1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2    Copyright (C) 1991-2013 Free Software Foundation, Inc.
3 
4    This file is part of the GNU C Library.
5 
6    The GNU C Library is free software; you can redistribute it and/or
7    modify it under the terms of the GNU Lesser General Public
8    License as published by the Free Software Foundation; either
9    version 2.1 of the License, or (at your option) any later version.
10 
11    In addition to the permissions in the GNU Lesser General Public
12    License, the Free Software Foundation gives you unlimited
13    permission to link the compiled version of this file into
14    combinations with other programs, and to distribute those
15    combinations without any restriction coming from the use of this
16    file.  (The Lesser General Public License restrictions do apply in
17    other respects; for example, they cover modification of the file,
18    and distribution when not linked into a combine executable.)
19 
20    The GNU C Library is distributed in the hope that it will be useful,
21    but WITHOUT ANY WARRANTY; without even the implied warranty of
22    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
23    Lesser General Public License for more details.
24 
25    You should have received a copy of the GNU Lesser General Public
26    License along with the GNU C Library; if not, see
27    <http://www.gnu.org/licenses/>.  */
28 
29 /* You have to define the following before including this file:
30 
31    UWtype -- An unsigned type, default type for operations (typically a "word")
32    UHWtype -- An unsigned type, at least half the size of UWtype.
33    UDWtype -- An unsigned type, at least twice as large a UWtype
34    W_TYPE_SIZE -- size in bits of UWtype
35 
36    UQItype -- Unsigned 8 bit type.
37    SItype, USItype -- Signed and unsigned 32 bit types.
38    DItype, UDItype -- Signed and unsigned 64 bit types.
39 
40    On a 32 bit machine UWtype should typically be USItype;
41    on a 64 bit machine, UWtype should typically be UDItype.  */
42 
43 #define __BITS4 (W_TYPE_SIZE / 4)
44 #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
45 #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
46 #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
47 
48 #ifndef W_TYPE_SIZE
49 #define W_TYPE_SIZE	32
50 #define UWtype		USItype
51 #define UHWtype		USItype
52 #define UDWtype		UDItype
53 #endif
54 
55 /* Used in glibc only.  */
56 #ifndef attribute_hidden
57 #define attribute_hidden
58 #endif
59 
60 extern const UQItype __clz_tab[256] attribute_hidden;
61 
62 /* Define auxiliary asm macros.
63 
64    1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
65    UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
66    word product in HIGH_PROD and LOW_PROD.
67 
68    2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
69    UDWtype product.  This is just a variant of umul_ppmm.
70 
71    3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
72    denominator) divides a UDWtype, composed by the UWtype integers
73    HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
74    in QUOTIENT and the remainder in REMAINDER.  HIGH_NUMERATOR must be less
75    than DENOMINATOR for correct operation.  If, in addition, the most
76    significant bit of DENOMINATOR must be 1, then the pre-processor symbol
77    UDIV_NEEDS_NORMALIZATION is defined to 1.
78 
79    4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
80    denominator).  Like udiv_qrnnd but the numbers are signed.  The quotient
81    is rounded towards 0.
82 
83    5) count_leading_zeros(count, x) counts the number of zero-bits from the
84    msb to the first nonzero bit in the UWtype X.  This is the number of
85    steps X needs to be shifted left to set the msb.  Undefined for X == 0,
86    unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
87 
88    6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
89    from the least significant end.
90 
91    7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
92    high_addend_2, low_addend_2) adds two UWtype integers, composed by
93    HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
94    respectively.  The result is placed in HIGH_SUM and LOW_SUM.  Overflow
95    (i.e. carry out) is not stored anywhere, and is lost.
96 
97    8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
98    high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
99    composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
100    LOW_SUBTRAHEND_2 respectively.  The result is placed in HIGH_DIFFERENCE
101    and LOW_DIFFERENCE.  Overflow (i.e. carry out) is not stored anywhere,
102    and is lost.
103 
104    If any of these macros are left undefined for a particular CPU,
105    C macros are used.  */
106 
107 /* The CPUs come in alphabetical order below.
108 
109    Please add support for more CPUs here, or improve the current support
110    for the CPUs below!
111    (E.g. WE32100, IBM360.)  */
112 
113 #if defined (__GNUC__) && !defined (NO_ASM)
114 
115 /* We sometimes need to clobber "cc" with gcc2, but that would not be
116    understood by gcc1.  Use cpp to avoid major code duplication.  */
117 #if __GNUC__ < 2
118 #define __CLOBBER_CC
119 #define __AND_CLOBBER_CC
120 #else /* __GNUC__ >= 2 */
121 #define __CLOBBER_CC : "cc"
122 #define __AND_CLOBBER_CC , "cc"
123 #endif /* __GNUC__ < 2 */
124 
125 #if defined (__alpha) && W_TYPE_SIZE == 64
126 #define umul_ppmm(ph, pl, m0, m1) \
127   do {									\
128     UDItype __m0 = (m0), __m1 = (m1);					\
129     (ph) = __builtin_alpha_umulh (__m0, __m1);				\
130     (pl) = __m0 * __m1;							\
131   } while (0)
132 #define UMUL_TIME 46
133 #ifndef LONGLONG_STANDALONE
134 #define udiv_qrnnd(q, r, n1, n0, d) \
135   do { UDItype __r;							\
136     (q) = __udiv_qrnnd (&__r, (n1), (n0), (d));				\
137     (r) = __r;								\
138   } while (0)
139 extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
140 #define UDIV_TIME 220
141 #endif /* LONGLONG_STANDALONE */
142 #ifdef __alpha_cix__
143 #define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clzl (X))
144 #define count_trailing_zeros(COUNT,X)	((COUNT) = __builtin_ctzl (X))
145 #define COUNT_LEADING_ZEROS_0 64
146 #else
147 #define count_leading_zeros(COUNT,X) \
148   do {									\
149     UDItype __xr = (X), __t, __a;					\
150     __t = __builtin_alpha_cmpbge (0, __xr);				\
151     __a = __clz_tab[__t ^ 0xff] - 1;					\
152     __t = __builtin_alpha_extbl (__xr, __a);				\
153     (COUNT) = 64 - (__clz_tab[__t] + __a*8);				\
154   } while (0)
155 #define count_trailing_zeros(COUNT,X) \
156   do {									\
157     UDItype __xr = (X), __t, __a;					\
158     __t = __builtin_alpha_cmpbge (0, __xr);				\
159     __t = ~__t & -~__t;							\
160     __a = ((__t & 0xCC) != 0) * 2;					\
161     __a += ((__t & 0xF0) != 0) * 4;					\
162     __a += ((__t & 0xAA) != 0);						\
163     __t = __builtin_alpha_extbl (__xr, __a);				\
164     __a <<= 3;								\
165     __t &= -__t;							\
166     __a += ((__t & 0xCC) != 0) * 2;					\
167     __a += ((__t & 0xF0) != 0) * 4;					\
168     __a += ((__t & 0xAA) != 0);						\
169     (COUNT) = __a;							\
170   } while (0)
171 #endif /* __alpha_cix__ */
172 #endif /* __alpha */
173 
174 #if defined (__arc__) && W_TYPE_SIZE == 32
175 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
176   __asm__ ("add.f	%1, %4, %5\n\tadc	%0, %2, %3"		\
177 	   : "=r" ((USItype) (sh)),					\
178 	     "=&r" ((USItype) (sl))					\
179 	   : "%r" ((USItype) (ah)),					\
180 	     "rIJ" ((USItype) (bh)),					\
181 	     "%r" ((USItype) (al)),					\
182 	     "rIJ" ((USItype) (bl)))
183 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
184   __asm__ ("sub.f	%1, %4, %5\n\tsbc	%0, %2, %3"		\
185 	   : "=r" ((USItype) (sh)),					\
186 	     "=&r" ((USItype) (sl))					\
187 	   : "r" ((USItype) (ah)),					\
188 	     "rIJ" ((USItype) (bh)),					\
189 	     "r" ((USItype) (al)),					\
190 	     "rIJ" ((USItype) (bl)))
191 /* Call libgcc routine.  */
192 #define umul_ppmm(w1, w0, u, v) \
193 do {									\
194   DWunion __w;								\
195   __w.ll = __umulsidi3 (u, v);						\
196   w1 = __w.s.high;							\
197   w0 = __w.s.low;							\
198 } while (0)
199 #define __umulsidi3 __umulsidi3
200 UDItype __umulsidi3 (USItype, USItype);
201 #endif
202 
203 #if defined (__arm__) && (defined (__thumb2__) || !defined (__thumb__)) \
204  && W_TYPE_SIZE == 32
205 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
206   __asm__ ("adds	%1, %4, %5\n\tadc	%0, %2, %3"		\
207 	   : "=r" ((USItype) (sh)),					\
208 	     "=&r" ((USItype) (sl))					\
209 	   : "%r" ((USItype) (ah)),					\
210 	     "rI" ((USItype) (bh)),					\
211 	     "%r" ((USItype) (al)),					\
212 	     "rI" ((USItype) (bl)) __CLOBBER_CC)
213 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
214   __asm__ ("subs	%1, %4, %5\n\tsbc	%0, %2, %3"		\
215 	   : "=r" ((USItype) (sh)),					\
216 	     "=&r" ((USItype) (sl))					\
217 	   : "r" ((USItype) (ah)),					\
218 	     "rI" ((USItype) (bh)),					\
219 	     "r" ((USItype) (al)),					\
220 	     "rI" ((USItype) (bl)) __CLOBBER_CC)
221 # if defined(__ARM_ARCH_2__) || defined(__ARM_ARCH_2A__) \
222      || defined(__ARM_ARCH_3__)
223 #  define umul_ppmm(xh, xl, a, b)					\
224   do {									\
225     register USItype __t0, __t1, __t2;					\
226     __asm__ ("%@ Inlined umul_ppmm\n"					\
227 	   "	mov	%2, %5, lsr #16\n"				\
228 	   "	mov	%0, %6, lsr #16\n"				\
229 	   "	bic	%3, %5, %2, lsl #16\n"				\
230 	   "	bic	%4, %6, %0, lsl #16\n"				\
231 	   "	mul	%1, %3, %4\n"					\
232 	   "	mul	%4, %2, %4\n"					\
233 	   "	mul	%3, %0, %3\n"					\
234 	   "	mul	%0, %2, %0\n"					\
235 	   "	adds	%3, %4, %3\n"					\
236 	   "	addcs	%0, %0, #65536\n"				\
237 	   "	adds	%1, %1, %3, lsl #16\n"				\
238 	   "	adc	%0, %0, %3, lsr #16"				\
239 	   : "=&r" ((USItype) (xh)),					\
240 	     "=r" ((USItype) (xl)),					\
241 	     "=&r" (__t0), "=&r" (__t1), "=r" (__t2)			\
242 	   : "r" ((USItype) (a)),					\
243 	     "r" ((USItype) (b)) __CLOBBER_CC );			\
244   } while (0)
245 #  define UMUL_TIME 20
246 # else
247 #  define umul_ppmm(xh, xl, a, b)					\
248   do {									\
249     /* Generate umull, under compiler control.  */			\
250     register UDItype __t0 = (UDItype)(USItype)(a) * (USItype)(b);	\
251     (xl) = (USItype)__t0;						\
252     (xh) = (USItype)(__t0 >> 32);					\
253   } while (0)
254 #  define UMUL_TIME 3
255 # endif
256 # define UDIV_TIME 100
257 #endif /* __arm__ */
258 
259 #if defined(__arm__)
260 /* Let gcc decide how best to implement count_leading_zeros.  */
261 #define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clz (X))
262 #define count_trailing_zeros(COUNT,X)   ((COUNT) = __builtin_ctz (X))
263 #define COUNT_LEADING_ZEROS_0 32
264 #endif
265 
266 #if defined (__AVR__)
267 
268 #if W_TYPE_SIZE == 16
269 #define count_leading_zeros(COUNT,X)  ((COUNT) = __builtin_clz (X))
270 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X))
271 #define COUNT_LEADING_ZEROS_0 16
272 #endif /* W_TYPE_SIZE == 16 */
273 
274 #if W_TYPE_SIZE == 32
275 #define count_leading_zeros(COUNT,X)  ((COUNT) = __builtin_clzl (X))
276 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
277 #define COUNT_LEADING_ZEROS_0 32
278 #endif /* W_TYPE_SIZE == 32 */
279 
280 #if W_TYPE_SIZE == 64
281 #define count_leading_zeros(COUNT,X)  ((COUNT) = __builtin_clzll (X))
282 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzll (X))
283 #define COUNT_LEADING_ZEROS_0 64
284 #endif /* W_TYPE_SIZE == 64 */
285 
286 #endif /* defined (__AVR__) */
287 
288 #if defined (__CRIS__) && __CRIS_arch_version >= 3
289 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
290 #if __CRIS_arch_version >= 8
291 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
292 #endif
293 #endif /* __CRIS__ */
294 
295 #if defined (__hppa) && W_TYPE_SIZE == 32
296 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
297   __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0"				\
298 	   : "=r" ((USItype) (sh)),					\
299 	     "=&r" ((USItype) (sl))					\
300 	   : "%rM" ((USItype) (ah)),					\
301 	     "rM" ((USItype) (bh)),					\
302 	     "%rM" ((USItype) (al)),					\
303 	     "rM" ((USItype) (bl)))
304 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
305   __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0"				\
306 	   : "=r" ((USItype) (sh)),					\
307 	     "=&r" ((USItype) (sl))					\
308 	   : "rM" ((USItype) (ah)),					\
309 	     "rM" ((USItype) (bh)),					\
310 	     "rM" ((USItype) (al)),					\
311 	     "rM" ((USItype) (bl)))
312 #if defined (_PA_RISC1_1)
313 #define umul_ppmm(w1, w0, u, v) \
314   do {									\
315     union								\
316       {									\
317 	UDItype __f;							\
318 	struct {USItype __w1, __w0;} __w1w0;				\
319       } __t;								\
320     __asm__ ("xmpyu %1,%2,%0"						\
321 	     : "=x" (__t.__f)						\
322 	     : "x" ((USItype) (u)),					\
323 	       "x" ((USItype) (v)));					\
324     (w1) = __t.__w1w0.__w1;						\
325     (w0) = __t.__w1w0.__w0;						\
326      } while (0)
327 #define UMUL_TIME 8
328 #else
329 #define UMUL_TIME 30
330 #endif
331 #define UDIV_TIME 40
332 #define count_leading_zeros(count, x) \
333   do {									\
334     USItype __tmp;							\
335     __asm__ (								\
336        "ldi		1,%0\n"						\
337 "	extru,=		%1,15,16,%%r0		; Bits 31..16 zero?\n"	\
338 "	extru,tr	%1,15,16,%1		; No.  Shift down, skip add.\n"\
339 "	ldo		16(%0),%0		; Yes.  Perform add.\n"	\
340 "	extru,=		%1,23,8,%%r0		; Bits 15..8 zero?\n"	\
341 "	extru,tr	%1,23,8,%1		; No.  Shift down, skip add.\n"\
342 "	ldo		8(%0),%0		; Yes.  Perform add.\n"	\
343 "	extru,=		%1,27,4,%%r0		; Bits 7..4 zero?\n"	\
344 "	extru,tr	%1,27,4,%1		; No.  Shift down, skip add.\n"\
345 "	ldo		4(%0),%0		; Yes.  Perform add.\n"	\
346 "	extru,=		%1,29,2,%%r0		; Bits 3..2 zero?\n"	\
347 "	extru,tr	%1,29,2,%1		; No.  Shift down, skip add.\n"\
348 "	ldo		2(%0),%0		; Yes.  Perform add.\n"	\
349 "	extru		%1,30,1,%1		; Extract bit 1.\n"	\
350 "	sub		%0,%1,%0		; Subtract it.\n"	\
351 	: "=r" (count), "=r" (__tmp) : "1" (x));			\
352   } while (0)
353 #endif
354 
355 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
356 #if !defined (__zarch__)
357 #define smul_ppmm(xh, xl, m0, m1) \
358   do {									\
359     union {DItype __ll;							\
360 	   struct {USItype __h, __l;} __i;				\
361 	  } __x;							\
362     __asm__ ("lr %N0,%1\n\tmr %0,%2"					\
363 	     : "=&r" (__x.__ll)						\
364 	     : "r" (m0), "r" (m1));					\
365     (xh) = __x.__i.__h; (xl) = __x.__i.__l;				\
366   } while (0)
367 #define sdiv_qrnnd(q, r, n1, n0, d) \
368   do {									\
369     union {DItype __ll;							\
370 	   struct {USItype __h, __l;} __i;				\
371 	  } __x;							\
372     __x.__i.__h = n1; __x.__i.__l = n0;					\
373     __asm__ ("dr %0,%2"							\
374 	     : "=r" (__x.__ll)						\
375 	     : "0" (__x.__ll), "r" (d));				\
376     (q) = __x.__i.__l; (r) = __x.__i.__h;				\
377   } while (0)
378 #else
379 #define smul_ppmm(xh, xl, m0, m1) \
380   do {                                                                  \
381     register SItype __r0 __asm__ ("0");					\
382     register SItype __r1 __asm__ ("1") = (m0);				\
383 									\
384     __asm__ ("mr\t%%r0,%3"                                              \
385 	     : "=r" (__r0), "=r" (__r1)					\
386 	     : "r"  (__r1),  "r" (m1));					\
387     (xh) = __r0; (xl) = __r1;						\
388   } while (0)
389 
390 #define sdiv_qrnnd(q, r, n1, n0, d) \
391   do {									\
392     register SItype __r0 __asm__ ("0") = (n1);				\
393     register SItype __r1 __asm__ ("1") = (n0);				\
394 									\
395     __asm__ ("dr\t%%r0,%4"                                              \
396 	     : "=r" (__r0), "=r" (__r1)					\
397 	     : "r" (__r0), "r" (__r1), "r" (d));			\
398     (q) = __r1; (r) = __r0;						\
399   } while (0)
400 #endif /* __zarch__ */
401 #endif
402 
403 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
404 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
405   __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}"		\
406 	   : "=r" ((USItype) (sh)),					\
407 	     "=&r" ((USItype) (sl))					\
408 	   : "%0" ((USItype) (ah)),					\
409 	     "g" ((USItype) (bh)),					\
410 	     "%1" ((USItype) (al)),					\
411 	     "g" ((USItype) (bl)))
412 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
413   __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}"		\
414 	   : "=r" ((USItype) (sh)),					\
415 	     "=&r" ((USItype) (sl))					\
416 	   : "0" ((USItype) (ah)),					\
417 	     "g" ((USItype) (bh)),					\
418 	     "1" ((USItype) (al)),					\
419 	     "g" ((USItype) (bl)))
420 #define umul_ppmm(w1, w0, u, v) \
421   __asm__ ("mul{l} %3"							\
422 	   : "=a" ((USItype) (w0)),					\
423 	     "=d" ((USItype) (w1))					\
424 	   : "%0" ((USItype) (u)),					\
425 	     "rm" ((USItype) (v)))
426 #define udiv_qrnnd(q, r, n1, n0, dv) \
427   __asm__ ("div{l} %4"							\
428 	   : "=a" ((USItype) (q)),					\
429 	     "=d" ((USItype) (r))					\
430 	   : "0" ((USItype) (n0)),					\
431 	     "1" ((USItype) (n1)),					\
432 	     "rm" ((USItype) (dv)))
433 #define count_leading_zeros(count, x)	((count) = __builtin_clz (x))
434 #define count_trailing_zeros(count, x)	((count) = __builtin_ctz (x))
435 #define UMUL_TIME 40
436 #define UDIV_TIME 40
437 #endif /* 80x86 */
438 
439 #if (defined (__x86_64__) || defined (__i386__)) && W_TYPE_SIZE == 64
440 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
441   __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}"		\
442 	   : "=r" ((UDItype) (sh)),					\
443 	     "=&r" ((UDItype) (sl))					\
444 	   : "%0" ((UDItype) (ah)),					\
445 	     "rme" ((UDItype) (bh)),					\
446 	     "%1" ((UDItype) (al)),					\
447 	     "rme" ((UDItype) (bl)))
448 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
449   __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}"		\
450 	   : "=r" ((UDItype) (sh)),					\
451 	     "=&r" ((UDItype) (sl))					\
452 	   : "0" ((UDItype) (ah)),					\
453 	     "rme" ((UDItype) (bh)),					\
454 	     "1" ((UDItype) (al)),					\
455 	     "rme" ((UDItype) (bl)))
456 #define umul_ppmm(w1, w0, u, v) \
457   __asm__ ("mul{q} %3"							\
458 	   : "=a" ((UDItype) (w0)),					\
459 	     "=d" ((UDItype) (w1))					\
460 	   : "%0" ((UDItype) (u)),					\
461 	     "rm" ((UDItype) (v)))
462 #define udiv_qrnnd(q, r, n1, n0, dv) \
463   __asm__ ("div{q} %4"							\
464 	   : "=a" ((UDItype) (q)),					\
465 	     "=d" ((UDItype) (r))					\
466 	   : "0" ((UDItype) (n0)),					\
467 	     "1" ((UDItype) (n1)),					\
468 	     "rm" ((UDItype) (dv)))
469 #define count_leading_zeros(count, x)	((count) = __builtin_clzll (x))
470 #define count_trailing_zeros(count, x)	((count) = __builtin_ctzll (x))
471 #define UMUL_TIME 40
472 #define UDIV_TIME 40
473 #endif /* x86_64 */
474 
475 #if defined (__i960__) && W_TYPE_SIZE == 32
476 #define umul_ppmm(w1, w0, u, v) \
477   ({union {UDItype __ll;						\
478 	   struct {USItype __l, __h;} __i;				\
479 	  } __xx;							\
480   __asm__ ("emul	%2,%1,%0"					\
481 	   : "=d" (__xx.__ll)						\
482 	   : "%dI" ((USItype) (u)),					\
483 	     "dI" ((USItype) (v)));					\
484   (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
485 #define __umulsidi3(u, v) \
486   ({UDItype __w;							\
487     __asm__ ("emul	%2,%1,%0"					\
488 	     : "=d" (__w)						\
489 	     : "%dI" ((USItype) (u)),					\
490 	       "dI" ((USItype) (v)));					\
491     __w; })
492 #endif /* __i960__ */
493 
494 #if defined (__ia64) && W_TYPE_SIZE == 64
495 /* This form encourages gcc (pre-release 3.4 at least) to emit predicated
496    "sub r=r,r" and "sub r=r,r,1", giving a 2 cycle latency.  The generic
497    code using "al<bl" arithmetically comes out making an actual 0 or 1 in a
498    register, which takes an extra cycle.  */
499 #define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
500   do {									\
501     UWtype __x;								\
502     __x = (al) - (bl);							\
503     if ((al) < (bl))							\
504       (sh) = (ah) - (bh) - 1;						\
505     else								\
506       (sh) = (ah) - (bh);						\
507     (sl) = __x;								\
508   } while (0)
509 
510 /* Do both product parts in assembly, since that gives better code with
511    all gcc versions.  Some callers will just use the upper part, and in
512    that situation we waste an instruction, but not any cycles.  */
513 #define umul_ppmm(ph, pl, m0, m1)					\
514   __asm__ ("xma.hu %0 = %2, %3, f0\n\txma.l %1 = %2, %3, f0"		\
515 	   : "=&f" (ph), "=f" (pl)					\
516 	   : "f" (m0), "f" (m1))
517 #define count_leading_zeros(count, x)					\
518   do {									\
519     UWtype _x = (x), _y, _a, _c;					\
520     __asm__ ("mux1 %0 = %1, @rev" : "=r" (_y) : "r" (_x));		\
521     __asm__ ("czx1.l %0 = %1" : "=r" (_a) : "r" (-_y | _y));		\
522     _c = (_a - 1) << 3;							\
523     _x >>= _c;								\
524     if (_x >= 1 << 4)							\
525       _x >>= 4, _c += 4;						\
526     if (_x >= 1 << 2)							\
527       _x >>= 2, _c += 2;						\
528     _c += _x >> 1;							\
529     (count) =  W_TYPE_SIZE - 1 - _c;					\
530   } while (0)
531 /* similar to what gcc does for __builtin_ffs, but 0 based rather than 1
532    based, and we don't need a special case for x==0 here */
533 #define count_trailing_zeros(count, x)					\
534   do {									\
535     UWtype __ctz_x = (x);						\
536     __asm__ ("popcnt %0 = %1"						\
537 	     : "=r" (count)						\
538 	     : "r" ((__ctz_x-1) & ~__ctz_x));				\
539   } while (0)
540 #define UMUL_TIME 14
541 #endif
542 
543 #if defined (__M32R__) && W_TYPE_SIZE == 32
544 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
545   /* The cmp clears the condition bit.  */ \
546   __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3"			\
547 	   : "=r" ((USItype) (sh)),					\
548 	     "=&r" ((USItype) (sl))					\
549 	   : "0" ((USItype) (ah)),					\
550 	     "r" ((USItype) (bh)),					\
551 	     "1" ((USItype) (al)),					\
552 	     "r" ((USItype) (bl))					\
553 	   : "cbit")
554 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
555   /* The cmp clears the condition bit.  */ \
556   __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3"			\
557 	   : "=r" ((USItype) (sh)),					\
558 	     "=&r" ((USItype) (sl))					\
559 	   : "0" ((USItype) (ah)),					\
560 	     "r" ((USItype) (bh)),					\
561 	     "1" ((USItype) (al)),					\
562 	     "r" ((USItype) (bl))					\
563 	   : "cbit")
564 #endif /* __M32R__ */
565 
566 #if defined (__mc68000__) && W_TYPE_SIZE == 32
567 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
568   __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0"				\
569 	   : "=d" ((USItype) (sh)),					\
570 	     "=&d" ((USItype) (sl))					\
571 	   : "%0" ((USItype) (ah)),					\
572 	     "d" ((USItype) (bh)),					\
573 	     "%1" ((USItype) (al)),					\
574 	     "g" ((USItype) (bl)))
575 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
576   __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0"				\
577 	   : "=d" ((USItype) (sh)),					\
578 	     "=&d" ((USItype) (sl))					\
579 	   : "0" ((USItype) (ah)),					\
580 	     "d" ((USItype) (bh)),					\
581 	     "1" ((USItype) (al)),					\
582 	     "g" ((USItype) (bl)))
583 
584 /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r.  */
585 #if (defined (__mc68020__) && !defined (__mc68060__))
586 #define umul_ppmm(w1, w0, u, v) \
587   __asm__ ("mulu%.l %3,%1:%0"						\
588 	   : "=d" ((USItype) (w0)),					\
589 	     "=d" ((USItype) (w1))					\
590 	   : "%0" ((USItype) (u)),					\
591 	     "dmi" ((USItype) (v)))
592 #define UMUL_TIME 45
593 #define udiv_qrnnd(q, r, n1, n0, d) \
594   __asm__ ("divu%.l %4,%1:%0"						\
595 	   : "=d" ((USItype) (q)),					\
596 	     "=d" ((USItype) (r))					\
597 	   : "0" ((USItype) (n0)),					\
598 	     "1" ((USItype) (n1)),					\
599 	     "dmi" ((USItype) (d)))
600 #define UDIV_TIME 90
601 #define sdiv_qrnnd(q, r, n1, n0, d) \
602   __asm__ ("divs%.l %4,%1:%0"						\
603 	   : "=d" ((USItype) (q)),					\
604 	     "=d" ((USItype) (r))					\
605 	   : "0" ((USItype) (n0)),					\
606 	     "1" ((USItype) (n1)),					\
607 	     "dmi" ((USItype) (d)))
608 
609 #elif defined (__mcoldfire__) /* not mc68020 */
610 
611 #define umul_ppmm(xh, xl, a, b) \
612   __asm__ ("| Inlined umul_ppmm\n"					\
613 	   "	move%.l	%2,%/d0\n"					\
614 	   "	move%.l	%3,%/d1\n"					\
615 	   "	move%.l	%/d0,%/d2\n"					\
616 	   "	swap	%/d0\n"						\
617 	   "	move%.l	%/d1,%/d3\n"					\
618 	   "	swap	%/d1\n"						\
619 	   "	move%.w	%/d2,%/d4\n"					\
620 	   "	mulu	%/d3,%/d4\n"					\
621 	   "	mulu	%/d1,%/d2\n"					\
622 	   "	mulu	%/d0,%/d3\n"					\
623 	   "	mulu	%/d0,%/d1\n"					\
624 	   "	move%.l	%/d4,%/d0\n"					\
625 	   "	clr%.w	%/d0\n"						\
626 	   "	swap	%/d0\n"						\
627 	   "	add%.l	%/d0,%/d2\n"					\
628 	   "	add%.l	%/d3,%/d2\n"					\
629 	   "	jcc	1f\n"						\
630 	   "	add%.l	%#65536,%/d1\n"					\
631 	   "1:	swap	%/d2\n"						\
632 	   "	moveq	%#0,%/d0\n"					\
633 	   "	move%.w	%/d2,%/d0\n"					\
634 	   "	move%.w	%/d4,%/d2\n"					\
635 	   "	move%.l	%/d2,%1\n"					\
636 	   "	add%.l	%/d1,%/d0\n"					\
637 	   "	move%.l	%/d0,%0"					\
638 	   : "=g" ((USItype) (xh)),					\
639 	     "=g" ((USItype) (xl))					\
640 	   : "g" ((USItype) (a)),					\
641 	     "g" ((USItype) (b))					\
642 	   : "d0", "d1", "d2", "d3", "d4")
643 #define UMUL_TIME 100
644 #define UDIV_TIME 400
645 #else /* not ColdFire */
646 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX.  */
647 #define umul_ppmm(xh, xl, a, b) \
648   __asm__ ("| Inlined umul_ppmm\n"					\
649 	   "	move%.l	%2,%/d0\n"					\
650 	   "	move%.l	%3,%/d1\n"					\
651 	   "	move%.l	%/d0,%/d2\n"					\
652 	   "	swap	%/d0\n"						\
653 	   "	move%.l	%/d1,%/d3\n"					\
654 	   "	swap	%/d1\n"						\
655 	   "	move%.w	%/d2,%/d4\n"					\
656 	   "	mulu	%/d3,%/d4\n"					\
657 	   "	mulu	%/d1,%/d2\n"					\
658 	   "	mulu	%/d0,%/d3\n"					\
659 	   "	mulu	%/d0,%/d1\n"					\
660 	   "	move%.l	%/d4,%/d0\n"					\
661 	   "	eor%.w	%/d0,%/d0\n"					\
662 	   "	swap	%/d0\n"						\
663 	   "	add%.l	%/d0,%/d2\n"					\
664 	   "	add%.l	%/d3,%/d2\n"					\
665 	   "	jcc	1f\n"						\
666 	   "	add%.l	%#65536,%/d1\n"					\
667 	   "1:	swap	%/d2\n"						\
668 	   "	moveq	%#0,%/d0\n"					\
669 	   "	move%.w	%/d2,%/d0\n"					\
670 	   "	move%.w	%/d4,%/d2\n"					\
671 	   "	move%.l	%/d2,%1\n"					\
672 	   "	add%.l	%/d1,%/d0\n"					\
673 	   "	move%.l	%/d0,%0"					\
674 	   : "=g" ((USItype) (xh)),					\
675 	     "=g" ((USItype) (xl))					\
676 	   : "g" ((USItype) (a)),					\
677 	     "g" ((USItype) (b))					\
678 	   : "d0", "d1", "d2", "d3", "d4")
679 #define UMUL_TIME 100
680 #define UDIV_TIME 400
681 
682 #endif /* not mc68020 */
683 
684 /* The '020, '030, '040 and '060 have bitfield insns.
685    cpu32 disguises as a 68020, but lacks them.  */
686 #if defined (__mc68020__) && !defined (__mcpu32__)
687 #define count_leading_zeros(count, x) \
688   __asm__ ("bfffo %1{%b2:%b2},%0"					\
689 	   : "=d" ((USItype) (count))					\
690 	   : "od" ((USItype) (x)), "n" (0))
691 /* Some ColdFire architectures have a ff1 instruction supported via
692    __builtin_clz. */
693 #elif defined (__mcfisaaplus__) || defined (__mcfisac__)
694 #define count_leading_zeros(count,x) ((count) = __builtin_clz (x))
695 #define COUNT_LEADING_ZEROS_0 32
696 #endif
697 #endif /* mc68000 */
698 
699 #if defined (__m88000__) && W_TYPE_SIZE == 32
700 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
701   __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3"			\
702 	   : "=r" ((USItype) (sh)),					\
703 	     "=&r" ((USItype) (sl))					\
704 	   : "%rJ" ((USItype) (ah)),					\
705 	     "rJ" ((USItype) (bh)),					\
706 	     "%rJ" ((USItype) (al)),					\
707 	     "rJ" ((USItype) (bl)))
708 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
709   __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3"			\
710 	   : "=r" ((USItype) (sh)),					\
711 	     "=&r" ((USItype) (sl))					\
712 	   : "rJ" ((USItype) (ah)),					\
713 	     "rJ" ((USItype) (bh)),					\
714 	     "rJ" ((USItype) (al)),					\
715 	     "rJ" ((USItype) (bl)))
716 #define count_leading_zeros(count, x) \
717   do {									\
718     USItype __cbtmp;							\
719     __asm__ ("ff1 %0,%1"						\
720 	     : "=r" (__cbtmp)						\
721 	     : "r" ((USItype) (x)));					\
722     (count) = __cbtmp ^ 31;						\
723   } while (0)
724 #define COUNT_LEADING_ZEROS_0 63 /* sic */
725 #if defined (__mc88110__)
726 #define umul_ppmm(wh, wl, u, v) \
727   do {									\
728     union {UDItype __ll;						\
729 	   struct {USItype __h, __l;} __i;				\
730 	  } __xx;							\
731     __asm__ ("mulu.d	%0,%1,%2"					\
732 	     : "=r" (__xx.__ll)						\
733 	     : "r" ((USItype) (u)),					\
734 	       "r" ((USItype) (v)));					\
735     (wh) = __xx.__i.__h;						\
736     (wl) = __xx.__i.__l;						\
737   } while (0)
738 #define udiv_qrnnd(q, r, n1, n0, d) \
739   ({union {UDItype __ll;						\
740 	   struct {USItype __h, __l;} __i;				\
741 	  } __xx;							\
742   USItype __q;								\
743   __xx.__i.__h = (n1); __xx.__i.__l = (n0);				\
744   __asm__ ("divu.d %0,%1,%2"						\
745 	   : "=r" (__q)							\
746 	   : "r" (__xx.__ll),						\
747 	     "r" ((USItype) (d)));					\
748   (r) = (n0) - __q * (d); (q) = __q; })
749 #define UMUL_TIME 5
750 #define UDIV_TIME 25
751 #else
752 #define UMUL_TIME 17
753 #define UDIV_TIME 150
754 #endif /* __mc88110__ */
755 #endif /* __m88000__ */
756 
757 #if defined (__mn10300__)
758 # if defined (__AM33__)
759 #  define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clz (X))
760 #  define umul_ppmm(w1, w0, u, v)		\
761     asm("mulu %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
762 #  define smul_ppmm(w1, w0, u, v)		\
763     asm("mul %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
764 # else
765 #  define umul_ppmm(w1, w0, u, v)		\
766     asm("nop; nop; mulu %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
767 #  define smul_ppmm(w1, w0, u, v)		\
768     asm("nop; nop; mul %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
769 # endif
770 # define add_ssaaaa(sh, sl, ah, al, bh, bl)	\
771   do {						\
772     DWunion __s, __a, __b;			\
773     __a.s.low = (al); __a.s.high = (ah);	\
774     __b.s.low = (bl); __b.s.high = (bh);	\
775     __s.ll = __a.ll + __b.ll;			\
776     (sl) = __s.s.low; (sh) = __s.s.high;	\
777   } while (0)
778 # define sub_ddmmss(sh, sl, ah, al, bh, bl)	\
779   do {						\
780     DWunion __s, __a, __b;			\
781     __a.s.low = (al); __a.s.high = (ah);	\
782     __b.s.low = (bl); __b.s.high = (bh);	\
783     __s.ll = __a.ll - __b.ll;			\
784     (sl) = __s.s.low; (sh) = __s.s.high;	\
785   } while (0)
786 # define udiv_qrnnd(q, r, nh, nl, d)		\
787   asm("divu %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
788 # define sdiv_qrnnd(q, r, nh, nl, d)		\
789   asm("div %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
790 # define UMUL_TIME 3
791 # define UDIV_TIME 38
792 #endif
793 
794 #if defined (__mips__) && W_TYPE_SIZE == 32
795 #define umul_ppmm(w1, w0, u, v)						\
796   do {									\
797     UDItype __x = (UDItype) (USItype) (u) * (USItype) (v);		\
798     (w1) = (USItype) (__x >> 32);					\
799     (w0) = (USItype) (__x);						\
800   } while (0)
801 #define UMUL_TIME 10
802 #define UDIV_TIME 100
803 
804 #if (__mips == 32 || __mips == 64) && ! __mips16
805 #define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clz (X))
806 #define COUNT_LEADING_ZEROS_0 32
807 #endif
808 #endif /* __mips__ */
809 
810 #if defined (__ns32000__) && W_TYPE_SIZE == 32
811 #define umul_ppmm(w1, w0, u, v) \
812   ({union {UDItype __ll;						\
813 	   struct {USItype __l, __h;} __i;				\
814 	  } __xx;							\
815   __asm__ ("meid %2,%0"							\
816 	   : "=g" (__xx.__ll)						\
817 	   : "%0" ((USItype) (u)),					\
818 	     "g" ((USItype) (v)));					\
819   (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
820 #define __umulsidi3(u, v) \
821   ({UDItype __w;							\
822     __asm__ ("meid %2,%0"						\
823 	     : "=g" (__w)						\
824 	     : "%0" ((USItype) (u)),					\
825 	       "g" ((USItype) (v)));					\
826     __w; })
827 #define udiv_qrnnd(q, r, n1, n0, d) \
828   ({union {UDItype __ll;						\
829 	   struct {USItype __l, __h;} __i;				\
830 	  } __xx;							\
831   __xx.__i.__h = (n1); __xx.__i.__l = (n0);				\
832   __asm__ ("deid %2,%0"							\
833 	   : "=g" (__xx.__ll)						\
834 	   : "0" (__xx.__ll),						\
835 	     "g" ((USItype) (d)));					\
836   (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
837 #define count_trailing_zeros(count,x) \
838   do {									\
839     __asm__ ("ffsd     %2,%0"						\
840 	    : "=r" ((USItype) (count))					\
841 	    : "0" ((USItype) 0),					\
842 	      "r" ((USItype) (x)));					\
843   } while (0)
844 #endif /* __ns32000__ */
845 
846 /* FIXME: We should test _IBMR2 here when we add assembly support for the
847    system vendor compilers.
848    FIXME: What's needed for gcc PowerPC VxWorks?  __vxworks__ is not good
849    enough, since that hits ARM and m68k too.  */
850 #if (defined (_ARCH_PPC)	/* AIX */				\
851      || defined (__powerpc__)	/* gcc */				\
852      || defined (__POWERPC__)	/* BEOS */				\
853      || defined (__ppc__)	/* Darwin */				\
854      || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */    \
855      || (defined (PPC) && defined (CPU_FAMILY)    /* VxWorks */               \
856 	 && CPU_FAMILY == PPC)                                                \
857      ) && W_TYPE_SIZE == 32
858 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
859   do {									\
860     if (__builtin_constant_p (bh) && (bh) == 0)				\
861       __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2"		\
862 	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
863     else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0)		\
864       __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2"		\
865 	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
866     else								\
867       __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3"		\
868 	     : "=r" (sh), "=&r" (sl)					\
869 	     : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl));		\
870   } while (0)
871 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
872   do {									\
873     if (__builtin_constant_p (ah) && (ah) == 0)				\
874       __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2"	\
875 	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
876     else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0)		\
877       __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2"	\
878 	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
879     else if (__builtin_constant_p (bh) && (bh) == 0)			\
880       __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2"		\
881 	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
882     else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0)		\
883       __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2"		\
884 	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
885     else								\
886       __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2"	\
887 	       : "=r" (sh), "=&r" (sl)					\
888 	       : "r" (ah), "r" (bh), "rI" (al), "r" (bl));		\
889   } while (0)
890 #define count_leading_zeros(count, x) \
891   __asm__ ("cntlzw %0,%1" : "=r" (count) : "r" (x))
892 #define COUNT_LEADING_ZEROS_0 32
893 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
894   || defined (__ppc__)                                                    \
895   || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */       \
896   || (defined (PPC) && defined (CPU_FAMILY)    /* VxWorks */                  \
897 	 && CPU_FAMILY == PPC)
898 #define umul_ppmm(ph, pl, m0, m1) \
899   do {									\
900     USItype __m0 = (m0), __m1 = (m1);					\
901     __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
902     (pl) = __m0 * __m1;							\
903   } while (0)
904 #define UMUL_TIME 15
905 #define smul_ppmm(ph, pl, m0, m1) \
906   do {									\
907     SItype __m0 = (m0), __m1 = (m1);					\
908     __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
909     (pl) = __m0 * __m1;							\
910   } while (0)
911 #define SMUL_TIME 14
912 #define UDIV_TIME 120
913 #endif
914 #endif /* 32-bit POWER architecture variants.  */
915 
916 /* We should test _IBMR2 here when we add assembly support for the system
917    vendor compilers.  */
918 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
919 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
920   do {									\
921     if (__builtin_constant_p (bh) && (bh) == 0)				\
922       __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2"		\
923 	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
924     else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0)		\
925       __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2"		\
926 	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
927     else								\
928       __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3"		\
929 	     : "=r" (sh), "=&r" (sl)					\
930 	     : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl));		\
931   } while (0)
932 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
933   do {									\
934     if (__builtin_constant_p (ah) && (ah) == 0)				\
935       __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2"	\
936 	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
937     else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0)		\
938       __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2"	\
939 	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
940     else if (__builtin_constant_p (bh) && (bh) == 0)			\
941       __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2"		\
942 	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
943     else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0)		\
944       __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2"		\
945 	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
946     else								\
947       __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2"	\
948 	       : "=r" (sh), "=&r" (sl)					\
949 	       : "r" (ah), "r" (bh), "rI" (al), "r" (bl));		\
950   } while (0)
951 #define count_leading_zeros(count, x) \
952   __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
953 #define COUNT_LEADING_ZEROS_0 64
954 #define umul_ppmm(ph, pl, m0, m1) \
955   do {									\
956     UDItype __m0 = (m0), __m1 = (m1);					\
957     __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
958     (pl) = __m0 * __m1;							\
959   } while (0)
960 #define UMUL_TIME 15
961 #define smul_ppmm(ph, pl, m0, m1) \
962   do {									\
963     DItype __m0 = (m0), __m1 = (m1);					\
964     __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
965     (pl) = __m0 * __m1;							\
966   } while (0)
967 #define SMUL_TIME 14  /* ??? */
968 #define UDIV_TIME 120 /* ??? */
969 #endif /* 64-bit PowerPC.  */
970 
971 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
972 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
973   __asm__ ("a %1,%5\n\tae %0,%3"					\
974 	   : "=r" ((USItype) (sh)),					\
975 	     "=&r" ((USItype) (sl))					\
976 	   : "%0" ((USItype) (ah)),					\
977 	     "r" ((USItype) (bh)),					\
978 	     "%1" ((USItype) (al)),					\
979 	     "r" ((USItype) (bl)))
980 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
981   __asm__ ("s %1,%5\n\tse %0,%3"					\
982 	   : "=r" ((USItype) (sh)),					\
983 	     "=&r" ((USItype) (sl))					\
984 	   : "0" ((USItype) (ah)),					\
985 	     "r" ((USItype) (bh)),					\
986 	     "1" ((USItype) (al)),					\
987 	     "r" ((USItype) (bl)))
988 #define umul_ppmm(ph, pl, m0, m1) \
989   do {									\
990     USItype __m0 = (m0), __m1 = (m1);					\
991     __asm__ (								\
992        "s	r2,r2\n"						\
993 "	mts	r10,%2\n"						\
994 "	m	r2,%3\n"						\
995 "	m	r2,%3\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 "	cas	%0,r2,r0\n"						\
1011 "	mfs	r10,%1"							\
1012 	     : "=r" ((USItype) (ph)),					\
1013 	       "=r" ((USItype) (pl))					\
1014 	     : "%r" (__m0),						\
1015 		"r" (__m1)						\
1016 	     : "r2");							\
1017     (ph) += ((((SItype) __m0 >> 31) & __m1)				\
1018 	     + (((SItype) __m1 >> 31) & __m0));				\
1019   } while (0)
1020 #define UMUL_TIME 20
1021 #define UDIV_TIME 200
1022 #define count_leading_zeros(count, x) \
1023   do {									\
1024     if ((x) >= 0x10000)							\
1025       __asm__ ("clz	%0,%1"						\
1026 	       : "=r" ((USItype) (count))				\
1027 	       : "r" ((USItype) (x) >> 16));				\
1028     else								\
1029       {									\
1030 	__asm__ ("clz	%0,%1"						\
1031 		 : "=r" ((USItype) (count))				\
1032 		 : "r" ((USItype) (x)));					\
1033 	(count) += 16;							\
1034       }									\
1035   } while (0)
1036 #endif
1037 
1038 #if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
1039 #ifndef __sh1__
1040 #define umul_ppmm(w1, w0, u, v) \
1041   __asm__ (								\
1042        "dmulu.l	%2,%3\n\tsts%M1	macl,%1\n\tsts%M0	mach,%0"	\
1043 	   : "=r<" ((USItype)(w1)),					\
1044 	     "=r<" ((USItype)(w0))					\
1045 	   : "r" ((USItype)(u)),					\
1046 	     "r" ((USItype)(v))						\
1047 	   : "macl", "mach")
1048 #define UMUL_TIME 5
1049 #endif
1050 
1051 /* This is the same algorithm as __udiv_qrnnd_c.  */
1052 #define UDIV_NEEDS_NORMALIZATION 1
1053 
1054 #define udiv_qrnnd(q, r, n1, n0, d) \
1055   do {									\
1056     extern UWtype __udiv_qrnnd_16 (UWtype, UWtype)			\
1057 			__attribute__ ((visibility ("hidden")));	\
1058     /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */	\
1059     __asm__ (								\
1060 	"mov%M4 %4,r5\n"						\
1061 "	swap.w %3,r4\n"							\
1062 "	swap.w r5,r6\n"							\
1063 "	jsr @%5\n"							\
1064 "	shll16 r6\n"							\
1065 "	swap.w r4,r4\n"							\
1066 "	jsr @%5\n"							\
1067 "	swap.w r1,%0\n"							\
1068 "	or r1,%0"							\
1069 	: "=r" (q), "=&z" (r)						\
1070 	: "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16)		\
1071 	: "r1", "r2", "r4", "r5", "r6", "pr", "t");			\
1072   } while (0)
1073 
1074 #define UDIV_TIME 80
1075 
1076 #define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
1077   __asm__ ("clrt;subc %5,%1; subc %4,%0"				\
1078 	   : "=r" (sh), "=r" (sl)					\
1079 	   : "0" (ah), "1" (al), "r" (bh), "r" (bl) : "t")
1080 
1081 #endif /* __sh__ */
1082 
1083 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
1084 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1085 #define count_leading_zeros(count, x) \
1086   do									\
1087     {									\
1088       UDItype x_ = (USItype)(x);					\
1089       SItype c_;							\
1090 									\
1091       __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_));			\
1092       (count) = c_ - 31;						\
1093     }									\
1094   while (0)
1095 #define COUNT_LEADING_ZEROS_0 32
1096 #endif
1097 
1098 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
1099     && W_TYPE_SIZE == 32
1100 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1101   __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0"				\
1102 	   : "=r" ((USItype) (sh)),					\
1103 	     "=&r" ((USItype) (sl))					\
1104 	   : "%rJ" ((USItype) (ah)),					\
1105 	     "rI" ((USItype) (bh)),					\
1106 	     "%rJ" ((USItype) (al)),					\
1107 	     "rI" ((USItype) (bl))					\
1108 	   __CLOBBER_CC)
1109 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1110   __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0"				\
1111 	   : "=r" ((USItype) (sh)),					\
1112 	     "=&r" ((USItype) (sl))					\
1113 	   : "rJ" ((USItype) (ah)),					\
1114 	     "rI" ((USItype) (bh)),					\
1115 	     "rJ" ((USItype) (al)),					\
1116 	     "rI" ((USItype) (bl))					\
1117 	   __CLOBBER_CC)
1118 #if defined (__sparc_v9__)
1119 #define umul_ppmm(w1, w0, u, v) \
1120   do {									\
1121     register USItype __g1 asm ("g1");					\
1122     __asm__ ("umul\t%2,%3,%1\n\t"					\
1123 	     "srlx\t%1, 32, %0"						\
1124 	     : "=r" ((USItype) (w1)),					\
1125 	       "=r" (__g1)						\
1126 	     : "r" ((USItype) (u)),					\
1127 	       "r" ((USItype) (v)));					\
1128     (w0) = __g1;							\
1129   } while (0)
1130 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1131   __asm__ ("mov\t%2,%%y\n\t"						\
1132 	   "udiv\t%3,%4,%0\n\t"						\
1133 	   "umul\t%0,%4,%1\n\t"						\
1134 	   "sub\t%3,%1,%1"						\
1135 	   : "=&r" ((USItype) (__q)),					\
1136 	     "=&r" ((USItype) (__r))					\
1137 	   : "r" ((USItype) (__n1)),					\
1138 	     "r" ((USItype) (__n0)),					\
1139 	     "r" ((USItype) (__d)))
1140 #else
1141 #if defined (__sparc_v8__)
1142 #define umul_ppmm(w1, w0, u, v) \
1143   __asm__ ("umul %2,%3,%1;rd %%y,%0"					\
1144 	   : "=r" ((USItype) (w1)),					\
1145 	     "=r" ((USItype) (w0))					\
1146 	   : "r" ((USItype) (u)),					\
1147 	     "r" ((USItype) (v)))
1148 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1149   __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
1150 	   : "=&r" ((USItype) (__q)),					\
1151 	     "=&r" ((USItype) (__r))					\
1152 	   : "r" ((USItype) (__n1)),					\
1153 	     "r" ((USItype) (__n0)),					\
1154 	     "r" ((USItype) (__d)))
1155 #else
1156 #if defined (__sparclite__)
1157 /* This has hardware multiply but not divide.  It also has two additional
1158    instructions scan (ffs from high bit) and divscc.  */
1159 #define umul_ppmm(w1, w0, u, v) \
1160   __asm__ ("umul %2,%3,%1;rd %%y,%0"					\
1161 	   : "=r" ((USItype) (w1)),					\
1162 	     "=r" ((USItype) (w0))					\
1163 	   : "r" ((USItype) (u)),					\
1164 	     "r" ((USItype) (v)))
1165 #define udiv_qrnnd(q, r, n1, n0, d) \
1166   __asm__ ("! Inlined udiv_qrnnd\n"					\
1167 "	wr	%%g0,%2,%%y	! Not a delayed write for sparclite\n"	\
1168 "	tst	%%g0\n"							\
1169 "	divscc	%3,%4,%%g1\n"						\
1170 "	divscc	%%g1,%4,%%g1\n"						\
1171 "	divscc	%%g1,%4,%%g1\n"						\
1172 "	divscc	%%g1,%4,%%g1\n"						\
1173 "	divscc	%%g1,%4,%%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,%0\n"						\
1201 "	rd	%%y,%1\n"						\
1202 "	bl,a 1f\n"							\
1203 "	add	%1,%4,%1\n"						\
1204 "1:	! End of inline udiv_qrnnd"					\
1205 	   : "=r" ((USItype) (q)),					\
1206 	     "=r" ((USItype) (r))					\
1207 	   : "r" ((USItype) (n1)),					\
1208 	     "r" ((USItype) (n0)),					\
1209 	     "rI" ((USItype) (d))					\
1210 	   : "g1" __AND_CLOBBER_CC)
1211 #define UDIV_TIME 37
1212 #define count_leading_zeros(count, x) \
1213   do {                                                                  \
1214   __asm__ ("scan %1,1,%0"                                               \
1215 	   : "=r" ((USItype) (count))                                   \
1216 	   : "r" ((USItype) (x)));					\
1217   } while (0)
1218 /* Early sparclites return 63 for an argument of 0, but they warn that future
1219    implementations might change this.  Therefore, leave COUNT_LEADING_ZEROS_0
1220    undefined.  */
1221 #else
1222 /* SPARC without integer multiplication and divide instructions.
1223    (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
1224 #define umul_ppmm(w1, w0, u, v) \
1225   __asm__ ("! Inlined umul_ppmm\n"					\
1226 "	wr	%%g0,%2,%%y	! SPARC has 0-3 delay insn after a wr\n"\
1227 "	sra	%3,31,%%o5	! Don't move this insn\n"		\
1228 "	and	%2,%%o5,%%o5	! Don't move this insn\n"		\
1229 "	andcc	%%g0,0,%%g1	! Don't move this insn\n"		\
1230 "	mulscc	%%g1,%3,%%g1\n"						\
1231 "	mulscc	%%g1,%3,%%g1\n"						\
1232 "	mulscc	%%g1,%3,%%g1\n"						\
1233 "	mulscc	%%g1,%3,%%g1\n"						\
1234 "	mulscc	%%g1,%3,%%g1\n"						\
1235 "	mulscc	%%g1,%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	%%g1,0,%%g1\n"						\
1263 "	add	%%g1,%%o5,%0\n"						\
1264 "	rd	%%y,%1"							\
1265 	   : "=r" ((USItype) (w1)),					\
1266 	     "=r" ((USItype) (w0))					\
1267 	   : "%rI" ((USItype) (u)),					\
1268 	     "r" ((USItype) (v))						\
1269 	   : "g1", "o5" __AND_CLOBBER_CC)
1270 #define UMUL_TIME 39		/* 39 instructions */
1271 /* It's quite necessary to add this much assembler for the sparc.
1272    The default udiv_qrnnd (in C) is more than 10 times slower!  */
1273 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1274   __asm__ ("! Inlined udiv_qrnnd\n"					\
1275 "	mov	32,%%g1\n"						\
1276 "	subcc	%1,%2,%%g0\n"						\
1277 "1:	bcs	5f\n"							\
1278 "	 addxcc %0,%0,%0	! shift n1n0 and a q-bit in lsb\n"	\
1279 "	sub	%1,%2,%1	! this kills msb of n\n"		\
1280 "	addx	%1,%1,%1	! so this can't give carry\n"		\
1281 "	subcc	%%g1,1,%%g1\n"						\
1282 "2:	bne	1b\n"							\
1283 "	 subcc	%1,%2,%%g0\n"						\
1284 "	bcs	3f\n"							\
1285 "	 addxcc %0,%0,%0	! shift n1n0 and a q-bit in lsb\n"	\
1286 "	b	3f\n"							\
1287 "	 sub	%1,%2,%1	! this kills msb of n\n"		\
1288 "4:	sub	%1,%2,%1\n"						\
1289 "5:	addxcc	%1,%1,%1\n"						\
1290 "	bcc	2b\n"							\
1291 "	 subcc	%%g1,1,%%g1\n"						\
1292 "! Got carry from n.  Subtract next step to cancel this carry.\n"	\
1293 "	bne	4b\n"							\
1294 "	 addcc	%0,%0,%0	! shift n1n0 and a 0-bit in lsb\n"	\
1295 "	sub	%1,%2,%1\n"						\
1296 "3:	xnor	%0,0,%0\n"						\
1297 "	! End of inline udiv_qrnnd"					\
1298 	   : "=&r" ((USItype) (__q)),					\
1299 	     "=&r" ((USItype) (__r))					\
1300 	   : "r" ((USItype) (__d)),					\
1301 	     "1" ((USItype) (__n1)),					\
1302 	     "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1303 #define UDIV_TIME (3+7*32)	/* 7 instructions/iteration. 32 iterations.  */
1304 #endif /* __sparclite__ */
1305 #endif /* __sparc_v8__ */
1306 #endif /* __sparc_v9__ */
1307 #endif /* sparc32 */
1308 
1309 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1310     && W_TYPE_SIZE == 64
1311 #define add_ssaaaa(sh, sl, ah, al, bh, bl)				\
1312   do {									\
1313     UDItype __carry = 0;						\
1314     __asm__ ("addcc\t%r5,%6,%1\n\t"					\
1315 	     "add\t%r3,%4,%0\n\t"					\
1316 	     "movcs\t%%xcc, 1, %2\n\t"					\
1317 	     "add\t%0, %2, %0"						\
1318 	     : "=r" ((UDItype)(sh)),				      	\
1319 	       "=&r" ((UDItype)(sl)),				      	\
1320 	       "+r" (__carry)				      		\
1321 	     : "%rJ" ((UDItype)(ah)),				     	\
1322 	       "rI" ((UDItype)(bh)),				      	\
1323 	       "%rJ" ((UDItype)(al)),				     	\
1324 	       "rI" ((UDItype)(bl))				       	\
1325 	     __CLOBBER_CC);						\
1326   } while (0)
1327 
1328 #define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
1329   do {									\
1330     UDItype __carry = 0;						\
1331     __asm__ ("subcc\t%r5,%6,%1\n\t"					\
1332 	     "sub\t%r3,%4,%0\n\t"					\
1333 	     "movcs\t%%xcc, 1, %2\n\t"					\
1334 	     "sub\t%0, %2, %0"						\
1335 	     : "=r" ((UDItype)(sh)),				      	\
1336 	       "=&r" ((UDItype)(sl)),				      	\
1337 	       "+r" (__carry)				      		\
1338 	     : "%rJ" ((UDItype)(ah)),				     	\
1339 	       "rI" ((UDItype)(bh)),				      	\
1340 	       "%rJ" ((UDItype)(al)),				     	\
1341 	       "rI" ((UDItype)(bl))				       	\
1342 	     __CLOBBER_CC);						\
1343   } while (0)
1344 
1345 #define umul_ppmm(wh, wl, u, v)						\
1346   do {									\
1347 	  UDItype tmp1, tmp2, tmp3, tmp4;				\
1348 	  __asm__ __volatile__ (					\
1349 		   "srl %7,0,%3\n\t"					\
1350 		   "mulx %3,%6,%1\n\t"					\
1351 		   "srlx %6,32,%2\n\t"					\
1352 		   "mulx %2,%3,%4\n\t"					\
1353 		   "sllx %4,32,%5\n\t"					\
1354 		   "srl %6,0,%3\n\t"					\
1355 		   "sub %1,%5,%5\n\t"					\
1356 		   "srlx %5,32,%5\n\t"					\
1357 		   "addcc %4,%5,%4\n\t"					\
1358 		   "srlx %7,32,%5\n\t"					\
1359 		   "mulx %3,%5,%3\n\t"					\
1360 		   "mulx %2,%5,%5\n\t"					\
1361 		   "sethi %%hi(0x80000000),%2\n\t"			\
1362 		   "addcc %4,%3,%4\n\t"					\
1363 		   "srlx %4,32,%4\n\t"					\
1364 		   "add %2,%2,%2\n\t"					\
1365 		   "movcc %%xcc,%%g0,%2\n\t"				\
1366 		   "addcc %5,%4,%5\n\t"					\
1367 		   "sllx %3,32,%3\n\t"					\
1368 		   "add %1,%3,%1\n\t"					\
1369 		   "add %5,%2,%0"					\
1370 	   : "=r" ((UDItype)(wh)),					\
1371 	     "=&r" ((UDItype)(wl)),					\
1372 	     "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4)	\
1373 	   : "r" ((UDItype)(u)),					\
1374 	     "r" ((UDItype)(v))						\
1375 	   __CLOBBER_CC);						\
1376   } while (0)
1377 #define UMUL_TIME 96
1378 #define UDIV_TIME 230
1379 #endif /* sparc64 */
1380 
1381 #if defined (__vax__) && W_TYPE_SIZE == 32
1382 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1383   __asm__ ("addl2 %5,%1\n\tadwc %3,%0"					\
1384 	   : "=g" ((USItype) (sh)),					\
1385 	     "=&g" ((USItype) (sl))					\
1386 	   : "%0" ((USItype) (ah)),					\
1387 	     "g" ((USItype) (bh)),					\
1388 	     "%1" ((USItype) (al)),					\
1389 	     "g" ((USItype) (bl)))
1390 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1391   __asm__ ("subl2 %5,%1\n\tsbwc %3,%0"					\
1392 	   : "=g" ((USItype) (sh)),					\
1393 	     "=&g" ((USItype) (sl))					\
1394 	   : "0" ((USItype) (ah)),					\
1395 	     "g" ((USItype) (bh)),					\
1396 	     "1" ((USItype) (al)),					\
1397 	     "g" ((USItype) (bl)))
1398 #define umul_ppmm(xh, xl, m0, m1) \
1399   do {									\
1400     union {								\
1401 	UDItype __ll;							\
1402 	struct {USItype __l, __h;} __i;					\
1403       } __xx;								\
1404     USItype __m0 = (m0), __m1 = (m1);					\
1405     __asm__ ("emul %1,%2,$0,%0"						\
1406 	     : "=r" (__xx.__ll)						\
1407 	     : "g" (__m0),						\
1408 	       "g" (__m1));						\
1409     (xh) = __xx.__i.__h;						\
1410     (xl) = __xx.__i.__l;						\
1411     (xh) += ((((SItype) __m0 >> 31) & __m1)				\
1412 	     + (((SItype) __m1 >> 31) & __m0));				\
1413   } while (0)
1414 #define sdiv_qrnnd(q, r, n1, n0, d) \
1415   do {									\
1416     union {DItype __ll;							\
1417 	   struct {SItype __l, __h;} __i;				\
1418 	  } __xx;							\
1419     __xx.__i.__h = n1; __xx.__i.__l = n0;				\
1420     __asm__ ("ediv %3,%2,%0,%1"						\
1421 	     : "=g" (q), "=g" (r)					\
1422 	     : "g" (__xx.__ll), "g" (d));				\
1423   } while (0)
1424 #endif /* __vax__ */
1425 
1426 #ifdef _TMS320C6X
1427 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1428   do									\
1429     {									\
1430       UDItype __ll;							\
1431       __asm__ ("addu .l1 %1, %2, %0"					\
1432 	       : "=a" (__ll) : "a" (al), "a" (bl));			\
1433       (sl) = (USItype)__ll;						\
1434       (sh) = ((USItype)(__ll >> 32)) + (ah) + (bh);			\
1435     }									\
1436   while (0)
1437 
1438 #ifdef _TMS320C6400_PLUS
1439 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1440 #define umul_ppmm(w1, w0, u, v)						\
1441   do {									\
1442     UDItype __x = (UDItype) (USItype) (u) * (USItype) (v);		\
1443     (w1) = (USItype) (__x >> 32);					\
1444     (w0) = (USItype) (__x);						\
1445   } while (0)
1446 #endif  /* _TMS320C6400_PLUS */
1447 
1448 #define count_leading_zeros(count, x)	((count) = __builtin_clz (x))
1449 #ifdef _TMS320C6400
1450 #define count_trailing_zeros(count, x)	((count) = __builtin_ctz (x))
1451 #endif
1452 #define UMUL_TIME 4
1453 #define UDIV_TIME 40
1454 #endif /* _TMS320C6X */
1455 
1456 #if defined (__xtensa__) && W_TYPE_SIZE == 32
1457 /* This code is not Xtensa-configuration-specific, so rely on the compiler
1458    to expand builtin functions depending on what configuration features
1459    are available.  This avoids library calls when the operation can be
1460    performed in-line.  */
1461 #define umul_ppmm(w1, w0, u, v)						\
1462   do {									\
1463     DWunion __w;							\
1464     __w.ll = __builtin_umulsidi3 (u, v);				\
1465     w1 = __w.s.high;							\
1466     w0 = __w.s.low;							\
1467   } while (0)
1468 #define __umulsidi3(u, v)		__builtin_umulsidi3 (u, v)
1469 #define count_leading_zeros(COUNT, X)	((COUNT) = __builtin_clz (X))
1470 #define count_trailing_zeros(COUNT, X)	((COUNT) = __builtin_ctz (X))
1471 #endif /* __xtensa__ */
1472 
1473 #if defined xstormy16
1474 extern UHItype __stormy16_count_leading_zeros (UHItype);
1475 #define count_leading_zeros(count, x)					\
1476   do									\
1477     {									\
1478       UHItype size;							\
1479 									\
1480       /* We assume that W_TYPE_SIZE is a multiple of 16...  */		\
1481       for ((count) = 0, size = W_TYPE_SIZE; size; size -= 16)		\
1482 	{								\
1483 	  UHItype c;							\
1484 									\
1485 	  c = __clzhi2 ((x) >> (size - 16));				\
1486 	  (count) += c;							\
1487 	  if (c != 16)							\
1488 	    break;							\
1489 	}								\
1490     }									\
1491   while (0)
1492 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1493 #endif
1494 
1495 #if defined (__z8000__) && W_TYPE_SIZE == 16
1496 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1497   __asm__ ("add	%H1,%H5\n\tadc	%H0,%H3"				\
1498 	   : "=r" ((unsigned int)(sh)),					\
1499 	     "=&r" ((unsigned int)(sl))					\
1500 	   : "%0" ((unsigned int)(ah)),					\
1501 	     "r" ((unsigned int)(bh)),					\
1502 	     "%1" ((unsigned int)(al)),					\
1503 	     "rQR" ((unsigned int)(bl)))
1504 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1505   __asm__ ("sub	%H1,%H5\n\tsbc	%H0,%H3"				\
1506 	   : "=r" ((unsigned int)(sh)),					\
1507 	     "=&r" ((unsigned int)(sl))					\
1508 	   : "0" ((unsigned int)(ah)),					\
1509 	     "r" ((unsigned int)(bh)),					\
1510 	     "1" ((unsigned int)(al)),					\
1511 	     "rQR" ((unsigned int)(bl)))
1512 #define umul_ppmm(xh, xl, m0, m1) \
1513   do {									\
1514     union {long int __ll;						\
1515 	   struct {unsigned int __h, __l;} __i;				\
1516 	  } __xx;							\
1517     unsigned int __m0 = (m0), __m1 = (m1);				\
1518     __asm__ ("mult	%S0,%H3"					\
1519 	     : "=r" (__xx.__i.__h),					\
1520 	       "=r" (__xx.__i.__l)					\
1521 	     : "%1" (__m0),						\
1522 	       "rQR" (__m1));						\
1523     (xh) = __xx.__i.__h; (xl) = __xx.__i.__l;				\
1524     (xh) += ((((signed int) __m0 >> 15) & __m1)				\
1525 	     + (((signed int) __m1 >> 15) & __m0));			\
1526   } while (0)
1527 #endif /* __z8000__ */
1528 
1529 #endif /* __GNUC__ */
1530 
1531 /* If this machine has no inline assembler, use C macros.  */
1532 
1533 #if !defined (add_ssaaaa)
1534 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1535   do {									\
1536     UWtype __x;								\
1537     __x = (al) + (bl);							\
1538     (sh) = (ah) + (bh) + (__x < (al));					\
1539     (sl) = __x;								\
1540   } while (0)
1541 #endif
1542 
1543 #if !defined (sub_ddmmss)
1544 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1545   do {									\
1546     UWtype __x;								\
1547     __x = (al) - (bl);							\
1548     (sh) = (ah) - (bh) - (__x > (al));					\
1549     (sl) = __x;								\
1550   } while (0)
1551 #endif
1552 
1553 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1554    smul_ppmm.  */
1555 #if !defined (umul_ppmm) && defined (smul_ppmm)
1556 #define umul_ppmm(w1, w0, u, v)						\
1557   do {									\
1558     UWtype __w1;							\
1559     UWtype __xm0 = (u), __xm1 = (v);					\
1560     smul_ppmm (__w1, w0, __xm0, __xm1);					\
1561     (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1)		\
1562 		+ (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0);		\
1563   } while (0)
1564 #endif
1565 
1566 /* If we still don't have umul_ppmm, define it using plain C.  */
1567 #if !defined (umul_ppmm)
1568 #define umul_ppmm(w1, w0, u, v)						\
1569   do {									\
1570     UWtype __x0, __x1, __x2, __x3;					\
1571     UHWtype __ul, __vl, __uh, __vh;					\
1572 									\
1573     __ul = __ll_lowpart (u);						\
1574     __uh = __ll_highpart (u);						\
1575     __vl = __ll_lowpart (v);						\
1576     __vh = __ll_highpart (v);						\
1577 									\
1578     __x0 = (UWtype) __ul * __vl;					\
1579     __x1 = (UWtype) __ul * __vh;					\
1580     __x2 = (UWtype) __uh * __vl;					\
1581     __x3 = (UWtype) __uh * __vh;					\
1582 									\
1583     __x1 += __ll_highpart (__x0);/* this can't give carry */		\
1584     __x1 += __x2;		/* but this indeed can */		\
1585     if (__x1 < __x2)		/* did we get it? */			\
1586       __x3 += __ll_B;		/* yes, add it in the proper pos.  */	\
1587 									\
1588     (w1) = __x3 + __ll_highpart (__x1);					\
1589     (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0);		\
1590   } while (0)
1591 #endif
1592 
1593 #if !defined (__umulsidi3)
1594 #define __umulsidi3(u, v) \
1595   ({DWunion __w;							\
1596     umul_ppmm (__w.s.high, __w.s.low, u, v);				\
1597     __w.ll; })
1598 #endif
1599 
1600 /* Define this unconditionally, so it can be used for debugging.  */
1601 #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1602   do {									\
1603     UWtype __d1, __d0, __q1, __q0;					\
1604     UWtype __r1, __r0, __m;						\
1605     __d1 = __ll_highpart (d);						\
1606     __d0 = __ll_lowpart (d);						\
1607 									\
1608     __r1 = (n1) % __d1;							\
1609     __q1 = (n1) / __d1;							\
1610     __m = (UWtype) __q1 * __d0;						\
1611     __r1 = __r1 * __ll_B | __ll_highpart (n0);				\
1612     if (__r1 < __m)							\
1613       {									\
1614 	__q1--, __r1 += (d);						\
1615 	if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1616 	  if (__r1 < __m)						\
1617 	    __q1--, __r1 += (d);					\
1618       }									\
1619     __r1 -= __m;							\
1620 									\
1621     __r0 = __r1 % __d1;							\
1622     __q0 = __r1 / __d1;							\
1623     __m = (UWtype) __q0 * __d0;						\
1624     __r0 = __r0 * __ll_B | __ll_lowpart (n0);				\
1625     if (__r0 < __m)							\
1626       {									\
1627 	__q0--, __r0 += (d);						\
1628 	if (__r0 >= (d))						\
1629 	  if (__r0 < __m)						\
1630 	    __q0--, __r0 += (d);					\
1631       }									\
1632     __r0 -= __m;							\
1633 									\
1634     (q) = (UWtype) __q1 * __ll_B | __q0;				\
1635     (r) = __r0;								\
1636   } while (0)
1637 
1638 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1639    __udiv_w_sdiv (defined in libgcc or elsewhere).  */
1640 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1641 #define udiv_qrnnd(q, r, nh, nl, d) \
1642   do {									\
1643     USItype __r;							\
1644     (q) = __udiv_w_sdiv (&__r, nh, nl, d);				\
1645     (r) = __r;								\
1646   } while (0)
1647 #endif
1648 
1649 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c.  */
1650 #if !defined (udiv_qrnnd)
1651 #define UDIV_NEEDS_NORMALIZATION 1
1652 #define udiv_qrnnd __udiv_qrnnd_c
1653 #endif
1654 
1655 #if !defined (count_leading_zeros)
1656 #define count_leading_zeros(count, x) \
1657   do {									\
1658     UWtype __xr = (x);							\
1659     UWtype __a;								\
1660 									\
1661     if (W_TYPE_SIZE <= 32)						\
1662       {									\
1663 	__a = __xr < ((UWtype)1<<2*__BITS4)				\
1664 	  ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4)			\
1665 	  : (__xr < ((UWtype)1<<3*__BITS4) ?  2*__BITS4 : 3*__BITS4);	\
1666       }									\
1667     else								\
1668       {									\
1669 	for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8)			\
1670 	  if (((__xr >> __a) & 0xff) != 0)				\
1671 	    break;							\
1672       }									\
1673 									\
1674     (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a);		\
1675   } while (0)
1676 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1677 #endif
1678 
1679 #if !defined (count_trailing_zeros)
1680 /* Define count_trailing_zeros using count_leading_zeros.  The latter might be
1681    defined in asm, but if it is not, the C version above is good enough.  */
1682 #define count_trailing_zeros(count, x) \
1683   do {									\
1684     UWtype __ctz_x = (x);						\
1685     UWtype __ctz_c;							\
1686     count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x);			\
1687     (count) = W_TYPE_SIZE - 1 - __ctz_c;				\
1688   } while (0)
1689 #endif
1690 
1691 #ifndef UDIV_NEEDS_NORMALIZATION
1692 #define UDIV_NEEDS_NORMALIZATION 0
1693 #endif
1694