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 && HAVE_GETSTDHANDLE
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_qpel", checkasm_check_hevc_qpel },
120 { "hevc_qpel_uni", checkasm_check_hevc_qpel_uni },
121 { "hevc_qpel_uni_w", checkasm_check_hevc_qpel_uni_w },
122 { "hevc_qpel_bi", checkasm_check_hevc_qpel_bi },
123 { "hevc_qpel_bi_w", checkasm_check_hevc_qpel_bi_w },
124 { "hevc_epel", checkasm_check_hevc_epel },
125 { "hevc_epel_uni", checkasm_check_hevc_epel_uni },
126 { "hevc_epel_uni_w", checkasm_check_hevc_epel_uni_w },
127 { "hevc_epel_bi", checkasm_check_hevc_epel_bi },
128 { "hevc_epel_bi_w", checkasm_check_hevc_epel_bi_w },
129 { "hevc_sao", checkasm_check_hevc_sao },
130 #endif
131 #if CONFIG_HUFFYUV_DECODER
132 { "huffyuvdsp", checkasm_check_huffyuvdsp },
133 #endif
134 #if CONFIG_JPEG2000_DECODER
135 { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
136 #endif
137 #if CONFIG_HUFFYUVDSP
138 { "llviddsp", checkasm_check_llviddsp },
139 #endif
140 #if CONFIG_LLVIDENCDSP
141 { "llviddspenc", checkasm_check_llviddspenc },
142 #endif
143 #if CONFIG_OPUS_DECODER
144 { "opusdsp", checkasm_check_opusdsp },
145 #endif
146 #if CONFIG_PIXBLOCKDSP
147 { "pixblockdsp", checkasm_check_pixblockdsp },
148 #endif
149 #if CONFIG_UTVIDEO_DECODER
150 { "utvideodsp", checkasm_check_utvideodsp },
151 #endif
152 #if CONFIG_V210_DECODER
153 { "v210dec", checkasm_check_v210dec },
154 #endif
155 #if CONFIG_V210_ENCODER
156 { "v210enc", checkasm_check_v210enc },
157 #endif
158 #if CONFIG_VP8DSP
159 { "vp8dsp", checkasm_check_vp8dsp },
160 #endif
161 #if CONFIG_VP9_DECODER
162 { "vp9dsp", checkasm_check_vp9dsp },
163 #endif
164 #if CONFIG_VIDEODSP
165 { "videodsp", checkasm_check_videodsp },
166 #endif
167 #endif
168 #if CONFIG_AVFILTER
169 #if CONFIG_AFIR_FILTER
170 { "af_afir", checkasm_check_afir },
171 #endif
172 #if CONFIG_BLEND_FILTER
173 { "vf_blend", checkasm_check_blend },
174 #endif
175 #if CONFIG_COLORSPACE_FILTER
176 { "vf_colorspace", checkasm_check_colorspace },
177 #endif
178 #if CONFIG_EQ_FILTER
179 { "vf_eq", checkasm_check_vf_eq },
180 #endif
181 #if CONFIG_GBLUR_FILTER
182 { "vf_gblur", checkasm_check_vf_gblur },
183 #endif
184 #if CONFIG_HFLIP_FILTER
185 { "vf_hflip", checkasm_check_vf_hflip },
186 #endif
187 #if CONFIG_NLMEANS_FILTER
188 { "vf_nlmeans", checkasm_check_nlmeans },
189 #endif
190 #if CONFIG_THRESHOLD_FILTER
191 { "vf_threshold", checkasm_check_vf_threshold },
192 #endif
193 #endif
194 #if CONFIG_SWSCALE
195 { "sw_rgb", checkasm_check_sw_rgb },
196 { "sw_scale", checkasm_check_sw_scale },
197 #endif
198 #if CONFIG_AVUTIL
199 { "fixed_dsp", checkasm_check_fixed_dsp },
200 { "float_dsp", checkasm_check_float_dsp },
201 #endif
202 { NULL }
203 };
204
205 /* List of cpu flags to check */
206 static const struct {
207 const char *name;
208 const char *suffix;
209 int flag;
210 } cpus[] = {
211 #if ARCH_AARCH64
212 { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
213 { "NEON", "neon", AV_CPU_FLAG_NEON },
214 #elif ARCH_ARM
215 { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
216 { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
217 { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
218 { "VFP", "vfp", AV_CPU_FLAG_VFP },
219 { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
220 { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
221 { "NEON", "neon", AV_CPU_FLAG_NEON },
222 #elif ARCH_PPC
223 { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
224 { "VSX", "vsx", AV_CPU_FLAG_VSX },
225 { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
226 #elif ARCH_MIPS
227 { "MMI", "mmi", AV_CPU_FLAG_MMI },
228 { "MSA", "msa", AV_CPU_FLAG_MSA },
229 #elif ARCH_X86
230 { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
231 { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
232 { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
233 { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
234 { "SSE", "sse", AV_CPU_FLAG_SSE },
235 { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
236 { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
237 { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
238 { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
239 { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
240 { "AES-NI", "aesni", AV_CPU_FLAG_AESNI },
241 { "AVX", "avx", AV_CPU_FLAG_AVX },
242 { "XOP", "xop", AV_CPU_FLAG_XOP },
243 { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
244 { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
245 { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
246 { "AVX-512", "avx512", AV_CPU_FLAG_AVX512 },
247 #endif
248 { NULL }
249 };
250
251 typedef struct CheckasmFuncVersion {
252 struct CheckasmFuncVersion *next;
253 void *func;
254 int ok;
255 int cpu;
256 CheckasmPerf perf;
257 } CheckasmFuncVersion;
258
259 /* Binary search tree node */
260 typedef struct CheckasmFunc {
261 struct CheckasmFunc *child[2];
262 CheckasmFuncVersion versions;
263 uint8_t color; /* 0 = red, 1 = black */
264 char name[1];
265 } CheckasmFunc;
266
267 /* Internal state */
268 static struct {
269 CheckasmFunc *funcs;
270 CheckasmFunc *current_func;
271 CheckasmFuncVersion *current_func_ver;
272 const char *current_test_name;
273 const char *bench_pattern;
274 int bench_pattern_len;
275 int num_checked;
276 int num_failed;
277
278 /* perf */
279 int nop_time;
280 int sysfd;
281
282 int cpu_flag;
283 const char *cpu_flag_name;
284 const char *test_name;
285 int verbose;
286 } state;
287
288 /* PRNG state */
289 AVLFG checkasm_lfg;
290
291 /* float compare support code */
is_negative(union av_intfloat32 u)292 static int is_negative(union av_intfloat32 u)
293 {
294 return u.i >> 31;
295 }
296
float_near_ulp(float a,float b,unsigned max_ulp)297 int float_near_ulp(float a, float b, unsigned max_ulp)
298 {
299 union av_intfloat32 x, y;
300
301 x.f = a;
302 y.f = b;
303
304 if (is_negative(x) != is_negative(y)) {
305 // handle -0.0 == +0.0
306 return a == b;
307 }
308
309 if (llabs((int64_t)x.i - y.i) <= max_ulp)
310 return 1;
311
312 return 0;
313 }
314
float_near_ulp_array(const float * a,const float * b,unsigned max_ulp,unsigned len)315 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
316 unsigned len)
317 {
318 unsigned i;
319
320 for (i = 0; i < len; i++) {
321 if (!float_near_ulp(a[i], b[i], max_ulp))
322 return 0;
323 }
324 return 1;
325 }
326
float_near_abs_eps(float a,float b,float eps)327 int float_near_abs_eps(float a, float b, float eps)
328 {
329 float abs_diff = fabsf(a - b);
330 if (abs_diff < eps)
331 return 1;
332
333 fprintf(stderr, "test failed comparing %g with %g (abs diff=%g with EPS=%g)\n", a, b, abs_diff, eps);
334
335 return 0;
336 }
337
float_near_abs_eps_array(const float * a,const float * b,float eps,unsigned len)338 int float_near_abs_eps_array(const float *a, const float *b, float eps,
339 unsigned len)
340 {
341 unsigned i;
342
343 for (i = 0; i < len; i++) {
344 if (!float_near_abs_eps(a[i], b[i], eps))
345 return 0;
346 }
347 return 1;
348 }
349
float_near_abs_eps_ulp(float a,float b,float eps,unsigned max_ulp)350 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
351 {
352 return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
353 }
354
float_near_abs_eps_array_ulp(const float * a,const float * b,float eps,unsigned max_ulp,unsigned len)355 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
356 unsigned max_ulp, unsigned len)
357 {
358 unsigned i;
359
360 for (i = 0; i < len; i++) {
361 if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
362 return 0;
363 }
364 return 1;
365 }
366
double_near_abs_eps(double a,double b,double eps)367 int double_near_abs_eps(double a, double b, double eps)
368 {
369 double abs_diff = fabs(a - b);
370
371 return abs_diff < eps;
372 }
373
double_near_abs_eps_array(const double * a,const double * b,double eps,unsigned len)374 int double_near_abs_eps_array(const double *a, const double *b, double eps,
375 unsigned len)
376 {
377 unsigned i;
378
379 for (i = 0; i < len; i++) {
380 if (!double_near_abs_eps(a[i], b[i], eps))
381 return 0;
382 }
383 return 1;
384 }
385
386 /* Print colored text to stderr if the terminal supports it */
color_printf(int color,const char * fmt,...)387 static void color_printf(int color, const char *fmt, ...)
388 {
389 static int use_color = -1;
390 va_list arg;
391
392 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
393 static HANDLE con;
394 static WORD org_attributes;
395
396 if (use_color < 0) {
397 CONSOLE_SCREEN_BUFFER_INFO con_info;
398 con = GetStdHandle(STD_ERROR_HANDLE);
399 if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
400 org_attributes = con_info.wAttributes;
401 use_color = 1;
402 } else
403 use_color = 0;
404 }
405 if (use_color)
406 SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
407 #else
408 if (use_color < 0) {
409 const char *term = getenv("TERM");
410 use_color = term && strcmp(term, "dumb") && isatty(2);
411 }
412 if (use_color)
413 fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
414 #endif
415
416 va_start(arg, fmt);
417 vfprintf(stderr, fmt, arg);
418 va_end(arg);
419
420 if (use_color) {
421 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
422 SetConsoleTextAttribute(con, org_attributes);
423 #else
424 fprintf(stderr, "\x1b[0m");
425 #endif
426 }
427 }
428
429 /* Deallocate a tree */
destroy_func_tree(CheckasmFunc * f)430 static void destroy_func_tree(CheckasmFunc *f)
431 {
432 if (f) {
433 CheckasmFuncVersion *v = f->versions.next;
434 while (v) {
435 CheckasmFuncVersion *next = v->next;
436 free(v);
437 v = next;
438 }
439
440 destroy_func_tree(f->child[0]);
441 destroy_func_tree(f->child[1]);
442 free(f);
443 }
444 }
445
446 /* Allocate a zero-initialized block, clean up and exit on failure */
checkasm_malloc(size_t size)447 static void *checkasm_malloc(size_t size)
448 {
449 void *ptr = calloc(1, size);
450 if (!ptr) {
451 fprintf(stderr, "checkasm: malloc failed\n");
452 destroy_func_tree(state.funcs);
453 exit(1);
454 }
455 return ptr;
456 }
457
458 /* Get the suffix of the specified cpu flag */
cpu_suffix(int cpu)459 static const char *cpu_suffix(int cpu)
460 {
461 int i = FF_ARRAY_ELEMS(cpus);
462
463 while (--i >= 0)
464 if (cpu & cpus[i].flag)
465 return cpus[i].suffix;
466
467 return "c";
468 }
469
cmp_nop(const void * a,const void * b)470 static int cmp_nop(const void *a, const void *b)
471 {
472 return *(const uint16_t*)a - *(const uint16_t*)b;
473 }
474
475 /* Measure the overhead of the timing code (in decicycles) */
measure_nop_time(void)476 static int measure_nop_time(void)
477 {
478 uint16_t nops[10000];
479 int i, nop_sum = 0;
480 av_unused const int sysfd = state.sysfd;
481
482 uint64_t t = 0;
483 for (i = 0; i < 10000; i++) {
484 PERF_START(t);
485 PERF_STOP(t);
486 nops[i] = t;
487 }
488
489 qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
490 for (i = 2500; i < 7500; i++)
491 nop_sum += nops[i];
492
493 return nop_sum / 500;
494 }
495
496 /* Print benchmark results */
print_benchs(CheckasmFunc * f)497 static void print_benchs(CheckasmFunc *f)
498 {
499 if (f) {
500 print_benchs(f->child[0]);
501
502 /* Only print functions with at least one assembly version */
503 if (f->versions.cpu || f->versions.next) {
504 CheckasmFuncVersion *v = &f->versions;
505 do {
506 CheckasmPerf *p = &v->perf;
507 if (p->iterations) {
508 int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
509 printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
510 }
511 } while ((v = v->next));
512 }
513
514 print_benchs(f->child[1]);
515 }
516 }
517
518 /* ASCIIbetical sort except preserving natural order for numbers */
cmp_func_names(const char * a,const char * b)519 static int cmp_func_names(const char *a, const char *b)
520 {
521 const char *start = a;
522 int ascii_diff, digit_diff;
523
524 for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
525 for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
526
527 if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
528 return digit_diff;
529
530 return ascii_diff;
531 }
532
533 /* Perform a tree rotation in the specified direction and return the new root */
rotate_tree(CheckasmFunc * f,int dir)534 static CheckasmFunc *rotate_tree(CheckasmFunc *f, int dir)
535 {
536 CheckasmFunc *r = f->child[dir^1];
537 f->child[dir^1] = r->child[dir];
538 r->child[dir] = f;
539 r->color = f->color;
540 f->color = 0;
541 return r;
542 }
543
544 #define is_red(f) ((f) && !(f)->color)
545
546 /* Balance a left-leaning red-black tree at the specified node */
balance_tree(CheckasmFunc ** root)547 static void balance_tree(CheckasmFunc **root)
548 {
549 CheckasmFunc *f = *root;
550
551 if (is_red(f->child[0]) && is_red(f->child[1])) {
552 f->color ^= 1;
553 f->child[0]->color = f->child[1]->color = 1;
554 }
555
556 if (!is_red(f->child[0]) && is_red(f->child[1]))
557 *root = rotate_tree(f, 0); /* Rotate left */
558 else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
559 *root = rotate_tree(f, 1); /* Rotate right */
560 }
561
562 /* Get a node with the specified name, creating it if it doesn't exist */
get_func(CheckasmFunc ** root,const char * name)563 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
564 {
565 CheckasmFunc *f = *root;
566
567 if (f) {
568 /* Search the tree for a matching node */
569 int cmp = cmp_func_names(name, f->name);
570 if (cmp) {
571 f = get_func(&f->child[cmp > 0], name);
572
573 /* Rebalance the tree on the way up if a new node was inserted */
574 if (!f->versions.func)
575 balance_tree(root);
576 }
577 } else {
578 /* Allocate and insert a new node into the tree */
579 int name_length = strlen(name);
580 f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
581 memcpy(f->name, name, name_length + 1);
582 }
583
584 return f;
585 }
586
587 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
check_cpu_flag(const char * name,int flag)588 static void check_cpu_flag(const char *name, int flag)
589 {
590 int old_cpu_flag = state.cpu_flag;
591
592 flag |= old_cpu_flag;
593 av_force_cpu_flags(-1);
594 state.cpu_flag = flag & av_get_cpu_flags();
595 av_force_cpu_flags(state.cpu_flag);
596
597 if (!flag || state.cpu_flag != old_cpu_flag) {
598 int i;
599
600 state.cpu_flag_name = name;
601 for (i = 0; tests[i].func; i++) {
602 if (state.test_name && strcmp(tests[i].name, state.test_name))
603 continue;
604 state.current_test_name = tests[i].name;
605 tests[i].func();
606 }
607 }
608 }
609
610 /* Print the name of the current CPU flag, but only do it once */
print_cpu_name(void)611 static void print_cpu_name(void)
612 {
613 if (state.cpu_flag_name) {
614 color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
615 state.cpu_flag_name = NULL;
616 }
617 }
618
619 #if CONFIG_LINUX_PERF
bench_init_linux(void)620 static int bench_init_linux(void)
621 {
622 struct perf_event_attr attr = {
623 .type = PERF_TYPE_HARDWARE,
624 .size = sizeof(struct perf_event_attr),
625 .config = PERF_COUNT_HW_CPU_CYCLES,
626 .disabled = 1, // start counting only on demand
627 .exclude_kernel = 1,
628 .exclude_hv = 1,
629 };
630
631 printf("benchmarking with Linux Perf Monitoring API\n");
632
633 state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
634 if (state.sysfd == -1) {
635 perror("syscall");
636 return -1;
637 }
638 return 0;
639 }
640 #endif
641
642 #if !CONFIG_LINUX_PERF
bench_init_ffmpeg(void)643 static int bench_init_ffmpeg(void)
644 {
645 #ifdef AV_READ_TIME
646 printf("benchmarking with native FFmpeg timers\n");
647 return 0;
648 #else
649 fprintf(stderr, "checkasm: --bench is not supported on your system\n");
650 return -1;
651 #endif
652 }
653 #endif
654
bench_init(void)655 static int bench_init(void)
656 {
657 #if CONFIG_LINUX_PERF
658 int ret = bench_init_linux();
659 #else
660 int ret = bench_init_ffmpeg();
661 #endif
662 if (ret < 0)
663 return ret;
664
665 state.nop_time = measure_nop_time();
666 printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
667 return 0;
668 }
669
bench_uninit(void)670 static void bench_uninit(void)
671 {
672 #if CONFIG_LINUX_PERF
673 if (state.sysfd > 0)
674 close(state.sysfd);
675 #endif
676 }
677
main(int argc,char * argv[])678 int main(int argc, char *argv[])
679 {
680 unsigned int seed = av_get_random_seed();
681 int i, ret = 0;
682
683 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
684 if (have_vfp(av_get_cpu_flags()) || have_neon(av_get_cpu_flags()))
685 checkasm_checked_call = checkasm_checked_call_vfp;
686 #endif
687
688 if (!tests[0].func || !cpus[0].flag) {
689 fprintf(stderr, "checkasm: no tests to perform\n");
690 return 0;
691 }
692
693 while (argc > 1) {
694 if (!strncmp(argv[1], "--bench", 7)) {
695 if (bench_init() < 0)
696 return 1;
697 if (argv[1][7] == '=') {
698 state.bench_pattern = argv[1] + 8;
699 state.bench_pattern_len = strlen(state.bench_pattern);
700 } else
701 state.bench_pattern = "";
702 } else if (!strncmp(argv[1], "--test=", 7)) {
703 state.test_name = argv[1] + 7;
704 } else if (!strcmp(argv[1], "--verbose") || !strcmp(argv[1], "-v")) {
705 state.verbose = 1;
706 } else {
707 seed = strtoul(argv[1], NULL, 10);
708 }
709
710 argc--;
711 argv++;
712 }
713
714 fprintf(stderr, "checkasm: using random seed %u\n", seed);
715 av_lfg_init(&checkasm_lfg, seed);
716
717 check_cpu_flag(NULL, 0);
718 for (i = 0; cpus[i].flag; i++)
719 check_cpu_flag(cpus[i].name, cpus[i].flag);
720
721 if (state.num_failed) {
722 fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
723 ret = 1;
724 } else {
725 fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
726 if (state.bench_pattern) {
727 print_benchs(state.funcs);
728 }
729 }
730
731 destroy_func_tree(state.funcs);
732 bench_uninit();
733 return ret;
734 }
735
736 /* Decide whether or not the specified function needs to be tested and
737 * allocate/initialize data structures if needed. Returns a pointer to a
738 * reference function if the function should be tested, otherwise NULL */
checkasm_check_func(void * func,const char * name,...)739 void *checkasm_check_func(void *func, const char *name, ...)
740 {
741 char name_buf[256];
742 void *ref = func;
743 CheckasmFuncVersion *v;
744 int name_length;
745 va_list arg;
746
747 va_start(arg, name);
748 name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
749 va_end(arg);
750
751 if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
752 return NULL;
753
754 state.current_func = get_func(&state.funcs, name_buf);
755 state.funcs->color = 1;
756 v = &state.current_func->versions;
757
758 if (v->func) {
759 CheckasmFuncVersion *prev;
760 do {
761 /* Only test functions that haven't already been tested */
762 if (v->func == func)
763 return NULL;
764
765 if (v->ok)
766 ref = v->func;
767
768 prev = v;
769 } while ((v = v->next));
770
771 v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
772 }
773
774 v->func = func;
775 v->ok = 1;
776 v->cpu = state.cpu_flag;
777 state.current_func_ver = v;
778
779 if (state.cpu_flag)
780 state.num_checked++;
781
782 return ref;
783 }
784
785 /* Decide whether or not the current function needs to be benchmarked */
checkasm_bench_func(void)786 int checkasm_bench_func(void)
787 {
788 return !state.num_failed && state.bench_pattern &&
789 !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
790 }
791
792 /* Indicate that the current test has failed */
checkasm_fail_func(const char * msg,...)793 void checkasm_fail_func(const char *msg, ...)
794 {
795 if (state.current_func_ver->cpu && state.current_func_ver->ok) {
796 va_list arg;
797
798 print_cpu_name();
799 fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
800 va_start(arg, msg);
801 vfprintf(stderr, msg, arg);
802 va_end(arg);
803 fprintf(stderr, ")\n");
804
805 state.current_func_ver->ok = 0;
806 state.num_failed++;
807 }
808 }
809
810 /* Get the benchmark context of the current function */
checkasm_get_perf_context(void)811 CheckasmPerf *checkasm_get_perf_context(void)
812 {
813 CheckasmPerf *perf = &state.current_func_ver->perf;
814 memset(perf, 0, sizeof(*perf));
815 perf->sysfd = state.sysfd;
816 return perf;
817 }
818
819 /* Print the outcome of all tests performed since the last time this function was called */
checkasm_report(const char * name,...)820 void checkasm_report(const char *name, ...)
821 {
822 static int prev_checked, prev_failed, max_length;
823
824 if (state.num_checked > prev_checked) {
825 int pad_length = max_length + 4;
826 va_list arg;
827
828 print_cpu_name();
829 pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
830 va_start(arg, name);
831 pad_length -= vfprintf(stderr, name, arg);
832 va_end(arg);
833 fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
834
835 if (state.num_failed == prev_failed)
836 color_printf(COLOR_GREEN, "OK");
837 else
838 color_printf(COLOR_RED, "FAILED");
839 fprintf(stderr, "]\n");
840
841 prev_checked = state.num_checked;
842 prev_failed = state.num_failed;
843 } else if (!state.cpu_flag) {
844 /* Calculate the amount of padding required to make the output vertically aligned */
845 int length = strlen(state.current_test_name);
846 va_list arg;
847
848 va_start(arg, name);
849 length += vsnprintf(NULL, 0, name, arg);
850 va_end(arg);
851
852 if (length > max_length)
853 max_length = length;
854 }
855 }
856
857 #define DEF_CHECKASM_CHECK_FUNC(type, fmt) \
858 int checkasm_check_##type(const char *const file, const int line, \
859 const type *buf1, ptrdiff_t stride1, \
860 const type *buf2, ptrdiff_t stride2, \
861 const int w, int h, const char *const name) \
862 { \
863 int y = 0; \
864 stride1 /= sizeof(*buf1); \
865 stride2 /= sizeof(*buf2); \
866 for (y = 0; y < h; y++) \
867 if (memcmp(&buf1[y*stride1], &buf2[y*stride2], w*sizeof(*buf1))) \
868 break; \
869 if (y == h) \
870 return 0; \
871 checkasm_fail_func("%s:%d", file, line); \
872 if (!state.verbose) \
873 return 1; \
874 fprintf(stderr, "%s:\n", name); \
875 while (h--) { \
876 for (int x = 0; x < w; x++) \
877 fprintf(stderr, " " fmt, buf1[x]); \
878 fprintf(stderr, " "); \
879 for (int x = 0; x < w; x++) \
880 fprintf(stderr, " " fmt, buf2[x]); \
881 fprintf(stderr, " "); \
882 for (int x = 0; x < w; x++) \
883 fprintf(stderr, "%c", buf1[x] != buf2[x] ? 'x' : '.'); \
884 buf1 += stride1; \
885 buf2 += stride2; \
886 fprintf(stderr, "\n"); \
887 } \
888 return 1; \
889 }
890
891 DEF_CHECKASM_CHECK_FUNC(uint8_t, "%02x")
892 DEF_CHECKASM_CHECK_FUNC(uint16_t, "%04x")
893 DEF_CHECKASM_CHECK_FUNC(int16_t, "%6d")
894 DEF_CHECKASM_CHECK_FUNC(int32_t, "%9d")
895