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#include "config_components.h"
25
26#if CONFIG_LINUX_PERF
27# ifndef _GNU_SOURCE
28#  define _GNU_SOURCE // for syscall (performance monitoring API)
29# endif
30#endif
31
32#include <stdarg.h>
33#include <stdio.h>
34#include <stdlib.h>
35#include <string.h>
36#include "checkasm.h"
37#include "libavutil/common.h"
38#include "libavutil/cpu.h"
39#include "libavutil/intfloat.h"
40#include "libavutil/random_seed.h"
41
42#if HAVE_IO_H
43#include <io.h>
44#endif
45
46#if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
47#include <windows.h>
48#define COLOR_RED    FOREGROUND_RED
49#define COLOR_GREEN  FOREGROUND_GREEN
50#define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
51#else
52#define COLOR_RED    1
53#define COLOR_GREEN  2
54#define COLOR_YELLOW 3
55#endif
56
57#if HAVE_UNISTD_H
58#include <unistd.h>
59#endif
60
61#if !HAVE_ISATTY
62#define isatty(fd) 1
63#endif
64
65#if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
66#include "libavutil/arm/cpu.h"
67
68void (*checkasm_checked_call)(void *func, int dummy, ...) = checkasm_checked_call_novfp;
69#endif
70
71/* List of tests to invoke */
72static const struct {
73    const char *name;
74    void (*func)(void);
75} tests[] = {
76#if CONFIG_AVCODEC
77    #if CONFIG_AAC_DECODER
78        { "aacpsdsp", checkasm_check_aacpsdsp },
79        { "sbrdsp",   checkasm_check_sbrdsp },
80    #endif
81    #if CONFIG_ALAC_DECODER
82        { "alacdsp", checkasm_check_alacdsp },
83    #endif
84    #if CONFIG_AUDIODSP
85        { "audiodsp", checkasm_check_audiodsp },
86    #endif
87    #if CONFIG_BLOCKDSP
88        { "blockdsp", checkasm_check_blockdsp },
89    #endif
90    #if CONFIG_BSWAPDSP
91        { "bswapdsp", checkasm_check_bswapdsp },
92    #endif
93    #if CONFIG_DCA_DECODER
94        { "synth_filter", checkasm_check_synth_filter },
95    #endif
96    #if CONFIG_EXR_DECODER
97        { "exrdsp", checkasm_check_exrdsp },
98    #endif
99    #if CONFIG_FLACDSP
100        { "flacdsp", checkasm_check_flacdsp },
101    #endif
102    #if CONFIG_FMTCONVERT
103        { "fmtconvert", checkasm_check_fmtconvert },
104    #endif
105    #if CONFIG_G722DSP
106        { "g722dsp", checkasm_check_g722dsp },
107    #endif
108    #if CONFIG_H264DSP
109        { "h264dsp", checkasm_check_h264dsp },
110    #endif
111    #if CONFIG_H264PRED
112        { "h264pred", checkasm_check_h264pred },
113    #endif
114    #if CONFIG_H264QPEL
115        { "h264qpel", checkasm_check_h264qpel },
116    #endif
117    #if CONFIG_HEVC_DECODER
118        { "hevc_add_res", checkasm_check_hevc_add_res },
119        { "hevc_idct", checkasm_check_hevc_idct },
120        { "hevc_pel", checkasm_check_hevc_pel },
121        { "hevc_sao", checkasm_check_hevc_sao },
122    #endif
123    #if CONFIG_HUFFYUV_DECODER
124        { "huffyuvdsp", checkasm_check_huffyuvdsp },
125    #endif
126    #if CONFIG_IDCTDSP
127        { "idctdsp", checkasm_check_idctdsp },
128    #endif
129    #if CONFIG_JPEG2000_DECODER
130        { "jpeg2000dsp", checkasm_check_jpeg2000dsp },
131    #endif
132    #if CONFIG_HUFFYUVDSP
133        { "llviddsp", checkasm_check_llviddsp },
134    #endif
135    #if CONFIG_LLVIDENCDSP
136        { "llviddspenc", checkasm_check_llviddspenc },
137    #endif
138    #if CONFIG_ME_CMP
139        { "motion", checkasm_check_motion },
140    #endif
141    #if CONFIG_OPUS_DECODER
142        { "opusdsp", checkasm_check_opusdsp },
143    #endif
144    #if CONFIG_PIXBLOCKDSP
145        { "pixblockdsp", checkasm_check_pixblockdsp },
146    #endif
147    #if CONFIG_UTVIDEO_DECODER
148        { "utvideodsp", checkasm_check_utvideodsp },
149    #endif
150    #if CONFIG_V210_DECODER
151        { "v210dec", checkasm_check_v210dec },
152    #endif
153    #if CONFIG_V210_ENCODER
154        { "v210enc", checkasm_check_v210enc },
155    #endif
156    #if CONFIG_VC1DSP
157        { "vc1dsp", checkasm_check_vc1dsp },
158    #endif
159    #if CONFIG_VP8DSP
160        { "vp8dsp", checkasm_check_vp8dsp },
161    #endif
162    #if CONFIG_VP9_DECODER
163        { "vp9dsp", checkasm_check_vp9dsp },
164    #endif
165    #if CONFIG_VIDEODSP
166        { "videodsp", checkasm_check_videodsp },
167    #endif
168#endif
169#if CONFIG_AVFILTER
170    #if CONFIG_AFIR_FILTER
171        { "af_afir", checkasm_check_afir },
172    #endif
173    #if CONFIG_BLEND_FILTER
174        { "vf_blend", checkasm_check_blend },
175    #endif
176    #if CONFIG_COLORSPACE_FILTER
177        { "vf_colorspace", checkasm_check_colorspace },
178    #endif
179    #if CONFIG_EQ_FILTER
180        { "vf_eq", checkasm_check_vf_eq },
181    #endif
182    #if CONFIG_GBLUR_FILTER
183        { "vf_gblur", checkasm_check_vf_gblur },
184    #endif
185    #if CONFIG_HFLIP_FILTER
186        { "vf_hflip", checkasm_check_vf_hflip },
187    #endif
188    #if CONFIG_NLMEANS_FILTER
189        { "vf_nlmeans", checkasm_check_nlmeans },
190    #endif
191    #if CONFIG_THRESHOLD_FILTER
192        { "vf_threshold", checkasm_check_vf_threshold },
193    #endif
194#endif
195#if CONFIG_SWSCALE
196    { "sw_gbrp", checkasm_check_sw_gbrp },
197    { "sw_rgb", checkasm_check_sw_rgb },
198    { "sw_scale", checkasm_check_sw_scale },
199#endif
200#if CONFIG_AVUTIL
201        { "fixed_dsp", checkasm_check_fixed_dsp },
202        { "float_dsp", checkasm_check_float_dsp },
203        { "av_tx",     checkasm_check_av_tx },
204#endif
205    { NULL }
206};
207
208/* List of cpu flags to check */
209static const struct {
210    const char *name;
211    const char *suffix;
212    int flag;
213} cpus[] = {
214#if   ARCH_AARCH64
215    { "ARMV8",    "armv8",    AV_CPU_FLAG_ARMV8 },
216    { "NEON",     "neon",     AV_CPU_FLAG_NEON },
217#elif ARCH_ARM
218    { "ARMV5TE",  "armv5te",  AV_CPU_FLAG_ARMV5TE },
219    { "ARMV6",    "armv6",    AV_CPU_FLAG_ARMV6 },
220    { "ARMV6T2",  "armv6t2",  AV_CPU_FLAG_ARMV6T2 },
221    { "VFP",      "vfp",      AV_CPU_FLAG_VFP },
222    { "VFP_VM",   "vfp_vm",   AV_CPU_FLAG_VFP_VM },
223    { "VFPV3",    "vfp3",     AV_CPU_FLAG_VFPV3 },
224    { "NEON",     "neon",     AV_CPU_FLAG_NEON },
225#elif ARCH_PPC
226    { "ALTIVEC",  "altivec",  AV_CPU_FLAG_ALTIVEC },
227    { "VSX",      "vsx",      AV_CPU_FLAG_VSX },
228    { "POWER8",   "power8",   AV_CPU_FLAG_POWER8 },
229#elif ARCH_MIPS
230    { "MMI",      "mmi",      AV_CPU_FLAG_MMI },
231    { "MSA",      "msa",      AV_CPU_FLAG_MSA },
232#elif ARCH_X86
233    { "MMX",        "mmx",       AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
234    { "MMXEXT",     "mmxext",    AV_CPU_FLAG_MMXEXT },
235    { "3DNOW",      "3dnow",     AV_CPU_FLAG_3DNOW },
236    { "3DNOWEXT",   "3dnowext",  AV_CPU_FLAG_3DNOWEXT },
237    { "SSE",        "sse",       AV_CPU_FLAG_SSE },
238    { "SSE2",       "sse2",      AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
239    { "SSE3",       "sse3",      AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
240    { "SSSE3",      "ssse3",     AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
241    { "SSE4.1",     "sse4",      AV_CPU_FLAG_SSE4 },
242    { "SSE4.2",     "sse42",     AV_CPU_FLAG_SSE42 },
243    { "AES-NI",     "aesni",     AV_CPU_FLAG_AESNI },
244    { "AVX",        "avx",       AV_CPU_FLAG_AVX },
245    { "XOP",        "xop",       AV_CPU_FLAG_XOP },
246    { "FMA3",       "fma3",      AV_CPU_FLAG_FMA3 },
247    { "FMA4",       "fma4",      AV_CPU_FLAG_FMA4 },
248    { "AVX2",       "avx2",      AV_CPU_FLAG_AVX2 },
249    { "AVX-512",    "avx512",    AV_CPU_FLAG_AVX512 },
250    { "AVX-512ICL", "avx512icl", AV_CPU_FLAG_AVX512ICL },
251#elif ARCH_LOONGARCH
252    { "LSX",      "lsx",      AV_CPU_FLAG_LSX },
253    { "LASX",     "lasx",     AV_CPU_FLAG_LASX },
254#endif
255    { NULL }
256};
257
258typedef struct CheckasmFuncVersion {
259    struct CheckasmFuncVersion *next;
260    void *func;
261    int ok;
262    int cpu;
263    CheckasmPerf perf;
264} CheckasmFuncVersion;
265
266/* Binary search tree node */
267typedef struct CheckasmFunc {
268    struct CheckasmFunc *child[2];
269    CheckasmFuncVersion versions;
270    uint8_t color; /* 0 = red, 1 = black */
271    char name[1];
272} CheckasmFunc;
273
274/* Internal state */
275static struct {
276    CheckasmFunc *funcs;
277    CheckasmFunc *current_func;
278    CheckasmFuncVersion *current_func_ver;
279    const char *current_test_name;
280    const char *bench_pattern;
281    int bench_pattern_len;
282    int num_checked;
283    int num_failed;
284
285    /* perf */
286    int nop_time;
287    int sysfd;
288
289    int cpu_flag;
290    const char *cpu_flag_name;
291    const char *test_name;
292    int verbose;
293} state;
294
295/* PRNG state */
296AVLFG checkasm_lfg;
297
298/* float compare support code */
299static int is_negative(union av_intfloat32 u)
300{
301    return u.i >> 31;
302}
303
304int float_near_ulp(float a, float b, unsigned max_ulp)
305{
306    union av_intfloat32 x, y;
307
308    x.f = a;
309    y.f = b;
310
311    if (is_negative(x) != is_negative(y)) {
312        // handle -0.0 == +0.0
313        return a == b;
314    }
315
316    if (llabs((int64_t)x.i - y.i) <= max_ulp)
317        return 1;
318
319    return 0;
320}
321
322int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
323                         unsigned len)
324{
325    unsigned i;
326
327    for (i = 0; i < len; i++) {
328        if (!float_near_ulp(a[i], b[i], max_ulp))
329            return 0;
330    }
331    return 1;
332}
333
334int float_near_abs_eps(float a, float b, float eps)
335{
336    float abs_diff = fabsf(a - b);
337    if (abs_diff < eps)
338        return 1;
339
340    fprintf(stderr, "test failed comparing %g with %g (abs diff=%g with EPS=%g)\n", a, b, abs_diff, eps);
341
342    return 0;
343}
344
345int float_near_abs_eps_array(const float *a, const float *b, float eps,
346                         unsigned len)
347{
348    unsigned i;
349
350    for (i = 0; i < len; i++) {
351        if (!float_near_abs_eps(a[i], b[i], eps))
352            return 0;
353    }
354    return 1;
355}
356
357int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
358{
359    return float_near_ulp(a, b, max_ulp) || float_near_abs_eps(a, b, eps);
360}
361
362int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
363                         unsigned max_ulp, unsigned len)
364{
365    unsigned i;
366
367    for (i = 0; i < len; i++) {
368        if (!float_near_abs_eps_ulp(a[i], b[i], eps, max_ulp))
369            return 0;
370    }
371    return 1;
372}
373
374int double_near_abs_eps(double a, double b, double eps)
375{
376    double abs_diff = fabs(a - b);
377
378    return abs_diff < eps;
379}
380
381int double_near_abs_eps_array(const double *a, const double *b, double eps,
382                              unsigned len)
383{
384    unsigned i;
385
386    for (i = 0; i < len; i++) {
387        if (!double_near_abs_eps(a[i], b[i], eps))
388            return 0;
389    }
390    return 1;
391}
392
393/* Print colored text to stderr if the terminal supports it */
394static void color_printf(int color, const char *fmt, ...)
395{
396    static int use_color = -1;
397    va_list arg;
398
399#if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
400    static HANDLE con;
401    static WORD org_attributes;
402
403    if (use_color < 0) {
404        CONSOLE_SCREEN_BUFFER_INFO con_info;
405        con = GetStdHandle(STD_ERROR_HANDLE);
406        if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
407            org_attributes = con_info.wAttributes;
408            use_color = 1;
409        } else
410            use_color = 0;
411    }
412    if (use_color)
413        SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
414#else
415    if (use_color < 0) {
416        const char *term = getenv("TERM");
417        use_color = term && strcmp(term, "dumb") && isatty(2);
418    }
419    if (use_color)
420        fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
421#endif
422
423    va_start(arg, fmt);
424    vfprintf(stderr, fmt, arg);
425    va_end(arg);
426
427    if (use_color) {
428#if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
429        SetConsoleTextAttribute(con, org_attributes);
430#else
431        fprintf(stderr, "\x1b[0m");
432#endif
433    }
434}
435
436/* Deallocate a tree */
437static void destroy_func_tree(CheckasmFunc *f)
438{
439    if (f) {
440        CheckasmFuncVersion *v = f->versions.next;
441        while (v) {
442            CheckasmFuncVersion *next = v->next;
443            free(v);
444            v = next;
445        }
446
447        destroy_func_tree(f->child[0]);
448        destroy_func_tree(f->child[1]);
449        free(f);
450    }
451}
452
453/* Allocate a zero-initialized block, clean up and exit on failure */
454static void *checkasm_malloc(size_t size)
455{
456    void *ptr = calloc(1, size);
457    if (!ptr) {
458        fprintf(stderr, "checkasm: malloc failed\n");
459        destroy_func_tree(state.funcs);
460        exit(1);
461    }
462    return ptr;
463}
464
465/* Get the suffix of the specified cpu flag */
466static const char *cpu_suffix(int cpu)
467{
468    int i = FF_ARRAY_ELEMS(cpus);
469
470    while (--i >= 0)
471        if (cpu & cpus[i].flag)
472            return cpus[i].suffix;
473
474    return "c";
475}
476
477static int cmp_nop(const void *a, const void *b)
478{
479    return *(const uint16_t*)a - *(const uint16_t*)b;
480}
481
482/* Measure the overhead of the timing code (in decicycles) */
483static int measure_nop_time(void)
484{
485    uint16_t nops[10000];
486    int i, nop_sum = 0;
487    av_unused const int sysfd = state.sysfd;
488
489    uint64_t t = 0;
490    for (i = 0; i < 10000; i++) {
491        PERF_START(t);
492        PERF_STOP(t);
493        nops[i] = t;
494    }
495
496    qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
497    for (i = 2500; i < 7500; i++)
498        nop_sum += nops[i];
499
500    return nop_sum / 500;
501}
502
503/* Print benchmark results */
504static void print_benchs(CheckasmFunc *f)
505{
506    if (f) {
507        print_benchs(f->child[0]);
508
509        /* Only print functions with at least one assembly version */
510        if (f->versions.cpu || f->versions.next) {
511            CheckasmFuncVersion *v = &f->versions;
512            do {
513                CheckasmPerf *p = &v->perf;
514                if (p->iterations) {
515                    int decicycles = (10*p->cycles/p->iterations - state.nop_time) / 4;
516                    printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
517                }
518            } while ((v = v->next));
519        }
520
521        print_benchs(f->child[1]);
522    }
523}
524
525/* ASCIIbetical sort except preserving natural order for numbers */
526static int cmp_func_names(const char *a, const char *b)
527{
528    const char *start = a;
529    int ascii_diff, digit_diff;
530
531    for (; !(ascii_diff = *(const unsigned char*)a - *(const unsigned char*)b) && *a; a++, b++);
532    for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
533
534    if (a > start && av_isdigit(a[-1]) && (digit_diff = av_isdigit(*a) - av_isdigit(*b)))
535        return digit_diff;
536
537    return ascii_diff;
538}
539
540/* Perform a tree rotation in the specified direction and return the new root */
541static CheckasmFunc *rotate_tree(CheckasmFunc *f, int dir)
542{
543    CheckasmFunc *r = f->child[dir^1];
544    f->child[dir^1] = r->child[dir];
545    r->child[dir] = f;
546    r->color = f->color;
547    f->color = 0;
548    return r;
549}
550
551#define is_red(f) ((f) && !(f)->color)
552
553/* Balance a left-leaning red-black tree at the specified node */
554static void balance_tree(CheckasmFunc **root)
555{
556    CheckasmFunc *f = *root;
557
558    if (is_red(f->child[0]) && is_red(f->child[1])) {
559        f->color ^= 1;
560        f->child[0]->color = f->child[1]->color = 1;
561    }
562
563    if (!is_red(f->child[0]) && is_red(f->child[1]))
564        *root = rotate_tree(f, 0); /* Rotate left */
565    else if (is_red(f->child[0]) && is_red(f->child[0]->child[0]))
566        *root = rotate_tree(f, 1); /* Rotate right */
567}
568
569/* Get a node with the specified name, creating it if it doesn't exist */
570static CheckasmFunc *get_func(CheckasmFunc **root, const char *name)
571{
572    CheckasmFunc *f = *root;
573
574    if (f) {
575        /* Search the tree for a matching node */
576        int cmp = cmp_func_names(name, f->name);
577        if (cmp) {
578            f = get_func(&f->child[cmp > 0], name);
579
580            /* Rebalance the tree on the way up if a new node was inserted */
581            if (!f->versions.func)
582                balance_tree(root);
583        }
584    } else {
585        /* Allocate and insert a new node into the tree */
586        int name_length = strlen(name);
587        f = *root = checkasm_malloc(sizeof(CheckasmFunc) + name_length);
588        memcpy(f->name, name, name_length + 1);
589    }
590
591    return f;
592}
593
594/* Perform tests and benchmarks for the specified cpu flag if supported by the host */
595static void check_cpu_flag(const char *name, int flag)
596{
597    int old_cpu_flag = state.cpu_flag;
598
599    flag |= old_cpu_flag;
600    av_force_cpu_flags(-1);
601    state.cpu_flag = flag & av_get_cpu_flags();
602    av_force_cpu_flags(state.cpu_flag);
603
604    if (!flag || state.cpu_flag != old_cpu_flag) {
605        int i;
606
607        state.cpu_flag_name = name;
608        for (i = 0; tests[i].func; i++) {
609            if (state.test_name && strcmp(tests[i].name, state.test_name))
610                continue;
611            state.current_test_name = tests[i].name;
612            tests[i].func();
613        }
614    }
615}
616
617/* Print the name of the current CPU flag, but only do it once */
618static void print_cpu_name(void)
619{
620    if (state.cpu_flag_name) {
621        color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
622        state.cpu_flag_name = NULL;
623    }
624}
625
626#if CONFIG_LINUX_PERF
627static int bench_init_linux(void)
628{
629    struct perf_event_attr attr = {
630        .type           = PERF_TYPE_HARDWARE,
631        .size           = sizeof(struct perf_event_attr),
632        .config         = PERF_COUNT_HW_CPU_CYCLES,
633        .disabled       = 1, // start counting only on demand
634        .exclude_kernel = 1,
635        .exclude_hv     = 1,
636    };
637
638    printf("benchmarking with Linux Perf Monitoring API\n");
639
640    state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
641    if (state.sysfd == -1) {
642        perror("syscall");
643        return -1;
644    }
645    return 0;
646}
647#elif CONFIG_MACOS_KPERF
648static int bench_init_kperf(void)
649{
650    ff_kperf_init();
651    return 0;
652}
653#else
654static int bench_init_ffmpeg(void)
655{
656#ifdef AV_READ_TIME
657    printf("benchmarking with native FFmpeg timers\n");
658    return 0;
659#else
660    fprintf(stderr, "checkasm: --bench is not supported on your system\n");
661    return -1;
662#endif
663}
664#endif
665
666static int bench_init(void)
667{
668#if CONFIG_LINUX_PERF
669    int ret = bench_init_linux();
670#elif CONFIG_MACOS_KPERF
671    int ret = bench_init_kperf();
672#else
673    int ret = bench_init_ffmpeg();
674#endif
675    if (ret < 0)
676        return ret;
677
678    state.nop_time = measure_nop_time();
679    printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
680    return 0;
681}
682
683static void bench_uninit(void)
684{
685#if CONFIG_LINUX_PERF
686    if (state.sysfd > 0)
687        close(state.sysfd);
688#endif
689}
690
691int main(int argc, char *argv[])
692{
693    unsigned int seed = av_get_random_seed();
694    int i, ret = 0;
695
696#if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
697    if (have_vfp(av_get_cpu_flags()) || have_neon(av_get_cpu_flags()))
698        checkasm_checked_call = checkasm_checked_call_vfp;
699#endif
700
701    if (!tests[0].func || !cpus[0].flag) {
702        fprintf(stderr, "checkasm: no tests to perform\n");
703        return 0;
704    }
705
706    while (argc > 1) {
707        if (!strncmp(argv[1], "--bench", 7)) {
708            if (bench_init() < 0)
709                return 1;
710            if (argv[1][7] == '=') {
711                state.bench_pattern = argv[1] + 8;
712                state.bench_pattern_len = strlen(state.bench_pattern);
713            } else
714                state.bench_pattern = "";
715        } else if (!strncmp(argv[1], "--test=", 7)) {
716            state.test_name = argv[1] + 7;
717        } else if (!strcmp(argv[1], "--verbose") || !strcmp(argv[1], "-v")) {
718            state.verbose = 1;
719        } else {
720            seed = strtoul(argv[1], NULL, 10);
721        }
722
723        argc--;
724        argv++;
725    }
726
727    fprintf(stderr, "checkasm: using random seed %u\n", seed);
728    av_lfg_init(&checkasm_lfg, seed);
729
730    check_cpu_flag(NULL, 0);
731    for (i = 0; cpus[i].flag; i++)
732        check_cpu_flag(cpus[i].name, cpus[i].flag);
733
734    if (state.num_failed) {
735        fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
736        ret = 1;
737    } else {
738        fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
739        if (state.bench_pattern) {
740            print_benchs(state.funcs);
741        }
742    }
743
744    destroy_func_tree(state.funcs);
745    bench_uninit();
746    return ret;
747}
748
749/* Decide whether or not the specified function needs to be tested and
750 * allocate/initialize data structures if needed. Returns a pointer to a
751 * reference function if the function should be tested, otherwise NULL */
752void *checkasm_check_func(void *func, const char *name, ...)
753{
754    char name_buf[256];
755    void *ref = func;
756    CheckasmFuncVersion *v;
757    int name_length;
758    va_list arg;
759
760    va_start(arg, name);
761    name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
762    va_end(arg);
763
764    if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
765        return NULL;
766
767    state.current_func = get_func(&state.funcs, name_buf);
768    state.funcs->color = 1;
769    v = &state.current_func->versions;
770
771    if (v->func) {
772        CheckasmFuncVersion *prev;
773        do {
774            /* Only test functions that haven't already been tested */
775            if (v->func == func)
776                return NULL;
777
778            if (v->ok)
779                ref = v->func;
780
781            prev = v;
782        } while ((v = v->next));
783
784        v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
785    }
786
787    v->func = func;
788    v->ok = 1;
789    v->cpu = state.cpu_flag;
790    state.current_func_ver = v;
791
792    if (state.cpu_flag)
793        state.num_checked++;
794
795    return ref;
796}
797
798/* Decide whether or not the current function needs to be benchmarked */
799int checkasm_bench_func(void)
800{
801    return !state.num_failed && state.bench_pattern &&
802           !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
803}
804
805/* Indicate that the current test has failed */
806void checkasm_fail_func(const char *msg, ...)
807{
808    if (state.current_func_ver->cpu && state.current_func_ver->ok) {
809        va_list arg;
810
811        print_cpu_name();
812        fprintf(stderr, "   %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
813        va_start(arg, msg);
814        vfprintf(stderr, msg, arg);
815        va_end(arg);
816        fprintf(stderr, ")\n");
817
818        state.current_func_ver->ok = 0;
819        state.num_failed++;
820    }
821}
822
823/* Get the benchmark context of the current function */
824CheckasmPerf *checkasm_get_perf_context(void)
825{
826    CheckasmPerf *perf = &state.current_func_ver->perf;
827    memset(perf, 0, sizeof(*perf));
828    perf->sysfd = state.sysfd;
829    return perf;
830}
831
832/* Print the outcome of all tests performed since the last time this function was called */
833void checkasm_report(const char *name, ...)
834{
835    static int prev_checked, prev_failed, max_length;
836
837    if (state.num_checked > prev_checked) {
838        int pad_length = max_length + 4;
839        va_list arg;
840
841        print_cpu_name();
842        pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
843        va_start(arg, name);
844        pad_length -= vfprintf(stderr, name, arg);
845        va_end(arg);
846        fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
847
848        if (state.num_failed == prev_failed)
849            color_printf(COLOR_GREEN, "OK");
850        else
851            color_printf(COLOR_RED, "FAILED");
852        fprintf(stderr, "]\n");
853
854        prev_checked = state.num_checked;
855        prev_failed  = state.num_failed;
856    } else if (!state.cpu_flag) {
857        /* Calculate the amount of padding required to make the output vertically aligned */
858        int length = strlen(state.current_test_name);
859        va_list arg;
860
861        va_start(arg, name);
862        length += vsnprintf(NULL, 0, name, arg);
863        va_end(arg);
864
865        if (length > max_length)
866            max_length = length;
867    }
868}
869
870#define DEF_CHECKASM_CHECK_FUNC(type, fmt) \
871int checkasm_check_##type(const char *const file, const int line, \
872                          const type *buf1, ptrdiff_t stride1, \
873                          const type *buf2, ptrdiff_t stride2, \
874                          const int w, int h, const char *const name) \
875{ \
876    int y = 0; \
877    stride1 /= sizeof(*buf1); \
878    stride2 /= sizeof(*buf2); \
879    for (y = 0; y < h; y++) \
880        if (memcmp(&buf1[y*stride1], &buf2[y*stride2], w*sizeof(*buf1))) \
881            break; \
882    if (y == h) \
883        return 0; \
884    checkasm_fail_func("%s:%d", file, line); \
885    if (!state.verbose) \
886        return 1; \
887    fprintf(stderr, "%s:\n", name); \
888    while (h--) { \
889        for (int x = 0; x < w; x++) \
890            fprintf(stderr, " " fmt, buf1[x]); \
891        fprintf(stderr, "    "); \
892        for (int x = 0; x < w; x++) \
893            fprintf(stderr, " " fmt, buf2[x]); \
894        fprintf(stderr, "    "); \
895        for (int x = 0; x < w; x++) \
896            fprintf(stderr, "%c", buf1[x] != buf2[x] ? 'x' : '.'); \
897        buf1 += stride1; \
898        buf2 += stride2; \
899        fprintf(stderr, "\n"); \
900    } \
901    return 1; \
902}
903
904DEF_CHECKASM_CHECK_FUNC(uint8_t,  "%02x")
905DEF_CHECKASM_CHECK_FUNC(uint16_t, "%04x")
906DEF_CHECKASM_CHECK_FUNC(int16_t,  "%6d")
907DEF_CHECKASM_CHECK_FUNC(int32_t,  "%9d")
908