2 * Assembly testing and benchmarking tool
3 * Copyright (c) 2015 Henrik Gramner
4 * Copyright (c) 2008 Loren Merritt
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License along
19 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
28 #include "libavutil/common.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/random_seed.h"
33 #include "libavutil/x86/cpu.h"
36 #if HAVE_SETCONSOLETEXTATTRIBUTE
38 #define COLOR_RED FOREGROUND_RED
39 #define COLOR_GREEN FOREGROUND_GREEN
40 #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
44 #define COLOR_YELLOW 3
55 /* List of tests to invoke */
56 static void (* const tests[])(void) = {
58 checkasm_check_h264pred,
61 checkasm_check_h264qpel,
66 /* List of cpu flags to check */
73 { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
74 { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
75 { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
76 { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
77 { "SSE", "sse", AV_CPU_FLAG_SSE },
78 { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
79 { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
80 { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
81 { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
82 { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
83 { "AVX", "avx", AV_CPU_FLAG_AVX },
84 { "XOP", "xop", AV_CPU_FLAG_XOP },
85 { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
86 { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
87 { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
92 typedef struct CheckasmFuncVersion {
93 struct CheckasmFuncVersion *next;
99 } CheckasmFuncVersion;
101 /* Binary search tree node */
102 typedef struct CheckasmFunc {
103 struct CheckasmFunc *child[2];
104 CheckasmFuncVersion versions;
111 CheckasmFunc *current_func;
112 CheckasmFuncVersion *current_func_ver;
113 const char *bench_pattern;
114 int bench_pattern_len;
119 const char *cpu_flag_name;
125 /* Print colored text to stderr if the terminal supports it */
126 static void color_printf(int color, const char *fmt, ...)
128 static int use_color = -1;
131 #if HAVE_SETCONSOLETEXTATTRIBUTE
133 static WORD org_attributes;
136 CONSOLE_SCREEN_BUFFER_INFO con_info;
137 con = GetStdHandle(STD_ERROR_HANDLE);
138 if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
139 org_attributes = con_info.wAttributes;
145 SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
148 const char *term = getenv("TERM");
149 use_color = term && strcmp(term, "dumb") && isatty(2);
152 fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
156 vfprintf(stderr, fmt, arg);
160 #if HAVE_SETCONSOLETEXTATTRIBUTE
161 SetConsoleTextAttribute(con, org_attributes);
163 fprintf(stderr, "\x1b[0m");
168 /* Deallocate a tree */
169 static void destroy_func_tree(CheckasmFunc *f)
172 CheckasmFuncVersion *v = f->versions.next;
174 CheckasmFuncVersion *next = v->next;
179 destroy_func_tree(f->child[0]);
180 destroy_func_tree(f->child[1]);
185 /* Allocate a zero-initialized block, clean up and exit on failure */
186 static void *checkasm_malloc(size_t size)
188 void *ptr = calloc(1, size);
190 fprintf(stderr, "checkasm: malloc failed\n");
191 destroy_func_tree(state.funcs);
197 /* Get the suffix of the specified cpu flag */
198 static const char *cpu_suffix(int cpu)
200 int i = FF_ARRAY_ELEMS(cpus);
203 if (cpu & cpus[i].flag)
204 return cpus[i].suffix;
210 static int cmp_nop(const void *a, const void *b)
212 return *(const uint16_t*)a - *(const uint16_t*)b;
215 /* Measure the overhead of the timing code (in decicycles) */
216 static int measure_nop_time(void)
218 uint16_t nops[10000];
221 for (i = 0; i < 10000; i++) {
222 uint64_t t = AV_READ_TIME();
223 nops[i] = AV_READ_TIME() - t;
226 qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
227 for (i = 2500; i < 7500; i++)
230 return nop_sum / 500;
233 /* Print benchmark results */
234 static void print_benchs(CheckasmFunc *f)
237 print_benchs(f->child[0]);
239 /* Only print functions with at least one assembly version */
240 if (f->versions.cpu || f->versions.next) {
241 CheckasmFuncVersion *v = &f->versions;
244 int decicycles = (10*v->cycles/v->iterations - state.nop_time) / 4;
245 printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
247 } while ((v = v->next));
250 print_benchs(f->child[1]);
255 /* ASCIIbetical sort except preserving natural order for numbers */
256 static int cmp_func_names(const char *a, const char *b)
258 int ascii_diff, digit_diff;
260 for (; !(ascii_diff = *a - *b) && *a; a++, b++);
261 for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
263 return (digit_diff = av_isdigit(*a) - av_isdigit(*b)) ? digit_diff : ascii_diff;
266 /* Get a node with the specified name, creating it if it doesn't exist */
267 static CheckasmFunc *get_func(const char *name, int length)
269 CheckasmFunc *f, **f_ptr = &state.funcs;
271 /* Search the tree for a matching node */
272 while ((f = *f_ptr)) {
273 int cmp = cmp_func_names(name, f->name);
277 f_ptr = &f->child[(cmp > 0)];
280 /* Allocate and insert a new node into the tree */
281 f = *f_ptr = checkasm_malloc(sizeof(CheckasmFunc) + length);
282 memcpy(f->name, name, length+1);
287 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
288 static void check_cpu_flag(const char *name, int flag)
290 int old_cpu_flag = state.cpu_flag;
292 flag |= old_cpu_flag;
293 av_set_cpu_flags_mask(flag);
294 state.cpu_flag = av_get_cpu_flags();
296 if (!flag || state.cpu_flag != old_cpu_flag) {
299 state.cpu_flag_name = name;
300 for (i = 0; tests[i]; i++)
305 /* Print the name of the current CPU flag, but only do it once */
306 static void print_cpu_name(void)
308 if (state.cpu_flag_name) {
309 color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
310 state.cpu_flag_name = NULL;
314 int main(int argc, char *argv[])
316 int i, seed, ret = 0;
318 if (!tests[0] || !cpus[0].flag) {
319 fprintf(stderr, "checkasm: no tests to perform\n");
323 if (argc > 1 && !strncmp(argv[1], "--bench", 7)) {
325 fprintf(stderr, "checkasm: --bench is not supported on your system\n");
328 if (argv[1][7] == '=') {
329 state.bench_pattern = argv[1] + 8;
330 state.bench_pattern_len = strlen(state.bench_pattern);
332 state.bench_pattern = "";
338 seed = (argc > 1) ? atoi(argv[1]) : av_get_random_seed();
339 fprintf(stderr, "checkasm: using random seed %u\n", seed);
340 av_lfg_init(&checkasm_lfg, seed);
342 check_cpu_flag(NULL, 0);
343 for (i = 0; cpus[i].flag; i++)
344 check_cpu_flag(cpus[i].name, cpus[i].flag);
346 if (state.num_failed) {
347 fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
350 fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
352 if (state.bench_pattern) {
353 state.nop_time = measure_nop_time();
354 printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
355 print_benchs(state.funcs);
360 destroy_func_tree(state.funcs);
364 /* Decide whether or not the specified function needs to be tested and
365 * allocate/initialize data structures if needed. Returns a pointer to a
366 * reference function if the function should be tested, otherwise NULL */
367 intptr_t (*checkasm_check_func(intptr_t (*func)(), const char *name, ...))()
370 intptr_t (*ref)() = func;
371 CheckasmFuncVersion *v;
376 name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
379 if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
382 state.current_func = get_func(name_buf, name_length);
383 v = &state.current_func->versions;
386 CheckasmFuncVersion *prev;
388 /* Only test functions that haven't already been tested */
396 } while ((v = v->next));
398 v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
403 v->cpu = state.cpu_flag;
404 state.current_func_ver = v;
412 /* Decide whether or not the current function needs to be benchmarked */
413 int checkasm_bench_func(void)
415 return !state.num_failed && state.bench_pattern &&
416 !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
419 /* Indicate that the current test has failed */
420 void checkasm_fail_func(const char *msg, ...)
422 if (state.current_func_ver->cpu && state.current_func_ver->ok) {
426 fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
428 vfprintf(stderr, msg, arg);
430 fprintf(stderr, ")\n");
432 state.current_func_ver->ok = 0;
437 /* Update benchmark results of the current function */
438 void checkasm_update_bench(int iterations, uint64_t cycles)
440 state.current_func_ver->iterations += iterations;
441 state.current_func_ver->cycles += cycles;
444 /* Print the outcome of all tests performed since the last time this function was called */
445 void checkasm_report(const char *name, ...)
447 static int prev_checked, prev_failed, max_length;
449 if (state.num_checked > prev_checked) {
453 int pad_length = max_length;
456 fprintf(stderr, " - ");
458 pad_length -= vfprintf(stderr, name, arg);
460 fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
462 fprintf(stderr, " - %-*s [", max_length, state.current_func->name);
464 if (state.num_failed == prev_failed)
465 color_printf(COLOR_GREEN, "OK");
467 color_printf(COLOR_RED, "FAILED");
468 fprintf(stderr, "]\n");
470 prev_checked = state.num_checked;
471 prev_failed = state.num_failed;
472 } else if (!state.cpu_flag) {
475 /* Calculate the amount of padding required to make the output vertically aligned */
479 length = vsnprintf(NULL, 0, name, arg);
482 length = strlen(state.current_func->name);
484 if (length > max_length)