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 <stdarg.h>
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include "checkasm.h"
28 #include "libavutil/common.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/intfloat.h"
31 #include "libavutil/random_seed.h"
32 
33 #if HAVE_IO_H
34 #include <io.h>
35 #endif
36 
37 #if HAVE_SETCONSOLETEXTATTRIBUTE
38 #include <windows.h>
39 #define COLOR_RED FOREGROUND_RED
40 #define COLOR_GREEN FOREGROUND_GREEN
41 #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
42 #else
43 #define COLOR_RED 1
44 #define COLOR_GREEN 2
45 #define COLOR_YELLOW 3
46 #endif
47 
48 #if HAVE_UNISTD_H
49 #include <unistd.h>
50 #endif
51 
52 #if !HAVE_ISATTY
53 #define isatty(fd) 1
54 #endif
55 
56 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
57 #include "libavutil/arm/cpu.h"
58 
59 void (*checkasm_checked_call)(void *func, int dummy, ...) = checkasm_checked_call_novfp;
60 #endif
61 
62 /* List of tests to invoke */
63 static const struct {
64  const char *name;
65  void (*func)(void);
66 } tests[] = {
67 #if CONFIG_AVCODEC
68  #if CONFIG_ALAC_DECODER
69  { "alacdsp", checkasm_check_alacdsp },
70  #endif
71  #if CONFIG_AUDIODSP
72  { "audiodsp", checkasm_check_audiodsp },
73  #endif
74  #if CONFIG_BLOCKDSP
75  { "blockdsp", checkasm_check_blockdsp },
76  #endif
77  #if CONFIG_BSWAPDSP
78  { "bswapdsp", checkasm_check_bswapdsp },
79  #endif
80  #if CONFIG_DCA_DECODER
81  { "synth_filter", checkasm_check_synth_filter },
82  #endif
83  #if CONFIG_FLACDSP
84  { "flacdsp", checkasm_check_flacdsp },
85  #endif
86  #if CONFIG_FMTCONVERT
87  { "fmtconvert", checkasm_check_fmtconvert },
88  #endif
89  #if CONFIG_H264DSP
90  { "h264dsp", checkasm_check_h264dsp },
91  #endif
92  #if CONFIG_H264PRED
93  { "h264pred", checkasm_check_h264pred },
94  #endif
95  #if CONFIG_H264QPEL
96  { "h264qpel", checkasm_check_h264qpel },
97  #endif
98  #if CONFIG_HEVC_DECODER
99  { "hevc_add_res", checkasm_check_hevc_add_res },
100  { "hevc_idct", checkasm_check_hevc_idct },
101  #endif
102  #if CONFIG_JPEG2000_DECODER
103  { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
104  #endif
105  #if CONFIG_HUFFYUVDSP
106  { "llviddsp", checkasm_check_llviddsp },
107  #endif
108  #if CONFIG_PIXBLOCKDSP
109  { "pixblockdsp", checkasm_check_pixblockdsp },
110  #endif
111  #if CONFIG_V210_ENCODER
112  { "v210enc", checkasm_check_v210enc },
113  #endif
114  #if CONFIG_VP8DSP
115  { "vp8dsp", checkasm_check_vp8dsp },
116  #endif
117  #if CONFIG_VP9_DECODER
118  { "vp9dsp", checkasm_check_vp9dsp },
119  #endif
120  #if CONFIG_VIDEODSP
121  { "videodsp", checkasm_check_videodsp },
122  #endif
123 #endif
124 #if CONFIG_AVFILTER
125  #if CONFIG_BLEND_FILTER
126  { "vf_blend", checkasm_check_blend },
127  #endif
128  #if CONFIG_COLORSPACE_FILTER
129  { "vf_colorspace", checkasm_check_colorspace },
130  #endif
131 #endif
132  { NULL }
133 };
134 
135 /* List of cpu flags to check */
136 static const struct {
137  const char *name;
138  const char *suffix;
139  int flag;
140 } cpus[] = {
141 #if ARCH_AARCH64
142  { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
143  { "NEON", "neon", AV_CPU_FLAG_NEON },
144 #elif ARCH_ARM
145  { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
146  { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
147  { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
148  { "VFP", "vfp", AV_CPU_FLAG_VFP },
149  { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
150  { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
151  { "NEON", "neon", AV_CPU_FLAG_NEON },
152 #elif ARCH_PPC
153  { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
154  { "VSX", "vsx", AV_CPU_FLAG_VSX },
155  { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
156 #elif ARCH_X86
157  { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
158  { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
159  { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
160  { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
161  { "SSE", "sse", AV_CPU_FLAG_SSE },
162  { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
163  { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
164  { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
165  { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
166  { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
167  { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
168  { "AVX", "avx", AV_CPU_FLAG_AVX },
169  { "XOP", "xop", AV_CPU_FLAG_XOP },
170  { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
171  { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
172  { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
173 #endif
174  { NULL }
175 };
176 
177 typedef struct CheckasmFuncVersion {
179  void *func;
180  int ok;
181  int cpu;
183  uint64_t cycles;
185 
186 /* Binary search tree node */
187 typedef struct CheckasmFunc {
188  struct CheckasmFunc *child[2];
190  uint8_t color; /* 0 = red, 1 = black */
191  char name[1];
192 } CheckasmFunc;
193 
194 /* Internal state */
195 static struct {
199  const char *current_test_name;
200  const char *bench_pattern;
204  int nop_time;
205  int cpu_flag;
206  const char *cpu_flag_name;
207  const char *test_name;
208 } state;
209 
210 /* PRNG state */
212 
213 /* float compare support code */
214 static int is_negative(union av_intfloat32 u)
215 {
216  return u.i >> 31;
217 }
218 
219 int float_near_ulp(float a, float b, unsigned max_ulp)
220 {
221  union av_intfloat32 x, y;
222 
223  x.f = a;
224  y.f = b;
225 
226  if (is_negative(x) != is_negative(y)) {
227  // handle -0.0 == +0.0
228  return a == b;
229  }
230 
231  if (llabs((int64_t)x.i - y.i) <= max_ulp)
232  return 1;
233 
234  return 0;
235 }
236 
237 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
238  unsigned len)
239 {
240  unsigned i;
241 
242  for (i = 0; i < len; i++) {
243  if (!float_near_ulp(a[i], b[i], max_ulp))
244  return 0;
245  }
246  return 1;
247 }
248 
249 int float_near_abs_eps(float a, float b, float eps)
250 {
251  float abs_diff = fabsf(a - b);
252 
253  return abs_diff < eps;
254 }
255 
256 int float_near_abs_eps_array(const float *a, const float *b, float eps,
257  unsigned len)
258 {
259  unsigned i;
260 
261  for (i = 0; i < len; i++) {
262  if (!float_near_abs_eps(a[i], b[i], eps))
263  return 0;
264  }
265  return 1;
266 }
267 
268 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
269 {
270  return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
271 }
272 
273 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
274  unsigned max_ulp, unsigned len)
275 {
276  unsigned i;
277 
278  for (i = 0; i < len; i++) {
279  if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
280  return 0;
281  }
282  return 1;
283 }
284 
285 /* Print colored text to stderr if the terminal supports it */
286 static void color_printf(int color, const char *fmt, ...)
287 {
288  static int use_color = -1;
289  va_list arg;
290 
291 #if HAVE_SETCONSOLETEXTATTRIBUTE
292  static HANDLE con;
293  static WORD org_attributes;
294 
295  if (use_color < 0) {
296  CONSOLE_SCREEN_BUFFER_INFO con_info;
297  con = GetStdHandle(STD_ERROR_HANDLE);
298  if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
299  org_attributes = con_info.wAttributes;
300  use_color = 1;
301  } else
302  use_color = 0;
303  }
304  if (use_color)
305  SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
306 #else
307  if (use_color < 0) {
308  const char *term = getenv("TERM");
309  use_color = term && strcmp(term, "dumb") && isatty(2);
310  }
311  if (use_color)
312  fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
313 #endif
314 
315  va_start(arg, fmt);
316  vfprintf(stderr, fmt, arg);
317  va_end(arg);
318 
319  if (use_color) {
320 #if HAVE_SETCONSOLETEXTATTRIBUTE
321  SetConsoleTextAttribute(con, org_attributes);
322 #else
323  fprintf(stderr, "\x1b[0m");
324 #endif
325  }
326 }
327 
328 /* Deallocate a tree */
330 {
331  if (f) {
333  while (v) {
334  CheckasmFuncVersion *next = v->next;
335  free(v);
336  v = next;
337  }
338 
339  destroy_func_tree(f->child[0]);
340  destroy_func_tree(f->child[1]);
341  free(f);
342  }
343 }
344 
345 /* Allocate a zero-initialized block, clean up and exit on failure */
346 static void *checkasm_malloc(size_t size)
347 {
348  void *ptr = calloc(1, size);
349  if (!ptr) {
350  fprintf(stderr, "checkasm: malloc failed\n");
351  destroy_func_tree(state.funcs);
352  exit(1);
353  }
354  return ptr;
355 }
356 
357 /* Get the suffix of the specified cpu flag */
358 static const char *cpu_suffix(int cpu)
359 {
360  int i = FF_ARRAY_ELEMS(cpus);
361 
362  while (--i >= 0)
363  if (cpu & cpus[i].flag)
364  return cpus[i].suffix;
365 
366  return "c";
367 }
368 
369 #ifdef AV_READ_TIME
370 static int cmp_nop(const void *a, const void *b)
371 {
372  return *(const uint16_t*)a - *(const uint16_t*)b;
373 }
374 
375 /* Measure the overhead of the timing code (in decicycles) */
376 static int measure_nop_time(void)
377 {
378  uint16_t nops[10000];
379  int i, nop_sum = 0;
380 
381  for (i = 0; i < 10000; i++) {
382  uint64_t t = AV_READ_TIME();
383  nops[i] = AV_READ_TIME() - t;
384  }
385 
386  qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
387  for (i = 2500; i < 7500; i++)
388  nop_sum += nops[i];
389 
390  return nop_sum / 500;
391 }
392 
393 /* Print benchmark results */
394 static void print_benchs(CheckasmFunc *f)
395 {
396  if (f) {
397  print_benchs(f->child[0]);
398 
399  /* Only print functions with at least one assembly version */
400  if (f->versions.cpu || f->versions.next) {
401  CheckasmFuncVersion *v = &f->versions;
402  do {
403  if (v->iterations) {
404  int decicycles = (10*v->cycles/v->iterations - state.nop_time) / 4;
405  printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
406  }
407  } while ((v = v->next));
408  }
409 
410  print_benchs(f->child[1]);
411  }
412 }
413 #endif
414 
415 /* ASCIIbetical sort except preserving natural order for numbers */
416 static int cmp_func_names(const char *a, const char *b)
417 {
418  const char *start = a;
419  int ascii_diff, digit_diff;
420 
421  for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
422  for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
423 
424  if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
425  return digit_diff;
426 
427  return ascii_diff;
428 }
429 
430 /* Perform a tree rotation in the specified direction and return the new root */
432 {
433  CheckasmFunc *r = f->child[dir^1];
434  f->child[dir^1] = r->child[dir];
435  r->child[dir] = f;
436  r->color = f->color;
437  f->color = 0;
438  return r;
439 }
440 
441 #define is_red(f) ((f) && !(f)->color)
442 
443 /* Balance a left-leaning red-black tree at the specified node */
444 static void balance_tree(CheckasmFunc **root)
445 {
446  CheckasmFunc *f = *root;
447 
448  if (is_red(f->child[0]) && is_red(f->child[1])) {
449  f->color ^= 1;
450  f->child[0]->color = f->child[1]->color = 1;
451  }
452 
453  if (!is_red(f->child[0]) && is_red(f->child[1]))
454  *root = rotate_tree(f, 0); /* Rotate left */
455  else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
456  *root = rotate_tree(f, 1); /* Rotate right */
457 }
458 
459 /* Get a node with the specified name, creating it if it doesn't exist */
460 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
461 {
462  CheckasmFunc *f = *root;
463 
464  if (f) {
465  /* Search the tree for a matching node */
466  int cmp = cmp_func_names(name, f->name);
467  if (cmp) {
468  f = get_func(&f->child[cmp > 0], name);
469 
470  /* Rebalance the tree on the way up if a new node was inserted */
471  if (!f->versions.func)
472  balance_tree(root);
473  }
474  } else {
475  /* Allocate and insert a new node into the tree */
476  int name_length = strlen(name);
477  f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
478  memcpy(f->name, name, name_length + 1);
479  }
480 
481  return f;
482 }
483 
484 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
485 static void check_cpu_flag(const char *name, int flag)
486 {
487  int old_cpu_flag = state.cpu_flag;
488 
489  flag |= old_cpu_flag;
490  av_force_cpu_flags(-1);
491  state.cpu_flag = flag & av_get_cpu_flags();
492  av_force_cpu_flags(state.cpu_flag);
493 
494  if (!flag || state.cpu_flag != old_cpu_flag) {
495  int i;
496 
497  state.cpu_flag_name = name;
498  for (i = 0; tests[i].func; i++) {
499  if (state.test_name && strcmp(tests[i].name, state.test_name))
500  continue;
501  state.current_test_name = tests[i].name;
502  tests[i].func();
503  }
504  }
505 }
506 
507 /* Print the name of the current CPU flag, but only do it once */
508 static void print_cpu_name(void)
509 {
510  if (state.cpu_flag_name) {
511  color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
512  state.cpu_flag_name = NULL;
513  }
514 }
515 
516 int main(int argc, char *argv[])
517 {
518  unsigned int seed = av_get_random_seed();
519  int i, ret = 0;
520 
521 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
523  checkasm_checked_call = checkasm_checked_call_vfp;
524 #endif
525 
526  if (!tests[0].func || !cpus[0].flag) {
527  fprintf(stderr, "checkasm: no tests to perform\n");
528  return 0;
529  }
530 
531  while (argc > 1) {
532  if (!strncmp(argv[1], "--bench", 7)) {
533 #ifndef AV_READ_TIME
534  fprintf(stderr, "checkasm: --bench is not supported on your system\n");
535  return 1;
536 #endif
537  if (argv[1][7] == '=') {
538  state.bench_pattern = argv[1] + 8;
539  state.bench_pattern_len = strlen(state.bench_pattern);
540  } else
541  state.bench_pattern = "";
542  } else if (!strncmp(argv[1], "--test=", 7)) {
543  state.test_name = argv[1] + 7;
544  } else {
545  seed = strtoul(argv[1], NULL, 10);
546  }
547 
548  argc--;
549  argv++;
550  }
551 
552  fprintf(stderr, "checkasm: using random seed %u\n", seed);
553  av_lfg_init(&checkasm_lfg, seed);
554 
555  check_cpu_flag(NULL, 0);
556  for (i = 0; cpus[i].flag; i++)
557  check_cpu_flag(cpus[i].name, cpus[i].flag);
558 
559  if (state.num_failed) {
560  fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
561  ret = 1;
562  } else {
563  fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
564 #ifdef AV_READ_TIME
565  if (state.bench_pattern) {
566  state.nop_time = measure_nop_time();
567  printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
568  print_benchs(state.funcs);
569  }
570 #endif
571  }
572 
573  destroy_func_tree(state.funcs);
574  return ret;
575 }
576 
577 /* Decide whether or not the specified function needs to be tested and
578  * allocate/initialize data structures if needed. Returns a pointer to a
579  * reference function if the function should be tested, otherwise NULL */
580 void *checkasm_check_func(void *func, const char *name, ...)
581 {
582  char name_buf[256];
583  void *ref = func;
585  int name_length;
586  va_list arg;
587 
588  va_start(arg, name);
589  name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
590  va_end(arg);
591 
592  if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
593  return NULL;
594 
595  state.current_func = get_func(&state.funcs, name_buf);
596  state.funcs->color = 1;
597  v = &state.current_func->versions;
598 
599  if (v->func) {
600  CheckasmFuncVersion *prev;
601  do {
602  /* Only test functions that haven't already been tested */
603  if (v->func == func)
604  return NULL;
605 
606  if (v->ok)
607  ref = v->func;
608 
609  prev = v;
610  } while ((v = v->next));
611 
612  v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
613  }
614 
615  v->func = func;
616  v->ok = 1;
617  v->cpu = state.cpu_flag;
618  state.current_func_ver = v;
619 
620  if (state.cpu_flag)
621  state.num_checked++;
622 
623  return ref;
624 }
625 
626 /* Decide whether or not the current function needs to be benchmarked */
628 {
629  return !state.num_failed && state.bench_pattern &&
630  !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
631 }
632 
633 /* Indicate that the current test has failed */
634 void checkasm_fail_func(const char *msg, ...)
635 {
636  if (state.current_func_ver->cpu && state.current_func_ver->ok) {
637  va_list arg;
638 
639  print_cpu_name();
640  fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
641  va_start(arg, msg);
642  vfprintf(stderr, msg, arg);
643  va_end(arg);
644  fprintf(stderr, ")\n");
645 
646  state.current_func_ver->ok = 0;
647  state.num_failed++;
648  }
649 }
650 
651 /* Update benchmark results of the current function */
652 void checkasm_update_bench(int iterations, uint64_t cycles)
653 {
654  state.current_func_ver->iterations += iterations;
655  state.current_func_ver->cycles += cycles;
656 }
657 
658 /* Print the outcome of all tests performed since the last time this function was called */
659 void checkasm_report(const char *name, ...)
660 {
661  static int prev_checked, prev_failed, max_length;
662 
663  if (state.num_checked > prev_checked) {
664  int pad_length = max_length + 4;
665  va_list arg;
666 
667  print_cpu_name();
668  pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
669  va_start(arg, name);
670  pad_length -= vfprintf(stderr, name, arg);
671  va_end(arg);
672  fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
673 
674  if (state.num_failed == prev_failed)
675  color_printf(COLOR_GREEN, "OK");
676  else
677  color_printf(COLOR_RED, "FAILED");
678  fprintf(stderr, "]\n");
679 
680  prev_checked = state.num_checked;
681  prev_failed = state.num_failed;
682  } else if (!state.cpu_flag) {
683  /* Calculate the amount of padding required to make the output vertically aligned */
684  int length = strlen(state.current_test_name);
685  va_list arg;
686 
687  va_start(arg, name);
688  length += vsnprintf(NULL, 0, name, arg);
689  va_end(arg);
690 
691  if (length > max_length)
692  max_length = length;
693  }
694 }
Definition: lfg.h:27
const char * name
Definition: checkasm.c:64
#define AV_CPU_FLAG_AVX
AVX functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:47
#define NULL
Definition: coverity.c:32
#define AV_CPU_FLAG_ALTIVEC
standard
Definition: cpu.h:57
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
Definition: avstring.h:206
CheckasmFunc * current_func
Definition: checkasm.c:197
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:256
static int is_negative(union av_intfloat32 u)
Definition: checkasm.c:214
const char * fmt
Definition: avisynth_c.h:769
static const struct @252 cpus[]
#define AV_CPU_FLAG_SSE
SSE functions.
Definition: cpu.h:33
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:631
#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:51
void checkasm_check_h264pred(void)
Definition: h264pred.c:232
#define AV_CPU_FLAG_VFP
Definition: cpu.h:64
static const char * cpu_suffix(int cpu)
Definition: checkasm.c:358
CheckasmFuncVersion * current_func_ver
Definition: checkasm.c:198
const char * test_name
Definition: checkasm.c:207
void(* func)(void)
Definition: checkasm.c:65
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:115
static int cmp_func_names(const char *a, const char *b)
Definition: checkasm.c:416
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:249
void * checkasm_check_func(void *func, const char *name,...)
Definition: checkasm.c:580
void checkasm_check_colorspace(void)
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
struct CheckasmFunc * child[2]
Definition: checkasm.c:188
static const struct @251 tests[]
uint8_t
const char * cpu_flag_name
Definition: checkasm.c:206
int checkasm_bench_func(void)
Definition: checkasm.c:627
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
Definition: cpu.h:53
char name[1]
Definition: checkasm.c:191
#define AV_CPU_FLAG_NEON
Definition: cpu.h:66
#define AV_CPU_FLAG_MMXEXT
SSE integer functions or AMD MMX ext.
Definition: cpu.h:30
int main(int argc, char *argv[])
Definition: checkasm.c:516
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
Definition: cpu.h:43
#define COLOR_YELLOW
Definition: checkasm.c:45
#define AV_READ_TIME
Definition: timer.h:26
#define AV_CPU_FLAG_AVX2
AVX2 functions: requires OS support even if YMM registers aren't used.
Definition: cpu.h:52
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
Definition: cpu.h:35
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:50
ptrdiff_t size
Definition: opengl_enc.c:101
CheckasmFunc * funcs
Definition: checkasm.c:196
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:58
#define AV_CPU_FLAG_XOP
Bulldozer XOP functions.
Definition: cpu.h:49
int flag
Definition: checkasm.c:139
#define AV_CPU_FLAG_SSE42
Nehalem SSE4.2 functions.
Definition: cpu.h:45
#define isatty(fd)
Definition: checkasm.c:53
static CheckasmFunc * get_func(CheckasmFunc **root, const char *name)
Definition: checkasm.c:460
CheckasmFuncVersion versions
Definition: checkasm.c:189
const char * suffix
Definition: checkasm.c:138
#define AV_CPU_FLAG_SSSE3
Conroe SSSE3 functions.
Definition: cpu.h:41
unsigned short WORD
int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:219
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:229
int num_checked
Definition: checkasm.c:202
#define COLOR_GREEN
Definition: checkasm.c:44
static struct @253 state
const char * r
Definition: vf_curves.c:111
const char * arg
Definition: jacosubdec.c:66
#define AV_CPU_FLAG_ARMV6T2
Definition: cpu.h:63
GLsizei GLsizei * length
Definition: opengl_enc.c:115
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:74
AVLFG checkasm_lfg
Definition: checkasm.c:211
#define AV_CPU_FLAG_VFP_VM
VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations.
Definition: cpu.h:68
#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:61
#define have_neon(flags)
Definition: cpu.h:27
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
Definition: cpu.h:38
PVOID HANDLE
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
#define is_red(f)
Definition: checkasm.c:441
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:65
static void * checkasm_malloc(size_t size)
Definition: checkasm.c:346
const char * bench_pattern
Definition: checkasm.c:200
int bench_pattern_len
Definition: checkasm.c:201
void checkasm_report(const char *name,...)
Definition: checkasm.c:659
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
#define FF_ARRAY_ELEMS(a)
int cpu_flag
Definition: checkasm.c:205
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:237
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
Definition: cpu.h:32
int nop_time
Definition: checkasm.c:204
#define AV_CPU_FLAG_ARMV6
Definition: cpu.h:62
int num_failed
Definition: checkasm.c:203
#define AV_CPU_FLAG_SSE3SLOW
SSE3 supported, but usually not faster.
Definition: cpu.h:39
#define COLOR_RED
Definition: checkasm.c:43
#define AV_CPU_FLAG_MMX
standard MMX
Definition: cpu.h:29
static unsigned int seed
Definition: videogen.c:78
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:44
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
Definition: checkasm.c:431
#define AV_CPU_FLAG_FMA4
Bulldozer FMA4 functions.
Definition: cpu.h:50
#define AV_CPU_FLAG_SSE4
Penryn SSE4.1 functions.
Definition: cpu.h:44
#define AV_CPU_FLAG_AESNI
Advanced Encryption Standard functions.
Definition: cpu.h:46
#define AV_CPU_FLAG_VSX
ISA 2.06.
Definition: cpu.h:58
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:32
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:66
#define u(width,...)
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:89
static void color_printf(int color, const char *fmt,...)
Definition: checkasm.c:286
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
Definition: cpu.h:37
#define AV_CPU_FLAG_POWER8
ISA 2.07.
Definition: cpu.h:59
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:268
void checkasm_update_bench(int iterations, uint64_t cycles)
Definition: checkasm.c:652
static void destroy_func_tree(CheckasmFunc *f)
Definition: checkasm.c:329
common internal and external API header
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
struct CheckasmFuncVersion * next
Definition: checkasm.c:178
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:67
const char * current_test_name
Definition: checkasm.c:199
static void check_cpu_flag(const char *name, int flag)
Definition: checkasm.c:485
static int use_color
Definition: log.c:123
#define have_vfp(flags)
Definition: cpu.h:28
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
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:273
#define AV_CPU_FLAG_SSE2
PIV SSE2 functions.
Definition: cpu.h:34
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:634
#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:63
static void balance_tree(CheckasmFunc **root)
Definition: checkasm.c:444
void checkasm_check_blend(void)
Definition: vf_blend.c:87
const char * name
Definition: opengl_enc.c:103
static void print_cpu_name(void)
Definition: checkasm.c:508
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:54
uint8_t color
Definition: checkasm.c:190