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_BLEND_FILTER
154  { "vf_blend", checkasm_check_blend },
155  #endif
156  #if CONFIG_COLORSPACE_FILTER
157  { "vf_colorspace", checkasm_check_colorspace },
158  #endif
159  #if CONFIG_HFLIP_FILTER
160  { "vf_hflip", checkasm_check_vf_hflip },
161  #endif
162  #if CONFIG_THRESHOLD_FILTER
163  { "vf_threshold", checkasm_check_vf_threshold },
164  #endif
165 #endif
166 #if CONFIG_SWSCALE
167  { "sw_rgb", checkasm_check_sw_rgb },
168 #endif
169 #if CONFIG_AVUTIL
170  { "fixed_dsp", checkasm_check_fixed_dsp },
171  { "float_dsp", checkasm_check_float_dsp },
172 #endif
173  { NULL }
174 };
175 
176 /* List of cpu flags to check */
177 static const struct {
178  const char *name;
179  const char *suffix;
180  int flag;
181 } cpus[] = {
182 #if ARCH_AARCH64
183  { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
184  { "NEON", "neon", AV_CPU_FLAG_NEON },
185 #elif ARCH_ARM
186  { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
187  { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
188  { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
189  { "VFP", "vfp", AV_CPU_FLAG_VFP },
190  { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
191  { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
192  { "NEON", "neon", AV_CPU_FLAG_NEON },
193 #elif ARCH_PPC
194  { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
195  { "VSX", "vsx", AV_CPU_FLAG_VSX },
196  { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
197 #elif ARCH_X86
198  { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
199  { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
200  { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
201  { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
202  { "SSE", "sse", AV_CPU_FLAG_SSE },
203  { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
204  { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
205  { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
206  { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
207  { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
208  { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
209  { "AVX", "avx", AV_CPU_FLAG_AVX },
210  { "XOP", "xop", AV_CPU_FLAG_XOP },
211  { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
212  { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
213  { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
214  { "AVX-512", "avx512", AV_CPU_FLAG_AVX512 },
215 #endif
216  { NULL }
217 };
218 
219 typedef struct CheckasmFuncVersion {
221  void *func;
222  int ok;
223  int cpu;
226 
227 /* Binary search tree node */
228 typedef struct CheckasmFunc {
229  struct CheckasmFunc *child[2];
231  uint8_t color; /* 0 = red, 1 = black */
232  char name[1];
233 } CheckasmFunc;
234 
235 /* Internal state */
236 static struct {
240  const char *current_test_name;
241  const char *bench_pattern;
245 
246  /* perf */
247  int nop_time;
248  int sysfd;
249 
250  int cpu_flag;
251  const char *cpu_flag_name;
252  const char *test_name;
253 } state;
254 
255 /* PRNG state */
257 
258 /* float compare support code */
259 static int is_negative(union av_intfloat32 u)
260 {
261  return u.i >> 31;
262 }
263 
264 int float_near_ulp(float a, float b, unsigned max_ulp)
265 {
266  union av_intfloat32 x, y;
267 
268  x.f = a;
269  y.f = b;
270 
271  if (is_negative(x) != is_negative(y)) {
272  // handle -0.0 == +0.0
273  return a == b;
274  }
275 
276  if (llabs((int64_t)x.i - y.i) <= max_ulp)
277  return 1;
278 
279  return 0;
280 }
281 
282 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
283  unsigned len)
284 {
285  unsigned i;
286 
287  for (i = 0; i < len; i++) {
288  if (!float_near_ulp(a[i], b[i], max_ulp))
289  return 0;
290  }
291  return 1;
292 }
293 
294 int float_near_abs_eps(float a, float b, float eps)
295 {
296  float abs_diff = fabsf(a - b);
297  if (abs_diff < eps)
298  return 1;
299 
300  fprintf(stderr, "test failed comparing %g with %g (abs diff=%g with EPS=%g)\n", a, b, abs_diff, eps);
301 
302  return 0;
303 }
304 
305 int float_near_abs_eps_array(const float *a, const float *b, float eps,
306  unsigned len)
307 {
308  unsigned i;
309 
310  for (i = 0; i < len; i++) {
311  if (!float_near_abs_eps(a[i], b[i], eps))
312  return 0;
313  }
314  return 1;
315 }
316 
317 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
318 {
319  return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
320 }
321 
322 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
323  unsigned max_ulp, unsigned len)
324 {
325  unsigned i;
326 
327  for (i = 0; i < len; i++) {
328  if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
329  return 0;
330  }
331  return 1;
332 }
333 
334 int double_near_abs_eps(double a, double b, double eps)
335 {
336  double abs_diff = fabs(a - b);
337 
338  return abs_diff < eps;
339 }
340 
341 int double_near_abs_eps_array(const double *a, const double *b, double eps,
342  unsigned len)
343 {
344  unsigned i;
345 
346  for (i = 0; i < len; i++) {
347  if (!double_near_abs_eps(a[i], b[i], eps))
348  return 0;
349  }
350  return 1;
351 }
352 
353 /* Print colored text to stderr if the terminal supports it */
354 static void color_printf(int color, const char *fmt, ...)
355 {
356  static int use_color = -1;
357  va_list arg;
358 
359 #if HAVE_SETCONSOLETEXTATTRIBUTE
360  static HANDLE con;
361  static WORD org_attributes;
362 
363  if (use_color < 0) {
364  CONSOLE_SCREEN_BUFFER_INFO con_info;
365  con = GetStdHandle(STD_ERROR_HANDLE);
366  if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
367  org_attributes = con_info.wAttributes;
368  use_color = 1;
369  } else
370  use_color = 0;
371  }
372  if (use_color)
373  SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
374 #else
375  if (use_color < 0) {
376  const char *term = getenv("TERM");
377  use_color = term && strcmp(term, "dumb") && isatty(2);
378  }
379  if (use_color)
380  fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
381 #endif
382 
383  va_start(arg, fmt);
384  vfprintf(stderr, fmt, arg);
385  va_end(arg);
386 
387  if (use_color) {
388 #if HAVE_SETCONSOLETEXTATTRIBUTE
389  SetConsoleTextAttribute(con, org_attributes);
390 #else
391  fprintf(stderr, "\x1b[0m");
392 #endif
393  }
394 }
395 
396 /* Deallocate a tree */
398 {
399  if (f) {
401  while (v) {
402  CheckasmFuncVersion *next = v->next;
403  free(v);
404  v = next;
405  }
406 
407  destroy_func_tree(f->child[0]);
408  destroy_func_tree(f->child[1]);
409  free(f);
410  }
411 }
412 
413 /* Allocate a zero-initialized block, clean up and exit on failure */
414 static void *checkasm_malloc(size_t size)
415 {
416  void *ptr = calloc(1, size);
417  if (!ptr) {
418  fprintf(stderr, "checkasm: malloc failed\n");
419  destroy_func_tree(state.funcs);
420  exit(1);
421  }
422  return ptr;
423 }
424 
425 /* Get the suffix of the specified cpu flag */
426 static const char *cpu_suffix(int cpu)
427 {
428  int i = FF_ARRAY_ELEMS(cpus);
429 
430  while (--i >= 0)
431  if (cpu & cpus[i].flag)
432  return cpus[i].suffix;
433 
434  return "c";
435 }
436 
437 static int cmp_nop(const void *a, const void *b)
438 {
439  return *(const uint16_t*)a - *(const uint16_t*)b;
440 }
441 
442 /* Measure the overhead of the timing code (in decicycles) */
443 static int measure_nop_time(void)
444 {
445  uint16_t nops[10000];
446  int i, nop_sum = 0;
447  av_unused const int sysfd = state.sysfd;
448 
449  uint64_t t = 0;
450  for (i = 0; i < 10000; i++) {
451  PERF_START(t);
452  PERF_STOP(t);
453  nops[i] = t;
454  }
455 
456  qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
457  for (i = 2500; i < 7500; i++)
458  nop_sum += nops[i];
459 
460  return nop_sum / 500;
461 }
462 
463 /* Print benchmark results */
465 {
466  if (f) {
467  print_benchs(f->child[0]);
468 
469  /* Only print functions with at least one assembly version */
470  if (f->versions.cpu || f->versions.next) {
471  CheckasmFuncVersion *v = &f->versions;
472  do {
473  CheckasmPerf *p = &v->perf;
474  if (p->iterations) {
475  int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
476  printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
477  }
478  } while ((v = v->next));
479  }
480 
481  print_benchs(f->child[1]);
482  }
483 }
484 
485 /* ASCIIbetical sort except preserving natural order for numbers */
486 static int cmp_func_names(const char *a, const char *b)
487 {
488  const char *start = a;
489  int ascii_diff, digit_diff;
490 
491  for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
492  for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
493 
494  if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
495  return digit_diff;
496 
497  return ascii_diff;
498 }
499 
500 /* Perform a tree rotation in the specified direction and return the new root */
502 {
503  CheckasmFunc *r = f->child[dir^1];
504  f->child[dir^1] = r->child[dir];
505  r->child[dir] = f;
506  r->color = f->color;
507  f->color = 0;
508  return r;
509 }
510 
511 #define is_red(f) ((f) && !(f)->color)
512 
513 /* Balance a left-leaning red-black tree at the specified node */
514 static void balance_tree(CheckasmFunc **root)
515 {
516  CheckasmFunc *f = *root;
517 
518  if (is_red(f->child[0]) && is_red(f->child[1])) {
519  f->color ^= 1;
520  f->child[0]->color = f->child[1]->color = 1;
521  }
522 
523  if (!is_red(f->child[0]) && is_red(f->child[1]))
524  *root = rotate_tree(f, 0); /* Rotate left */
525  else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
526  *root = rotate_tree(f, 1); /* Rotate right */
527 }
528 
529 /* Get a node with the specified name, creating it if it doesn't exist */
530 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
531 {
532  CheckasmFunc *f = *root;
533 
534  if (f) {
535  /* Search the tree for a matching node */
536  int cmp = cmp_func_names(name, f->name);
537  if (cmp) {
538  f = get_func(&f->child[cmp > 0], name);
539 
540  /* Rebalance the tree on the way up if a new node was inserted */
541  if (!f->versions.func)
542  balance_tree(root);
543  }
544  } else {
545  /* Allocate and insert a new node into the tree */
546  int name_length = strlen(name);
547  f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
548  memcpy(f->name, name, name_length + 1);
549  }
550 
551  return f;
552 }
553 
554 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
555 static void check_cpu_flag(const char *name, int flag)
556 {
557  int old_cpu_flag = state.cpu_flag;
558 
559  flag |= old_cpu_flag;
560  av_force_cpu_flags(-1);
561  state.cpu_flag = flag & av_get_cpu_flags();
562  av_force_cpu_flags(state.cpu_flag);
563 
564  if (!flag || state.cpu_flag != old_cpu_flag) {
565  int i;
566 
567  state.cpu_flag_name = name;
568  for (i = 0; tests[i].func; i++) {
569  if (state.test_name && strcmp(tests[i].name, state.test_name))
570  continue;
571  state.current_test_name = tests[i].name;
572  tests[i].func();
573  }
574  }
575 }
576 
577 /* Print the name of the current CPU flag, but only do it once */
578 static void print_cpu_name(void)
579 {
580  if (state.cpu_flag_name) {
581  color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
582  state.cpu_flag_name = NULL;
583  }
584 }
585 
586 #if CONFIG_LINUX_PERF
587 static int bench_init_linux(void)
588 {
589  struct perf_event_attr attr = {
590  .type = PERF_TYPE_HARDWARE,
591  .size = sizeof(struct perf_event_attr),
592  .config = PERF_COUNT_HW_CPU_CYCLES,
593  .disabled = 1, // start counting only on demand
594  .exclude_kernel = 1,
595  .exclude_hv = 1,
596  };
597 
598  printf("benchmarking with Linux Perf Monitoring API\n");
599 
600  state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
601  if (state.sysfd == -1) {
602  perror("syscall");
603  return -1;
604  }
605  return 0;
606 }
607 #endif
608 
609 static int bench_init_ffmpeg(void)
610 {
611 #ifdef AV_READ_TIME
612  printf("benchmarking with native FFmpeg timers\n");
613  return 0;
614 #else
615  fprintf(stderr, "checkasm: --bench is not supported on your system\n");
616  return -1;
617 #endif
618 }
619 
620 static int bench_init(void)
621 {
622 #if CONFIG_LINUX_PERF
623  int ret = bench_init_linux();
624 #else
625  int ret = bench_init_ffmpeg();
626 #endif
627  if (ret < 0)
628  return ret;
629 
630  state.nop_time = measure_nop_time();
631  printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
632  return 0;
633 }
634 
635 static void bench_uninit(void)
636 {
637 #if CONFIG_LINUX_PERF
638  if (state.sysfd > 0)
639  close(state.sysfd);
640 #endif
641 }
642 
643 int main(int argc, char *argv[])
644 {
645  unsigned int seed = av_get_random_seed();
646  int i, ret = 0;
647 
648 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
650  checkasm_checked_call = checkasm_checked_call_vfp;
651 #endif
652 
653  if (!tests[0].func || !cpus[0].flag) {
654  fprintf(stderr, "checkasm: no tests to perform\n");
655  return 0;
656  }
657 
658  while (argc > 1) {
659  if (!strncmp(argv[1], "--bench", 7)) {
660  if (bench_init() < 0)
661  return 1;
662  if (argv[1][7] == '=') {
663  state.bench_pattern = argv[1] + 8;
664  state.bench_pattern_len = strlen(state.bench_pattern);
665  } else
666  state.bench_pattern = "";
667  } else if (!strncmp(argv[1], "--test=", 7)) {
668  state.test_name = argv[1] + 7;
669  } else {
670  seed = strtoul(argv[1], NULL, 10);
671  }
672 
673  argc--;
674  argv++;
675  }
676 
677  fprintf(stderr, "checkasm: using random seed %u\n", seed);
678  av_lfg_init(&checkasm_lfg, seed);
679 
680  check_cpu_flag(NULL, 0);
681  for (i = 0; cpus[i].flag; i++)
682  check_cpu_flag(cpus[i].name, cpus[i].flag);
683 
684  if (state.num_failed) {
685  fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
686  ret = 1;
687  } else {
688  fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
689  if (state.bench_pattern) {
690  print_benchs(state.funcs);
691  }
692  }
693 
694  destroy_func_tree(state.funcs);
695  bench_uninit();
696  return ret;
697 }
698 
699 /* Decide whether or not the specified function needs to be tested and
700  * allocate/initialize data structures if needed. Returns a pointer to a
701  * reference function if the function should be tested, otherwise NULL */
702 void *checkasm_check_func(void *func, const char *name, ...)
703 {
704  char name_buf[256];
705  void *ref = func;
707  int name_length;
708  va_list arg;
709 
710  va_start(arg, name);
711  name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
712  va_end(arg);
713 
714  if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
715  return NULL;
716 
717  state.current_func = get_func(&state.funcs, name_buf);
718  state.funcs->color = 1;
719  v = &state.current_func->versions;
720 
721  if (v->func) {
722  CheckasmFuncVersion *prev;
723  do {
724  /* Only test functions that haven't already been tested */
725  if (v->func == func)
726  return NULL;
727 
728  if (v->ok)
729  ref = v->func;
730 
731  prev = v;
732  } while ((v = v->next));
733 
734  v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
735  }
736 
737  v->func = func;
738  v->ok = 1;
739  v->cpu = state.cpu_flag;
740  state.current_func_ver = v;
741 
742  if (state.cpu_flag)
743  state.num_checked++;
744 
745  return ref;
746 }
747 
748 /* Decide whether or not the current function needs to be benchmarked */
750 {
751  return !state.num_failed && state.bench_pattern &&
752  !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
753 }
754 
755 /* Indicate that the current test has failed */
756 void checkasm_fail_func(const char *msg, ...)
757 {
758  if (state.current_func_ver->cpu && state.current_func_ver->ok) {
759  va_list arg;
760 
761  print_cpu_name();
762  fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
763  va_start(arg, msg);
764  vfprintf(stderr, msg, arg);
765  va_end(arg);
766  fprintf(stderr, ")\n");
767 
768  state.current_func_ver->ok = 0;
769  state.num_failed++;
770  }
771 }
772 
773 /* Get the benchmark context of the current function */
775 {
776  CheckasmPerf *perf = &state.current_func_ver->perf;
777  memset(perf, 0, sizeof(*perf));
778  perf->sysfd = state.sysfd;
779  return perf;
780 }
781 
782 /* Print the outcome of all tests performed since the last time this function was called */
783 void checkasm_report(const char *name, ...)
784 {
785  static int prev_checked, prev_failed, max_length;
786 
787  if (state.num_checked > prev_checked) {
788  int pad_length = max_length + 4;
789  va_list arg;
790 
791  print_cpu_name();
792  pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
793  va_start(arg, name);
794  pad_length -= vfprintf(stderr, name, arg);
795  va_end(arg);
796  fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
797 
798  if (state.num_failed == prev_failed)
799  color_printf(COLOR_GREEN, "OK");
800  else
801  color_printf(COLOR_RED, "FAILED");
802  fprintf(stderr, "]\n");
803 
804  prev_checked = state.num_checked;
805  prev_failed = state.num_failed;
806  } else if (!state.cpu_flag) {
807  /* Calculate the amount of padding required to make the output vertically aligned */
808  int length = strlen(state.current_test_name);
809  va_list arg;
810 
811  va_start(arg, name);
812  length += vsnprintf(NULL, 0, name, arg);
813  va_end(arg);
814 
815  if (length > max_length)
816  max_length = length;
817  }
818 }
Definition: lfg.h:27
const char * name
Definition: checkasm.c:72
static int bench_init_ffmpeg(void)
Definition: checkasm.c:609
static int cmp_nop(const void *a, const void *b)
Definition: checkasm.c:437
#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:443
static void bench_uninit(void)
Definition: checkasm.c:635
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
Definition: avstring.h:206
CheckasmFunc * current_func
Definition: checkasm.c:238
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:305
static int is_negative(union av_intfloat32 u)
Definition: checkasm.c:259
uint64_t cycles
Definition: checkasm.h:201
CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:774
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:113
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:56
#define AV_CPU_FLAG_CMOV
supports cmov instruction
Definition: cpu.h:53
void checkasm_check_h264pred(void)
Definition: h264pred.c:232
#define AV_CPU_FLAG_VFP
Definition: cpu.h:67
static const char * cpu_suffix(int cpu)
Definition: checkasm.c:426
CheckasmFuncVersion * current_func_ver
Definition: checkasm.c:239
const char * test_name
Definition: checkasm.c:252
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:486
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:294
void * checkasm_check_func(void *func, const char *name,...)
Definition: checkasm.c:702
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:215
CheckasmPerf perf
Definition: checkasm.c:224
void checkasm_check_colorspace(void)
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
static const struct @269 tests[]
struct CheckasmFunc * child[2]
Definition: checkasm.c:229
uint8_t
const char * cpu_flag_name
Definition: checkasm.c:251
int checkasm_bench_func(void)
Definition: checkasm.c:749
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
Definition: cpu.h:55
char name[1]
Definition: checkasm.c:232
#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:643
#define u(width, name, range_min, range_max)
Definition: cbs_h2645.c:344
#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:237
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:180
#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:530
CheckasmFuncVersion versions
Definition: checkasm.c:230
const char * suffix
Definition: checkasm.c:179
#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:264
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:315
int num_checked
Definition: checkasm.c:243
#define COLOR_GREEN
Definition: checkasm.c:52
const char * r
Definition: vf_curves.c:111
const char * arg
Definition: jacosubdec.c:66
#define AV_CPU_FLAG_ARMV6T2
Definition: cpu.h: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:620
AVLFG checkasm_lfg
Definition: checkasm.c:256
#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:464
#define have_neon(flags)
Definition: cpu.h:26
void checkasm_check_utvideodsp(void)
Definition: utvideodsp.c:90
static const struct @270 cpus[]
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
Definition: cpu.h:40
PVOID HANDLE
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:67
#define is_red(f)
Definition: checkasm.c:511
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:414
const char * bench_pattern
Definition: checkasm.c:241
static struct @271 state
int iterations
Definition: checkasm.h:202
int bench_pattern_len
Definition: checkasm.c:242
void checkasm_report(const char *name,...)
Definition: checkasm.c:783
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
#define FF_ARRAY_ELEMS(a)
int cpu_flag
Definition: checkasm.c:250
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:282
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
Definition: cpu.h:34
int nop_time
Definition: checkasm.c:247
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:334
#define AV_CPU_FLAG_ARMV6
Definition: cpu.h:65
int num_failed
Definition: checkasm.c:244
#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:250
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
Definition: checkasm.c:501
#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:248
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:341
static void color_printf(int color, const char *fmt,...)
Definition: checkasm.c:354
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
Definition: cpu.h:39
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:255
#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:317
static void destroy_func_tree(CheckasmFunc *f)
Definition: checkasm.c:397
common internal and external API header
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
struct CheckasmFuncVersion * next
Definition: checkasm.c:220
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:70
const char * current_test_name
Definition: checkasm.c:240
static void check_cpu_flag(const char *name, int flag)
Definition: checkasm.c:555
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:251
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:322
#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:756
#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:514
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:69
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:578
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:54
uint8_t color
Definition: checkasm.c:231