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"
37 #include "libavutil/x86/cpu.h"
40 #if HAVE_SETCONSOLETEXTATTRIBUTE
42 #define COLOR_RED FOREGROUND_RED
43 #define COLOR_GREEN FOREGROUND_GREEN
44 #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
48 #define COLOR_YELLOW 3
59 /* List of tests to invoke */
65 { "bswapdsp", checkasm_check_bswapdsp },
68 { "h264pred", checkasm_check_h264pred },
71 { "h264qpel", checkasm_check_h264qpel },
76 /* List of cpu flags to check */
83 { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
84 { "NEON", "neon", AV_CPU_FLAG_NEON },
86 { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
87 { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
88 { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
89 { "VFP", "vfp", AV_CPU_FLAG_VFP },
90 { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
91 { "NEON", "neon", AV_CPU_FLAG_NEON },
93 { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
94 { "VSX", "vsx", AV_CPU_FLAG_VSX },
95 { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
97 { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
98 { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
99 { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
100 { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
101 { "SSE", "sse", AV_CPU_FLAG_SSE },
102 { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
103 { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
104 { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
105 { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
106 { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
107 { "AVX", "avx", AV_CPU_FLAG_AVX },
108 { "XOP", "xop", AV_CPU_FLAG_XOP },
109 { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
110 { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
111 { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
116 typedef struct CheckasmFuncVersion {
117 struct CheckasmFuncVersion *next;
123 } CheckasmFuncVersion;
125 /* Binary search tree node */
126 typedef struct CheckasmFunc {
127 struct CheckasmFunc *child[2];
128 CheckasmFuncVersion versions;
135 CheckasmFunc *current_func;
136 CheckasmFuncVersion *current_func_ver;
137 const char *current_test_name;
138 const char *bench_pattern;
139 int bench_pattern_len;
144 const char *cpu_flag_name;
150 /* Print colored text to stderr if the terminal supports it */
151 static void color_printf(int color, const char *fmt, ...)
153 static int use_color = -1;
156 #if HAVE_SETCONSOLETEXTATTRIBUTE
158 static WORD org_attributes;
161 CONSOLE_SCREEN_BUFFER_INFO con_info;
162 con = GetStdHandle(STD_ERROR_HANDLE);
163 if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
164 org_attributes = con_info.wAttributes;
170 SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
173 const char *term = getenv("TERM");
174 use_color = term && strcmp(term, "dumb") && isatty(2);
177 fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
181 vfprintf(stderr, fmt, arg);
185 #if HAVE_SETCONSOLETEXTATTRIBUTE
186 SetConsoleTextAttribute(con, org_attributes);
188 fprintf(stderr, "\x1b[0m");
193 /* Deallocate a tree */
194 static void destroy_func_tree(CheckasmFunc *f)
197 CheckasmFuncVersion *v = f->versions.next;
199 CheckasmFuncVersion *next = v->next;
204 destroy_func_tree(f->child[0]);
205 destroy_func_tree(f->child[1]);
210 /* Allocate a zero-initialized block, clean up and exit on failure */
211 static void *checkasm_malloc(size_t size)
213 void *ptr = calloc(1, size);
215 fprintf(stderr, "checkasm: malloc failed\n");
216 destroy_func_tree(state.funcs);
222 /* Get the suffix of the specified cpu flag */
223 static const char *cpu_suffix(int cpu)
225 int i = FF_ARRAY_ELEMS(cpus);
228 if (cpu & cpus[i].flag)
229 return cpus[i].suffix;
235 static int cmp_nop(const void *a, const void *b)
237 return *(const uint16_t*)a - *(const uint16_t*)b;
240 /* Measure the overhead of the timing code (in decicycles) */
241 static int measure_nop_time(void)
243 uint16_t nops[10000];
246 for (i = 0; i < 10000; i++) {
247 uint64_t t = AV_READ_TIME();
248 nops[i] = AV_READ_TIME() - t;
251 qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
252 for (i = 2500; i < 7500; i++)
255 return nop_sum / 500;
258 /* Print benchmark results */
259 static void print_benchs(CheckasmFunc *f)
262 print_benchs(f->child[0]);
264 /* Only print functions with at least one assembly version */
265 if (f->versions.cpu || f->versions.next) {
266 CheckasmFuncVersion *v = &f->versions;
269 int decicycles = (10*v->cycles/v->iterations - state.nop_time) / 4;
270 printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
272 } while ((v = v->next));
275 print_benchs(f->child[1]);
280 /* ASCIIbetical sort except preserving natural order for numbers */
281 static int cmp_func_names(const char *a, const char *b)
283 int ascii_diff, digit_diff;
285 for (; !(ascii_diff = *a - *b) && *a; a++, b++);
286 for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
288 return (digit_diff = av_isdigit(*a) - av_isdigit(*b)) ? digit_diff : ascii_diff;
291 /* Get a node with the specified name, creating it if it doesn't exist */
292 static CheckasmFunc *get_func(const char *name, int length)
294 CheckasmFunc *f, **f_ptr = &state.funcs;
296 /* Search the tree for a matching node */
297 while ((f = *f_ptr)) {
298 int cmp = cmp_func_names(name, f->name);
302 f_ptr = &f->child[(cmp > 0)];
305 /* Allocate and insert a new node into the tree */
306 f = *f_ptr = checkasm_malloc(sizeof(CheckasmFunc) + length);
307 memcpy(f->name, name, length+1);
312 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
313 static void check_cpu_flag(const char *name, int flag)
315 int old_cpu_flag = state.cpu_flag;
317 flag |= old_cpu_flag;
318 av_set_cpu_flags_mask(flag);
319 state.cpu_flag = av_get_cpu_flags();
321 if (!flag || state.cpu_flag != old_cpu_flag) {
324 state.cpu_flag_name = name;
325 for (i = 0; tests[i].func; i++) {
326 state.current_test_name = tests[i].name;
332 /* Print the name of the current CPU flag, but only do it once */
333 static void print_cpu_name(void)
335 if (state.cpu_flag_name) {
336 color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
337 state.cpu_flag_name = NULL;
341 int main(int argc, char *argv[])
343 int i, seed, ret = 0;
345 if (!tests[0].func || !cpus[0].flag) {
346 fprintf(stderr, "checkasm: no tests to perform\n");
350 if (argc > 1 && !strncmp(argv[1], "--bench", 7)) {
352 fprintf(stderr, "checkasm: --bench is not supported on your system\n");
355 if (argv[1][7] == '=') {
356 state.bench_pattern = argv[1] + 8;
357 state.bench_pattern_len = strlen(state.bench_pattern);
359 state.bench_pattern = "";
365 seed = (argc > 1) ? atoi(argv[1]) : av_get_random_seed();
366 fprintf(stderr, "checkasm: using random seed %u\n", seed);
367 av_lfg_init(&checkasm_lfg, seed);
369 check_cpu_flag(NULL, 0);
370 for (i = 0; cpus[i].flag; i++)
371 check_cpu_flag(cpus[i].name, cpus[i].flag);
373 if (state.num_failed) {
374 fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
377 fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
379 if (state.bench_pattern) {
380 state.nop_time = measure_nop_time();
381 printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
382 print_benchs(state.funcs);
387 destroy_func_tree(state.funcs);
391 /* Decide whether or not the specified function needs to be tested and
392 * allocate/initialize data structures if needed. Returns a pointer to a
393 * reference function if the function should be tested, otherwise NULL */
394 intptr_t (*checkasm_check_func(intptr_t (*func)(), const char *name, ...))()
397 intptr_t (*ref)() = func;
398 CheckasmFuncVersion *v;
403 name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
406 if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
409 state.current_func = get_func(name_buf, name_length);
410 v = &state.current_func->versions;
413 CheckasmFuncVersion *prev;
415 /* Only test functions that haven't already been tested */
423 } while ((v = v->next));
425 v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
430 v->cpu = state.cpu_flag;
431 state.current_func_ver = v;
439 /* Decide whether or not the current function needs to be benchmarked */
440 int checkasm_bench_func(void)
442 return !state.num_failed && state.bench_pattern &&
443 !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
446 /* Indicate that the current test has failed */
447 void checkasm_fail_func(const char *msg, ...)
449 if (state.current_func_ver->cpu && state.current_func_ver->ok) {
453 fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
455 vfprintf(stderr, msg, arg);
457 fprintf(stderr, ")\n");
459 state.current_func_ver->ok = 0;
464 /* Update benchmark results of the current function */
465 void checkasm_update_bench(int iterations, uint64_t cycles)
467 state.current_func_ver->iterations += iterations;
468 state.current_func_ver->cycles += cycles;
471 /* Print the outcome of all tests performed since the last time this function was called */
472 void checkasm_report(const char *name, ...)
474 static int prev_checked, prev_failed, max_length;
476 if (state.num_checked > prev_checked) {
477 int pad_length = max_length + 4;
481 pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
483 pad_length -= vfprintf(stderr, name, arg);
485 fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
487 if (state.num_failed == prev_failed)
488 color_printf(COLOR_GREEN, "OK");
490 color_printf(COLOR_RED, "FAILED");
491 fprintf(stderr, "]\n");
493 prev_checked = state.num_checked;
494 prev_failed = state.num_failed;
495 } else if (!state.cpu_flag) {
496 /* Calculate the amount of padding required to make the output vertically aligned */
497 int length = strlen(state.current_test_name);
501 length += vsnprintf(NULL, 0, name, arg);
504 if (length > max_length)