summaryrefslogtreecommitdiffstats
path: root/src/util/fast_idiv_by_const.c
blob: 0bc9b60878bd050384790596f80f739c046163b8 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
/*
 * Copyright © 2018 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

/* Imported from:
 *   https://raw.githubusercontent.com/ridiculousfish/libdivide/master/divide_by_constants_codegen_reference.c
 * Paper:
 *   http://ridiculousfish.com/files/faster_unsigned_division_by_constants.pdf
 *
 * The author, ridiculous_fish, wrote:
 *
 *  ''Reference implementations of computing and using the "magic number"
 *    approach to dividing by constants, including codegen instructions.
 *    The unsigned division incorporates the "round down" optimization per
 *    ridiculous_fish.
 *
 *    This is free and unencumbered software. Any copyright is dedicated
 *    to the Public Domain.''
 */

#include "fast_idiv_by_const.h"
#include "u_math.h"
#include <limits.h>
#include <assert.h>

/* uint_t and sint_t can be replaced by different integer types and the code
 * will work as-is. The only requirement is that sizeof(uintN) == sizeof(intN).
 */

struct util_fast_udiv_info
util_compute_fast_udiv_info(uint_t D, unsigned num_bits)
{
   /* The numerator must fit in a uint_t */
   assert(num_bits > 0 && num_bits <= sizeof(uint_t) * CHAR_BIT);
   assert(D != 0);

   /* The eventual result */
   struct util_fast_udiv_info result;

   /* Bits in a uint_t */
   const unsigned UINT_BITS = sizeof(uint_t) * CHAR_BIT;

   /* The extra shift implicit in the difference between UINT_BITS and num_bits
    */
   const unsigned extra_shift = UINT_BITS - num_bits;

   /* The initial power of 2 is one less than the first one that can possibly
    * work.
    */
   const uint_t initial_power_of_2 = (uint_t)1 << (UINT_BITS-1);

   /* The remainder and quotient of our power of 2 divided by d */
   uint_t quotient = initial_power_of_2 / D;
   uint_t remainder = initial_power_of_2 % D;

   /* ceil(log_2 D) */
   unsigned ceil_log_2_D;

   /* The magic info for the variant "round down" algorithm */
   uint_t down_multiplier = 0;
   unsigned down_exponent = 0;
   int has_magic_down = 0;

   /* Compute ceil(log_2 D) */
   ceil_log_2_D = 0;
   uint_t tmp;
   for (tmp = D; tmp > 0; tmp >>= 1)
      ceil_log_2_D += 1;


   /* Begin a loop that increments the exponent, until we find a power of 2
    * that works.
    */
   unsigned exponent;
   for (exponent = 0; ; exponent++) {
      /* Quotient and remainder is from previous exponent; compute it for this
       * exponent.
       */
      if (remainder >= D - remainder) {
         /* Doubling remainder will wrap around D */
         quotient = quotient * 2 + 1;
         remainder = remainder * 2 - D;
      } else {
         /* Remainder will not wrap */
         quotient = quotient * 2;
         remainder = remainder * 2;
      }

      /* We're done if this exponent works for the round_up algorithm.
       * Note that exponent may be larger than the maximum shift supported,
       * so the check for >= ceil_log_2_D is critical.
       */
      if ((exponent + extra_shift >= ceil_log_2_D) ||
          (D - remainder) <= ((uint_t)1 << (exponent + extra_shift)))
         break;

      /* Set magic_down if we have not set it yet and this exponent works for
       * the round_down algorithm
       */
      if (!has_magic_down &&
          remainder <= ((uint_t)1 << (exponent + extra_shift))) {
         has_magic_down = 1;
         down_multiplier = quotient;
         down_exponent = exponent;
      }
   }

   if (exponent < ceil_log_2_D) {
      /* magic_up is efficient */
      result.multiplier = quotient + 1;
      result.pre_shift = 0;
      result.post_shift = exponent;
      result.increment = 0;
   } else if (D & 1) {
      /* Odd divisor, so use magic_down, which must have been set */
      assert(has_magic_down);
      result.multiplier = down_multiplier;
      result.pre_shift = 0;
      result.post_shift = down_exponent;
      result.increment = 1;
   } else {
      /* Even divisor, so use a prefix-shifted dividend */
      unsigned pre_shift = 0;
      uint_t shifted_D = D;
      while ((shifted_D & 1) == 0) {
         shifted_D >>= 1;
         pre_shift += 1;
      }
      result = util_compute_fast_udiv_info(shifted_D, num_bits - pre_shift);
      /* expect no increment or pre_shift in this path */
      assert(result.increment == 0 && result.pre_shift == 0);
      result.pre_shift = pre_shift;
   }
   return result;
}

struct util_fast_sdiv_info
util_compute_fast_sdiv_info(sint_t D)
{
   /* D must not be zero. */
   assert(D != 0);
   /* The result is not correct for these divisors. */
   assert(D != 1 && D != -1);

   /* Our result */
   struct util_fast_sdiv_info result;

   /* Bits in an sint_t */
   const unsigned SINT_BITS = sizeof(sint_t) * CHAR_BIT;

   /* Absolute value of D (we know D is not the most negative value since
    * that's a power of 2)
    */
   const uint_t abs_d = (D < 0 ? -D : D);

   /* The initial power of 2 is one less than the first one that can possibly
    * work */
   /* "two31" in Warren */
   unsigned exponent = SINT_BITS - 1;
   const uint_t initial_power_of_2 = (uint_t)1 << exponent;

   /* Compute the absolute value of our "test numerator,"
    * which is the largest dividend whose remainder with d is d-1.
    * This is called anc in Warren.
    */
   const uint_t tmp = initial_power_of_2 + (D < 0);
   const uint_t abs_test_numer = tmp - 1 - tmp % abs_d;

   /* Initialize our quotients and remainders (q1, r1, q2, r2 in Warren) */
   uint_t quotient1 = initial_power_of_2 / abs_test_numer;
   uint_t remainder1 = initial_power_of_2 % abs_test_numer;
   uint_t quotient2 = initial_power_of_2 / abs_d;
   uint_t remainder2 = initial_power_of_2 % abs_d;
   uint_t delta;

   /* Begin our loop */
   do {
      /* Update the exponent */
      exponent++;

      /* Update quotient1 and remainder1 */
      quotient1 *= 2;
      remainder1 *= 2;
      if (remainder1 >= abs_test_numer) {
         quotient1 += 1;
         remainder1 -= abs_test_numer;
      }

      /* Update quotient2 and remainder2 */
      quotient2 *= 2;
      remainder2 *= 2;
      if (remainder2 >= abs_d) {
         quotient2 += 1;
         remainder2 -= abs_d;
      }

      /* Keep going as long as (2**exponent) / abs_d <= delta */
      delta = abs_d - remainder2;
   } while (quotient1 < delta || (quotient1 == delta && remainder1 == 0));

   result.multiplier = quotient2 + 1;
   if (D < 0) result.multiplier = -result.multiplier;
   result.shift = exponent - SINT_BITS;
   return result;
}