1: /* taken from gcc-4.1.1 */
2: /* longlong.h -- definitions for mixed size 32/64 bit arithmetic.
3: Copyright (C) 1991, 1992, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2004, 2005,2006,2007 Free Software Foundation, Inc.
4:
5: This definition file is free software: you can redistribute it and/or
6: modify it under the terms of the GNU General Public License
7: as published by the Free Software Foundation, either version 3
8: of the License, 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 warranty
12: of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13: 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, see http://www.gnu.org/licenses/.
17: */
18:
19: /* You have to define the following before including this file:
20:
21: UWtype -- An unsigned type, default type for operations (typically a "word")
22: UHWtype -- An unsigned type, at least half the size of UWtype.
23: UDWtype -- An unsigned type, at least twice as large a UWtype
24: W_TYPE_SIZE -- size in bits of UWtype
25:
26: UQItype -- Unsigned 8 bit type.
27: SItype, USItype -- Signed and unsigned 32 bit types.
28: DItype, UDItype -- Signed and unsigned 64 bit types.
29:
30: On a 32 bit machine UWtype should typically be USItype;
31: on a 64 bit machine, UWtype should typically be UDItype. */
32:
33: #define __BITS4 (W_TYPE_SIZE / 4)
34: #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
35: #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
36: #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
37:
38: #ifndef W_TYPE_SIZE
39: #define W_TYPE_SIZE 32
40: #define UWtype USItype
41: #define UHWtype USItype
42: #define UDWtype UDItype
43: #endif
44:
45: extern const UQItype __clz_tab[256];
46:
47: /* Define auxiliary asm macros.
48:
49: 1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two
50: UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype
51: word product in HIGH_PROD and LOW_PROD.
52:
53: 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a
54: UDWtype product. This is just a variant of umul_ppmm.
55:
56: 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
57: denominator) divides a UDWtype, composed by the UWtype integers
58: HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
59: in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less
60: than DENOMINATOR for correct operation. If, in addition, the most
61: significant bit of DENOMINATOR must be 1, then the pre-processor symbol
62: UDIV_NEEDS_NORMALIZATION is defined to 1.
63:
64: 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
65: denominator). Like udiv_qrnnd but the numbers are signed. The quotient
66: is rounded towards 0.
67:
68: 5) count_leading_zeros(count, x) counts the number of zero-bits from the
69: msb to the first nonzero bit in the UWtype X. This is the number of
70: steps X needs to be shifted left to set the msb. Undefined for X == 0,
71: unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value.
72:
73: 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts
74: from the least significant end.
75:
76: 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
77: high_addend_2, low_addend_2) adds two UWtype integers, composed by
78: HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
79: respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow
80: (i.e. carry out) is not stored anywhere, and is lost.
81:
82: 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
83: high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
84: composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
85: LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE
86: and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
87: and is lost.
88:
89: If any of these macros are left undefined for a particular CPU,
90: C macros are used. */
91:
92: /* The CPUs come in alphabetical order below.
93:
94: Please add support for more CPUs here, or improve the current support
95: for the CPUs below!
96: (E.g. WE32100, IBM360.) */
97:
98: #if defined (__GNUC__) && !defined (NO_ASM)
99:
100: /* We sometimes need to clobber "cc" with gcc2, but that would not be
101: understood by gcc1. Use cpp to avoid major code duplication. */
102: #if __GNUC__ < 2
103: #define __CLOBBER_CC
104: #define __AND_CLOBBER_CC
105: #else /* __GNUC__ >= 2 */
106: #define __CLOBBER_CC : "cc"
107: #define __AND_CLOBBER_CC , "cc"
108: #endif /* __GNUC__ < 2 */
109:
110: #if defined (__alpha) && W_TYPE_SIZE == 64 && __GNUC__ >= 3
111: #define umul_ppmm(ph, pl, m0, m1) \
112: do { \
113: UDItype __m0 = (m0), __m1 = (m1); \
114: (ph) = __builtin_alpha_umulh (__m0, __m1); \
115: (pl) = __m0 * __m1; \
116: } while (0)
117: #define UMUL_TIME 46
118: #ifndef LONGLONG_STANDALONE
119: #define udiv_qrnnd(q, r, n1, n0, d) \
120: do { UDItype __r; \
121: (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \
122: (r) = __r; \
123: } while (0)
124: extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype);
125: #define UDIV_TIME 220
126: #endif /* LONGLONG_STANDALONE */
127: #ifdef __alpha_cix__
128: #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X))
129: #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X))
130: #define COUNT_LEADING_ZEROS_0 64
131: #else
132: #define count_leading_zeros(COUNT,X) \
133: do { \
134: UDItype __xr = (X), __t, __a; \
135: __t = __builtin_alpha_cmpbge (0, __xr); \
136: __a = __clz_tab[__t ^ 0xff] - 1; \
137: __t = __builtin_alpha_extbl (__xr, __a); \
138: (COUNT) = 64 - (__clz_tab[__t] + __a*8); \
139: } while (0)
140: #define count_trailing_zeros(COUNT,X) \
141: do { \
142: UDItype __xr = (X), __t, __a; \
143: __t = __builtin_alpha_cmpbge (0, __xr); \
144: __t = ~__t & -~__t; \
145: __a = ((__t & 0xCC) != 0) * 2; \
146: __a += ((__t & 0xF0) != 0) * 4; \
147: __a += ((__t & 0xAA) != 0); \
148: __t = __builtin_alpha_extbl (__xr, __a); \
149: __a <<= 3; \
150: __t &= -__t; \
151: __a += ((__t & 0xCC) != 0) * 2; \
152: __a += ((__t & 0xF0) != 0) * 4; \
153: __a += ((__t & 0xAA) != 0); \
154: (COUNT) = __a; \
155: } while (0)
156: #endif /* __alpha_cix__ */
157: #endif /* __alpha */
158:
159: #if defined (__arc__) && W_TYPE_SIZE == 32
160: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
161: __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \
162: : "=r" ((USItype) (sh)), \
163: "=&r" ((USItype) (sl)) \
164: : "%r" ((USItype) (ah)), \
165: "rIJ" ((USItype) (bh)), \
166: "%r" ((USItype) (al)), \
167: "rIJ" ((USItype) (bl)))
168: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
169: __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \
170: : "=r" ((USItype) (sh)), \
171: "=&r" ((USItype) (sl)) \
172: : "r" ((USItype) (ah)), \
173: "rIJ" ((USItype) (bh)), \
174: "r" ((USItype) (al)), \
175: "rIJ" ((USItype) (bl)))
176: /* Call libgcc routine. */
177: #define umul_ppmm(w1, w0, u, v) \
178: do { \
179: DWunion __w; \
180: __w.ll = __umulsidi3 (u, v); \
181: w1 = __w.s.high; \
182: w0 = __w.s.low; \
183: } while (0)
184: #define __umulsidi3 __umulsidi3
185: UDItype __umulsidi3 (USItype, USItype);
186: #endif
187:
188: #if defined (__arm__) && !defined (__thumb__) && W_TYPE_SIZE == 32
189: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
190: __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \
191: : "=r" ((USItype) (sh)), \
192: "=&r" ((USItype) (sl)) \
193: : "%r" ((USItype) (ah)), \
194: "rI" ((USItype) (bh)), \
195: "%r" ((USItype) (al)), \
196: "rI" ((USItype) (bl)) __CLOBBER_CC)
197: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
198: __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \
199: : "=r" ((USItype) (sh)), \
200: "=&r" ((USItype) (sl)) \
201: : "r" ((USItype) (ah)), \
202: "rI" ((USItype) (bh)), \
203: "r" ((USItype) (al)), \
204: "rI" ((USItype) (bl)) __CLOBBER_CC)
205: #define umul_ppmm(xh, xl, a, b) \
206: {register USItype __t0, __t1, __t2; \
207: __asm__ ("%@ Inlined umul_ppmm\n" \
208: " mov %2, %5, lsr #16\n" \
209: " mov %0, %6, lsr #16\n" \
210: " bic %3, %5, %2, lsl #16\n" \
211: " bic %4, %6, %0, lsl #16\n" \
212: " mul %1, %3, %4\n" \
213: " mul %4, %2, %4\n" \
214: " mul %3, %0, %3\n" \
215: " mul %0, %2, %0\n" \
216: " adds %3, %4, %3\n" \
217: " addcs %0, %0, #65536\n" \
218: " adds %1, %1, %3, lsl #16\n" \
219: " adc %0, %0, %3, lsr #16" \
220: : "=&r" ((USItype) (xh)), \
221: "=r" ((USItype) (xl)), \
222: "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \
223: : "r" ((USItype) (a)), \
224: "r" ((USItype) (b)) __CLOBBER_CC );}
225: #define UMUL_TIME 20
226: #define UDIV_TIME 100
227: #endif /* __arm__ */
228:
229: #if defined (__hppa) && W_TYPE_SIZE == 32
230: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
231: __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \
232: : "=r" ((USItype) (sh)), \
233: "=&r" ((USItype) (sl)) \
234: : "%rM" ((USItype) (ah)), \
235: "rM" ((USItype) (bh)), \
236: "%rM" ((USItype) (al)), \
237: "rM" ((USItype) (bl)))
238: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
239: __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \
240: : "=r" ((USItype) (sh)), \
241: "=&r" ((USItype) (sl)) \
242: : "rM" ((USItype) (ah)), \
243: "rM" ((USItype) (bh)), \
244: "rM" ((USItype) (al)), \
245: "rM" ((USItype) (bl)))
246: #if defined (_PA_RISC1_1)
247: #define umul_ppmm(w1, w0, u, v) \
248: do { \
249: union \
250: { \
251: UDItype __f; \
252: struct {USItype __w1, __w0;} __w1w0; \
253: } __t; \
254: __asm__ ("xmpyu %1,%2,%0" \
255: : "=x" (__t.__f) \
256: : "x" ((USItype) (u)), \
257: "x" ((USItype) (v))); \
258: (w1) = __t.__w1w0.__w1; \
259: (w0) = __t.__w1w0.__w0; \
260: } while (0)
261: #define UMUL_TIME 8
262: #else
263: #define UMUL_TIME 30
264: #endif
265: #define UDIV_TIME 40
266: #define count_leading_zeros(count, x) \
267: do { \
268: USItype __tmp; \
269: __asm__ ( \
270: "ldi 1,%0\n" \
271: " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \
272: " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\
273: " ldo 16(%0),%0 ; Yes. Perform add.\n" \
274: " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \
275: " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\
276: " ldo 8(%0),%0 ; Yes. Perform add.\n" \
277: " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \
278: " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\
279: " ldo 4(%0),%0 ; Yes. Perform add.\n" \
280: " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \
281: " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\
282: " ldo 2(%0),%0 ; Yes. Perform add.\n" \
283: " extru %1,30,1,%1 ; Extract bit 1.\n" \
284: " sub %0,%1,%0 ; Subtract it.\n" \
285: : "=r" (count), "=r" (__tmp) : "1" (x)); \
286: } while (0)
287: #endif
288:
289: #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32
290: #define smul_ppmm(xh, xl, m0, m1) \
291: do { \
292: union {DItype __ll; \
293: struct {USItype __h, __l;} __i; \
294: } __x; \
295: __asm__ ("lr %N0,%1\n\tmr %0,%2" \
296: : "=&r" (__x.__ll) \
297: : "r" (m0), "r" (m1)); \
298: (xh) = __x.__i.__h; (xl) = __x.__i.__l; \
299: } while (0)
300: #define sdiv_qrnnd(q, r, n1, n0, d) \
301: do { \
302: union {DItype __ll; \
303: struct {USItype __h, __l;} __i; \
304: } __x; \
305: __x.__i.__h = n1; __x.__i.__l = n0; \
306: __asm__ ("dr %0,%2" \
307: : "=r" (__x.__ll) \
308: : "0" (__x.__ll), "r" (d)); \
309: (q) = __x.__i.__l; (r) = __x.__i.__h; \
310: } while (0)
311: #endif
312:
313: #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32
314: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
315: __asm__ ("addl %5,%1\n\tadcl %3,%0" \
316: : "=r" ((USItype) (sh)), \
317: "=&r" ((USItype) (sl)) \
318: : "%0" ((USItype) (ah)), \
319: "g" ((USItype) (bh)), \
320: "%1" ((USItype) (al)), \
321: "g" ((USItype) (bl)))
322: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
323: __asm__ ("subl %5,%1\n\tsbbl %3,%0" \
324: : "=r" ((USItype) (sh)), \
325: "=&r" ((USItype) (sl)) \
326: : "0" ((USItype) (ah)), \
327: "g" ((USItype) (bh)), \
328: "1" ((USItype) (al)), \
329: "g" ((USItype) (bl)))
330: #define umul_ppmm(w1, w0, u, v) \
331: __asm__ ("mull %3" \
332: : "=a" ((USItype) (w0)), \
333: "=d" ((USItype) (w1)) \
334: : "%0" ((USItype) (u)), \
335: "rm" ((USItype) (v)))
336: #define udiv_qrnnd(q, r, n1, n0, dv) \
337: __asm__ ("divl %4" \
338: : "=a" ((USItype) (q)), \
339: "=d" ((USItype) (r)) \
340: : "0" ((USItype) (n0)), \
341: "1" ((USItype) (n1)), \
342: "rm" ((USItype) (dv)))
343: #define count_leading_zeros(count, x) \
344: do { \
345: USItype __cbtmp; \
346: __asm__ ("bsrl %1,%0" \
347: : "=r" (__cbtmp) : "rm" ((USItype) (x))); \
348: (count) = __cbtmp ^ 31; \
349: } while (0)
350: #define count_trailing_zeros(count, x) \
351: __asm__ ("bsfl %1,%0" : "=r" (count) : "rm" ((USItype)(x)))
352: #define UMUL_TIME 40
353: #define UDIV_TIME 40
354: #endif /* 80x86 */
355:
356: #if defined (__i960__) && W_TYPE_SIZE == 32
357: #define umul_ppmm(w1, w0, u, v) \
358: ({union {UDItype __ll; \
359: struct {USItype __l, __h;} __i; \
360: } __xx; \
361: __asm__ ("emul %2,%1,%0" \
362: : "=d" (__xx.__ll) \
363: : "%dI" ((USItype) (u)), \
364: "dI" ((USItype) (v))); \
365: (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
366: #define __umulsidi3(u, v) \
367: ({UDItype __w; \
368: __asm__ ("emul %2,%1,%0" \
369: : "=d" (__w) \
370: : "%dI" ((USItype) (u)), \
371: "dI" ((USItype) (v))); \
372: __w; })
373: #endif /* __i960__ */
374:
375: #if defined (__M32R__) && W_TYPE_SIZE == 32
376: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
377: /* The cmp clears the condition bit. */ \
378: __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \
379: : "=r" ((USItype) (sh)), \
380: "=&r" ((USItype) (sl)) \
381: : "0" ((USItype) (ah)), \
382: "r" ((USItype) (bh)), \
383: "1" ((USItype) (al)), \
384: "r" ((USItype) (bl)) \
385: : "cbit")
386: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
387: /* The cmp clears the condition bit. */ \
388: __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \
389: : "=r" ((USItype) (sh)), \
390: "=&r" ((USItype) (sl)) \
391: : "0" ((USItype) (ah)), \
392: "r" ((USItype) (bh)), \
393: "1" ((USItype) (al)), \
394: "r" ((USItype) (bl)) \
395: : "cbit")
396: #endif /* __M32R__ */
397:
398: #if defined (__mc68000__) && W_TYPE_SIZE == 32
399: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
400: __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \
401: : "=d" ((USItype) (sh)), \
402: "=&d" ((USItype) (sl)) \
403: : "%0" ((USItype) (ah)), \
404: "d" ((USItype) (bh)), \
405: "%1" ((USItype) (al)), \
406: "g" ((USItype) (bl)))
407: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
408: __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \
409: : "=d" ((USItype) (sh)), \
410: "=&d" ((USItype) (sl)) \
411: : "0" ((USItype) (ah)), \
412: "d" ((USItype) (bh)), \
413: "1" ((USItype) (al)), \
414: "g" ((USItype) (bl)))
415:
416: /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r. */
417: #if (defined (__mc68020__) && !defined (__mc68060__))
418: #define umul_ppmm(w1, w0, u, v) \
419: __asm__ ("mulu%.l %3,%1:%0" \
420: : "=d" ((USItype) (w0)), \
421: "=d" ((USItype) (w1)) \
422: : "%0" ((USItype) (u)), \
423: "dmi" ((USItype) (v)))
424: #define UMUL_TIME 45
425: #define udiv_qrnnd(q, r, n1, n0, d) \
426: __asm__ ("divu%.l %4,%1:%0" \
427: : "=d" ((USItype) (q)), \
428: "=d" ((USItype) (r)) \
429: : "0" ((USItype) (n0)), \
430: "1" ((USItype) (n1)), \
431: "dmi" ((USItype) (d)))
432: #define UDIV_TIME 90
433: #define sdiv_qrnnd(q, r, n1, n0, d) \
434: __asm__ ("divs%.l %4,%1:%0" \
435: : "=d" ((USItype) (q)), \
436: "=d" ((USItype) (r)) \
437: : "0" ((USItype) (n0)), \
438: "1" ((USItype) (n1)), \
439: "dmi" ((USItype) (d)))
440:
441: #elif defined (__mcoldfire__) /* not mc68020 */
442:
443: #define umul_ppmm(xh, xl, a, b) \
444: __asm__ ("| Inlined umul_ppmm\n" \
445: " move%.l %2,%/d0\n" \
446: " move%.l %3,%/d1\n" \
447: " move%.l %/d0,%/d2\n" \
448: " swap %/d0\n" \
449: " move%.l %/d1,%/d3\n" \
450: " swap %/d1\n" \
451: " move%.w %/d2,%/d4\n" \
452: " mulu %/d3,%/d4\n" \
453: " mulu %/d1,%/d2\n" \
454: " mulu %/d0,%/d3\n" \
455: " mulu %/d0,%/d1\n" \
456: " move%.l %/d4,%/d0\n" \
457: " clr%.w %/d0\n" \
458: " swap %/d0\n" \
459: " add%.l %/d0,%/d2\n" \
460: " add%.l %/d3,%/d2\n" \
461: " jcc 1f\n" \
462: " add%.l %#65536,%/d1\n" \
463: "1: swap %/d2\n" \
464: " moveq %#0,%/d0\n" \
465: " move%.w %/d2,%/d0\n" \
466: " move%.w %/d4,%/d2\n" \
467: " move%.l %/d2,%1\n" \
468: " add%.l %/d1,%/d0\n" \
469: " move%.l %/d0,%0" \
470: : "=g" ((USItype) (xh)), \
471: "=g" ((USItype) (xl)) \
472: : "g" ((USItype) (a)), \
473: "g" ((USItype) (b)) \
474: : "d0", "d1", "d2", "d3", "d4")
475: #define UMUL_TIME 100
476: #define UDIV_TIME 400
477: #else /* not ColdFire */
478: /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */
479: #define umul_ppmm(xh, xl, a, b) \
480: __asm__ ("| Inlined umul_ppmm\n" \
481: " move%.l %2,%/d0\n" \
482: " move%.l %3,%/d1\n" \
483: " move%.l %/d0,%/d2\n" \
484: " swap %/d0\n" \
485: " move%.l %/d1,%/d3\n" \
486: " swap %/d1\n" \
487: " move%.w %/d2,%/d4\n" \
488: " mulu %/d3,%/d4\n" \
489: " mulu %/d1,%/d2\n" \
490: " mulu %/d0,%/d3\n" \
491: " mulu %/d0,%/d1\n" \
492: " move%.l %/d4,%/d0\n" \
493: " eor%.w %/d0,%/d0\n" \
494: " swap %/d0\n" \
495: " add%.l %/d0,%/d2\n" \
496: " add%.l %/d3,%/d2\n" \
497: " jcc 1f\n" \
498: " add%.l %#65536,%/d1\n" \
499: "1: swap %/d2\n" \
500: " moveq %#0,%/d0\n" \
501: " move%.w %/d2,%/d0\n" \
502: " move%.w %/d4,%/d2\n" \
503: " move%.l %/d2,%1\n" \
504: " add%.l %/d1,%/d0\n" \
505: " move%.l %/d0,%0" \
506: : "=g" ((USItype) (xh)), \
507: "=g" ((USItype) (xl)) \
508: : "g" ((USItype) (a)), \
509: "g" ((USItype) (b)) \
510: : "d0", "d1", "d2", "d3", "d4")
511: #define UMUL_TIME 100
512: #define UDIV_TIME 400
513:
514: #endif /* not mc68020 */
515:
516: /* The '020, '030, '040 and '060 have bitfield insns.
517: cpu32 disguises as a 68020, but lacks them. */
518: #if defined (__mc68020__) && !defined (__mcpu32__)
519: #define count_leading_zeros(count, x) \
520: __asm__ ("bfffo %1{%b2:%b2},%0" \
521: : "=d" ((USItype) (count)) \
522: : "od" ((USItype) (x)), "n" (0))
523: #endif
524: #endif /* mc68000 */
525:
526: #if defined (__m88000__) && W_TYPE_SIZE == 32
527: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
528: __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \
529: : "=r" ((USItype) (sh)), \
530: "=&r" ((USItype) (sl)) \
531: : "%rJ" ((USItype) (ah)), \
532: "rJ" ((USItype) (bh)), \
533: "%rJ" ((USItype) (al)), \
534: "rJ" ((USItype) (bl)))
535: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
536: __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \
537: : "=r" ((USItype) (sh)), \
538: "=&r" ((USItype) (sl)) \
539: : "rJ" ((USItype) (ah)), \
540: "rJ" ((USItype) (bh)), \
541: "rJ" ((USItype) (al)), \
542: "rJ" ((USItype) (bl)))
543: #define count_leading_zeros(count, x) \
544: do { \
545: USItype __cbtmp; \
546: __asm__ ("ff1 %0,%1" \
547: : "=r" (__cbtmp) \
548: : "r" ((USItype) (x))); \
549: (count) = __cbtmp ^ 31; \
550: } while (0)
551: #define COUNT_LEADING_ZEROS_0 63 /* sic */
552: #if defined (__mc88110__)
553: #define umul_ppmm(wh, wl, u, v) \
554: do { \
555: union {UDItype __ll; \
556: struct {USItype __h, __l;} __i; \
557: } __xx; \
558: __asm__ ("mulu.d %0,%1,%2" \
559: : "=r" (__xx.__ll) \
560: : "r" ((USItype) (u)), \
561: "r" ((USItype) (v))); \
562: (wh) = __xx.__i.__h; \
563: (wl) = __xx.__i.__l; \
564: } while (0)
565: #define udiv_qrnnd(q, r, n1, n0, d) \
566: ({union {UDItype __ll; \
567: struct {USItype __h, __l;} __i; \
568: } __xx; \
569: USItype __q; \
570: __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
571: __asm__ ("divu.d %0,%1,%2" \
572: : "=r" (__q) \
573: : "r" (__xx.__ll), \
574: "r" ((USItype) (d))); \
575: (r) = (n0) - __q * (d); (q) = __q; })
576: #define UMUL_TIME 5
577: #define UDIV_TIME 25
578: #else
579: #define UMUL_TIME 17
580: #define UDIV_TIME 150
581: #endif /* __mc88110__ */
582: #endif /* __m88000__ */
583:
584: #if defined (__mips__) && W_TYPE_SIZE == 32
585: #define umul_ppmm(w1, w0, u, v) \
586: __asm__ ("multu %2,%3" \
587: : "=l" ((USItype) (w0)), \
588: "=h" ((USItype) (w1)) \
589: : "d" ((USItype) (u)), \
590: "d" ((USItype) (v)))
591: #define UMUL_TIME 10
592: #define UDIV_TIME 100
593: #endif /* __mips__ */
594:
595: #if defined (__ns32000__) && W_TYPE_SIZE == 32
596: #define umul_ppmm(w1, w0, u, v) \
597: ({union {UDItype __ll; \
598: struct {USItype __l, __h;} __i; \
599: } __xx; \
600: __asm__ ("meid %2,%0" \
601: : "=g" (__xx.__ll) \
602: : "%0" ((USItype) (u)), \
603: "g" ((USItype) (v))); \
604: (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;})
605: #define __umulsidi3(u, v) \
606: ({UDItype __w; \
607: __asm__ ("meid %2,%0" \
608: : "=g" (__w) \
609: : "%0" ((USItype) (u)), \
610: "g" ((USItype) (v))); \
611: __w; })
612: #define udiv_qrnnd(q, r, n1, n0, d) \
613: ({union {UDItype __ll; \
614: struct {USItype __l, __h;} __i; \
615: } __xx; \
616: __xx.__i.__h = (n1); __xx.__i.__l = (n0); \
617: __asm__ ("deid %2,%0" \
618: : "=g" (__xx.__ll) \
619: : "0" (__xx.__ll), \
620: "g" ((USItype) (d))); \
621: (r) = __xx.__i.__l; (q) = __xx.__i.__h; })
622: #define count_trailing_zeros(count,x) \
623: do { \
624: __asm__ ("ffsd %2,%0" \
625: : "=r" ((USItype) (count)) \
626: : "0" ((USItype) 0), \
627: "r" ((USItype) (x))); \
628: } while (0)
629: #endif /* __ns32000__ */
630:
631: /* FIXME: We should test _IBMR2 here when we add assembly support for the
632: system vendor compilers.
633: FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good
634: enough, since that hits ARM and m68k too. */
635: #if (defined (_ARCH_PPC) /* AIX */ \
636: || defined (_ARCH_PWR) /* AIX */ \
637: || defined (_ARCH_COM) /* AIX */ \
638: || defined (__powerpc__) /* gcc */ \
639: || defined (__POWERPC__) /* BEOS */ \
640: || defined (__ppc__) /* Darwin */ \
641: || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
642: || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
643: && CPU_FAMILY == PPC) \
644: ) && W_TYPE_SIZE == 32
645: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
646: do { \
647: if (__builtin_constant_p (bh) && (bh) == 0) \
648: __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
649: : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
650: else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
651: __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
652: : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
653: else \
654: __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
655: : "=r" (sh), "=&r" (sl) \
656: : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
657: } while (0)
658: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
659: do { \
660: if (__builtin_constant_p (ah) && (ah) == 0) \
661: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
662: : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
663: else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \
664: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
665: : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
666: else if (__builtin_constant_p (bh) && (bh) == 0) \
667: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
668: : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
669: else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \
670: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
671: : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
672: else \
673: __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
674: : "=r" (sh), "=&r" (sl) \
675: : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
676: } while (0)
677: #define count_leading_zeros(count, x) \
678: __asm__ ("{cntlz|cntlzw} %0,%1" : "=r" (count) : "r" (x))
679: #define COUNT_LEADING_ZEROS_0 32
680: #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \
681: || defined (__ppc__) \
682: || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \
683: || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \
684: && CPU_FAMILY == PPC)
685: #define umul_ppmm(ph, pl, m0, m1) \
686: do { \
687: USItype __m0 = (m0), __m1 = (m1); \
688: __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
689: (pl) = __m0 * __m1; \
690: } while (0)
691: #define UMUL_TIME 15
692: #define smul_ppmm(ph, pl, m0, m1) \
693: do { \
694: SItype __m0 = (m0), __m1 = (m1); \
695: __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
696: (pl) = __m0 * __m1; \
697: } while (0)
698: #define SMUL_TIME 14
699: #define UDIV_TIME 120
700: #elif defined (_ARCH_PWR)
701: #define UMUL_TIME 8
702: #define smul_ppmm(xh, xl, m0, m1) \
703: __asm__ ("mul %0,%2,%3" : "=r" (xh), "=q" (xl) : "r" (m0), "r" (m1))
704: #define SMUL_TIME 4
705: #define sdiv_qrnnd(q, r, nh, nl, d) \
706: __asm__ ("div %0,%2,%4" : "=r" (q), "=q" (r) : "r" (nh), "1" (nl), "r" (d))
707: #define UDIV_TIME 100
708: #endif
709: #endif /* 32-bit POWER architecture variants. */
710:
711: /* We should test _IBMR2 here when we add assembly support for the system
712: vendor compilers. */
713: #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64
714: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
715: do { \
716: if (__builtin_constant_p (bh) && (bh) == 0) \
717: __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
718: : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
719: else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
720: __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
721: : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\
722: else \
723: __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
724: : "=r" (sh), "=&r" (sl) \
725: : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \
726: } while (0)
727: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
728: do { \
729: if (__builtin_constant_p (ah) && (ah) == 0) \
730: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
731: : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
732: else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \
733: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
734: : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\
735: else if (__builtin_constant_p (bh) && (bh) == 0) \
736: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
737: : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
738: else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \
739: __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
740: : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\
741: else \
742: __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
743: : "=r" (sh), "=&r" (sl) \
744: : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \
745: } while (0)
746: #define count_leading_zeros(count, x) \
747: __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x))
748: #define COUNT_LEADING_ZEROS_0 64
749: #define umul_ppmm(ph, pl, m0, m1) \
750: do { \
751: UDItype __m0 = (m0), __m1 = (m1); \
752: __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
753: (pl) = __m0 * __m1; \
754: } while (0)
755: #define UMUL_TIME 15
756: #define smul_ppmm(ph, pl, m0, m1) \
757: do { \
758: DItype __m0 = (m0), __m1 = (m1); \
759: __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \
760: (pl) = __m0 * __m1; \
761: } while (0)
762: #define SMUL_TIME 14 /* ??? */
763: #define UDIV_TIME 120 /* ??? */
764: #endif /* 64-bit PowerPC. */
765:
766: #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32
767: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
768: __asm__ ("a %1,%5\n\tae %0,%3" \
769: : "=r" ((USItype) (sh)), \
770: "=&r" ((USItype) (sl)) \
771: : "%0" ((USItype) (ah)), \
772: "r" ((USItype) (bh)), \
773: "%1" ((USItype) (al)), \
774: "r" ((USItype) (bl)))
775: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
776: __asm__ ("s %1,%5\n\tse %0,%3" \
777: : "=r" ((USItype) (sh)), \
778: "=&r" ((USItype) (sl)) \
779: : "0" ((USItype) (ah)), \
780: "r" ((USItype) (bh)), \
781: "1" ((USItype) (al)), \
782: "r" ((USItype) (bl)))
783: #define umul_ppmm(ph, pl, m0, m1) \
784: do { \
785: USItype __m0 = (m0), __m1 = (m1); \
786: __asm__ ( \
787: "s r2,r2\n" \
788: " mts r10,%2\n" \
789: " m r2,%3\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: " cas %0,r2,r0\n" \
806: " mfs r10,%1" \
807: : "=r" ((USItype) (ph)), \
808: "=r" ((USItype) (pl)) \
809: : "%r" (__m0), \
810: "r" (__m1) \
811: : "r2"); \
812: (ph) += ((((SItype) __m0 >> 31) & __m1) \
813: + (((SItype) __m1 >> 31) & __m0)); \
814: } while (0)
815: #define UMUL_TIME 20
816: #define UDIV_TIME 200
817: #define count_leading_zeros(count, x) \
818: do { \
819: if ((x) >= 0x10000) \
820: __asm__ ("clz %0,%1" \
821: : "=r" ((USItype) (count)) \
822: : "r" ((USItype) (x) >> 16)); \
823: else \
824: { \
825: __asm__ ("clz %0,%1" \
826: : "=r" ((USItype) (count)) \
827: : "r" ((USItype) (x))); \
828: (count) += 16; \
829: } \
830: } while (0)
831: #endif
832:
833: #if defined (__sh2__) && W_TYPE_SIZE == 32
834: #define umul_ppmm(w1, w0, u, v) \
835: __asm__ ( \
836: "dmulu.l %2,%3\n\tsts macl,%1\n\tsts mach,%0" \
837: : "=r" ((USItype)(w1)), \
838: "=r" ((USItype)(w0)) \
839: : "r" ((USItype)(u)), \
840: "r" ((USItype)(v)) \
841: : "macl", "mach")
842: #define UMUL_TIME 5
843: #endif
844:
845: #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32
846: #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v)
847: #define count_leading_zeros(count, x) \
848: do \
849: { \
850: UDItype x_ = (USItype)(x); \
851: SItype c_; \
852: \
853: __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \
854: (count) = c_ - 31; \
855: } \
856: while (0)
857: #define COUNT_LEADING_ZEROS_0 32
858: #endif
859:
860: #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \
861: && W_TYPE_SIZE == 32
862: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
863: __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \
864: : "=r" ((USItype) (sh)), \
865: "=&r" ((USItype) (sl)) \
866: : "%rJ" ((USItype) (ah)), \
867: "rI" ((USItype) (bh)), \
868: "%rJ" ((USItype) (al)), \
869: "rI" ((USItype) (bl)) \
870: __CLOBBER_CC)
871: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
872: __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \
873: : "=r" ((USItype) (sh)), \
874: "=&r" ((USItype) (sl)) \
875: : "rJ" ((USItype) (ah)), \
876: "rI" ((USItype) (bh)), \
877: "rJ" ((USItype) (al)), \
878: "rI" ((USItype) (bl)) \
879: __CLOBBER_CC)
880: #if defined (__sparc_v8__)
881: #define umul_ppmm(w1, w0, u, v) \
882: __asm__ ("umul %2,%3,%1;rd %%y,%0" \
883: : "=r" ((USItype) (w1)), \
884: "=r" ((USItype) (w0)) \
885: : "r" ((USItype) (u)), \
886: "r" ((USItype) (v)))
887: #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
888: __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\
889: : "=&r" ((USItype) (__q)), \
890: "=&r" ((USItype) (__r)) \
891: : "r" ((USItype) (__n1)), \
892: "r" ((USItype) (__n0)), \
893: "r" ((USItype) (__d)))
894: #else
895: #if defined (__sparclite__)
896: /* This has hardware multiply but not divide. It also has two additional
897: instructions scan (ffs from high bit) and divscc. */
898: #define umul_ppmm(w1, w0, u, v) \
899: __asm__ ("umul %2,%3,%1;rd %%y,%0" \
900: : "=r" ((USItype) (w1)), \
901: "=r" ((USItype) (w0)) \
902: : "r" ((USItype) (u)), \
903: "r" ((USItype) (v)))
904: #define udiv_qrnnd(q, r, n1, n0, d) \
905: __asm__ ("! Inlined udiv_qrnnd\n" \
906: " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \
907: " tst %%g0\n" \
908: " divscc %3,%4,%%g1\n" \
909: " divscc %%g1,%4,%%g1\n" \
910: " divscc %%g1,%4,%%g1\n" \
911: " divscc %%g1,%4,%%g1\n" \
912: " divscc %%g1,%4,%%g1\n" \
913: " divscc %%g1,%4,%%g1\n" \
914: " divscc %%g1,%4,%%g1\n" \
915: " divscc %%g1,%4,%%g1\n" \
916: " divscc %%g1,%4,%%g1\n" \
917: " divscc %%g1,%4,%%g1\n" \
918: " divscc %%g1,%4,%%g1\n" \
919: " divscc %%g1,%4,%%g1\n" \
920: " divscc %%g1,%4,%%g1\n" \
921: " divscc %%g1,%4,%%g1\n" \
922: " divscc %%g1,%4,%%g1\n" \
923: " divscc %%g1,%4,%%g1\n" \
924: " divscc %%g1,%4,%%g1\n" \
925: " divscc %%g1,%4,%%g1\n" \
926: " divscc %%g1,%4,%%g1\n" \
927: " divscc %%g1,%4,%%g1\n" \
928: " divscc %%g1,%4,%%g1\n" \
929: " divscc %%g1,%4,%%g1\n" \
930: " divscc %%g1,%4,%%g1\n" \
931: " divscc %%g1,%4,%%g1\n" \
932: " divscc %%g1,%4,%%g1\n" \
933: " divscc %%g1,%4,%%g1\n" \
934: " divscc %%g1,%4,%%g1\n" \
935: " divscc %%g1,%4,%%g1\n" \
936: " divscc %%g1,%4,%%g1\n" \
937: " divscc %%g1,%4,%%g1\n" \
938: " divscc %%g1,%4,%%g1\n" \
939: " divscc %%g1,%4,%0\n" \
940: " rd %%y,%1\n" \
941: " bl,a 1f\n" \
942: " add %1,%4,%1\n" \
943: "1: ! End of inline udiv_qrnnd" \
944: : "=r" ((USItype) (q)), \
945: "=r" ((USItype) (r)) \
946: : "r" ((USItype) (n1)), \
947: "r" ((USItype) (n0)), \
948: "rI" ((USItype) (d)) \
949: : "g1" __AND_CLOBBER_CC)
950: #define UDIV_TIME 37
951: #define count_leading_zeros(count, x) \
952: do { \
953: __asm__ ("scan %1,1,%0" \
954: : "=r" ((USItype) (count)) \
955: : "r" ((USItype) (x))); \
956: } while (0)
957: /* Early sparclites return 63 for an argument of 0, but they warn that future
958: implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0
959: undefined. */
960: #else
961: /* SPARC without integer multiplication and divide instructions.
962: (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */
963: #define umul_ppmm(w1, w0, u, v) \
964: __asm__ ("! Inlined umul_ppmm\n" \
965: " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\
966: " sra %3,31,%%o5 ! Don't move this insn\n" \
967: " and %2,%%o5,%%o5 ! Don't move this insn\n" \
968: " andcc %%g0,0,%%g1 ! Don't move this insn\n" \
969: " mulscc %%g1,%3,%%g1\n" \
970: " mulscc %%g1,%3,%%g1\n" \
971: " mulscc %%g1,%3,%%g1\n" \
972: " mulscc %%g1,%3,%%g1\n" \
973: " mulscc %%g1,%3,%%g1\n" \
974: " mulscc %%g1,%3,%%g1\n" \
975: " mulscc %%g1,%3,%%g1\n" \
976: " mulscc %%g1,%3,%%g1\n" \
977: " mulscc %%g1,%3,%%g1\n" \
978: " mulscc %%g1,%3,%%g1\n" \
979: " mulscc %%g1,%3,%%g1\n" \
980: " mulscc %%g1,%3,%%g1\n" \
981: " mulscc %%g1,%3,%%g1\n" \
982: " mulscc %%g1,%3,%%g1\n" \
983: " mulscc %%g1,%3,%%g1\n" \
984: " mulscc %%g1,%3,%%g1\n" \
985: " mulscc %%g1,%3,%%g1\n" \
986: " mulscc %%g1,%3,%%g1\n" \
987: " mulscc %%g1,%3,%%g1\n" \
988: " mulscc %%g1,%3,%%g1\n" \
989: " mulscc %%g1,%3,%%g1\n" \
990: " mulscc %%g1,%3,%%g1\n" \
991: " mulscc %%g1,%3,%%g1\n" \
992: " mulscc %%g1,%3,%%g1\n" \
993: " mulscc %%g1,%3,%%g1\n" \
994: " mulscc %%g1,%3,%%g1\n" \
995: " mulscc %%g1,%3,%%g1\n" \
996: " mulscc %%g1,%3,%%g1\n" \
997: " mulscc %%g1,%3,%%g1\n" \
998: " mulscc %%g1,%3,%%g1\n" \
999: " mulscc %%g1,%3,%%g1\n" \
1000: " mulscc %%g1,%3,%%g1\n" \
1001: " mulscc %%g1,0,%%g1\n" \
1002: " add %%g1,%%o5,%0\n" \
1003: " rd %%y,%1" \
1004: : "=r" ((USItype) (w1)), \
1005: "=r" ((USItype) (w0)) \
1006: : "%rI" ((USItype) (u)), \
1007: "r" ((USItype) (v)) \
1008: : "g1", "o5" __AND_CLOBBER_CC)
1009: #define UMUL_TIME 39 /* 39 instructions */
1010: /* It's quite necessary to add this much assembler for the sparc.
1011: The default udiv_qrnnd (in C) is more than 10 times slower! */
1012: #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \
1013: __asm__ ("! Inlined udiv_qrnnd\n" \
1014: " mov 32,%%g1\n" \
1015: " subcc %1,%2,%%g0\n" \
1016: "1: bcs 5f\n" \
1017: " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1018: " sub %1,%2,%1 ! this kills msb of n\n" \
1019: " addx %1,%1,%1 ! so this can't give carry\n" \
1020: " subcc %%g1,1,%%g1\n" \
1021: "2: bne 1b\n" \
1022: " subcc %1,%2,%%g0\n" \
1023: " bcs 3f\n" \
1024: " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \
1025: " b 3f\n" \
1026: " sub %1,%2,%1 ! this kills msb of n\n" \
1027: "4: sub %1,%2,%1\n" \
1028: "5: addxcc %1,%1,%1\n" \
1029: " bcc 2b\n" \
1030: " subcc %%g1,1,%%g1\n" \
1031: "! Got carry from n. Subtract next step to cancel this carry.\n" \
1032: " bne 4b\n" \
1033: " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \
1034: " sub %1,%2,%1\n" \
1035: "3: xnor %0,0,%0\n" \
1036: " ! End of inline udiv_qrnnd" \
1037: : "=&r" ((USItype) (__q)), \
1038: "=&r" ((USItype) (__r)) \
1039: : "r" ((USItype) (__d)), \
1040: "1" ((USItype) (__n1)), \
1041: "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC)
1042: #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */
1043: #endif /* __sparclite__ */
1044: #endif /* __sparc_v8__ */
1045: #endif /* sparc32 */
1046:
1047: #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \
1048: && W_TYPE_SIZE == 64
1049: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1050: __asm__ ("addcc %r4,%5,%1\n\t" \
1051: "add %r2,%3,%0\n\t" \
1052: "bcs,a,pn %%xcc, 1f\n\t" \
1053: "add %0, 1, %0\n" \
1054: "1:" \
1055: : "=r" ((UDItype)(sh)), \
1056: "=&r" ((UDItype)(sl)) \
1057: : "%rJ" ((UDItype)(ah)), \
1058: "rI" ((UDItype)(bh)), \
1059: "%rJ" ((UDItype)(al)), \
1060: "rI" ((UDItype)(bl)) \
1061: __CLOBBER_CC)
1062:
1063: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1064: __asm__ ("subcc %r4,%5,%1\n\t" \
1065: "sub %r2,%3,%0\n\t" \
1066: "bcs,a,pn %%xcc, 1f\n\t" \
1067: "sub %0, 1, %0\n\t" \
1068: "1:" \
1069: : "=r" ((UDItype)(sh)), \
1070: "=&r" ((UDItype)(sl)) \
1071: : "rJ" ((UDItype)(ah)), \
1072: "rI" ((UDItype)(bh)), \
1073: "rJ" ((UDItype)(al)), \
1074: "rI" ((UDItype)(bl)) \
1075: __CLOBBER_CC)
1076:
1077: #define umul_ppmm(wh, wl, u, v) \
1078: do { \
1079: UDItype tmp1, tmp2, tmp3, tmp4; \
1080: __asm__ __volatile__ ( \
1081: "srl %7,0,%3\n\t" \
1082: "mulx %3,%6,%1\n\t" \
1083: "srlx %6,32,%2\n\t" \
1084: "mulx %2,%3,%4\n\t" \
1085: "sllx %4,32,%5\n\t" \
1086: "srl %6,0,%3\n\t" \
1087: "sub %1,%5,%5\n\t" \
1088: "srlx %5,32,%5\n\t" \
1089: "addcc %4,%5,%4\n\t" \
1090: "srlx %7,32,%5\n\t" \
1091: "mulx %3,%5,%3\n\t" \
1092: "mulx %2,%5,%5\n\t" \
1093: "sethi %%hi(0x80000000),%2\n\t" \
1094: "addcc %4,%3,%4\n\t" \
1095: "srlx %4,32,%4\n\t" \
1096: "add %2,%2,%2\n\t" \
1097: "movcc %%xcc,%%g0,%2\n\t" \
1098: "addcc %5,%4,%5\n\t" \
1099: "sllx %3,32,%3\n\t" \
1100: "add %1,%3,%1\n\t" \
1101: "add %5,%2,%0" \
1102: : "=r" ((UDItype)(wh)), \
1103: "=&r" ((UDItype)(wl)), \
1104: "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \
1105: : "r" ((UDItype)(u)), \
1106: "r" ((UDItype)(v)) \
1107: __CLOBBER_CC); \
1108: } while (0)
1109: #define UMUL_TIME 96
1110: #define UDIV_TIME 230
1111: #endif /* sparc64 */
1112:
1113: #if defined (__vax__) && W_TYPE_SIZE == 32
1114: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1115: __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \
1116: : "=g" ((USItype) (sh)), \
1117: "=&g" ((USItype) (sl)) \
1118: : "%0" ((USItype) (ah)), \
1119: "g" ((USItype) (bh)), \
1120: "%1" ((USItype) (al)), \
1121: "g" ((USItype) (bl)))
1122: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1123: __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \
1124: : "=g" ((USItype) (sh)), \
1125: "=&g" ((USItype) (sl)) \
1126: : "0" ((USItype) (ah)), \
1127: "g" ((USItype) (bh)), \
1128: "1" ((USItype) (al)), \
1129: "g" ((USItype) (bl)))
1130: #define umul_ppmm(xh, xl, m0, m1) \
1131: do { \
1132: union { \
1133: UDItype __ll; \
1134: struct {USItype __l, __h;} __i; \
1135: } __xx; \
1136: USItype __m0 = (m0), __m1 = (m1); \
1137: __asm__ ("emul %1,%2,$0,%0" \
1138: : "=r" (__xx.__ll) \
1139: : "g" (__m0), \
1140: "g" (__m1)); \
1141: (xh) = __xx.__i.__h; \
1142: (xl) = __xx.__i.__l; \
1143: (xh) += ((((SItype) __m0 >> 31) & __m1) \
1144: + (((SItype) __m1 >> 31) & __m0)); \
1145: } while (0)
1146: #define sdiv_qrnnd(q, r, n1, n0, d) \
1147: do { \
1148: union {DItype __ll; \
1149: struct {SItype __l, __h;} __i; \
1150: } __xx; \
1151: __xx.__i.__h = n1; __xx.__i.__l = n0; \
1152: __asm__ ("ediv %3,%2,%0,%1" \
1153: : "=g" (q), "=g" (r) \
1154: : "g" (__xx.__ll), "g" (d)); \
1155: } while (0)
1156: #endif /* __vax__ */
1157:
1158: #if defined (__z8000__) && W_TYPE_SIZE == 16
1159: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1160: __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \
1161: : "=r" ((unsigned int)(sh)), \
1162: "=&r" ((unsigned int)(sl)) \
1163: : "%0" ((unsigned int)(ah)), \
1164: "r" ((unsigned int)(bh)), \
1165: "%1" ((unsigned int)(al)), \
1166: "rQR" ((unsigned int)(bl)))
1167: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1168: __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \
1169: : "=r" ((unsigned int)(sh)), \
1170: "=&r" ((unsigned int)(sl)) \
1171: : "0" ((unsigned int)(ah)), \
1172: "r" ((unsigned int)(bh)), \
1173: "1" ((unsigned int)(al)), \
1174: "rQR" ((unsigned int)(bl)))
1175: #define umul_ppmm(xh, xl, m0, m1) \
1176: do { \
1177: union {long int __ll; \
1178: struct {unsigned int __h, __l;} __i; \
1179: } __xx; \
1180: unsigned int __m0 = (m0), __m1 = (m1); \
1181: __asm__ ("mult %S0,%H3" \
1182: : "=r" (__xx.__i.__h), \
1183: "=r" (__xx.__i.__l) \
1184: : "%1" (__m0), \
1185: "rQR" (__m1)); \
1186: (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \
1187: (xh) += ((((signed int) __m0 >> 15) & __m1) \
1188: + (((signed int) __m1 >> 15) & __m0)); \
1189: } while (0)
1190: #endif /* __z8000__ */
1191:
1192: #endif /* __GNUC__ */
1193:
1194: /* If this machine has no inline assembler, use C macros. */
1195:
1196: #if !defined (add_ssaaaa)
1197: #define add_ssaaaa(sh, sl, ah, al, bh, bl) \
1198: do { \
1199: UWtype __x; \
1200: __x = (al) + (bl); \
1201: (sh) = (ah) + (bh) + (__x < (al)); \
1202: (sl) = __x; \
1203: } while (0)
1204: #endif
1205:
1206: #if !defined (sub_ddmmss)
1207: #define sub_ddmmss(sh, sl, ah, al, bh, bl) \
1208: do { \
1209: UWtype __x; \
1210: __x = (al) - (bl); \
1211: (sh) = (ah) - (bh) - (__x > (al)); \
1212: (sl) = __x; \
1213: } while (0)
1214: #endif
1215:
1216: /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of
1217: smul_ppmm. */
1218: #if !defined (umul_ppmm) && defined (smul_ppmm)
1219: #define umul_ppmm(w1, w0, u, v) \
1220: do { \
1221: UWtype __w1; \
1222: UWtype __xm0 = (u), __xm1 = (v); \
1223: smul_ppmm (__w1, w0, __xm0, __xm1); \
1224: (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \
1225: + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \
1226: } while (0)
1227: #endif
1228:
1229: /* If we still don't have umul_ppmm, define it using plain C. */
1230: #if !defined (umul_ppmm)
1231: #define umul_ppmm(w1, w0, u, v) \
1232: do { \
1233: UWtype __x0, __x1, __x2, __x3; \
1234: UHWtype __ul, __vl, __uh, __vh; \
1235: \
1236: __ul = __ll_lowpart (u); \
1237: __uh = __ll_highpart (u); \
1238: __vl = __ll_lowpart (v); \
1239: __vh = __ll_highpart (v); \
1240: \
1241: __x0 = (UWtype) __ul * __vl; \
1242: __x1 = (UWtype) __ul * __vh; \
1243: __x2 = (UWtype) __uh * __vl; \
1244: __x3 = (UWtype) __uh * __vh; \
1245: \
1246: __x1 += __ll_highpart (__x0);/* this can't give carry */ \
1247: __x1 += __x2; /* but this indeed can */ \
1248: if (__x1 < __x2) /* did we get it? */ \
1249: __x3 += __ll_B; /* yes, add it in the proper pos. */ \
1250: \
1251: (w1) = __x3 + __ll_highpart (__x1); \
1252: (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \
1253: } while (0)
1254: #endif
1255:
1256: #if !defined (__umulsidi3)
1257: #define __umulsidi3(u, v) \
1258: ({DWunion __w; \
1259: umul_ppmm (__w.s.high, __w.s.low, u, v); \
1260: __w.ll; })
1261: #endif
1262:
1263: /* Define this unconditionally, so it can be used for debugging. */
1264: #define __udiv_qrnnd_c(q, r, n1, n0, d) \
1265: do { \
1266: UWtype __d1, __d0, __q1, __q0; \
1267: UWtype __r1, __r0, __m; \
1268: __d1 = __ll_highpart (d); \
1269: __d0 = __ll_lowpart (d); \
1270: \
1271: __r1 = (n1) % __d1; \
1272: __q1 = (n1) / __d1; \
1273: __m = (UWtype) __q1 * __d0; \
1274: __r1 = __r1 * __ll_B | __ll_highpart (n0); \
1275: if (__r1 < __m) \
1276: { \
1277: __q1--, __r1 += (d); \
1278: if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\
1279: if (__r1 < __m) \
1280: __q1--, __r1 += (d); \
1281: } \
1282: __r1 -= __m; \
1283: \
1284: __r0 = __r1 % __d1; \
1285: __q0 = __r1 / __d1; \
1286: __m = (UWtype) __q0 * __d0; \
1287: __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
1288: if (__r0 < __m) \
1289: { \
1290: __q0--, __r0 += (d); \
1291: if (__r0 >= (d)) \
1292: if (__r0 < __m) \
1293: __q0--, __r0 += (d); \
1294: } \
1295: __r0 -= __m; \
1296: \
1297: (q) = (UWtype) __q1 * __ll_B | __q0; \
1298: (r) = __r0; \
1299: } while (0)
1300:
1301: /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through
1302: __udiv_w_sdiv (defined in libgcc or elsewhere). */
1303: #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd)
1304: #define udiv_qrnnd(q, r, nh, nl, d) \
1305: do { \
1306: USItype __r; \
1307: (q) = __udiv_w_sdiv (&__r, nh, nl, d); \
1308: (r) = __r; \
1309: } while (0)
1310: #endif
1311:
1312: /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */
1313: #if !defined (udiv_qrnnd)
1314: #define UDIV_NEEDS_NORMALIZATION 1
1315: #define udiv_qrnnd __udiv_qrnnd_c
1316: #endif
1317:
1318: #if !defined (count_leading_zeros)
1319: #define count_leading_zeros(count, x) \
1320: do { \
1321: UWtype __xr = (x); \
1322: UWtype __a; \
1323: \
1324: if (W_TYPE_SIZE <= 32) \
1325: { \
1326: __a = __xr < ((UWtype)1<<2*__BITS4) \
1327: ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \
1328: : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \
1329: } \
1330: else \
1331: { \
1332: for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \
1333: if (((__xr >> __a) & 0xff) != 0) \
1334: break; \
1335: } \
1336: \
1337: (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \
1338: } while (0)
1339: #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE
1340: #endif
1341:
1342: #if !defined (count_trailing_zeros)
1343: /* Define count_trailing_zeros using count_leading_zeros. The latter might be
1344: defined in asm, but if it is not, the C version above is good enough. */
1345: #define count_trailing_zeros(count, x) \
1346: do { \
1347: UWtype __ctz_x = (x); \
1348: UWtype __ctz_c; \
1349: count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \
1350: (count) = W_TYPE_SIZE - 1 - __ctz_c; \
1351: } while (0)
1352: #endif
1353:
1354: #ifndef UDIV_NEEDS_NORMALIZATION
1355: #define UDIV_NEEDS_NORMALIZATION 0
1356: #endif
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>