FFmpeg
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 #include "config_components.h"
25 
26 #ifndef _GNU_SOURCE
27 # define _GNU_SOURCE // for syscall (performance monitoring API), strsignal()
28 #endif
29 
30 #include <signal.h>
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 defined(_WIN32) && !defined(SIGBUS)
46 /* non-standard, use the same value as mingw-w64 */
47 #define SIGBUS 10
48 #endif
49 
50 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
51 #include <windows.h>
52 #define COLOR_RED FOREGROUND_RED
53 #define COLOR_GREEN FOREGROUND_GREEN
54 #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
55 #else
56 #define COLOR_RED 1
57 #define COLOR_GREEN 2
58 #define COLOR_YELLOW 3
59 #endif
60 
61 #if HAVE_UNISTD_H
62 #include <unistd.h>
63 #endif
64 
65 #if !HAVE_ISATTY
66 #define isatty(fd) 1
67 #endif
68 
69 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
70 #include "libavutil/arm/cpu.h"
71 
72 void (*checkasm_checked_call)(void *func, int dummy, ...) = checkasm_checked_call_novfp;
73 #endif
74 
75 /* List of tests to invoke */
76 static const struct {
77  const char *name;
78  void (*func)(void);
79 } tests[] = {
80 #if CONFIG_AVCODEC
81  #if CONFIG_AAC_DECODER
82  { "aacpsdsp", checkasm_check_aacpsdsp },
83  { "sbrdsp", checkasm_check_sbrdsp },
84  #endif
85  #if CONFIG_AAC_ENCODER
86  { "aacencdsp", checkasm_check_aacencdsp },
87  #endif
88  #if CONFIG_AC3DSP
89  { "ac3dsp", checkasm_check_ac3dsp },
90  #endif
91  #if CONFIG_ALAC_DECODER
92  { "alacdsp", checkasm_check_alacdsp },
93  #endif
94  #if CONFIG_AUDIODSP
95  { "audiodsp", checkasm_check_audiodsp },
96  #endif
97  #if CONFIG_BLOCKDSP
98  { "blockdsp", checkasm_check_blockdsp },
99  #endif
100  #if CONFIG_BSWAPDSP
101  { "bswapdsp", checkasm_check_bswapdsp },
102  #endif
103  #if CONFIG_DCA_DECODER
104  { "synth_filter", checkasm_check_synth_filter },
105  #endif
106  #if CONFIG_EXR_DECODER
107  { "exrdsp", checkasm_check_exrdsp },
108  #endif
109  #if CONFIG_FDCTDSP
110  { "fdctdsp", checkasm_check_fdctdsp },
111  #endif
112  #if CONFIG_FLAC_DECODER
113  { "flacdsp", checkasm_check_flacdsp },
114  #endif
115  #if CONFIG_FMTCONVERT
116  { "fmtconvert", checkasm_check_fmtconvert },
117  #endif
118  #if CONFIG_G722DSP
119  { "g722dsp", checkasm_check_g722dsp },
120  #endif
121  #if CONFIG_H264CHROMA
122  { "h264chroma", checkasm_check_h264chroma },
123  #endif
124  #if CONFIG_H264DSP
125  { "h264dsp", checkasm_check_h264dsp },
126  #endif
127  #if CONFIG_H264PRED
128  { "h264pred", checkasm_check_h264pred },
129  #endif
130  #if CONFIG_H264QPEL
131  { "h264qpel", checkasm_check_h264qpel },
132  #endif
133  #if CONFIG_HEVC_DECODER
134  { "hevc_add_res", checkasm_check_hevc_add_res },
135  { "hevc_deblock", checkasm_check_hevc_deblock },
136  { "hevc_idct", checkasm_check_hevc_idct },
137  { "hevc_pel", checkasm_check_hevc_pel },
138  { "hevc_sao", checkasm_check_hevc_sao },
139  #endif
140  #if CONFIG_HUFFYUV_DECODER
141  { "huffyuvdsp", checkasm_check_huffyuvdsp },
142  #endif
143  #if CONFIG_IDCTDSP
144  { "idctdsp", checkasm_check_idctdsp },
145  #endif
146  #if CONFIG_JPEG2000_DECODER
147  { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
148  #endif
149  #if CONFIG_LLAUDDSP
150  { "llauddsp", checkasm_check_llauddsp },
151  #endif
152  #if CONFIG_HUFFYUVDSP
153  { "llviddsp", checkasm_check_llviddsp },
154  #endif
155  #if CONFIG_LLVIDENCDSP
156  { "llviddspenc", checkasm_check_llviddspenc },
157  #endif
158  #if CONFIG_LPC
159  { "lpc", checkasm_check_lpc },
160  #endif
161  #if CONFIG_ME_CMP
162  { "motion", checkasm_check_motion },
163  #endif
164  #if CONFIG_OPUS_DECODER
165  { "opusdsp", checkasm_check_opusdsp },
166  #endif
167  #if CONFIG_PIXBLOCKDSP
168  { "pixblockdsp", checkasm_check_pixblockdsp },
169  #endif
170  #if CONFIG_RV34DSP
171  { "rv34dsp", checkasm_check_rv34dsp },
172  #endif
173  #if CONFIG_RV40_DECODER
174  { "rv40dsp", checkasm_check_rv40dsp },
175  #endif
176  #if CONFIG_SVQ1_ENCODER
177  { "svq1enc", checkasm_check_svq1enc },
178  #endif
179  #if CONFIG_TAK_DECODER
180  { "takdsp", checkasm_check_takdsp },
181  #endif
182  #if CONFIG_UTVIDEO_DECODER
183  { "utvideodsp", checkasm_check_utvideodsp },
184  #endif
185  #if CONFIG_V210_DECODER
186  { "v210dec", checkasm_check_v210dec },
187  #endif
188  #if CONFIG_V210_ENCODER
189  { "v210enc", checkasm_check_v210enc },
190  #endif
191  #if CONFIG_VC1DSP
192  { "vc1dsp", checkasm_check_vc1dsp },
193  #endif
194  #if CONFIG_VP8DSP
195  { "vp8dsp", checkasm_check_vp8dsp },
196  #endif
197  #if CONFIG_VP9_DECODER
198  { "vp9dsp", checkasm_check_vp9dsp },
199  #endif
200  #if CONFIG_VIDEODSP
201  { "videodsp", checkasm_check_videodsp },
202  #endif
203  #if CONFIG_VORBIS_DECODER
204  { "vorbisdsp", checkasm_check_vorbisdsp },
205  #endif
206  #if CONFIG_VVC_DECODER
207  { "vvc_alf", checkasm_check_vvc_alf },
208  { "vvc_mc", checkasm_check_vvc_mc },
209  #endif
210 #endif
211 #if CONFIG_AVFILTER
212  #if CONFIG_AFIR_FILTER
213  { "af_afir", checkasm_check_afir },
214  #endif
215  #if CONFIG_BLEND_FILTER
216  { "vf_blend", checkasm_check_blend },
217  #endif
218  #if CONFIG_BWDIF_FILTER
219  { "vf_bwdif", checkasm_check_vf_bwdif },
220  #endif
221  #if CONFIG_COLORSPACE_FILTER
222  { "vf_colorspace", checkasm_check_colorspace },
223  #endif
224  #if CONFIG_EQ_FILTER
225  { "vf_eq", checkasm_check_vf_eq },
226  #endif
227  #if CONFIG_GBLUR_FILTER
228  { "vf_gblur", checkasm_check_vf_gblur },
229  #endif
230  #if CONFIG_HFLIP_FILTER
231  { "vf_hflip", checkasm_check_vf_hflip },
232  #endif
233  #if CONFIG_NLMEANS_FILTER
234  { "vf_nlmeans", checkasm_check_nlmeans },
235  #endif
236  #if CONFIG_THRESHOLD_FILTER
237  { "vf_threshold", checkasm_check_vf_threshold },
238  #endif
239  #if CONFIG_SOBEL_FILTER
240  { "vf_sobel", checkasm_check_vf_sobel },
241  #endif
242 #endif
243 #if CONFIG_SWSCALE
244  { "sw_gbrp", checkasm_check_sw_gbrp },
245  { "sw_rgb", checkasm_check_sw_rgb },
246  { "sw_scale", checkasm_check_sw_scale },
247 #endif
248 #if CONFIG_AVUTIL
249  { "fixed_dsp", checkasm_check_fixed_dsp },
250  { "float_dsp", checkasm_check_float_dsp },
251  { "av_tx", checkasm_check_av_tx },
252 #endif
253  { NULL }
254 };
255 
256 /* List of cpu flags to check */
257 static const struct {
258  const char *name;
259  const char *suffix;
260  int flag;
261 } cpus[] = {
262 #if ARCH_AARCH64
263  { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
264  { "NEON", "neon", AV_CPU_FLAG_NEON },
265  { "DOTPROD", "dotprod", AV_CPU_FLAG_DOTPROD },
266  { "I8MM", "i8mm", AV_CPU_FLAG_I8MM },
267 #elif ARCH_ARM
268  { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
269  { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
270  { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
271  { "VFP", "vfp", AV_CPU_FLAG_VFP },
272  { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
273  { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
274  { "NEON", "neon", AV_CPU_FLAG_NEON },
275 #elif ARCH_PPC
276  { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
277  { "VSX", "vsx", AV_CPU_FLAG_VSX },
278  { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
279 #elif ARCH_RISCV
280  { "RVI", "rvi", AV_CPU_FLAG_RVI },
281  { "RVF", "rvf", AV_CPU_FLAG_RVF },
282  { "RVD", "rvd", AV_CPU_FLAG_RVD },
283  { "RVBaddr", "rvb_a", AV_CPU_FLAG_RVB_ADDR },
284  { "RVBbasic", "rvb_b", AV_CPU_FLAG_RVB_BASIC },
285  { "RVVi32", "rvv_i32", AV_CPU_FLAG_RVV_I32 },
286  { "RVVf32", "rvv_f32", AV_CPU_FLAG_RVV_F32 },
287  { "RVVi64", "rvv_i64", AV_CPU_FLAG_RVV_I64 },
288  { "RVVf64", "rvv_f64", AV_CPU_FLAG_RVV_F64 },
289  { "RV_Zvbb", "rv_zvbb", AV_CPU_FLAG_RV_ZVBB },
290  { "misaligned", "misaligned", AV_CPU_FLAG_RV_MISALIGNED },
291 #elif ARCH_MIPS
292  { "MMI", "mmi", AV_CPU_FLAG_MMI },
293  { "MSA", "msa", AV_CPU_FLAG_MSA },
294 #elif ARCH_X86
295  { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
296  { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
297  { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
298  { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
299  { "SSE", "sse", AV_CPU_FLAG_SSE },
300  { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
301  { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
302  { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
303  { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
304  { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
305  { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
306  { "AVX", "avx", AV_CPU_FLAG_AVX },
307  { "XOP", "xop", AV_CPU_FLAG_XOP },
308  { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
309  { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
310  { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
311  { "AVX-512", "avx512", AV_CPU_FLAG_AVX512 },
312  { "AVX-512ICL", "avx512icl", AV_CPU_FLAG_AVX512ICL },
313 #elif ARCH_LOONGARCH
314  { "LSX", "lsx", AV_CPU_FLAG_LSX },
315  { "LASX", "lasx", AV_CPU_FLAG_LASX },
316 #endif
317  { NULL }
318 };
319 
320 typedef struct CheckasmFuncVersion {
322  void *func;
323  int ok;
324  int cpu;
327 
328 /* Binary search tree node */
329 typedef struct CheckasmFunc {
330  struct CheckasmFunc *child[2];
332  uint8_t color; /* 0 = red, 1 = black */
333  char name[1];
334 } CheckasmFunc;
335 
336 /* Internal state */
337 static struct {
341  const char *current_test_name;
342  const char *bench_pattern;
346 
347  /* perf */
348  int nop_time;
349  int sysfd;
350 
351  int cpu_flag;
352  const char *cpu_flag_name;
353  const char *test_name;
354  int verbose;
355  volatile sig_atomic_t catch_signals;
356 } state;
357 
358 /* PRNG state */
360 
361 /* float compare support code */
362 static int is_negative(union av_intfloat32 u)
363 {
364  return u.i >> 31;
365 }
366 
367 int float_near_ulp(float a, float b, unsigned max_ulp)
368 {
369  union av_intfloat32 x, y;
370 
371  x.f = a;
372  y.f = b;
373 
374  if (is_negative(x) != is_negative(y)) {
375  // handle -0.0 == +0.0
376  return a == b;
377  }
378 
379  if (llabs((int64_t)x.i - y.i) <= max_ulp)
380  return 1;
381 
382  return 0;
383 }
384 
385 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
386  unsigned len)
387 {
388  unsigned i;
389 
390  for (i = 0; i < len; i++) {
391  if (!float_near_ulp(a[i], b[i], max_ulp))
392  return 0;
393  }
394  return 1;
395 }
396 
397 int float_near_abs_eps(float a, float b, float eps)
398 {
399  float abs_diff = fabsf(a - b);
400  if (abs_diff < eps)
401  return 1;
402 
403  fprintf(stderr, "test failed comparing %g with %g (abs diff=%g with EPS=%g)\n", a, b, abs_diff, eps);
404 
405  return 0;
406 }
407 
408 int float_near_abs_eps_array(const float *a, const float *b, float eps,
409  unsigned len)
410 {
411  unsigned i;
412 
413  for (i = 0; i < len; i++) {
414  if (!float_near_abs_eps(a[i], b[i], eps))
415  return 0;
416  }
417  return 1;
418 }
419 
420 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
421 {
422  return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
423 }
424 
425 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
426  unsigned max_ulp, unsigned len)
427 {
428  unsigned i;
429 
430  for (i = 0; i < len; i++) {
431  if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
432  return 0;
433  }
434  return 1;
435 }
436 
437 int double_near_abs_eps(double a, double b, double eps)
438 {
439  double abs_diff = fabs(a - b);
440 
441  return abs_diff < eps;
442 }
443 
444 int double_near_abs_eps_array(const double *a, const double *b, double eps,
445  unsigned len)
446 {
447  unsigned i;
448 
449  for (i = 0; i < len; i++) {
450  if (!double_near_abs_eps(a[i], b[i], eps))
451  return 0;
452  }
453  return 1;
454 }
455 
456 /* Print colored text to stderr if the terminal supports it */
457 static void color_printf(int color, const char *fmt, ...)
458 {
459  static int use_color = -1;
460  va_list arg;
461 
462 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
463  static HANDLE con;
464  static WORD org_attributes;
465 
466  if (use_color < 0) {
467  CONSOLE_SCREEN_BUFFER_INFO con_info;
468  con = GetStdHandle(STD_ERROR_HANDLE);
469  if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
470  org_attributes = con_info.wAttributes;
471  use_color = 1;
472  } else
473  use_color = 0;
474  }
475  if (use_color)
476  SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
477 #else
478  if (use_color < 0) {
479  const char *term = getenv("TERM");
480  use_color = term && strcmp(term, "dumb") && isatty(2);
481  }
482  if (use_color)
483  fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
484 #endif
485 
486  va_start(arg, fmt);
487  vfprintf(stderr, fmt, arg);
488  va_end(arg);
489 
490  if (use_color) {
491 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
492  SetConsoleTextAttribute(con, org_attributes);
493 #else
494  fprintf(stderr, "\x1b[0m");
495 #endif
496  }
497 }
498 
499 /* Deallocate a tree */
501 {
502  if (f) {
503  CheckasmFuncVersion *v = f->versions.next;
504  while (v) {
505  CheckasmFuncVersion *next = v->next;
506  free(v);
507  v = next;
508  }
509 
510  destroy_func_tree(f->child[0]);
511  destroy_func_tree(f->child[1]);
512  free(f);
513  }
514 }
515 
516 /* Allocate a zero-initialized block, clean up and exit on failure */
517 static void *checkasm_malloc(size_t size)
518 {
519  void *ptr = calloc(1, size);
520  if (!ptr) {
521  fprintf(stderr, "checkasm: malloc failed\n");
522  destroy_func_tree(state.funcs);
523  exit(1);
524  }
525  return ptr;
526 }
527 
528 /* Get the suffix of the specified cpu flag */
529 static const char *cpu_suffix(int cpu)
530 {
531  int i = FF_ARRAY_ELEMS(cpus);
532 
533  while (--i >= 0)
534  if (cpu & cpus[i].flag)
535  return cpus[i].suffix;
536 
537  return "c";
538 }
539 
540 static int cmp_nop(const void *a, const void *b)
541 {
542  return *(const uint16_t*)a - *(const uint16_t*)b;
543 }
544 
545 /* Measure the overhead of the timing code (in decicycles) */
546 static int measure_nop_time(void)
547 {
548  uint16_t nops[10000];
549  int i, nop_sum = 0;
550  av_unused const int sysfd = state.sysfd;
551 
552  uint64_t t = 0;
553  for (i = 0; i < 10000; i++) {
554  PERF_START(t);
555  PERF_STOP(t);
556  nops[i] = t;
557  }
558 
559  qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
560  for (i = 2500; i < 7500; i++)
561  nop_sum += nops[i];
562 
563  return nop_sum / 500;
564 }
565 
566 /* Print benchmark results */
568 {
569  if (f) {
570  print_benchs(f->child[0]);
571 
572  /* Only print functions with at least one assembly version */
573  if (f->versions.cpu || f->versions.next) {
574  CheckasmFuncVersion *v = &f->versions;
575  do {
576  CheckasmPerf *p = &v->perf;
577  if (p->iterations) {
578  int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
579  printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
580  }
581  } while ((v = v->next));
582  }
583 
584  print_benchs(f->child[1]);
585  }
586 }
587 
588 /* ASCIIbetical sort except preserving natural order for numbers */
589 static int cmp_func_names(const char *a, const char *b)
590 {
591  const char *start = a;
592  int ascii_diff, digit_diff;
593 
594  for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
595  for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
596 
597  if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
598  return digit_diff;
599 
600  return ascii_diff;
601 }
602 
603 /* Perform a tree rotation in the specified direction and return the new root */
605 {
606  CheckasmFunc *r = f->child[dir^1];
607  f->child[dir^1] = r->child[dir];
608  r->child[dir] = f;
609  r->color = f->color;
610  f->color = 0;
611  return r;
612 }
613 
614 #define is_red(f) ((f) && !(f)->color)
615 
616 /* Balance a left-leaning red-black tree at the specified node */
617 static void balance_tree(CheckasmFunc **root)
618 {
619  CheckasmFunc *f = *root;
620 
621  if (is_red(f->child[0]) && is_red(f->child[1])) {
622  f->color ^= 1;
623  f->child[0]->color = f->child[1]->color = 1;
624  }
625 
626  if (!is_red(f->child[0]) && is_red(f->child[1]))
627  *root = rotate_tree(f, 0); /* Rotate left */
628  else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
629  *root = rotate_tree(f, 1); /* Rotate right */
630 }
631 
632 /* Get a node with the specified name, creating it if it doesn't exist */
633 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
634 {
635  CheckasmFunc *f = *root;
636 
637  if (f) {
638  /* Search the tree for a matching node */
639  int cmp = cmp_func_names(name, f->name);
640  if (cmp) {
641  f = get_func(&f->child[cmp > 0], name);
642 
643  /* Rebalance the tree on the way up if a new node was inserted */
644  if (!f->versions.func)
645  balance_tree(root);
646  }
647  } else {
648  /* Allocate and insert a new node into the tree */
649  int name_length = strlen(name);
650  f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
651  memcpy(f->name, name, name_length + 1);
652  }
653 
654  return f;
655 }
656 
658 
659 /* Crash handling: attempt to catch crashes and handle them
660  * gracefully instead of just aborting abruptly. */
661 #ifdef _WIN32
662 #if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
663 static LONG NTAPI signal_handler(EXCEPTION_POINTERS *e) {
664  int s;
665 
666  if (!state.catch_signals)
667  return EXCEPTION_CONTINUE_SEARCH;
668 
669  switch (e->ExceptionRecord->ExceptionCode) {
670  case EXCEPTION_FLT_DIVIDE_BY_ZERO:
671  case EXCEPTION_INT_DIVIDE_BY_ZERO:
672  s = SIGFPE;
673  break;
674  case EXCEPTION_ILLEGAL_INSTRUCTION:
675  case EXCEPTION_PRIV_INSTRUCTION:
676  s = SIGILL;
677  break;
678  case EXCEPTION_ACCESS_VIOLATION:
679  case EXCEPTION_ARRAY_BOUNDS_EXCEEDED:
680  case EXCEPTION_DATATYPE_MISALIGNMENT:
681  case EXCEPTION_STACK_OVERFLOW:
682  s = SIGSEGV;
683  break;
684  case EXCEPTION_IN_PAGE_ERROR:
685  s = SIGBUS;
686  break;
687  default:
688  return EXCEPTION_CONTINUE_SEARCH;
689  }
690  state.catch_signals = 0;
692  return EXCEPTION_CONTINUE_EXECUTION; /* never reached, but shuts up gcc */
693 }
694 #endif
695 #else
696 static void signal_handler(int s);
697 
698 static const struct sigaction signal_handler_act = {
699  .sa_handler = signal_handler,
700  .sa_flags = SA_RESETHAND,
701 };
702 
703 static void signal_handler(int s) {
704  if (state.catch_signals) {
705  state.catch_signals = 0;
706  sigaction(s, &signal_handler_act, NULL);
708  }
709 }
710 #endif
711 
712 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
713 static void check_cpu_flag(const char *name, int flag)
714 {
715  int old_cpu_flag = state.cpu_flag;
716 
717  flag |= old_cpu_flag;
718  av_force_cpu_flags(-1);
719  state.cpu_flag = flag & av_get_cpu_flags();
720  av_force_cpu_flags(state.cpu_flag);
721 
722  if (!flag || state.cpu_flag != old_cpu_flag) {
723  int i;
724 
725  state.cpu_flag_name = name;
726  for (i = 0; tests[i].func; i++) {
727  if (state.test_name && strcmp(tests[i].name, state.test_name))
728  continue;
729  state.current_test_name = tests[i].name;
730  tests[i].func();
731  }
732  }
733 }
734 
735 /* Print the name of the current CPU flag, but only do it once */
736 static void print_cpu_name(void)
737 {
738  if (state.cpu_flag_name) {
739  color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
740  state.cpu_flag_name = NULL;
741  }
742 }
743 
744 #if CONFIG_LINUX_PERF
745 static int bench_init_linux(void)
746 {
747  struct perf_event_attr attr = {
748  .type = PERF_TYPE_HARDWARE,
749  .size = sizeof(struct perf_event_attr),
750  .config = PERF_COUNT_HW_CPU_CYCLES,
751  .disabled = 1, // start counting only on demand
752  .exclude_kernel = 1,
753  .exclude_hv = 1,
754 #if !ARCH_X86
755  .exclude_guest = 1,
756 #endif
757  };
758 
759  printf("benchmarking with Linux Perf Monitoring API\n");
760 
761  state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
762  if (state.sysfd == -1) {
763  perror("perf_event_open");
764  return -1;
765  }
766  return 0;
767 }
768 #elif CONFIG_MACOS_KPERF
769 static int bench_init_kperf(void)
770 {
771  ff_kperf_init();
772  return 0;
773 }
774 #else
775 static int bench_init_ffmpeg(void)
776 {
777 #ifdef AV_READ_TIME
778  if (!checkasm_save_context()) {
780  AV_READ_TIME();
782  } else {
783  fprintf(stderr, "checkasm: unable to execute platform specific timer\n");
784  return -1;
785  }
786  printf("benchmarking with native FFmpeg timers\n");
787  return 0;
788 #else
789  fprintf(stderr, "checkasm: --bench is not supported on your system\n");
790  return -1;
791 #endif
792 }
793 #endif
794 
795 static int bench_init(void)
796 {
797 #if CONFIG_LINUX_PERF
798  int ret = bench_init_linux();
799 #elif CONFIG_MACOS_KPERF
800  int ret = bench_init_kperf();
801 #else
802  int ret = bench_init_ffmpeg();
803 #endif
804  if (ret < 0)
805  return ret;
806 
807  state.nop_time = measure_nop_time();
808  printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
809  return 0;
810 }
811 
812 static void bench_uninit(void)
813 {
814 #if CONFIG_LINUX_PERF
815  if (state.sysfd > 0)
816  close(state.sysfd);
817 #endif
818 }
819 
820 static int usage(const char *path)
821 {
822  fprintf(stderr,
823  "Usage: %s [--bench] [--test=<pattern>] [--verbose] [seed]\n",
824  path);
825  return 1;
826 }
827 
828 int main(int argc, char *argv[])
829 {
830  unsigned int seed = av_get_random_seed();
831  int i, ret = 0;
832 
833 #ifdef _WIN32
834 #if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
835  AddVectoredExceptionHandler(0, signal_handler);
836 #endif
837 #else
838  sigaction(SIGBUS, &signal_handler_act, NULL);
839  sigaction(SIGFPE, &signal_handler_act, NULL);
840  sigaction(SIGILL, &signal_handler_act, NULL);
841  sigaction(SIGSEGV, &signal_handler_act, NULL);
842 #endif
843 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
845  checkasm_checked_call = checkasm_checked_call_vfp;
846 #endif
847 
848  if (!tests[0].func || !cpus[0].flag) {
849  fprintf(stderr, "checkasm: no tests to perform\n");
850  return 0;
851  }
852 
853  for (i = 1; i < argc; i++) {
854  const char *arg = argv[i];
855  unsigned long l;
856  char *end;
857 
858  if (!strncmp(arg, "--bench", 7)) {
859  if (bench_init() < 0)
860  return 1;
861  if (arg[7] == '=') {
862  state.bench_pattern = arg + 8;
863  state.bench_pattern_len = strlen(state.bench_pattern);
864  } else
865  state.bench_pattern = "";
866  } else if (!strncmp(arg, "--test=", 7)) {
867  state.test_name = arg + 7;
868  } else if (!strcmp(arg, "--verbose") || !strcmp(arg, "-v")) {
869  state.verbose = 1;
870  } else if ((l = strtoul(arg, &end, 10)) <= UINT_MAX &&
871  *end == '\0') {
872  seed = l;
873  } else {
874  return usage(argv[0]);
875  }
876  }
877 
878  fprintf(stderr, "checkasm: using random seed %u\n", seed);
880 
881  check_cpu_flag(NULL, 0);
882  for (i = 0; cpus[i].flag; i++)
884 
885  if (state.num_failed) {
886  fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
887  ret = 1;
888  } else {
889  fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
890  if (state.bench_pattern) {
891  print_benchs(state.funcs);
892  }
893  }
894 
895  destroy_func_tree(state.funcs);
896  bench_uninit();
897  return ret;
898 }
899 
900 /* Decide whether or not the specified function needs to be tested and
901  * allocate/initialize data structures if needed. Returns a pointer to a
902  * reference function if the function should be tested, otherwise NULL */
903 void *checkasm_check_func(void *func, const char *name, ...)
904 {
905  char name_buf[256];
906  void *ref = func;
908  int name_length;
909  va_list arg;
910 
911  va_start(arg, name);
912  name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
913  va_end(arg);
914 
915  if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
916  return NULL;
917 
918  state.current_func = get_func(&state.funcs, name_buf);
919  state.funcs->color = 1;
920  v = &state.current_func->versions;
921 
922  if (v->func) {
923  CheckasmFuncVersion *prev;
924  do {
925  /* Only test functions that haven't already been tested */
926  if (v->func == func)
927  return NULL;
928 
929  if (v->ok)
930  ref = v->func;
931 
932  prev = v;
933  } while ((v = v->next));
934 
935  v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
936  }
937 
938  v->func = func;
939  v->ok = 1;
940  v->cpu = state.cpu_flag;
941  state.current_func_ver = v;
942 
943  if (state.cpu_flag)
944  state.num_checked++;
945 
946  return ref;
947 }
948 
949 /* Decide whether or not the current function needs to be benchmarked */
951 {
952  return !state.num_failed && state.bench_pattern &&
953  !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
954 }
955 
956 /* Indicate that the current test has failed */
957 void checkasm_fail_func(const char *msg, ...)
958 {
959  if (state.current_func_ver && state.current_func_ver->cpu &&
960  state.current_func_ver->ok)
961  {
962  va_list arg;
963 
964  print_cpu_name();
965  fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
966  va_start(arg, msg);
967  vfprintf(stderr, msg, arg);
968  va_end(arg);
969  fprintf(stderr, ")\n");
970 
971  state.current_func_ver->ok = 0;
972  state.num_failed++;
973  }
974 }
975 
977  state.catch_signals = enabled;
978 }
979 
981  if (s) {
982 #ifdef __GLIBC__
983  checkasm_fail_func("fatal signal %d: %s", s, strsignal(s));
984 #else
985  checkasm_fail_func(s == SIGFPE ? "fatal arithmetic error" :
986  s == SIGILL ? "illegal instruction" :
987  s == SIGBUS ? "bus error" :
988  "segmentation fault");
989 #endif
990  }
991  return s;
992 }
993 
994 /* Get the benchmark context of the current function */
996 {
997  CheckasmPerf *perf = &state.current_func_ver->perf;
998  memset(perf, 0, sizeof(*perf));
999  perf->sysfd = state.sysfd;
1000  return perf;
1001 }
1002 
1003 /* Print the outcome of all tests performed since the last time this function was called */
1004 void checkasm_report(const char *name, ...)
1005 {
1006  static int prev_checked, prev_failed, max_length;
1007 
1008  if (state.num_checked > prev_checked) {
1009  int pad_length = max_length + 4;
1010  va_list arg;
1011 
1012  print_cpu_name();
1013  pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
1014  va_start(arg, name);
1015  pad_length -= vfprintf(stderr, name, arg);
1016  va_end(arg);
1017  fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
1018 
1019  if (state.num_failed == prev_failed)
1020  color_printf(COLOR_GREEN, "OK");
1021  else
1022  color_printf(COLOR_RED, "FAILED");
1023  fprintf(stderr, "]\n");
1024 
1025  prev_checked = state.num_checked;
1026  prev_failed = state.num_failed;
1027  } else if (!state.cpu_flag) {
1028  /* Calculate the amount of padding required to make the output vertically aligned */
1029  int length = strlen(state.current_test_name);
1030  va_list arg;
1031 
1032  va_start(arg, name);
1033  length += vsnprintf(NULL, 0, name, arg);
1034  va_end(arg);
1035 
1036  if (length > max_length)
1037  max_length = length;
1038  }
1039 }
1040 
1041 #define DEF_CHECKASM_CHECK_FUNC(type, fmt) \
1042 int checkasm_check_##type(const char *file, int line, \
1043  const type *buf1, ptrdiff_t stride1, \
1044  const type *buf2, ptrdiff_t stride2, \
1045  int w, int h, const char *name) \
1046 { \
1047  int y = 0; \
1048  stride1 /= sizeof(*buf1); \
1049  stride2 /= sizeof(*buf2); \
1050  for (y = 0; y < h; y++) \
1051  if (memcmp(&buf1[y*stride1], &buf2[y*stride2], w*sizeof(*buf1))) \
1052  break; \
1053  if (y == h) \
1054  return 0; \
1055  checkasm_fail_func("%s:%d", file, line); \
1056  if (!state.verbose) \
1057  return 1; \
1058  fprintf(stderr, "%s:\n", name); \
1059  while (h--) { \
1060  for (int x = 0; x < w; x++) \
1061  fprintf(stderr, " " fmt, buf1[x]); \
1062  fprintf(stderr, " "); \
1063  for (int x = 0; x < w; x++) \
1064  fprintf(stderr, " " fmt, buf2[x]); \
1065  fprintf(stderr, " "); \
1066  for (int x = 0; x < w; x++) \
1067  fprintf(stderr, "%c", buf1[x] != buf2[x] ? 'x' : '.'); \
1068  buf1 += stride1; \
1069  buf2 += stride2; \
1070  fprintf(stderr, "\n"); \
1071  } \
1072  return 1; \
1073 }
1074 
1075 DEF_CHECKASM_CHECK_FUNC(uint8_t, "%02x")
1076 DEF_CHECKASM_CHECK_FUNC(uint16_t, "%04x")
1077 DEF_CHECKASM_CHECK_FUNC(uint32_t, "%08x")
1078 DEF_CHECKASM_CHECK_FUNC(int16_t, "%6d")
checkasm_check_nlmeans
void checkasm_check_nlmeans(void)
Definition: vf_nlmeans.c:32
AV_CPU_FLAG_VFP
#define AV_CPU_FLAG_VFP
Definition: cpu.h:67
checkasm_check_sw_gbrp
void checkasm_check_sw_gbrp(void)
Definition: sw_gbrp.c:401
bench_pattern_len
int bench_pattern_len
Definition: checkasm.c:343
av_force_cpu_flags
void av_force_cpu_flags(int arg)
Disables cpu detection and forces the specified flags.
Definition: cpu.c:75
checkasm_check_vp8dsp
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
checkasm_check_blockdsp
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:78
AV_CPU_FLAG_SSE3
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
Definition: cpu.h:38
checkasm_check_fdctdsp
void checkasm_check_fdctdsp(void)
Definition: fdctdsp.c:67
destroy_func_tree
static void destroy_func_tree(CheckasmFunc *f)
Definition: checkasm.c:500
r
const char * r
Definition: vf_curves.c:127
checkasm_check_videodsp
void checkasm_check_videodsp(void)
Definition: videodsp.c:81
cmp_nop
static int cmp_nop(const void *a, const void *b)
Definition: checkasm.c:540
checkasm_check_vf_eq
void checkasm_check_vf_eq(void)
Definition: vf_eq.c:76
AV_CPU_FLAG_RVB_BASIC
#define AV_CPU_FLAG_RVB_BASIC
Basic bit-manipulations.
Definition: cpu.h:91
COLOR_RED
#define COLOR_RED
Definition: checkasm.c:56
checkasm_lfg
AVLFG checkasm_lfg
Definition: checkasm.c:359
color
Definition: vf_paletteuse.c:512
av_lfg_init
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:32
u
#define u(width, name, range_min, range_max)
Definition: cbs_h2645.c:251
checkasm_check_alacdsp
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:116
AV_CPU_FLAG_LSX
#define AV_CPU_FLAG_LSX
Definition: cpu.h:80
checkasm_check_v210dec
void checkasm_check_v210dec(void)
Definition: v210dec.c:47
float_near_abs_eps_array
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:408
AV_CPU_FLAG_SSE3SLOW
#define AV_CPU_FLAG_SSE3SLOW
SSE3 supported, but usually not faster.
Definition: cpu.h:39
CheckasmFunc
Definition: checkasm.c:329
name
const char * name
Definition: checkasm.c:77
bench_pattern
const char * bench_pattern
Definition: checkasm.c:342
AV_CPU_FLAG_RV_ZVBB
#define AV_CPU_FLAG_RV_ZVBB
Vector basic bit-manipulations.
Definition: cpu.h:93
av_unused
#define av_unused
Definition: attributes.h:131
checkasm_check_hevc_idct
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:108
checkasm_check_sw_scale
void checkasm_check_sw_scale(void)
Definition: sw_scale.c:362
CheckasmFunc::child
struct CheckasmFunc * child[2]
Definition: checkasm.c:330
checkasm_check_afir
void checkasm_check_afir(void)
Definition: af_afir.c:144
print_cpu_name
static void print_cpu_name(void)
Definition: checkasm.c:736
AV_CPU_FLAG_3DNOW
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
Definition: cpu.h:32
AV_CPU_FLAG_RVF
#define AV_CPU_FLAG_RVF
F (single precision FP)
Definition: cpu.h:85
av_intfloat32::i
uint32_t i
Definition: intfloat.h:28
float_near_abs_eps
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:397
b
#define b
Definition: input.c:41
checkasm_check_aacpsdsp
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:234
bench_uninit
static void bench_uninit(void)
Definition: checkasm.c:812
CheckasmFuncVersion::cpu
int cpu
Definition: checkasm.c:324
cpu_flag
int cpu_flag
Definition: checkasm.c:351
AV_CPU_FLAG_DOTPROD
#define AV_CPU_FLAG_DOTPROD
Definition: cpu.h:72
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:103
cpu_suffix
static const char * cpu_suffix(int cpu)
Definition: checkasm.c:529
CheckasmPerf::sysfd
int sysfd
Definition: checkasm.h:305
checkasm_check_llviddspenc
void checkasm_check_llviddspenc(void)
Definition: llviddspenc.c:104
intfloat.h
checkasm_check_h264dsp
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:443
tf_sess_config.config
config
Definition: tf_sess_config.py:33
color_printf
static void color_printf(int color, const char *fmt,...)
Definition: checkasm.c:457
AV_CPU_FLAG_RVV_F64
#define AV_CPU_FLAG_RVV_F64
Vectors of double's.
Definition: cpu.h:90
av_get_random_seed
uint32_t av_get_random_seed(void)
Get a seed to use in conjunction with random functions.
Definition: random_seed.c:167
main
int main(int argc, char *argv[])
Definition: checkasm.c:828
checkasm_check_motion
void checkasm_check_motion(void)
Definition: motion.c:146
AV_CPU_FLAG_SSSE3
#define AV_CPU_FLAG_SSSE3
Conroe SSSE3 functions.
Definition: cpu.h:41
double_near_abs_eps
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:437
CheckasmPerf
Definition: checkasm.h:304
current_test_name
const char * current_test_name
Definition: checkasm.c:341
PERF_START
#define PERF_START(t)
Definition: checkasm.h:362
rotate_tree
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
Definition: checkasm.c:604
dummy
int dummy
Definition: motion.c:66
AV_CPU_FLAG_XOP
#define AV_CPU_FLAG_XOP
Bulldozer XOP functions.
Definition: cpu.h:49
checkasm_check_vp9dsp
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:627
checkasm.h
checkasm_check_vf_threshold
void checkasm_check_vf_threshold(void)
Definition: vf_threshold.c:79
checkasm_check_vf_gblur
void checkasm_check_vf_gblur(void)
Definition: vf_gblur.c:82
AV_CPU_FLAG_3DNOWEXT
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
Definition: cpu.h:37
float_near_abs_eps_ulp
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:420
checkasm_check_vvc_mc
void checkasm_check_vvc_mc(void)
Definition: vvc_mc.c:325
AV_CPU_FLAG_VSX
#define AV_CPU_FLAG_VSX
ISA 2.06.
Definition: cpu.h:61
fabsf
static __device__ float fabsf(float a)
Definition: cuda_runtime.h:181
AV_CPU_FLAG_RVV_F32
#define AV_CPU_FLAG_RVV_F32
Vectors of float's *‍/.
Definition: cpu.h:88
num_failed
int num_failed
Definition: checkasm.c:345
checkasm_check_float_dsp
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:281
AV_CPU_FLAG_AVX512
#define AV_CPU_FLAG_AVX512
AVX-512 functions: requires OS support even if YMM/ZMM registers aren't used.
Definition: cpu.h:56
bench_init
static int bench_init(void)
Definition: checkasm.c:795
checkasm_check_pixblockdsp
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:82
FF_ARRAY_ELEMS
#define FF_ARRAY_ELEMS(a)
Definition: sinewin_tablegen.c:29
CheckasmFuncVersion
Definition: checkasm.c:320
COLOR_YELLOW
#define COLOR_YELLOW
Definition: checkasm.c:58
nop_time
int nop_time
Definition: checkasm.c:348
CheckasmFuncVersion::perf
CheckasmPerf perf
Definition: checkasm.c:325
s
#define s(width, name)
Definition: cbs_vp9.c:198
checkasm_check_vf_bwdif
void checkasm_check_vf_bwdif(void)
Definition: vf_bwdif.c:72
checkasm_report
void checkasm_report(const char *name,...)
Definition: checkasm.c:1004
get_func
static CheckasmFunc * get_func(CheckasmFunc **root, const char *name)
Definition: checkasm.c:633
checkasm_fail_func
void checkasm_fail_func(const char *msg,...)
Definition: checkasm.c:957
checkasm_check_sw_rgb
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:182
checkasm_check_hevc_sao
void checkasm_check_hevc_sao(void)
Definition: hevc_sao.c:141
num_checked
int num_checked
Definition: checkasm.c:344
AV_CPU_FLAG_ARMV6
#define AV_CPU_FLAG_ARMV6
Definition: cpu.h:65
AV_READ_TIME
#define AV_READ_TIME
Definition: timer.h:28
AV_CPU_FLAG_SSE4
#define AV_CPU_FLAG_SSE4
Penryn SSE4.1 functions.
Definition: cpu.h:44
AV_CPU_FLAG_RVB_ADDR
#define AV_CPU_FLAG_RVB_ADDR
Address bit-manipulations.
Definition: cpu.h:92
checkasm_check_hevc_pel
void checkasm_check_hevc_pel(void)
Definition: hevc_pel.c:607
checkasm_set_signal_handler_state
void checkasm_set_signal_handler_state(int enabled)
Definition: checkasm.c:976
cmp
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:262
CheckasmPerf::iterations
int iterations
Definition: checkasm.h:307
CheckasmFuncVersion::ok
int ok
Definition: checkasm.c:323
arg
const char * arg
Definition: jacosubdec.c:67
func
void(* func)(void)
Definition: checkasm.c:78
AV_CPU_FLAG_RVD
#define AV_CPU_FLAG_RVD
D (double precision FP)
Definition: cpu.h:86
fabs
static __device__ float fabs(float a)
Definition: cuda_runtime.h:182
NULL
#define NULL
Definition: coverity.c:32
measure_nop_time
static int measure_nop_time(void)
Definition: checkasm.c:546
checkasm_check_takdsp
void checkasm_check_takdsp(void)
Definition: takdsp.c:145
double_near_abs_eps_array
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:444
state
static struct @414 state
AV_CPU_FLAG_LASX
#define AV_CPU_FLAG_LASX
Definition: cpu.h:81
cpus
static const struct @413 cpus[]
AV_CPU_FLAG_AVX512ICL
#define AV_CPU_FLAG_AVX512ICL
F/CD/BW/DQ/VL/VNNI/IFMA/VBMI/VBMI2/VPOPCNTDQ/BITALG/GFNI/VAES/VPCLMULQDQ.
Definition: cpu.h:57
checkasm_check_fmtconvert
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:46
checkasm_load_context
#define checkasm_load_context(s)
Definition: checkasm.h:73
AV_CPU_FLAG_MSA
#define AV_CPU_FLAG_MSA
Definition: cpu.h:77
checkasm_save_context
#define checkasm_save_context()
Definition: checkasm.h:72
COLOR_GREEN
#define COLOR_GREEN
Definition: checkasm.c:57
checkasm_check_aacencdsp
void checkasm_check_aacencdsp(void)
Definition: aacencdsp.c:64
checkasm_check_vorbisdsp
void checkasm_check_vorbisdsp(void)
Definition: vorbisdsp.c:76
cpu.h
av_intfloat32
Definition: intfloat.h:27
AV_CPU_FLAG_CMOV
#define AV_CPU_FLAG_CMOV
supports cmov instruction
Definition: cpu.h:51
seed
static unsigned int seed
Definition: videogen.c:78
AV_CPU_FLAG_ALTIVEC
#define AV_CPU_FLAG_ALTIVEC
standard
Definition: cpu.h:60
checkasm_check_rv40dsp
void checkasm_check_rv40dsp(void)
Definition: rv40dsp.c:71
print_benchs
static void print_benchs(CheckasmFunc *f)
Definition: checkasm.c:567
current_func
CheckasmFunc * current_func
Definition: checkasm.c:339
AV_CPU_FLAG_I8MM
#define AV_CPU_FLAG_I8MM
Definition: cpu.h:73
checkasm_check_vc1dsp
void checkasm_check_vc1dsp(void)
Definition: vc1dsp.c:476
catch_signals
volatile sig_atomic_t catch_signals
Definition: checkasm.c:355
checkasm_check_h264chroma
void checkasm_check_h264chroma(void)
Definition: h264chroma.c:81
AV_CPU_FLAG_SSE2
#define AV_CPU_FLAG_SSE2
PIV SSE2 functions.
Definition: cpu.h:34
CheckasmFunc::color
uint8_t color
Definition: checkasm.c:332
AVLFG
Context structure for the Lagged Fibonacci PRNG.
Definition: lfg.h:33
f
f
Definition: af_crystalizer.c:121
checkasm_check_llauddsp
void checkasm_check_llauddsp(void)
Definition: llauddsp.c:108
AV_CPU_FLAG_AVX
#define AV_CPU_FLAG_AVX
AVX functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:47
AV_CPU_FLAG_FMA4
#define AV_CPU_FLAG_FMA4
Bulldozer FMA4 functions.
Definition: cpu.h:50
cpu.h
AV_CPU_FLAG_AVX2
#define AV_CPU_FLAG_AVX2
AVX2 functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:52
checkasm_check_llviddsp
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:195
checkasm_handle_signal
int checkasm_handle_signal(int s)
Definition: checkasm.c:980
AV_CPU_FLAG_SSE2SLOW
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
Definition: cpu.h:35
AV_CPU_FLAG_NEON
#define AV_CPU_FLAG_NEON
Definition: cpu.h:69
size
int size
Definition: twinvq_data.h:10344
DEF_CHECKASM_CHECK_FUNC
#define DEF_CHECKASM_CHECK_FUNC(type, fmt)
Definition: checkasm.c:1041
printf
printf("static const uint8_t my_array[100] = {\n")
sysfd
int sysfd
Definition: checkasm.c:349
have_neon
#define have_neon(flags)
Definition: cpu.h:26
CheckasmFuncVersion::func
void * func
Definition: checkasm.c:322
checkasm_check_h264pred
void checkasm_check_h264pred(void)
Definition: h264pred.c:233
av_isdigit
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
Definition: avstring.h:202
AV_CPU_FLAG_FMA3
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
Definition: cpu.h:53
a
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
Definition: undefined.txt:41
ff_kperf_init
void ff_kperf_init(void)
Definition: macos_kperf.c:99
checkasm_check_huffyuvdsp
void checkasm_check_huffyuvdsp(void)
Definition: huffyuvdsp.c:95
float_near_abs_eps_array_ulp
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:425
test_name
const char * test_name
Definition: checkasm.c:353
CheckasmPerf::cycles
uint64_t cycles
Definition: checkasm.h:306
checkasm_get_perf_context
CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:995
checkasm_check_fixed_dsp
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:133
checkasm_check_colorspace
void checkasm_check_colorspace(void)
Definition: vf_colorspace.c:309
checkasm_check_utvideodsp
void checkasm_check_utvideodsp(void)
Definition: utvideodsp.c:90
is_negative
static int is_negative(union av_intfloat32 u)
Definition: checkasm.c:362
checkasm_check_audiodsp
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:52
AV_CPU_FLAG_RV_MISALIGNED
#define AV_CPU_FLAG_RV_MISALIGNED
Fast misaligned accesses.
Definition: cpu.h:94
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:256
checkasm_check_flacdsp
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:126
checkasm_check_func
void * checkasm_check_func(void *func, const char *name,...)
Definition: checkasm.c:903
AV_CPU_FLAG_SSE42
#define AV_CPU_FLAG_SSE42
Nehalem SSE4.2 functions.
Definition: cpu.h:45
checkasm_check_idctdsp
void checkasm_check_idctdsp(void)
Definition: idctdsp.c:95
CheckasmFuncVersion::next
struct CheckasmFuncVersion * next
Definition: checkasm.c:321
vsnprintf
#define vsnprintf
Definition: snprintf.h:36
common.h
use_color
static int use_color
Definition: log.c:123
AV_CPU_FLAG_RVV_I32
#define AV_CPU_FLAG_RVV_I32
Vectors of 8/16/32-bit int's *‍/.
Definition: cpu.h:87
checkasm_check_synth_filter
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:47
cmp_func_names
static int cmp_func_names(const char *a, const char *b)
Definition: checkasm.c:589
AV_CPU_FLAG_ARMV8
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:70
checkasm_check_lpc
void checkasm_check_lpc(void)
Definition: lpc.c:60
len
int len
Definition: vorbis_enc_data.h:426
checkasm_check_g722dsp
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
checkasm_check_ac3dsp
void checkasm_check_ac3dsp(void)
Definition: ac3dsp.c:193
PERF_STOP
#define PERF_STOP(t)
Definition: checkasm.h:363
is_red
#define is_red(f)
Definition: checkasm.c:614
AV_CPU_FLAG_ATOM
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
Definition: cpu.h:43
have_vfp
#define have_vfp(flags)
Definition: cpu.h:27
checkasm_check_svq1enc
void checkasm_check_svq1enc(void)
Definition: svq1enc.c:63
checkasm_check_vf_hflip
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:72
ret
ret
Definition: filter_design.txt:187
checkasm_check_blend
void checkasm_check_blend(void)
Definition: vf_blend.c:89
verbose
int verbose
Definition: checkasm.c:354
AV_CPU_FLAG_VFPV3
#define AV_CPU_FLAG_VFPV3
Definition: cpu.h:68
checkasm_check_bswapdsp
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:60
checkasm_context_buf
checkasm_context checkasm_context_buf
Definition: checkasm.c:657
AV_CPU_FLAG_ARMV5TE
#define AV_CPU_FLAG_ARMV5TE
Definition: cpu.h:64
AV_CPU_FLAG_MMX
#define AV_CPU_FLAG_MMX
standard MMX
Definition: cpu.h:29
suffix
const char * suffix
Definition: checkasm.c:259
checkasm_malloc
static void * checkasm_malloc(size_t size)
Definition: checkasm.c:517
AV_CPU_FLAG_RVI
#define AV_CPU_FLAG_RVI
I (full GPR bank)
Definition: cpu.h:84
checkasm_check_h264qpel
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:51
random_seed.h
signal_handler
static void signal_handler(int s)
Definition: checkasm.c:703
AV_CPU_FLAG_AESNI
#define AV_CPU_FLAG_AESNI
Advanced Encryption Standard functions.
Definition: cpu.h:46
checkasm_bench_func
int checkasm_bench_func(void)
Definition: checkasm.c:950
AV_CPU_FLAG_POWER8
#define AV_CPU_FLAG_POWER8
ISA 2.07.
Definition: cpu.h:62
cpu_flag_name
const char * cpu_flag_name
Definition: checkasm.c:352
checkasm_check_vf_sobel
void checkasm_check_vf_sobel(void)
Definition: vf_convolution.c:100
AV_CPU_FLAG_SSE
#define AV_CPU_FLAG_SSE
SSE functions.
Definition: cpu.h:33
ref
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:112
float_near_ulp_array
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:385
signal_handler_act
static const struct sigaction signal_handler_act
Definition: checkasm.c:698
AV_CPU_FLAG_MMXEXT
#define AV_CPU_FLAG_MMXEXT
SSE integer functions or AMD MMX ext.
Definition: cpu.h:30
checkasm_check_av_tx
void checkasm_check_av_tx(void)
Definition: av_tx.c:102
av_intfloat32::f
float f
Definition: intfloat.h:29
flag
int flag
Definition: checkasm.c:260
checkasm_check_v210enc
void checkasm_check_v210enc(void)
Definition: v210enc.c:84
checkasm_check_sbrdsp
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:257
checkasm_check_hevc_deblock
void checkasm_check_hevc_deblock(void)
Definition: hevc_deblock.c:262
usage
static int usage(const char *path)
Definition: checkasm.c:820
AV_CPU_FLAG_VFP_VM
#define AV_CPU_FLAG_VFP_VM
VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations.
Definition: cpu.h:71
AV_CPU_FLAG_MMI
#define AV_CPU_FLAG_MMI
Definition: cpu.h:76
checkasm_check_jpeg2000dsp
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:92
d
d
Definition: ffmpeg_filter.c:424
checkasm_check_hevc_add_res
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:87
int32_t
int32_t
Definition: audioconvert.c:56
float_near_ulp
int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:367
current_func_ver
CheckasmFuncVersion * current_func_ver
Definition: checkasm.c:340
check_cpu_flag
static void check_cpu_flag(const char *name, int flag)
Definition: checkasm.c:713
checkasm_context
sigjmp_buf checkasm_context
Definition: checkasm.h:71
bench_init_ffmpeg
static int bench_init_ffmpeg(void)
Definition: checkasm.c:775
checkasm_check_rv34dsp
void checkasm_check_rv34dsp(void)
Definition: rv34dsp.c:82
CheckasmFunc::versions
CheckasmFuncVersion versions
Definition: checkasm.c:331
isatty
#define isatty(fd)
Definition: checkasm.c:66
tests
static const struct @412 tests[]
AV_CPU_FLAG_RVV_I64
#define AV_CPU_FLAG_RVV_I64
Vectors of 64-bit int's *‍/.
Definition: cpu.h:89
CheckasmFunc::name
char name[1]
Definition: checkasm.c:333
checkasm_check_vvc_alf
void checkasm_check_vvc_alf(void)
Definition: vvc_alf.c:165
checkasm_check_exrdsp
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:77
balance_tree
static void balance_tree(CheckasmFunc **root)
Definition: checkasm.c:617
AV_CPU_FLAG_ARMV6T2
#define AV_CPU_FLAG_ARMV6T2
Definition: cpu.h:66
funcs
CheckasmFunc * funcs
Definition: checkasm.c:338
checkasm_check_opusdsp
void checkasm_check_opusdsp(void)
Definition: opusdsp.c:86