Go to the documentation of this file.
29 #include "config_components.h"
55 #if CONFIG_PRORES_DECODER
76 #if CONFIG_PRORES_DECODER
77 static void ff_prores_idct_wrap(int16_t *
dst){
97 #if CONFIG_PRORES_DECODER
103 #if CONFIG_MPEG4_DECODER
121 #define AANSCALE_BITS 12
124 #define NB_ITS_SPEED 50000
137 for (
i = 0;
i < 64;
i++)
141 for (
i = 0;
i < 64;
i++)
147 for (
i = 0;
i < j;
i++) {
171 for (
i = 0;
i < 64;
i++)
172 dst[(
i & 0x38) | ((
i & 6) >> 1) | ((
i & 1) << 2)] =
src[
i];
175 for (
i = 0;
i < 64;
i++)
176 dst[(
i & 0x24) | ((
i & 3) << 3) | ((
i >> 3) & 3)] =
src[
i];
179 for (
i = 0;
i < 64;
i++)
183 for (
i = 0;
i < 64;
i++)
194 int64_t err2, ti, ti1, it1, err_sum = 0;
195 int64_t sysErr[64], sysErrMax = 0;
196 int64_t err2_matrix[64], err2_max = 0;
198 int blockSumErrMax = 0, blockSumErr;
200 const int vals=1<<
bits;
208 for (
i = 0;
i < 64;
i++)
209 err2_matrix[
i] = sysErr[
i] = 0;
217 if (!strcmp(
dct->name,
"IJG-AAN-INT")) {
218 for (
i = 0;
i < 64;
i++) {
225 if (!strcmp(
dct->name,
"PR-SSE2"))
226 for (
i = 0;
i < 64;
i++)
230 for (
i = 0;
i < 64;
i++) {
243 if (blockSumErrMax < blockSumErr)
244 blockSumErrMax = blockSumErr;
246 for (
i = 0;
i < 64;
i++) {
248 err2_max =
FFMAX(err2_max ,
FFABS(err2_matrix[
i]));
251 for (
i = 0;
i < 64;
i++) {
254 printf(
"%7d ", (
int) sysErr[
i]);
261 spec_err = is_idct && (err_inf > 1 || omse > 0.02 ||
fabs(ome) > 0.0015);
263 spec_err = is_idct && ((
double) err2_max /
NB_ITS > 0.06 || (
double) sysErrMax /
NB_ITS > 0.015);
265 printf(
"%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
266 is_idct ?
"IDCT" :
"DCT",
dct->name, err_inf,
267 omse, ome, (
double) sysErrMax /
NB_ITS,
268 maxout, blockSumErrMax);
270 if (spec_err && !
dct->nonspec) {
293 }
while (ti1 < 1000000);
295 printf(
"%s %s: %0.1f kdct/s\n", is_idct ?
"IDCT" :
"DCT",
dct->name,
296 (
double) it1 * 1000.0 / (
double) ti1);
307 static double c8[8][8];
308 static double c4[4][4];
309 double block1[64], block2[64], block3[64];
316 for (
i = 0;
i < 8;
i++) {
318 for (j = 0; j < 8; j++) {
319 s = (
i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
320 c8[
i][j] =
s * cos(
M_PI *
i * (j + 0.5) / 8.0);
321 sum += c8[
i][j] * c8[
i][j];
325 for (
i = 0;
i < 4;
i++) {
327 for (j = 0; j < 4; j++) {
328 s = (
i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
329 c4[
i][j] =
s * cos(
M_PI *
i * (j + 0.5) / 4.0);
330 sum += c4[
i][j] * c4[
i][j];
337 for (
i = 0;
i < 4;
i++) {
338 for (j = 0; j < 8; j++) {
347 for (
i = 0;
i < 8;
i++) {
348 for (j = 0; j < 8; j++) {
350 for (k = 0; k < 8; k++)
351 sum += c8[k][j] *
block1[8 *
i + k];
352 block2[8 *
i + j] = sum;
357 for (
i = 0;
i < 8;
i++) {
358 for (j = 0; j < 4; j++) {
361 for (k = 0; k < 4; k++)
362 sum += c4[k][j] * block2[8 * (2 * k) +
i];
363 block3[8 * (2 * j) +
i] = sum;
367 for (k = 0; k < 4; k++)
368 sum += c4[k][j] * block2[8 * (2 * k + 1) +
i];
369 block3[8 * (2 * j + 1) +
i] = sum;
374 for (
i = 0;
i < 8;
i++) {
375 for (j = 0; j < 8; j++) {
376 v = block3[8 *
i + j];
378 else if (v > 255) v = 255;
379 dest[
i * linesize + j] = (int)
rint(v);
385 void (*idct248_put)(uint8_t *dest,
390 int it,
i, it1, ti, ti1, err_max, v;
400 for (
i = 0;
i < 64;
i++)
404 for (
i = 0;
i < 64;
i++)
408 for (
i = 0;
i < 64;
i++)
412 for (
i = 0;
i < 64;
i++) {
439 printf(
"%s %s: err_inf=%d\n", 1 ?
"IDCT248" :
"DCT248",
name, err_max);
448 for (
i = 0;
i < 64;
i++)
455 }
while (ti1 < 1000000);
457 printf(
"%s %s: %0.1f kdct/s\n", 1 ?
"IDCT248" :
"DCT248",
name,
458 (
double) it1 * 1000.0 / (
double) ti1);
463 printf(
"dct-test [-i] [<test-number>] [<bits>]\n"
464 "test-number 0 -> test with random matrixes\n"
465 " 1 -> test with random sparse matrixes\n"
466 " 2 -> do 3. test from MPEG-4 std\n"
467 "bits Number of time domain bits to use, 8 is default\n"
468 "-i test IDCT implementations\n"
469 "-4 test IDCT248 implementations\n"
477 int main(
int argc,
char **argv)
479 int test_idct = 0, test_248_dct = 0;
489 c =
getopt(argc, argv,
"ih4t");
513 printf(
"ffmpeg DCT/IDCT test\n");
540 printf(
"Error: %d.\n", err);
int64_t av_gettime_relative(void)
Get the current time in microseconds since some unspecified starting point.
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default minimum maximum flags name is the option name
__device__ int printf(const char *,...)
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
av_cold void ff_ref_dct_init(void)
Initialize the double precision discrete cosine transform functions fdct & idct.
static void permute(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
static int permute_x86(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
static const struct algo fdct_tab[]
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
static atomic_int cpu_flags
static uint8_t img_dest1[64]
void ff_simple_idct_int16_10bit(int16_t *block)
void ff_j_rev_dct(int16_t data[64])
static void prores_idct_10(int16_t *restrict block, const int16_t *restrict qmat)
Special version of ff_simple_idct_int16_10bit() which does dequantization and scales by a factor of 2...
void ff_simple_idct248_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
#define FF_ARRAY_ELEMS(a)
static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
static int getopt(int argc, char *argv[], char *opts)
#define LOCAL_ALIGNED(a, t, v,...)
int main(int argc, char **argv)
static unsigned int av_lfg_get(AVLFG *c)
Get the next random unsigned 32-bit number using an ALFG.
void ff_xvid_idct(int16_t *const in)
void ff_faanidct(int16_t block[64])
static void idct248_error(const char *name, void(*idct248_put)(uint8_t *dest, ptrdiff_t line_size, int16_t *block), int speed)
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
static __device__ float fabs(float a)
void ff_faandct(int16_t *data)
static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Context structure for the Lagged Fibonacci PRNG.
int(* init)(AVBSFContext *ctx)
#define DECLARE_ALIGNED(n, t, v)
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
static const struct algo fdct_tab_arch[]
static int16_t block1[64]
void(* func)(int16_t *block)
static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
void ff_simple_idct_int16_8bit(int16_t *block)
#define i(width, name, range_min, range_max)
void ff_jpeg_fdct_islow_8(int16_t *data)
static void dct(AudioRNNContext *s, float *out, const float *in)
void ff_fdct_ifast(int16_t *data)
static uint8_t img_dest[64]
static const struct algo idct_tab[]
void ff_simple_idct_int16_12bit(int16_t *block)
static int ref[MAX_W *MAX_W]
void ff_ref_fdct(short *block)
Transform 8x8 block of data with a double precision forward DCT This is a reference implementation.
static const struct algo idct_tab_arch[]
s EdgeDetect Foobar g libavfilter vf_edgedetect c libavfilter vf_foobar c edit libavfilter and add an entry for foobar following the pattern of the other filters edit libavfilter allfilters and add an entry for foobar following the pattern of the other filters configure make j< whatever > ffmpeg ffmpeg i you should get a foobar png with Lena edge detected That s it
void ff_ref_idct(short *block)
Transform 8x8 block of data with a double precision inverse DCT This is a reference implementation.
enum idct_permutation_type perm_type
static void scale(int *out, const int *in, const int w, const int h, const int shift)
const uint16_t ff_aanscales[64]