Go to the documentation of this file.
92 #if CONFIG_MPEG4_DECODER
110 #define AANSCALE_BITS 12
113 #define NB_ITS_SPEED 50000
126 for (
i = 0;
i < 64;
i++)
130 for (
i = 0;
i < 64;
i++)
136 for (
i = 0;
i < j;
i++) {
160 for (
i = 0;
i < 64;
i++)
161 dst[(
i & 0x38) | ((
i & 6) >> 1) | ((
i & 1) << 2)] =
src[
i];
164 for (
i = 0;
i < 64;
i++)
165 dst[(
i & 0x24) | ((
i & 3) << 3) | ((
i >> 3) & 3)] =
src[
i];
168 for (
i = 0;
i < 64;
i++)
169 dst[(
i>>3) | ((
i<<3)&0x38)] =
src[
i];
172 for (
i = 0;
i < 64;
i++)
183 int64_t err2, ti, ti1, it1, err_sum = 0;
184 int64_t sysErr[64], sysErrMax = 0;
185 int64_t err2_matrix[64], err2_max = 0;
187 int blockSumErrMax = 0, blockSumErr;
189 const int vals=1<<
bits;
197 for (
i = 0;
i < 64;
i++)
198 err2_matrix[
i] = sysErr[
i] = 0;
206 if (!strcmp(
dct->name,
"IJG-AAN-INT")) {
207 for (
i = 0;
i < 64;
i++) {
214 if (!strcmp(
dct->name,
"PR-SSE2"))
215 for (
i = 0;
i < 64;
i++)
219 for (
i = 0;
i < 64;
i++) {
225 err2_matrix[
i] += v * v;
232 if (blockSumErrMax < blockSumErr)
233 blockSumErrMax = blockSumErr;
235 for (
i = 0;
i < 64;
i++) {
237 err2_max =
FFMAX(err2_max ,
FFABS(err2_matrix[
i]));
240 for (
i = 0;
i < 64;
i++) {
243 printf(
"%7d ", (
int) sysErr[
i]);
247 omse = (double) err2 /
NB_ITS / 64;
248 ome = (double) err_sum /
NB_ITS / 64;
250 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
252 spec_err = is_idct && ((double) err2_max /
NB_ITS > 0.06 || (
double) sysErrMax /
NB_ITS > 0.015);
254 printf(
"%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
255 is_idct ?
"IDCT" :
"DCT",
dct->name, err_inf,
256 omse, ome, (
double) sysErrMax /
NB_ITS,
257 maxout, blockSumErrMax);
259 if (spec_err && !
dct->nonspec) {
282 }
while (ti1 < 1000000);
284 printf(
"%s %s: %0.1f kdct/s\n", is_idct ?
"IDCT" :
"DCT",
dct->name,
285 (
double) it1 * 1000.0 / (
double) ti1);
296 static double c8[8][8];
297 static double c4[4][4];
298 double block1[64], block2[64], block3[64];
305 for (
i = 0;
i < 8;
i++) {
307 for (j = 0; j < 8; j++) {
308 s = (
i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
309 c8[
i][j] =
s * cos(
M_PI *
i * (j + 0.5) / 8.0);
310 sum += c8[
i][j] * c8[
i][j];
314 for (
i = 0;
i < 4;
i++) {
316 for (j = 0; j < 4; j++) {
317 s = (
i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
318 c4[
i][j] =
s * cos(
M_PI *
i * (j + 0.5) / 4.0);
319 sum += c4[
i][j] * c4[
i][j];
326 for (
i = 0;
i < 4;
i++) {
327 for (j = 0; j < 8; j++) {
336 for (
i = 0;
i < 8;
i++) {
337 for (j = 0; j < 8; j++) {
339 for (k = 0; k < 8; k++)
340 sum += c8[k][j] *
block1[8 *
i + k];
341 block2[8 *
i + j] = sum;
346 for (
i = 0;
i < 8;
i++) {
347 for (j = 0; j < 4; j++) {
350 for (k = 0; k < 4; k++)
351 sum += c4[k][j] * block2[8 * (2 * k) +
i];
352 block3[8 * (2 * j) +
i] = sum;
356 for (k = 0; k < 4; k++)
357 sum += c4[k][j] * block2[8 * (2 * k + 1) +
i];
358 block3[8 * (2 * j + 1) +
i] = sum;
363 for (
i = 0;
i < 8;
i++) {
364 for (j = 0; j < 8; j++) {
365 v = block3[8 *
i + j];
367 else if (v > 255) v = 255;
368 dest[
i * linesize + j] = (
int)
rint(v);
374 void (*idct248_put)(
uint8_t *dest,
379 int it,
i, it1, ti, ti1, err_max, v;
389 for (
i = 0;
i < 64;
i++)
393 for (
i = 0;
i < 64;
i++)
397 for (
i = 0;
i < 64;
i++)
401 for (
i = 0;
i < 64;
i++) {
428 printf(
"%s %s: err_inf=%d\n", 1 ?
"IDCT248" :
"DCT248",
name, err_max);
437 for (
i = 0;
i < 64;
i++)
444 }
while (ti1 < 1000000);
446 printf(
"%s %s: %0.1f kdct/s\n", 1 ?
"IDCT248" :
"DCT248",
name,
447 (
double) it1 * 1000.0 / (
double) ti1);
452 printf(
"dct-test [-i] [<test-number>] [<bits>]\n"
453 "test-number 0 -> test with random matrixes\n"
454 " 1 -> test with random sparse matrixes\n"
455 " 2 -> do 3. test from MPEG-4 std\n"
456 "bits Number of time domain bits to use, 8 is default\n"
457 "-i test IDCT implementations\n"
458 "-4 test IDCT248 implementations\n"
466 int main(
int argc,
char **argv)
468 int test_idct = 0, test_248_dct = 0;
478 c =
getopt(argc, argv,
"ih4t");
502 printf(
"ffmpeg DCT/IDCT test\n");
529 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
static av_cold int init(AVCodecContext *avctx)
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 ff_prores_idct_wrap(int16_t *dst)
static uint8_t img_dest[64]
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
static atomic_int cpu_flags
void ff_simple_idct_int16_10bit(int16_t *block)
int main(int argc, char **argv)
static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
void ff_simple_idct248_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
static int getopt(int argc, char *argv[], char *opts)
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])
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
static const struct algo idct_tab[]
void ff_faandct(int16_t *data)
void ff_fdct_ifast(int16_t *data)
void ff_jpeg_fdct_islow_8(int16_t *data)
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.
printf("static const uint8_t my_array[100] = {\n")
void(* func)(int16_t *block)
static const struct algo fdct_tab_arch[]
#define DECLARE_ALIGNED(n, t, v)
static int permute_x86(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
void ff_simple_idct_int16_8bit(int16_t *block)
#define i(width, name, range_min, range_max)
static void permute(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
static const struct algo fdct_tab[]
static void dct(AudioRNNContext *s, float *out, const float *in)
void ff_j_rev_dct(int16_t *data)
static void idct248_error(const char *name, void(*idct248_put)(uint8_t *dest, ptrdiff_t line_size, int16_t *block), int speed)
#define FF_ARRAY_ELEMS(a)
static void test(const char *pattern, const char *host)
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[]
void ff_prores_idct_10(int16_t *block, const int16_t *qmat)
Special version of ff_simple_idct_int16_10bit() which does dequantization and scales by a factor of 2...
#define LOCAL_ALIGNED(a, t, v,...)
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 uint8_t img_dest1[64]
static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
static int16_t block1[64]
const uint16_t ff_aanscales[64]