xref: /linux-6.15/include/linux/math64.h (revision 9d64fc08)
1 #ifndef _LINUX_MATH64_H
2 #define _LINUX_MATH64_H
3 
4 #include <linux/types.h>
5 #include <asm/div64.h>
6 
7 #if BITS_PER_LONG == 64
8 
9 #define div64_long(x, y) div64_s64((x), (y))
10 #define div64_ul(x, y)   div64_u64((x), (y))
11 
12 /**
13  * div_u64_rem - unsigned 64bit divide with 32bit divisor with remainder
14  *
15  * This is commonly provided by 32bit archs to provide an optimized 64bit
16  * divide.
17  */
18 static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
19 {
20 	*remainder = dividend % divisor;
21 	return dividend / divisor;
22 }
23 
24 /**
25  * div_s64_rem - signed 64bit divide with 32bit divisor with remainder
26  */
27 static inline s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder)
28 {
29 	*remainder = dividend % divisor;
30 	return dividend / divisor;
31 }
32 
33 /**
34  * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
35  */
36 static inline u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
37 {
38 	*remainder = dividend % divisor;
39 	return dividend / divisor;
40 }
41 
42 /**
43  * div64_u64 - unsigned 64bit divide with 64bit divisor
44  */
45 static inline u64 div64_u64(u64 dividend, u64 divisor)
46 {
47 	return dividend / divisor;
48 }
49 
50 /**
51  * div64_s64 - signed 64bit divide with 64bit divisor
52  */
53 static inline s64 div64_s64(s64 dividend, s64 divisor)
54 {
55 	return dividend / divisor;
56 }
57 
58 #elif BITS_PER_LONG == 32
59 
60 #define div64_long(x, y) div_s64((x), (y))
61 #define div64_ul(x, y)   div_u64((x), (y))
62 
63 #ifndef div_u64_rem
64 static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
65 {
66 	*remainder = do_div(dividend, divisor);
67 	return dividend;
68 }
69 #endif
70 
71 #ifndef div_s64_rem
72 extern s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder);
73 #endif
74 
75 #ifndef div64_u64_rem
76 extern u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder);
77 #endif
78 
79 #ifndef div64_u64
80 extern u64 div64_u64(u64 dividend, u64 divisor);
81 #endif
82 
83 #ifndef div64_s64
84 extern s64 div64_s64(s64 dividend, s64 divisor);
85 #endif
86 
87 #endif /* BITS_PER_LONG */
88 
89 /**
90  * div_u64 - unsigned 64bit divide with 32bit divisor
91  *
92  * This is the most common 64bit divide and should be used if possible,
93  * as many 32bit archs can optimize this variant better than a full 64bit
94  * divide.
95  */
96 #ifndef div_u64
97 static inline u64 div_u64(u64 dividend, u32 divisor)
98 {
99 	u32 remainder;
100 	return div_u64_rem(dividend, divisor, &remainder);
101 }
102 #endif
103 
104 /**
105  * div_s64 - signed 64bit divide with 32bit divisor
106  */
107 #ifndef div_s64
108 static inline s64 div_s64(s64 dividend, s32 divisor)
109 {
110 	s32 remainder;
111 	return div_s64_rem(dividend, divisor, &remainder);
112 }
113 #endif
114 
115 u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder);
116 
117 static __always_inline u32
118 __iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
119 {
120 	u32 ret = 0;
121 
122 	while (dividend >= divisor) {
123 		/* The following asm() prevents the compiler from
124 		   optimising this loop into a modulo operation.  */
125 		asm("" : "+rm"(dividend));
126 
127 		dividend -= divisor;
128 		ret++;
129 	}
130 
131 	*remainder = dividend;
132 
133 	return ret;
134 }
135 
136 #ifndef mul_u32_u32
137 /*
138  * Many a GCC version messes this up and generates a 64x64 mult :-(
139  */
140 static inline u64 mul_u32_u32(u32 a, u32 b)
141 {
142 	return (u64)a * b;
143 }
144 #endif
145 
146 #if defined(CONFIG_ARCH_SUPPORTS_INT128) && defined(__SIZEOF_INT128__)
147 
148 #ifndef mul_u64_u32_shr
149 static inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
150 {
151 	return (u64)(((unsigned __int128)a * mul) >> shift);
152 }
153 #endif /* mul_u64_u32_shr */
154 
155 #ifndef mul_u64_u64_shr
156 static inline u64 mul_u64_u64_shr(u64 a, u64 mul, unsigned int shift)
157 {
158 	return (u64)(((unsigned __int128)a * mul) >> shift);
159 }
160 #endif /* mul_u64_u64_shr */
161 
162 #else
163 
164 #ifndef mul_u64_u32_shr
165 static inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
166 {
167 	u32 ah, al;
168 	u64 ret;
169 
170 	al = a;
171 	ah = a >> 32;
172 
173 	ret = mul_u32_u32(al, mul) >> shift;
174 	if (ah)
175 		ret += mul_u32_u32(ah, mul) << (32 - shift);
176 
177 	return ret;
178 }
179 #endif /* mul_u64_u32_shr */
180 
181 #ifndef mul_u64_u64_shr
182 static inline u64 mul_u64_u64_shr(u64 a, u64 b, unsigned int shift)
183 {
184 	union {
185 		u64 ll;
186 		struct {
187 #ifdef __BIG_ENDIAN
188 			u32 high, low;
189 #else
190 			u32 low, high;
191 #endif
192 		} l;
193 	} rl, rm, rn, rh, a0, b0;
194 	u64 c;
195 
196 	a0.ll = a;
197 	b0.ll = b;
198 
199 	rl.ll = mul_u32_u32(a0.l.low, b0.l.low);
200 	rm.ll = mul_u32_u32(a0.l.low, b0.l.high);
201 	rn.ll = mul_u32_u32(a0.l.high, b0.l.low);
202 	rh.ll = mul_u32_u32(a0.l.high, b0.l.high);
203 
204 	/*
205 	 * Each of these lines computes a 64-bit intermediate result into "c",
206 	 * starting at bits 32-95.  The low 32-bits go into the result of the
207 	 * multiplication, the high 32-bits are carried into the next step.
208 	 */
209 	rl.l.high = c = (u64)rl.l.high + rm.l.low + rn.l.low;
210 	rh.l.low = c = (c >> 32) + rm.l.high + rn.l.high + rh.l.low;
211 	rh.l.high = (c >> 32) + rh.l.high;
212 
213 	/*
214 	 * The 128-bit result of the multiplication is in rl.ll and rh.ll,
215 	 * shift it right and throw away the high part of the result.
216 	 */
217 	if (shift == 0)
218 		return rl.ll;
219 	if (shift < 64)
220 		return (rl.ll >> shift) | (rh.ll << (64 - shift));
221 	return rh.ll >> (shift & 63);
222 }
223 #endif /* mul_u64_u64_shr */
224 
225 #endif
226 
227 #ifndef mul_u64_u32_div
228 static inline u64 mul_u64_u32_div(u64 a, u32 mul, u32 divisor)
229 {
230 	union {
231 		u64 ll;
232 		struct {
233 #ifdef __BIG_ENDIAN
234 			u32 high, low;
235 #else
236 			u32 low, high;
237 #endif
238 		} l;
239 	} u, rl, rh;
240 
241 	u.ll = a;
242 	rl.ll = mul_u32_u32(u.l.low, mul);
243 	rh.ll = mul_u32_u32(u.l.high, mul) + rl.l.high;
244 
245 	/* Bits 32-63 of the result will be in rh.l.low. */
246 	rl.l.high = do_div(rh.ll, divisor);
247 
248 	/* Bits 0-31 of the result will be in rl.l.low.	*/
249 	do_div(rl.ll, divisor);
250 
251 	rl.l.high = rh.l.low;
252 	return rl.ll;
253 }
254 #endif /* mul_u64_u32_div */
255 
256 #endif /* _LINUX_MATH64_H */
257