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  { "hevc_sao", checkasm_check_hevc_sao },
120  #endif
121  #if CONFIG_HUFFYUV_DECODER
122  { "huffyuvdsp", checkasm_check_huffyuvdsp },
123  #endif
124  #if CONFIG_JPEG2000_DECODER
125  { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
126  #endif
127  #if CONFIG_HUFFYUVDSP
128  { "llviddsp", checkasm_check_llviddsp },
129  #endif
130  #if CONFIG_LLVIDENCDSP
131  { "llviddspenc", checkasm_check_llviddspenc },
132  #endif
133  #if CONFIG_PIXBLOCKDSP
134  { "pixblockdsp", checkasm_check_pixblockdsp },
135  #endif
136  #if CONFIG_UTVIDEO_DECODER
137  { "utvideodsp", checkasm_check_utvideodsp },
138  #endif
139  #if CONFIG_V210_ENCODER
140  { "v210enc", checkasm_check_v210enc },
141  #endif
142  #if CONFIG_VP8DSP
143  { "vp8dsp", checkasm_check_vp8dsp },
144  #endif
145  #if CONFIG_VP9_DECODER
146  { "vp9dsp", checkasm_check_vp9dsp },
147  #endif
148  #if CONFIG_VIDEODSP
149  { "videodsp", checkasm_check_videodsp },
150  #endif
151 #endif
152 #if CONFIG_AVFILTER
153  #if CONFIG_AFIR_FILTER
154  { "af_afir", checkasm_check_afir },
155  #endif
156  #if CONFIG_BLEND_FILTER
157  { "vf_blend", checkasm_check_blend },
158  #endif
159  #if CONFIG_COLORSPACE_FILTER
160  { "vf_colorspace", checkasm_check_colorspace },
161  #endif
162  #if CONFIG_HFLIP_FILTER
163  { "vf_hflip", checkasm_check_vf_hflip },
164  #endif
165  #if CONFIG_NLMEANS_FILTER
166  { "vf_nlmeans", checkasm_check_nlmeans },
167  #endif
168  #if CONFIG_THRESHOLD_FILTER
169  { "vf_threshold", checkasm_check_vf_threshold },
170  #endif
171 #endif
172 #if CONFIG_SWSCALE
173  { "sw_rgb", checkasm_check_sw_rgb },
174 #endif
175 #if CONFIG_AVUTIL
176  { "fixed_dsp", checkasm_check_fixed_dsp },
177  { "float_dsp", checkasm_check_float_dsp },
178 #endif
179  { NULL }
180 };
181 
182 /* List of cpu flags to check */
183 static const struct {
184  const char *name;
185  const char *suffix;
186  int flag;
187 } cpus[] = {
188 #if ARCH_AARCH64
189  { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
190  { "NEON", "neon", AV_CPU_FLAG_NEON },
191 #elif ARCH_ARM
192  { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
193  { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
194  { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
195  { "VFP", "vfp", AV_CPU_FLAG_VFP },
196  { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
197  { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
198  { "NEON", "neon", AV_CPU_FLAG_NEON },
199 #elif ARCH_PPC
200  { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
201  { "VSX", "vsx", AV_CPU_FLAG_VSX },
202  { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
203 #elif ARCH_X86
204  { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
205  { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
206  { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
207  { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
208  { "SSE", "sse", AV_CPU_FLAG_SSE },
209  { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
210  { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
211  { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
212  { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
213  { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
214  { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
215  { "AVX", "avx", AV_CPU_FLAG_AVX },
216  { "XOP", "xop", AV_CPU_FLAG_XOP },
217  { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
218  { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
219  { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
220  { "AVX-512", "avx512", AV_CPU_FLAG_AVX512 },
221 #endif
222  { NULL }
223 };
224 
225 typedef struct CheckasmFuncVersion {
227  void *func;
228  int ok;
229  int cpu;
232 
233 /* Binary search tree node */
234 typedef struct CheckasmFunc {
235  struct CheckasmFunc *child[2];
237  uint8_t color; /* 0 = red, 1 = black */
238  char name[1];
239 } CheckasmFunc;
240 
241 /* Internal state */
242 static struct {
246  const char *current_test_name;
247  const char *bench_pattern;
251 
252  /* perf */
253  int nop_time;
254  int sysfd;
255 
256  int cpu_flag;
257  const char *cpu_flag_name;
258  const char *test_name;
259 } state;
260 
261 /* PRNG state */
263 
264 /* float compare support code */
265 static int is_negative(union av_intfloat32 u)
266 {
267  return u.i >> 31;
268 }
269 
270 int float_near_ulp(float a, float b, unsigned max_ulp)
271 {
272  union av_intfloat32 x, y;
273 
274  x.f = a;
275  y.f = b;
276 
277  if (is_negative(x) != is_negative(y)) {
278  // handle -0.0 == +0.0
279  return a == b;
280  }
281 
282  if (llabs((int64_t)x.i - y.i) <= max_ulp)
283  return 1;
284 
285  return 0;
286 }
287 
288 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
289  unsigned len)
290 {
291  unsigned i;
292 
293  for (i = 0; i < len; i++) {
294  if (!float_near_ulp(a[i], b[i], max_ulp))
295  return 0;
296  }
297  return 1;
298 }
299 
300 int float_near_abs_eps(float a, float b, float eps)
301 {
302  float abs_diff = fabsf(a - b);
303  if (abs_diff < eps)
304  return 1;
305 
306  fprintf(stderr, "test failed comparing %g with %g (abs diff=%g with EPS=%g)\n", a, b, abs_diff, eps);
307 
308  return 0;
309 }
310 
311 int float_near_abs_eps_array(const float *a, const float *b, float eps,
312  unsigned len)
313 {
314  unsigned i;
315 
316  for (i = 0; i < len; i++) {
317  if (!float_near_abs_eps(a[i], b[i], eps))
318  return 0;
319  }
320  return 1;
321 }
322 
323 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
324 {
325  return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
326 }
327 
328 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
329  unsigned max_ulp, unsigned len)
330 {
331  unsigned i;
332 
333  for (i = 0; i < len; i++) {
334  if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
335  return 0;
336  }
337  return 1;
338 }
339 
340 int double_near_abs_eps(double a, double b, double eps)
341 {
342  double abs_diff = fabs(a - b);
343 
344  return abs_diff < eps;
345 }
346 
347 int double_near_abs_eps_array(const double *a, const double *b, double eps,
348  unsigned len)
349 {
350  unsigned i;
351 
352  for (i = 0; i < len; i++) {
353  if (!double_near_abs_eps(a[i], b[i], eps))
354  return 0;
355  }
356  return 1;
357 }
358 
359 /* Print colored text to stderr if the terminal supports it */
360 static void color_printf(int color, const char *fmt, ...)
361 {
362  static int use_color = -1;
363  va_list arg;
364 
365 #if HAVE_SETCONSOLETEXTATTRIBUTE
366  static HANDLE con;
367  static WORD org_attributes;
368 
369  if (use_color < 0) {
370  CONSOLE_SCREEN_BUFFER_INFO con_info;
371  con = GetStdHandle(STD_ERROR_HANDLE);
372  if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
373  org_attributes = con_info.wAttributes;
374  use_color = 1;
375  } else
376  use_color = 0;
377  }
378  if (use_color)
379  SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
380 #else
381  if (use_color < 0) {
382  const char *term = getenv("TERM");
383  use_color = term && strcmp(term, "dumb") && isatty(2);
384  }
385  if (use_color)
386  fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
387 #endif
388 
389  va_start(arg, fmt);
390  vfprintf(stderr, fmt, arg);
391  va_end(arg);
392 
393  if (use_color) {
394 #if HAVE_SETCONSOLETEXTATTRIBUTE
395  SetConsoleTextAttribute(con, org_attributes);
396 #else
397  fprintf(stderr, "\x1b[0m");
398 #endif
399  }
400 }
401 
402 /* Deallocate a tree */
404 {
405  if (f) {
407  while (v) {
408  CheckasmFuncVersion *next = v->next;
409  free(v);
410  v = next;
411  }
412 
413  destroy_func_tree(f->child[0]);
414  destroy_func_tree(f->child[1]);
415  free(f);
416  }
417 }
418 
419 /* Allocate a zero-initialized block, clean up and exit on failure */
420 static void *checkasm_malloc(size_t size)
421 {
422  void *ptr = calloc(1, size);
423  if (!ptr) {
424  fprintf(stderr, "checkasm: malloc failed\n");
425  destroy_func_tree(state.funcs);
426  exit(1);
427  }
428  return ptr;
429 }
430 
431 /* Get the suffix of the specified cpu flag */
432 static const char *cpu_suffix(int cpu)
433 {
434  int i = FF_ARRAY_ELEMS(cpus);
435 
436  while (--i >= 0)
437  if (cpu & cpus[i].flag)
438  return cpus[i].suffix;
439 
440  return "c";
441 }
442 
443 static int cmp_nop(const void *a, const void *b)
444 {
445  return *(const uint16_t*)a - *(const uint16_t*)b;
446 }
447 
448 /* Measure the overhead of the timing code (in decicycles) */
449 static int measure_nop_time(void)
450 {
451  uint16_t nops[10000];
452  int i, nop_sum = 0;
453  av_unused const int sysfd = state.sysfd;
454 
455  uint64_t t = 0;
456  for (i = 0; i < 10000; i++) {
457  PERF_START(t);
458  PERF_STOP(t);
459  nops[i] = t;
460  }
461 
462  qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
463  for (i = 2500; i < 7500; i++)
464  nop_sum += nops[i];
465 
466  return nop_sum / 500;
467 }
468 
469 /* Print benchmark results */
471 {
472  if (f) {
473  print_benchs(f->child[0]);
474 
475  /* Only print functions with at least one assembly version */
476  if (f->versions.cpu || f->versions.next) {
477  CheckasmFuncVersion *v = &f->versions;
478  do {
479  CheckasmPerf *p = &v->perf;
480  if (p->iterations) {
481  int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
482  printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
483  }
484  } while ((v = v->next));
485  }
486 
487  print_benchs(f->child[1]);
488  }
489 }
490 
491 /* ASCIIbetical sort except preserving natural order for numbers */
492 static int cmp_func_names(const char *a, const char *b)
493 {
494  const char *start = a;
495  int ascii_diff, digit_diff;
496 
497  for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
498  for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
499 
500  if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
501  return digit_diff;
502 
503  return ascii_diff;
504 }
505 
506 /* Perform a tree rotation in the specified direction and return the new root */
508 {
509  CheckasmFunc *r = f->child[dir^1];
510  f->child[dir^1] = r->child[dir];
511  r->child[dir] = f;
512  r->color = f->color;
513  f->color = 0;
514  return r;
515 }
516 
517 #define is_red(f) ((f) && !(f)->color)
518 
519 /* Balance a left-leaning red-black tree at the specified node */
520 static void balance_tree(CheckasmFunc **root)
521 {
522  CheckasmFunc *f = *root;
523 
524  if (is_red(f->child[0]) && is_red(f->child[1])) {
525  f->color ^= 1;
526  f->child[0]->color = f->child[1]->color = 1;
527  }
528 
529  if (!is_red(f->child[0]) && is_red(f->child[1]))
530  *root = rotate_tree(f, 0); /* Rotate left */
531  else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
532  *root = rotate_tree(f, 1); /* Rotate right */
533 }
534 
535 /* Get a node with the specified name, creating it if it doesn't exist */
536 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
537 {
538  CheckasmFunc *f = *root;
539 
540  if (f) {
541  /* Search the tree for a matching node */
542  int cmp = cmp_func_names(name, f->name);
543  if (cmp) {
544  f = get_func(&f->child[cmp > 0], name);
545 
546  /* Rebalance the tree on the way up if a new node was inserted */
547  if (!f->versions.func)
548  balance_tree(root);
549  }
550  } else {
551  /* Allocate and insert a new node into the tree */
552  int name_length = strlen(name);
553  f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
554  memcpy(f->name, name, name_length + 1);
555  }
556 
557  return f;
558 }
559 
560 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
561 static void check_cpu_flag(const char *name, int flag)
562 {
563  int old_cpu_flag = state.cpu_flag;
564 
565  flag |= old_cpu_flag;
566  av_force_cpu_flags(-1);
567  state.cpu_flag = flag & av_get_cpu_flags();
568  av_force_cpu_flags(state.cpu_flag);
569 
570  if (!flag || state.cpu_flag != old_cpu_flag) {
571  int i;
572 
573  state.cpu_flag_name = name;
574  for (i = 0; tests[i].func; i++) {
575  if (state.test_name && strcmp(tests[i].name, state.test_name))
576  continue;
577  state.current_test_name = tests[i].name;
578  tests[i].func();
579  }
580  }
581 }
582 
583 /* Print the name of the current CPU flag, but only do it once */
584 static void print_cpu_name(void)
585 {
586  if (state.cpu_flag_name) {
587  color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
588  state.cpu_flag_name = NULL;
589  }
590 }
591 
592 #if CONFIG_LINUX_PERF
593 static int bench_init_linux(void)
594 {
595  struct perf_event_attr attr = {
596  .type = PERF_TYPE_HARDWARE,
597  .size = sizeof(struct perf_event_attr),
598  .config = PERF_COUNT_HW_CPU_CYCLES,
599  .disabled = 1, // start counting only on demand
600  .exclude_kernel = 1,
601  .exclude_hv = 1,
602  };
603 
604  printf("benchmarking with Linux Perf Monitoring API\n");
605 
606  state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
607  if (state.sysfd == -1) {
608  perror("syscall");
609  return -1;
610  }
611  return 0;
612 }
613 #endif
614 
615 static int bench_init_ffmpeg(void)
616 {
617 #ifdef AV_READ_TIME
618  printf("benchmarking with native FFmpeg timers\n");
619  return 0;
620 #else
621  fprintf(stderr, "checkasm: --bench is not supported on your system\n");
622  return -1;
623 #endif
624 }
625 
626 static int bench_init(void)
627 {
628 #if CONFIG_LINUX_PERF
629  int ret = bench_init_linux();
630 #else
631  int ret = bench_init_ffmpeg();
632 #endif
633  if (ret < 0)
634  return ret;
635 
636  state.nop_time = measure_nop_time();
637  printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
638  return 0;
639 }
640 
641 static void bench_uninit(void)
642 {
643 #if CONFIG_LINUX_PERF
644  if (state.sysfd > 0)
645  close(state.sysfd);
646 #endif
647 }
648 
649 int main(int argc, char *argv[])
650 {
651  unsigned int seed = av_get_random_seed();
652  int i, ret = 0;
653 
654 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
656  checkasm_checked_call = checkasm_checked_call_vfp;
657 #endif
658 
659  if (!tests[0].func || !cpus[0].flag) {
660  fprintf(stderr, "checkasm: no tests to perform\n");
661  return 0;
662  }
663 
664  while (argc > 1) {
665  if (!strncmp(argv[1], "--bench", 7)) {
666  if (bench_init() < 0)
667  return 1;
668  if (argv[1][7] == '=') {
669  state.bench_pattern = argv[1] + 8;
670  state.bench_pattern_len = strlen(state.bench_pattern);
671  } else
672  state.bench_pattern = "";
673  } else if (!strncmp(argv[1], "--test=", 7)) {
674  state.test_name = argv[1] + 7;
675  } else {
676  seed = strtoul(argv[1], NULL, 10);
677  }
678 
679  argc--;
680  argv++;
681  }
682 
683  fprintf(stderr, "checkasm: using random seed %u\n", seed);
684  av_lfg_init(&checkasm_lfg, seed);
685 
686  check_cpu_flag(NULL, 0);
687  for (i = 0; cpus[i].flag; i++)
688  check_cpu_flag(cpus[i].name, cpus[i].flag);
689 
690  if (state.num_failed) {
691  fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
692  ret = 1;
693  } else {
694  fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
695  if (state.bench_pattern) {
696  print_benchs(state.funcs);
697  }
698  }
699 
700  destroy_func_tree(state.funcs);
701  bench_uninit();
702  return ret;
703 }
704 
705 /* Decide whether or not the specified function needs to be tested and
706  * allocate/initialize data structures if needed. Returns a pointer to a
707  * reference function if the function should be tested, otherwise NULL */
708 void *checkasm_check_func(void *func, const char *name, ...)
709 {
710  char name_buf[256];
711  void *ref = func;
713  int name_length;
714  va_list arg;
715 
716  va_start(arg, name);
717  name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
718  va_end(arg);
719 
720  if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
721  return NULL;
722 
723  state.current_func = get_func(&state.funcs, name_buf);
724  state.funcs->color = 1;
725  v = &state.current_func->versions;
726 
727  if (v->func) {
728  CheckasmFuncVersion *prev;
729  do {
730  /* Only test functions that haven't already been tested */
731  if (v->func == func)
732  return NULL;
733 
734  if (v->ok)
735  ref = v->func;
736 
737  prev = v;
738  } while ((v = v->next));
739 
740  v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
741  }
742 
743  v->func = func;
744  v->ok = 1;
745  v->cpu = state.cpu_flag;
746  state.current_func_ver = v;
747 
748  if (state.cpu_flag)
749  state.num_checked++;
750 
751  return ref;
752 }
753 
754 /* Decide whether or not the current function needs to be benchmarked */
756 {
757  return !state.num_failed && state.bench_pattern &&
758  !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
759 }
760 
761 /* Indicate that the current test has failed */
762 void checkasm_fail_func(const char *msg, ...)
763 {
764  if (state.current_func_ver->cpu && state.current_func_ver->ok) {
765  va_list arg;
766 
767  print_cpu_name();
768  fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
769  va_start(arg, msg);
770  vfprintf(stderr, msg, arg);
771  va_end(arg);
772  fprintf(stderr, ")\n");
773 
774  state.current_func_ver->ok = 0;
775  state.num_failed++;
776  }
777 }
778 
779 /* Get the benchmark context of the current function */
781 {
782  CheckasmPerf *perf = &state.current_func_ver->perf;
783  memset(perf, 0, sizeof(*perf));
784  perf->sysfd = state.sysfd;
785  return perf;
786 }
787 
788 /* Print the outcome of all tests performed since the last time this function was called */
789 void checkasm_report(const char *name, ...)
790 {
791  static int prev_checked, prev_failed, max_length;
792 
793  if (state.num_checked > prev_checked) {
794  int pad_length = max_length + 4;
795  va_list arg;
796 
797  print_cpu_name();
798  pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
799  va_start(arg, name);
800  pad_length -= vfprintf(stderr, name, arg);
801  va_end(arg);
802  fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
803 
804  if (state.num_failed == prev_failed)
805  color_printf(COLOR_GREEN, "OK");
806  else
807  color_printf(COLOR_RED, "FAILED");
808  fprintf(stderr, "]\n");
809 
810  prev_checked = state.num_checked;
811  prev_failed = state.num_failed;
812  } else if (!state.cpu_flag) {
813  /* Calculate the amount of padding required to make the output vertically aligned */
814  int length = strlen(state.current_test_name);
815  va_list arg;
816 
817  va_start(arg, name);
818  length += vsnprintf(NULL, 0, name, arg);
819  va_end(arg);
820 
821  if (length > max_length)
822  max_length = length;
823  }
824 }
Definition: lfg.h:27
static const struct @303 cpus[]
const char * name
Definition: checkasm.c:72
static int bench_init_ffmpeg(void)
Definition: checkasm.c:615
static int cmp_nop(const void *a, const void *b)
Definition: checkasm.c:443
#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:60
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
static int measure_nop_time(void)
Definition: checkasm.c:449
static void bench_uninit(void)
Definition: checkasm.c:641
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
Definition: avstring.h:206
CheckasmFunc * current_func
Definition: checkasm.c:244
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:311
static int is_negative(union av_intfloat32 u)
Definition: checkasm.c:265
uint64_t cycles
Definition: checkasm.h:203
CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:780
const char * fmt
Definition: avisynth_c.h:769
#define AV_CPU_FLAG_SSE
SSE functions.
Definition: cpu.h:35
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:625
void checkasm_check_vf_threshold(void)
Definition: vf_threshold.c:78
#define vsnprintf
Definition: snprintf.h:36
const char * b
Definition: vf_curves.c:116
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:67
static const char * cpu_suffix(int cpu)
Definition: checkasm.c:432
void checkasm_check_afir(void)
Definition: af_afir.c:67
CheckasmFuncVersion * current_func_ver
Definition: checkasm.c:245
const char * test_name
Definition: checkasm.c:258
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:492
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:300
void * checkasm_check_func(void *func, const char *name,...)
Definition: checkasm.c:708
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:215
CheckasmPerf perf
Definition: checkasm.c:230
void checkasm_check_colorspace(void)
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
struct CheckasmFunc * child[2]
Definition: checkasm.c:235
uint8_t
const char * cpu_flag_name
Definition: checkasm.c:257
int checkasm_bench_func(void)
Definition: checkasm.c:755
#define f(width, name)
Definition: cbs_vp9.c:255
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
Definition: cpu.h:55
char name[1]
Definition: checkasm.c:238
#define AV_CPU_FLAG_NEON
Definition: cpu.h:69
#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:649
#define u(width, name, range_min, range_max)
Definition: cbs_h2645.c:253
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
Definition: cpu.h:45
#define COLOR_YELLOW
Definition: checkasm.c:53
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:250
#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:243
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:91
#define AV_CPU_FLAG_XOP
Bulldozer XOP functions.
Definition: cpu.h:51
void checkasm_check_llviddspenc(void)
Definition: llviddspenc.c:104
int flag
Definition: checkasm.c:186
#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:536
CheckasmFuncVersion versions
Definition: checkasm.c:236
const char * suffix
Definition: checkasm.c:185
#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:270
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:315
int num_checked
Definition: checkasm.c:249
#define COLOR_GREEN
Definition: checkasm.c:52
const char * r
Definition: vf_curves.c:114
const char * arg
Definition: jacosubdec.c:66
#define AV_CPU_FLAG_ARMV6T2
Definition: cpu.h:66
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:626
AVLFG checkasm_lfg
Definition: checkasm.c:262
#define AV_CPU_FLAG_AVX512
AVX-512 functions: requires OS support even if YMM/ZMM registers aren't used.
Definition: cpu.h:58
#define AV_CPU_FLAG_VFP_VM
VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations.
Definition: cpu.h:71
#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:64
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:132
static void print_benchs(CheckasmFunc *f)
Definition: checkasm.c:470
#define have_neon(flags)
Definition: cpu.h:26
void checkasm_check_utvideodsp(void)
Definition: utvideodsp.c:90
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
Definition: cpu.h:40
void checkasm_check_nlmeans(void)
Definition: vf_nlmeans.c:31
PVOID HANDLE
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:114
#define is_red(f)
Definition: checkasm.c:517
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:68
static void * checkasm_malloc(size_t size)
Definition: checkasm.c:420
const char * bench_pattern
Definition: checkasm.c:247
int iterations
Definition: checkasm.h:204
int bench_pattern_len
Definition: checkasm.c:248
void checkasm_report(const char *name,...)
Definition: checkasm.c:789
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
#define FF_ARRAY_ELEMS(a)
int cpu_flag
Definition: checkasm.c:256
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:288
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
Definition: cpu.h:34
int nop_time
Definition: checkasm.c:253
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:340
#define AV_CPU_FLAG_ARMV6
Definition: cpu.h:65
int num_failed
Definition: checkasm.c:250
#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:252
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
Definition: checkasm.c:507
#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:61
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:32
int sysfd
Definition: checkasm.c:254
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:195
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:93
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:347
static const struct @302 tests[]
static void color_printf(int color, const char *fmt,...)
Definition: checkasm.c:360
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
Definition: cpu.h:39
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:277
#define AV_CPU_FLAG_POWER8
ISA 2.07.
Definition: cpu.h:62
void checkasm_check_huffyuvdsp(void)
Definition: huffyuvdsp.c:67
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:323
static void destroy_func_tree(CheckasmFunc *f)
Definition: checkasm.c:403
common internal and external API header
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
struct CheckasmFuncVersion * next
Definition: checkasm.c:226
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:70
const char * current_test_name
Definition: checkasm.c:246
static void check_cpu_flag(const char *name, int flag)
Definition: checkasm.c:561
static int use_color
Definition: log.c:121
#define have_vfp(flags)
Definition: cpu.h:27
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
void checkasm_check_hevc_sao(void)
Definition: hevc_sao.c:131
#define PERF_STOP(t)
Definition: checkasm.h:253
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:328
#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:762
#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:520
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:69
static struct @304 state
void checkasm_check_blend(void)
Definition: vf_blend.c:88
#define av_unused
Definition: attributes.h:125
const char * name
Definition: opengl_enc.c:103
static void print_cpu_name(void)
Definition: checkasm.c:584
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:54
uint8_t color
Definition: checkasm.c:237