1/* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
2   Copyright (C) 1991-2014 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
60extern 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 (__aarch64__)
126
127#if W_TYPE_SIZE == 32
128#define count_leading_zeros(COUNT, X)	((COUNT) = __builtin_clz (X))
129#define count_trailing_zeros(COUNT, X)   ((COUNT) = __builtin_ctz (X))
130#define COUNT_LEADING_ZEROS_0 32
131#endif /* W_TYPE_SIZE == 32 */
132
133#if W_TYPE_SIZE == 64
134#define count_leading_zeros(COUNT, X)	((COUNT) = __builtin_clzll (X))
135#define count_trailing_zeros(COUNT, X)   ((COUNT) = __builtin_ctzll (X))
136#define COUNT_LEADING_ZEROS_0 64
137#endif /* W_TYPE_SIZE == 64 */
138
139#endif /* __aarch64__ */
140
141#if defined (__alpha) && W_TYPE_SIZE == 64
142/* There is a bug in g++ before version 5 that
143   errors on __builtin_alpha_umulh.  */
144#if !defined(__cplusplus) || __GNUC__ >= 5
145#define umul_ppmm(ph, pl, m0, m1) \
146  do {									\
147    UDItype __m0 = (m0), __m1 = (m1);					\
148    (ph) = __builtin_alpha_umulh (__m0, __m1);				\
149    (pl) = __m0 * __m1;							\
150  } while (0)
151#define UMUL_TIME 46
152#endif /* !c++ */
153#ifndef LONGLONG_STANDALONE
154#define udiv_qrnnd(q, r, n1, n0, d) \
155  do { UDItype __r;							\
156    (q) = __udiv_qrnnd (&__r, (n1), (n0), (d));				\
157    (r) = __r;								\
158  } while (0)
159extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
160#define UDIV_TIME 220
161#endif /* LONGLONG_STANDALONE */
162#ifdef __alpha_cix__
163#define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clzl (X))
164#define count_trailing_zeros(COUNT,X)	((COUNT) = __builtin_ctzl (X))
165#define COUNT_LEADING_ZEROS_0 64
166#else
167#define count_leading_zeros(COUNT,X) \
168  do {									\
169    UDItype __xr = (X), __t, __a;					\
170    __t = __builtin_alpha_cmpbge (0, __xr);				\
171    __a = __clz_tab[__t ^ 0xff] - 1;					\
172    __t = __builtin_alpha_extbl (__xr, __a);				\
173    (COUNT) = 64 - (__clz_tab[__t] + __a*8);				\
174  } while (0)
175#define count_trailing_zeros(COUNT,X) \
176  do {									\
177    UDItype __xr = (X), __t, __a;					\
178    __t = __builtin_alpha_cmpbge (0, __xr);				\
179    __t = ~__t & -~__t;							\
180    __a = ((__t & 0xCC) != 0) * 2;					\
181    __a += ((__t & 0xF0) != 0) * 4;					\
182    __a += ((__t & 0xAA) != 0);						\
183    __t = __builtin_alpha_extbl (__xr, __a);				\
184    __a <<= 3;								\
185    __t &= -__t;							\
186    __a += ((__t & 0xCC) != 0) * 2;					\
187    __a += ((__t & 0xF0) != 0) * 4;					\
188    __a += ((__t & 0xAA) != 0);						\
189    (COUNT) = __a;							\
190  } while (0)
191#endif /* __alpha_cix__ */
192#endif /* __alpha */
193
194#if defined (__arc__) && W_TYPE_SIZE == 32
195#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
196  __asm__ ("add.f	%1, %4, %5\n\tadc	%0, %2, %3"		\
197	   : "=r" ((USItype) (sh)),					\
198	     "=&r" ((USItype) (sl))					\
199	   : "%r" ((USItype) (ah)),					\
200	     "rIJ" ((USItype) (bh)),					\
201	     "%r" ((USItype) (al)),					\
202	     "rIJ" ((USItype) (bl)))
203#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
204  __asm__ ("sub.f	%1, %4, %5\n\tsbc	%0, %2, %3"		\
205	   : "=r" ((USItype) (sh)),					\
206	     "=&r" ((USItype) (sl))					\
207	   : "r" ((USItype) (ah)),					\
208	     "rIJ" ((USItype) (bh)),					\
209	     "r" ((USItype) (al)),					\
210	     "rIJ" ((USItype) (bl)))
211
212#define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
213#ifdef __ARC_NORM__
214#define count_leading_zeros(count, x) \
215  do									\
216    {									\
217      SItype c_;							\
218									\
219      __asm__ ("norm.f\t%0,%1\n\tmov.mi\t%0,-1" : "=r" (c_) : "r" (x) : "cc");\
220      (count) = c_ + 1;							\
221    }									\
222  while (0)
223#define COUNT_LEADING_ZEROS_0 32
224#endif
225#endif
226
227#if defined (__arm__) && (defined (__thumb2__) || !defined (__thumb__)) \
228 && W_TYPE_SIZE == 32
229#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
230  __asm__ ("adds	%1, %4, %5\n\tadc	%0, %2, %3"		\
231	   : "=r" ((USItype) (sh)),					\
232	     "=&r" ((USItype) (sl))					\
233	   : "%r" ((USItype) (ah)),					\
234	     "rI" ((USItype) (bh)),					\
235	     "%r" ((USItype) (al)),					\
236	     "rI" ((USItype) (bl)) __CLOBBER_CC)
237#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
238  __asm__ ("subs	%1, %4, %5\n\tsbc	%0, %2, %3"		\
239	   : "=r" ((USItype) (sh)),					\
240	     "=&r" ((USItype) (sl))					\
241	   : "r" ((USItype) (ah)),					\
242	     "rI" ((USItype) (bh)),					\
243	     "r" ((USItype) (al)),					\
244	     "rI" ((USItype) (bl)) __CLOBBER_CC)
245# if defined(__ARM_ARCH_2__) || defined(__ARM_ARCH_2A__) \
246     || defined(__ARM_ARCH_3__)
247#  define umul_ppmm(xh, xl, a, b)					\
248  do {									\
249    register USItype __t0, __t1, __t2;					\
250    __asm__ ("%@ Inlined umul_ppmm\n"					\
251	   "	mov	%2, %5, lsr #16\n"				\
252	   "	mov	%0, %6, lsr #16\n"				\
253	   "	bic	%3, %5, %2, lsl #16\n"				\
254	   "	bic	%4, %6, %0, lsl #16\n"				\
255	   "	mul	%1, %3, %4\n"					\
256	   "	mul	%4, %2, %4\n"					\
257	   "	mul	%3, %0, %3\n"					\
258	   "	mul	%0, %2, %0\n"					\
259	   "	adds	%3, %4, %3\n"					\
260	   "	addcs	%0, %0, #65536\n"				\
261	   "	adds	%1, %1, %3, lsl #16\n"				\
262	   "	adc	%0, %0, %3, lsr #16"				\
263	   : "=&r" ((USItype) (xh)),					\
264	     "=r" ((USItype) (xl)),					\
265	     "=&r" (__t0), "=&r" (__t1), "=r" (__t2)			\
266	   : "r" ((USItype) (a)),					\
267	     "r" ((USItype) (b)) __CLOBBER_CC );			\
268  } while (0)
269#  define UMUL_TIME 20
270# else
271#  define umul_ppmm(xh, xl, a, b)					\
272  do {									\
273    /* Generate umull, under compiler control.  */			\
274    register UDItype __t0 = (UDItype)(USItype)(a) * (USItype)(b);	\
275    (xl) = (USItype)__t0;						\
276    (xh) = (USItype)(__t0 >> 32);					\
277  } while (0)
278#  define UMUL_TIME 3
279# endif
280# define UDIV_TIME 100
281#endif /* __arm__ */
282
283#if defined(__arm__)
284/* Let gcc decide how best to implement count_leading_zeros.  */
285#define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clz (X))
286#define count_trailing_zeros(COUNT,X)   ((COUNT) = __builtin_ctz (X))
287#define COUNT_LEADING_ZEROS_0 32
288#endif
289
290#if defined (__AVR__)
291
292#if W_TYPE_SIZE == 16
293#define count_leading_zeros(COUNT,X)  ((COUNT) = __builtin_clz (X))
294#define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X))
295#define COUNT_LEADING_ZEROS_0 16
296#endif /* W_TYPE_SIZE == 16 */
297
298#if W_TYPE_SIZE == 32
299#define count_leading_zeros(COUNT,X)  ((COUNT) = __builtin_clzl (X))
300#define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
301#define COUNT_LEADING_ZEROS_0 32
302#endif /* W_TYPE_SIZE == 32 */
303
304#if W_TYPE_SIZE == 64
305#define count_leading_zeros(COUNT,X)  ((COUNT) = __builtin_clzll (X))
306#define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzll (X))
307#define COUNT_LEADING_ZEROS_0 64
308#endif /* W_TYPE_SIZE == 64 */
309
310#endif /* defined (__AVR__) */
311
312#if defined (__CRIS__)
313
314#if __CRIS_arch_version >= 3
315#define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X))
316#define COUNT_LEADING_ZEROS_0 32
317#endif /* __CRIS_arch_version >= 3 */
318
319#if __CRIS_arch_version >= 8
320#define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X))
321#endif /* __CRIS_arch_version >= 8 */
322
323#if __CRIS_arch_version >= 10
324#define __umulsidi3(u,v) ((UDItype)(USItype) (u) * (UDItype)(USItype) (v))
325#else
326#define __umulsidi3 __umulsidi3
327extern UDItype __umulsidi3 (USItype, USItype);
328#endif /* __CRIS_arch_version >= 10 */
329
330#define umul_ppmm(w1, w0, u, v)		\
331  do {					\
332    UDItype __x = __umulsidi3 (u, v);	\
333    (w0) = (USItype) (__x);		\
334    (w1) = (USItype) (__x >> 32);	\
335  } while (0)
336
337/* FIXME: defining add_ssaaaa and sub_ddmmss should be advantageous for
338   DFmode ("double" intrinsics, avoiding two of the three insns handling
339   carry), but defining them as open-code C composing and doing the
340   operation in DImode (UDImode) shows that the DImode needs work:
341   register pressure from requiring neighboring registers and the
342   traffic to and from them come to dominate, in the 4.7 series.  */
343
344#endif /* defined (__CRIS__) */
345
346#if defined (__hppa) && W_TYPE_SIZE == 32
347#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
348  __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0"				\
349	   : "=r" ((USItype) (sh)),					\
350	     "=&r" ((USItype) (sl))					\
351	   : "%rM" ((USItype) (ah)),					\
352	     "rM" ((USItype) (bh)),					\
353	     "%rM" ((USItype) (al)),					\
354	     "rM" ((USItype) (bl)))
355#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
356  __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0"				\
357	   : "=r" ((USItype) (sh)),					\
358	     "=&r" ((USItype) (sl))					\
359	   : "rM" ((USItype) (ah)),					\
360	     "rM" ((USItype) (bh)),					\
361	     "rM" ((USItype) (al)),					\
362	     "rM" ((USItype) (bl)))
363#if defined (_PA_RISC1_1)
364#define umul_ppmm(w1, w0, u, v) \
365  do {									\
366    union								\
367      {									\
368	UDItype __f;							\
369	struct {USItype __w1, __w0;} __w1w0;				\
370      } __t;								\
371    __asm__ ("xmpyu %1,%2,%0"						\
372	     : "=x" (__t.__f)						\
373	     : "x" ((USItype) (u)),					\
374	       "x" ((USItype) (v)));					\
375    (w1) = __t.__w1w0.__w1;						\
376    (w0) = __t.__w1w0.__w0;						\
377     } while (0)
378#define UMUL_TIME 8
379#else
380#define UMUL_TIME 30
381#endif
382#define UDIV_TIME 40
383#define count_leading_zeros(count, x) \
384  do {									\
385    USItype __tmp;							\
386    __asm__ (								\
387       "ldi		1,%0\n"						\
388"	extru,=		%1,15,16,%%r0		; Bits 31..16 zero?\n"	\
389"	extru,tr	%1,15,16,%1		; No.  Shift down, skip add.\n"\
390"	ldo		16(%0),%0		; Yes.  Perform add.\n"	\
391"	extru,=		%1,23,8,%%r0		; Bits 15..8 zero?\n"	\
392"	extru,tr	%1,23,8,%1		; No.  Shift down, skip add.\n"\
393"	ldo		8(%0),%0		; Yes.  Perform add.\n"	\
394"	extru,=		%1,27,4,%%r0		; Bits 7..4 zero?\n"	\
395"	extru,tr	%1,27,4,%1		; No.  Shift down, skip add.\n"\
396"	ldo		4(%0),%0		; Yes.  Perform add.\n"	\
397"	extru,=		%1,29,2,%%r0		; Bits 3..2 zero?\n"	\
398"	extru,tr	%1,29,2,%1		; No.  Shift down, skip add.\n"\
399"	ldo		2(%0),%0		; Yes.  Perform add.\n"	\
400"	extru		%1,30,1,%1		; Extract bit 1.\n"	\
401"	sub		%0,%1,%0		; Subtract it.\n"	\
402	: "=r" (count), "=r" (__tmp) : "1" (x));			\
403  } while (0)
404#endif
405
406#if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
407#if !defined (__zarch__)
408#define smul_ppmm(xh, xl, m0, m1) \
409  do {									\
410    union {DItype __ll;							\
411	   struct {USItype __h, __l;} __i;				\
412	  } __x;							\
413    __asm__ ("lr %N0,%1\n\tmr %0,%2"					\
414	     : "=&r" (__x.__ll)						\
415	     : "r" (m0), "r" (m1));					\
416    (xh) = __x.__i.__h; (xl) = __x.__i.__l;				\
417  } while (0)
418#define sdiv_qrnnd(q, r, n1, n0, d) \
419  do {									\
420    union {DItype __ll;							\
421	   struct {USItype __h, __l;} __i;				\
422	  } __x;							\
423    __x.__i.__h = n1; __x.__i.__l = n0;					\
424    __asm__ ("dr %0,%2"							\
425	     : "=r" (__x.__ll)						\
426	     : "0" (__x.__ll), "r" (d));				\
427    (q) = __x.__i.__l; (r) = __x.__i.__h;				\
428  } while (0)
429#else
430#define smul_ppmm(xh, xl, m0, m1) \
431  do {                                                                  \
432    register SItype __r0 __asm__ ("0");					\
433    register SItype __r1 __asm__ ("1") = (m0);				\
434									\
435    __asm__ ("mr\t%%r0,%3"                                              \
436	     : "=r" (__r0), "=r" (__r1)					\
437	     : "r"  (__r1),  "r" (m1));					\
438    (xh) = __r0; (xl) = __r1;						\
439  } while (0)
440
441#define sdiv_qrnnd(q, r, n1, n0, d) \
442  do {									\
443    register SItype __r0 __asm__ ("0") = (n1);				\
444    register SItype __r1 __asm__ ("1") = (n0);				\
445									\
446    __asm__ ("dr\t%%r0,%4"                                              \
447	     : "=r" (__r0), "=r" (__r1)					\
448	     : "r" (__r0), "r" (__r1), "r" (d));			\
449    (q) = __r1; (r) = __r0;						\
450  } while (0)
451#endif /* __zarch__ */
452#endif
453
454#if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
455#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
456  __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}"		\
457	   : "=r" ((USItype) (sh)),					\
458	     "=&r" ((USItype) (sl))					\
459	   : "%0" ((USItype) (ah)),					\
460	     "g" ((USItype) (bh)),					\
461	     "%1" ((USItype) (al)),					\
462	     "g" ((USItype) (bl)))
463#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
464  __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}"		\
465	   : "=r" ((USItype) (sh)),					\
466	     "=&r" ((USItype) (sl))					\
467	   : "0" ((USItype) (ah)),					\
468	     "g" ((USItype) (bh)),					\
469	     "1" ((USItype) (al)),					\
470	     "g" ((USItype) (bl)))
471#define umul_ppmm(w1, w0, u, v) \
472  __asm__ ("mul{l} %3"							\
473	   : "=a" ((USItype) (w0)),					\
474	     "=d" ((USItype) (w1))					\
475	   : "%0" ((USItype) (u)),					\
476	     "rm" ((USItype) (v)))
477#define udiv_qrnnd(q, r, n1, n0, dv) \
478  __asm__ ("div{l} %4"							\
479	   : "=a" ((USItype) (q)),					\
480	     "=d" ((USItype) (r))					\
481	   : "0" ((USItype) (n0)),					\
482	     "1" ((USItype) (n1)),					\
483	     "rm" ((USItype) (dv)))
484#define count_leading_zeros(count, x)	((count) = __builtin_clz (x))
485#define count_trailing_zeros(count, x)	((count) = __builtin_ctz (x))
486#define UMUL_TIME 40
487#define UDIV_TIME 40
488#endif /* 80x86 */
489
490#if defined (__x86_64__) && W_TYPE_SIZE == 64
491#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
492  __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}"		\
493	   : "=r" ((UDItype) (sh)),					\
494	     "=&r" ((UDItype) (sl))					\
495	   : "%0" ((UDItype) (ah)),					\
496	     "rme" ((UDItype) (bh)),					\
497	     "%1" ((UDItype) (al)),					\
498	     "rme" ((UDItype) (bl)))
499#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
500  __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}"		\
501	   : "=r" ((UDItype) (sh)),					\
502	     "=&r" ((UDItype) (sl))					\
503	   : "0" ((UDItype) (ah)),					\
504	     "rme" ((UDItype) (bh)),					\
505	     "1" ((UDItype) (al)),					\
506	     "rme" ((UDItype) (bl)))
507#define umul_ppmm(w1, w0, u, v) \
508  __asm__ ("mul{q} %3"							\
509	   : "=a" ((UDItype) (w0)),					\
510	     "=d" ((UDItype) (w1))					\
511	   : "%0" ((UDItype) (u)),					\
512	     "rm" ((UDItype) (v)))
513#define udiv_qrnnd(q, r, n1, n0, dv) \
514  __asm__ ("div{q} %4"							\
515	   : "=a" ((UDItype) (q)),					\
516	     "=d" ((UDItype) (r))					\
517	   : "0" ((UDItype) (n0)),					\
518	     "1" ((UDItype) (n1)),					\
519	     "rm" ((UDItype) (dv)))
520#define count_leading_zeros(count, x)	((count) = __builtin_clzll (x))
521#define count_trailing_zeros(count, x)	((count) = __builtin_ctzll (x))
522#define UMUL_TIME 40
523#define UDIV_TIME 40
524#endif /* x86_64 */
525
526#if defined (__i960__) && W_TYPE_SIZE == 32
527#define umul_ppmm(w1, w0, u, v) \
528  ({union {UDItype __ll;						\
529	   struct {USItype __l, __h;} __i;				\
530	  } __xx;							\
531  __asm__ ("emul	%2,%1,%0"					\
532	   : "=d" (__xx.__ll)						\
533	   : "%dI" ((USItype) (u)),					\
534	     "dI" ((USItype) (v)));					\
535  (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
536#define __umulsidi3(u, v) \
537  ({UDItype __w;							\
538    __asm__ ("emul	%2,%1,%0"					\
539	     : "=d" (__w)						\
540	     : "%dI" ((USItype) (u)),					\
541	       "dI" ((USItype) (v)));					\
542    __w; })
543#endif /* __i960__ */
544
545#if defined (__ia64) && W_TYPE_SIZE == 64
546/* This form encourages gcc (pre-release 3.4 at least) to emit predicated
547   "sub r=r,r" and "sub r=r,r,1", giving a 2 cycle latency.  The generic
548   code using "al<bl" arithmetically comes out making an actual 0 or 1 in a
549   register, which takes an extra cycle.  */
550#define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
551  do {									\
552    UWtype __x;								\
553    __x = (al) - (bl);							\
554    if ((al) < (bl))							\
555      (sh) = (ah) - (bh) - 1;						\
556    else								\
557      (sh) = (ah) - (bh);						\
558    (sl) = __x;								\
559  } while (0)
560
561/* Do both product parts in assembly, since that gives better code with
562   all gcc versions.  Some callers will just use the upper part, and in
563   that situation we waste an instruction, but not any cycles.  */
564#define umul_ppmm(ph, pl, m0, m1)					\
565  __asm__ ("xma.hu %0 = %2, %3, f0\n\txma.l %1 = %2, %3, f0"		\
566	   : "=&f" (ph), "=f" (pl)					\
567	   : "f" (m0), "f" (m1))
568#define count_leading_zeros(count, x)					\
569  do {									\
570    UWtype _x = (x), _y, _a, _c;					\
571    __asm__ ("mux1 %0 = %1, @rev" : "=r" (_y) : "r" (_x));		\
572    __asm__ ("czx1.l %0 = %1" : "=r" (_a) : "r" (-_y | _y));		\
573    _c = (_a - 1) << 3;							\
574    _x >>= _c;								\
575    if (_x >= 1 << 4)							\
576      _x >>= 4, _c += 4;						\
577    if (_x >= 1 << 2)							\
578      _x >>= 2, _c += 2;						\
579    _c += _x >> 1;							\
580    (count) =  W_TYPE_SIZE - 1 - _c;					\
581  } while (0)
582/* similar to what gcc does for __builtin_ffs, but 0 based rather than 1
583   based, and we don't need a special case for x==0 here */
584#define count_trailing_zeros(count, x)					\
585  do {									\
586    UWtype __ctz_x = (x);						\
587    __asm__ ("popcnt %0 = %1"						\
588	     : "=r" (count)						\
589	     : "r" ((__ctz_x-1) & ~__ctz_x));				\
590  } while (0)
591#define UMUL_TIME 14
592#endif
593
594#if defined (__M32R__) && W_TYPE_SIZE == 32
595#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
596  /* The cmp clears the condition bit.  */ \
597  __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3"			\
598	   : "=r" ((USItype) (sh)),					\
599	     "=&r" ((USItype) (sl))					\
600	   : "0" ((USItype) (ah)),					\
601	     "r" ((USItype) (bh)),					\
602	     "1" ((USItype) (al)),					\
603	     "r" ((USItype) (bl))					\
604	   : "cbit")
605#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
606  /* The cmp clears the condition bit.  */ \
607  __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3"			\
608	   : "=r" ((USItype) (sh)),					\
609	     "=&r" ((USItype) (sl))					\
610	   : "0" ((USItype) (ah)),					\
611	     "r" ((USItype) (bh)),					\
612	     "1" ((USItype) (al)),					\
613	     "r" ((USItype) (bl))					\
614	   : "cbit")
615#endif /* __M32R__ */
616
617#if defined (__mc68000__) && W_TYPE_SIZE == 32
618#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
619  __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0"				\
620	   : "=d" ((USItype) (sh)),					\
621	     "=&d" ((USItype) (sl))					\
622	   : "%0" ((USItype) (ah)),					\
623	     "d" ((USItype) (bh)),					\
624	     "%1" ((USItype) (al)),					\
625	     "g" ((USItype) (bl)))
626#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
627  __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0"				\
628	   : "=d" ((USItype) (sh)),					\
629	     "=&d" ((USItype) (sl))					\
630	   : "0" ((USItype) (ah)),					\
631	     "d" ((USItype) (bh)),					\
632	     "1" ((USItype) (al)),					\
633	     "g" ((USItype) (bl)))
634
635/* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r.  */
636#if (defined (__mc68020__) && !defined (__mc68060__))
637#define umul_ppmm(w1, w0, u, v) \
638  __asm__ ("mulu%.l %3,%1:%0"						\
639	   : "=d" ((USItype) (w0)),					\
640	     "=d" ((USItype) (w1))					\
641	   : "%0" ((USItype) (u)),					\
642	     "dmi" ((USItype) (v)))
643#define UMUL_TIME 45
644#define udiv_qrnnd(q, r, n1, n0, d) \
645  __asm__ ("divu%.l %4,%1:%0"						\
646	   : "=d" ((USItype) (q)),					\
647	     "=d" ((USItype) (r))					\
648	   : "0" ((USItype) (n0)),					\
649	     "1" ((USItype) (n1)),					\
650	     "dmi" ((USItype) (d)))
651#define UDIV_TIME 90
652#define sdiv_qrnnd(q, r, n1, n0, d) \
653  __asm__ ("divs%.l %4,%1:%0"						\
654	   : "=d" ((USItype) (q)),					\
655	     "=d" ((USItype) (r))					\
656	   : "0" ((USItype) (n0)),					\
657	     "1" ((USItype) (n1)),					\
658	     "dmi" ((USItype) (d)))
659
660#elif defined (__mcoldfire__) /* not mc68020 */
661
662#define umul_ppmm(xh, xl, a, b) \
663  __asm__ ("| Inlined umul_ppmm\n"					\
664	   "	move%.l	%2,%/d0\n"					\
665	   "	move%.l	%3,%/d1\n"					\
666	   "	move%.l	%/d0,%/d2\n"					\
667	   "	swap	%/d0\n"						\
668	   "	move%.l	%/d1,%/d3\n"					\
669	   "	swap	%/d1\n"						\
670	   "	move%.w	%/d2,%/d4\n"					\
671	   "	mulu	%/d3,%/d4\n"					\
672	   "	mulu	%/d1,%/d2\n"					\
673	   "	mulu	%/d0,%/d3\n"					\
674	   "	mulu	%/d0,%/d1\n"					\
675	   "	move%.l	%/d4,%/d0\n"					\
676	   "	clr%.w	%/d0\n"						\
677	   "	swap	%/d0\n"						\
678	   "	add%.l	%/d0,%/d2\n"					\
679	   "	add%.l	%/d3,%/d2\n"					\
680	   "	jcc	1f\n"						\
681	   "	add%.l	%#65536,%/d1\n"					\
682	   "1:	swap	%/d2\n"						\
683	   "	moveq	%#0,%/d0\n"					\
684	   "	move%.w	%/d2,%/d0\n"					\
685	   "	move%.w	%/d4,%/d2\n"					\
686	   "	move%.l	%/d2,%1\n"					\
687	   "	add%.l	%/d1,%/d0\n"					\
688	   "	move%.l	%/d0,%0"					\
689	   : "=g" ((USItype) (xh)),					\
690	     "=g" ((USItype) (xl))					\
691	   : "g" ((USItype) (a)),					\
692	     "g" ((USItype) (b))					\
693	   : "d0", "d1", "d2", "d3", "d4")
694#define UMUL_TIME 100
695#define UDIV_TIME 400
696#else /* not ColdFire */
697/* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX.  */
698#define umul_ppmm(xh, xl, a, b) \
699  __asm__ ("| Inlined umul_ppmm\n"					\
700	   "	move%.l	%2,%/d0\n"					\
701	   "	move%.l	%3,%/d1\n"					\
702	   "	move%.l	%/d0,%/d2\n"					\
703	   "	swap	%/d0\n"						\
704	   "	move%.l	%/d1,%/d3\n"					\
705	   "	swap	%/d1\n"						\
706	   "	move%.w	%/d2,%/d4\n"					\
707	   "	mulu	%/d3,%/d4\n"					\
708	   "	mulu	%/d1,%/d2\n"					\
709	   "	mulu	%/d0,%/d3\n"					\
710	   "	mulu	%/d0,%/d1\n"					\
711	   "	move%.l	%/d4,%/d0\n"					\
712	   "	eor%.w	%/d0,%/d0\n"					\
713	   "	swap	%/d0\n"						\
714	   "	add%.l	%/d0,%/d2\n"					\
715	   "	add%.l	%/d3,%/d2\n"					\
716	   "	jcc	1f\n"						\
717	   "	add%.l	%#65536,%/d1\n"					\
718	   "1:	swap	%/d2\n"						\
719	   "	moveq	%#0,%/d0\n"					\
720	   "	move%.w	%/d2,%/d0\n"					\
721	   "	move%.w	%/d4,%/d2\n"					\
722	   "	move%.l	%/d2,%1\n"					\
723	   "	add%.l	%/d1,%/d0\n"					\
724	   "	move%.l	%/d0,%0"					\
725	   : "=g" ((USItype) (xh)),					\
726	     "=g" ((USItype) (xl))					\
727	   : "g" ((USItype) (a)),					\
728	     "g" ((USItype) (b))					\
729	   : "d0", "d1", "d2", "d3", "d4")
730#define UMUL_TIME 100
731#define UDIV_TIME 400
732
733#endif /* not mc68020 */
734
735/* The '020, '030, '040 and '060 have bitfield insns.
736   cpu32 disguises as a 68020, but lacks them.  */
737#if defined (__mc68020__) && !defined (__mcpu32__)
738#define count_leading_zeros(count, x) \
739  __asm__ ("bfffo %1{%b2:%b2},%0"					\
740	   : "=d" ((USItype) (count))					\
741	   : "od" ((USItype) (x)), "n" (0))
742/* Some ColdFire architectures have a ff1 instruction supported via
743   __builtin_clz. */
744#elif defined (__mcfisaaplus__) || defined (__mcfisac__)
745#define count_leading_zeros(count,x) ((count) = __builtin_clz (x))
746#define COUNT_LEADING_ZEROS_0 32
747#endif
748#endif /* mc68000 */
749
750#if defined (__m88000__) && W_TYPE_SIZE == 32
751#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
752  __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3"			\
753	   : "=r" ((USItype) (sh)),					\
754	     "=&r" ((USItype) (sl))					\
755	   : "%rJ" ((USItype) (ah)),					\
756	     "rJ" ((USItype) (bh)),					\
757	     "%rJ" ((USItype) (al)),					\
758	     "rJ" ((USItype) (bl)))
759#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
760  __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3"			\
761	   : "=r" ((USItype) (sh)),					\
762	     "=&r" ((USItype) (sl))					\
763	   : "rJ" ((USItype) (ah)),					\
764	     "rJ" ((USItype) (bh)),					\
765	     "rJ" ((USItype) (al)),					\
766	     "rJ" ((USItype) (bl)))
767#define count_leading_zeros(count, x) \
768  do {									\
769    USItype __cbtmp;							\
770    __asm__ ("ff1 %0,%1"						\
771	     : "=r" (__cbtmp)						\
772	     : "r" ((USItype) (x)));					\
773    (count) = __cbtmp ^ 31;						\
774  } while (0)
775#define COUNT_LEADING_ZEROS_0 63 /* sic */
776#if defined (__mc88110__)
777#define umul_ppmm(wh, wl, u, v) \
778  do {									\
779    union {UDItype __ll;						\
780	   struct {USItype __h, __l;} __i;				\
781	  } __xx;							\
782    __asm__ ("mulu.d	%0,%1,%2"					\
783	     : "=r" (__xx.__ll)						\
784	     : "r" ((USItype) (u)),					\
785	       "r" ((USItype) (v)));					\
786    (wh) = __xx.__i.__h;						\
787    (wl) = __xx.__i.__l;						\
788  } while (0)
789#define udiv_qrnnd(q, r, n1, n0, d) \
790  ({union {UDItype __ll;						\
791	   struct {USItype __h, __l;} __i;				\
792	  } __xx;							\
793  USItype __q;								\
794  __xx.__i.__h = (n1); __xx.__i.__l = (n0);				\
795  __asm__ ("divu.d %0,%1,%2"						\
796	   : "=r" (__q)							\
797	   : "r" (__xx.__ll),						\
798	     "r" ((USItype) (d)));					\
799  (r) = (n0) - __q * (d); (q) = __q; })
800#define UMUL_TIME 5
801#define UDIV_TIME 25
802#else
803#define UMUL_TIME 17
804#define UDIV_TIME 150
805#endif /* __mc88110__ */
806#endif /* __m88000__ */
807
808#if defined (__mn10300__)
809# if defined (__AM33__)
810#  define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clz (X))
811#  define umul_ppmm(w1, w0, u, v)		\
812    asm("mulu %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
813#  define smul_ppmm(w1, w0, u, v)		\
814    asm("mul %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v))
815# else
816#  define umul_ppmm(w1, w0, u, v)		\
817    asm("nop; nop; mulu %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
818#  define smul_ppmm(w1, w0, u, v)		\
819    asm("nop; nop; mul %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v))
820# endif
821# define add_ssaaaa(sh, sl, ah, al, bh, bl)	\
822  do {						\
823    DWunion __s, __a, __b;			\
824    __a.s.low = (al); __a.s.high = (ah);	\
825    __b.s.low = (bl); __b.s.high = (bh);	\
826    __s.ll = __a.ll + __b.ll;			\
827    (sl) = __s.s.low; (sh) = __s.s.high;	\
828  } while (0)
829# define sub_ddmmss(sh, sl, ah, al, bh, bl)	\
830  do {						\
831    DWunion __s, __a, __b;			\
832    __a.s.low = (al); __a.s.high = (ah);	\
833    __b.s.low = (bl); __b.s.high = (bh);	\
834    __s.ll = __a.ll - __b.ll;			\
835    (sl) = __s.s.low; (sh) = __s.s.high;	\
836  } while (0)
837# define udiv_qrnnd(q, r, nh, nl, d)		\
838  asm("divu %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
839# define sdiv_qrnnd(q, r, nh, nl, d)		\
840  asm("div %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh))
841# define UMUL_TIME 3
842# define UDIV_TIME 38
843#endif
844
845#if defined (__mips__) && W_TYPE_SIZE == 32
846#define umul_ppmm(w1, w0, u, v)						\
847  do {									\
848    UDItype __x = (UDItype) (USItype) (u) * (USItype) (v);		\
849    (w1) = (USItype) (__x >> 32);					\
850    (w0) = (USItype) (__x);						\
851  } while (0)
852#define UMUL_TIME 10
853#define UDIV_TIME 100
854
855#if (__mips == 32 || __mips == 64) && ! defined (__mips16)
856#define count_leading_zeros(COUNT,X)	((COUNT) = __builtin_clz (X))
857#define COUNT_LEADING_ZEROS_0 32
858#endif
859#endif /* __mips__ */
860
861#if defined (__ns32000__) && W_TYPE_SIZE == 32
862#define umul_ppmm(w1, w0, u, v) \
863  ({union {UDItype __ll;						\
864	   struct {USItype __l, __h;} __i;				\
865	  } __xx;							\
866  __asm__ ("meid %2,%0"							\
867	   : "=g" (__xx.__ll)						\
868	   : "%0" ((USItype) (u)),					\
869	     "g" ((USItype) (v)));					\
870  (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
871#define __umulsidi3(u, v) \
872  ({UDItype __w;							\
873    __asm__ ("meid %2,%0"						\
874	     : "=g" (__w)						\
875	     : "%0" ((USItype) (u)),					\
876	       "g" ((USItype) (v)));					\
877    __w; })
878#define udiv_qrnnd(q, r, n1, n0, d) \
879  ({union {UDItype __ll;						\
880	   struct {USItype __l, __h;} __i;				\
881	  } __xx;							\
882  __xx.__i.__h = (n1); __xx.__i.__l = (n0);				\
883  __asm__ ("deid %2,%0"							\
884	   : "=g" (__xx.__ll)						\
885	   : "0" (__xx.__ll),						\
886	     "g" ((USItype) (d)));					\
887  (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
888#define count_trailing_zeros(count,x) \
889  do {									\
890    __asm__ ("ffsd     %2,%0"						\
891	    : "=r" ((USItype) (count))					\
892	    : "0" ((USItype) 0),					\
893	      "r" ((USItype) (x)));					\
894  } while (0)
895#endif /* __ns32000__ */
896
897/* FIXME: We should test _IBMR2 here when we add assembly support for the
898   system vendor compilers.
899   FIXME: What's needed for gcc PowerPC VxWorks?  __vxworks__ is not good
900   enough, since that hits ARM and m68k too.  */
901#if (defined (_ARCH_PPC)	/* AIX */				\
902     || defined (__powerpc__)	/* gcc */				\
903     || defined (__POWERPC__)	/* BEOS */				\
904     || defined (__ppc__)	/* Darwin */				\
905     || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */    \
906     || (defined (PPC) && defined (CPU_FAMILY)    /* VxWorks */               \
907	 && CPU_FAMILY == PPC)                                                \
908     ) && W_TYPE_SIZE == 32
909#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
910  do {									\
911    if (__builtin_constant_p (bh) && (bh) == 0)				\
912      __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2"		\
913	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
914    else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0)		\
915      __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2"		\
916	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
917    else								\
918      __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3"		\
919	     : "=r" (sh), "=&r" (sl)					\
920	     : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl));		\
921  } while (0)
922#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
923  do {									\
924    if (__builtin_constant_p (ah) && (ah) == 0)				\
925      __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2"	\
926	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
927    else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0)		\
928      __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2"	\
929	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
930    else if (__builtin_constant_p (bh) && (bh) == 0)			\
931      __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2"		\
932	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
933    else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0)		\
934      __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2"		\
935	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
936    else								\
937      __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2"	\
938	       : "=r" (sh), "=&r" (sl)					\
939	       : "r" (ah), "r" (bh), "rI" (al), "r" (bl));		\
940  } while (0)
941#define count_leading_zeros(count, x) \
942  __asm__ ("cntlzw %0,%1" : "=r" (count) : "r" (x))
943#define COUNT_LEADING_ZEROS_0 32
944#if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
945  || defined (__ppc__)                                                    \
946  || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */       \
947  || (defined (PPC) && defined (CPU_FAMILY)    /* VxWorks */                  \
948	 && CPU_FAMILY == PPC)
949#define umul_ppmm(ph, pl, m0, m1) \
950  do {									\
951    USItype __m0 = (m0), __m1 = (m1);					\
952    __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
953    (pl) = __m0 * __m1;							\
954  } while (0)
955#define UMUL_TIME 15
956#define smul_ppmm(ph, pl, m0, m1) \
957  do {									\
958    SItype __m0 = (m0), __m1 = (m1);					\
959    __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
960    (pl) = __m0 * __m1;							\
961  } while (0)
962#define SMUL_TIME 14
963#define UDIV_TIME 120
964#endif
965#endif /* 32-bit POWER architecture variants.  */
966
967/* We should test _IBMR2 here when we add assembly support for the system
968   vendor compilers.  */
969#if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
970#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
971  do {									\
972    if (__builtin_constant_p (bh) && (bh) == 0)				\
973      __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2"		\
974	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
975    else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0)		\
976      __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2"		\
977	     : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
978    else								\
979      __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3"		\
980	     : "=r" (sh), "=&r" (sl)					\
981	     : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl));		\
982  } while (0)
983#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
984  do {									\
985    if (__builtin_constant_p (ah) && (ah) == 0)				\
986      __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2"	\
987	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
988    else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0)		\
989      __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2"	\
990	       : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
991    else if (__builtin_constant_p (bh) && (bh) == 0)			\
992      __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2"		\
993	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
994    else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0)		\
995      __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2"		\
996	       : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
997    else								\
998      __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2"	\
999	       : "=r" (sh), "=&r" (sl)					\
1000	       : "r" (ah), "r" (bh), "rI" (al), "r" (bl));		\
1001  } while (0)
1002#define count_leading_zeros(count, x) \
1003  __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
1004#define COUNT_LEADING_ZEROS_0 64
1005#define umul_ppmm(ph, pl, m0, m1) \
1006  do {									\
1007    UDItype __m0 = (m0), __m1 = (m1);					\
1008    __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
1009    (pl) = __m0 * __m1;							\
1010  } while (0)
1011#define UMUL_TIME 15
1012#define smul_ppmm(ph, pl, m0, m1) \
1013  do {									\
1014    DItype __m0 = (m0), __m1 = (m1);					\
1015    __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1));	\
1016    (pl) = __m0 * __m1;							\
1017  } while (0)
1018#define SMUL_TIME 14  /* ??? */
1019#define UDIV_TIME 120 /* ??? */
1020#endif /* 64-bit PowerPC.  */
1021
1022#if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
1023#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1024  __asm__ ("a %1,%5\n\tae %0,%3"					\
1025	   : "=r" ((USItype) (sh)),					\
1026	     "=&r" ((USItype) (sl))					\
1027	   : "%0" ((USItype) (ah)),					\
1028	     "r" ((USItype) (bh)),					\
1029	     "%1" ((USItype) (al)),					\
1030	     "r" ((USItype) (bl)))
1031#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1032  __asm__ ("s %1,%5\n\tse %0,%3"					\
1033	   : "=r" ((USItype) (sh)),					\
1034	     "=&r" ((USItype) (sl))					\
1035	   : "0" ((USItype) (ah)),					\
1036	     "r" ((USItype) (bh)),					\
1037	     "1" ((USItype) (al)),					\
1038	     "r" ((USItype) (bl)))
1039#define umul_ppmm(ph, pl, m0, m1) \
1040  do {									\
1041    USItype __m0 = (m0), __m1 = (m1);					\
1042    __asm__ (								\
1043       "s	r2,r2\n"						\
1044"	mts	r10,%2\n"						\
1045"	m	r2,%3\n"						\
1046"	m	r2,%3\n"						\
1047"	m	r2,%3\n"						\
1048"	m	r2,%3\n"						\
1049"	m	r2,%3\n"						\
1050"	m	r2,%3\n"						\
1051"	m	r2,%3\n"						\
1052"	m	r2,%3\n"						\
1053"	m	r2,%3\n"						\
1054"	m	r2,%3\n"						\
1055"	m	r2,%3\n"						\
1056"	m	r2,%3\n"						\
1057"	m	r2,%3\n"						\
1058"	m	r2,%3\n"						\
1059"	m	r2,%3\n"						\
1060"	m	r2,%3\n"						\
1061"	cas	%0,r2,r0\n"						\
1062"	mfs	r10,%1"							\
1063	     : "=r" ((USItype) (ph)),					\
1064	       "=r" ((USItype) (pl))					\
1065	     : "%r" (__m0),						\
1066		"r" (__m1)						\
1067	     : "r2");							\
1068    (ph) += ((((SItype) __m0 >> 31) & __m1)				\
1069	     + (((SItype) __m1 >> 31) & __m0));				\
1070  } while (0)
1071#define UMUL_TIME 20
1072#define UDIV_TIME 200
1073#define count_leading_zeros(count, x) \
1074  do {									\
1075    if ((x) >= 0x10000)							\
1076      __asm__ ("clz	%0,%1"						\
1077	       : "=r" ((USItype) (count))				\
1078	       : "r" ((USItype) (x) >> 16));				\
1079    else								\
1080      {									\
1081	__asm__ ("clz	%0,%1"						\
1082		 : "=r" ((USItype) (count))				\
1083		 : "r" ((USItype) (x)));					\
1084	(count) += 16;							\
1085      }									\
1086  } while (0)
1087#endif
1088
1089#if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32
1090#ifndef __sh1__
1091#define umul_ppmm(w1, w0, u, v) \
1092  __asm__ (								\
1093       "dmulu.l	%2,%3\n\tsts%M1	macl,%1\n\tsts%M0	mach,%0"	\
1094	   : "=r<" ((USItype)(w1)),					\
1095	     "=r<" ((USItype)(w0))					\
1096	   : "r" ((USItype)(u)),					\
1097	     "r" ((USItype)(v))						\
1098	   : "macl", "mach")
1099#define UMUL_TIME 5
1100#endif
1101
1102/* This is the same algorithm as __udiv_qrnnd_c.  */
1103#define UDIV_NEEDS_NORMALIZATION 1
1104
1105#define udiv_qrnnd(q, r, n1, n0, d) \
1106  do {									\
1107    extern UWtype __udiv_qrnnd_16 (UWtype, UWtype)			\
1108			__attribute__ ((visibility ("hidden")));	\
1109    /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */	\
1110    __asm__ (								\
1111	"mov%M4 %4,r5\n"						\
1112"	swap.w %3,r4\n"							\
1113"	swap.w r5,r6\n"							\
1114"	jsr @%5\n"							\
1115"	shll16 r6\n"							\
1116"	swap.w r4,r4\n"							\
1117"	jsr @%5\n"							\
1118"	swap.w r1,%0\n"							\
1119"	or r1,%0"							\
1120	: "=r" (q), "=&z" (r)						\
1121	: "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16)		\
1122	: "r1", "r2", "r4", "r5", "r6", "pr", "t");			\
1123  } while (0)
1124
1125#define UDIV_TIME 80
1126
1127#define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
1128  __asm__ ("clrt;subc %5,%1; subc %4,%0"				\
1129	   : "=r" (sh), "=r" (sl)					\
1130	   : "0" (ah), "1" (al), "r" (bh), "r" (bl) : "t")
1131
1132#endif /* __sh__ */
1133
1134#if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
1135#define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1136#define count_leading_zeros(count, x) \
1137  do									\
1138    {									\
1139      UDItype x_ = (USItype)(x);					\
1140      SItype c_;							\
1141									\
1142      __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_));			\
1143      (count) = c_ - 31;						\
1144    }									\
1145  while (0)
1146#define COUNT_LEADING_ZEROS_0 32
1147#endif
1148
1149#if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
1150    && W_TYPE_SIZE == 32
1151#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1152  __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0"				\
1153	   : "=r" ((USItype) (sh)),					\
1154	     "=&r" ((USItype) (sl))					\
1155	   : "%rJ" ((USItype) (ah)),					\
1156	     "rI" ((USItype) (bh)),					\
1157	     "%rJ" ((USItype) (al)),					\
1158	     "rI" ((USItype) (bl))					\
1159	   __CLOBBER_CC)
1160#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1161  __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0"				\
1162	   : "=r" ((USItype) (sh)),					\
1163	     "=&r" ((USItype) (sl))					\
1164	   : "rJ" ((USItype) (ah)),					\
1165	     "rI" ((USItype) (bh)),					\
1166	     "rJ" ((USItype) (al)),					\
1167	     "rI" ((USItype) (bl))					\
1168	   __CLOBBER_CC)
1169#if defined (__sparc_v9__)
1170#define umul_ppmm(w1, w0, u, v) \
1171  do {									\
1172    register USItype __g1 asm ("g1");					\
1173    __asm__ ("umul\t%2,%3,%1\n\t"					\
1174	     "srlx\t%1, 32, %0"						\
1175	     : "=r" ((USItype) (w1)),					\
1176	       "=r" (__g1)						\
1177	     : "r" ((USItype) (u)),					\
1178	       "r" ((USItype) (v)));					\
1179    (w0) = __g1;							\
1180  } while (0)
1181#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1182  __asm__ ("mov\t%2,%%y\n\t"						\
1183	   "udiv\t%3,%4,%0\n\t"						\
1184	   "umul\t%0,%4,%1\n\t"						\
1185	   "sub\t%3,%1,%1"						\
1186	   : "=&r" ((USItype) (__q)),					\
1187	     "=&r" ((USItype) (__r))					\
1188	   : "r" ((USItype) (__n1)),					\
1189	     "r" ((USItype) (__n0)),					\
1190	     "r" ((USItype) (__d)))
1191#else
1192#if defined (__sparc_v8__)
1193#define umul_ppmm(w1, w0, u, v) \
1194  __asm__ ("umul %2,%3,%1;rd %%y,%0"					\
1195	   : "=r" ((USItype) (w1)),					\
1196	     "=r" ((USItype) (w0))					\
1197	   : "r" ((USItype) (u)),					\
1198	     "r" ((USItype) (v)))
1199#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1200  __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
1201	   : "=&r" ((USItype) (__q)),					\
1202	     "=&r" ((USItype) (__r))					\
1203	   : "r" ((USItype) (__n1)),					\
1204	     "r" ((USItype) (__n0)),					\
1205	     "r" ((USItype) (__d)))
1206#else
1207#if defined (__sparclite__)
1208/* This has hardware multiply but not divide.  It also has two additional
1209   instructions scan (ffs from high bit) and divscc.  */
1210#define umul_ppmm(w1, w0, u, v) \
1211  __asm__ ("umul %2,%3,%1;rd %%y,%0"					\
1212	   : "=r" ((USItype) (w1)),					\
1213	     "=r" ((USItype) (w0))					\
1214	   : "r" ((USItype) (u)),					\
1215	     "r" ((USItype) (v)))
1216#define udiv_qrnnd(q, r, n1, n0, d) \
1217  __asm__ ("! Inlined udiv_qrnnd\n"					\
1218"	wr	%%g0,%2,%%y	! Not a delayed write for sparclite\n"	\
1219"	tst	%%g0\n"							\
1220"	divscc	%3,%4,%%g1\n"						\
1221"	divscc	%%g1,%4,%%g1\n"						\
1222"	divscc	%%g1,%4,%%g1\n"						\
1223"	divscc	%%g1,%4,%%g1\n"						\
1224"	divscc	%%g1,%4,%%g1\n"						\
1225"	divscc	%%g1,%4,%%g1\n"						\
1226"	divscc	%%g1,%4,%%g1\n"						\
1227"	divscc	%%g1,%4,%%g1\n"						\
1228"	divscc	%%g1,%4,%%g1\n"						\
1229"	divscc	%%g1,%4,%%g1\n"						\
1230"	divscc	%%g1,%4,%%g1\n"						\
1231"	divscc	%%g1,%4,%%g1\n"						\
1232"	divscc	%%g1,%4,%%g1\n"						\
1233"	divscc	%%g1,%4,%%g1\n"						\
1234"	divscc	%%g1,%4,%%g1\n"						\
1235"	divscc	%%g1,%4,%%g1\n"						\
1236"	divscc	%%g1,%4,%%g1\n"						\
1237"	divscc	%%g1,%4,%%g1\n"						\
1238"	divscc	%%g1,%4,%%g1\n"						\
1239"	divscc	%%g1,%4,%%g1\n"						\
1240"	divscc	%%g1,%4,%%g1\n"						\
1241"	divscc	%%g1,%4,%%g1\n"						\
1242"	divscc	%%g1,%4,%%g1\n"						\
1243"	divscc	%%g1,%4,%%g1\n"						\
1244"	divscc	%%g1,%4,%%g1\n"						\
1245"	divscc	%%g1,%4,%%g1\n"						\
1246"	divscc	%%g1,%4,%%g1\n"						\
1247"	divscc	%%g1,%4,%%g1\n"						\
1248"	divscc	%%g1,%4,%%g1\n"						\
1249"	divscc	%%g1,%4,%%g1\n"						\
1250"	divscc	%%g1,%4,%%g1\n"						\
1251"	divscc	%%g1,%4,%0\n"						\
1252"	rd	%%y,%1\n"						\
1253"	bl,a 1f\n"							\
1254"	add	%1,%4,%1\n"						\
1255"1:	! End of inline udiv_qrnnd"					\
1256	   : "=r" ((USItype) (q)),					\
1257	     "=r" ((USItype) (r))					\
1258	   : "r" ((USItype) (n1)),					\
1259	     "r" ((USItype) (n0)),					\
1260	     "rI" ((USItype) (d))					\
1261	   : "g1" __AND_CLOBBER_CC)
1262#define UDIV_TIME 37
1263#define count_leading_zeros(count, x) \
1264  do {                                                                  \
1265  __asm__ ("scan %1,1,%0"                                               \
1266	   : "=r" ((USItype) (count))                                   \
1267	   : "r" ((USItype) (x)));					\
1268  } while (0)
1269/* Early sparclites return 63 for an argument of 0, but they warn that future
1270   implementations might change this.  Therefore, leave COUNT_LEADING_ZEROS_0
1271   undefined.  */
1272#else
1273/* SPARC without integer multiplication and divide instructions.
1274   (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
1275#define umul_ppmm(w1, w0, u, v) \
1276  __asm__ ("! Inlined umul_ppmm\n"					\
1277"	wr	%%g0,%2,%%y	! SPARC has 0-3 delay insn after a wr\n"\
1278"	sra	%3,31,%%o5	! Don't move this insn\n"		\
1279"	and	%2,%%o5,%%o5	! Don't move this insn\n"		\
1280"	andcc	%%g0,0,%%g1	! Don't move this insn\n"		\
1281"	mulscc	%%g1,%3,%%g1\n"						\
1282"	mulscc	%%g1,%3,%%g1\n"						\
1283"	mulscc	%%g1,%3,%%g1\n"						\
1284"	mulscc	%%g1,%3,%%g1\n"						\
1285"	mulscc	%%g1,%3,%%g1\n"						\
1286"	mulscc	%%g1,%3,%%g1\n"						\
1287"	mulscc	%%g1,%3,%%g1\n"						\
1288"	mulscc	%%g1,%3,%%g1\n"						\
1289"	mulscc	%%g1,%3,%%g1\n"						\
1290"	mulscc	%%g1,%3,%%g1\n"						\
1291"	mulscc	%%g1,%3,%%g1\n"						\
1292"	mulscc	%%g1,%3,%%g1\n"						\
1293"	mulscc	%%g1,%3,%%g1\n"						\
1294"	mulscc	%%g1,%3,%%g1\n"						\
1295"	mulscc	%%g1,%3,%%g1\n"						\
1296"	mulscc	%%g1,%3,%%g1\n"						\
1297"	mulscc	%%g1,%3,%%g1\n"						\
1298"	mulscc	%%g1,%3,%%g1\n"						\
1299"	mulscc	%%g1,%3,%%g1\n"						\
1300"	mulscc	%%g1,%3,%%g1\n"						\
1301"	mulscc	%%g1,%3,%%g1\n"						\
1302"	mulscc	%%g1,%3,%%g1\n"						\
1303"	mulscc	%%g1,%3,%%g1\n"						\
1304"	mulscc	%%g1,%3,%%g1\n"						\
1305"	mulscc	%%g1,%3,%%g1\n"						\
1306"	mulscc	%%g1,%3,%%g1\n"						\
1307"	mulscc	%%g1,%3,%%g1\n"						\
1308"	mulscc	%%g1,%3,%%g1\n"						\
1309"	mulscc	%%g1,%3,%%g1\n"						\
1310"	mulscc	%%g1,%3,%%g1\n"						\
1311"	mulscc	%%g1,%3,%%g1\n"						\
1312"	mulscc	%%g1,%3,%%g1\n"						\
1313"	mulscc	%%g1,0,%%g1\n"						\
1314"	add	%%g1,%%o5,%0\n"						\
1315"	rd	%%y,%1"							\
1316	   : "=r" ((USItype) (w1)),					\
1317	     "=r" ((USItype) (w0))					\
1318	   : "%rI" ((USItype) (u)),					\
1319	     "r" ((USItype) (v))						\
1320	   : "g1", "o5" __AND_CLOBBER_CC)
1321#define UMUL_TIME 39		/* 39 instructions */
1322/* It's quite necessary to add this much assembler for the sparc.
1323   The default udiv_qrnnd (in C) is more than 10 times slower!  */
1324#define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1325  __asm__ ("! Inlined udiv_qrnnd\n"					\
1326"	mov	32,%%g1\n"						\
1327"	subcc	%1,%2,%%g0\n"						\
1328"1:	bcs	5f\n"							\
1329"	 addxcc %0,%0,%0	! shift n1n0 and a q-bit in lsb\n"	\
1330"	sub	%1,%2,%1	! this kills msb of n\n"		\
1331"	addx	%1,%1,%1	! so this can't give carry\n"		\
1332"	subcc	%%g1,1,%%g1\n"						\
1333"2:	bne	1b\n"							\
1334"	 subcc	%1,%2,%%g0\n"						\
1335"	bcs	3f\n"							\
1336"	 addxcc %0,%0,%0	! shift n1n0 and a q-bit in lsb\n"	\
1337"	b	3f\n"							\
1338"	 sub	%1,%2,%1	! this kills msb of n\n"		\
1339"4:	sub	%1,%2,%1\n"						\
1340"5:	addxcc	%1,%1,%1\n"						\
1341"	bcc	2b\n"							\
1342"	 subcc	%%g1,1,%%g1\n"						\
1343"! Got carry from n.  Subtract next step to cancel this carry.\n"	\
1344"	bne	4b\n"							\
1345"	 addcc	%0,%0,%0	! shift n1n0 and a 0-bit in lsb\n"	\
1346"	sub	%1,%2,%1\n"						\
1347"3:	xnor	%0,0,%0\n"						\
1348"	! End of inline udiv_qrnnd"					\
1349	   : "=&r" ((USItype) (__q)),					\
1350	     "=&r" ((USItype) (__r))					\
1351	   : "r" ((USItype) (__d)),					\
1352	     "1" ((USItype) (__n1)),					\
1353	     "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1354#define UDIV_TIME (3+7*32)	/* 7 instructions/iteration. 32 iterations.  */
1355#endif /* __sparclite__ */
1356#endif /* __sparc_v8__ */
1357#endif /* __sparc_v9__ */
1358#endif /* sparc32 */
1359
1360#if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1361    && W_TYPE_SIZE == 64
1362#define add_ssaaaa(sh, sl, ah, al, bh, bl)				\
1363  do {									\
1364    UDItype __carry = 0;						\
1365    __asm__ ("addcc\t%r5,%6,%1\n\t"					\
1366	     "add\t%r3,%4,%0\n\t"					\
1367	     "movcs\t%%xcc, 1, %2\n\t"					\
1368	     "add\t%0, %2, %0"						\
1369	     : "=r" ((UDItype)(sh)),				      	\
1370	       "=&r" ((UDItype)(sl)),				      	\
1371	       "+r" (__carry)				      		\
1372	     : "%rJ" ((UDItype)(ah)),				     	\
1373	       "rI" ((UDItype)(bh)),				      	\
1374	       "%rJ" ((UDItype)(al)),				     	\
1375	       "rI" ((UDItype)(bl))				       	\
1376	     __CLOBBER_CC);						\
1377  } while (0)
1378
1379#define sub_ddmmss(sh, sl, ah, al, bh, bl)				\
1380  do {									\
1381    UDItype __carry = 0;						\
1382    __asm__ ("subcc\t%r5,%6,%1\n\t"					\
1383	     "sub\t%r3,%4,%0\n\t"					\
1384	     "movcs\t%%xcc, 1, %2\n\t"					\
1385	     "sub\t%0, %2, %0"						\
1386	     : "=r" ((UDItype)(sh)),				      	\
1387	       "=&r" ((UDItype)(sl)),				      	\
1388	       "+r" (__carry)				      		\
1389	     : "%rJ" ((UDItype)(ah)),				     	\
1390	       "rI" ((UDItype)(bh)),				      	\
1391	       "%rJ" ((UDItype)(al)),				     	\
1392	       "rI" ((UDItype)(bl))				       	\
1393	     __CLOBBER_CC);						\
1394  } while (0)
1395
1396#define umul_ppmm(wh, wl, u, v)						\
1397  do {									\
1398	  UDItype tmp1, tmp2, tmp3, tmp4;				\
1399	  __asm__ __volatile__ (					\
1400		   "srl %7,0,%3\n\t"					\
1401		   "mulx %3,%6,%1\n\t"					\
1402		   "srlx %6,32,%2\n\t"					\
1403		   "mulx %2,%3,%4\n\t"					\
1404		   "sllx %4,32,%5\n\t"					\
1405		   "srl %6,0,%3\n\t"					\
1406		   "sub %1,%5,%5\n\t"					\
1407		   "srlx %5,32,%5\n\t"					\
1408		   "addcc %4,%5,%4\n\t"					\
1409		   "srlx %7,32,%5\n\t"					\
1410		   "mulx %3,%5,%3\n\t"					\
1411		   "mulx %2,%5,%5\n\t"					\
1412		   "sethi %%hi(0x80000000),%2\n\t"			\
1413		   "addcc %4,%3,%4\n\t"					\
1414		   "srlx %4,32,%4\n\t"					\
1415		   "add %2,%2,%2\n\t"					\
1416		   "movcc %%xcc,%%g0,%2\n\t"				\
1417		   "addcc %5,%4,%5\n\t"					\
1418		   "sllx %3,32,%3\n\t"					\
1419		   "add %1,%3,%1\n\t"					\
1420		   "add %5,%2,%0"					\
1421	   : "=r" ((UDItype)(wh)),					\
1422	     "=&r" ((UDItype)(wl)),					\
1423	     "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4)	\
1424	   : "r" ((UDItype)(u)),					\
1425	     "r" ((UDItype)(v))						\
1426	   __CLOBBER_CC);						\
1427  } while (0)
1428#define UMUL_TIME 96
1429#define UDIV_TIME 230
1430#endif /* sparc64 */
1431
1432#if defined (__vax__) && W_TYPE_SIZE == 32
1433#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1434  __asm__ ("addl2 %5,%1\n\tadwc %3,%0"					\
1435	   : "=g" ((USItype) (sh)),					\
1436	     "=&g" ((USItype) (sl))					\
1437	   : "%0" ((USItype) (ah)),					\
1438	     "g" ((USItype) (bh)),					\
1439	     "%1" ((USItype) (al)),					\
1440	     "g" ((USItype) (bl)))
1441#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1442  __asm__ ("subl2 %5,%1\n\tsbwc %3,%0"					\
1443	   : "=g" ((USItype) (sh)),					\
1444	     "=&g" ((USItype) (sl))					\
1445	   : "0" ((USItype) (ah)),					\
1446	     "g" ((USItype) (bh)),					\
1447	     "1" ((USItype) (al)),					\
1448	     "g" ((USItype) (bl)))
1449#define umul_ppmm(xh, xl, m0, m1) \
1450  do {									\
1451    union {								\
1452	UDItype __ll;							\
1453	struct {USItype __l, __h;} __i;					\
1454      } __xx;								\
1455    USItype __m0 = (m0), __m1 = (m1);					\
1456    __asm__ ("emul %1,%2,$0,%0"						\
1457	     : "=r" (__xx.__ll)						\
1458	     : "g" (__m0),						\
1459	       "g" (__m1));						\
1460    (xh) = __xx.__i.__h;						\
1461    (xl) = __xx.__i.__l;						\
1462    (xh) += ((((SItype) __m0 >> 31) & __m1)				\
1463	     + (((SItype) __m1 >> 31) & __m0));				\
1464  } while (0)
1465#define sdiv_qrnnd(q, r, n1, n0, d) \
1466  do {									\
1467    union {DItype __ll;							\
1468	   struct {SItype __l, __h;} __i;				\
1469	  } __xx;							\
1470    __xx.__i.__h = n1; __xx.__i.__l = n0;				\
1471    __asm__ ("ediv %3,%2,%0,%1"						\
1472	     : "=g" (q), "=g" (r)					\
1473	     : "g" (__xx.__ll), "g" (d));				\
1474  } while (0)
1475#endif /* __vax__ */
1476
1477#ifdef _TMS320C6X
1478#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1479  do									\
1480    {									\
1481      UDItype __ll;							\
1482      __asm__ ("addu .l1 %1, %2, %0"					\
1483	       : "=a" (__ll) : "a" (al), "a" (bl));			\
1484      (sl) = (USItype)__ll;						\
1485      (sh) = ((USItype)(__ll >> 32)) + (ah) + (bh);			\
1486    }									\
1487  while (0)
1488
1489#ifdef _TMS320C6400_PLUS
1490#define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
1491#define umul_ppmm(w1, w0, u, v)						\
1492  do {									\
1493    UDItype __x = (UDItype) (USItype) (u) * (USItype) (v);		\
1494    (w1) = (USItype) (__x >> 32);					\
1495    (w0) = (USItype) (__x);						\
1496  } while (0)
1497#endif  /* _TMS320C6400_PLUS */
1498
1499#define count_leading_zeros(count, x)	((count) = __builtin_clz (x))
1500#ifdef _TMS320C6400
1501#define count_trailing_zeros(count, x)	((count) = __builtin_ctz (x))
1502#endif
1503#define UMUL_TIME 4
1504#define UDIV_TIME 40
1505#endif /* _TMS320C6X */
1506
1507#if defined (__xtensa__) && W_TYPE_SIZE == 32
1508/* This code is not Xtensa-configuration-specific, so rely on the compiler
1509   to expand builtin functions depending on what configuration features
1510   are available.  This avoids library calls when the operation can be
1511   performed in-line.  */
1512#define umul_ppmm(w1, w0, u, v)						\
1513  do {									\
1514    DWunion __w;							\
1515    __w.ll = __builtin_umulsidi3 (u, v);				\
1516    w1 = __w.s.high;							\
1517    w0 = __w.s.low;							\
1518  } while (0)
1519#define __umulsidi3(u, v)		__builtin_umulsidi3 (u, v)
1520#define count_leading_zeros(COUNT, X)	((COUNT) = __builtin_clz (X))
1521#define count_trailing_zeros(COUNT, X)	((COUNT) = __builtin_ctz (X))
1522#endif /* __xtensa__ */
1523
1524#if defined xstormy16
1525extern UHItype __stormy16_count_leading_zeros (UHItype);
1526#define count_leading_zeros(count, x)					\
1527  do									\
1528    {									\
1529      UHItype size;							\
1530									\
1531      /* We assume that W_TYPE_SIZE is a multiple of 16...  */		\
1532      for ((count) = 0, size = W_TYPE_SIZE; size; size -= 16)		\
1533	{								\
1534	  UHItype c;							\
1535									\
1536	  c = __clzhi2 ((x) >> (size - 16));				\
1537	  (count) += c;							\
1538	  if (c != 16)							\
1539	    break;							\
1540	}								\
1541    }									\
1542  while (0)
1543#define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1544#endif
1545
1546#if defined (__z8000__) && W_TYPE_SIZE == 16
1547#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1548  __asm__ ("add	%H1,%H5\n\tadc	%H0,%H3"				\
1549	   : "=r" ((unsigned int)(sh)),					\
1550	     "=&r" ((unsigned int)(sl))					\
1551	   : "%0" ((unsigned int)(ah)),					\
1552	     "r" ((unsigned int)(bh)),					\
1553	     "%1" ((unsigned int)(al)),					\
1554	     "rQR" ((unsigned int)(bl)))
1555#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1556  __asm__ ("sub	%H1,%H5\n\tsbc	%H0,%H3"				\
1557	   : "=r" ((unsigned int)(sh)),					\
1558	     "=&r" ((unsigned int)(sl))					\
1559	   : "0" ((unsigned int)(ah)),					\
1560	     "r" ((unsigned int)(bh)),					\
1561	     "1" ((unsigned int)(al)),					\
1562	     "rQR" ((unsigned int)(bl)))
1563#define umul_ppmm(xh, xl, m0, m1) \
1564  do {									\
1565    union {long int __ll;						\
1566	   struct {unsigned int __h, __l;} __i;				\
1567	  } __xx;							\
1568    unsigned int __m0 = (m0), __m1 = (m1);				\
1569    __asm__ ("mult	%S0,%H3"					\
1570	     : "=r" (__xx.__i.__h),					\
1571	       "=r" (__xx.__i.__l)					\
1572	     : "%1" (__m0),						\
1573	       "rQR" (__m1));						\
1574    (xh) = __xx.__i.__h; (xl) = __xx.__i.__l;				\
1575    (xh) += ((((signed int) __m0 >> 15) & __m1)				\
1576	     + (((signed int) __m1 >> 15) & __m0));			\
1577  } while (0)
1578#endif /* __z8000__ */
1579
1580#endif /* __GNUC__ */
1581
1582/* If this machine has no inline assembler, use C macros.  */
1583
1584#if !defined (add_ssaaaa)
1585#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1586  do {									\
1587    UWtype __x;								\
1588    __x = (al) + (bl);							\
1589    (sh) = (ah) + (bh) + (__x < (al));					\
1590    (sl) = __x;								\
1591  } while (0)
1592#endif
1593
1594#if !defined (sub_ddmmss)
1595#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1596  do {									\
1597    UWtype __x;								\
1598    __x = (al) - (bl);							\
1599    (sh) = (ah) - (bh) - (__x > (al));					\
1600    (sl) = __x;								\
1601  } while (0)
1602#endif
1603
1604/* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1605   smul_ppmm.  */
1606#if !defined (umul_ppmm) && defined (smul_ppmm)
1607#define umul_ppmm(w1, w0, u, v)						\
1608  do {									\
1609    UWtype __w1;							\
1610    UWtype __xm0 = (u), __xm1 = (v);					\
1611    smul_ppmm (__w1, w0, __xm0, __xm1);					\
1612    (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1)		\
1613		+ (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0);		\
1614  } while (0)
1615#endif
1616
1617/* If we still don't have umul_ppmm, define it using plain C.  */
1618#if !defined (umul_ppmm)
1619#define umul_ppmm(w1, w0, u, v)						\
1620  do {									\
1621    UWtype __x0, __x1, __x2, __x3;					\
1622    UHWtype __ul, __vl, __uh, __vh;					\
1623									\
1624    __ul = __ll_lowpart (u);						\
1625    __uh = __ll_highpart (u);						\
1626    __vl = __ll_lowpart (v);						\
1627    __vh = __ll_highpart (v);						\
1628									\
1629    __x0 = (UWtype) __ul * __vl;					\
1630    __x1 = (UWtype) __ul * __vh;					\
1631    __x2 = (UWtype) __uh * __vl;					\
1632    __x3 = (UWtype) __uh * __vh;					\
1633									\
1634    __x1 += __ll_highpart (__x0);/* this can't give carry */		\
1635    __x1 += __x2;		/* but this indeed can */		\
1636    if (__x1 < __x2)		/* did we get it? */			\
1637      __x3 += __ll_B;		/* yes, add it in the proper pos.  */	\
1638									\
1639    (w1) = __x3 + __ll_highpart (__x1);					\
1640    (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0);		\
1641  } while (0)
1642#endif
1643
1644#if !defined (__umulsidi3)
1645#define __umulsidi3(u, v) \
1646  ({DWunion __w;							\
1647    umul_ppmm (__w.s.high, __w.s.low, u, v);				\
1648    __w.ll; })
1649#endif
1650
1651/* Define this unconditionally, so it can be used for debugging.  */
1652#define __udiv_qrnnd_c(q, r, n1, n0, d) \
1653  do {									\
1654    UWtype __d1, __d0, __q1, __q0;					\
1655    UWtype __r1, __r0, __m;						\
1656    __d1 = __ll_highpart (d);						\
1657    __d0 = __ll_lowpart (d);						\
1658									\
1659    __r1 = (n1) % __d1;							\
1660    __q1 = (n1) / __d1;							\
1661    __m = (UWtype) __q1 * __d0;						\
1662    __r1 = __r1 * __ll_B | __ll_highpart (n0);				\
1663    if (__r1 < __m)							\
1664      {									\
1665	__q1--, __r1 += (d);						\
1666	if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1667	  if (__r1 < __m)						\
1668	    __q1--, __r1 += (d);					\
1669      }									\
1670    __r1 -= __m;							\
1671									\
1672    __r0 = __r1 % __d1;							\
1673    __q0 = __r1 / __d1;							\
1674    __m = (UWtype) __q0 * __d0;						\
1675    __r0 = __r0 * __ll_B | __ll_lowpart (n0);				\
1676    if (__r0 < __m)							\
1677      {									\
1678	__q0--, __r0 += (d);						\
1679	if (__r0 >= (d))						\
1680	  if (__r0 < __m)						\
1681	    __q0--, __r0 += (d);					\
1682      }									\
1683    __r0 -= __m;							\
1684									\
1685    (q) = (UWtype) __q1 * __ll_B | __q0;				\
1686    (r) = __r0;								\
1687  } while (0)
1688
1689/* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1690   __udiv_w_sdiv (defined in libgcc or elsewhere).  */
1691#if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1692#define udiv_qrnnd(q, r, nh, nl, d) \
1693  do {									\
1694    extern UWtype __udiv_w_sdiv (UWtype *, UWtype, UWtype, UWtype);	\
1695    UWtype __r;								\
1696    (q) = __udiv_w_sdiv (&__r, nh, nl, d);				\
1697    (r) = __r;								\
1698  } while (0)
1699#endif
1700
1701/* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c.  */
1702#if !defined (udiv_qrnnd)
1703#define UDIV_NEEDS_NORMALIZATION 1
1704#define udiv_qrnnd __udiv_qrnnd_c
1705#endif
1706
1707#if !defined (count_leading_zeros)
1708#define count_leading_zeros(count, x) \
1709  do {									\
1710    UWtype __xr = (x);							\
1711    UWtype __a;								\
1712									\
1713    if (W_TYPE_SIZE <= 32)						\
1714      {									\
1715	__a = __xr < ((UWtype)1<<2*__BITS4)				\
1716	  ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4)			\
1717	  : (__xr < ((UWtype)1<<3*__BITS4) ?  2*__BITS4 : 3*__BITS4);	\
1718      }									\
1719    else								\
1720      {									\
1721	for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8)			\
1722	  if (((__xr >> __a) & 0xff) != 0)				\
1723	    break;							\
1724      }									\
1725									\
1726    (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a);		\
1727  } while (0)
1728#define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1729#endif
1730
1731#if !defined (count_trailing_zeros)
1732/* Define count_trailing_zeros using count_leading_zeros.  The latter might be
1733   defined in asm, but if it is not, the C version above is good enough.  */
1734#define count_trailing_zeros(count, x) \
1735  do {									\
1736    UWtype __ctz_x = (x);						\
1737    UWtype __ctz_c;							\
1738    count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x);			\
1739    (count) = W_TYPE_SIZE - 1 - __ctz_c;				\
1740  } while (0)
1741#endif
1742
1743#ifndef UDIV_NEEDS_NORMALIZATION
1744#define UDIV_NEEDS_NORMALIZATION 0
1745#endif
1746