Go to the documentation of this file.
24 #include "config_components.h"
28 # define _GNU_SOURCE // for syscall (performance monitoring API)
46 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
48 #define COLOR_RED FOREGROUND_RED
49 #define COLOR_GREEN FOREGROUND_GREEN
50 #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
54 #define COLOR_YELLOW 3
65 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
68 void (*checkasm_checked_call)(
void *
func,
int dummy, ...) = checkasm_checked_call_novfp;
77 #if CONFIG_AAC_DECODER
81 #if CONFIG_ALAC_DECODER
93 #if CONFIG_DCA_DECODER
96 #if CONFIG_EXR_DECODER
99 #if CONFIG_FLAC_DECODER
102 #if CONFIG_FMTCONVERT
117 #if CONFIG_HEVC_DECODER
123 #if CONFIG_HUFFYUV_DECODER
129 #if CONFIG_JPEG2000_DECODER
132 #if CONFIG_HUFFYUVDSP
135 #if CONFIG_LLVIDENCDSP
144 #if CONFIG_OPUS_DECODER
147 #if CONFIG_PIXBLOCKDSP
150 #if CONFIG_UTVIDEO_DECODER
153 #if CONFIG_V210_DECODER
156 #if CONFIG_V210_ENCODER
165 #if CONFIG_VP9_DECODER
171 #if CONFIG_VORBIS_DECODER
176 #if CONFIG_AFIR_FILTER
179 #if CONFIG_BLEND_FILTER
182 #if CONFIG_COLORSPACE_FILTER
188 #if CONFIG_GBLUR_FILTER
191 #if CONFIG_HFLIP_FILTER
194 #if CONFIG_NLMEANS_FILTER
197 #if CONFIG_THRESHOLD_FILTER
200 #if CONFIG_SOBEL_FILTER
218 static const struct {
334 if (llabs((int64_t)x.
i - y.
i) <= max_ulp)
345 for (
i = 0;
i <
len;
i++) {
358 fprintf(stderr,
"test failed comparing %g with %g (abs diff=%g with EPS=%g)\n",
a,
b, abs_diff, eps);
368 for (
i = 0;
i <
len;
i++) {
381 unsigned max_ulp,
unsigned len)
385 for (
i = 0;
i <
len;
i++) {
394 double abs_diff =
fabs(
a -
b);
396 return abs_diff < eps;
404 for (
i = 0;
i <
len;
i++) {
417 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
419 static WORD org_attributes;
422 CONSOLE_SCREEN_BUFFER_INFO con_info;
423 con = GetStdHandle(STD_ERROR_HANDLE);
424 if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
425 org_attributes = con_info.wAttributes;
431 SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (
color & 0x0f));
434 const char *term = getenv(
"TERM");
438 fprintf(stderr,
"\x1b[%d;3%dm", (
color & 0x08) >> 3,
color & 0x07);
442 vfprintf(stderr, fmt,
arg);
446 #if HAVE_SETCONSOLETEXTATTRIBUTE && HAVE_GETSTDHANDLE
447 SetConsoleTextAttribute(con, org_attributes);
449 fprintf(stderr,
"\x1b[0m");
474 void *ptr = calloc(1,
size);
476 fprintf(stderr,
"checkasm: malloc failed\n");
490 return cpus[
i].suffix;
497 return *(
const uint16_t*)
a - *(
const uint16_t*)
b;
503 uint16_t nops[10000];
508 for (
i = 0;
i < 10000;
i++) {
514 qsort(nops, 10000,
sizeof(uint16_t),
cmp_nop);
515 for (
i = 2500;
i < 7500;
i++)
518 return nop_sum / 500;
528 if (
f->versions.cpu ||
f->versions.next) {
536 }
while ((v = v->
next));
546 const char *start =
a;
547 int ascii_diff, digit_diff;
549 for (; !(ascii_diff = *(
const unsigned char*)
a - *(
const unsigned char*)
b) && *
a;
a++,
b++);
562 f->child[dir^1] =
r->child[dir];
569 #define is_red(f) ((f) && !(f)->color)
578 f->child[0]->color =
f->child[1]->color = 1;
599 if (!
f->versions.func)
604 int name_length = strlen(
name);
606 memcpy(
f->name,
name, name_length + 1);
615 int old_cpu_flag =
state.cpu_flag;
617 flag |= old_cpu_flag;
622 if (!
flag ||
state.cpu_flag != old_cpu_flag) {
638 if (
state.cpu_flag_name) {
644 #if CONFIG_LINUX_PERF
645 static int bench_init_linux(
void)
647 struct perf_event_attr attr = {
648 .type = PERF_TYPE_HARDWARE,
649 .size =
sizeof(
struct perf_event_attr),
650 .
config = PERF_COUNT_HW_CPU_CYCLES,
656 printf(
"benchmarking with Linux Perf Monitoring API\n");
658 state.sysfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
659 if (
state.sysfd == -1) {
665 #elif CONFIG_MACOS_KPERF
666 static int bench_init_kperf(
void)
675 printf(
"benchmarking with native FFmpeg timers\n");
678 fprintf(stderr,
"checkasm: --bench is not supported on your system\n");
686 #if CONFIG_LINUX_PERF
687 int ret = bench_init_linux();
688 #elif CONFIG_MACOS_KPERF
689 int ret = bench_init_kperf();
703 #if CONFIG_LINUX_PERF
709 int main(
int argc,
char *argv[])
714 #if ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
716 checkasm_checked_call = checkasm_checked_call_vfp;
720 fprintf(stderr,
"checkasm: no tests to perform\n");
725 if (!strncmp(argv[1],
"--bench", 7)) {
728 if (argv[1][7] ==
'=') {
729 state.bench_pattern = argv[1] + 8;
730 state.bench_pattern_len = strlen(
state.bench_pattern);
732 state.bench_pattern =
"";
733 }
else if (!strncmp(argv[1],
"--test=", 7)) {
734 state.test_name = argv[1] + 7;
735 }
else if (!strcmp(argv[1],
"--verbose") || !strcmp(argv[1],
"-v")) {
745 fprintf(stderr,
"checkasm: using random seed %u\n",
seed);
752 if (
state.num_failed) {
753 fprintf(stderr,
"checkasm: %d of %d tests have failed\n",
state.num_failed,
state.num_checked);
756 fprintf(stderr,
"checkasm: all %d tests passed\n",
state.num_checked);
757 if (
state.bench_pattern) {
782 if (!
func || name_length <= 0 || name_length >=
sizeof(name_buf))
786 state.funcs->color = 1;
787 v = &
state.current_func->versions;
800 }
while ((v = v->
next));
808 state.current_func_ver = v;
819 return !
state.num_failed &&
state.bench_pattern &&
820 !strncmp(
state.current_func->name,
state.bench_pattern,
state.bench_pattern_len);
826 if (
state.current_func_ver->cpu &&
state.current_func_ver->ok) {
832 vfprintf(stderr, msg,
arg);
834 fprintf(stderr,
")\n");
836 state.current_func_ver->ok = 0;
845 memset(perf, 0,
sizeof(*perf));
853 static int prev_checked, prev_failed, max_length;
855 if (
state.num_checked > prev_checked) {
856 int pad_length = max_length + 4;
860 pad_length -= fprintf(stderr,
" - %s.",
state.current_test_name);
862 pad_length -= vfprintf(stderr,
name,
arg);
864 fprintf(stderr,
"%*c",
FFMAX(pad_length, 0) + 2,
'[');
866 if (
state.num_failed == prev_failed)
870 fprintf(stderr,
"]\n");
872 prev_checked =
state.num_checked;
873 prev_failed =
state.num_failed;
874 }
else if (!
state.cpu_flag) {
876 int length = strlen(
state.current_test_name);
883 if (length > max_length)
888 #define DEF_CHECKASM_CHECK_FUNC(type, fmt) \
889 int checkasm_check_##type(const char *const file, const int line, \
890 const type *buf1, ptrdiff_t stride1, \
891 const type *buf2, ptrdiff_t stride2, \
892 const int w, int h, const char *const name) \
895 stride1 /= sizeof(*buf1); \
896 stride2 /= sizeof(*buf2); \
897 for (y = 0; y < h; y++) \
898 if (memcmp(&buf1[y*stride1], &buf2[y*stride2], w*sizeof(*buf1))) \
902 checkasm_fail_func("%s:%d", file, line); \
903 if (!state.verbose) \
905 fprintf(stderr, "%s:\n", name); \
907 for (int x = 0; x < w; x++) \
908 fprintf(stderr, " " fmt, buf1[x]); \
909 fprintf(stderr, " "); \
910 for (int x = 0; x < w; x++) \
911 fprintf(stderr, " " fmt, buf2[x]); \
912 fprintf(stderr, " "); \
913 for (int x = 0; x < w; x++) \
914 fprintf(stderr, "%c", buf1[x] != buf2[x] ? 'x' : '.'); \
917 fprintf(stderr, "\n"); \
void checkasm_check_nlmeans(void)
void checkasm_check_sw_gbrp(void)
void av_force_cpu_flags(int arg)
Disables cpu detection and forces the specified flags.
void checkasm_check_vp8dsp(void)
void checkasm_check_blockdsp(void)
#define AV_CPU_FLAG_SSE3
Prescott SSE3 functions.
static void destroy_func_tree(CheckasmFunc *f)
void checkasm_check_videodsp(void)
static int cmp_nop(const void *a, const void *b)
void checkasm_check_vf_eq(void)
#define AV_CPU_FLAG_RVB_BASIC
Basic bit-manipulations.
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
#define u(width, name, range_min, range_max)
void checkasm_check_alacdsp(void)
void checkasm_check_v210dec(void)
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
#define AV_CPU_FLAG_SSE3SLOW
SSE3 supported, but usually not faster.
const char * bench_pattern
void checkasm_check_hevc_idct(void)
void checkasm_check_sw_scale(void)
struct CheckasmFunc * child[2]
void checkasm_check_afir(void)
static void print_cpu_name(void)
#define AV_CPU_FLAG_3DNOW
AMD 3DNOW.
#define AV_CPU_FLAG_RVF
F (single precision FP)
int float_near_abs_eps(float a, float b, float eps)
void checkasm_check_aacpsdsp(void)
static void bench_uninit(void)
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
static const char * cpu_suffix(int cpu)
void checkasm_check_llviddspenc(void)
void checkasm_check_h264dsp(void)
static void color_printf(int color, const char *fmt,...)
#define AV_CPU_FLAG_RVV_F64
Vectors of double's.
uint32_t av_get_random_seed(void)
Get a seed to use in conjunction with random functions.
int main(int argc, char *argv[])
void checkasm_check_motion(void)
#define AV_CPU_FLAG_SSSE3
Conroe SSSE3 functions.
int double_near_abs_eps(double a, double b, double eps)
const char * current_test_name
static CheckasmFunc * rotate_tree(CheckasmFunc *f, int dir)
#define AV_CPU_FLAG_XOP
Bulldozer XOP functions.
void checkasm_check_vp9dsp(void)
void checkasm_check_vf_threshold(void)
void checkasm_check_vf_gblur(void)
#define AV_CPU_FLAG_3DNOWEXT
AMD 3DNowExt.
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
#define AV_CPU_FLAG_VSX
ISA 2.06.
static __device__ float fabsf(float a)
#define AV_CPU_FLAG_RVV_F32
Vectors of float's */.
void checkasm_check_float_dsp(void)
#define AV_CPU_FLAG_AVX512
AVX-512 functions: requires OS support even if YMM/ZMM registers aren't used.
static int bench_init(void)
void checkasm_check_pixblockdsp(void)
#define FF_ARRAY_ELEMS(a)
void checkasm_report(const char *name,...)
static CheckasmFunc * get_func(CheckasmFunc **root, const char *name)
void checkasm_fail_func(const char *msg,...)
void checkasm_check_sw_rgb(void)
void checkasm_check_hevc_sao(void)
#define AV_CPU_FLAG_ARMV6
#define AV_CPU_FLAG_SSE4
Penryn SSE4.1 functions.
void checkasm_check_hevc_pel(void)
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...
#define AV_CPU_FLAG_RVD
D (double precision FP)
static __device__ float fabs(float a)
static int measure_nop_time(void)
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
#define AV_CPU_FLAG_AVX512ICL
F/CD/BW/DQ/VL/VNNI/IFMA/VBMI/VBMI2/VPOPCNTDQ/BITALG/GFNI/VAES/VPCLMULQDQ.
void checkasm_check_fmtconvert(void)
void checkasm_check_vorbisdsp(void)
#define AV_CPU_FLAG_CMOV
supports cmov instruction
#define AV_CPU_FLAG_ALTIVEC
standard
static void print_benchs(CheckasmFunc *f)
CheckasmFunc * current_func
void checkasm_check_vc1dsp(void)
#define AV_CPU_FLAG_SSE2
PIV SSE2 functions.
Context structure for the Lagged Fibonacci PRNG.
#define AV_CPU_FLAG_AVX
AVX functions: requires OS support even if YMM registers aren't used.
#define AV_CPU_FLAG_FMA4
Bulldozer FMA4 functions.
#define AV_CPU_FLAG_AVX2
AVX2 functions: requires OS support even if YMM registers aren't used.
void checkasm_check_llviddsp(void)
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
static const struct @343 tests[]
#define DEF_CHECKASM_CHECK_FUNC(type, fmt)
printf("static const uint8_t my_array[100] = {\n")
void checkasm_check_h264pred(void)
static av_const int av_isdigit(int c)
Locale-independent conversion of ASCII isdigit.
#define AV_CPU_FLAG_FMA3
Haswell FMA3 functions.
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
void checkasm_check_huffyuvdsp(void)
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
CheckasmPerf * checkasm_get_perf_context(void)
void checkasm_check_fixed_dsp(void)
void checkasm_check_colorspace(void)
void checkasm_check_utvideodsp(void)
static int is_negative(union av_intfloat32 u)
void checkasm_check_audiodsp(void)
static const struct @344 cpus[]
#define i(width, name, range_min, range_max)
void checkasm_check_flacdsp(void)
void * checkasm_check_func(void *func, const char *name,...)
#define AV_CPU_FLAG_SSE42
Nehalem SSE4.2 functions.
void checkasm_check_idctdsp(void)
struct CheckasmFuncVersion * next
#define AV_CPU_FLAG_RVV_I32
Vectors of 8/16/32-bit int's */.
void checkasm_check_synth_filter(void)
static int cmp_func_names(const char *a, const char *b)
#define AV_CPU_FLAG_ARMV8
void checkasm_check_lpc(void)
void checkasm_check_g722dsp(void)
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
void checkasm_check_vf_hflip(void)
void checkasm_check_blend(void)
#define AV_CPU_FLAG_VFPV3
void checkasm_check_bswapdsp(void)
#define AV_CPU_FLAG_ARMV5TE
#define AV_CPU_FLAG_MMX
standard MMX
static void * checkasm_malloc(size_t size)
#define AV_CPU_FLAG_RVI
I (full GPR bank)
void checkasm_check_h264qpel(void)
#define AV_CPU_FLAG_AESNI
Advanced Encryption Standard functions.
int checkasm_bench_func(void)
#define AV_CPU_FLAG_POWER8
ISA 2.07.
const char * cpu_flag_name
void checkasm_check_vf_sobel(void)
#define AV_CPU_FLAG_SSE
SSE functions.
static int ref[MAX_W *MAX_W]
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
#define AV_CPU_FLAG_MMXEXT
SSE integer functions or AMD MMX ext.
void checkasm_check_av_tx(void)
void checkasm_check_v210enc(void)
void checkasm_check_sbrdsp(void)
#define AV_CPU_FLAG_VFP_VM
VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations.
void checkasm_check_jpeg2000dsp(void)
void checkasm_check_hevc_add_res(void)
int float_near_ulp(float a, float b, unsigned max_ulp)
CheckasmFuncVersion * current_func_ver
static void check_cpu_flag(const char *name, int flag)
static int bench_init_ffmpeg(void)
CheckasmFuncVersion versions
#define AV_CPU_FLAG_RVV_I64
Vectors of 64-bit int's */.
void checkasm_check_exrdsp(void)
static void balance_tree(CheckasmFunc **root)
#define AV_CPU_FLAG_ARMV6T2
void checkasm_check_opusdsp(void)