "r" ((USItype) (d))); \
750 (r) = (n0) - __q * (d); (q) = __q; })
751 #define UMUL_TIME 5
752 #define UDIV_TIME 25
753 #else
754 #define UMUL_TIME 17
755 #define UDIV_TIME 150
756 #endif /* __mc88110__ */
757 #endif /* __m88000__ */
758
759 #if defined (__mn10300__)
760 # if defined (__AM33__)
761 # define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
762 # define umul_ppmm(w1, w0, u, v) \
763 asm("mulu %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
764 # define smul_ppmm(w1, w0, u, v) \
765 asm("mul %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
766 # else
767 # define umul_ppmm(w1, w0, u, v) \
768 asm("nop; nop; mulu %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
769 # define smul_ppmm(w1, w0, u, v) \
770 asm("nop; nop; mul %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
771 # endif
772 # define add_ssaaaa(sh, sl, ah, al, bh, bl) \
773 do { \
774 DWunion __s, __a, __b; \
775 __a.s.low = (al); __a.s.high = (ah); \
776 __b.s.low = (bl); __b.s.high = (bh); \
777 __s.ll = __a.ll + __b.ll; \
778 (sl) = __s.s.low; (sh) = __s.s.high; \
779 } while (0)
780 # define sub_ddmmss(sh, sl, ah, al, bh, bl) \
781 do { \
782 DWunion __s, __a, __b; \
783 __a.s.low = (al); __a.s.high = (ah); \
784 __b.s.low = (bl); __b.s.high = (bh); \
785 __s.ll = __a.ll - __b.ll; \
786 (sl) = __s.s.low; (sh) = __s.s.high; \
787 } while (0)
788 # define udiv_qrnnd(q, r, nh, nl, d) \
789 asm("divu %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
790 # define sdiv_qrnnd(q, r, nh, nl, d) \
791 asm("div %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
792 # define UMUL_TIME 3
793 # define UDIV_TIME 38
794 #endif
795
796 #if defined (__mips__) && W_TYPE_SIZE == 32
797 #define umul_ppmm(w1, w0, u, v) \
798 do { \
799 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \
800 (w1) = (USItype) (__x >> 32); \
801 (w0) = (USItype) (__x); \
802 } while (0)
803 #define UMUL_TIME 10
804 #define UDIV_TIME 100
805
806 #if (__mips == 32 || __mips == 64) && ! __mips16
807 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X))
808 #define COUNT_LEADING_ZEROS_0 32
809 #endif
810 #endif /* __mips__ */
811
812 #if defined (__ns32000__) && W_TYPE_SIZE == 32
813 #define umul_ppmm(w1, w0, u, v) \
814 ({union {UDItype __ll; \
815 struct {USItype __l, __h;} __i; \
816 } __xx; \
817 __asm__ ("meid %2,%0" \
818 : "=g" (__xx.__ll) \
819 : "%0" ((USItype) (u)), \
820 "g" ((USItype) (v))); \
821 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
822 #define __umulsidi3(u, v) \
823 ({UDItype __w; \
824 __asm__ ("meid %2,%0" \
825 : "=g" (__w) \
826 : "%0" ((USItype) (u)), \
827 "g" ((USItype) (v))); \
828 __w; })
829 #define udiv_qrnnd(q, r, n1, n0, d) \
830 ({union {UDItype __ll; \
831 struct {USItype __l, __h;} __i; \
832 } __xx; \
833 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
834 __asm__ ("deid %2,%0" \
835 : "=g" (__xx.__ll) \
836 : "0" (__xx.__ll), \
837 "g" ((USItype) (d))); \
838 (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
839 #define count_trailing_zeros(count,x) \
840 do { \
841 __asm__ ("ffsd %2,%0" \
842 : "=r" ((USItype) (count)) \
843 : "0" ((USItype) 0), \
844 "r" ((USItype) (x))); \
845 } while (0)
846 #endif /* __ns32000__ */
847
848 /* FIXME: We should test _IBMR2 here when we add assembly support for the
849 system vendor compilers.
850 FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good
851 enough, since that hits ARM and m68k too. */
852 #if (defined (_ARCH_PPC) /* AIX */ \
853 || defined (__powerpc__) /* gcc */ \
854 || defined (__POWERPC__) /* BEOS */ \
855 || defined (__ppc__) /* Darwin */ \
856 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
857 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
858 && CPU_FAMILY == PPC) \
859 ) && W_TYPE_SIZE == 32
860 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
861 do { \
862 if (__builtin_constant_p (bh) && (bh) == 0) \
863 __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2" \
864 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
865 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
866 __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2" \
867 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
868 else \
869 __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3" \
870 : "=r" (sh), "=&r" (sl) \
871 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
872 } while (0)
873 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
874 do { \
875 if (__builtin_constant_p (ah) && (ah) == 0) \
876 __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2" \
877 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
878 else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \
879 __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2" \
880 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
881 else if (__builtin_constant_p (bh) && (bh) == 0) \
882 __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2" \
883 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
884 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
885 __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2" \
886 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
887 else \
888 __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2" \
889 : "=r" (sh), "=&r" (sl) \
890 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
891 } while (0)
892 #define count_leading_zeros(count, x) \
893 __asm__ ("cntlzw %0,%1" : "=r" (count) : "r" (x))
894 #define COUNT_LEADING_ZEROS_0 32
895 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
896 || defined (__ppc__) \
897 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
898 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
899 && CPU_FAMILY == PPC)
900 #define umul_ppmm(ph, pl, m0, m1) \
901 do { \
902 USItype __m0 = (m0), __m1 = (m1); \
903 __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
904 (pl) = __m0 * __m1; \
905 } while (0)
906 #define UMUL_TIME 15
907 #define smul_ppmm(ph, pl, m0, m1) \
908 do { \
909 SItype __m0 = (m0), __m1 = (m1); \
910 __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
911 (pl) = __m0 * __m1; \
912 } while (0)
913 #define SMUL_TIME 14
914 #define UDIV_TIME 120
915 #endif
916 #endif /* 32-bit POWER architecture variants. */
917
918 /* We should test _IBMR2 here when we add assembly support for the system
919 vendor compilers. */
920 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
921 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
922 do { \
923 if (__builtin_constant_p (bh) && (bh) == 0) \
924 __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2" \
925 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
926 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
927 __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2" \
928 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
929 else \
930 __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3" \
931 : "=r" (sh), "=&r" (sl) \
932 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
933 } while (0)
934 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
935 do { \
936 if (__builtin_constant_p (ah) && (ah) == 0) \
937 __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2" \
938 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
939 else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \
940 __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2" \
941 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
942 else if (__builtin_constant_p (bh) && (bh) == 0) \
943 __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2" \
944 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
945 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
946 __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2" \
947 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
948 else \
949 __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2" \
950 : "=r" (sh), "=&r" (sl) \
951 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
952 } while (0)
953 #define count_leading_zeros(count, x) \
954 __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
955 #define COUNT_LEADING_ZEROS_0 64
956 #define umul_ppmm(ph, pl, m0, m1) \
957 do { \
958 UDItype __m0 = (m0), __m1 = (m1); \
959 __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
960 (pl) = __m0 * __m1; \
961 } while (0)
962 #define UMUL_TIME 15
963 #define smul_ppmm(ph, pl, m0, m1) \
964 do { \
965 DItype __m0 = (m0), __m1 = (m1); \
966 __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
967 (pl) = __m0 * __m1; \
968 } while (0)
969 #define SMUL_TIME 14 /* ??? */
970 #define UDIV_TIME 120 /* ??? */
971 #endif /* 64-bit PowerPC. */
972
973 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
974 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
975 __asm__ ("a %1,%5\n\tae %0,%3" \
976 : "=r" ((USItype) (sh)), \
977 "=&r" ((USItype) (sl)) \
978 : "%0" ((USItype) (ah)), \
979 "r" ((USItype) (bh)), \
980 "%1" ((USItype) (al)), \
981 "r" ((USItype) (bl)))
982 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
983 __asm__ ("s %1,%5\n\tse %0,%3" \
984 : "=r" ((USItype) (sh)), \
985 "=&r" ((USItype) (sl)) \
986 : "0" ((USItype) (ah)), \
987 "r" ((USItype) (bh)), \
988 "1" ((USItype) (al)), \
989 "r" ((USItype) (bl)))
990 #define umul_ppmm(ph, pl, m0, m1) \
991 do { \
992 USItype __m0 = (m0), __m1 = (m1); \
993 __asm__ ( \
994 "s r2,r2\n" \
995 " mts r10,%2\n" \
996 " m r2,%3\n" \
997 " m r2,%3\n" \
998 " m r2,%3\n" \
999 " m r2,%3\n" \
1000 " m r2,%3\n" \
1001 " m r2,%3\n" \
1002 " m r2,%3\n" \
1003 " m r2,%3\n" \
1004 " m r2,%3\n" \
1005 " m r2,%3\n" \
1006 " m r2,%3\n" \
1007 " m r2,%3\n" \
1008 " m r2,%3\n" \
1009 " m r2,%3\n" \
1010 " m r2,%3\n" \
1011 " m r2,%3\n" \
1012 " cas %0,r2,r0\n" \
1013 " mfs r10,%1" \
1014 : "=r" ((USItype) (ph)), \
1015 "=r" ((USItype) (pl)) \
1016 : "%r" (__m0), \
1017 "r" (__m1) \
1018 : "r2"); \
1019 (ph) += ((((SItype) __m0 >> 31) & __m1) \
1020 + (((SItype) __m1 >> 31) & __m0)); \
1021 } while (0)
1022 #define UMUL_TIME 20
1023 #define UDIV_TIME 200
1024 #define count_leading_zeros(count, x) \
1025 do { \
1026 if ((x) >= 0x10000) \
1027 __asm__ ("clz %0,%1" \
1028 : "=r" ((USItype) (count)) \
1029 : "r" ((USItype) (x) >> 16)); \
1030 else \
1031 { \
1032 __asm__ ("clz %0,%1" \
1033 : "=r" ((USItype) (count)) \
1034 : "r" ((USItype) (x))); \
1035 (count) += 16; \
1036 } \
1037 } while (0)
1038 #endif
1039
1040 #if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
1041 #ifndef __sh1__
1042 #define umul_ppmm(w1, w0, u, v) \
1043 __asm__ ( \
1044 "dmulu.l %2,%3\n\tsts%M1 macl,%1\n\tsts%M0 mach,%0" \
1045 : "=r<" ((USItype)(w1)), \
1046 "=r<" ((USItype)(w0)) \
1047 : "r" ((USItype)(u)), \
1048 "r" ((USItype)(v)) \
1049 : "macl", "mach")
1050 #define UMUL_TIME 5
1051 #endif
1052
1053 /* This is the same algorithm as __udiv_qrnnd_c. */
1054 #define UDIV_NEEDS_NORMALIZATION 1
1055
1056 #define udiv_qrnnd(q, r, n1, n0, d) \
1057 do { \
1058 extern UWtype __udiv_qrnnd_16 (UWtype, UWtype) \
1059 __attribute__ ((visibility ("hidden"))); \
1060 /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */ \
1061 __asm__ ( \
1062 "mov%M4 %4,r5\n" \
1063 " swap.w %3,r4\n" \
1064 " swap.w r5,r6\n" \
1065 " jsr @%5\n" \
1066 " shll16 r6\n" \
1067 " swap.w r4,r4\n" \
1068 " jsr @%5\n" \
1069 " swap.w r1,%0\n" \
1070 " or r1,%0" \
1071 : "=r" (q), "=&z" (r) \
1072 : "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16) \
1073 : "r1", "r2", "r4", "r5", "r6", "pr", "t"); \
1074 } while (0)
1075
1076 #define UDIV_TIME 80
1077
1078 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1079 __asm__ ("clrt;subc %5,%1; subc %4,%0" \
1080 : "=r" (sh), "=r" (sl) \
1081 : "0" (ah), "1" (al), "r" (bh), "r" (bl) : "t")
1082
1083 #endif /* __sh__ */
1084
1085 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
1086 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1087 #define count_leading_zeros(count, x) \
1088 do \
1089 { \
1090 UDItype x_ = (USItype)(x); \
1091 SItype c_; \
1092 \
1093 __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \
1094 (count) = c_ - 31; \
1095 } \
1096 while (0)
1097 #define COUNT_LEADING_ZEROS_0 32
1098 #endif
1099
1100 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
1101 && W_TYPE_SIZE == 32
1102 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1103 __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \
1104 : "=r" ((USItype) (sh)), \
1105 "=&r" ((USItype) (sl)) \
1106 : "%rJ" ((USItype) (ah)), \
1107 "rI" ((USItype) (bh)), \
1108 "%rJ" ((USItype) (al)), \
1109 "rI" ((USItype) (bl)) \
1110 __CLOBBER_CC)
1111 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1112 __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \
1113 : "=r" ((USItype) (sh)), \
1114 "=&r" ((USItype) (sl)) \
1115 : "rJ" ((USItype) (ah)), \
1116 "rI" ((USItype) (bh)), \
1117 "rJ" ((USItype) (al)), \
1118 "rI" ((USItype) (bl)) \
1119 __CLOBBER_CC)
1120 #if defined (__sparc_v9__)
1121 #define umul_ppmm(w1, w0, u, v) \
1122 do { \
1123 register USItype __g1 asm ("g1"); \
1124 __asm__ ("umul\t%2,%3,%1\n\t" \
1125 "srlx\t%1, 32, %0" \
1126 : "=r" ((USItype) (w1)), \
1127 "=r" (__g1) \
1128 : "r" ((USItype) (u)), \
1129 "r" ((USItype) (v))); \
1130 (w0) = __g1; \
1131 } while (0)
1132 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1133 __asm__ ("mov\t%2,%%y\n\t" \
1134 "udiv\t%3,%4,%0\n\t" \
1135 "umul\t%0,%4,%1\n\t" \
1136 "sub\t%3,%1,%1" \
1137 : "=&r" ((USItype) (__q)), \
1138 "=&r" ((USItype) (__r)) \
1139 : "r" ((USItype) (__n1)), \
1140 "r" ((USItype) (__n0)), \
1141 "r" ((USItype) (__d)))
1142 #else
1143 #if defined (__sparc_v8__)
1144 #define umul_ppmm(w1, w0, u, v) \
1145 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1146 : "=r" ((USItype) (w1)), \
1147 "=r" ((USItype) (w0)) \
1148 : "r" ((USItype) (u)), \
1149 "r" ((USItype) (v)))
1150 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1151 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
1152 : "=&r" ((USItype) (__q)), \
1153 "=&r" ((USItype) (__r)) \
1154 : "r" ((USItype) (__n1)), \
1155 "r" ((USItype) (__n0)), \
1156 "r" ((USItype) (__d)))
1157 #else
1158 #if defined (__sparclite__)
1159 /* This has hardware multiply but not divide. It also has two additional
1160 instructions scan (ffs from high bit) and divscc. */
1161 #define umul_ppmm(w1, w0, u, v) \
1162 __asm__ ("umul %2,%3,%1;rd %%y,%0" \
1163 : "=r" ((USItype) (w1)), \
1164 "=r" ((USItype) (w0)) \
1165 : "r" ((USItype) (u)), \
1166 "r" ((USItype) (v)))
1167 #define udiv_qrnnd(q, r, n1, n0, d) \
1168 __asm__ ("! Inlined udiv_qrnnd\n" \
1169 " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
1170 " tst %%g0\n" \
1171 " divscc %3,%4,%%g1\n" \
1172 " divscc %%g1,%4,%%g1\n" \
1173 " divscc %%g1,%4,%%g1\n" \
1174 " divscc %%g1,%4,%%g1\n" \
1175 " divscc %%g1,%4,%%g1\n" \
1176 " divscc %%g1,%4,%%g1\n" \
1177 " divscc %%g1,%4,%%g1\n" \
1178 " divscc %%g1,%4,%%g1\n" \
1179 " divscc %%g1,%4,%%g1\n" \
1180 " divscc %%g1,%4,%%g1\n" \
1181 " divscc %%g1,%4,%%g1\n" \
1182 " divscc %%g1,%4,%%g1\n" \
1183 " divscc %%g1,%4,%%g1\n" \
1184 " divscc %%g1,%4,%%g1\n" \
1185 " divscc %%g1,%4,%%g1\n" \
1186 " divscc %%g1,%4,%%g1\n" \
1187 " divscc %%g1,%4,%%g1\n" \
1188 " divscc %%g1,%4,%%g1\n" \
1189 " divscc %%g1,%4,%%g1\n" \
1190 " divscc %%g1,%4,%%g1\n" \
1191 " divscc %%g1,%4,%%g1\n" \
1192 " divscc %%g1,%4,%%g1\n" \
1193 " divscc %%g1,%4,%%g1\n" \
1194 " divscc %%g1,%4,%%g1\n" \
1195 " divscc %%g1,%4,%%g1\n" \
1196 " divscc %%g1,%4,%%g1\n" \
1197 " divscc %%g1,%4,%%g1\n" \
1198 " divscc %%g1,%4,%%g1\n" \
1199 " divscc %%g1,%4,%%g1\n" \
1200 " divscc %%g1,%4,%%g1\n" \
1201 " divscc %%g1,%4,%%g1\n" \
1202 " divscc %%g1,%4,%0\n" \
1203 " rd %%y,%1\n" \
1204 " bl,a 1f\n" \
1205 " add %1,%4,%1\n" \
1206 "1: ! End of inline udiv_qrnnd" \
1207 : "=r" ((USItype) (q)), \
1208 "=r" ((USItype) (r)) \
1209 : "r" ((USItype) (n1)), \
1210 "r" ((USItype) (n0)), \
1211 "rI" ((USItype) (d)) \
1212 : "g1" __AND_CLOBBER_CC)
1213 #define UDIV_TIME 37
1214 #define count_leading_zeros(count, x) \
1215 do { \
1216 __asm__ ("scan %1,1,%0" \
1217 : "=r" ((USItype) (count)) \
1218 : "r" ((USItype) (x))); \
1219 } while (0)
1220 /* Early sparclites return 63 for an argument of 0, but they warn that future
1221 implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0
1222 undefined. */
1223 #else
1224 /* SPARC without integer multiplication and divide instructions.
1225 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
1226 #define umul_ppmm(w1, w0, u, v) \
1227 __asm__ ("! Inlined umul_ppmm\n" \
1228 " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\
1229 " sra %3,31,%%o5 ! Don't move this insn\n" \
1230 " and %2,%%o5,%%o5 ! Don't move this insn\n" \
1231 " andcc %%g0,0,%%g1 ! Don't move this insn\n" \
1232 " mulscc %%g1,%3,%%g1\n" \
1233 " mulscc %%g1,%3,%%g1\n" \
1234 " mulscc %%g1,%3,%%g1\n" \
1235 " mulscc %%g1,%3,%%g1\n" \
1236 " mulscc %%g1,%3,%%g1\n" \
1237 " mulscc %%g1,%3,%%g1\n" \
1238 " mulscc %%g1,%3,%%g1\n" \
1239 " mulscc %%g1,%3,%%g1\n" \
1240 " mulscc %%g1,%3,%%g1\n" \
1241 " mulscc %%g1,%3,%%g1\n" \
1242 " mulscc %%g1,%3,%%g1\n" \
1243 " mulscc %%g1,%3,%%g1\n" \
1244 " mulscc %%g1,%3,%%g1\n" \
1245 " mulscc %%g1,%3,%%g1\n" \
1246 " mulscc %%g1,%3,%%g1\n" \
1247 " mulscc %%g1,%3,%%g1\n" \
1248 " mulscc %%g1,%3,%%g1\n" \
1249 " mulscc %%g1,%3,%%g1\n" \
1250 " mulscc %%g1,%3,%%g1\n" \
1251 " mulscc %%g1,%3,%%g1\n" \
1252 " mulscc %%g1,%3,%%g1\n" \
1253 " mulscc %%g1,%3,%%g1\n" \
1254 " mulscc %%g1,%3,%%g1\n" \
1255 " mulscc %%g1,%3,%%g1\n" \
1256 " mulscc %%g1,%3,%%g1\n" \
1257 " mulscc %%g1,%3,%%g1\n" \
1258 " mulscc %%g1,%3,%%g1\n" \
1259 " mulscc %%g1,%3,%%g1\n" \
1260 " mulscc %%g1,%3,%%g1\n" \
1261 " mulscc %%g1,%3,%%g1\n" \
1262 " mulscc %%g1,%3,%%g1\n" \
1263 " mulscc %%g1,%3,%%g1\n" \
1264 " mulscc %%g1,0,%%g1\n" \
1265 " add %%g1,%%o5,%0\n" \
1266 " rd %%y,%1" \
1267 : "=r" ((USItype) (w1)), \
1268 "=r" ((USItype) (w0)) \
1269 : "%rI" ((USItype) (u)), \
1270 "r" ((USItype) (v)) \
1271 : "g1", "o5" __AND_CLOBBER_CC)
1272 #define UMUL_TIME 39 /* 39 instructions */
1273 /* It's quite necessary to add this much assembler for the sparc.
1274 The default udiv_qrnnd (in C) is more than 10 times slower! */
1275 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1276 __asm__ ("! Inlined udiv_qrnnd\n" \
1277 " mov 32,%%g1\n" \
1278 " subcc %1,%2,%%g0\n" \
1279 "1: bcs 5f\n" \
1280 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1281 " sub %1,%2,%1 ! this kills msb of n\n" \
1282 " addx %1,%1,%1 ! so this can't give carry\n" \
1283 " subcc %%g1,1,%%g1\n" \
1284 "2: bne 1b\n" \
1285 " subcc %1,%2,%%g0\n" \
1286 " bcs 3f\n" \
1287 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1288 " b 3f\n" \
1289 " sub %1,%2,%1 ! this kills msb of n\n" \
1290 "4: sub %1,%2,%1\n" \
1291 "5: addxcc %1,%1,%1\n" \
1292 " bcc 2b\n" \
1293 " subcc %%g1,1,%%g1\n" \
1294 "! Got carry from n. Subtract next step to cancel this carry.\n" \
1295 " bne 4b\n" \
1296 " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \
1297 " sub %1,%2,%1\n" \
1298 "3: xnor %0,0,%0\n" \
1299 " ! End of inline udiv_qrnnd" \
1300 : "=&r" ((USItype) (__q)), \
1301 "=&r" ((USItype) (__r)) \
1302 : "r" ((USItype) (__d)), \
1303 "1" ((USItype) (__n1)), \
1304 "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1305 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1306 #endif /* __sparclite__ */
1307 #endif /* __sparc_v8__ */
1308 #endif /* __sparc_v9__ */
1309 #endif /* sparc32 */
1310
1311 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1312 && W_TYPE_SIZE == 64
1313 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1314 do { \
1315 UDItype __carry = 0; \
1316 __asm__ ("addcc\t%r5,%6,%1\n\t" \
1317 "add\t%r3,%4,%0\n\t" \
1318 "movcs\t%%xcc, 1, %2\n\t" \
1319 "add\t%0, %2, %0" \
1320 : "=r" ((UDItype)(sh)), \
1321 "=&r" ((UDItype)(sl)), \
1322 "+r" (__carry) \
1323 : "%rJ" ((UDItype)(ah)), \
1324 "rI" ((UDItype)(bh)), \
1325 "%rJ" ((UDItype)(al)), \
1326 "rI" ((UDItype)(bl)) \
1327 __CLOBBER_CC); \
1328 } while (0)
1329
1330 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1331 do { \
1332 UDItype __carry = 0; \
1333 __asm__ ("subcc\t%r5,%6,%1\n\t" \
1334 "sub\t%r3,%4,%0\n\t" \
1335 "movcs\t%%xcc, 1, %2\n\t" \
1336 "sub\t%0, %2, %0" \
1337 : "=r" ((UDItype)(sh)), \
1338 "=&r" ((UDItype)(sl)), \
1339 "+r" (__carry) \
1340 : "%rJ" ((UDItype)(ah)), \
1341 "rI" ((UDItype)(bh)), \
1342 "%rJ" ((UDItype)(al)), \
1343 "rI" ((UDItype)(bl)) \
1344 __CLOBBER_CC); \
1345 } while (0)
1346
1347 #define umul_ppmm(wh, wl, u, v) \
1348 do { \
1349 UDItype tmp1, tmp2, tmp3, tmp4; \
1350 __asm__ __volatile__ ( \
1351 "srl %7,0,%3\n\t" \
1352 "mulx %3,%6,%1\n\t" \
1353 "srlx %6,32,%2\n\t" \
1354 "mulx %2,%3,%4\n\t" \
1355 "sllx %4,32,%5\n\t" \
1356 "srl %6,0,%3\n\t" \
1357 "sub %1,%5,%5\n\t" \
1358 "srlx %5,32,%5\n\t" \
1359 "addcc %4,%5,%4\n\t" \
1360 "srlx %7,32,%5\n\t" \
1361 "mulx %3,%5,%3\n\t" \
1362 "mulx %2,%5,%5\n\t" \
1363 "sethi %%hi(0x80000000),%2\n\t" \
1364 "addcc %4,%3,%4\n\t" \
1365 "srlx %4,32,%4\n\t" \
1366 "add %2,%2,%2\n\t" \
1367 "movcc %%xcc,%%g0,%2\n\t" \
1368 "addcc %5,%4,%5\n\t" \
1369 "sllx %3,32,%3\n\t" \
1370 "add %1,%3,%1\n\t" \
1371 "add %5,%2,%0" \
1372 : "=r" ((UDItype)(wh)), \
1373 "=&r" ((UDItype)(wl)), \
1374 "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \
1375 : "r" ((UDItype)(u)), \
1376 "r" ((UDItype)(v)) \
1377 __CLOBBER_CC); \
1378 } while (0)
1379 #define UMUL_TIME 96
1380 #define UDIV_TIME 230
1381 #endif /* sparc64 */
1382
1383 #if defined (__vax__) && W_TYPE_SIZE == 32
1384 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1385 __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \
1386 : "=g" ((USItype) (sh)), \
1387 "=&g" ((USItype) (sl)) \
1388 : "%0" ((USItype) (ah)), \
1389 "g" ((USItype) (bh)), \
1390 "%1" ((USItype) (al)), \
1391 "g" ((USItype) (bl)))
1392 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1393 __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \
1394 : "=g" ((USItype) (sh)), \
1395 "=&g" ((USItype) (sl)) \
1396 : "0" ((USItype) (ah)), \
1397 "g" ((USItype) (bh)), \
1398 "1" ((USItype) (al)), \
1399 "g" ((USItype) (bl)))
1400 #define umul_ppmm(xh, xl, m0, m1) \
1401 do { \
1402 union { \
1403 UDItype __ll; \
1404 struct {USItype __l, __h;} __i; \
1405 } __xx; \
1406 USItype __m0 = (m0), __m1 = (m1); \
1407 __asm__ ("emul %1,%2,$0,%0" \
1408 : "=r" (__xx.__ll) \
1409 : "g" (__m0), \
1410 "g" (__m1)); \
1411 (xh) = __xx.__i.__h; \
1412 (xl) = __xx.__i.__l; \
1413 (xh) += ((((SItype) __m0 >> 31) & __m1) \
1414 + (((SItype) __m1 >> 31) & __m0)); \
1415 } while (0)
1416 #define sdiv_qrnnd(q, r, n1, n0, d) \
1417 do { \
1418 union {DItype __ll; \
1419 struct {SItype __l, __h;} __i; \
1420 } __xx; \
1421 __xx.__i.__h = n1; __xx.__i.__l = n0; \
1422 __asm__ ("ediv %3,%2,%0,%1" \
1423 : "=g" (q), "=g" (r) \
1424 : "g" (__xx.__ll), "g" (d)); \
1425 } while (0)
1426 #endif /* __vax__ */
1427
1428 #ifdef _TMS320C6X
1429 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1430 do \
1431 { \
1432 UDItype __ll; \
1433 __asm__ ("addu .l1 %1, %2, %0" \
1434 : "=a" (__ll) : "a" (al), "a" (bl)); \
1435 (sl) = (USItype)__ll; \
1436 (sh) = ((USItype)(__ll >> 32)) + (ah) + (bh); \
1437 } \
1438 while (0)
1439
1440 #ifdef _TMS320C6400_PLUS
1441 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1442 #define umul_ppmm(w1, w0, u, v) \
1443 do { \
1444 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \
1445 (w1) = (USItype) (__x >> 32); \
1446 (w0) = (USItype) (__x); \
1447 } while (0)
1448 #endif /* _TMS320C6400_PLUS */
1449
1450 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x))
1451 #ifdef _TMS320C6400
1452 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x))
1453 #endif
1454 #define UMUL_TIME 4
1455 #define UDIV_TIME 40
1456 #endif /* _TMS320C6X */
1457
1458 #if defined (__xtensa__) && W_TYPE_SIZE == 32
1459 /* This code is not Xtensa-configuration-specific, so rely on the compiler
1460 to expand builtin functions depending on what configuration features
1461 are available. This avoids library calls when the operation can be
1462 performed in-line. */
1463 #define umul_ppmm(w1, w0, u, v) \
1464 do { \
1465 DWunion __w; \
1466 __w.ll = __builtin_umulsidi3 (u, v); \
1467 w1 = __w.s.high; \
1468 w0 = __w.s.low; \
1469 } while (0)
1470 #define __umulsidi3(u, v) __builtin_umulsidi3 (u, v)
1471 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
1472 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
1473 #endif /* __xtensa__ */
1474
1475 #if defined xstormy16
1476 extern UHItype __stormy16_count_leading_zeros (UHItype);
1477 #define count_leading_zeros(count, x) \
1478 do \
1479 { \
1480 UHItype size; \
1481 \
1482 /* We assume that W_TYPE_SIZE is a multiple of 16... */ \
1483 for ((count) = 0, size = W_TYPE_SIZE; size; size -= 16) \
1484 { \
1485 UHItype c; \
1486 \
1487 c = __clzhi2 ((x) >> (size - 16)); \
1488 (count) += c; \
1489 if (c != 16) \
1490 break; \
1491 } \
1492 } \
1493 while (0)
1494 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1495 #endif
1496
1497 #if defined (__z8000__) && W_TYPE_SIZE == 16
1498 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1499 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1500 : "=r" ((unsigned int)(sh)), \
1501 "=&r" ((unsigned int)(sl)) \
1502 : "%0" ((unsigned int)(ah)), \
1503 "r" ((unsigned int)(bh)), \
1504 "%1" ((unsigned int)(al)), \
1505 "rQR" ((unsigned int)(bl)))
1506 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1507 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1508 : "=r" ((unsigned int)(sh)), \
1509 "=&r" ((unsigned int)(sl)) \
1510 : "0" ((unsigned int)(ah)), \
1511 "r" ((unsigned int)(bh)), \
1512 "1" ((unsigned int)(al)), \
1513 "rQR" ((unsigned int)(bl)))
1514 #define umul_ppmm(xh, xl, m0, m1) \
1515 do { \
1516 union {long int __ll; \
1517 struct {unsigned int __h, __l;} __i; \
1518 } __xx; \
1519 unsigned int __m0 = (m0), __m1 = (m1); \
1520 __asm__ ("mult %S0,%H3" \
1521 : "=r" (__xx.__i.__h), \
1522 "=r" (__xx.__i.__l) \
1523 : "%1" (__m0), \
1524 "rQR" (__m1)); \
1525 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1526 (xh) += ((((signed int) __m0 >> 15) & __m1) \
1527 + (((signed int) __m1 >> 15) & __m0)); \
1528 } while (0)
1529 #endif /* __z8000__ */
1530
1531 #endif /* __GNUC__ */
1532
1533 /* If this machine has no inline assembler, use C macros. */
1534
1535 #if !defined (add_ssaaaa)
1536 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1537 do { \
1538 UWtype __x; \
1539 __x = (al) + (bl); \
1540 (sh) = (ah) + (bh) + (__x < (al)); \
1541 (sl) = __x; \
1542 } while (0)
1543 #endif
1544
1545 #if !defined (sub_ddmmss)
1546 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1547 do { \
1548 UWtype __x; \
1549 __x = (al) - (bl); \
1550 (sh) = (ah) - (bh) - (__x > (al)); \
1551 (sl) = __x; \
1552 } while (0)
1553 #endif
1554
1555 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1556 smul_ppmm. */
1557 #if !defined (umul_ppmm) && defined (smul_ppmm)
1558 #define umul_ppmm(w1, w0, u, v) \
1559 do { \
1560 UWtype __w1; \
1561 UWtype __xm0 = (u), __xm1 = (v); \
1562 smul_ppmm (__w1, w0, __xm0, __xm1); \
1563 (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \
1564 + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \
1565 } while (0)
1566 #endif
1567
1568 /* If we still don't have umul_ppmm, define it using plain C. */
1569 #if !defined (umul_ppmm)
1570 #define umul_ppmm(w1, w0, u, v) \
1571 do { \
1572 UWtype __x0, __x1, __x2, __x3; \
1573 UHWtype __ul, __vl, __uh, __vh; \
1574 \
1575 __ul = __ll_lowpart (u); \
1576 __uh = __ll_highpart (u); \
1577 __vl = __ll_lowpart (v); \
1578 __vh = __ll_highpart (v); \
1579 \
1580 __x0 = (UWtype) __ul * __vl; \
1581 __x1 = (UWtype) __ul * __vh; \
1582 __x2 = (UWtype) __uh * __vl; \
1583 __x3 = (UWtype) __uh * __vh; \
1584 \
1585 __x1 += __ll_highpart (__x0);/* this can't give carry */ \
1586 __x1 += __x2; /* but this indeed can */ \
1587 if (__x1 < __x2) /* did we get it? */ \
1588 __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1589 \
1590 (w1) = __x3 + __ll_highpart (__x1); \
1591 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
1592 } while (0)
1593 #endif
1594
1595 #if !defined (__umulsidi3)
1596 #define __umulsidi3(u, v) \
1597 ({DWunion __w; \
1598 umul_ppmm (__w.s.high, __w.s.low, u, v); \
1599 __w.ll; })
1600 #endif
1601
1602 /* Define this unconditionally, so it can be used for debugging. */
1603 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1604 do { \
1605 UWtype __d1, __d0, __q1, __q0; \
1606 UWtype __r1, __r0, __m; \
1607 __d1 = __ll_highpart (d); \
1608 __d0 = __ll_lowpart (d); \
1609 \
1610 __r1 = (n1) % __d1; \
1611 __q1 = (n1) / __d1; \
1612 __m = (UWtype) __q1 * __d0; \
1613 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
1614 if (__r1 < __m) \
1615 { \
1616 __q1--, __r1 += (d); \
1617 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1618 if (__r1 < __m) \
1619 __q1--, __r1 += (d); \
1620 } \
1621 __r1 -= __m; \
1622 \
1623 __r0 = __r1 % __d1; \
1624 __q0 = __r1 / __d1; \
1625 __m = (UWtype) __q0 * __d0; \
1626 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
1627 if (__r0 < __m) \
1628 { \
1629 __q0--, __r0 += (d); \
1630 if (__r0 >= (d)) \
1631 if (__r0 < __m) \
1632 __q0--, __r0 += (d); \
1633 } \
1634 __r0 -= __m; \
1635 \
1636 (q) = (UWtype) __q1 * __ll_B | __q0; \
1637 (r) = __r0; \
1638 } while (0)
1639
1640 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1641 __udiv_w_sdiv (defined in libgcc or elsewhere). */
1642 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1643 #define udiv_qrnnd(q, r, nh, nl, d) \
1644 do { \
1645 USItype __r; \
1646 (q) = __udiv_w_sdiv (&__r, nh, nl, d); \
1647 (r) = __r; \
1648 } while (0)
1649 #endif
1650
1651 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1652 #if !defined (udiv_qrnnd)
1653 #define UDIV_NEEDS_NORMALIZATION 1
1654 #define udiv_qrnnd __udiv_qrnnd_c
1655 #endif
1656
1657 #if !defined (count_leading_zeros)
1658 #define count_leading_zeros(count, x) \
1659 do { \
1660 UWtype __xr = (x); \
1661 UWtype __a; \
1662 \
1663 if (W_TYPE_SIZE <= 32) \
1664 { \
1665 __a = __xr < ((UWtype)1<<2*__BITS4) \
1666 ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \
1667 : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1668 } \
1669 else \
1670 { \
1671 for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1672 if (((__xr >> __a) & 0xff) != 0) \
1673 break; \
1674 } \
1675 \
1676 (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1677 } while (0)
1678 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1679 #endif
1680
1681 #if !defined (count_trailing_zeros)
1682 /* Define count_trailing_zeros using count_leading_zeros. The latter might be
1683 defined in asm, but if it is not, the C version above is good enough. */
1684 #define count_trailing_zeros(count, x) \
1685 do { \
1686 UWtype __ctz_x = (x); \
1687 UWtype __ctz_c; \
1688 count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \
1689 (count) = W_TYPE_SIZE - 1 - __ctz_c; \
1690 } while (0)
1691 #endif
1692
1693 #ifndef UDIV_NEEDS_NORMALIZATION
1694 #define UDIV_NEEDS_NORMALIZATION 0
1695 #endif
Mirror of the offical gcc git repository hosted at gcc.gnu.org