2 * Assembly testing and benchmarking tool
3 * Copyright (c) 2015 Henrik Gramner
4 * Copyright (c) 2008 Loren Merritt
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; 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_bswapdsp,
61 checkasm_check_h264pred,
64 checkasm_check_h264qpel,
69 /* List of cpu flags to check */
76 { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
77 { "NEON", "neon", AV_CPU_FLAG_NEON },
79 { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
80 { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
81 { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
82 { "VFP", "vfp", AV_CPU_FLAG_VFP },
83 { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
84 { "NEON", "neon", AV_CPU_FLAG_NEON },
86 { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
87 { "VSX", "vsx", AV_CPU_FLAG_VSX },
88 { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
90 { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
91 { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
92 { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
93 { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
94 { "SSE", "sse", AV_CPU_FLAG_SSE },
95 { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
96 { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
97 { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
98 { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
99 { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
100 { "AVX", "avx", AV_CPU_FLAG_AVX },
101 { "XOP", "xop", AV_CPU_FLAG_XOP },
102 { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
103 { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
104 { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
109 typedef struct CheckasmFuncVersion {
110 struct CheckasmFuncVersion *next;
116 } CheckasmFuncVersion;
118 /* Binary search tree node */
119 typedef struct CheckasmFunc {
120 struct CheckasmFunc *child[2];
121 CheckasmFuncVersion versions;
128 CheckasmFunc *current_func;
129 CheckasmFuncVersion *current_func_ver;
130 const char *bench_pattern;
131 int bench_pattern_len;
136 const char *cpu_flag_name;
142 /* Print colored text to stderr if the terminal supports it */
143 static void color_printf(int color, const char *fmt, ...)
145 static int use_color = -1;
148 #if HAVE_SETCONSOLETEXTATTRIBUTE
150 static WORD org_attributes;
153 CONSOLE_SCREEN_BUFFER_INFO con_info;
154 con = GetStdHandle(STD_ERROR_HANDLE);
155 if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
156 org_attributes = con_info.wAttributes;
162 SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
165 const char *term = getenv("TERM");
166 use_color = term && strcmp(term, "dumb") && isatty(2);
169 fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
173 vfprintf(stderr, fmt, arg);
177 #if HAVE_SETCONSOLETEXTATTRIBUTE
178 SetConsoleTextAttribute(con, org_attributes);
180 fprintf(stderr, "\x1b[0m");
185 /* Deallocate a tree */
186 static void destroy_func_tree(CheckasmFunc *f)
189 CheckasmFuncVersion *v = f->versions.next;
191 CheckasmFuncVersion *next = v->next;
196 destroy_func_tree(f->child[0]);
197 destroy_func_tree(f->child[1]);
202 /* Allocate a zero-initialized block, clean up and exit on failure */
203 static void *checkasm_malloc(size_t size)
205 void *ptr = calloc(1, size);
207 fprintf(stderr, "checkasm: malloc failed\n");
208 destroy_func_tree(state.funcs);
214 /* Get the suffix of the specified cpu flag */
215 static const char *cpu_suffix(int cpu)
217 int i = FF_ARRAY_ELEMS(cpus);
220 if (cpu & cpus[i].flag)
221 return cpus[i].suffix;
227 static int cmp_nop(const void *a, const void *b)
229 return *(const uint16_t*)a - *(const uint16_t*)b;
232 /* Measure the overhead of the timing code (in decicycles) */
233 static int measure_nop_time(void)
235 uint16_t nops[10000];
238 for (i = 0; i < 10000; i++) {
239 uint64_t t = AV_READ_TIME();
240 nops[i] = AV_READ_TIME() - t;
243 qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
244 for (i = 2500; i < 7500; i++)
247 return nop_sum / 500;
250 /* Print benchmark results */
251 static void print_benchs(CheckasmFunc *f)
254 print_benchs(f->child[0]);
256 /* Only print functions with at least one assembly version */
257 if (f->versions.cpu || f->versions.next) {
258 CheckasmFuncVersion *v = &f->versions;
261 int decicycles = (10*v->cycles/v->iterations - state.nop_time) / 4;
262 printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
264 } while ((v = v->next));
267 print_benchs(f->child[1]);
272 /* ASCIIbetical sort except preserving natural order for numbers */
273 static int cmp_func_names(const char *a, const char *b)
275 int ascii_diff, digit_diff;
277 for (; !(ascii_diff = *a - *b) && *a; a++, b++);
278 for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
280 return (digit_diff = av_isdigit(*a) - av_isdigit(*b)) ? digit_diff : ascii_diff;
283 /* Get a node with the specified name, creating it if it doesn't exist */
284 static CheckasmFunc *get_func(const char *name, int length)
286 CheckasmFunc *f, **f_ptr = &state.funcs;
288 /* Search the tree for a matching node */
289 while ((f = *f_ptr)) {
290 int cmp = cmp_func_names(name, f->name);
294 f_ptr = &f->child[(cmp > 0)];
297 /* Allocate and insert a new node into the tree */
298 f = *f_ptr = checkasm_malloc(sizeof(CheckasmFunc) + length);
299 memcpy(f->name, name, length+1);
304 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
305 static void check_cpu_flag(const char *name, int flag)
307 int old_cpu_flag = state.cpu_flag;
309 flag |= old_cpu_flag;
310 av_set_cpu_flags_mask(flag);
311 state.cpu_flag = av_get_cpu_flags();
313 if (!flag || state.cpu_flag != old_cpu_flag) {
316 state.cpu_flag_name = name;
317 for (i = 0; tests[i]; i++)
322 /* Print the name of the current CPU flag, but only do it once */
323 static void print_cpu_name(void)
325 if (state.cpu_flag_name) {
326 color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
327 state.cpu_flag_name = NULL;
331 int main(int argc, char *argv[])
333 int i, seed, ret = 0;
335 if (!tests[0] || !cpus[0].flag) {
336 fprintf(stderr, "checkasm: no tests to perform\n");
340 if (argc > 1 && !strncmp(argv[1], "--bench", 7)) {
342 fprintf(stderr, "checkasm: --bench is not supported on your system\n");
345 if (argv[1][7] == '=') {
346 state.bench_pattern = argv[1] + 8;
347 state.bench_pattern_len = strlen(state.bench_pattern);
349 state.bench_pattern = "";
355 seed = (argc > 1) ? atoi(argv[1]) : av_get_random_seed();
356 fprintf(stderr, "checkasm: using random seed %u\n", seed);
357 av_lfg_init(&checkasm_lfg, seed);
359 check_cpu_flag(NULL, 0);
360 for (i = 0; cpus[i].flag; i++)
361 check_cpu_flag(cpus[i].name, cpus[i].flag);
363 if (state.num_failed) {
364 fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
367 fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
369 if (state.bench_pattern) {
370 state.nop_time = measure_nop_time();
371 printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
372 print_benchs(state.funcs);
377 destroy_func_tree(state.funcs);
381 /* Decide whether or not the specified function needs to be tested and
382 * allocate/initialize data structures if needed. Returns a pointer to a
383 * reference function if the function should be tested, otherwise NULL */
384 intptr_t (*checkasm_check_func(intptr_t (*func)(), const char *name, ...))()
387 intptr_t (*ref)() = func;
388 CheckasmFuncVersion *v;
393 name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
396 if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
399 state.current_func = get_func(name_buf, name_length);
400 v = &state.current_func->versions;
403 CheckasmFuncVersion *prev;
405 /* Only test functions that haven't already been tested */
413 } while ((v = v->next));
415 v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
420 v->cpu = state.cpu_flag;
421 state.current_func_ver = v;
429 /* Decide whether or not the current function needs to be benchmarked */
430 int checkasm_bench_func(void)
432 return !state.num_failed && state.bench_pattern &&
433 !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
436 /* Indicate that the current test has failed */
437 void checkasm_fail_func(const char *msg, ...)
439 if (state.current_func_ver->cpu && state.current_func_ver->ok) {
443 fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
445 vfprintf(stderr, msg, arg);
447 fprintf(stderr, ")\n");
449 state.current_func_ver->ok = 0;
454 /* Update benchmark results of the current function */
455 void checkasm_update_bench(int iterations, uint64_t cycles)
457 state.current_func_ver->iterations += iterations;
458 state.current_func_ver->cycles += cycles;
461 /* Print the outcome of all tests performed since the last time this function was called */
462 void checkasm_report(const char *name, ...)
464 static int prev_checked, prev_failed, max_length;
466 if (state.num_checked > prev_checked) {
470 int pad_length = max_length;
473 fprintf(stderr, " - ");
475 pad_length -= vfprintf(stderr, name, arg);
477 fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
479 fprintf(stderr, " - %-*s [", max_length, state.current_func->name);
481 if (state.num_failed == prev_failed)
482 color_printf(COLOR_GREEN, "OK");
484 color_printf(COLOR_RED, "FAILED");
485 fprintf(stderr, "]\n");
487 prev_checked = state.num_checked;
488 prev_failed = state.num_failed;
489 } else if (!state.cpu_flag) {
492 /* Calculate the amount of padding required to make the output vertically aligned */
496 length = vsnprintf(NULL, 0, name, arg);
499 length = strlen(state.current_func->name);
501 if (length > max_length)