1/* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2   Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2004,
3   2005  Free Software Foundation, Inc.
4
5   This definition file is free software; you can redistribute it
6   and/or modify it under the terms of the GNU General Public
7   License as published by the Free Software Foundation; either
8   version 2, or (at your option) any later version.
9
10   This definition file is distributed in the hope that it will be
11   useful, but WITHOUT ANY WARRANTY; without even the implied
12   warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
13   See the GNU General Public License for more details.
14
15   You should have received a copy of the GNU General Public License
16   along with this program; if not, write to the Free Software
17   Foundation, Inc., 51 Franklin Street, Fifth Floor,
18   Boston, MA 02110-1301, USA.  */
19
20/* You have to define the following before including this file:
21
22   UWtype -- An unsigned type, default type for operations (typically a "word")
23   UHWtype -- An unsigned type, at least half the size of UWtype.
24   UDWtype -- An unsigned type, at least twice as large a UWtype
25   W_TYPE_SIZE -- size in bits of UWtype
26
27   UQItype -- Unsigned 8 bit type.
28   SItype, USItype -- Signed and unsigned 32 bit types.
29   DItype, UDItype -- Signed and unsigned 64 bit types.
30
31   On a 32 bit machine UWtype should typically be USItype;
32   on a 64 bit machine, UWtype should typically be UDItype.  */
33
34#define __BITS4 (W_TYPE_SIZE / 4)
35#define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
36#define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
37#define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
38
39#ifndef W_TYPE_SIZE
40#define W_TYPE_SIZE	32
41#define UWtype		USItype
42#define UHWtype		USItype
43#define UDWtype		UDItype
44#endif
45
46extern const UQItype __clz_tab[256];
47
48/* Define auxiliary asm macros.
49
50   1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
51   UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
52   word product in HIGH_PROD and LOW_PROD.
53
54   2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
55   UDWtype product.  This is just a variant of umul_ppmm.
56
57   3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
58   denominator) divides a UDWtype, composed by the UWtype integers
59   HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
60   in QUOTIENT and the remainder in REMAINDER.  HIGH_NUMERATOR must be less
61   than DENOMINATOR for correct operation.  If, in addition, the most
62   significant bit of DENOMINATOR must be 1, then the pre-processor symbol
63   UDIV_NEEDS_NORMALIZATION is defined to 1.
64
65   4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
66   denominator).  Like udiv_qrnnd but the numbers are signed.  The quotient
67   is rounded towards 0.
68
69   5) count_leading_zeros(count, x) counts the number of zero-bits from the
70   msb to the first nonzero bit in the UWtype X.  This is the number of
71   steps X needs to be shifted left to set the msb.  Undefined for X == 0,
72   unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
73
74   6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
75   from the least significant end.
76
77   7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
78   high_addend_2, low_addend_2) adds two UWtype integers, composed by
79   HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
80   respectively.  The result is placed in HIGH_SUM and LOW_SUM.  Overflow
81   (i.e. carry out) is not stored anywhere, and is lost.
82
83   8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
84   high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
85   composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
86   LOW_SUBTRAHEND_2 respectively.  The result is placed in HIGH_DIFFERENCE
87   and LOW_DIFFERENCE.  Overflow (i.e. carry out) is not stored anywhere,
88   and is lost.
89
90   If any of these macros are left undefined for a particular CPU,
91   C macros are used.  */
92
93/* The CPUs come in alphabetical order below.
94
95   Please add support for more CPUs here, or improve the current support
96   for the CPUs below!
97   (E.g. WE32100, IBM360.)  */
98
99#if defined (__GNUC__) && !defined (NO_ASM)
100
101/* We sometimes need to clobber "cc" with gcc2, but that would not be
102   understood by gcc1.  Use cpp to avoid major code duplication.  */
103#if __GNUC__ < 2
104#define __CLOBBER_CC
105#define __AND_CLOBBER_CC
106#else /* __GNUC__ >= 2 */
107#define __CLOBBER_CC : "cc"
108#define __AND_CLOBBER_CC , "cc"
109#endif /* __GNUC__ < 2 */
110
111#if defined (__alpha) && W_TYPE_SIZE == 64
112#define umul_ppmm(ph, pl, m0, m1) \
113  do {									\
114    UDItype __m0 = (m0), __m1 = (m1);					\
115    (ph) = __builtin_alpha_umulh (__m0, __m1);				\
116    (pl) = __m0 * __m1;							\
117  } while (0)
118#define UMUL_TIME 46
119#ifndef LONGLONG_STANDALONE
120#define udiv_qrnnd(q, r, n1, n0, d) \
121  do { UDItype __r;							\
122    (q) = __udiv_qrnnd (&__r, (n1), (n0), (d));				\
123    (r) = __r;								\
124  } while (0)
125extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
126#define UDIV_TIME 220
127#endif /* LONGLONG_STANDALONE */
128#ifdef __alpha_cix__
129#define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clzl (X))
130#define count_trailing_zeros(COUNT,X)	((COUNT) = __builtin_ctzl (X))
131#define COUNT_LEADING_ZEROS_0 64
132#else
133#define count_leading_zeros(COUNT,X) \
134  do {									\
135    UDItype __xr = (X), __t, __a;					\
136    __t = __builtin_alpha_cmpbge (0, __xr);				\
137    __a = __clz_tab[__t ^ 0xff] - 1;					\
138    __t = __builtin_alpha_extbl (__xr, __a);				\
139    (COUNT) = 64 - (__clz_tab[__t] + __a*8);				\
140  } while (0)
141#define count_trailing_zeros(COUNT,X) \
142  do {									\
143    UDItype __xr = (X), __t, __a;					\
144    __t = __builtin_alpha_cmpbge (0, __xr);				\
145    __t = ~__t & -~__t;							\
146    __a = ((__t & 0xCC) != 0) * 2;					\
147    __a += ((__t & 0xF0) != 0) * 4;					\
148    __a += ((__t & 0xAA) != 0);						\
149    __t = __builtin_alpha_extbl (__xr, __a);				\
150    __a <<= 3;								\
151    __t &= -__t;							\
152    __a += ((__t & 0xCC) != 0) * 2;					\
153    __a += ((__t & 0xF0) != 0) * 4;					\
154    __a += ((__t & 0xAA) != 0);						\
155    (COUNT) = __a;							\
156  } while (0)
157#endif /* __alpha_cix__ */
158#endif /* __alpha */
159
160#if defined (__arc__) && W_TYPE_SIZE == 32
161#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
162  __asm__ ("add.f	%1, %4, %5\n\tadc	%0, %2, %3"		\
163	   : "=r" ((USItype) (sh)),					\
164	     "=&r" ((USItype) (sl))					\
165	   : "%r" ((USItype) (ah)),					\
166	     "rIJ" ((USItype) (bh)),					\
167	     "%r" ((USItype) (al)),					\
168	     "rIJ" ((USItype) (bl)))
169#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
170  __asm__ ("sub.f	%1, %4, %5\n\tsbc	%0, %2, %3"		\
171	   : "=r" ((USItype) (sh)),					\
172	     "=&r" ((USItype) (sl))					\
173	   : "r" ((USItype) (ah)),					\
174	     "rIJ" ((USItype) (bh)),					\
175	     "r" ((USItype) (al)),					\
176	     "rIJ" ((USItype) (bl)))
177/* Call libgcc routine.  */
178#define umul_ppmm(w1, w0, u, v) \
179do {									\
180  DWunion __w;								\
181  __w.ll = __umulsidi3 (u, v);						\
182  w1 = __w.s.high;							\
183  w0 = __w.s.low;							\
184} while (0)
185#define __umulsidi3 __umulsidi3
186UDItype __umulsidi3 (USItype, USItype);
187#endif
188
189#if defined (__arm__) && !defined (__thumb__) && W_TYPE_SIZE == 32
190#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
191  __asm__ ("adds	%1, %4, %5\n\tadc	%0, %2, %3"		\
192	   : "=r" ((USItype) (sh)),					\
193	     "=&r" ((USItype) (sl))					\
194	   : "%r" ((USItype) (ah)),					\
195	     "rI" ((USItype) (bh)),					\
196	     "%r" ((USItype) (al)),					\
197	     "rI" ((USItype) (bl)) __CLOBBER_CC)
198#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
199  __asm__ ("subs	%1, %4, %5\n\tsbc	%0, %2, %3"		\
200	   : "=r" ((USItype) (sh)),					\
201	     "=&r" ((USItype) (sl))					\
202	   : "r" ((USItype) (ah)),					\
203	     "rI" ((USItype) (bh)),					\
204	     "r" ((USItype) (al)),					\
205	     "rI" ((USItype) (bl)) __CLOBBER_CC)
206#define umul_ppmm(xh, xl, a, b) \
207{register USItype __t0, __t1, __t2;					\
208  __asm__ ("/* Inlined umul_ppmm */\n"					\
209	   "	mov	%2, %5, lsr #16\n"				\
210	   "	mov	%0, %6, lsr #16\n"				\
211	   "	bic	%3, %5, %2, lsl #16\n"				\
212	   "	bic	%4, %6, %0, lsl #16\n"				\
213	   "	mul	%1, %3, %4\n"					\
214	   "	mul	%4, %2, %4\n"					\
215	   "	mul	%3, %0, %3\n"					\
216	   "	mul	%0, %2, %0\n"					\
217	   "	adds	%3, %4, %3\n"					\
218	   "	addcs	%0, %0, #65536\n"				\
219	   "	adds	%1, %1, %3, lsl #16\n"				\
220	   "	adc	%0, %0, %3, lsr #16"				\
221	   : "=&r" ((USItype) (xh)),					\
222	     "=r" ((USItype) (xl)),					\
223	     "=&r" (__t0), "=&r" (__t1), "=r" (__t2)			\
224	   : "r" ((USItype) (a)),					\
225	     "r" ((USItype) (b)) __CLOBBER_CC );}
226#define UMUL_TIME 20
227#define UDIV_TIME 100
228#endif /* __arm__ */
229
230#if defined (__hppa) && W_TYPE_SIZE == 32
231#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
232  __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0"				\
233	   : "=r" ((USItype) (sh)),					\
234	     "=&r" ((USItype) (sl))					\
235	   : "%rM" ((USItype) (ah)),					\
236	     "rM" ((USItype) (bh)),					\
237	     "%rM" ((USItype) (al)),					\
238	     "rM" ((USItype) (bl)))
239#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
240  __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0"				\
241	   : "=r" ((USItype) (sh)),					\
242	     "=&r" ((USItype) (sl))					\
243	   : "rM" ((USItype) (ah)),					\
244	     "rM" ((USItype) (bh)),					\
245	     "rM" ((USItype) (al)),					\
246	     "rM" ((USItype) (bl)))
247#if defined (_PA_RISC1_1)
248#define umul_ppmm(w1, w0, u, v) \
249  do {									\
250    union								\
251      {									\
252	UDItype __f;							\
253	struct {USItype __w1, __w0;} __w1w0;				\
254      } __t;								\
255    __asm__ ("xmpyu %1,%2,%0"						\
256	     : "=x" (__t.__f)						\
257	     : "x" ((USItype) (u)),					\
258	       "x" ((USItype) (v)));					\
259    (w1) = __t.__w1w0.__w1;						\
260    (w0) = __t.__w1w0.__w0;						\
261     } while (0)
262#define UMUL_TIME 8
263#else
264#define UMUL_TIME 30
265#endif
266#define UDIV_TIME 40
267#define count_leading_zeros(count, x) \
268  do {									\
269    USItype __tmp;							\
270    __asm__ (								\
271       "ldi		1,%0\n"						\
272"	extru,=		%1,15,16,%%r0		; Bits 31..16 zero?\n"	\
273"	extru,tr	%1,15,16,%1		; No.  Shift down, skip add.\n"\
274"	ldo		16(%0),%0		; Yes.  Perform add.\n"	\
275"	extru,=		%1,23,8,%%r0		; Bits 15..8 zero?\n"	\
276"	extru,tr	%1,23,8,%1		; No.  Shift down, skip add.\n"\
277"	ldo		8(%0),%0		; Yes.  Perform add.\n"	\
278"	extru,=		%1,27,4,%%r0		; Bits 7..4 zero?\n"	\
279"	extru,tr	%1,27,4,%1		; No.  Shift down, skip add.\n"\
280"	ldo		4(%0),%0		; Yes.  Perform add.\n"	\
281"	extru,=		%1,29,2,%%r0		; Bits 3..2 zero?\n"	\
282"	extru,tr	%1,29,2,%1		; No.  Shift down, skip add.\n"\
283"	ldo		2(%0),%0		; Yes.  Perform add.\n"	\
284"	extru		%1,30,1,%1		; Extract bit 1.\n"	\
285"	sub		%0,%1,%0		; Subtract it.\n"	\
286	: "=r" (count), "=r" (__tmp) : "1" (x));			\
287  } while (0)
288#endif
289
290#if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
291#define smul_ppmm(xh, xl, m0, m1) \
292  do {									\
293    union {DItype __ll;							\
294	   struct {USItype __h, __l;} __i;				\
295	  } __x;							\
296    __asm__ ("lr %N0,%1\n\tmr %0,%2"					\
297	     : "=&r" (__x.__ll)						\
298	     : "r" (m0), "r" (m1));					\
299    (xh) = __x.__i.__h; (xl) = __x.__i.__l;				\
300  } while (0)
301#define sdiv_qrnnd(q, r, n1, n0, d) \
302  do {									\
303    union {DItype __ll;							\
304	   struct {USItype __h, __l;} __i;				\
305	  } __x;							\
306    __x.__i.__h = n1; __x.__i.__l = n0;					\
307    __asm__ ("dr %0,%2"							\
308	     : "=r" (__x.__ll)						\
309	     : "0" (__x.__ll), "r" (d));				\
310    (q) = __x.__i.__l; (r) = __x.__i.__h;				\
311  } while (0)
312#endif
313
314#if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
315#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
316  __asm__ ("addl %5,%1\n\tadcl %3,%0"					\
317	   : "=r" ((USItype) (sh)),					\
318	     "=&r" ((USItype) (sl))					\
319	   : "%0" ((USItype) (ah)),					\
320	     "g" ((USItype) (bh)),					\
321	     "%1" ((USItype) (al)),					\
322	     "g" ((USItype) (bl)))
323#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
324  __asm__ ("subl %5,%1\n\tsbbl %3,%0"					\
325	   : "=r" (sh),							\
326	     "=&r" (sl)							\
327	   : "0" ((USItype) (ah)),					\
328	     "g" ((USItype) (bh)),					\
329	     "1" ((USItype) (al)),					\
330	     "g" ((USItype) (bl)))
331#define umul_ppmm(w1, w0, u, v) \
332  __asm__ ("mull %3"							\
333	   : "=a" (w0),							\
334	     "=d" (w1)							\
335	   : "%0" ((USItype) (u)),					\
336	     "rm" ((USItype) (v)))
337#define udiv_qrnnd(q, r, n1, n0, dv) \
338  __asm__ ("divl %4"							\
339	   : "=a" (q),							\
340	     "=d" (r)							\
341	   : "0" ((USItype) (n0)),					\
342	     "1" ((USItype) (n1)),					\
343	     "rm" ((USItype) (dv)))
344#define count_leading_zeros(count, x) \
345  do {									\
346    USItype __cbtmp;							\
347    __asm__ ("bsrl %1,%0"						\
348	     : "=r" (__cbtmp) : "rm" ((USItype) (x)));			\
349    (count) = __cbtmp ^ 31;						\
350  } while (0)
351#define count_trailing_zeros(count, x) \
352  __asm__ ("bsfl %1,%0" : "=r" (count) : "rm" ((USItype)(x)))
353#define UMUL_TIME 40
354#define UDIV_TIME 40
355#endif /* 80x86 */
356
357#if defined (__i960__) && W_TYPE_SIZE == 32
358#define umul_ppmm(w1, w0, u, v) \
359  ({union {UDItype __ll;						\
360	   struct {USItype __l, __h;} __i;				\
361	  } __xx;							\
362  __asm__ ("emul	%2,%1,%0"					\
363	   : "=d" (__xx.__ll)						\
364	   : "%dI" ((USItype) (u)),					\
365	     "dI" ((USItype) (v)));					\
366  (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
367#define __umulsidi3(u, v) \
368  ({UDItype __w;							\
369    __asm__ ("emul	%2,%1,%0"					\
370	     : "=d" (__w)						\
371	     : "%dI" ((USItype) (u)),					\
372	       "dI" ((USItype) (v)));					\
373    __w; })
374#endif /* __i960__ */
375
376#if defined (__M32R__) && W_TYPE_SIZE == 32
377#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
378  /* The cmp clears the condition bit.  */ \
379  __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3"			\
380	   : "=r" ((USItype) (sh)),					\
381	     "=&r" ((USItype) (sl))					\
382	   : "0" ((USItype) (ah)),					\
383	     "r" ((USItype) (bh)),					\
384	     "1" ((USItype) (al)),					\
385	     "r" ((USItype) (bl))					\
386	   : "cbit")
387#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
388  /* The cmp clears the condition bit.  */ \
389  __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3"			\
390	   : "=r" ((USItype) (sh)),					\
391	     "=&r" ((USItype) (sl))					\
392	   : "0" ((USItype) (ah)),					\
393	     "r" ((USItype) (bh)),					\
394	     "1" ((USItype) (al)),					\
395	     "r" ((USItype) (bl))					\
396	   : "cbit")
397#endif /* __M32R__ */
398
399#if defined (__mc68000__) && W_TYPE_SIZE == 32
400#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
401  __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0"				\
402	   : "=d" ((USItype) (sh)),					\
403	     "=&d" ((USItype) (sl))					\
404	   : "%0" ((USItype) (ah)),					\
405	     "d" ((USItype) (bh)),					\
406	     "%1" ((USItype) (al)),					\
407	     "g" ((USItype) (bl)))
408#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
409  __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0"				\
410	   : "=d" ((USItype) (sh)),					\
411	     "=&d" ((USItype) (sl))					\
412	   : "0" ((USItype) (ah)),					\
413	     "d" ((USItype) (bh)),					\
414	     "1" ((USItype) (al)),					\
415	     "g" ((USItype) (bl)))
416
417/* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r.  */
418#if (defined (__mc68020__) && !defined (__mc68060__))
419#define umul_ppmm(w1, w0, u, v) \
420  __asm__ ("mulu%.l %3,%1:%0"						\
421	   : "=d" ((USItype) (w0)),					\
422	     "=d" ((USItype) (w1))					\
423	   : "%0" ((USItype) (u)),					\
424	     "dmi" ((USItype) (v)))
425#define UMUL_TIME 45
426#define udiv_qrnnd(q, r, n1, n0, d) \
427  __asm__ ("divu%.l %4,%1:%0"						\
428	   : "=d" ((USItype) (q)),					\
429	     "=d" ((USItype) (r))					\
430	   : "0" ((USItype) (n0)),					\
431	     "1" ((USItype) (n1)),					\
432	     "dmi" ((USItype) (d)))
433#define UDIV_TIME 90
434#define sdiv_qrnnd(q, r, n1, n0, d) \
435  __asm__ ("divs%.l %4,%1:%0"						\
436	   : "=d" ((USItype) (q)),					\
437	     "=d" ((USItype) (r))					\
438	   : "0" ((USItype) (n0)),					\
439	     "1" ((USItype) (n1)),					\
440	     "dmi" ((USItype) (d)))
441
442#elif defined (__mcoldfire__) /* not mc68020 */
443
444#define umul_ppmm(xh, xl, a, b) \
445  __asm__ ("| Inlined umul_ppmm\n"					\
446	   "	move%.l	%2,%/d0\n"					\
447	   "	move%.l	%3,%/d1\n"					\
448	   "	move%.l	%/d0,%/d2\n"					\
449	   "	swap	%/d0\n"						\
450	   "	move%.l	%/d1,%/d3\n"					\
451	   "	swap	%/d1\n"						\
452	   "	move%.w	%/d2,%/d4\n"					\
453	   "	mulu	%/d3,%/d4\n"					\
454	   "	mulu	%/d1,%/d2\n"					\
455	   "	mulu	%/d0,%/d3\n"					\
456	   "	mulu	%/d0,%/d1\n"					\
457	   "	move%.l	%/d4,%/d0\n"					\
458	   "	clr%.w	%/d0\n"						\
459	   "	swap	%/d0\n"						\
460	   "	add%.l	%/d0,%/d2\n"					\
461	   "	add%.l	%/d3,%/d2\n"					\
462	   "	jcc	1f\n"						\
463	   "	add%.l	%#65536,%/d1\n"					\
464	   "1:	swap	%/d2\n"						\
465	   "	moveq	%#0,%/d0\n"					\
466	   "	move%.w	%/d2,%/d0\n"					\
467	   "	move%.w	%/d4,%/d2\n"					\
468	   "	move%.l	%/d2,%1\n"					\
469	   "	add%.l	%/d1,%/d0\n"					\
470	   "	move%.l	%/d0,%0"					\
471	   : "=g" ((USItype) (xh)),					\
472	     "=g" ((USItype) (xl))					\
473	   : "g" ((USItype) (a)),					\
474	     "g" ((USItype) (b))					\
475	   : "d0", "d1", "d2", "d3", "d4")
476#define UMUL_TIME 100
477#define UDIV_TIME 400
478#else /* not ColdFire */
479/* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX.  */
480#define umul_ppmm(xh, xl, a, b) \
481  __asm__ ("| Inlined umul_ppmm\n"					\
482	   "	move%.l	%2,%/d0\n"					\
483	   "	move%.l	%3,%/d1\n"					\
484	   "	move%.l	%/d0,%/d2\n"					\
485	   "	swap	%/d0\n"						\
486	   "	move%.l	%/d1,%/d3\n"					\
487	   "	swap	%/d1\n"						\
488	   "	move%.w	%/d2,%/d4\n"					\
489	   "	mulu	%/d3,%/d4\n"					\
490	   "	mulu	%/d1,%/d2\n"					\
491	   "	mulu	%/d0,%/d3\n"					\
492	   "	mulu	%/d0,%/d1\n"					\
493	   "	move%.l	%/d4,%/d0\n"					\
494	   "	eor%.w	%/d0,%/d0\n"					\
495	   "	swap	%/d0\n"						\
496	   "	add%.l	%/d0,%/d2\n"					\
497	   "	add%.l	%/d3,%/d2\n"					\
498	   "	jcc	1f\n"						\
499	   "	add%.l	%#65536,%/d1\n"					\
500	   "1:	swap	%/d2\n"						\
501	   "	moveq	%#0,%/d0\n"					\
502	   "	move%.w	%/d2,%/d0\n"					\
503	   "	move%.w	%/d4,%/d2\n"					\
504	   "	move%.l	%/d2,%1\n"					\
505	   "	add%.l	%/d1,%/d0\n"					\
506	   "	move%.l	%/d0,%0"					\
507	   : "=g" ((USItype) (xh)),					\
508	     "=g" ((USItype) (xl))					\
509	   : "g" ((USItype) (a)),					\
510	     "g" ((USItype) (b))					\
511	   : "d0", "d1", "d2", "d3", "d4")
512#define UMUL_TIME 100
513#define UDIV_TIME 400
514
515#endif /* not mc68020 */
516
517/* The '020, '030, '040 and '060 have bitfield insns.
518   cpu32 disguises as a 68020, but lacks them.  */
519#if defined (__mc68020__) && !defined (__mcpu32__)
520#define count_leading_zeros(count, x) \
521  __asm__ ("bfffo %1{%b2:%b2},%0"					\
522	   : "=d" ((USItype) (count))					\
523	   : "od" ((USItype) (x)), "n" (0))
524#endif
525#endif /* mc68000 */
526
527#if defined (__m88000__) && W_TYPE_SIZE == 32
528#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
529  __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3"			\
530	   : "=r" ((USItype) (sh)),					\
531	     "=&r" ((USItype) (sl))					\
532	   : "%rJ" ((USItype) (ah)),					\
533	     "rJ" ((USItype) (bh)),					\
534	     "%rJ" ((USItype) (al)),					\
535	     "rJ" ((USItype) (bl)))
536#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
537  __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3"			\
538	   : "=r" ((USItype) (sh)),					\
539	     "=&r" ((USItype) (sl))					\
540	   : "rJ" ((USItype) (ah)),					\
541	     "rJ" ((USItype) (bh)),					\
542	     "rJ" ((USItype) (al)),					\
543	     "rJ" ((USItype) (bl)))
544#define count_leading_zeros(count, x) \
545  do {									\
546    USItype __cbtmp;							\
547    __asm__ ("ff1 %0,%1"						\
548	     : "=r" (__cbtmp)						\
549	     : "r" ((USItype) (x)));					\
550    (count) = __cbtmp ^ 31;						\
551  } while (0)
552#define COUNT_LEADING_ZEROS_0 63 /* sic */
553#if defined (__mc88110__)
554#define umul_ppmm(wh, wl, u, v) \
555  do {									\
556    union {UDItype __ll;						\
557	   struct {USItype __h, __l;} __i;				\
558	  } __xx;							\
559    __asm__ ("mulu.d	%0,%1,%2"					\
560	     : "=r" (__xx.__ll)						\
561	     : "r" ((USItype) (u)),					\
562	       "r" ((USItype) (v)));					\
563    (wh) = __xx.__i.__h;						\
564    (wl) = __xx.__i.__l;						\
565  } while (0)
566#define udiv_qrnnd(q, r, n1, n0, d) \
567  ({union {UDItype __ll;						\
568	   struct {USItype __h, __l;} __i;				\
569	  } __xx;							\
570  USItype __q;								\
571  __xx.__i.__h = (n1); __xx.__i.__l = (n0);				\
572  __asm__ ("divu.d %0,%1,%2"						\
573	   : "=r" (__q)							\
574	   : "r" (__xx.__ll),						\
575	     "r" ((USItype) (d)));					\
576  (r) = (n0) - __q * (d); (q) = __q; })
577#define UMUL_TIME 5
578#define UDIV_TIME 25
579#else
580#define UMUL_TIME 17
581#define UDIV_TIME 150
582#endif /* __mc88110__ */
583#endif /* __m88000__ */
584
585#if defined (__mips__) && W_TYPE_SIZE == 32
586#define umul_ppmm(w1, w0, u, v) \
587  __asm__ ("multu %2,%3"						\
588	   : "=l" ((USItype) (w0)),					\
589	     "=h" ((USItype) (w1))					\
590	   : "d" ((USItype) (u)),					\
591	     "d" ((USItype) (v)))
592#define UMUL_TIME 10
593#define UDIV_TIME 100
594#endif /* __mips__ */
595
596#if defined (__ns32000__) && W_TYPE_SIZE == 32
597#define umul_ppmm(w1, w0, u, v) \
598  ({union {UDItype __ll;						\
599	   struct {USItype __l, __h;} __i;				\
600	  } __xx;							\
601  __asm__ ("meid %2,%0"							\
602	   : "=g" (__xx.__ll)						\
603	   : "%0" ((USItype) (u)),					\
604	     "g" ((USItype) (v)));					\
605  (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
606#define __umulsidi3(u, v) \
607  ({UDItype __w;							\
608    __asm__ ("meid %2,%0"						\
609	     : "=g" (__w)						\
610	     : "%0" ((USItype) (u)),					\
611	       "g" ((USItype) (v)));					\
612    __w; })
613#define udiv_qrnnd(q, r, n1, n0, d) \
614  ({union {UDItype __ll;						\
615	   struct {USItype __l, __h;} __i;				\
616	  } __xx;							\
617  __xx.__i.__h = (n1); __xx.__i.__l = (n0);				\
618  __asm__ ("deid %2,%0"							\
619	   : "=g" (__xx.__ll)						\
620	   : "0" (__xx.__ll),						\
621	     "g" ((USItype) (d)));					\
622  (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
623#define count_trailing_zeros(count,x) \
624  do {									\
625    __asm__ ("ffsd     %2,%0"						\
626            : "=r" ((USItype) (count))					\
627            : "0" ((USItype) 0),					\
628              "r" ((USItype) (x)));					\
629  } while (0)
630#endif /* __ns32000__ */
631
632/* FIXME: We should test _IBMR2 here when we add assembly support for the
633   system vendor compilers.
634   FIXME: What's needed for gcc PowerPC VxWorks?  __vxworks__ is not good
635   enough, since that hits ARM and m68k too.  */
636#if (defined (_ARCH_PPC)	/* AIX */				\
637     || defined (_ARCH_PWR)	/* AIX */				\
638     || defined (_ARCH_COM)	/* AIX */				\
639     || defined (__powerpc__)	/* gcc */				\
640     || defined (__POWERPC__)	/* BEOS */				\
641     || defined (__ppc__)	/* Darwin */				\
642     || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */    \
643     || (defined (PPC) && defined (CPU_FAMILY)    /* VxWorks */               \
644         && CPU_FAMILY == PPC)                                                \
645     ) && W_TYPE_SIZE == 32
646#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
647  do {									\
648    if (__builtin_constant_p (bh) && (bh) == 0)				\
649      __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2"		\
650	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
651    else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0)		\
652      __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2"		\
653	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
654    else								\
655      __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3"		\
656	     : "=r" (sh), "=&r" (sl)					\
657	     : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl));		\
658  } while (0)
659#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
660  do {									\
661    if (__builtin_constant_p (ah) && (ah) == 0)				\
662      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2"	\
663	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
664    else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0)		\
665      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2"	\
666	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
667    else if (__builtin_constant_p (bh) && (bh) == 0)			\
668      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2"		\
669	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
670    else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0)		\
671      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2"		\
672	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
673    else								\
674      __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2"	\
675	       : "=r" (sh), "=&r" (sl)					\
676	       : "r" (ah), "r" (bh), "rI" (al), "r" (bl));		\
677  } while (0)
678#define count_leading_zeros(count, x) \
679  __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x))
680#define COUNT_LEADING_ZEROS_0 32
681#if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
682  || defined (__ppc__)                                                    \
683  || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */       \
684  || (defined (PPC) && defined (CPU_FAMILY)    /* VxWorks */                  \
685         && CPU_FAMILY == PPC)
686#define umul_ppmm(ph, pl, m0, m1) \
687  do {									\
688    USItype __m0 = (m0), __m1 = (m1);					\
689    __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
690    (pl) = __m0 * __m1;							\
691  } while (0)
692#define UMUL_TIME 15
693#define smul_ppmm(ph, pl, m0, m1) \
694  do {									\
695    SItype __m0 = (m0), __m1 = (m1);					\
696    __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
697    (pl) = __m0 * __m1;							\
698  } while (0)
699#define SMUL_TIME 14
700#define UDIV_TIME 120
701#elif defined (_ARCH_PWR)
702#define UMUL_TIME 8
703#define smul_ppmm(xh, xl, m0, m1) \
704  __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1))
705#define SMUL_TIME 4
706#define sdiv_qrnnd(q, r, nh, nl, d) \
707  __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d))
708#define UDIV_TIME 100
709#endif
710#endif /* 32-bit POWER architecture variants.  */
711
712/* We should test _IBMR2 here when we add assembly support for the system
713   vendor compilers.  */
714#if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
715#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
716  do {									\
717    if (__builtin_constant_p (bh) && (bh) == 0)				\
718      __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2"		\
719	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
720    else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0)		\
721      __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2"		\
722	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
723    else								\
724      __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3"		\
725	     : "=r" (sh), "=&r" (sl)					\
726	     : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl));		\
727  } while (0)
728#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
729  do {									\
730    if (__builtin_constant_p (ah) && (ah) == 0)				\
731      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2"	\
732	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
733    else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0)		\
734      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2"	\
735	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
736    else if (__builtin_constant_p (bh) && (bh) == 0)			\
737      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2"		\
738	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
739    else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0)		\
740      __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2"		\
741	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
742    else								\
743      __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2"	\
744	       : "=r" (sh), "=&r" (sl)					\
745	       : "r" (ah), "r" (bh), "rI" (al), "r" (bl));		\
746  } while (0)
747#define count_leading_zeros(count, x) \
748  __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
749#define COUNT_LEADING_ZEROS_0 64
750#define umul_ppmm(ph, pl, m0, m1) \
751  do {									\
752    UDItype __m0 = (m0), __m1 = (m1);					\
753    __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
754    (pl) = __m0 * __m1;							\
755  } while (0)
756#define UMUL_TIME 15
757#define smul_ppmm(ph, pl, m0, m1) \
758  do {									\
759    DItype __m0 = (m0), __m1 = (m1);					\
760    __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
761    (pl) = __m0 * __m1;							\
762  } while (0)
763#define SMUL_TIME 14  /* ??? */
764#define UDIV_TIME 120 /* ??? */
765#endif /* 64-bit PowerPC.  */
766
767#if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
768#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
769  __asm__ ("a %1,%5\n\tae %0,%3"					\
770	   : "=r" ((USItype) (sh)),					\
771	     "=&r" ((USItype) (sl))					\
772	   : "%0" ((USItype) (ah)),					\
773	     "r" ((USItype) (bh)),					\
774	     "%1" ((USItype) (al)),					\
775	     "r" ((USItype) (bl)))
776#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
777  __asm__ ("s %1,%5\n\tse %0,%3"					\
778	   : "=r" ((USItype) (sh)),					\
779	     "=&r" ((USItype) (sl))					\
780	   : "0" ((USItype) (ah)),					\
781	     "r" ((USItype) (bh)),					\
782	     "1" ((USItype) (al)),					\
783	     "r" ((USItype) (bl)))
784#define umul_ppmm(ph, pl, m0, m1) \
785  do {									\
786    USItype __m0 = (m0), __m1 = (m1);					\
787    __asm__ (								\
788       "s	r2,r2\n"						\
789"	mts	r10,%2\n"						\
790"	m	r2,%3\n"						\
791"	m	r2,%3\n"						\
792"	m	r2,%3\n"						\
793"	m	r2,%3\n"						\
794"	m	r2,%3\n"						\
795"	m	r2,%3\n"						\
796"	m	r2,%3\n"						\
797"	m	r2,%3\n"						\
798"	m	r2,%3\n"						\
799"	m	r2,%3\n"						\
800"	m	r2,%3\n"						\
801"	m	r2,%3\n"						\
802"	m	r2,%3\n"						\
803"	m	r2,%3\n"						\
804"	m	r2,%3\n"						\
805"	m	r2,%3\n"						\
806"	cas	%0,r2,r0\n"						\
807"	mfs	r10,%1"							\
808	     : "=r" ((USItype) (ph)),					\
809	       "=r" ((USItype) (pl))					\
810	     : "%r" (__m0),						\
811		"r" (__m1)						\
812	     : "r2");							\
813    (ph) += ((((SItype) __m0 >> 31) & __m1)				\
814	     + (((SItype) __m1 >> 31) & __m0));				\
815  } while (0)
816#define UMUL_TIME 20
817#define UDIV_TIME 200
818#define count_leading_zeros(count, x) \
819  do {									\
820    if ((x) >= 0x10000)							\
821      __asm__ ("clz	%0,%1"						\
822	       : "=r" ((USItype) (count))				\
823	       : "r" ((USItype) (x) >> 16));				\
824    else								\
825      {									\
826	__asm__ ("clz	%0,%1"						\
827		 : "=r" ((USItype) (count))				\
828		 : "r" ((USItype) (x)));					\
829	(count) += 16;							\
830      }									\
831  } while (0)
832#endif
833
834#if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
835#ifndef __sh1__
836#define umul_ppmm(w1, w0, u, v) \
837  __asm__ (								\
838       "dmulu.l	%2,%3\n\tsts%M1	macl,%1\n\tsts%M0	mach,%0"	\
839	   : "=r<" ((USItype)(w1)),					\
840	     "=r<" ((USItype)(w0))					\
841	   : "r" ((USItype)(u)),					\
842	     "r" ((USItype)(v))						\
843	   : "macl", "mach")
844#define UMUL_TIME 5
845#endif
846
847/* This is the same algorithm as __udiv_qrnnd_c.  */
848#define UDIV_NEEDS_NORMALIZATION 1
849
850#define udiv_qrnnd(q, r, n1, n0, d) \
851  do {									\
852    extern UWtype __udiv_qrnnd_16 (UWtype, UWtype)			\
853                        __attribute__ ((visibility ("hidden")));	\
854    /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */	\
855    __asm__ (								\
856	"mov%M4 %4,r5\n"						\
857"	swap.w %3,r4\n"							\
858"	swap.w r5,r6\n"							\
859"	jsr @%5\n"							\
860"	shll16 r6\n"							\
861"	swap.w r4,r4\n"							\
862"	jsr @%5\n"							\
863"	swap.w r1,%0\n"							\
864"	or r1,%0"							\
865	: "=r" (q), "=&z" (r)						\
866	: "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16)		\
867	: "r1", "r2", "r4", "r5", "r6", "pr");				\
868  } while (0)
869
870#define UDIV_TIME 80
871
872#define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
873  __asm__ ("clrt;subc %5,%1; subc %4,%0"				\
874	   : "=r" (sh), "=r" (sl)					\
875	   : "0" (ah), "1" (al), "r" (bh), "r" (bl))
876
877#endif /* __sh__ */
878
879#if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
880#define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
881#define count_leading_zeros(count, x) \
882  do									\
883    {									\
884      UDItype x_ = (USItype)(x);					\
885      SItype c_;							\
886									\
887      __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_));			\
888      (count) = c_ - 31;						\
889    }									\
890  while (0)
891#define COUNT_LEADING_ZEROS_0 32
892#endif
893
894#if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
895    && W_TYPE_SIZE == 32
896#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
897  __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0"				\
898	   : "=r" ((USItype) (sh)),					\
899	     "=&r" ((USItype) (sl))					\
900	   : "%rJ" ((USItype) (ah)),					\
901	     "rI" ((USItype) (bh)),					\
902	     "%rJ" ((USItype) (al)),					\
903	     "rI" ((USItype) (bl))					\
904	   __CLOBBER_CC)
905#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
906  __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0"				\
907	   : "=r" ((USItype) (sh)),					\
908	     "=&r" ((USItype) (sl))					\
909	   : "rJ" ((USItype) (ah)),					\
910	     "rI" ((USItype) (bh)),					\
911	     "rJ" ((USItype) (al)),					\
912	     "rI" ((USItype) (bl))					\
913	   __CLOBBER_CC)
914#if defined (__sparc_v8__)
915#define umul_ppmm(w1, w0, u, v) \
916  __asm__ ("umul %2,%3,%1;rd %%y,%0"					\
917	   : "=r" ((USItype) (w1)),					\
918	     "=r" ((USItype) (w0))					\
919	   : "r" ((USItype) (u)),					\
920	     "r" ((USItype) (v)))
921#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
922  __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
923	   : "=&r" ((USItype) (__q)),					\
924	     "=&r" ((USItype) (__r))					\
925	   : "r" ((USItype) (__n1)),					\
926	     "r" ((USItype) (__n0)),					\
927	     "r" ((USItype) (__d)))
928#else
929#if defined (__sparclite__)
930/* This has hardware multiply but not divide.  It also has two additional
931   instructions scan (ffs from high bit) and divscc.  */
932#define umul_ppmm(w1, w0, u, v) \
933  __asm__ ("umul %2,%3,%1;rd %%y,%0"					\
934	   : "=r" ((USItype) (w1)),					\
935	     "=r" ((USItype) (w0))					\
936	   : "r" ((USItype) (u)),					\
937	     "r" ((USItype) (v)))
938#define udiv_qrnnd(q, r, n1, n0, d) \
939  __asm__ ("! Inlined udiv_qrnnd\n"					\
940"	wr	%%g0,%2,%%y	! Not a delayed write for sparclite\n"	\
941"	tst	%%g0\n"							\
942"	divscc	%3,%4,%%g1\n"						\
943"	divscc	%%g1,%4,%%g1\n"						\
944"	divscc	%%g1,%4,%%g1\n"						\
945"	divscc	%%g1,%4,%%g1\n"						\
946"	divscc	%%g1,%4,%%g1\n"						\
947"	divscc	%%g1,%4,%%g1\n"						\
948"	divscc	%%g1,%4,%%g1\n"						\
949"	divscc	%%g1,%4,%%g1\n"						\
950"	divscc	%%g1,%4,%%g1\n"						\
951"	divscc	%%g1,%4,%%g1\n"						\
952"	divscc	%%g1,%4,%%g1\n"						\
953"	divscc	%%g1,%4,%%g1\n"						\
954"	divscc	%%g1,%4,%%g1\n"						\
955"	divscc	%%g1,%4,%%g1\n"						\
956"	divscc	%%g1,%4,%%g1\n"						\
957"	divscc	%%g1,%4,%%g1\n"						\
958"	divscc	%%g1,%4,%%g1\n"						\
959"	divscc	%%g1,%4,%%g1\n"						\
960"	divscc	%%g1,%4,%%g1\n"						\
961"	divscc	%%g1,%4,%%g1\n"						\
962"	divscc	%%g1,%4,%%g1\n"						\
963"	divscc	%%g1,%4,%%g1\n"						\
964"	divscc	%%g1,%4,%%g1\n"						\
965"	divscc	%%g1,%4,%%g1\n"						\
966"	divscc	%%g1,%4,%%g1\n"						\
967"	divscc	%%g1,%4,%%g1\n"						\
968"	divscc	%%g1,%4,%%g1\n"						\
969"	divscc	%%g1,%4,%%g1\n"						\
970"	divscc	%%g1,%4,%%g1\n"						\
971"	divscc	%%g1,%4,%%g1\n"						\
972"	divscc	%%g1,%4,%%g1\n"						\
973"	divscc	%%g1,%4,%0\n"						\
974"	rd	%%y,%1\n"						\
975"	bl,a 1f\n"							\
976"	add	%1,%4,%1\n"						\
977"1:	! End of inline udiv_qrnnd"					\
978	   : "=r" ((USItype) (q)),					\
979	     "=r" ((USItype) (r))					\
980	   : "r" ((USItype) (n1)),					\
981	     "r" ((USItype) (n0)),					\
982	     "rI" ((USItype) (d))					\
983	   : "g1" __AND_CLOBBER_CC)
984#define UDIV_TIME 37
985#define count_leading_zeros(count, x) \
986  do {                                                                  \
987  __asm__ ("scan %1,1,%0"                                               \
988           : "=r" ((USItype) (count))                                   \
989           : "r" ((USItype) (x)));					\
990  } while (0)
991/* Early sparclites return 63 for an argument of 0, but they warn that future
992   implementations might change this.  Therefore, leave COUNT_LEADING_ZEROS_0
993   undefined.  */
994#else
995/* SPARC without integer multiplication and divide instructions.
996   (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
997#define umul_ppmm(w1, w0, u, v) \
998  __asm__ ("! Inlined umul_ppmm\n"					\
999"	wr	%%g0,%2,%%y	! SPARC has 0-3 delay insn after a wr\n"\
1000"	sra	%3,31,%%o5	! Don't move this insn\n"		\
1001"	and	%2,%%o5,%%o5	! Don't move this insn\n"		\
1002"	andcc	%%g0,0,%%g1	! Don't move this insn\n"		\
1003"	mulscc	%%g1,%3,%%g1\n"						\
1004"	mulscc	%%g1,%3,%%g1\n"						\
1005"	mulscc	%%g1,%3,%%g1\n"						\
1006"	mulscc	%%g1,%3,%%g1\n"						\
1007"	mulscc	%%g1,%3,%%g1\n"						\
1008"	mulscc	%%g1,%3,%%g1\n"						\
1009"	mulscc	%%g1,%3,%%g1\n"						\
1010"	mulscc	%%g1,%3,%%g1\n"						\
1011"	mulscc	%%g1,%3,%%g1\n"						\
1012"	mulscc	%%g1,%3,%%g1\n"						\
1013"	mulscc	%%g1,%3,%%g1\n"						\
1014"	mulscc	%%g1,%3,%%g1\n"						\
1015"	mulscc	%%g1,%3,%%g1\n"						\
1016"	mulscc	%%g1,%3,%%g1\n"						\
1017"	mulscc	%%g1,%3,%%g1\n"						\
1018"	mulscc	%%g1,%3,%%g1\n"						\
1019"	mulscc	%%g1,%3,%%g1\n"						\
1020"	mulscc	%%g1,%3,%%g1\n"						\
1021"	mulscc	%%g1,%3,%%g1\n"						\
1022"	mulscc	%%g1,%3,%%g1\n"						\
1023"	mulscc	%%g1,%3,%%g1\n"						\
1024"	mulscc	%%g1,%3,%%g1\n"						\
1025"	mulscc	%%g1,%3,%%g1\n"						\
1026"	mulscc	%%g1,%3,%%g1\n"						\
1027"	mulscc	%%g1,%3,%%g1\n"						\
1028"	mulscc	%%g1,%3,%%g1\n"						\
1029"	mulscc	%%g1,%3,%%g1\n"						\
1030"	mulscc	%%g1,%3,%%g1\n"						\
1031"	mulscc	%%g1,%3,%%g1\n"						\
1032"	mulscc	%%g1,%3,%%g1\n"						\
1033"	mulscc	%%g1,%3,%%g1\n"						\
1034"	mulscc	%%g1,%3,%%g1\n"						\
1035"	mulscc	%%g1,0,%%g1\n"						\
1036"	add	%%g1,%%o5,%0\n"						\
1037"	rd	%%y,%1"							\
1038	   : "=r" ((USItype) (w1)),					\
1039	     "=r" ((USItype) (w0))					\
1040	   : "%rI" ((USItype) (u)),					\
1041	     "r" ((USItype) (v))						\
1042	   : "g1", "o5" __AND_CLOBBER_CC)
1043#define UMUL_TIME 39		/* 39 instructions */
1044/* It's quite necessary to add this much assembler for the sparc.
1045   The default udiv_qrnnd (in C) is more than 10 times slower!  */
1046#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1047  __asm__ ("! Inlined udiv_qrnnd\n"					\
1048"	mov	32,%%g1\n"						\
1049"	subcc	%1,%2,%%g0\n"						\
1050"1:	bcs	5f\n"							\
1051"	 addxcc %0,%0,%0	! shift n1n0 and a q-bit in lsb\n"	\
1052"	sub	%1,%2,%1	! this kills msb of n\n"		\
1053"	addx	%1,%1,%1	! so this can't give carry\n"		\
1054"	subcc	%%g1,1,%%g1\n"						\
1055"2:	bne	1b\n"							\
1056"	 subcc	%1,%2,%%g0\n"						\
1057"	bcs	3f\n"							\
1058"	 addxcc %0,%0,%0	! shift n1n0 and a q-bit in lsb\n"	\
1059"	b	3f\n"							\
1060"	 sub	%1,%2,%1	! this kills msb of n\n"		\
1061"4:	sub	%1,%2,%1\n"						\
1062"5:	addxcc	%1,%1,%1\n"						\
1063"	bcc	2b\n"							\
1064"	 subcc	%%g1,1,%%g1\n"						\
1065"! Got carry from n.  Subtract next step to cancel this carry.\n"	\
1066"	bne	4b\n"							\
1067"	 addcc	%0,%0,%0	! shift n1n0 and a 0-bit in lsb\n"	\
1068"	sub	%1,%2,%1\n"						\
1069"3:	xnor	%0,0,%0\n"						\
1070"	! End of inline udiv_qrnnd"					\
1071	   : "=&r" ((USItype) (__q)),					\
1072	     "=&r" ((USItype) (__r))					\
1073	   : "r" ((USItype) (__d)),					\
1074	     "1" ((USItype) (__n1)),					\
1075	     "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1076#define UDIV_TIME (3+7*32)	/* 7 instructions/iteration. 32 iterations.  */
1077#endif /* __sparclite__ */
1078#endif /* __sparc_v8__ */
1079#endif /* sparc32 */
1080
1081#if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1082    && W_TYPE_SIZE == 64
1083#define add_ssaaaa(sh, sl, ah, al, bh, bl)				\
1084  __asm__ ("addcc %r4,%5,%1\n\t"					\
1085   	   "add %r2,%3,%0\n\t"						\
1086   	   "bcs,a,pn %%xcc, 1f\n\t"					\
1087   	   "add %0, 1, %0\n"						\
1088	   "1:"								\
1089	   : "=r" ((UDItype)(sh)),				      	\
1090	     "=&r" ((UDItype)(sl))				      	\
1091	   : "%rJ" ((UDItype)(ah)),				     	\
1092	     "rI" ((UDItype)(bh)),				      	\
1093	     "%rJ" ((UDItype)(al)),				     	\
1094	     "rI" ((UDItype)(bl))				       	\
1095	   __CLOBBER_CC)
1096
1097#define sub_ddmmss(sh, sl, ah, al, bh, bl) 				\
1098  __asm__ ("subcc %r4,%5,%1\n\t"					\
1099   	   "sub %r2,%3,%0\n\t"						\
1100   	   "bcs,a,pn %%xcc, 1f\n\t"					\
1101   	   "sub %0, 1, %0\n\t"						\
1102	   "1:"								\
1103	   : "=r" ((UDItype)(sh)),				      	\
1104	     "=&r" ((UDItype)(sl))				      	\
1105	   : "rJ" ((UDItype)(ah)),				     	\
1106	     "rI" ((UDItype)(bh)),				      	\
1107	     "rJ" ((UDItype)(al)),				     	\
1108	     "rI" ((UDItype)(bl))				       	\
1109	   __CLOBBER_CC)
1110
1111#define umul_ppmm(wh, wl, u, v)						\
1112  do {									\
1113	  UDItype tmp1, tmp2, tmp3, tmp4;				\
1114	  __asm__ __volatile__ (					\
1115		   "srl %7,0,%3\n\t"					\
1116		   "mulx %3,%6,%1\n\t"					\
1117		   "srlx %6,32,%2\n\t"					\
1118		   "mulx %2,%3,%4\n\t"					\
1119		   "sllx %4,32,%5\n\t"					\
1120		   "srl %6,0,%3\n\t"					\
1121		   "sub %1,%5,%5\n\t"					\
1122		   "srlx %5,32,%5\n\t"					\
1123		   "addcc %4,%5,%4\n\t"					\
1124		   "srlx %7,32,%5\n\t"					\
1125		   "mulx %3,%5,%3\n\t"					\
1126		   "mulx %2,%5,%5\n\t"					\
1127		   "sethi %%hi(0x80000000),%2\n\t"			\
1128		   "addcc %4,%3,%4\n\t"					\
1129		   "srlx %4,32,%4\n\t"					\
1130		   "add %2,%2,%2\n\t"					\
1131		   "movcc %%xcc,%%g0,%2\n\t"				\
1132		   "addcc %5,%4,%5\n\t"					\
1133		   "sllx %3,32,%3\n\t"					\
1134		   "add %1,%3,%1\n\t"					\
1135		   "add %5,%2,%0"					\
1136	   : "=r" ((UDItype)(wh)),					\
1137	     "=&r" ((UDItype)(wl)),					\
1138	     "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4)	\
1139	   : "r" ((UDItype)(u)),					\
1140	     "r" ((UDItype)(v))						\
1141	   __CLOBBER_CC);						\
1142  } while (0)
1143#define UMUL_TIME 96
1144#define UDIV_TIME 230
1145#endif /* sparc64 */
1146
1147#if defined (__vax__) && W_TYPE_SIZE == 32
1148#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1149  __asm__ ("addl2 %5,%1\n\tadwc %3,%0"					\
1150	   : "=g" ((USItype) (sh)),					\
1151	     "=&g" ((USItype) (sl))					\
1152	   : "%0" ((USItype) (ah)),					\
1153	     "g" ((USItype) (bh)),					\
1154	     "%1" ((USItype) (al)),					\
1155	     "g" ((USItype) (bl)))
1156#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1157  __asm__ ("subl2 %5,%1\n\tsbwc %3,%0"					\
1158	   : "=g" ((USItype) (sh)),					\
1159	     "=&g" ((USItype) (sl))					\
1160	   : "0" ((USItype) (ah)),					\
1161	     "g" ((USItype) (bh)),					\
1162	     "1" ((USItype) (al)),					\
1163	     "g" ((USItype) (bl)))
1164#define umul_ppmm(xh, xl, m0, m1) \
1165  do {									\
1166    union {								\
1167	UDItype __ll;							\
1168	struct {USItype __l, __h;} __i;					\
1169      } __xx;								\
1170    USItype __m0 = (m0), __m1 = (m1);					\
1171    __asm__ ("emul %1,%2,$0,%0"						\
1172	     : "=r" (__xx.__ll)						\
1173	     : "g" (__m0),						\
1174	       "g" (__m1));						\
1175    (xh) = __xx.__i.__h;						\
1176    (xl) = __xx.__i.__l;						\
1177    (xh) += ((((SItype) __m0 >> 31) & __m1)				\
1178	     + (((SItype) __m1 >> 31) & __m0));				\
1179  } while (0)
1180#define sdiv_qrnnd(q, r, n1, n0, d) \
1181  do {									\
1182    union {DItype __ll;							\
1183	   struct {SItype __l, __h;} __i;				\
1184	  } __xx;							\
1185    __xx.__i.__h = n1; __xx.__i.__l = n0;				\
1186    __asm__ ("ediv %3,%2,%0,%1"						\
1187	     : "=g" (q), "=g" (r)					\
1188	     : "g" (__xx.__ll), "g" (d));				\
1189  } while (0)
1190#endif /* __vax__ */
1191
1192#if defined (__z8000__) && W_TYPE_SIZE == 16
1193#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1194  __asm__ ("add	%H1,%H5\n\tadc	%H0,%H3"				\
1195	   : "=r" ((unsigned int)(sh)),					\
1196	     "=&r" ((unsigned int)(sl))					\
1197	   : "%0" ((unsigned int)(ah)),					\
1198	     "r" ((unsigned int)(bh)),					\
1199	     "%1" ((unsigned int)(al)),					\
1200	     "rQR" ((unsigned int)(bl)))
1201#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1202  __asm__ ("sub	%H1,%H5\n\tsbc	%H0,%H3"				\
1203	   : "=r" ((unsigned int)(sh)),					\
1204	     "=&r" ((unsigned int)(sl))					\
1205	   : "0" ((unsigned int)(ah)),					\
1206	     "r" ((unsigned int)(bh)),					\
1207	     "1" ((unsigned int)(al)),					\
1208	     "rQR" ((unsigned int)(bl)))
1209#define umul_ppmm(xh, xl, m0, m1) \
1210  do {									\
1211    union {long int __ll;						\
1212	   struct {unsigned int __h, __l;} __i;				\
1213	  } __xx;							\
1214    unsigned int __m0 = (m0), __m1 = (m1);				\
1215    __asm__ ("mult	%S0,%H3"					\
1216	     : "=r" (__xx.__i.__h),					\
1217	       "=r" (__xx.__i.__l)					\
1218	     : "%1" (__m0),						\
1219	       "rQR" (__m1));						\
1220    (xh) = __xx.__i.__h; (xl) = __xx.__i.__l;				\
1221    (xh) += ((((signed int) __m0 >> 15) & __m1)				\
1222	     + (((signed int) __m1 >> 15) & __m0));			\
1223  } while (0)
1224#endif /* __z8000__ */
1225
1226#endif /* __GNUC__ */
1227
1228/* If this machine has no inline assembler, use C macros.  */
1229
1230#if !defined (add_ssaaaa)
1231#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1232  do {									\
1233    UWtype __x;								\
1234    __x = (al) + (bl);							\
1235    (sh) = (ah) + (bh) + (__x < (al));					\
1236    (sl) = __x;								\
1237  } while (0)
1238#endif
1239
1240#if !defined (sub_ddmmss)
1241#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1242  do {									\
1243    UWtype __x;								\
1244    __x = (al) - (bl);							\
1245    (sh) = (ah) - (bh) - (__x > (al));					\
1246    (sl) = __x;								\
1247  } while (0)
1248#endif
1249
1250/* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1251   smul_ppmm.  */
1252#if !defined (umul_ppmm) && defined (smul_ppmm)
1253#define umul_ppmm(w1, w0, u, v)						\
1254  do {									\
1255    UWtype __w1;							\
1256    UWtype __xm0 = (u), __xm1 = (v);					\
1257    smul_ppmm (__w1, w0, __xm0, __xm1);					\
1258    (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1)		\
1259		+ (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0);		\
1260  } while (0)
1261#endif
1262
1263/* If we still don't have umul_ppmm, define it using plain C.  */
1264#if !defined (umul_ppmm)
1265#define umul_ppmm(w1, w0, u, v)						\
1266  do {									\
1267    UWtype __x0, __x1, __x2, __x3;					\
1268    UHWtype __ul, __vl, __uh, __vh;					\
1269									\
1270    __ul = __ll_lowpart (u);						\
1271    __uh = __ll_highpart (u);						\
1272    __vl = __ll_lowpart (v);						\
1273    __vh = __ll_highpart (v);						\
1274									\
1275    __x0 = (UWtype) __ul * __vl;					\
1276    __x1 = (UWtype) __ul * __vh;					\
1277    __x2 = (UWtype) __uh * __vl;					\
1278    __x3 = (UWtype) __uh * __vh;					\
1279									\
1280    __x1 += __ll_highpart (__x0);/* this can't give carry */		\
1281    __x1 += __x2;		/* but this indeed can */		\
1282    if (__x1 < __x2)		/* did we get it? */			\
1283      __x3 += __ll_B;		/* yes, add it in the proper pos.  */	\
1284									\
1285    (w1) = __x3 + __ll_highpart (__x1);					\
1286    (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0);		\
1287  } while (0)
1288#endif
1289
1290#if !defined (__umulsidi3)
1291#define __umulsidi3(u, v) \
1292  ({DWunion __w;							\
1293    umul_ppmm (__w.s.high, __w.s.low, u, v);				\
1294    __w.ll; })
1295#endif
1296
1297/* Define this unconditionally, so it can be used for debugging.  */
1298#define __udiv_qrnnd_c(q, r, n1, n0, d) \
1299  do {									\
1300    UWtype __d1, __d0, __q1, __q0;					\
1301    UWtype __r1, __r0, __m;						\
1302    __d1 = __ll_highpart (d);						\
1303    __d0 = __ll_lowpart (d);						\
1304									\
1305    __r1 = (n1) % __d1;							\
1306    __q1 = (n1) / __d1;							\
1307    __m = (UWtype) __q1 * __d0;						\
1308    __r1 = __r1 * __ll_B | __ll_highpart (n0);				\
1309    if (__r1 < __m)							\
1310      {									\
1311	__q1--, __r1 += (d);						\
1312	if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1313	  if (__r1 < __m)						\
1314	    __q1--, __r1 += (d);					\
1315      }									\
1316    __r1 -= __m;							\
1317									\
1318    __r0 = __r1 % __d1;							\
1319    __q0 = __r1 / __d1;							\
1320    __m = (UWtype) __q0 * __d0;						\
1321    __r0 = __r0 * __ll_B | __ll_lowpart (n0);				\
1322    if (__r0 < __m)							\
1323      {									\
1324	__q0--, __r0 += (d);						\
1325	if (__r0 >= (d))						\
1326	  if (__r0 < __m)						\
1327	    __q0--, __r0 += (d);					\
1328      }									\
1329    __r0 -= __m;							\
1330									\
1331    (q) = (UWtype) __q1 * __ll_B | __q0;				\
1332    (r) = __r0;								\
1333  } while (0)
1334
1335/* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1336   __udiv_w_sdiv (defined in libgcc or elsewhere).  */
1337#if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1338#define udiv_qrnnd(q, r, nh, nl, d) \
1339  do {									\
1340    USItype __r;							\
1341    (q) = __udiv_w_sdiv (&__r, nh, nl, d);				\
1342    (r) = __r;								\
1343  } while (0)
1344#endif
1345
1346/* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c.  */
1347#if !defined (udiv_qrnnd)
1348#define UDIV_NEEDS_NORMALIZATION 1
1349#define udiv_qrnnd __udiv_qrnnd_c
1350#endif
1351
1352#if !defined (count_leading_zeros)
1353#define count_leading_zeros(count, x) \
1354  do {									\
1355    UWtype __xr = (x);							\
1356    UWtype __a;								\
1357									\
1358    if (W_TYPE_SIZE <= 32)						\
1359      {									\
1360	__a = __xr < ((UWtype)1<<2*__BITS4)				\
1361	  ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4)			\
1362	  : (__xr < ((UWtype)1<<3*__BITS4) ?  2*__BITS4 : 3*__BITS4);	\
1363      }									\
1364    else								\
1365      {									\
1366	for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8)			\
1367	  if (((__xr >> __a) & 0xff) != 0)				\
1368	    break;							\
1369      }									\
1370									\
1371    (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a);		\
1372  } while (0)
1373#define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1374#endif
1375
1376#if !defined (count_trailing_zeros)
1377/* Define count_trailing_zeros using count_leading_zeros.  The latter might be
1378   defined in asm, but if it is not, the C version above is good enough.  */
1379#define count_trailing_zeros(count, x) \
1380  do {									\
1381    UWtype __ctz_x = (x);						\
1382    UWtype __ctz_c;							\
1383    count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x);			\
1384    (count) = W_TYPE_SIZE - 1 - __ctz_c;				\
1385  } while (0)
1386#endif
1387
1388#ifndef UDIV_NEEDS_NORMALIZATION
1389#define UDIV_NEEDS_NORMALIZATION 0
1390#endif
1391