diff options
author | Diego Biurrun <diego@biurrun.de> | 2016-04-20 11:40:40 +0200 |
---|---|---|
committer | Diego Biurrun <diego@biurrun.de> | 2016-05-13 14:55:56 +0200 |
commit | a6a750c7ef240b72ce01e9653343a0ddf247d196 (patch) | |
tree | 5c472bb92b5264f19b9270b0caf8a8565b071cef /libavcodec/tests/dct.c | |
parent | 257f00ec1ab06a2a161f535036c6512f3fc8e801 (diff) | |
download | ffmpeg-a6a750c7ef240b72ce01e9653343a0ddf247d196.tar.gz |
tests: Move all test programs to a subdirectory
Diffstat (limited to 'libavcodec/tests/dct.c')
-rw-r--r-- | libavcodec/tests/dct.c | 472 |
1 files changed, 472 insertions, 0 deletions
diff --git a/libavcodec/tests/dct.c b/libavcodec/tests/dct.c new file mode 100644 index 0000000000..6c6a4e0d0d --- /dev/null +++ b/libavcodec/tests/dct.c @@ -0,0 +1,472 @@ +/* + * (c) 2001 Fabrice Bellard + * 2007 Marc Hoffman <marc.hoffman@analog.com> + * + * This file is part of Libav. + * + * Libav is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * Libav is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * DCT test (c) 2001 Fabrice Bellard + * Started from sample code by Juan J. Sierralta P. + */ + +#include "config.h" +#include <stdlib.h> +#include <stdio.h> +#include <string.h> +#if HAVE_UNISTD_H +#include <unistd.h> +#endif +#include <math.h> + +#include "libavutil/cpu.h" +#include "libavutil/common.h" +#include "libavutil/internal.h" +#include "libavutil/lfg.h" +#include "libavutil/time.h" + +#include "libavcodec/aandcttab.h" +#include "libavcodec/dct.h" +#include "libavcodec/dctref.h" +#include "libavcodec/faandct.h" +#include "libavcodec/faanidct.h" +#include "libavcodec/idctdsp.h" +#include "libavcodec/simple_idct.h" +#include "libavcodec/xvididct.h" + +struct algo { + const char *name; + void (*func)(int16_t *block); + enum idct_permutation_type perm_type; + int cpu_flag; + int nonspec; +}; + +static const struct algo fdct_tab[] = { + { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE }, + { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE }, + { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE }, +#if CONFIG_FAANDCT + { "FAAN", ff_faandct, FF_IDCT_PERM_NONE }, +#endif /* CONFIG_FAANDCT */ +}; + +static const struct algo idct_tab[] = { + { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE }, + { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 }, + { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE }, +#if CONFIG_FAANIDCT + { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE }, +#endif /* CONFIG_FAANIDCT */ +#if CONFIG_MPEG4_DECODER + { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 }, +#endif /* CONFIG_MPEG4_DECODER */ +}; + +#if ARCH_ARM +#include "arm/dct.c" +#elif ARCH_PPC +#include "ppc/dct.c" +#elif ARCH_X86 +#include "x86/dct.c" +#else +static const struct algo fdct_tab_arch[] = { { 0 } }; +static const struct algo idct_tab_arch[] = { { 0 } }; +#endif + +#define AANSCALE_BITS 12 + +#define NB_ITS 20000 +#define NB_ITS_SPEED 50000 + +DECLARE_ALIGNED(16, static int16_t, block)[64]; +DECLARE_ALIGNED(8, static int16_t, block1)[64]; + +static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng) +{ + int i, j; + + memset(block, 0, 64 * sizeof(*block)); + + switch (test) { + case 0: + for (i = 0; i < 64; i++) + block[i] = (av_lfg_get(prng) % 512) - 256; + if (is_idct) { + ff_ref_fdct(block); + for (i = 0; i < 64; i++) + block[i] >>= 3; + } + break; + case 1: + j = av_lfg_get(prng) % 10 + 1; + for (i = 0; i < j; i++) + block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256; + break; + case 2: + block[ 0] = av_lfg_get(prng) % 4096 - 2048; + block[63] = (block[0] & 1) ^ 1; + break; + } +} + +static void permute(int16_t dst[64], const int16_t src[64], + enum idct_permutation_type perm_type) +{ + int i; + +#if ARCH_X86 + if (permute_x86(dst, src, perm_type)) + return; +#endif + + switch (perm_type) { + case FF_IDCT_PERM_LIBMPEG2: + for (i = 0; i < 64; i++) + dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i]; + break; + case FF_IDCT_PERM_PARTTRANS: + for (i = 0; i < 64; i++) + dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i]; + break; + default: + for (i = 0; i < 64; i++) + dst[i] = src[i]; + break; + } +} + +static int dct_error(const struct algo *dct, int test, int is_idct, int speed) +{ + void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct; + int it, i, scale; + int err_inf, v; + int64_t err2, ti, ti1, it1, err_sum = 0; + int64_t sysErr[64], sysErrMax = 0; + int maxout = 0; + int blockSumErrMax = 0, blockSumErr; + AVLFG prng; + double omse, ome; + int spec_err; + + av_lfg_init(&prng, 1); + + err_inf = 0; + err2 = 0; + for (i = 0; i < 64; i++) + sysErr[i] = 0; + for (it = 0; it < NB_ITS; it++) { + init_block(block1, test, is_idct, &prng); + permute(block, block1, dct->perm_type); + + dct->func(block); + emms_c(); + + if (!strcmp(dct->name, "IJG-AAN-INT")) { + for (i = 0; i < 64; i++) { + scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i]; + block[i] = (block[i] * scale) >> AANSCALE_BITS; + } + } + + ref(block1); + + blockSumErr = 0; + for (i = 0; i < 64; i++) { + int err = block[i] - block1[i]; + err_sum += err; + v = abs(err); + if (v > err_inf) + err_inf = v; + err2 += v * v; + sysErr[i] += block[i] - block1[i]; + blockSumErr += v; + if (abs(block[i]) > maxout) + maxout = abs(block[i]); + } + if (blockSumErrMax < blockSumErr) + blockSumErrMax = blockSumErr; + } + for (i = 0; i < 64; i++) + sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i])); + + for (i = 0; i < 64; i++) { + if (i % 8 == 0) + printf("\n"); + printf("%7d ", (int) sysErr[i]); + } + printf("\n"); + + omse = (double) err2 / NB_ITS / 64; + ome = (double) err_sum / NB_ITS / 64; + + spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015); + + printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n", + is_idct ? "IDCT" : "DCT", dct->name, err_inf, + omse, ome, (double) sysErrMax / NB_ITS, + maxout, blockSumErrMax); + + if (spec_err && !dct->nonspec) + return 1; + + if (!speed) + return 0; + + /* speed test */ + init_block(block, test, is_idct, &prng); + permute(block1, block, dct->perm_type); + + ti = av_gettime_relative(); + it1 = 0; + do { + for (it = 0; it < NB_ITS_SPEED; it++) { + memcpy(block, block1, sizeof(block)); + dct->func(block); + } + it1 += NB_ITS_SPEED; + ti1 = av_gettime_relative() - ti; + } while (ti1 < 1000000); + emms_c(); + + printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name, + (double) it1 * 1000.0 / (double) ti1); + + return 0; +} + +DECLARE_ALIGNED(8, static uint8_t, img_dest)[64]; +DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64]; + +static void idct248_ref(uint8_t *dest, int linesize, int16_t *block) +{ + static int init; + static double c8[8][8]; + static double c4[4][4]; + double block1[64], block2[64], block3[64]; + double s, sum, v; + int i, j, k; + + if (!init) { + init = 1; + + for (i = 0; i < 8; i++) { + sum = 0; + for (j = 0; j < 8; j++) { + s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0); + c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0); + sum += c8[i][j] * c8[i][j]; + } + } + + for (i = 0; i < 4; i++) { + sum = 0; + for (j = 0; j < 4; j++) { + s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0); + c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0); + sum += c4[i][j] * c4[i][j]; + } + } + } + + /* butterfly */ + s = 0.5 * sqrt(2.0); + for (i = 0; i < 4; i++) { + for (j = 0; j < 8; j++) { + block1[8 * (2 * i) + j] = + (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s; + block1[8 * (2 * i + 1) + j] = + (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s; + } + } + + /* idct8 on lines */ + for (i = 0; i < 8; i++) { + for (j = 0; j < 8; j++) { + sum = 0; + for (k = 0; k < 8; k++) + sum += c8[k][j] * block1[8 * i + k]; + block2[8 * i + j] = sum; + } + } + + /* idct4 */ + for (i = 0; i < 8; i++) { + for (j = 0; j < 4; j++) { + /* top */ + sum = 0; + for (k = 0; k < 4; k++) + sum += c4[k][j] * block2[8 * (2 * k) + i]; + block3[8 * (2 * j) + i] = sum; + + /* bottom */ + sum = 0; + for (k = 0; k < 4; k++) + sum += c4[k][j] * block2[8 * (2 * k + 1) + i]; + block3[8 * (2 * j + 1) + i] = sum; + } + } + + /* clamp and store the result */ + for (i = 0; i < 8; i++) { + for (j = 0; j < 8; j++) { + v = block3[8 * i + j]; + if (v < 0) v = 0; + else if (v > 255) v = 255; + dest[i * linesize + j] = (int) rint(v); + } + } +} + +static void idct248_error(const char *name, + void (*idct248_put)(uint8_t *dest, int line_size, + int16_t *block), + int speed) +{ + int it, i, it1, ti, ti1, err_max, v; + AVLFG prng; + + av_lfg_init(&prng, 1); + + /* just one test to see if code is correct (precision is less + important here) */ + err_max = 0; + for (it = 0; it < NB_ITS; it++) { + /* XXX: use forward transform to generate values */ + for (i = 0; i < 64; i++) + block1[i] = av_lfg_get(&prng) % 256 - 128; + block1[0] += 1024; + + for (i = 0; i < 64; i++) + block[i] = block1[i]; + idct248_ref(img_dest1, 8, block); + + for (i = 0; i < 64; i++) + block[i] = block1[i]; + idct248_put(img_dest, 8, block); + + for (i = 0; i < 64; i++) { + v = abs((int) img_dest[i] - (int) img_dest1[i]); + if (v == 255) + printf("%d %d\n", img_dest[i], img_dest1[i]); + if (v > err_max) + err_max = v; + } + } + printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max); + + if (!speed) + return; + + ti = av_gettime_relative(); + it1 = 0; + do { + for (it = 0; it < NB_ITS_SPEED; it++) { + for (i = 0; i < 64; i++) + block[i] = block1[i]; + idct248_put(img_dest, 8, block); + } + it1 += NB_ITS_SPEED; + ti1 = av_gettime_relative() - ti; + } while (ti1 < 1000000); + emms_c(); + + printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name, + (double) it1 * 1000.0 / (double) ti1); +} + +static void help(void) +{ + printf("dct-test [-i] [<test-number>]\n" + "test-number 0 -> test with random matrixes\n" + " 1 -> test with random sparse matrixes\n" + " 2 -> do 3. test from MPEG-4 std\n" + "-i test IDCT implementations\n" + "-4 test IDCT248 implementations\n" + "-t speed test\n"); +} + +#if !HAVE_GETOPT +#include "compat/getopt.c" +#endif + +int main(int argc, char **argv) +{ + int test_idct = 0, test_248_dct = 0; + int c, i; + int test = 1; + int speed = 0; + int err = 0; + + ff_ref_dct_init(); + + for (;;) { + c = getopt(argc, argv, "ih4t"); + if (c == -1) + break; + switch (c) { + case 'i': + test_idct = 1; + break; + case '4': + test_248_dct = 1; + break; + case 't': + speed = 1; + break; + default: + case 'h': + help(); + return 0; + } + } + + if (optind < argc) + test = atoi(argv[optind]); + + printf("Libav DCT/IDCT test\n"); + + if (test_248_dct) { + idct248_error("SIMPLE-C", ff_simple_idct248_put, speed); + } else { + const int cpu_flags = av_get_cpu_flags(); + if (test_idct) { + for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++) + err |= dct_error(&idct_tab[i], test, test_idct, speed); + + for (i = 0; idct_tab_arch[i].name; i++) + if (!(~cpu_flags & idct_tab_arch[i].cpu_flag)) + err |= dct_error(&idct_tab_arch[i], test, test_idct, speed); + } +#if CONFIG_FDCTDSP + else { + for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++) + err |= dct_error(&fdct_tab[i], test, test_idct, speed); + + for (i = 0; fdct_tab_arch[i].name; i++) + if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag)) + err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed); + } +#endif /* CONFIG_FDCTDSP */ + } + + if (err) + printf("Error: %d.\n", err); + + return !!err; +} |