xref: /linux/include/linux/math64.h (revision 509d3f45847627f4c5cdce004c3ec79262b5239c)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MATH64_H
3 #define _LINUX_MATH64_H
4 
5 #include <linux/types.h>
6 #include <linux/math.h>
7 #include <asm/div64.h>
8 #include <vdso/math64.h>
9 
10 #if BITS_PER_LONG == 64
11 
12 #define div64_long(x, y) div64_s64((x), (y))
13 #define div64_ul(x, y)   div64_u64((x), (y))
14 
15 /**
16  * div_u64_rem - unsigned 64bit divide with 32bit divisor with remainder
17  * @dividend: unsigned 64bit dividend
18  * @divisor: unsigned 32bit divisor
19  * @remainder: pointer to unsigned 32bit remainder
20  *
21  * Return: sets ``*remainder``, then returns dividend / divisor
22  *
23  * This is commonly provided by 32bit archs to provide an optimized 64bit
24  * divide.
25  */
div_u64_rem(u64 dividend,u32 divisor,u32 * remainder)26 static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
27 {
28 	*remainder = dividend % divisor;
29 	return dividend / divisor;
30 }
31 
32 /**
33  * div_s64_rem - signed 64bit divide with 32bit divisor with remainder
34  * @dividend: signed 64bit dividend
35  * @divisor: signed 32bit divisor
36  * @remainder: pointer to signed 32bit remainder
37  *
38  * Return: sets ``*remainder``, then returns dividend / divisor
39  */
div_s64_rem(s64 dividend,s32 divisor,s32 * remainder)40 static inline s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder)
41 {
42 	*remainder = dividend % divisor;
43 	return dividend / divisor;
44 }
45 
46 /**
47  * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
48  * @dividend: unsigned 64bit dividend
49  * @divisor: unsigned 64bit divisor
50  * @remainder: pointer to unsigned 64bit remainder
51  *
52  * Return: sets ``*remainder``, then returns dividend / divisor
53  */
div64_u64_rem(u64 dividend,u64 divisor,u64 * remainder)54 static inline u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
55 {
56 	*remainder = dividend % divisor;
57 	return dividend / divisor;
58 }
59 
60 /**
61  * div64_u64 - unsigned 64bit divide with 64bit divisor
62  * @dividend: unsigned 64bit dividend
63  * @divisor: unsigned 64bit divisor
64  *
65  * Return: dividend / divisor
66  */
div64_u64(u64 dividend,u64 divisor)67 static inline u64 div64_u64(u64 dividend, u64 divisor)
68 {
69 	return dividend / divisor;
70 }
71 
72 /**
73  * div64_s64 - signed 64bit divide with 64bit divisor
74  * @dividend: signed 64bit dividend
75  * @divisor: signed 64bit divisor
76  *
77  * Return: dividend / divisor
78  */
div64_s64(s64 dividend,s64 divisor)79 static inline s64 div64_s64(s64 dividend, s64 divisor)
80 {
81 	return dividend / divisor;
82 }
83 
84 #elif BITS_PER_LONG == 32
85 
86 #define div64_long(x, y) div_s64((x), (y))
87 #define div64_ul(x, y)   div_u64((x), (y))
88 
89 #ifndef div_u64_rem
div_u64_rem(u64 dividend,u32 divisor,u32 * remainder)90 static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
91 {
92 	*remainder = do_div(dividend, divisor);
93 	return dividend;
94 }
95 #endif
96 
97 #ifndef div_s64_rem
98 extern s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder);
99 #endif
100 
101 #ifndef div64_u64_rem
102 extern u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder);
103 #endif
104 
105 #ifndef div64_u64
106 extern u64 div64_u64(u64 dividend, u64 divisor);
107 #endif
108 
109 #ifndef div64_s64
110 extern s64 div64_s64(s64 dividend, s64 divisor);
111 #endif
112 
113 #endif /* BITS_PER_LONG */
114 
115 /**
116  * div_u64 - unsigned 64bit divide with 32bit divisor
117  * @dividend: unsigned 64bit dividend
118  * @divisor: unsigned 32bit divisor
119  *
120  * This is the most common 64bit divide and should be used if possible,
121  * as many 32bit archs can optimize this variant better than a full 64bit
122  * divide.
123  *
124  * Return: dividend / divisor
125  */
126 #ifndef div_u64
div_u64(u64 dividend,u32 divisor)127 static inline u64 div_u64(u64 dividend, u32 divisor)
128 {
129 	u32 remainder;
130 	return div_u64_rem(dividend, divisor, &remainder);
131 }
132 #endif
133 
134 /**
135  * div_s64 - signed 64bit divide with 32bit divisor
136  * @dividend: signed 64bit dividend
137  * @divisor: signed 32bit divisor
138  *
139  * Return: dividend / divisor
140  */
141 #ifndef div_s64
div_s64(s64 dividend,s32 divisor)142 static inline s64 div_s64(s64 dividend, s32 divisor)
143 {
144 	s32 remainder;
145 	return div_s64_rem(dividend, divisor, &remainder);
146 }
147 #endif
148 
149 u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder);
150 
151 #ifndef mul_u32_u32
152 /*
153  * Many a GCC version messes this up and generates a 64x64 mult :-(
154  */
mul_u32_u32(u32 a,u32 b)155 static inline u64 mul_u32_u32(u32 a, u32 b)
156 {
157 	return (u64)a * b;
158 }
159 #endif
160 
161 #ifndef add_u64_u32
162 /*
163  * Many a GCC version also messes this up.
164  * Zero extending b and then spilling everything to stack.
165  */
add_u64_u32(u64 a,u32 b)166 static inline u64 add_u64_u32(u64 a, u32 b)
167 {
168 	return a + b;
169 }
170 #endif
171 
172 #if defined(CONFIG_ARCH_SUPPORTS_INT128) && defined(__SIZEOF_INT128__)
173 
174 #ifndef mul_u64_u32_shr
mul_u64_u32_shr(u64 a,u32 mul,unsigned int shift)175 static __always_inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
176 {
177 	return (u64)(((unsigned __int128)a * mul) >> shift);
178 }
179 #endif /* mul_u64_u32_shr */
180 
181 #ifndef mul_u64_u64_shr
mul_u64_u64_shr(u64 a,u64 mul,unsigned int shift)182 static __always_inline u64 mul_u64_u64_shr(u64 a, u64 mul, unsigned int shift)
183 {
184 	return (u64)(((unsigned __int128)a * mul) >> shift);
185 }
186 #endif /* mul_u64_u64_shr */
187 
188 #else
189 
190 #ifndef mul_u64_u32_shr
mul_u64_u32_shr(u64 a,u32 mul,unsigned int shift)191 static __always_inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
192 {
193 	u32 ah = a >> 32, al = a;
194 	u64 ret;
195 
196 	ret = mul_u32_u32(al, mul) >> shift;
197 	if (ah)
198 		ret += mul_u32_u32(ah, mul) << (32 - shift);
199 	return ret;
200 }
201 #endif /* mul_u64_u32_shr */
202 
203 #ifndef mul_u64_u64_shr
mul_u64_u64_shr(u64 a,u64 b,unsigned int shift)204 static inline u64 mul_u64_u64_shr(u64 a, u64 b, unsigned int shift)
205 {
206 	union {
207 		u64 ll;
208 		struct {
209 #ifdef __BIG_ENDIAN
210 			u32 high, low;
211 #else
212 			u32 low, high;
213 #endif
214 		} l;
215 	} rl, rm, rn, rh, a0, b0;
216 	u64 c;
217 
218 	a0.ll = a;
219 	b0.ll = b;
220 
221 	rl.ll = mul_u32_u32(a0.l.low, b0.l.low);
222 	rm.ll = mul_u32_u32(a0.l.low, b0.l.high);
223 	rn.ll = mul_u32_u32(a0.l.high, b0.l.low);
224 	rh.ll = mul_u32_u32(a0.l.high, b0.l.high);
225 
226 	/*
227 	 * Each of these lines computes a 64-bit intermediate result into "c",
228 	 * starting at bits 32-95.  The low 32-bits go into the result of the
229 	 * multiplication, the high 32-bits are carried into the next step.
230 	 */
231 	rl.l.high = c = (u64)rl.l.high + rm.l.low + rn.l.low;
232 	rh.l.low = c = (c >> 32) + rm.l.high + rn.l.high + rh.l.low;
233 	rh.l.high = (c >> 32) + rh.l.high;
234 
235 	/*
236 	 * The 128-bit result of the multiplication is in rl.ll and rh.ll,
237 	 * shift it right and throw away the high part of the result.
238 	 */
239 	if (shift == 0)
240 		return rl.ll;
241 	if (shift < 64)
242 		return (rl.ll >> shift) | (rh.ll << (64 - shift));
243 	return rh.ll >> (shift & 63);
244 }
245 #endif /* mul_u64_u64_shr */
246 
247 #endif
248 
249 #ifndef mul_s64_u64_shr
mul_s64_u64_shr(s64 a,u64 b,unsigned int shift)250 static inline u64 mul_s64_u64_shr(s64 a, u64 b, unsigned int shift)
251 {
252 	u64 ret;
253 
254 	/*
255 	 * Extract the sign before the multiplication and put it back
256 	 * afterwards if needed.
257 	 */
258 	ret = mul_u64_u64_shr(abs(a), b, shift);
259 
260 	if (a < 0)
261 		ret = -((s64) ret);
262 
263 	return ret;
264 }
265 #endif /* mul_s64_u64_shr */
266 
267 #ifndef mul_u64_u32_div
mul_u64_u32_div(u64 a,u32 mul,u32 divisor)268 static inline u64 mul_u64_u32_div(u64 a, u32 mul, u32 divisor)
269 {
270 	union {
271 		u64 ll;
272 		struct {
273 #ifdef __BIG_ENDIAN
274 			u32 high, low;
275 #else
276 			u32 low, high;
277 #endif
278 		} l;
279 	} u, rl, rh;
280 
281 	u.ll = a;
282 	rl.ll = mul_u32_u32(u.l.low, mul);
283 	rh.ll = mul_u32_u32(u.l.high, mul) + rl.l.high;
284 
285 	/* Bits 32-63 of the result will be in rh.l.low. */
286 	rl.l.high = do_div(rh.ll, divisor);
287 
288 	/* Bits 0-31 of the result will be in rl.l.low.	*/
289 	do_div(rl.ll, divisor);
290 
291 	rl.l.high = rh.l.low;
292 	return rl.ll;
293 }
294 #endif /* mul_u64_u32_div */
295 
296 /**
297  * mul_u64_add_u64_div_u64 - unsigned 64bit multiply, add, and divide
298  * @a: first unsigned 64bit multiplicand
299  * @b: second unsigned 64bit multiplicand
300  * @c: unsigned 64bit addend
301  * @d: unsigned 64bit divisor
302  *
303  * Multiply two 64bit values together to generate a 128bit product
304  * add a third value and then divide by a fourth.
305  * The Generic code divides by 0 if @d is zero and returns ~0 on overflow.
306  * Architecture specific code may trap on zero or overflow.
307  *
308  * Return: (@a * @b + @c) / @d
309  */
310 u64 mul_u64_add_u64_div_u64(u64 a, u64 b, u64 c, u64 d);
311 
312 /**
313  * mul_u64_u64_div_u64 - unsigned 64bit multiply and divide
314  * @a: first unsigned 64bit multiplicand
315  * @b: second unsigned 64bit multiplicand
316  * @d: unsigned 64bit divisor
317  *
318  * Multiply two 64bit values together to generate a 128bit product
319  * and then divide by a third value.
320  * The Generic code divides by 0 if @d is zero and returns ~0 on overflow.
321  * Architecture specific code may trap on zero or overflow.
322  *
323  * Return: @a * @b / @d
324  */
325 #define mul_u64_u64_div_u64(a, b, d) mul_u64_add_u64_div_u64(a, b, 0, d)
326 
327 /**
328  * mul_u64_u64_div_u64_roundup - unsigned 64bit multiply and divide rounded up
329  * @a: first unsigned 64bit multiplicand
330  * @b: second unsigned 64bit multiplicand
331  * @d: unsigned 64bit divisor
332  *
333  * Multiply two 64bit values together to generate a 128bit product
334  * and then divide and round up.
335  * The Generic code divides by 0 if @d is zero and returns ~0 on overflow.
336  * Architecture specific code may trap on zero or overflow.
337  *
338  * Return: (@a * @b + @d - 1) / @d
339  */
340 #define mul_u64_u64_div_u64_roundup(a, b, d) \
341 	({ u64 _tmp = (d); mul_u64_add_u64_div_u64(a, b, _tmp - 1, _tmp); })
342 
343 
344 /**
345  * DIV64_U64_ROUND_UP - unsigned 64bit divide with 64bit divisor rounded up
346  * @ll: unsigned 64bit dividend
347  * @d: unsigned 64bit divisor
348  *
349  * Divide unsigned 64bit dividend by unsigned 64bit divisor
350  * and round up.
351  *
352  * Return: dividend / divisor rounded up
353  */
354 #define DIV64_U64_ROUND_UP(ll, d)	\
355 	({ u64 _tmp = (d); div64_u64((ll) + _tmp - 1, _tmp); })
356 
357 /**
358  * DIV_U64_ROUND_UP - unsigned 64bit divide with 32bit divisor rounded up
359  * @ll: unsigned 64bit dividend
360  * @d: unsigned 32bit divisor
361  *
362  * Divide unsigned 64bit dividend by unsigned 32bit divisor
363  * and round up.
364  *
365  * Return: dividend / divisor rounded up
366  */
367 #define DIV_U64_ROUND_UP(ll, d)		\
368 	({ u32 _tmp = (d); div_u64((ll) + _tmp - 1, _tmp); })
369 
370 /**
371  * DIV64_U64_ROUND_CLOSEST - unsigned 64bit divide with 64bit divisor rounded to nearest integer
372  * @dividend: unsigned 64bit dividend
373  * @divisor: unsigned 64bit divisor
374  *
375  * Divide unsigned 64bit dividend by unsigned 64bit divisor
376  * and round to closest integer.
377  *
378  * Return: dividend / divisor rounded to nearest integer
379  */
380 #define DIV64_U64_ROUND_CLOSEST(dividend, divisor)	\
381 	({ u64 _tmp = (divisor); div64_u64((dividend) + _tmp / 2, _tmp); })
382 
383 /**
384  * DIV_U64_ROUND_CLOSEST - unsigned 64bit divide with 32bit divisor rounded to nearest integer
385  * @dividend: unsigned 64bit dividend
386  * @divisor: unsigned 32bit divisor
387  *
388  * Divide unsigned 64bit dividend by unsigned 32bit divisor
389  * and round to closest integer.
390  *
391  * Return: dividend / divisor rounded to nearest integer
392  */
393 #define DIV_U64_ROUND_CLOSEST(dividend, divisor)	\
394 	({ u32 _tmp = (divisor); div_u64((u64)(dividend) + _tmp / 2, _tmp); })
395 
396 /**
397  * DIV_S64_ROUND_CLOSEST - signed 64bit divide with 32bit divisor rounded to nearest integer
398  * @dividend: signed 64bit dividend
399  * @divisor: signed 32bit divisor
400  *
401  * Divide signed 64bit dividend by signed 32bit divisor
402  * and round to closest integer.
403  *
404  * Return: dividend / divisor rounded to nearest integer
405  */
406 #define DIV_S64_ROUND_CLOSEST(dividend, divisor)(	\
407 {							\
408 	s64 __x = (dividend);				\
409 	s32 __d = (divisor);				\
410 	((__x > 0) == (__d > 0)) ?			\
411 		div_s64((__x + (__d / 2)), __d) :	\
412 		div_s64((__x - (__d / 2)), __d);	\
413 }							\
414 )
415 
416 /**
417  * roundup_u64 - Round up a 64bit value to the next specified 32bit multiple
418  * @x: the value to up
419  * @y: 32bit multiple to round up to
420  *
421  * Rounds @x to the next multiple of @y. For 32bit @x values, see roundup and
422  * the faster round_up() for powers of 2.
423  *
424  * Return: rounded up value.
425  */
roundup_u64(u64 x,u32 y)426 static inline u64 roundup_u64(u64 x, u32 y)
427 {
428 	return DIV_U64_ROUND_UP(x, y) * y;
429 }
430 #endif /* _LINUX_MATH64_H */
431