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 
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_DECODER
140  { "v210dec", checkasm_check_v210dec },
141  #endif
142  #if CONFIG_V210_ENCODER
143  { "v210enc", checkasm_check_v210enc },
144  #endif
145  #if CONFIG_VP8DSP
146  { "vp8dsp", checkasm_check_vp8dsp },
147  #endif
148  #if CONFIG_VP9_DECODER
149  { "vp9dsp", checkasm_check_vp9dsp },
150  #endif
151  #if CONFIG_VIDEODSP
152  { "videodsp", checkasm_check_videodsp },
153  #endif
154 #endif
155 #if CONFIG_AVFILTER
156  #if CONFIG_AFIR_FILTER
157  { "af_afir", checkasm_check_afir },
158  #endif
159  #if CONFIG_BLEND_FILTER
160  { "vf_blend", checkasm_check_blend },
161  #endif
162  #if CONFIG_COLORSPACE_FILTER
163  { "vf_colorspace", checkasm_check_colorspace },
164  #endif
165  #if CONFIG_GBLUR_FILTER
166  { "vf_gblur", checkasm_check_vf_gblur },
167  #endif
168  #if CONFIG_HFLIP_FILTER
169  { "vf_hflip", checkasm_check_vf_hflip },
170  #endif
171  #if CONFIG_NLMEANS_FILTER
172  { "vf_nlmeans", checkasm_check_nlmeans },
173  #endif
174  #if CONFIG_THRESHOLD_FILTER
175  { "vf_threshold", checkasm_check_vf_threshold },
176  #endif
177 #endif
178 #if CONFIG_SWSCALE
179  { "sw_rgb", checkasm_check_sw_rgb },
180 #endif
181 #if CONFIG_AVUTIL
182  { "fixed_dsp", checkasm_check_fixed_dsp },
183  { "float_dsp", checkasm_check_float_dsp },
184 #endif
185  { NULL }
186 };
187 
188 /* List of cpu flags to check */
189 static const struct {
190  const char *name;
191  const char *suffix;
192  int flag;
193 } cpus[] = {
194 #if ARCH_AARCH64
195  { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
196  { "NEON", "neon", AV_CPU_FLAG_NEON },
197 #elif ARCH_ARM
198  { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
199  { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
200  { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
201  { "VFP", "vfp", AV_CPU_FLAG_VFP },
202  { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
203  { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
204  { "NEON", "neon", AV_CPU_FLAG_NEON },
205 #elif ARCH_PPC
206  { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
207  { "VSX", "vsx", AV_CPU_FLAG_VSX },
208  { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
209 #elif ARCH_X86
210  { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
211  { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
212  { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
213  { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
214  { "SSE", "sse", AV_CPU_FLAG_SSE },
215  { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
216  { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
217  { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
218  { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
219  { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
220  { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
221  { "AVX", "avx", AV_CPU_FLAG_AVX },
222  { "XOP", "xop", AV_CPU_FLAG_XOP },
223  { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
224  { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
225  { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
226  { "AVX-512", "avx512", AV_CPU_FLAG_AVX512 },
227 #endif
228  { NULL }
229 };
230 
231 typedef struct CheckasmFuncVersion {
233  void *func;
234  int ok;
235  int cpu;
238 
239 /* Binary search tree node */
240 typedef struct CheckasmFunc {
241  struct CheckasmFunc *child[2];
243  uint8_t color; /* 0 = red, 1 = black */
244  char name[1];
245 } CheckasmFunc;
246 
247 /* Internal state */
248 static struct {
252  const char *current_test_name;
253  const char *bench_pattern;
257 
258  /* perf */
259  int nop_time;
260  int sysfd;
261 
262  int cpu_flag;
263  const char *cpu_flag_name;
264  const char *test_name;
265 } state;
266 
267 /* PRNG state */
269 
270 /* float compare support code */
271 static int is_negative(union av_intfloat32 u)
272 {
273  return u.i >> 31;
274 }
275 
276 int float_near_ulp(float a, float b, unsigned max_ulp)
277 {
278  union av_intfloat32 x, y;
279 
280  x.f = a;
281  y.f = b;
282 
283  if (is_negative(x) != is_negative(y)) {
284  // handle -0.0 == +0.0
285  return a == b;
286  }
287 
288  if (llabs((int64_t)x.i - y.i) <= max_ulp)
289  return 1;
290 
291  return 0;
292 }
293 
294 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
295  unsigned len)
296 {
297  unsigned i;
298 
299  for (i = 0; i < len; i++) {
300  if (!float_near_ulp(a[i], b[i], max_ulp))
301  return 0;
302  }
303  return 1;
304 }
305 
306 int float_near_abs_eps(float a, float b, float eps)
307 {
308  float abs_diff = fabsf(a - b);
309  if (abs_diff < eps)
310  return 1;
311 
312  fprintf(stderr, "test failed comparing %g with %g (abs diff=%g with EPS=%g)\n", a, b, abs_diff, eps);
313 
314  return 0;
315 }
316 
317 int float_near_abs_eps_array(const float *a, const float *b, float eps,
318  unsigned len)
319 {
320  unsigned i;
321 
322  for (i = 0; i < len; i++) {
323  if (!float_near_abs_eps(a[i], b[i], eps))
324  return 0;
325  }
326  return 1;
327 }
328 
329 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
330 {
331  return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
332 }
333 
334 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
335  unsigned max_ulp, unsigned len)
336 {
337  unsigned i;
338 
339  for (i = 0; i < len; i++) {
340  if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
341  return 0;
342  }
343  return 1;
344 }
345 
346 int double_near_abs_eps(double a, double b, double eps)
347 {
348  double abs_diff = fabs(a - b);
349 
350  return abs_diff < eps;
351 }
352 
353 int double_near_abs_eps_array(const double *a, const double *b, double eps,
354  unsigned len)
355 {
356  unsigned i;
357 
358  for (i = 0; i < len; i++) {
359  if (!double_near_abs_eps(a[i], b[i], eps))
360  return 0;
361  }
362  return 1;
363 }
364 
365 /* Print colored text to stderr if the terminal supports it */
366 static void color_printf(int color, const char *fmt, ...)
367 {
368  static int use_color = -1;
369  va_list arg;
370 
371 #if HAVE_SETCONSOLETEXTATTRIBUTE
372  static HANDLE con;
373  static WORD org_attributes;
374 
375  if (use_color < 0) {
376  CONSOLE_SCREEN_BUFFER_INFO con_info;
377  con = GetStdHandle(STD_ERROR_HANDLE);
378  if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
379  org_attributes = con_info.wAttributes;
380  use_color = 1;
381  } else
382  use_color = 0;
383  }
384  if (use_color)
385  SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
386 #else
387  if (use_color < 0) {
388  const char *term = getenv("TERM");
389  use_color = term && strcmp(term, "dumb") && isatty(2);
390  }
391  if (use_color)
392  fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
393 #endif
394 
395  va_start(arg, fmt);
396  vfprintf(stderr, fmt, arg);
397  va_end(arg);
398 
399  if (use_color) {
400 #if HAVE_SETCONSOLETEXTATTRIBUTE
401  SetConsoleTextAttribute(con, org_attributes);
402 #else
403  fprintf(stderr, "\x1b[0m");
404 #endif
405  }
406 }
407 
408 /* Deallocate a tree */
410 {
411  if (f) {
412  CheckasmFuncVersion *v = f->versions.next;
413  while (v) {
414  CheckasmFuncVersion *next = v->next;
415  free(v);
416  v = next;
417  }
418 
419  destroy_func_tree(f->child[0]);
420  destroy_func_tree(f->child[1]);
421  free(f);
422  }
423 }
424 
425 /* Allocate a zero-initialized block, clean up and exit on failure */
426 static void *checkasm_malloc(size_t size)
427 {
428  void *ptr = calloc(1, size);
429  if (!ptr) {
430  fprintf(stderr, "checkasm: malloc failed\n");
431  destroy_func_tree(state.funcs);
432  exit(1);
433  }
434  return ptr;
435 }
436 
437 /* Get the suffix of the specified cpu flag */
438 static const char *cpu_suffix(int cpu)
439 {
440  int i = FF_ARRAY_ELEMS(cpus);
441 
442  while (--i >= 0)
443  if (cpu & cpus[i].flag)
444  return cpus[i].suffix;
445 
446  return "c";
447 }
448 
449 static int cmp_nop(const void *a, const void *b)
450 {
451  return *(const uint16_t*)a - *(const uint16_t*)b;
452 }
453 
454 /* Measure the overhead of the timing code (in decicycles) */
455 static int measure_nop_time(void)
456 {
457  uint16_t nops[10000];
458  int i, nop_sum = 0;
459  av_unused const int sysfd = state.sysfd;
460 
461  uint64_t t = 0;
462  for (i = 0; i < 10000; i++) {
463  PERF_START(t);
464  PERF_STOP(t);
465  nops[i] = t;
466  }
467 
468  qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
469  for (i = 2500; i < 7500; i++)
470  nop_sum += nops[i];
471 
472  return nop_sum / 500;
473 }
474 
475 /* Print benchmark results */
477 {
478  if (f) {
479  print_benchs(f->child[0]);
480 
481  /* Only print functions with at least one assembly version */
482  if (f->versions.cpu || f->versions.next) {
483  CheckasmFuncVersion *v = &f->versions;
484  do {
485  CheckasmPerf *p = &v->perf;
486  if (p->iterations) {
487  int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
488  printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
489  }
490  } while ((v = v->next));
491  }
492 
493  print_benchs(f->child[1]);
494  }
495 }
496 
497 /* ASCIIbetical sort except preserving natural order for numbers */
498 static int cmp_func_names(const char *a, const char *b)
499 {
500  const char *start = a;
501  int ascii_diff, digit_diff;
502 
503  for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
504  for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
505 
506  if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
507  return digit_diff;
508 
509  return ascii_diff;
510 }
511 
512 /* Perform a tree rotation in the specified direction and return the new root */
514 {
515  CheckasmFunc *r = f->child[dir^1];
516  f->child[dir^1] = r->child[dir];
517  r->child[dir] = f;
518  r->color = f->color;
519  f->color = 0;
520  return r;
521 }
522 
523 #define is_red(f) ((f) && !(f)->color)
524 
525 /* Balance a left-leaning red-black tree at the specified node */
526 static void balance_tree(CheckasmFunc **root)
527 {
528  CheckasmFunc *f = *root;
529 
530  if (is_red(f->child[0]) && is_red(f->child[1])) {
531  f->color ^= 1;
532  f->child[0]->color = f->child[1]->color = 1;
533  }
534 
535  if (!is_red(f->child[0]) && is_red(f->child[1]))
536  *root = rotate_tree(f, 0); /* Rotate left */
537  else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
538  *root = rotate_tree(f, 1); /* Rotate right */
539 }
540 
541 /* Get a node with the specified name, creating it if it doesn't exist */
542 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
543 {
544  CheckasmFunc *f = *root;
545 
546  if (f) {
547  /* Search the tree for a matching node */
548  int cmp = cmp_func_names(name, f->name);
549  if (cmp) {
550  f = get_func(&f->child[cmp > 0], name);
551 
552  /* Rebalance the tree on the way up if a new node was inserted */
553  if (!f->versions.func)
554  balance_tree(root);
555  }
556  } else {
557  /* Allocate and insert a new node into the tree */
558  int name_length = strlen(name);
559  f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
560  memcpy(f->name, name, name_length + 1);
561  }
562 
563  return f;
564 }
565 
566 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
567 static void check_cpu_flag(const char *name, int flag)
568 {
569  int old_cpu_flag = state.cpu_flag;
570 
571  flag |= old_cpu_flag;
572  av_force_cpu_flags(-1);
573  state.cpu_flag = flag & av_get_cpu_flags();
574  av_force_cpu_flags(state.cpu_flag);
575 
576  if (!flag || state.cpu_flag != old_cpu_flag) {
577  int i;
578 
579  state.cpu_flag_name = name;
580  for (i = 0; tests[i].func; i++) {
581  if (state.test_name && strcmp(tests[i].name, state.test_name))
582  continue;
583  state.current_test_name = tests[i].name;
584  tests[i].func();
585  }
586  }
587 }
588 
589 /* Print the name of the current CPU flag, but only do it once */
590 static void print_cpu_name(void)
591 {
592  if (state.cpu_flag_name) {
593  color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
594  state.cpu_flag_name = NULL;
595  }
596 }
597 
598 #if CONFIG_LINUX_PERF
599 static int bench_init_linux(void)
600 {
601  struct perf_event_attr attr = {
602  .type = PERF_TYPE_HARDWARE,
603  .size = sizeof(struct perf_event_attr),
604  .config = PERF_COUNT_HW_CPU_CYCLES,
605  .disabled = 1, // start counting only on demand
606  .exclude_kernel = 1,
607  .exclude_hv = 1,
608  };
609 
610  printf("benchmarking with Linux Perf Monitoring API\n");
611 
612  state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
613  if (state.sysfd == -1) {
614  perror("syscall");
615  return -1;
616  }
617  return 0;
618 }
619 #endif
620 
621 #if !CONFIG_LINUX_PERF
622 static int bench_init_ffmpeg(void)
623 {
624 #ifdef AV_READ_TIME
625  printf("benchmarking with native FFmpeg timers\n");
626  return 0;
627 #else
628  fprintf(stderr, "checkasm: --bench is not supported on your system\n");
629  return -1;
630 #endif
631 }
632 #endif
633 
634 static int bench_init(void)
635 {
636 #if CONFIG_LINUX_PERF
637  int ret = bench_init_linux();
638 #else
639  int ret = bench_init_ffmpeg();
640 #endif
641  if (ret < 0)
642  return ret;
643 
644  state.nop_time = measure_nop_time();
645  printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
646  return 0;
647 }
648 
649 static void bench_uninit(void)
650 {
651 #if CONFIG_LINUX_PERF
652  if (state.sysfd > 0)
653  close(state.sysfd);
654 #endif
655 }
656 
657 int main(int argc, char *argv[])
658 {
659  unsigned int seed = av_get_random_seed();
660  int i, ret = 0;
661 
662 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
664  checkasm_checked_call = checkasm_checked_call_vfp;
665 #endif
666 
667  if (!tests[0].func || !cpus[0].flag) {
668  fprintf(stderr, "checkasm: no tests to perform\n");
669  return 0;
670  }
671 
672  while (argc > 1) {
673  if (!strncmp(argv[1], "--bench", 7)) {
674  if (bench_init() < 0)
675  return 1;
676  if (argv[1][7] == '=') {
677  state.bench_pattern = argv[1] + 8;
678  state.bench_pattern_len = strlen(state.bench_pattern);
679  } else
680  state.bench_pattern = "";
681  } else if (!strncmp(argv[1], "--test=", 7)) {
682  state.test_name = argv[1] + 7;
683  } else {
684  seed = strtoul(argv[1], NULL, 10);
685  }
686 
687  argc--;
688  argv++;
689  }
690 
691  fprintf(stderr, "checkasm: using random seed %u\n", seed);
693 
694  check_cpu_flag(NULL, 0);
695  for (i = 0; cpus[i].flag; i++)
697 
698  if (state.num_failed) {
699  fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
700  ret = 1;
701  } else {
702  fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
703  if (state.bench_pattern) {
704  print_benchs(state.funcs);
705  }
706  }
707 
708  destroy_func_tree(state.funcs);
709  bench_uninit();
710  return ret;
711 }
712 
713 /* Decide whether or not the specified function needs to be tested and
714  * allocate/initialize data structures if needed. Returns a pointer to a
715  * reference function if the function should be tested, otherwise NULL */
716 void *checkasm_check_func(void *func, const char *name, ...)
717 {
718  char name_buf[256];
719  void *ref = func;
721  int name_length;
722  va_list arg;
723 
724  va_start(arg, name);
725  name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
726  va_end(arg);
727 
728  if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
729  return NULL;
730 
731  state.current_func = get_func(&state.funcs, name_buf);
732  state.funcs->color = 1;
733  v = &state.current_func->versions;
734 
735  if (v->func) {
736  CheckasmFuncVersion *prev;
737  do {
738  /* Only test functions that haven't already been tested */
739  if (v->func == func)
740  return NULL;
741 
742  if (v->ok)
743  ref = v->func;
744 
745  prev = v;
746  } while ((v = v->next));
747 
748  v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
749  }
750 
751  v->func = func;
752  v->ok = 1;
753  v->cpu = state.cpu_flag;
754  state.current_func_ver = v;
755 
756  if (state.cpu_flag)
757  state.num_checked++;
758 
759  return ref;
760 }
761 
762 /* Decide whether or not the current function needs to be benchmarked */
764 {
765  return !state.num_failed && state.bench_pattern &&
766  !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
767 }
768 
769 /* Indicate that the current test has failed */
770 void checkasm_fail_func(const char *msg, ...)
771 {
772  if (state.current_func_ver->cpu && state.current_func_ver->ok) {
773  va_list arg;
774 
775  print_cpu_name();
776  fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
777  va_start(arg, msg);
778  vfprintf(stderr, msg, arg);
779  va_end(arg);
780  fprintf(stderr, ")\n");
781 
782  state.current_func_ver->ok = 0;
783  state.num_failed++;
784  }
785 }
786 
787 /* Get the benchmark context of the current function */
789 {
790  CheckasmPerf *perf = &state.current_func_ver->perf;
791  memset(perf, 0, sizeof(*perf));
792  perf->sysfd = state.sysfd;
793  return perf;
794 }
795 
796 /* Print the outcome of all tests performed since the last time this function was called */
797 void checkasm_report(const char *name, ...)
798 {
799  static int prev_checked, prev_failed, max_length;
800 
801  if (state.num_checked > prev_checked) {
802  int pad_length = max_length + 4;
803  va_list arg;
804 
805  print_cpu_name();
806  pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
807  va_start(arg, name);
808  pad_length -= vfprintf(stderr, name, arg);
809  va_end(arg);
810  fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
811 
812  if (state.num_failed == prev_failed)
813  color_printf(COLOR_GREEN, "OK");
814  else
815  color_printf(COLOR_RED, "FAILED");
816  fprintf(stderr, "]\n");
817 
818  prev_checked = state.num_checked;
819  prev_failed = state.num_failed;
820  } else if (!state.cpu_flag) {
821  /* Calculate the amount of padding required to make the output vertically aligned */
822  int length = strlen(state.current_test_name);
823  va_list arg;
824 
825  va_start(arg, name);
826  length += vsnprintf(NULL, 0, name, arg);
827  va_end(arg);
828 
829  if (length > max_length)
830  max_length = length;
831  }
832 }
checkasm_check_nlmeans
void checkasm_check_nlmeans(void)
Definition: vf_nlmeans.c:31
AV_CPU_FLAG_VFP
#define AV_CPU_FLAG_VFP
Definition: cpu.h:67
bench_pattern_len
int bench_pattern_len
Definition: checkasm.c:254
av_force_cpu_flags
void av_force_cpu_flags(int arg)
Disables cpu detection and forces the specified flags.
Definition: cpu.c:65
checkasm_check_vp8dsp
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
checkasm_check_blockdsp
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:54
AV_CPU_FLAG_SSE3
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
Definition: cpu.h:40
destroy_func_tree
static void destroy_func_tree(CheckasmFunc *f)
Definition: checkasm.c:409
r
const char * r
Definition: vf_curves.c:114
checkasm_check_videodsp
void checkasm_check_videodsp(void)
Definition: videodsp.c:80
cmp_nop
static int cmp_nop(const void *a, const void *b)
Definition: checkasm.c:449
COLOR_RED
#define COLOR_RED
Definition: checkasm.c:51
tests
static const struct @311 tests[]
checkasm_lfg
AVLFG checkasm_lfg
Definition: checkasm.c:268
color
Definition: vf_paletteuse.c:588
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:252
checkasm_check_alacdsp
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:115
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:317
AV_CPU_FLAG_SSE3SLOW
#define AV_CPU_FLAG_SSE3SLOW
SSE3 supported, but usually not faster.
Definition: cpu.h:41
CheckasmFunc
Definition: checkasm.c:240
name
const char * name
Definition: checkasm.c:72
bench_pattern
const char * bench_pattern
Definition: checkasm.c:253
av_unused
#define av_unused
Definition: attributes.h:125
checkasm_check_hevc_idct
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
CheckasmFunc::child
struct CheckasmFunc * child[2]
Definition: checkasm.c:241
checkasm_check_afir
void checkasm_check_afir(void)
Definition: af_afir.c:67
print_cpu_name
static void print_cpu_name(void)
Definition: checkasm.c:590
AV_CPU_FLAG_3DNOW
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
Definition: cpu.h:34
av_intfloat32::i
uint32_t i
Definition: intfloat.h:28
WORD
unsigned short WORD
Definition: basicDataTypeConversions.h:48
float_near_abs_eps
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:306
b
#define b
Definition: input.c:41
checkasm_check_aacpsdsp
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:215
bench_uninit
static void bench_uninit(void)
Definition: checkasm.c:649
HANDLE
PVOID HANDLE
Definition: basicDataTypeConversions.h:21
CheckasmFuncVersion::cpu
int cpu
Definition: checkasm.c:235
cpu_flag
int cpu_flag
Definition: checkasm.c:262
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:93
cpu_suffix
static const char * cpu_suffix(int cpu)
Definition: checkasm.c:438
CheckasmPerf::sysfd
int sysfd
Definition: checkasm.h:204
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:441
color_printf
static void color_printf(int color, const char *fmt,...)
Definition: checkasm.c:366
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:120
main
int main(int argc, char *argv[])
Definition: checkasm.c:657
AV_CPU_FLAG_SSSE3
#define AV_CPU_FLAG_SSSE3
Conroe SSSE3 functions.
Definition: cpu.h:43
fmt
const char * fmt
Definition: avisynth_c.h:861
double_near_abs_eps
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:346
CheckasmPerf
Definition: checkasm.h:203
start
void INT64 start
Definition: avisynth_c.h:767
current_test_name
const char * current_test_name
Definition: checkasm.c:252
PERF_START
#define PERF_START(t)
Definition: checkasm.h:254
rotate_tree
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
Definition: checkasm.c:513
AV_CPU_FLAG_XOP
#define AV_CPU_FLAG_XOP
Bulldozer XOP functions.
Definition: cpu.h:51
checkasm_check_vp9dsp
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:625
checkasm.h
checkasm_check_vf_threshold
void checkasm_check_vf_threshold(void)
Definition: vf_threshold.c:78
checkasm_check_vf_gblur
void checkasm_check_vf_gblur(void)
Definition: vf_gblur.c:36
AV_CPU_FLAG_3DNOWEXT
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
Definition: cpu.h:39
float_near_abs_eps_ulp
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:329
AV_CPU_FLAG_VSX
#define AV_CPU_FLAG_VSX
ISA 2.06.
Definition: cpu.h:61
num_failed
int num_failed
Definition: checkasm.c:256
checkasm_check_float_dsp
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:277
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:58
bench_init
static int bench_init(void)
Definition: checkasm.c:634
checkasm_check_pixblockdsp
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
state
static struct @313 state
CheckasmFuncVersion
Definition: checkasm.c:231
COLOR_YELLOW
#define COLOR_YELLOW
Definition: checkasm.c:53
nop_time
int nop_time
Definition: checkasm.c:259
CheckasmFuncVersion::perf
CheckasmPerf perf
Definition: checkasm.c:236
checkasm_report
void checkasm_report(const char *name,...)
Definition: checkasm.c:797
get_func
static CheckasmFunc * get_func(CheckasmFunc **root, const char *name)
Definition: checkasm.c:542
checkasm_fail_func
void checkasm_fail_func(const char *msg,...)
Definition: checkasm.c:770
checkasm_check_sw_rgb
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:114
checkasm_check_hevc_sao
void checkasm_check_hevc_sao(void)
Definition: hevc_sao.c:131
num_checked
int num_checked
Definition: checkasm.c:255
AV_CPU_FLAG_ARMV6
#define AV_CPU_FLAG_ARMV6
Definition: cpu.h:65
AV_CPU_FLAG_SSE4
#define AV_CPU_FLAG_SSE4
Penryn SSE4.1 functions.
Definition: cpu.h:46
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:260
f
#define f(width, name)
Definition: cbs_vp9.c:255
CheckasmPerf::iterations
int iterations
Definition: checkasm.h:206
CheckasmFuncVersion::ok
int ok
Definition: checkasm.c:234
arg
const char * arg
Definition: jacosubdec.c:66
func
void(* func)(void)
Definition: checkasm.c:73
NULL
#define NULL
Definition: coverity.c:32
measure_nop_time
static int measure_nop_time(void)
Definition: checkasm.c:455
double_near_abs_eps_array
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:353
checkasm_check_fmtconvert
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:44
COLOR_GREEN
#define COLOR_GREEN
Definition: checkasm.c:52
cpu.h
cpus
static const struct @312 cpus[]
av_intfloat32
Definition: intfloat.h:27
AV_CPU_FLAG_CMOV
#define AV_CPU_FLAG_CMOV
supports cmov instruction
Definition: cpu.h:53
seed
static unsigned int seed
Definition: videogen.c:78
AV_CPU_FLAG_ALTIVEC
#define AV_CPU_FLAG_ALTIVEC
standard
Definition: cpu.h:60
print_benchs
static void print_benchs(CheckasmFunc *f)
Definition: checkasm.c:476
current_func
CheckasmFunc * current_func
Definition: checkasm.c:250
AV_CPU_FLAG_SSE2
#define AV_CPU_FLAG_SSE2
PIV SSE2 functions.
Definition: cpu.h:36
CheckasmFunc::color
uint8_t color
Definition: checkasm.c:243
AVLFG
Context structure for the Lagged Fibonacci PRNG.
Definition: lfg.h:33
AV_CPU_FLAG_AVX
#define AV_CPU_FLAG_AVX
AVX functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:49
AV_CPU_FLAG_FMA4
#define AV_CPU_FLAG_FMA4
Bulldozer FMA4 functions.
Definition: cpu.h:52
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:54
checkasm_check_llviddsp
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:195
FFMAX
#define FFMAX(a, b)
Definition: common.h:94
AV_CPU_FLAG_SSE2SLOW
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
Definition: cpu.h:37
AV_CPU_FLAG_NEON
#define AV_CPU_FLAG_NEON
Definition: cpu.h:69
size
int size
Definition: twinvq_data.h:11134
printf
printf("static const uint8_t my_array[100] = {\n")
sysfd
int sysfd
Definition: checkasm.c:260
have_neon
#define have_neon(flags)
Definition: cpu.h:26
CheckasmFuncVersion::func
void * func
Definition: checkasm.c:233
checkasm_check_h264pred
void checkasm_check_h264pred(void)
Definition: h264pred.c:232
av_isdigit
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
Definition: avstring.h:206
AV_CPU_FLAG_FMA3
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
Definition: cpu.h:55
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
checkasm_check_huffyuvdsp
void checkasm_check_huffyuvdsp(void)
Definition: huffyuvdsp.c:67
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:334
test_name
const char * test_name
Definition: checkasm.c:264
CheckasmPerf::cycles
uint64_t cycles
Definition: checkasm.h:205
checkasm_get_perf_context
CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:788
checkasm_check_fixed_dsp
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:132
checkasm_check_colorspace
void checkasm_check_colorspace(void)
Definition: vf_colorspace.c:308
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:271
checkasm_check_audiodsp
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:259
checkasm_check_flacdsp
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:56
checkasm_check_func
void * checkasm_check_func(void *func, const char *name,...)
Definition: checkasm.c:716
AV_CPU_FLAG_SSE42
#define AV_CPU_FLAG_SSE42
Nehalem SSE4.2 functions.
Definition: cpu.h:47
CheckasmFuncVersion::next
struct CheckasmFuncVersion * next
Definition: checkasm.c:232
vsnprintf
#define vsnprintf
Definition: snprintf.h:36
common.h
use_color
static int use_color
Definition: log.c:121
checkasm_check_synth_filter
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:44
cmp_func_names
static int cmp_func_names(const char *a, const char *b)
Definition: checkasm.c:498
uint8_t
uint8_t
Definition: audio_convert.c:194
AV_CPU_FLAG_ARMV8
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:70
len
int len
Definition: vorbis_enc_data.h:452
checkasm_check_g722dsp
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
PERF_STOP
#define PERF_STOP(t)
Definition: checkasm.h:255
is_red
#define is_red(f)
Definition: checkasm.c:523
AV_CPU_FLAG_ATOM
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
Definition: cpu.h:45
have_vfp
#define have_vfp(flags)
Definition: cpu.h:27
checkasm_check_vf_hflip
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:69
ret
ret
Definition: filter_design.txt:187
checkasm_check_blend
void checkasm_check_blend(void)
Definition: vf_blend.c:88
AV_CPU_FLAG_VFPV3
#define AV_CPU_FLAG_VFPV3
Definition: cpu.h:68
checkasm_check_bswapdsp
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:59
FF_ARRAY_ELEMS
#define FF_ARRAY_ELEMS(a)
Definition: sinewin_tablegen_template.c:38
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:31
suffix
const char * suffix
Definition: checkasm.c:191
checkasm_malloc
static void * checkasm_malloc(size_t size)
Definition: checkasm.c:426
checkasm_check_h264qpel
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:50
random_seed.h
AV_CPU_FLAG_AESNI
#define AV_CPU_FLAG_AESNI
Advanced Encryption Standard functions.
Definition: cpu.h:48
checkasm_bench_func
int checkasm_bench_func(void)
Definition: checkasm.c:763
config.h
dummy
int dummy
Definition: motion.c:64
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:263
AV_CPU_FLAG_SSE
#define AV_CPU_FLAG_SSE
SSE functions.
Definition: cpu.h:35
ref
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
float_near_ulp_array
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:294
AV_CPU_FLAG_MMXEXT
#define AV_CPU_FLAG_MMXEXT
SSE integer functions or AMD MMX ext.
Definition: cpu.h:32
av_intfloat32::f
float f
Definition: intfloat.h:29
flag
int flag
Definition: checkasm.c:192
checkasm_check_v210enc
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
checkasm_check_sbrdsp
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:250
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
INVALID_HANDLE_VALUE
#define INVALID_HANDLE_VALUE
Definition: windows2linux.h:47
checkasm_check_jpeg2000dsp
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:91
checkasm_check_hevc_add_res
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:74
float_near_ulp
int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:276
current_func_ver
CheckasmFuncVersion * current_func_ver
Definition: checkasm.c:251
check_cpu_flag
static void check_cpu_flag(const char *name, int flag)
Definition: checkasm.c:567
length
const char int length
Definition: avisynth_c.h:860
bench_init_ffmpeg
static int bench_init_ffmpeg(void)
Definition: checkasm.c:622
CheckasmFunc::versions
CheckasmFuncVersion versions
Definition: checkasm.c:242
isatty
#define isatty(fd)
Definition: checkasm.c:61
CheckasmFunc::name
char name[1]
Definition: checkasm.c:244
checkasm_check_exrdsp
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:76
balance_tree
static void balance_tree(CheckasmFunc **root)
Definition: checkasm.c:526
AV_CPU_FLAG_ARMV6T2
#define AV_CPU_FLAG_ARMV6T2
Definition: cpu.h:66
funcs
CheckasmFunc * funcs
Definition: checkasm.c:249