| 1 | |
| 2 | |
| 3 | * Copyright (C) 2003 Bernardo Innocenti <[email protected]>
|
| 4 | *
|
| 5 | * Based on former do_div() implementation from asm-parisc/div64.h:
|
| 6 | * Copyright (C) 1999 Hewlett-Packard Co
|
| 7 | * Copyright (C) 1999 David Mosberger-Tang <[email protected]>
|
| 8 | *
|
| 9 | *
|
| 10 | * Generic C version of 64bit/32bit division and modulo, with
|
| 11 | * 64bit result and 32bit remainder.
|
| 12 | *
|
| 13 | * The fast case for (n>>32 == 0) is handled inline by do_div().
|
| 14 | *
|
| 15 | * Code generated for this function might be very inefficient
|
| 16 | * for some CPUs. __div64_32() can be overridden by linking arch-specific
|
| 17 | * assembly versions such as arch/ppc/lib/div64.S and arch/sh/lib/div64.S
|
| 18 | * or by defining a preprocessor macro in arch/include/asm/div64.h.
|
| 19 |
|
| 20 |
|
| 21 | #include <linux/bitops.h>
|
| 22 | #include <linux/export.h>
|
| 23 | #include <linux/math.h>
|
| 24 | #include <linux/math64.h>
|
| 25 | #include <linux/minmax.h>
|
| 26 | #include <linux/log2.h>
|
| 27 |
|
| 28 |
|
| 29 | #if BITS_PER_LONG == 32
|
| 30 |
|
| 31 | #ifndef __div64_32
|
| 32 | uint32_t __attribute__((weak)) __div64_32(uint64_t *n, uint32_t base)
|
| 33 | {
|
| 34 | uint64_t rem = *n;
|
| 35 | uint64_t b = base;
|
| 36 | uint64_t res, d = 1;
|
| 37 | uint32_t high = rem >> 32;
|
| 38 |
|
| 39 |
|
| 40 | res = 0;
|
| 41 | if (high >= base) {
|
| 42 | high /= base;
|
| 43 | res = (uint64_t) high << 32;
|
| 44 | rem -= (uint64_t) (high*base) << 32;
|
| 45 | }
|
| 46 |
|
| 47 | while ((int64_t)b > 0 && b < rem) {
|
| 48 | b = b+b;
|
| 49 | d = d+d;
|
| 50 | }
|
| 51 |
|
| 52 | do {
|
| 53 | if (rem >= b) {
|
| 54 | rem -= b;
|
| 55 | res += d;
|
| 56 | }
|
| 57 | b >>= 1;
|
| 58 | d >>= 1;
|
| 59 | } while (d);
|
| 60 |
|
| 61 | *n = res;
|
| 62 | return rem;
|
| 63 | }
|
| 64 | EXPORT_SYMBOL(__div64_32);
|
| 65 | #endif
|
| 66 |
|
| 67 | #ifndef div_s64_rem
|
| 68 | s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder)
|
| 69 | {
|
| 70 | u64 quotient;
|
| 71 |
|
| 72 | if (dividend < 0) {
|
| 73 | quotient = div_u64_rem(-dividend, abs(divisor), (u32 *)remainder);
|
| 74 | *remainder = -*remainder;
|
| 75 | if (divisor > 0)
|
| 76 | quotient = -quotient;
|
| 77 | } else {
|
| 78 | quotient = div_u64_rem(dividend, abs(divisor), (u32 *)remainder);
|
| 79 | if (divisor < 0)
|
| 80 | quotient = -quotient;
|
| 81 | }
|
| 82 | return quotient;
|
| 83 | }
|
| 84 | EXPORT_SYMBOL(div_s64_rem);
|
| 85 | #endif
|
| 86 |
|
| 87 | |
| 88 | * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
|
| 89 | * @dividend: 64bit dividend
|
| 90 | * @divisor: 64bit divisor
|
| 91 | * @remainder: 64bit remainder
|
| 92 | *
|
| 93 | * This implementation is a comparable to algorithm used by div64_u64.
|
| 94 | * But this operation, which includes math for calculating the remainder,
|
| 95 | * is kept distinct to avoid slowing down the div64_u64 operation on 32bit
|
| 96 | * systems.
|
| 97 |
|
| 98 | #ifndef div64_u64_rem
|
| 99 | u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
|
| 100 | {
|
| 101 | u32 high = divisor >> 32;
|
| 102 | u64 quot;
|
| 103 |
|
| 104 | if (high == 0) {
|
| 105 | u32 rem32;
|
| 106 | quot = div_u64_rem(dividend, divisor, &rem32);
|
| 107 | *remainder = rem32;
|
| 108 | } else {
|
| 109 | int n = fls(high);
|
| 110 | quot = div_u64(dividend >> n, divisor >> n);
|
| 111 |
|
| 112 | if (quot != 0)
|
| 113 | quot--;
|
| 114 |
|
| 115 | *remainder = dividend - quot * divisor;
|
| 116 | if (*remainder >= divisor) {
|
| 117 | quot++;
|
| 118 | *remainder -= divisor;
|
| 119 | }
|
| 120 | }
|
| 121 |
|
| 122 | return quot;
|
| 123 | }
|
| 124 | EXPORT_SYMBOL(div64_u64_rem);
|
| 125 | #endif
|
| 126 |
|
| 127 | |
| 128 | * div64_u64 - unsigned 64bit divide with 64bit divisor
|
| 129 | * @dividend: 64bit dividend
|
| 130 | * @divisor: 64bit divisor
|
| 131 | *
|
| 132 | * This implementation is a modified version of the algorithm proposed
|
| 133 | * by the book 'Hacker's Delight'. The original source and full proof
|
| 134 | * can be found here and is available for use without restriction.
|
| 135 | *
|
| 136 | * 'http://www.hackersdelight.org/hdcodetxt/divDouble.c.txt'
|
| 137 |
|
| 138 | #ifndef div64_u64
|
| 139 | u64 div64_u64(u64 dividend, u64 divisor)
|
| 140 | {
|
| 141 | u32 high = divisor >> 32;
|
| 142 | u64 quot;
|
| 143 |
|
| 144 | if (high == 0) {
|
| 145 | quot = div_u64(dividend, divisor);
|
| 146 | } else {
|
| 147 | int n = fls(high);
|
| 148 | quot = div_u64(dividend >> n, divisor >> n);
|
| 149 |
|
| 150 | if (quot != 0)
|
| 151 | quot--;
|
| 152 | if ((dividend - quot * divisor) >= divisor)
|
| 153 | quot++;
|
| 154 | }
|
| 155 |
|
| 156 | return quot;
|
| 157 | }
|
| 158 | EXPORT_SYMBOL(div64_u64);
|
| 159 | #endif
|
| 160 |
|
| 161 | #ifndef div64_s64
|
| 162 | s64 div64_s64(s64 dividend, s64 divisor)
|
| 163 | {
|
| 164 | s64 quot, t;
|
| 165 |
|
| 166 | quot = div64_u64(abs(dividend), abs(divisor));
|
| 167 | t = (dividend ^ divisor) >> 63;
|
| 168 |
|
| 169 | return (quot ^ t) - t;
|
| 170 | }
|
| 171 | EXPORT_SYMBOL(div64_s64);
|
| 172 | #endif
|
| 173 |
|
| 174 | #endif
|
| 175 |
|
| 176 | |
| 177 | * Iterative div/mod for use when dividend is not expected to be much
|
| 178 | * bigger than divisor.
|
| 179 |
|
| 180 | u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
|
| 181 | {
|
| 182 | return __iter_div_u64_rem(dividend, divisor, remainder);
|
| 183 | }
|
| 184 | EXPORT_SYMBOL(iter_div_u64_rem);
|
| 185 |
|
| 186 | #ifndef mul_u64_u64_div_u64
|
| 187 | u64 mul_u64_u64_div_u64(u64 a, u64 b, u64 c)
|
| 188 | {
|
| 189 | if (ilog2(a) + ilog2(b) <= 62)
|
| 190 | return div64_u64(a * b, c);
|
| 191 |
|
| 192 | #if defined(__SIZEOF_INT128__)
|
| 193 |
|
| 194 |
|
| 195 | u128 prod = (u128)a * b;
|
| 196 | u64 n_lo = prod, n_hi = prod >> 64;
|
| 197 |
|
| 198 | #else
|
| 199 |
|
| 200 |
|
| 201 | u32 a_lo = a, a_hi = a >> 32, b_lo = b, b_hi = b >> 32;
|
| 202 | u64 x, y, z;
|
| 203 |
|
| 204 | x = (u64)a_lo * b_lo;
|
| 205 | y = (u64)a_lo * b_hi + (u32)(x >> 32);
|
| 206 | z = (u64)a_hi * b_hi + (u32)(y >> 32);
|
| 207 | y = (u64)a_hi * b_lo + (u32)y;
|
| 208 | z += (u32)(y >> 32);
|
| 209 | x = (y << 32) + (u32)x;
|
| 210 |
|
| 211 | u64 n_lo = x, n_hi = z;
|
| 212 |
|
| 213 | #endif
|
| 214 |
|
| 215 |
|
| 216 | #pragma GCC diagnostic push
|
| 217 | #pragma GCC diagnostic ignored "-Wdiv-by-zero"
|
| 218 | if (unlikely(c == 0))
|
| 219 | return 1/0;
|
| 220 | #pragma GCC diagnostic pop
|
| 221 |
|
| 222 | int shift = __builtin_ctzll(c);
|
| 223 |
|
| 224 |
|
| 225 | if ((n_hi >> shift) == 0) {
|
| 226 | u64 n = shift ? (n_lo >> shift) | (n_hi << (64 - shift)) : n_lo;
|
| 227 |
|
| 228 | return div64_u64(n, c >> shift);
|
| 229 | |
| 230 | * The remainder value if needed would be:
|
| 231 | * res = div64_u64_rem(n, c >> shift, &rem);
|
| 232 | * rem = (rem << shift) + (n_lo - (n << shift));
|
| 233 |
|
| 234 | }
|
| 235 |
|
| 236 | if (n_hi >= c) {
|
| 237 |
|
| 238 | return -1;
|
| 239 | }
|
| 240 |
|
| 241 |
|
| 242 |
|
| 243 | shift = __builtin_clzll(c);
|
| 244 | c <<= shift;
|
| 245 |
|
| 246 | int p = 64 + shift;
|
| 247 | u64 res = 0;
|
| 248 | bool carry;
|
| 249 |
|
| 250 | do {
|
| 251 | carry = n_hi >> 63;
|
| 252 | shift = carry ? 1 : __builtin_clzll(n_hi);
|
| 253 | if (p < shift)
|
| 254 | break;
|
| 255 | p -= shift;
|
| 256 | n_hi <<= shift;
|
| 257 | n_hi |= n_lo >> (64 - shift);
|
| 258 | n_lo <<= shift;
|
| 259 | if (carry || (n_hi >= c)) {
|
| 260 | n_hi -= c;
|
| 261 | res |= 1ULL << p;
|
| 262 | }
|
| 263 | } while (n_hi);
|
| 264 |
|
| 265 |
|
| 266 | return res;
|
| 267 | }
|
| 268 | EXPORT_SYMBOL(mul_u64_u64_div_u64);
|
| 269 | #endif
|