Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* |
| 2 | * benchmark.c: |
| 3 | * Author: Konstantin Khlebnikov <koct9i@gmail.com> |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify it |
| 6 | * under the terms and conditions of the GNU General Public License, |
| 7 | * version 2, as published by the Free Software Foundation. |
| 8 | * |
| 9 | * This program is distributed in the hope it will be useful, but WITHOUT |
| 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
| 12 | * more details. |
| 13 | */ |
| 14 | #include <linux/radix-tree.h> |
| 15 | #include <linux/slab.h> |
| 16 | #include <linux/errno.h> |
| 17 | #include <time.h> |
| 18 | #include "test.h" |
| 19 | |
| 20 | #define for_each_index(i, base, order) \ |
| 21 | for (i = base; i < base + (1 << order); i++) |
| 22 | |
| 23 | #define NSEC_PER_SEC 1000000000L |
| 24 | |
| 25 | static long long benchmark_iter(struct radix_tree_root *root, bool tagged) |
| 26 | { |
| 27 | volatile unsigned long sink = 0; |
| 28 | struct radix_tree_iter iter; |
| 29 | struct timespec start, finish; |
| 30 | long long nsec; |
| 31 | int l, loops = 1; |
| 32 | void **slot; |
| 33 | |
| 34 | #ifdef BENCHMARK |
| 35 | again: |
| 36 | #endif |
| 37 | clock_gettime(CLOCK_MONOTONIC, &start); |
| 38 | for (l = 0; l < loops; l++) { |
| 39 | if (tagged) { |
| 40 | radix_tree_for_each_tagged(slot, root, &iter, 0, 0) |
| 41 | sink ^= (unsigned long)slot; |
| 42 | } else { |
| 43 | radix_tree_for_each_slot(slot, root, &iter, 0) |
| 44 | sink ^= (unsigned long)slot; |
| 45 | } |
| 46 | } |
| 47 | clock_gettime(CLOCK_MONOTONIC, &finish); |
| 48 | |
| 49 | nsec = (finish.tv_sec - start.tv_sec) * NSEC_PER_SEC + |
| 50 | (finish.tv_nsec - start.tv_nsec); |
| 51 | |
| 52 | #ifdef BENCHMARK |
| 53 | if (loops == 1 && nsec * 5 < NSEC_PER_SEC) { |
| 54 | loops = NSEC_PER_SEC / nsec / 4 + 1; |
| 55 | goto again; |
| 56 | } |
| 57 | #endif |
| 58 | |
| 59 | nsec /= loops; |
| 60 | return nsec; |
| 61 | } |
| 62 | |
| 63 | static void benchmark_insert(struct radix_tree_root *root, |
| 64 | unsigned long size, unsigned long step, int order) |
| 65 | { |
| 66 | struct timespec start, finish; |
| 67 | unsigned long index; |
| 68 | long long nsec; |
| 69 | |
| 70 | clock_gettime(CLOCK_MONOTONIC, &start); |
| 71 | |
| 72 | for (index = 0 ; index < size ; index += step) |
| 73 | item_insert_order(root, index, order); |
| 74 | |
| 75 | clock_gettime(CLOCK_MONOTONIC, &finish); |
| 76 | |
| 77 | nsec = (finish.tv_sec - start.tv_sec) * NSEC_PER_SEC + |
| 78 | (finish.tv_nsec - start.tv_nsec); |
| 79 | |
| 80 | printv(2, "Size: %8ld, step: %8ld, order: %d, insertion: %15lld ns\n", |
| 81 | size, step, order, nsec); |
| 82 | } |
| 83 | |
| 84 | static void benchmark_tagging(struct radix_tree_root *root, |
| 85 | unsigned long size, unsigned long step, int order) |
| 86 | { |
| 87 | struct timespec start, finish; |
| 88 | unsigned long index; |
| 89 | long long nsec; |
| 90 | |
| 91 | clock_gettime(CLOCK_MONOTONIC, &start); |
| 92 | |
| 93 | for (index = 0 ; index < size ; index += step) |
| 94 | radix_tree_tag_set(root, index, 0); |
| 95 | |
| 96 | clock_gettime(CLOCK_MONOTONIC, &finish); |
| 97 | |
| 98 | nsec = (finish.tv_sec - start.tv_sec) * NSEC_PER_SEC + |
| 99 | (finish.tv_nsec - start.tv_nsec); |
| 100 | |
| 101 | printv(2, "Size: %8ld, step: %8ld, order: %d, tagging: %17lld ns\n", |
| 102 | size, step, order, nsec); |
| 103 | } |
| 104 | |
| 105 | static void benchmark_delete(struct radix_tree_root *root, |
| 106 | unsigned long size, unsigned long step, int order) |
| 107 | { |
| 108 | struct timespec start, finish; |
| 109 | unsigned long index, i; |
| 110 | long long nsec; |
| 111 | |
| 112 | clock_gettime(CLOCK_MONOTONIC, &start); |
| 113 | |
| 114 | for (index = 0 ; index < size ; index += step) |
| 115 | for_each_index(i, index, order) |
| 116 | item_delete(root, i); |
| 117 | |
| 118 | clock_gettime(CLOCK_MONOTONIC, &finish); |
| 119 | |
| 120 | nsec = (finish.tv_sec - start.tv_sec) * NSEC_PER_SEC + |
| 121 | (finish.tv_nsec - start.tv_nsec); |
| 122 | |
| 123 | printv(2, "Size: %8ld, step: %8ld, order: %d, deletion: %16lld ns\n", |
| 124 | size, step, order, nsec); |
| 125 | } |
| 126 | |
| 127 | static void benchmark_size(unsigned long size, unsigned long step, int order) |
| 128 | { |
| 129 | RADIX_TREE(tree, GFP_KERNEL); |
| 130 | long long normal, tagged; |
| 131 | |
| 132 | benchmark_insert(&tree, size, step, order); |
| 133 | benchmark_tagging(&tree, size, step, order); |
| 134 | |
| 135 | tagged = benchmark_iter(&tree, true); |
| 136 | normal = benchmark_iter(&tree, false); |
| 137 | |
| 138 | printv(2, "Size: %8ld, step: %8ld, order: %d, tagged iteration: %8lld ns\n", |
| 139 | size, step, order, tagged); |
| 140 | printv(2, "Size: %8ld, step: %8ld, order: %d, normal iteration: %8lld ns\n", |
| 141 | size, step, order, normal); |
| 142 | |
| 143 | benchmark_delete(&tree, size, step, order); |
| 144 | |
| 145 | item_kill_tree(&tree); |
| 146 | rcu_barrier(); |
| 147 | } |
| 148 | |
| 149 | static long long __benchmark_split(unsigned long index, |
| 150 | int old_order, int new_order) |
| 151 | { |
| 152 | struct timespec start, finish; |
| 153 | long long nsec; |
| 154 | RADIX_TREE(tree, GFP_ATOMIC); |
| 155 | |
| 156 | item_insert_order(&tree, index, old_order); |
| 157 | |
| 158 | clock_gettime(CLOCK_MONOTONIC, &start); |
| 159 | radix_tree_split(&tree, index, new_order); |
| 160 | clock_gettime(CLOCK_MONOTONIC, &finish); |
| 161 | nsec = (finish.tv_sec - start.tv_sec) * NSEC_PER_SEC + |
| 162 | (finish.tv_nsec - start.tv_nsec); |
| 163 | |
| 164 | item_kill_tree(&tree); |
| 165 | |
| 166 | return nsec; |
| 167 | |
| 168 | } |
| 169 | |
| 170 | static void benchmark_split(unsigned long size, unsigned long step) |
| 171 | { |
| 172 | int i, j, idx; |
| 173 | long long nsec = 0; |
| 174 | |
| 175 | |
| 176 | for (idx = 0; idx < size; idx += step) { |
| 177 | for (i = 3; i < 11; i++) { |
| 178 | for (j = 0; j < i; j++) { |
| 179 | nsec += __benchmark_split(idx, i, j); |
| 180 | } |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | printv(2, "Size %8ld, step %8ld, split time %10lld ns\n", |
| 185 | size, step, nsec); |
| 186 | |
| 187 | } |
| 188 | |
| 189 | static long long __benchmark_join(unsigned long index, |
| 190 | unsigned order1, unsigned order2) |
| 191 | { |
| 192 | unsigned long loc; |
| 193 | struct timespec start, finish; |
| 194 | long long nsec; |
| 195 | void *item, *item2 = item_create(index + 1, order1); |
| 196 | RADIX_TREE(tree, GFP_KERNEL); |
| 197 | |
| 198 | item_insert_order(&tree, index, order2); |
| 199 | item = radix_tree_lookup(&tree, index); |
| 200 | |
| 201 | clock_gettime(CLOCK_MONOTONIC, &start); |
| 202 | radix_tree_join(&tree, index + 1, order1, item2); |
| 203 | clock_gettime(CLOCK_MONOTONIC, &finish); |
| 204 | nsec = (finish.tv_sec - start.tv_sec) * NSEC_PER_SEC + |
| 205 | (finish.tv_nsec - start.tv_nsec); |
| 206 | |
| 207 | loc = find_item(&tree, item); |
| 208 | if (loc == -1) |
| 209 | free(item); |
| 210 | |
| 211 | item_kill_tree(&tree); |
| 212 | |
| 213 | return nsec; |
| 214 | } |
| 215 | |
| 216 | static void benchmark_join(unsigned long step) |
| 217 | { |
| 218 | int i, j, idx; |
| 219 | long long nsec = 0; |
| 220 | |
| 221 | for (idx = 0; idx < 1 << 10; idx += step) { |
| 222 | for (i = 1; i < 15; i++) { |
| 223 | for (j = 0; j < i; j++) { |
| 224 | nsec += __benchmark_join(idx, i, j); |
| 225 | } |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | printv(2, "Size %8d, step %8ld, join time %10lld ns\n", |
| 230 | 1 << 10, step, nsec); |
| 231 | } |
| 232 | |
| 233 | void benchmark(void) |
| 234 | { |
| 235 | unsigned long size[] = {1 << 10, 1 << 20, 0}; |
| 236 | unsigned long step[] = {1, 2, 7, 15, 63, 64, 65, |
| 237 | 128, 256, 512, 12345, 0}; |
| 238 | int c, s; |
| 239 | |
| 240 | printv(1, "starting benchmarks\n"); |
| 241 | printv(1, "RADIX_TREE_MAP_SHIFT = %d\n", RADIX_TREE_MAP_SHIFT); |
| 242 | |
| 243 | for (c = 0; size[c]; c++) |
| 244 | for (s = 0; step[s]; s++) |
| 245 | benchmark_size(size[c], step[s], 0); |
| 246 | |
| 247 | for (c = 0; size[c]; c++) |
| 248 | for (s = 0; step[s]; s++) |
| 249 | benchmark_size(size[c], step[s] << 9, 9); |
| 250 | |
| 251 | for (c = 0; size[c]; c++) |
| 252 | for (s = 0; step[s]; s++) |
| 253 | benchmark_split(size[c], step[s]); |
| 254 | |
| 255 | for (s = 0; step[s]; s++) |
| 256 | benchmark_join(step[s]); |
| 257 | } |