FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
checkasm.c
Go to the documentation of this file.
1 /*
2  * Assembly testing and benchmarking tool
3  * Copyright (c) 2015 Henrik Gramner
4  * Copyright (c) 2008 Loren Merritt
5  *
6  * This file is part of FFmpeg.
7  *
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.
12  *
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.
17  *
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.
21  */
22 
23 #include "config.h"
24 
25 #if CONFIG_LINUX_PERF
26 # ifndef _GNU_SOURCE
27 # define _GNU_SOURCE // for syscall (performance monitoring API)
28 # endif
29 #endif
30 
31 #include <stdarg.h>
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <string.h>
35 #include "checkasm.h"
36 #include "libavutil/common.h"
37 #include "libavutil/cpu.h"
38 #include "libavutil/intfloat.h"
39 #include "libavutil/random_seed.h"
40 
41 #if HAVE_IO_H
42 #include <io.h>
43 #endif
44 
45 #if HAVE_SETCONSOLETEXTATTRIBUTE
46 #include <windows.h>
47 #define COLOR_RED FOREGROUND_RED
48 #define COLOR_GREEN FOREGROUND_GREEN
49 #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
50 #else
51 #define COLOR_RED 1
52 #define COLOR_GREEN 2
53 #define COLOR_YELLOW 3
54 #endif
55 
56 #if HAVE_UNISTD_H
57 #include <unistd.h>
58 #endif
59 
60 #if !HAVE_ISATTY
61 #define isatty(fd) 1
62 #endif
63 
64 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
65 #include "libavutil/arm/cpu.h"
66 
67 void (*checkasm_checked_call)(void *func, int dummy, ...) = checkasm_checked_call_novfp;
68 #endif
69 
70 /* List of tests to invoke */
71 static const struct {
72  const char *name;
73  void (*func)(void);
74 } tests[] = {
75 #if CONFIG_AVCODEC
76  #if CONFIG_AAC_DECODER
77  { "aacpsdsp", checkasm_check_aacpsdsp },
78  { "sbrdsp", checkasm_check_sbrdsp },
79  #endif
80  #if CONFIG_ALAC_DECODER
81  { "alacdsp", checkasm_check_alacdsp },
82  #endif
83  #if CONFIG_AUDIODSP
84  { "audiodsp", checkasm_check_audiodsp },
85  #endif
86  #if CONFIG_BLOCKDSP
87  { "blockdsp", checkasm_check_blockdsp },
88  #endif
89  #if CONFIG_BSWAPDSP
90  { "bswapdsp", checkasm_check_bswapdsp },
91  #endif
92  #if CONFIG_DCA_DECODER
93  { "synth_filter", checkasm_check_synth_filter },
94  #endif
95  #if CONFIG_EXR_DECODER
96  { "exrdsp", checkasm_check_exrdsp },
97  #endif
98  #if CONFIG_FLACDSP
99  { "flacdsp", checkasm_check_flacdsp },
100  #endif
101  #if CONFIG_FMTCONVERT
102  { "fmtconvert", checkasm_check_fmtconvert },
103  #endif
104  #if CONFIG_G722DSP
105  { "g722dsp", checkasm_check_g722dsp },
106  #endif
107  #if CONFIG_H264DSP
108  { "h264dsp", checkasm_check_h264dsp },
109  #endif
110  #if CONFIG_H264PRED
111  { "h264pred", checkasm_check_h264pred },
112  #endif
113  #if CONFIG_H264QPEL
114  { "h264qpel", checkasm_check_h264qpel },
115  #endif
116  #if CONFIG_HEVC_DECODER
117  { "hevc_add_res", checkasm_check_hevc_add_res },
118  { "hevc_idct", checkasm_check_hevc_idct },
119  #endif
120  #if CONFIG_JPEG2000_DECODER
121  { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
122  #endif
123  #if CONFIG_HUFFYUVDSP
124  { "llviddsp", checkasm_check_llviddsp },
125  #endif
126  #if CONFIG_PIXBLOCKDSP
127  { "pixblockdsp", checkasm_check_pixblockdsp },
128  #endif
129  #if CONFIG_V210_ENCODER
130  { "v210enc", checkasm_check_v210enc },
131  #endif
132  #if CONFIG_VP8DSP
133  { "vp8dsp", checkasm_check_vp8dsp },
134  #endif
135  #if CONFIG_VP9_DECODER
136  { "vp9dsp", checkasm_check_vp9dsp },
137  #endif
138  #if CONFIG_VIDEODSP
139  { "videodsp", checkasm_check_videodsp },
140  #endif
141 #endif
142 #if CONFIG_AVFILTER
143  #if CONFIG_BLEND_FILTER
144  { "vf_blend", checkasm_check_blend },
145  #endif
146  #if CONFIG_COLORSPACE_FILTER
147  { "vf_colorspace", checkasm_check_colorspace },
148  #endif
149 #endif
150 #if CONFIG_AVUTIL
151  { "fixed_dsp", checkasm_check_fixed_dsp },
152  { "float_dsp", checkasm_check_float_dsp },
153 #endif
154  { NULL }
155 };
156 
157 /* List of cpu flags to check */
158 static const struct {
159  const char *name;
160  const char *suffix;
161  int flag;
162 } cpus[] = {
163 #if ARCH_AARCH64
164  { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
165  { "NEON", "neon", AV_CPU_FLAG_NEON },
166 #elif ARCH_ARM
167  { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
168  { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
169  { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
170  { "VFP", "vfp", AV_CPU_FLAG_VFP },
171  { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
172  { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
173  { "NEON", "neon", AV_CPU_FLAG_NEON },
174 #elif ARCH_PPC
175  { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
176  { "VSX", "vsx", AV_CPU_FLAG_VSX },
177  { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
178 #elif ARCH_X86
179  { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
180  { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
181  { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
182  { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
183  { "SSE", "sse", AV_CPU_FLAG_SSE },
184  { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
185  { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
186  { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
187  { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
188  { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
189  { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
190  { "AVX", "avx", AV_CPU_FLAG_AVX },
191  { "XOP", "xop", AV_CPU_FLAG_XOP },
192  { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
193  { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
194  { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
195 #endif
196  { NULL }
197 };
198 
199 typedef struct CheckasmFuncVersion {
201  void *func;
202  int ok;
203  int cpu;
206 
207 /* Binary search tree node */
208 typedef struct CheckasmFunc {
209  struct CheckasmFunc *child[2];
211  uint8_t color; /* 0 = red, 1 = black */
212  char name[1];
213 } CheckasmFunc;
214 
215 /* Internal state */
216 static struct {
220  const char *current_test_name;
221  const char *bench_pattern;
225 
226  /* perf */
227  int nop_time;
228  int sysfd;
229 
230  int cpu_flag;
231  const char *cpu_flag_name;
232  const char *test_name;
233 } state;
234 
235 /* PRNG state */
237 
238 /* float compare support code */
239 static int is_negative(union av_intfloat32 u)
240 {
241  return u.i >> 31;
242 }
243 
244 int float_near_ulp(float a, float b, unsigned max_ulp)
245 {
246  union av_intfloat32 x, y;
247 
248  x.f = a;
249  y.f = b;
250 
251  if (is_negative(x) != is_negative(y)) {
252  // handle -0.0 == +0.0
253  return a == b;
254  }
255 
256  if (llabs((int64_t)x.i - y.i) <= max_ulp)
257  return 1;
258 
259  return 0;
260 }
261 
262 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
263  unsigned len)
264 {
265  unsigned i;
266 
267  for (i = 0; i < len; i++) {
268  if (!float_near_ulp(a[i], b[i], max_ulp))
269  return 0;
270  }
271  return 1;
272 }
273 
274 int float_near_abs_eps(float a, float b, float eps)
275 {
276  float abs_diff = fabsf(a - b);
277 
278  return abs_diff < eps;
279 }
280 
281 int float_near_abs_eps_array(const float *a, const float *b, float eps,
282  unsigned len)
283 {
284  unsigned i;
285 
286  for (i = 0; i < len; i++) {
287  if (!float_near_abs_eps(a[i], b[i], eps))
288  return 0;
289  }
290  return 1;
291 }
292 
293 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
294 {
295  return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
296 }
297 
298 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
299  unsigned max_ulp, unsigned len)
300 {
301  unsigned i;
302 
303  for (i = 0; i < len; i++) {
304  if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
305  return 0;
306  }
307  return 1;
308 }
309 
310 int double_near_abs_eps(double a, double b, double eps)
311 {
312  double abs_diff = fabs(a - b);
313 
314  return abs_diff < eps;
315 }
316 
317 int double_near_abs_eps_array(const double *a, const double *b, double eps,
318  unsigned len)
319 {
320  unsigned i;
321 
322  for (i = 0; i < len; i++) {
323  if (!double_near_abs_eps(a[i], b[i], eps))
324  return 0;
325  }
326  return 1;
327 }
328 
329 /* Print colored text to stderr if the terminal supports it */
330 static void color_printf(int color, const char *fmt, ...)
331 {
332  static int use_color = -1;
333  va_list arg;
334 
335 #if HAVE_SETCONSOLETEXTATTRIBUTE
336  static HANDLE con;
337  static WORD org_attributes;
338 
339  if (use_color < 0) {
340  CONSOLE_SCREEN_BUFFER_INFO con_info;
341  con = GetStdHandle(STD_ERROR_HANDLE);
342  if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
343  org_attributes = con_info.wAttributes;
344  use_color = 1;
345  } else
346  use_color = 0;
347  }
348  if (use_color)
349  SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
350 #else
351  if (use_color < 0) {
352  const char *term = getenv("TERM");
353  use_color = term && strcmp(term, "dumb") && isatty(2);
354  }
355  if (use_color)
356  fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
357 #endif
358 
359  va_start(arg, fmt);
360  vfprintf(stderr, fmt, arg);
361  va_end(arg);
362 
363  if (use_color) {
364 #if HAVE_SETCONSOLETEXTATTRIBUTE
365  SetConsoleTextAttribute(con, org_attributes);
366 #else
367  fprintf(stderr, "\x1b[0m");
368 #endif
369  }
370 }
371 
372 /* Deallocate a tree */
374 {
375  if (f) {
377  while (v) {
378  CheckasmFuncVersion *next = v->next;
379  free(v);
380  v = next;
381  }
382 
383  destroy_func_tree(f->child[0]);
384  destroy_func_tree(f->child[1]);
385  free(f);
386  }
387 }
388 
389 /* Allocate a zero-initialized block, clean up and exit on failure */
390 static void *checkasm_malloc(size_t size)
391 {
392  void *ptr = calloc(1, size);
393  if (!ptr) {
394  fprintf(stderr, "checkasm: malloc failed\n");
395  destroy_func_tree(state.funcs);
396  exit(1);
397  }
398  return ptr;
399 }
400 
401 /* Get the suffix of the specified cpu flag */
402 static const char *cpu_suffix(int cpu)
403 {
404  int i = FF_ARRAY_ELEMS(cpus);
405 
406  while (--i >= 0)
407  if (cpu & cpus[i].flag)
408  return cpus[i].suffix;
409 
410  return "c";
411 }
412 
413 static int cmp_nop(const void *a, const void *b)
414 {
415  return *(const uint16_t*)a - *(const uint16_t*)b;
416 }
417 
418 /* Measure the overhead of the timing code (in decicycles) */
419 static int measure_nop_time(void)
420 {
421  uint16_t nops[10000];
422  int i, nop_sum = 0;
423  av_unused const int sysfd = state.sysfd;
424 
425  uint64_t t = 0;
426  for (i = 0; i < 10000; i++) {
427  PERF_START(t);
428  PERF_STOP(t);
429  nops[i] = t;
430  }
431 
432  qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
433  for (i = 2500; i < 7500; i++)
434  nop_sum += nops[i];
435 
436  return nop_sum / 500;
437 }
438 
439 /* Print benchmark results */
441 {
442  if (f) {
443  print_benchs(f->child[0]);
444 
445  /* Only print functions with at least one assembly version */
446  if (f->versions.cpu || f->versions.next) {
447  CheckasmFuncVersion *v = &f->versions;
448  do {
449  CheckasmPerf *p = &v->perf;
450  if (p->iterations) {
451  int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
452  printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
453  }
454  } while ((v = v->next));
455  }
456 
457  print_benchs(f->child[1]);
458  }
459 }
460 
461 /* ASCIIbetical sort except preserving natural order for numbers */
462 static int cmp_func_names(const char *a, const char *b)
463 {
464  const char *start = a;
465  int ascii_diff, digit_diff;
466 
467  for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
468  for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
469 
470  if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
471  return digit_diff;
472 
473  return ascii_diff;
474 }
475 
476 /* Perform a tree rotation in the specified direction and return the new root */
478 {
479  CheckasmFunc *r = f->child[dir^1];
480  f->child[dir^1] = r->child[dir];
481  r->child[dir] = f;
482  r->color = f->color;
483  f->color = 0;
484  return r;
485 }
486 
487 #define is_red(f) ((f) && !(f)->color)
488 
489 /* Balance a left-leaning red-black tree at the specified node */
490 static void balance_tree(CheckasmFunc **root)
491 {
492  CheckasmFunc *f = *root;
493 
494  if (is_red(f->child[0]) && is_red(f->child[1])) {
495  f->color ^= 1;
496  f->child[0]->color = f->child[1]->color = 1;
497  }
498 
499  if (!is_red(f->child[0]) && is_red(f->child[1]))
500  *root = rotate_tree(f, 0); /* Rotate left */
501  else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
502  *root = rotate_tree(f, 1); /* Rotate right */
503 }
504 
505 /* Get a node with the specified name, creating it if it doesn't exist */
506 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
507 {
508  CheckasmFunc *f = *root;
509 
510  if (f) {
511  /* Search the tree for a matching node */
512  int cmp = cmp_func_names(name, f->name);
513  if (cmp) {
514  f = get_func(&f->child[cmp > 0], name);
515 
516  /* Rebalance the tree on the way up if a new node was inserted */
517  if (!f->versions.func)
518  balance_tree(root);
519  }
520  } else {
521  /* Allocate and insert a new node into the tree */
522  int name_length = strlen(name);
523  f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
524  memcpy(f->name, name, name_length + 1);
525  }
526 
527  return f;
528 }
529 
530 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
531 static void check_cpu_flag(const char *name, int flag)
532 {
533  int old_cpu_flag = state.cpu_flag;
534 
535  flag |= old_cpu_flag;
536  av_force_cpu_flags(-1);
537  state.cpu_flag = flag & av_get_cpu_flags();
538  av_force_cpu_flags(state.cpu_flag);
539 
540  if (!flag || state.cpu_flag != old_cpu_flag) {
541  int i;
542 
543  state.cpu_flag_name = name;
544  for (i = 0; tests[i].func; i++) {
545  if (state.test_name && strcmp(tests[i].name, state.test_name))
546  continue;
547  state.current_test_name = tests[i].name;
548  tests[i].func();
549  }
550  }
551 }
552 
553 /* Print the name of the current CPU flag, but only do it once */
554 static void print_cpu_name(void)
555 {
556  if (state.cpu_flag_name) {
557  color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
558  state.cpu_flag_name = NULL;
559  }
560 }
561 
562 #if CONFIG_LINUX_PERF
563 static int bench_init_linux(void)
564 {
565  struct perf_event_attr attr = {
566  .type = PERF_TYPE_HARDWARE,
567  .size = sizeof(struct perf_event_attr),
568  .config = PERF_COUNT_HW_CPU_CYCLES,
569  .disabled = 1, // start counting only on demand
570  .exclude_kernel = 1,
571  .exclude_hv = 1,
572  };
573 
574  printf("benchmarking with Linux Perf Monitoring API\n");
575 
576  state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
577  if (state.sysfd == -1) {
578  perror("syscall");
579  return -1;
580  }
581  return 0;
582 }
583 #endif
584 
585 static int bench_init_ffmpeg(void)
586 {
587 #ifdef AV_READ_TIME
588  printf("benchmarking with native FFmpeg timers\n");
589  return 0;
590 #else
591  fprintf(stderr, "checkasm: --bench is not supported on your system\n");
592  return -1;
593 #endif
594 }
595 
596 static int bench_init(void)
597 {
598 #if CONFIG_LINUX_PERF
599  int ret = bench_init_linux();
600 #else
601  int ret = bench_init_ffmpeg();
602 #endif
603  if (ret < 0)
604  return ret;
605 
606  state.nop_time = measure_nop_time();
607  printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
608  return 0;
609 }
610 
611 static void bench_uninit(void)
612 {
613 #if CONFIG_LINUX_PERF
614  if (state.sysfd > 0)
615  close(state.sysfd);
616 #endif
617 }
618 
619 int main(int argc, char *argv[])
620 {
621  unsigned int seed = av_get_random_seed();
622  int i, ret = 0;
623 
624 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
626  checkasm_checked_call = checkasm_checked_call_vfp;
627 #endif
628 
629  if (!tests[0].func || !cpus[0].flag) {
630  fprintf(stderr, "checkasm: no tests to perform\n");
631  return 0;
632  }
633 
634  while (argc > 1) {
635  if (!strncmp(argv[1], "--bench", 7)) {
636  if (bench_init() < 0)
637  return 1;
638  if (argv[1][7] == '=') {
639  state.bench_pattern = argv[1] + 8;
640  state.bench_pattern_len = strlen(state.bench_pattern);
641  } else
642  state.bench_pattern = "";
643  } else if (!strncmp(argv[1], "--test=", 7)) {
644  state.test_name = argv[1] + 7;
645  } else {
646  seed = strtoul(argv[1], NULL, 10);
647  }
648 
649  argc--;
650  argv++;
651  }
652 
653  fprintf(stderr, "checkasm: using random seed %u\n", seed);
654  av_lfg_init(&checkasm_lfg, seed);
655 
656  check_cpu_flag(NULL, 0);
657  for (i = 0; cpus[i].flag; i++)
658  check_cpu_flag(cpus[i].name, cpus[i].flag);
659 
660  if (state.num_failed) {
661  fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
662  ret = 1;
663  } else {
664  fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
665  if (state.bench_pattern) {
666  print_benchs(state.funcs);
667  }
668  }
669 
670  destroy_func_tree(state.funcs);
671  bench_uninit();
672  return ret;
673 }
674 
675 /* Decide whether or not the specified function needs to be tested and
676  * allocate/initialize data structures if needed. Returns a pointer to a
677  * reference function if the function should be tested, otherwise NULL */
678 void *checkasm_check_func(void *func, const char *name, ...)
679 {
680  char name_buf[256];
681  void *ref = func;
683  int name_length;
684  va_list arg;
685 
686  va_start(arg, name);
687  name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
688  va_end(arg);
689 
690  if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
691  return NULL;
692 
693  state.current_func = get_func(&state.funcs, name_buf);
694  state.funcs->color = 1;
695  v = &state.current_func->versions;
696 
697  if (v->func) {
698  CheckasmFuncVersion *prev;
699  do {
700  /* Only test functions that haven't already been tested */
701  if (v->func == func)
702  return NULL;
703 
704  if (v->ok)
705  ref = v->func;
706 
707  prev = v;
708  } while ((v = v->next));
709 
710  v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
711  }
712 
713  v->func = func;
714  v->ok = 1;
715  v->cpu = state.cpu_flag;
716  state.current_func_ver = v;
717 
718  if (state.cpu_flag)
719  state.num_checked++;
720 
721  return ref;
722 }
723 
724 /* Decide whether or not the current function needs to be benchmarked */
726 {
727  return !state.num_failed && state.bench_pattern &&
728  !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
729 }
730 
731 /* Indicate that the current test has failed */
732 void checkasm_fail_func(const char *msg, ...)
733 {
734  if (state.current_func_ver->cpu && state.current_func_ver->ok) {
735  va_list arg;
736 
737  print_cpu_name();
738  fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
739  va_start(arg, msg);
740  vfprintf(stderr, msg, arg);
741  va_end(arg);
742  fprintf(stderr, ")\n");
743 
744  state.current_func_ver->ok = 0;
745  state.num_failed++;
746  }
747 }
748 
749 /* Get the benchmark context of the current function */
751 {
752  CheckasmPerf *perf = &state.current_func_ver->perf;
753  memset(perf, 0, sizeof(*perf));
754  perf->sysfd = state.sysfd;
755  return perf;
756 }
757 
758 /* Print the outcome of all tests performed since the last time this function was called */
759 void checkasm_report(const char *name, ...)
760 {
761  static int prev_checked, prev_failed, max_length;
762 
763  if (state.num_checked > prev_checked) {
764  int pad_length = max_length + 4;
765  va_list arg;
766 
767  print_cpu_name();
768  pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
769  va_start(arg, name);
770  pad_length -= vfprintf(stderr, name, arg);
771  va_end(arg);
772  fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
773 
774  if (state.num_failed == prev_failed)
775  color_printf(COLOR_GREEN, "OK");
776  else
777  color_printf(COLOR_RED, "FAILED");
778  fprintf(stderr, "]\n");
779 
780  prev_checked = state.num_checked;
781  prev_failed = state.num_failed;
782  } else if (!state.cpu_flag) {
783  /* Calculate the amount of padding required to make the output vertically aligned */
784  int length = strlen(state.current_test_name);
785  va_list arg;
786 
787  va_start(arg, name);
788  length += vsnprintf(NULL, 0, name, arg);
789  va_end(arg);
790 
791  if (length > max_length)
792  max_length = length;
793  }
794 }
Definition: lfg.h:27
const char * name
Definition: checkasm.c:72
static int bench_init_ffmpeg(void)
Definition: checkasm.c:585
static int cmp_nop(const void *a, const void *b)
Definition: checkasm.c:413
#define AV_CPU_FLAG_AVX
AVX functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:49
#define NULL
Definition: coverity.c:32
#define AV_CPU_FLAG_ALTIVEC
standard
Definition: cpu.h:59
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
static int measure_nop_time(void)
Definition: checkasm.c:419
static void bench_uninit(void)
Definition: checkasm.c:611
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
Definition: avstring.h:206
CheckasmFunc * current_func
Definition: checkasm.c:218
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:281
static int is_negative(union av_intfloat32 u)
Definition: checkasm.c:239
uint64_t cycles
Definition: checkasm.h:194
CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:750
static struct @260 state
const char * fmt
Definition: avisynth_c.h:769
static FFServerConfig config
Definition: ffserver.c:193
#define AV_CPU_FLAG_SSE
SSE functions.
Definition: cpu.h:35
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:625
#define vsnprintf
Definition: snprintf.h:36
const char * b
Definition: vf_curves.c:113
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:56
#define AV_CPU_FLAG_CMOV
supports cmov instruction
Definition: cpu.h:53
void checkasm_check_h264pred(void)
Definition: h264pred.c:232
#define AV_CPU_FLAG_VFP
Definition: cpu.h:66
static const char * cpu_suffix(int cpu)
Definition: checkasm.c:402
CheckasmFuncVersion * current_func_ver
Definition: checkasm.c:219
const char * test_name
Definition: checkasm.c:232
void(* func)(void)
Definition: checkasm.c:73
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:115
static int cmp_func_names(const char *a, const char *b)
Definition: checkasm.c:462
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:274
void * checkasm_check_func(void *func, const char *name,...)
Definition: checkasm.c:678
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:215
CheckasmPerf perf
Definition: checkasm.c:204
void checkasm_check_colorspace(void)
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
struct CheckasmFunc * child[2]
Definition: checkasm.c:209
uint8_t
const char * cpu_flag_name
Definition: checkasm.c:231
int checkasm_bench_func(void)
Definition: checkasm.c:725
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
Definition: cpu.h:55
char name[1]
Definition: checkasm.c:212
#define AV_CPU_FLAG_NEON
Definition: cpu.h:68
#define AV_CPU_FLAG_MMXEXT
SSE integer functions or AMD MMX ext.
Definition: cpu.h:32
int main(int argc, char *argv[])
Definition: checkasm.c:619
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
Definition: cpu.h:45
#define COLOR_YELLOW
Definition: checkasm.c:53
static const struct @258 tests[]
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:250
static const struct @259 cpus[]
#define AV_CPU_FLAG_AVX2
AVX2 functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:54
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
Definition: cpu.h:37
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:50
ptrdiff_t size
Definition: opengl_enc.c:101
CheckasmFunc * funcs
Definition: checkasm.c:217
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:58
#define AV_CPU_FLAG_XOP
Bulldozer XOP functions.
Definition: cpu.h:51
int flag
Definition: checkasm.c:161
#define AV_CPU_FLAG_SSE42
Nehalem SSE4.2 functions.
Definition: cpu.h:47
#define isatty(fd)
Definition: checkasm.c:61
static CheckasmFunc * get_func(CheckasmFunc **root, const char *name)
Definition: checkasm.c:506
CheckasmFuncVersion versions
Definition: checkasm.c:210
const char * suffix
Definition: checkasm.c:160
#define AV_CPU_FLAG_SSSE3
Conroe SSSE3 functions.
Definition: cpu.h:43
unsigned short WORD
int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:244
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:229
int num_checked
Definition: checkasm.c:223
#define COLOR_GREEN
Definition: checkasm.c:52
const char * r
Definition: vf_curves.c:111
const char * arg
Definition: jacosubdec.c:66
#define AV_CPU_FLAG_ARMV6T2
Definition: cpu.h:65
GLsizei GLsizei * length
Definition: opengl_enc.c:115
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:74
static int bench_init(void)
Definition: checkasm.c:596
AVLFG checkasm_lfg
Definition: checkasm.c:236
#define AV_CPU_FLAG_VFP_VM
VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations.
Definition: cpu.h:70
#define FFMAX(a, b)
Definition: common.h:94
void checkasm_check_videodsp(void)
Definition: videodsp.c:80
#define AV_CPU_FLAG_ARMV5TE
Definition: cpu.h:63
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:132
static void print_benchs(CheckasmFunc *f)
Definition: checkasm.c:440
#define have_neon(flags)
Definition: cpu.h:27
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
Definition: cpu.h:40
PVOID HANDLE
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
#define is_red(f)
Definition: checkasm.c:487
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:76
uint32_t i
Definition: intfloat.h:28
static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby, const int size, const int h, int ref_index, int src_index, me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags)
compares a block (either a full macroblock or a partition thereof) against a proposed motion-compensa...
Definition: motion_est.c:260
int dummy
Definition: motion.c:64
#define AV_CPU_FLAG_VFPV3
Definition: cpu.h:67
static void * checkasm_malloc(size_t size)
Definition: checkasm.c:390
const char * bench_pattern
Definition: checkasm.c:221
int iterations
Definition: checkasm.h:195
int bench_pattern_len
Definition: checkasm.c:222
void checkasm_report(const char *name,...)
Definition: checkasm.c:759
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
#define FF_ARRAY_ELEMS(a)
int cpu_flag
Definition: checkasm.c:230
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:262
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
Definition: cpu.h:34
int nop_time
Definition: checkasm.c:227
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:310
#define AV_CPU_FLAG_ARMV6
Definition: cpu.h:64
int num_failed
Definition: checkasm.c:224
#define AV_CPU_FLAG_SSE3SLOW
SSE3 supported, but usually not faster.
Definition: cpu.h:41
#define COLOR_RED
Definition: checkasm.c:51
#define AV_CPU_FLAG_MMX
standard MMX
Definition: cpu.h:31
static unsigned int seed
Definition: videogen.c:78
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:44
#define PERF_START(t)
Definition: checkasm.h:243
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
Definition: checkasm.c:477
#define AV_CPU_FLAG_FMA4
Bulldozer FMA4 functions.
Definition: cpu.h:52
#define AV_CPU_FLAG_SSE4
Penryn SSE4.1 functions.
Definition: cpu.h:46
#define AV_CPU_FLAG_AESNI
Advanced Encryption Standard functions.
Definition: cpu.h:48
#define AV_CPU_FLAG_VSX
ISA 2.06.
Definition: cpu.h:60
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:32
int sysfd
Definition: checkasm.c:228
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:66
#define u(width,...)
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:92
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:317
static void color_printf(int color, const char *fmt,...)
Definition: checkasm.c:330
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
Definition: cpu.h:39
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:254
#define AV_CPU_FLAG_POWER8
ISA 2.07.
Definition: cpu.h:61
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:293
static void destroy_func_tree(CheckasmFunc *f)
Definition: checkasm.c:373
common internal and external API header
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
struct CheckasmFuncVersion * next
Definition: checkasm.c:200
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:69
const char * current_test_name
Definition: checkasm.c:220
static void check_cpu_flag(const char *name, int flag)
Definition: checkasm.c:531
static int use_color
Definition: log.c:123
#define have_vfp(flags)
Definition: cpu.h:28
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
#define PERF_STOP(t)
Definition: checkasm.h:244
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:44
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:298
#define AV_CPU_FLAG_SSE2
PIV SSE2 functions.
Definition: cpu.h:36
void INT64 start
Definition: avisynth_c.h:690
uint32_t av_get_random_seed(void)
Get a seed to use in conjunction with random functions.
Definition: random_seed.c:120
void checkasm_fail_func(const char *msg,...)
Definition: checkasm.c:732
#define INVALID_HANDLE_VALUE
Definition: windows2linux.h:47
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:59
void av_force_cpu_flags(int arg)
Disables cpu detection and forces the specified flags.
Definition: cpu.c:65
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
static void balance_tree(CheckasmFunc **root)
Definition: checkasm.c:490
void checkasm_check_blend(void)
Definition: vf_blend.c:87
#define av_unused
Definition: attributes.h:125
const char * name
Definition: opengl_enc.c:103
static void print_cpu_name(void)
Definition: checkasm.c:554
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:54
uint8_t color
Definition: checkasm.c:211