| 1 |
#include "is_utf8.h" |
| 2 |
#include "simdutf.h" |
| 3 |
#include <assert.h> |
| 4 |
#include <chrono> |
| 5 |
#include <stdbool.h> |
| 6 |
#include <stddef.h> |
| 7 |
#include <stdint.h> |
| 8 |
#include <stdio.h> |
| 9 |
#include <stdlib.h> |
| 10 |
#include <string.h> |
| 11 |
|
| 12 |
uint64_t nano() { |
| 13 |
return std::chrono::duration_cast<::std::chrono::nanoseconds>( |
| 14 |
std::chrono::steady_clock::now().time_since_epoch()) |
| 15 |
.count(); |
| 16 |
} |
| 17 |
|
| 18 |
#ifdef _MSC_VER |
| 19 |
#define never_inline __declspec(noinline) |
| 20 |
#else |
| 21 |
#define never_inline __attribute__((noinline)) |
| 22 |
#endif |
| 23 |
|
| 24 |
// generate a string having at least length N |
| 25 |
// can exceed by up to 3 chars, returns the actual length |
| 26 |
size_t populate_utf8(char *data, size_t N) { |
| 27 |
size_t i = 0; |
| 28 |
for (; i < N;) { |
| 29 |
int w = rand() & 0xFF; |
| 30 |
if (w < 0x80) { |
| 31 |
data[i++] = 0x20; // w; |
| 32 |
} else if (w < 0xE0) { |
| 33 |
data[i++] = 0xC2 + rand() % (0xDF - 0xC2 + 1); |
| 34 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 35 |
} else if (w == 0xE0) { |
| 36 |
data[i++] = w; |
| 37 |
data[i++] = 0xA0 + rand() % (0xBF - 0xA0 + 1); |
| 38 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 39 |
} else if (w <= 0xEC) { |
| 40 |
data[i++] = w; |
| 41 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 42 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 43 |
} else if (w == 0xED) { |
| 44 |
data[i++] = w; |
| 45 |
data[i++] = 0x80 + rand() % (0x9F - 0x80 + 1); |
| 46 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 47 |
} else if (w <= 0xEF) { |
| 48 |
data[i++] = w; |
| 49 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 50 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 51 |
} else if (w < 0xF0) { |
| 52 |
data[i++] = 0xF1 + rand() % (0xF3 - 0xF1 + 1); |
| 53 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 54 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 55 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 56 |
} else if (w == 0xF0) { |
| 57 |
data[i++] = w; |
| 58 |
data[i++] = 0x90 + rand() % (0xBF - 0x90 + 1); |
| 59 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 60 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 61 |
} else if (w <= 0xF3) { |
| 62 |
data[i++] = 0xF1 + rand() % (0xF3 - 0xF1 + 1); |
| 63 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 64 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 65 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 66 |
} else if (w == 0xF4) { |
| 67 |
data[i++] = w; |
| 68 |
data[i++] = 0x80 + rand() % (0x8F - 0x80 + 1); |
| 69 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 70 |
data[i++] = 0x80 + rand() % (0xBF - 0x80 + 1); |
| 71 |
} |
| 72 |
} |
| 73 |
data[i] = '\0'; |
| 74 |
return i; |
| 75 |
} |
| 76 |
|
| 77 |
// copied in part from Guava |
| 78 |
static never_inline bool basic_validate_utf8(const char *b, size_t length) { |
| 79 |
const unsigned char *bytes = (const unsigned char *)b; |
| 80 |
for (size_t index = 0;;) { |
| 81 |
unsigned char byte1; |
| 82 |
|
| 83 |
do { // fast ASCII Path |
| 84 |
if (index >= length) { |
| 85 |
return true; |
| 86 |
} |
| 87 |
byte1 = bytes[index++]; |
| 88 |
} while (byte1 < 0x80); |
| 89 |
if (byte1 < 0xE0) { |
| 90 |
// Two-byte form. |
| 91 |
if (index == length) { |
| 92 |
return false; |
| 93 |
} |
| 94 |
if (byte1 < 0xC2 || bytes[index++] > 0xBF) { |
| 95 |
return false; |
| 96 |
} |
| 97 |
} else if (byte1 < 0xF0) { |
| 98 |
// Three-byte form. |
| 99 |
if (index + 1 >= length) { |
| 100 |
return false; |
| 101 |
} |
| 102 |
unsigned char byte2 = bytes[index++]; |
| 103 |
if (byte2 > 0xBF |
| 104 |
// Overlong? 5 most significant bits must not all be zero. |
| 105 |
|| (byte1 == 0xE0 && byte2 < 0xA0) |
| 106 |
// Check for illegal surrogate codepoints. |
| 107 |
|| (byte1 == 0xED && 0xA0 <= byte2) |
| 108 |
// Third byte trailing-byte test. |
| 109 |
|| bytes[index++] > 0xBF) { |
| 110 |
return false; |
| 111 |
} |
| 112 |
} else { |
| 113 |
|
| 114 |
// Four-byte form. |
| 115 |
if (index + 2 >= length) { |
| 116 |
return false; |
| 117 |
} |
| 118 |
int byte2 = bytes[index++]; |
| 119 |
if (byte2 > 0xBF |
| 120 |
// Check that 1 <= plane <= 16. Tricky optimized form of: |
| 121 |
// if (byte1 > (byte) 0xF4 |
| 122 |
// || byte1 == (byte) 0xF0 && byte2 < (byte) 0x90 |
| 123 |
// || byte1 == (byte) 0xF4 && byte2 > (byte) 0x8F) |
| 124 |
|| (((byte1 << 28) + (byte2 - 0x90)) >> 30) != 0 |
| 125 |
// Third byte trailing-byte test |
| 126 |
|| bytes[index++] > 0xBF |
| 127 |
// Fourth byte trailing-byte test |
| 128 |
|| bytes[index++] > 0xBF) { |
| 129 |
return false; |
| 130 |
} |
| 131 |
} |
| 132 |
} |
| 133 |
} |
| 134 |
|
| 135 |
|
| 136 |
bool zerobuffer_bench(size_t N) { |
| 137 |
printf("zero buffer \n"); |
| 138 |
printf("string size = %zu \n", N); |
| 139 |
char *input = new char[N]{}; |
| 140 |
volatile bool isgood{true}; |
| 141 |
|
| 142 |
{ |
| 143 |
uint64_t start = nano(); |
| 144 |
uint64_t finish = start; |
| 145 |
size_t count{0}; |
| 146 |
uint64_t threshold = 500000000; |
| 147 |
for (; finish - start < threshold;) { |
| 148 |
count++; |
| 149 |
isgood &= basic_validate_utf8(input, N); |
| 150 |
finish = nano(); |
| 151 |
} |
| 152 |
double t = (N * count) / double(finish - start); |
| 153 |
|
| 154 |
printf("basic_validate_utf8 %f GB/s\n", t); |
| 155 |
} |
| 156 |
|
| 157 |
{ |
| 158 |
uint64_t start = nano(); |
| 159 |
uint64_t finish = start; |
| 160 |
size_t count{0}; |
| 161 |
uint64_t threshold = 500000000; |
| 162 |
for (; finish - start < threshold;) { |
| 163 |
count++; |
| 164 |
isgood &= simdutf::validate_utf8(input, N); |
| 165 |
finish = nano(); |
| 166 |
} |
| 167 |
double t = (N * count) / double(finish - start); |
| 168 |
|
| 169 |
printf("simdutf %f GB/s\n", t); |
| 170 |
} |
| 171 |
|
| 172 |
{ |
| 173 |
uint64_t start = nano(); |
| 174 |
uint64_t finish = start; |
| 175 |
size_t count{0}; |
| 176 |
uint64_t threshold = 500000000; |
| 177 |
for (; finish - start < threshold;) { |
| 178 |
count++; |
| 179 |
isgood &= is_utf8(input, N); |
| 180 |
finish = nano(); |
| 181 |
} |
| 182 |
double t = (N * count) / double(finish - start); |
| 183 |
|
| 184 |
printf("is_utf8 %f GB/s\n", t); |
| 185 |
} |
| 186 |
delete[] input; |
| 187 |
printf("\n"); |
| 188 |
return isgood; |
| 189 |
} |
| 190 |
|
| 191 |
bool bench(size_t N) { |
| 192 |
printf("random UTF-8\n"); |
| 193 |
printf("string size = %zu \n", N); |
| 194 |
char *input = new char[N]; |
| 195 |
populate_utf8(input, N); |
| 196 |
volatile bool isgood{true}; |
| 197 |
|
| 198 |
{ |
| 199 |
uint64_t start = nano(); |
| 200 |
uint64_t finish = start; |
| 201 |
size_t count{0}; |
| 202 |
uint64_t threshold = 500000000; |
| 203 |
for (; finish - start < threshold;) { |
| 204 |
count++; |
| 205 |
isgood &= basic_validate_utf8(input, N); |
| 206 |
finish = nano(); |
| 207 |
} |
| 208 |
double t = (N * count) / double(finish - start); |
| 209 |
|
| 210 |
printf("basic_validate_utf8 %f GB/s\n", t); |
| 211 |
} |
| 212 |
|
| 213 |
{ |
| 214 |
uint64_t start = nano(); |
| 215 |
uint64_t finish = start; |
| 216 |
size_t count{0}; |
| 217 |
uint64_t threshold = 500000000; |
| 218 |
for (; finish - start < threshold;) { |
| 219 |
count++; |
| 220 |
isgood &= simdutf::validate_utf8(input, N); |
| 221 |
finish = nano(); |
| 222 |
} |
| 223 |
double t = (N * count) / double(finish - start); |
| 224 |
|
| 225 |
printf("simdutf %f GB/s\n", t); |
| 226 |
} |
| 227 |
|
| 228 |
{ |
| 229 |
uint64_t start = nano(); |
| 230 |
uint64_t finish = start; |
| 231 |
size_t count{0}; |
| 232 |
uint64_t threshold = 500000000; |
| 233 |
for (; finish - start < threshold;) { |
| 234 |
count++; |
| 235 |
isgood &= is_utf8(input, N); |
| 236 |
finish = nano(); |
| 237 |
} |
| 238 |
double t = (N * count) / double(finish - start); |
| 239 |
|
| 240 |
printf("is_utf8 %f GB/s\n", t); |
| 241 |
} |
| 242 |
delete[] input; |
| 243 |
printf("\n"); |
| 244 |
return isgood; |
| 245 |
} |
| 246 |
|
| 247 |
int main() { |
| 248 |
return (bench(40096) & bench(100000) & bench(50000)) |
| 249 |
& (zerobuffer_bench(40096) & zerobuffer_bench(100000) & zerobuffer_bench(50000)) |
| 250 |
? EXIT_SUCCESS : EXIT_FAILURE; |
| 251 |
} |