include/linux/math64.h

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