checkasm: Read the unsigned value as it should
[libav.git] / tests / checkasm / checkasm.c
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 Libav.
7 *
8 * Libav 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 * Libav 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 Libav; 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_BSWAPDSP
68 { "bswapdsp", checkasm_check_bswapdsp },
69 #endif
70 #if CONFIG_DCA_DECODER
71 { "dcadsp", checkasm_check_dcadsp },
72 { "synth_filter", checkasm_check_synth_filter },
73 #endif
74 #if CONFIG_FMTCONVERT
75 { "fmtconvert", checkasm_check_fmtconvert },
76 #endif
77 #if CONFIG_H264DSP
78 { "h264dsp", checkasm_check_h264dsp },
79 #endif
80 #if CONFIG_H264PRED
81 { "h264pred", checkasm_check_h264pred },
82 #endif
83 #if CONFIG_H264QPEL
84 { "h264qpel", checkasm_check_h264qpel },
85 #endif
86 #if CONFIG_HEVC_DECODER
87 { "hevc_mc", checkasm_check_hevc_mc },
88 { "hevc_idct", checkasm_check_hevc_idct },
89 #endif
90 #if CONFIG_V210_ENCODER
91 { "v210enc", checkasm_check_v210enc },
92 #endif
93 #if CONFIG_VP8DSP
94 { "vp8dsp", checkasm_check_vp8dsp },
95 #endif
96 #if CONFIG_VP9_DECODER
97 { "vp9dsp", checkasm_check_vp9dsp },
98 #endif
99 { NULL }
100 };
101
102 /* List of cpu flags to check */
103 static const struct {
104 const char *name;
105 const char *suffix;
106 int flag;
107 } cpus[] = {
108 #if ARCH_AARCH64
109 { "ARMV8", "armv8", AV_CPU_FLAG_ARMV8 },
110 { "NEON", "neon", AV_CPU_FLAG_NEON },
111 #elif ARCH_ARM
112 { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
113 { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },
114 { "ARMV6T2", "armv6t2", AV_CPU_FLAG_ARMV6T2 },
115 { "VFP", "vfp", AV_CPU_FLAG_VFP },
116 { "VFP_VM", "vfp_vm", AV_CPU_FLAG_VFP_VM },
117 { "VFPV3", "vfp3", AV_CPU_FLAG_VFPV3 },
118 { "NEON", "neon", AV_CPU_FLAG_NEON },
119 #elif ARCH_PPC
120 { "ALTIVEC", "altivec", AV_CPU_FLAG_ALTIVEC },
121 { "VSX", "vsx", AV_CPU_FLAG_VSX },
122 { "POWER8", "power8", AV_CPU_FLAG_POWER8 },
123 #elif ARCH_X86
124 { "MMX", "mmx", AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
125 { "MMXEXT", "mmxext", AV_CPU_FLAG_MMXEXT },
126 { "3DNOW", "3dnow", AV_CPU_FLAG_3DNOW },
127 { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
128 { "SSE", "sse", AV_CPU_FLAG_SSE },
129 { "SSE2", "sse2", AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
130 { "SSE3", "sse3", AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
131 { "SSSE3", "ssse3", AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
132 { "SSE4.1", "sse4", AV_CPU_FLAG_SSE4 },
133 { "SSE4.2", "sse42", AV_CPU_FLAG_SSE42 },
134 { "AVX", "avx", AV_CPU_FLAG_AVX },
135 { "XOP", "xop", AV_CPU_FLAG_XOP },
136 { "FMA3", "fma3", AV_CPU_FLAG_FMA3 },
137 { "FMA4", "fma4", AV_CPU_FLAG_FMA4 },
138 { "AVX2", "avx2", AV_CPU_FLAG_AVX2 },
139 #endif
140 { NULL }
141 };
142
143 typedef struct CheckasmFuncVersion {
144 struct CheckasmFuncVersion *next;
145 void *func;
146 int ok;
147 int cpu;
148 int iterations;
149 uint64_t cycles;
150 } CheckasmFuncVersion;
151
152 /* Binary search tree node */
153 typedef struct CheckasmFunc {
154 struct CheckasmFunc *child[2];
155 CheckasmFuncVersion versions;
156 uint8_t color; /* 0 = red, 1 = black */
157 char name[1];
158 } CheckasmFunc;
159
160 /* Internal state */
161 static struct {
162 CheckasmFunc *funcs;
163 CheckasmFunc *current_func;
164 CheckasmFuncVersion *current_func_ver;
165 const char *current_test_name;
166 const char *bench_pattern;
167 int bench_pattern_len;
168 int num_checked;
169 int num_failed;
170 int nop_time;
171 int cpu_flag;
172 const char *cpu_flag_name;
173 } state;
174
175 /* PRNG state */
176 AVLFG checkasm_lfg;
177
178 /* float compare support code */
179 static int is_negative(union av_intfloat32 u)
180 {
181 return u.i >> 31;
182 }
183
184 int float_near_ulp(float a, float b, unsigned max_ulp)
185 {
186 union av_intfloat32 x, y;
187
188 x.f = a;
189 y.f = b;
190
191 if (is_negative(x) != is_negative(y)) {
192 // handle -0.0 == +0.0
193 return a == b;
194 }
195
196 if (llabs((int64_t)x.i - y.i) <= max_ulp)
197 return 1;
198
199 return 0;
200 }
201
202 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
203 unsigned len)
204 {
205 unsigned i;
206
207 for (i = 0; i < len; i++) {
208 if (!float_near_ulp(a[i], b[i], max_ulp))
209 return 0;
210 }
211 return 1;
212 }
213
214 int float_near_abs_eps(float a, float b, float eps)
215 {
216 float abs_diff = fabsf(a - b);
217
218 return abs_diff < eps;
219 }
220
221 int float_near_abs_eps_array(const float *a, const float *b, float eps,
222 unsigned len)
223 {
224 unsigned i;
225
226 for (i = 0; i < len; i++) {
227 if (!float_near_abs_eps(a[i], b[i], eps))
228 return 0;
229 }
230 return 1;
231 }
232
233 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
234 {
235 return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
236 }
237
238 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
239 unsigned max_ulp, unsigned len)
240 {
241 unsigned i;
242
243 for (i = 0; i < len; i++) {
244 if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
245 return 0;
246 }
247 return 1;
248 }
249
250 /* Print colored text to stderr if the terminal supports it */
251 static void color_printf(int color, const char *fmt, ...)
252 {
253 static int use_color = -1;
254 va_list arg;
255
256 #if HAVE_SETCONSOLETEXTATTRIBUTE
257 static HANDLE con;
258 static WORD org_attributes;
259
260 if (use_color < 0) {
261 CONSOLE_SCREEN_BUFFER_INFO con_info;
262 con = GetStdHandle(STD_ERROR_HANDLE);
263 if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
264 org_attributes = con_info.wAttributes;
265 use_color = 1;
266 } else
267 use_color = 0;
268 }
269 if (use_color)
270 SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
271 #else
272 if (use_color < 0) {
273 const char *term = getenv("TERM");
274 use_color = term && strcmp(term, "dumb") && isatty(2);
275 }
276 if (use_color)
277 fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
278 #endif
279
280 va_start(arg, fmt);
281 vfprintf(stderr, fmt, arg);
282 va_end(arg);
283
284 if (use_color) {
285 #if HAVE_SETCONSOLETEXTATTRIBUTE
286 SetConsoleTextAttribute(con, org_attributes);
287 #else
288 fprintf(stderr, "\x1b[0m");
289 #endif
290 }
291 }
292
293 /* Deallocate a tree */
294 static void destroy_func_tree(CheckasmFunc *f)
295 {
296 if (f) {
297 CheckasmFuncVersion *v = f->versions.next;
298 while (v) {
299 CheckasmFuncVersion *next = v->next;
300 free(v);
301 v = next;
302 }
303
304 destroy_func_tree(f->child[0]);
305 destroy_func_tree(f->child[1]);
306 free(f);
307 }
308 }
309
310 /* Allocate a zero-initialized block, clean up and exit on failure */
311 static void *checkasm_malloc(size_t size)
312 {
313 void *ptr = calloc(1, size);
314 if (!ptr) {
315 fprintf(stderr, "checkasm: malloc failed\n");
316 destroy_func_tree(state.funcs);
317 exit(1);
318 }
319 return ptr;
320 }
321
322 /* Get the suffix of the specified cpu flag */
323 static const char *cpu_suffix(int cpu)
324 {
325 int i = FF_ARRAY_ELEMS(cpus);
326
327 while (--i >= 0)
328 if (cpu & cpus[i].flag)
329 return cpus[i].suffix;
330
331 return "c";
332 }
333
334 #ifdef AV_READ_TIME
335 static int cmp_nop(const void *a, const void *b)
336 {
337 return *(const uint16_t*)a - *(const uint16_t*)b;
338 }
339
340 /* Measure the overhead of the timing code (in decicycles) */
341 static int measure_nop_time(void)
342 {
343 uint16_t nops[10000];
344 int i, nop_sum = 0;
345
346 for (i = 0; i < 10000; i++) {
347 uint64_t t = AV_READ_TIME();
348 nops[i] = AV_READ_TIME() - t;
349 }
350
351 qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
352 for (i = 2500; i < 7500; i++)
353 nop_sum += nops[i];
354
355 return nop_sum / 500;
356 }
357
358 /* Print benchmark results */
359 static void print_benchs(CheckasmFunc *f)
360 {
361 if (f) {
362 print_benchs(f->child[0]);
363
364 /* Only print functions with at least one assembly version */
365 if (f->versions.cpu || f->versions.next) {
366 CheckasmFuncVersion *v = &f->versions;
367 do {
368 if (v->iterations) {
369 int decicycles = (10*v->cycles/v->iterations - state.nop_time) / 4;
370 printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
371 }
372 } while ((v = v->next));
373 }
374
375 print_benchs(f->child[1]);
376 }
377 }
378 #endif
379
380 /* ASCIIbetical sort except preserving natural order for numbers */
381 static int cmp_func_names(const char *a, const char *b)
382 {
383 const char *start = a;
384 int ascii_diff, digit_diff;
385
386 for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
387 for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
388
389 if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
390 return digit_diff;
391
392 return ascii_diff;
393 }
394
395 /* Perform a tree rotation in the specified direction and return the new root */
396 static CheckasmFunc *rotate_tree(CheckasmFunc *f, int dir)
397 {
398 CheckasmFunc *r = f->child[dir^1];
399 f->child[dir^1] = r->child[dir];
400 r->child[dir] = f;
401 r->color = f->color;
402 f->color = 0;
403 return r;
404 }
405
406 #define is_red(f) ((f) && !(f)->color)
407
408 /* Balance a left-leaning red-black tree at the specified node */
409 static void balance_tree(CheckasmFunc **root)
410 {
411 CheckasmFunc *f = *root;
412
413 if (is_red(f->child[0]) && is_red(f->child[1])) {
414 f->color ^= 1;
415 f->child[0]->color = f->child[1]->color = 1;
416 }
417
418 if (!is_red(f->child[0]) && is_red(f->child[1]))
419 *root = rotate_tree(f, 0); /* Rotate left */
420 else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
421 *root = rotate_tree(f, 1); /* Rotate right */
422 }
423
424 /* Get a node with the specified name, creating it if it doesn't exist */
425 static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
426 {
427 CheckasmFunc *f = *root;
428
429 if (f) {
430 /* Search the tree for a matching node */
431 int cmp = cmp_func_names(name, f->name);
432 if (cmp) {
433 f = get_func(&f->child[cmp > 0], name);
434
435 /* Rebalance the tree on the way up if a new node was inserted */
436 if (!f->versions.func)
437 balance_tree(root);
438 }
439 } else {
440 /* Allocate and insert a new node into the tree */
441 int name_length = strlen(name);
442 f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
443 memcpy(f->name, name, name_length + 1);
444 }
445
446 return f;
447 }
448
449 /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
450 static void check_cpu_flag(const char *name, int flag)
451 {
452 int old_cpu_flag = state.cpu_flag;
453
454 flag |= old_cpu_flag;
455 av_set_cpu_flags_mask(flag);
456 state.cpu_flag = av_get_cpu_flags();
457
458 if (!flag || state.cpu_flag != old_cpu_flag) {
459 int i;
460
461 state.cpu_flag_name = name;
462 for (i = 0; tests[i].func; i++) {
463 state.current_test_name = tests[i].name;
464 tests[i].func();
465 }
466 }
467 }
468
469 /* Print the name of the current CPU flag, but only do it once */
470 static void print_cpu_name(void)
471 {
472 if (state.cpu_flag_name) {
473 color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
474 state.cpu_flag_name = NULL;
475 }
476 }
477
478 int main(int argc, char *argv[])
479 {
480 unsigned int seed;
481 int i, ret = 0;
482
483 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
484 if (have_vfp(av_get_cpu_flags()) || have_neon(av_get_cpu_flags()))
485 checkasm_checked_call = checkasm_checked_call_vfp;
486 #endif
487
488 if (!tests[0].func || !cpus[0].flag) {
489 fprintf(stderr, "checkasm: no tests to perform\n");
490 return 0;
491 }
492
493 if (argc > 1 && !strncmp(argv[1], "--bench", 7)) {
494 #ifndef AV_READ_TIME
495 fprintf(stderr, "checkasm: --bench is not supported on your system\n");
496 return 1;
497 #endif
498 if (argv[1][7] == '=') {
499 state.bench_pattern = argv[1] + 8;
500 state.bench_pattern_len = strlen(state.bench_pattern);
501 } else
502 state.bench_pattern = "";
503
504 argc--;
505 argv++;
506 }
507
508 seed = (argc > 1) ? strtoul(argv[1], NULL, 10) : av_get_random_seed();
509 fprintf(stderr, "checkasm: using random seed %u\n", seed);
510 av_lfg_init(&checkasm_lfg, seed);
511
512 check_cpu_flag(NULL, 0);
513 for (i = 0; cpus[i].flag; i++)
514 check_cpu_flag(cpus[i].name, cpus[i].flag);
515
516 if (state.num_failed) {
517 fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
518 ret = 1;
519 } else {
520 fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
521 #ifdef AV_READ_TIME
522 if (state.bench_pattern) {
523 state.nop_time = measure_nop_time();
524 printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
525 print_benchs(state.funcs);
526 }
527 #endif
528 }
529
530 destroy_func_tree(state.funcs);
531 return ret;
532 }
533
534 /* Decide whether or not the specified function needs to be tested and
535 * allocate/initialize data structures if needed. Returns a pointer to a
536 * reference function if the function should be tested, otherwise NULL */
537 void *checkasm_check_func(void *func, const char *name, ...)
538 {
539 char name_buf[256];
540 void *ref = func;
541 CheckasmFuncVersion *v;
542 int name_length;
543 va_list arg;
544
545 va_start(arg, name);
546 name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
547 va_end(arg);
548
549 if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
550 return NULL;
551
552 state.current_func = get_func(&state.funcs, name_buf);
553 state.funcs->color = 1;
554 v = &state.current_func->versions;
555
556 if (v->func) {
557 CheckasmFuncVersion *prev;
558 do {
559 /* Only test functions that haven't already been tested */
560 if (v->func == func)
561 return NULL;
562
563 if (v->ok)
564 ref = v->func;
565
566 prev = v;
567 } while ((v = v->next));
568
569 v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
570 }
571
572 v->func = func;
573 v->ok = 1;
574 v->cpu = state.cpu_flag;
575 state.current_func_ver = v;
576
577 if (state.cpu_flag)
578 state.num_checked++;
579
580 return ref;
581 }
582
583 /* Decide whether or not the current function needs to be benchmarked */
584 int checkasm_bench_func(void)
585 {
586 return !state.num_failed && state.bench_pattern &&
587 !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
588 }
589
590 /* Indicate that the current test has failed */
591 void checkasm_fail_func(const char *msg, ...)
592 {
593 if (state.current_func_ver->cpu && state.current_func_ver->ok) {
594 va_list arg;
595
596 print_cpu_name();
597 fprintf(stderr, " %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
598 va_start(arg, msg);
599 vfprintf(stderr, msg, arg);
600 va_end(arg);
601 fprintf(stderr, ")\n");
602
603 state.current_func_ver->ok = 0;
604 state.num_failed++;
605 }
606 }
607
608 /* Update benchmark results of the current function */
609 void checkasm_update_bench(int iterations, uint64_t cycles)
610 {
611 state.current_func_ver->iterations += iterations;
612 state.current_func_ver->cycles += cycles;
613 }
614
615 /* Print the outcome of all tests performed since the last time this function was called */
616 void checkasm_report(const char *name, ...)
617 {
618 static int prev_checked, prev_failed, max_length;
619
620 if (state.num_checked > prev_checked) {
621 int pad_length = max_length + 4;
622 va_list arg;
623
624 print_cpu_name();
625 pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
626 va_start(arg, name);
627 pad_length -= vfprintf(stderr, name, arg);
628 va_end(arg);
629 fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
630
631 if (state.num_failed == prev_failed)
632 color_printf(COLOR_GREEN, "OK");
633 else
634 color_printf(COLOR_RED, "FAILED");
635 fprintf(stderr, "]\n");
636
637 prev_checked = state.num_checked;
638 prev_failed = state.num_failed;
639 } else if (!state.cpu_flag) {
640 /* Calculate the amount of padding required to make the output vertically aligned */
641 int length = strlen(state.current_test_name);
642 va_list arg;
643
644 va_start(arg, name);
645 length += vsnprintf(NULL, 0, name, arg);
646 va_end(arg);
647
648 if (length > max_length)
649 max_length = length;
650 }
651 }