tests: Move all test programs to a subdirectory

10 files changed, 1384 insertions, 0 deletions

diff --git a/libavcodec/tests/.gitignore b/libavcodec/tests/.gitignore
new file mode 100644
index 0000000000..31fa59b018
--- /dev/null
+++ b/libavcodec/tests/.gitignore
@@ -0,0 +1,6 @@
+/dct
+/fft
+/fft-fixed
+/golomb
+/iirfilter
+/rangecoder
diff --git a/libavcodec/tests/arm/dct.c b/libavcodec/tests/arm/dct.c
new file mode 100644
index 0000000000..d18cb52b60
--- /dev/null
+++ b/libavcodec/tests/arm/dct.c
@@ -0,0 +1,40 @@
+/*
+ * 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
+ */
+
+#include "config.h"
+
+#include "libavcodec/arm/idct.h"
+
+static const struct algo fdct_tab_arch[] = {
+    { 0 }
+};
+
+static const struct algo idct_tab_arch[] = {
+    { "SIMPLE-ARM",     ff_simple_idct_arm,     FF_IDCT_PERM_NONE },
+    { "INT-ARM",        ff_j_rev_dct_arm,       FF_IDCT_PERM_LIBMPEG2 },
+#if HAVE_ARMV5TE
+    { "SIMPLE-ARMV5TE", ff_simple_idct_armv5te, FF_IDCT_PERM_NONE,      AV_CPU_FLAG_ARMV5TE },
+#endif
+#if HAVE_ARMV6
+    { "SIMPLE-ARMV6",   ff_simple_idct_armv6,   FF_IDCT_PERM_LIBMPEG2,  AV_CPU_FLAG_ARMV6 },
+#endif
+#if HAVE_NEON
+    { "SIMPLE-NEON",    ff_simple_idct_neon,    FF_IDCT_PERM_PARTTRANS, AV_CPU_FLAG_NEON },
+#endif
+    { 0 }
+};
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;
+}
diff --git a/libavcodec/tests/fft-fixed.c b/libavcodec/tests/fft-fixed.c
new file mode 100644
index 0000000000..6edd810b25
--- /dev/null
+++ b/libavcodec/tests/fft-fixed.c
@@ -0,0 +1,20 @@
+/*
+ * 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
+ */
+
+#define FFT_FLOAT 0
+#include "fft.c"
diff --git a/libavcodec/tests/fft.c b/libavcodec/tests/fft.c
new file mode 100644
index 0000000000..db1ce9807f
--- /dev/null
+++ b/libavcodec/tests/fft.c
@@ -0,0 +1,513 @@
+/*
+ * (c) 2002 Fabrice Bellard
+ *
+ * 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
+ * FFT and MDCT tests.
+ */
+
+#include "config.h"
+
+#include <math.h>
+#if HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "libavutil/cpu.h"
+#include "libavutil/lfg.h"
+#include "libavutil/log.h"
+#include "libavutil/mathematics.h"
+#include "libavutil/time.h"
+
+#include "libavcodec/fft.h"
+#if FFT_FLOAT
+#include "libavcodec/dct.h"
+#include "libavcodec/rdft.h"
+#endif
+
+/* reference fft */
+
+#define MUL16(a, b) ((a) * (b))
+
+#define CMAC(pre, pim, are, aim, bre, bim)          \
+    {                                               \
+        pre += (MUL16(are, bre) - MUL16(aim, bim)); \
+        pim += (MUL16(are, bim) + MUL16(bre, aim)); \
+    }
+
+#if FFT_FLOAT
+#define RANGE 1.0
+#define REF_SCALE(x, bits)  (x)
+#define FMT "%10.6f"
+#else
+#define RANGE 16384
+#define REF_SCALE(x, bits) ((x) / (1 << (bits)))
+#define FMT "%6d"
+#endif
+
+static struct {
+    float re, im;
+} *exptab;
+
+static int fft_ref_init(int nbits, int inverse)
+{
+    int i, n = 1 << nbits;
+
+    exptab = av_malloc((n / 2) * sizeof(*exptab));
+    if (!exptab)
+        return AVERROR(ENOMEM);
+
+    for (i = 0; i < (n / 2); i++) {
+        double alpha = 2 * M_PI * (float) i / (float) n;
+        double c1 = cos(alpha), s1 = sin(alpha);
+        if (!inverse)
+            s1 = -s1;
+        exptab[i].re = c1;
+        exptab[i].im = s1;
+    }
+    return 0;
+}
+
+static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
+{
+    int i, j;
+    int n  = 1 << nbits;
+    int n2 = n >> 1;
+
+    for (i = 0; i < n; i++) {
+        double tmp_re = 0, tmp_im = 0;
+        FFTComplex *q = tab;
+        for (j = 0; j < n; j++) {
+            double s, c;
+            int k = (i * j) & (n - 1);
+            if (k >= n2) {
+                c = -exptab[k - n2].re;
+                s = -exptab[k - n2].im;
+            } else {
+                c = exptab[k].re;
+                s = exptab[k].im;
+            }
+            CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
+            q++;
+        }
+        tabr[i].re = REF_SCALE(tmp_re, nbits);
+        tabr[i].im = REF_SCALE(tmp_im, nbits);
+    }
+}
+
+#if CONFIG_MDCT
+static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
+{
+    int i, k, n = 1 << nbits;
+
+    for (i = 0; i < n; i++) {
+        double sum = 0;
+        for (k = 0; k < n / 2; k++) {
+            int a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
+            double f = cos(M_PI * a / (double) (2 * n));
+            sum += f * in[k];
+        }
+        out[i] = REF_SCALE(-sum, nbits - 2);
+    }
+}
+
+/* NOTE: no normalisation by 1 / N is done */
+static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
+{
+    int i, k, n = 1 << nbits;
+
+    /* do it by hand */
+    for (k = 0; k < n / 2; k++) {
+        double s = 0;
+        for (i = 0; i < n; i++) {
+            double a = (2 * M_PI * (2 * i + 1 + n / 2) * (2 * k + 1) / (4 * n));
+            s += input[i] * cos(a);
+        }
+        output[k] = REF_SCALE(s, nbits - 1);
+    }
+}
+#endif /* CONFIG_MDCT */
+
+#if FFT_FLOAT
+#if CONFIG_DCT
+static void idct_ref(float *output, float *input, int nbits)
+{
+    int i, k, n = 1 << nbits;
+
+    /* do it by hand */
+    for (i = 0; i < n; i++) {
+        double s = 0.5 * input[0];
+        for (k = 1; k < n; k++) {
+            double a = M_PI * k * (i + 0.5) / n;
+            s += input[k] * cos(a);
+        }
+        output[i] = 2 * s / n;
+    }
+}
+
+static void dct_ref(float *output, float *input, int nbits)
+{
+    int i, k, n = 1 << nbits;
+
+    /* do it by hand */
+    for (k = 0; k < n; k++) {
+        double s = 0;
+        for (i = 0; i < n; i++) {
+            double a = M_PI * k * (i + 0.5) / n;
+            s += input[i] * cos(a);
+        }
+        output[k] = s;
+    }
+}
+#endif /* CONFIG_DCT */
+#endif /* FFT_FLOAT */
+
+static FFTSample frandom(AVLFG *prng)
+{
+    return (int16_t) av_lfg_get(prng) / 32768.0 * RANGE;
+}
+
+static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
+{
+    int i, err = 0;
+    double error = 0, max = 0;
+
+    for (i = 0; i < n; i++) {
+        double e = fabs(tab1[i] - (tab2[i] / scale)) / RANGE;
+        if (e >= 1e-3) {
+            av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
+                   i, tab1[i], tab2[i]);
+            err = 1;
+        }
+        error += e * e;
+        if (e > max)
+            max = e;
+    }
+    av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error) / n);
+    return err;
+}
+
+static void help(void)
+{
+    av_log(NULL, AV_LOG_INFO,
+           "usage: fft-test [-h] [-s] [-i] [-n b]\n"
+           "-h     print this help\n"
+           "-s     speed test\n"
+           "-m     (I)MDCT test\n"
+           "-d     (I)DCT test\n"
+           "-r     (I)RDFT test\n"
+           "-i     inverse transform test\n"
+           "-n b   set the transform size to 2^b\n"
+           "-f x   set scale factor for output data of (I)MDCT to x\n");
+}
+
+enum tf_transform {
+    TRANSFORM_FFT,
+    TRANSFORM_MDCT,
+    TRANSFORM_RDFT,
+    TRANSFORM_DCT,
+};
+
+#if !HAVE_GETOPT
+#include "compat/getopt.c"
+#endif
+
+int main(int argc, char **argv)
+{
+    FFTComplex *tab, *tab1, *tab_ref;
+    FFTSample *tab2;
+    enum tf_transform transform = TRANSFORM_FFT;
+    FFTContext m, s;
+#if FFT_FLOAT
+    RDFTContext r;
+    DCTContext d;
+#endif /* FFT_FLOAT */
+    int it, i, err = 1;
+    int do_speed = 0, do_inverse = 0;
+    int fft_nbits = 9, fft_size;
+    double scale = 1.0;
+    AVLFG prng;
+
+    av_lfg_init(&prng, 1);
+
+    for (;;) {
+        int c = getopt(argc, argv, "hsimrdn:f:c:");
+        if (c == -1)
+            break;
+        switch (c) {
+        case 'h':
+            help();
+            return 1;
+        case 's':
+            do_speed = 1;
+            break;
+        case 'i':
+            do_inverse = 1;
+            break;
+        case 'm':
+            transform = TRANSFORM_MDCT;
+            break;
+        case 'r':
+            transform = TRANSFORM_RDFT;
+            break;
+        case 'd':
+            transform = TRANSFORM_DCT;
+            break;
+        case 'n':
+            fft_nbits = atoi(optarg);
+            break;
+        case 'f':
+            scale = atof(optarg);
+            break;
+        case 'c':
+        {
+            int cpuflags = av_parse_cpu_flags(optarg);
+            if (cpuflags < 0)
+                return 1;
+            av_set_cpu_flags_mask(cpuflags);
+            break;
+        }
+        }
+    }
+
+    fft_size = 1 << fft_nbits;
+    tab      = av_malloc(fft_size * sizeof(FFTComplex));
+    tab1     = av_malloc(fft_size * sizeof(FFTComplex));
+    tab_ref  = av_malloc(fft_size * sizeof(FFTComplex));
+    tab2     = av_malloc(fft_size * sizeof(FFTSample));
+
+    if (!(tab && tab1 && tab_ref && tab2))
+        goto cleanup;
+
+    switch (transform) {
+#if CONFIG_MDCT
+    case TRANSFORM_MDCT:
+        av_log(NULL, AV_LOG_INFO, "Scale factor is set to %f\n", scale);
+        if (do_inverse)
+            av_log(NULL, AV_LOG_INFO, "IMDCT");
+        else
+            av_log(NULL, AV_LOG_INFO, "MDCT");
+        ff_mdct_init(&m, fft_nbits, do_inverse, scale);
+        break;
+#endif /* CONFIG_MDCT */
+    case TRANSFORM_FFT:
+        if (do_inverse)
+            av_log(NULL, AV_LOG_INFO, "IFFT");
+        else
+            av_log(NULL, AV_LOG_INFO, "FFT");
+        ff_fft_init(&s, fft_nbits, do_inverse);
+        if (err = fft_ref_init(fft_nbits, do_inverse) < 0)
+            goto cleanup;
+        break;
+#if FFT_FLOAT
+#if CONFIG_RDFT
+    case TRANSFORM_RDFT:
+        if (do_inverse)
+            av_log(NULL, AV_LOG_INFO, "IDFT_C2R");
+        else
+            av_log(NULL, AV_LOG_INFO, "DFT_R2C");
+        ff_rdft_init(&r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
+        if (err = fft_ref_init(fft_nbits, do_inverse) < 0)
+            goto cleanup;
+        break;
+#endif /* CONFIG_RDFT */
+#if CONFIG_DCT
+    case TRANSFORM_DCT:
+        if (do_inverse)
+            av_log(NULL, AV_LOG_INFO, "DCT_III");
+        else
+            av_log(NULL, AV_LOG_INFO, "DCT_II");
+        ff_dct_init(&d, fft_nbits, do_inverse ? DCT_III : DCT_II);
+        break;
+#endif /* CONFIG_DCT */
+#endif /* FFT_FLOAT */
+    default:
+        av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
+        goto cleanup;
+    }
+    av_log(NULL, AV_LOG_INFO, " %d test\n", fft_size);
+
+    /* generate random data */
+
+    for (i = 0; i < fft_size; i++) {
+        tab1[i].re = frandom(&prng);
+        tab1[i].im = frandom(&prng);
+    }
+
+    /* checking result */
+    av_log(NULL, AV_LOG_INFO, "Checking...\n");
+
+    switch (transform) {
+#if CONFIG_MDCT
+    case TRANSFORM_MDCT:
+        if (do_inverse) {
+            imdct_ref(&tab_ref->re, &tab1->re, fft_nbits);
+            m.imdct_calc(&m, tab2, &tab1->re);
+            err = check_diff(&tab_ref->re, tab2, fft_size, scale);
+        } else {
+            mdct_ref(&tab_ref->re, &tab1->re, fft_nbits);
+            m.mdct_calc(&m, tab2, &tab1->re);
+            err = check_diff(&tab_ref->re, tab2, fft_size / 2, scale);
+        }
+        break;
+#endif /* CONFIG_MDCT */
+    case TRANSFORM_FFT:
+        memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
+        s.fft_permute(&s, tab);
+        s.fft_calc(&s, tab);
+
+        fft_ref(tab_ref, tab1, fft_nbits);
+        err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0);
+        break;
+#if FFT_FLOAT
+#if CONFIG_RDFT
+    case TRANSFORM_RDFT:
+    {
+        int fft_size_2 = fft_size >> 1;
+        if (do_inverse) {
+            tab1[0].im          = 0;
+            tab1[fft_size_2].im = 0;
+            for (i = 1; i < fft_size_2; i++) {
+                tab1[fft_size_2 + i].re =  tab1[fft_size_2 - i].re;
+                tab1[fft_size_2 + i].im = -tab1[fft_size_2 - i].im;
+            }
+
+            memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
+            tab2[1] = tab1[fft_size_2].re;
+
+            r.rdft_calc(&r, tab2);
+            fft_ref(tab_ref, tab1, fft_nbits);
+            for (i = 0; i < fft_size; i++) {
+                tab[i].re = tab2[i];
+                tab[i].im = 0;
+            }
+            err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 0.5);
+        } else {
+            for (i = 0; i < fft_size; i++) {
+                tab2[i]    = tab1[i].re;
+                tab1[i].im = 0;
+            }
+            r.rdft_calc(&r, tab2);
+            fft_ref(tab_ref, tab1, fft_nbits);
+            tab_ref[0].im = tab_ref[fft_size_2].re;
+            err = check_diff(&tab_ref->re, tab2, fft_size, 1.0);
+        }
+        break;
+    }
+#endif /* CONFIG_RDFT */
+#if CONFIG_DCT
+    case TRANSFORM_DCT:
+        memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
+        d.dct_calc(&d, &tab->re);
+        if (do_inverse)
+            idct_ref(&tab_ref->re, &tab1->re, fft_nbits);
+        else
+            dct_ref(&tab_ref->re, &tab1->re, fft_nbits);
+        err = check_diff(&tab_ref->re, &tab->re, fft_size, 1.0);
+        break;
+#endif /* CONFIG_DCT */
+#endif /* FFT_FLOAT */
+    }
+
+    /* do a speed test */
+
+    if (do_speed) {
+        int64_t time_start, duration;
+        int nb_its;
+
+        av_log(NULL, AV_LOG_INFO, "Speed test...\n");
+        /* we measure during about 1 seconds */
+        nb_its = 1;
+        for (;;) {
+            time_start = av_gettime_relative();
+            for (it = 0; it < nb_its; it++) {
+                switch (transform) {
+                case TRANSFORM_MDCT:
+                    if (do_inverse)
+                        m.imdct_calc(&m, &tab->re, &tab1->re);
+                    else
+                        m.mdct_calc(&m, &tab->re, &tab1->re);
+                    break;
+                case TRANSFORM_FFT:
+                    memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
+                    s.fft_calc(&s, tab);
+                    break;
+#if FFT_FLOAT
+                case TRANSFORM_RDFT:
+                    memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
+                    r.rdft_calc(&r, tab2);
+                    break;
+                case TRANSFORM_DCT:
+                    memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
+                    d.dct_calc(&d, tab2);
+                    break;
+#endif /* FFT_FLOAT */
+                }
+            }
+            duration = av_gettime_relative() - time_start;
+            if (duration >= 1000000)
+                break;
+            nb_its *= 2;
+        }
+        av_log(NULL, AV_LOG_INFO,
+               "time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
+               (double) duration / nb_its,
+               (double) duration / 1000000.0,
+               nb_its);
+    }
+
+    switch (transform) {
+#if CONFIG_MDCT
+    case TRANSFORM_MDCT:
+        ff_mdct_end(&m);
+        break;
+#endif /* CONFIG_MDCT */
+    case TRANSFORM_FFT:
+        ff_fft_end(&s);
+        break;
+#if FFT_FLOAT
+#if CONFIG_RDFT
+    case TRANSFORM_RDFT:
+        ff_rdft_end(&r);
+        break;
+#endif /* CONFIG_RDFT */
+#if CONFIG_DCT
+    case TRANSFORM_DCT:
+        ff_dct_end(&d);
+        break;
+#endif /* CONFIG_DCT */
+#endif /* FFT_FLOAT */
+    }
+
+cleanup:
+    av_free(tab);
+    av_free(tab1);
+    av_free(tab2);
+    av_free(tab_ref);
+    av_free(exptab);
+
+    if (err)
+        printf("Error: %d.\n", err);
+
+    return !!err;
+}
diff --git a/libavcodec/tests/golomb.c b/libavcodec/tests/golomb.c
new file mode 100644
index 0000000000..7587402e24
--- /dev/null
+++ b/libavcodec/tests/golomb.c
@@ -0,0 +1,97 @@
+/*
+ * 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
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+
+#include "libavutil/mem.h"
+
+#include "libavcodec/get_bits.h"
+#include "libavcodec/put_bits.h"
+#include "libavcodec/golomb.h"
+
+#define COUNT 8191
+#define SIZE (COUNT * 4)
+
+int main(void)
+{
+    int i, ret = 0;
+    uint8_t *temp;
+    PutBitContext pb;
+    GetBitContext gb;
+
+    temp = av_malloc(SIZE);
+    if (!temp)
+        return 2;
+
+    init_put_bits(&pb, temp, SIZE);
+    for (i = 0; i < COUNT; i++)
+        set_ue_golomb(&pb, i);
+    flush_put_bits(&pb);
+
+    init_get_bits(&gb, temp, 8 * SIZE);
+    for (i = 0; i < COUNT; i++) {
+        int j, s = show_bits(&gb, 25);
+
+        j = get_ue_golomb(&gb);
+        if (j != i) {
+            fprintf(stderr, "get_ue_golomb: expected %d, got %d. bits: %7x\n",
+                    i, j, s);
+            ret = 1;
+        }
+    }
+
+#define EXTEND(i) (i << 3 | i & 7)
+    init_put_bits(&pb, temp, SIZE);
+    for (i = 0; i < COUNT; i++)
+        set_ue_golomb(&pb, EXTEND(i));
+    flush_put_bits(&pb);
+
+    init_get_bits(&gb, temp, 8 * SIZE);
+    for (i = 0; i < COUNT; i++) {
+        int j, s = show_bits_long(&gb, 32);
+
+        j = get_ue_golomb_long(&gb);
+        if (j != EXTEND(i)) {
+            fprintf(stderr, "get_ue_golomb_long: expected %d, got %d. "
+                    "bits: %8x\n", EXTEND(i), j, s);
+            ret = 1;
+        }
+    }
+
+    init_put_bits(&pb, temp, SIZE);
+    for (i = 0; i < COUNT; i++)
+        set_se_golomb(&pb, i - COUNT / 2);
+    flush_put_bits(&pb);
+
+    init_get_bits(&gb, temp, 8 * SIZE);
+    for (i = 0; i < COUNT; i++) {
+        int j, s = show_bits(&gb, 25);
+
+        j = get_se_golomb(&gb);
+        if (j != i - COUNT / 2) {
+            fprintf(stderr, "get_se_golomb: expected %d, got %d. bits: %7x\n",
+                    i - COUNT / 2, j, s);
+            ret = 1;
+        }
+    }
+
+    av_free(temp);
+
+    return ret;
+}
diff --git a/libavcodec/tests/iirfilter.c b/libavcodec/tests/iirfilter.c
new file mode 100644
index 0000000000..a6001a3270
--- /dev/null
+++ b/libavcodec/tests/iirfilter.c
@@ -0,0 +1,54 @@
+/*
+ * 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
+ */
+
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+
+#include "libavutil/libm.h"
+
+#include "libavcodec/iirfilter.h"
+
+#define FILT_ORDER 4
+#define SIZE 1024
+
+int main(void)
+{
+    struct FFIIRFilterCoeffs *fcoeffs = NULL;
+    struct FFIIRFilterState  *fstate  = NULL;
+    float cutoff_coeff = 0.4;
+    int16_t x[SIZE], y[SIZE];
+    int i;
+
+    fcoeffs = ff_iir_filter_init_coeffs(NULL, FF_FILTER_TYPE_BUTTERWORTH,
+                                        FF_FILTER_MODE_LOWPASS, FILT_ORDER,
+                                        cutoff_coeff, 0.0, 0.0);
+    fstate  = ff_iir_filter_init_state(FILT_ORDER);
+
+    for (i = 0; i < SIZE; i++)
+        x[i] = lrint(0.75 * INT16_MAX * sin(0.5 * M_PI * i * i / SIZE));
+
+    ff_iir_filter(fcoeffs, fstate, SIZE, x, 1, y, 1);
+
+    for (i = 0; i < SIZE; i++)
+        printf("%6d %6d\n", x[i], y[i]);
+
+    ff_iir_filter_free_coeffs(fcoeffs);
+    ff_iir_filter_free_state(fstate);
+    return 0;
+}
diff --git a/libavcodec/tests/ppc/dct.c b/libavcodec/tests/ppc/dct.c
new file mode 100644
index 0000000000..3d160d33e0
--- /dev/null
+++ b/libavcodec/tests/ppc/dct.c
@@ -0,0 +1,32 @@
+/*
+ * 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
+ */
+
+#include "config.h"
+
+#include "libavcodec/ppc/fdct.h"
+
+static const struct algo fdct_tab_arch[] = {
+#if HAVE_ALTIVEC && HAVE_BIGENDIAN
+    { "altivecfdct", ff_fdct_altivec, FF_IDCT_PERM_NONE, AV_CPU_FLAG_ALTIVEC },
+#endif
+    { 0 }
+};
+
+static const struct algo idct_tab_arch[] = {
+    { 0 }
+};
diff --git a/libavcodec/tests/rangecoder.c b/libavcodec/tests/rangecoder.c
new file mode 100644
index 0000000000..26bb589928
--- /dev/null
+++ b/libavcodec/tests/rangecoder.c
@@ -0,0 +1,64 @@
+/*
+ * 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
+ */
+
+#include <stdint.h>
+#include <string.h>
+
+#include "libavutil/lfg.h"
+#include "libavutil/log.h"
+
+#include "libavcodec/rangecoder.h"
+
+#define SIZE 10240
+
+int main(void)
+{
+    RangeCoder c;
+    uint8_t b[9 * SIZE];
+    uint8_t r[9 * SIZE];
+    int i;
+    uint8_t state[10];
+    AVLFG prng;
+
+    av_lfg_init(&prng, 1);
+
+    ff_init_range_encoder(&c, b, SIZE);
+    ff_build_rac_states(&c, 0.05 * (1LL << 32), 128 + 64 + 32 + 16);
+
+    memset(state, 128, sizeof(state));
+
+    for (i = 0; i < SIZE; i++)
+        r[i] = av_lfg_get(&prng) % 7;
+
+    for (i = 0; i < SIZE; i++)
+        put_rac(&c, state, r[i] & 1);
+
+    ff_rac_terminate(&c);
+
+    ff_init_range_decoder(&c, b, SIZE);
+
+    memset(state, 128, sizeof(state));
+
+    for (i = 0; i < SIZE; i++)
+        if ((r[i] & 1) != get_rac(&c, state)) {
+            av_log(NULL, AV_LOG_ERROR, "rac failure at %d\n", i);
+            return 1;
+        }
+
+    return 0;
+}
diff --git a/libavcodec/tests/x86/dct.c b/libavcodec/tests/x86/dct.c
new file mode 100644
index 0000000000..2ddb55510e
--- /dev/null
+++ b/libavcodec/tests/x86/dct.c
@@ -0,0 +1,86 @@
+/*
+ * 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
+ */
+
+#include "config.h"
+
+#include "libavcodec/x86/fdct.h"
+#include "libavcodec/x86/xvididct.h"
+#include "libavcodec/x86/simple_idct.h"
+
+static const struct algo fdct_tab_arch[] = {
+#if HAVE_MMX_INLINE
+    { "MMX",    ff_fdct_mmx,    FF_IDCT_PERM_NONE, AV_CPU_FLAG_MMX },
+#endif
+#if HAVE_MMXEXT_INLINE
+    { "MMXEXT", ff_fdct_mmxext, FF_IDCT_PERM_NONE, AV_CPU_FLAG_MMXEXT },
+#endif
+#if HAVE_SSE2_INLINE
+    { "SSE2",   ff_fdct_sse2,   FF_IDCT_PERM_NONE, AV_CPU_FLAG_SSE2 },
+#endif
+    { 0 }
+};
+
+static const struct algo idct_tab_arch[] = {
+#if HAVE_MMX_INLINE
+    { "SIMPLE-MMX",  ff_simple_idct_mmx,  FF_IDCT_PERM_SIMPLE, AV_CPU_FLAG_MMX },
+#endif
+#if CONFIG_MPEG4_DECODER
+#if HAVE_MMX_INLINE
+    { "XVID-MMX",    ff_xvid_idct_mmx,    FF_IDCT_PERM_NONE,   AV_CPU_FLAG_MMX,    1 },
+#endif
+#if HAVE_MMXEXT_INLINE
+    { "XVID-MMXEXT", ff_xvid_idct_mmxext, FF_IDCT_PERM_NONE,   AV_CPU_FLAG_MMXEXT, 1 },
+#endif
+#if HAVE_SSE2_INLINE
+    { "XVID-SSE2",   ff_xvid_idct_sse2,   FF_IDCT_PERM_SSE2,   AV_CPU_FLAG_SSE2,   1 },
+#endif
+#endif /* CONFIG_MPEG4_DECODER */
+    { 0 }
+};
+
+static short idct_simple_mmx_perm[64] = {
+    0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
+    0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
+    0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
+    0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
+    0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
+    0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
+    0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
+    0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
+};
+
+static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
+
+static int permute_x86(int16_t dst[64], const int16_t src[64],
+                       enum idct_permutation_type perm_type)
+{
+    int i;
+
+    switch (perm_type) {
+    case FF_IDCT_PERM_SIMPLE:
+        for (i = 0; i < 64; i++)
+            dst[idct_simple_mmx_perm[i]] = src[i];
+        return 1;
+    case FF_IDCT_PERM_SSE2:
+        for (i = 0; i < 64; i++)
+            dst[(i & 0x38) | idct_sse2_row_perm[i & 7]] = src[i];
+        return 1;
+    }
+
+    return 0;
+}