float_dsp-test.c 9.58 KB
Newer Older
1
/*
2
 * This file is part of FFmpeg.
3
 *
4
 * FFmpeg is free software; you can redistribute it and/or
5 6 7 8
 * 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.
 *
9
 * FFmpeg is distributed in the hope that it will be useful,
10 11 12 13 14
 * 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
15
 * License along with FFmpeg; if not, write to the Free Software
16 17 18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

19 20
#include "config.h"

21
#include <float.h>
22
#include <math.h>
23 24 25
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
26 27 28 29 30 31 32 33
#if HAVE_UNISTD_H
#include <unistd.h> /* for getopt */
#endif
#if !HAVE_GETOPT
#include "compat/getopt.c"
#endif

#include "common.h"
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
#include "cpu.h"
#include "internal.h"
#include "lfg.h"
#include "log.h"
#include "random_seed.h"
#include "float_dsp.h"

#define LEN 240

static void fill_float_array(AVLFG *lfg, float *a, int len)
{
    int i;
    double bmg[2], stddev = 10.0, mean = 0.0;

    for (i = 0; i < len; i += 2) {
        av_bmg_get(lfg, bmg);
        a[i]     = bmg[0] * stddev + mean;
        a[i + 1] = bmg[1] * stddev + mean;
    }
}
static int compare_floats(const float *a, const float *b, int len,
                          float max_diff)
{
    int i;
    for (i = 0; i < len; i++) {
        if (fabsf(a[i] - b[i]) > max_diff) {
            av_log(NULL, AV_LOG_ERROR, "%d: %- .12f - %- .12f = % .12g\n",
                   i, a[i], b[i], a[i] - b[i]);
            return -1;
        }
    }
    return 0;
}

static void fill_double_array(AVLFG *lfg, double *a, int len)
{
    int i;
    double bmg[2], stddev = 10.0, mean = 0.0;

    for (i = 0; i < len; i += 2) {
        av_bmg_get(lfg, bmg);
        a[i]     = bmg[0] * stddev + mean;
        a[i + 1] = bmg[1] * stddev + mean;
    }
}

static int compare_doubles(const double *a, const double *b, int len,
                           double max_diff)
{
    int i;

    for (i = 0; i < len; i++) {
        if (fabs(a[i] - b[i]) > max_diff) {
            av_log(NULL, AV_LOG_ERROR, "%d: %- .12f - %- .12f = % .12g\n",
                   i, a[i], b[i], a[i] - b[i]);
            return -1;
        }
    }
    return 0;
}

static int test_vector_fmul(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                            const float *v1, const float *v2)
{
    LOCAL_ALIGNED(32, float, cdst, [LEN]);
    LOCAL_ALIGNED(32, float, odst, [LEN]);
    int ret;

    cdsp->vector_fmul(cdst, v1, v2, LEN);
    fdsp->vector_fmul(odst, v1, v2, LEN);

    if (ret = compare_floats(cdst, odst, LEN, FLT_EPSILON))
        av_log(NULL, AV_LOG_ERROR, "vector_fmul failed\n");

    return ret;
}

#define ARBITRARY_FMAC_SCALAR_CONST 0.005
static int test_vector_fmac_scalar(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                   const float *v1, const float *src0, float scale)
{
    LOCAL_ALIGNED(32, float, cdst, [LEN]);
    LOCAL_ALIGNED(32, float, odst, [LEN]);
    int ret;

    memcpy(cdst, v1, LEN * sizeof(*v1));
    memcpy(odst, v1, LEN * sizeof(*v1));

    cdsp->vector_fmac_scalar(cdst, src0, scale, LEN);
    fdsp->vector_fmac_scalar(odst, src0, scale, LEN);

    if (ret = compare_floats(cdst, odst, LEN, ARBITRARY_FMAC_SCALAR_CONST))
        av_log(NULL, AV_LOG_ERROR, "vector_fmac_scalar failed\n");

    return ret;
}

static int test_vector_fmul_scalar(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                   const float *v1, float scale)
{
    LOCAL_ALIGNED(32, float, cdst, [LEN]);
    LOCAL_ALIGNED(32, float, odst, [LEN]);
    int ret;

    cdsp->vector_fmul_scalar(cdst, v1, scale, LEN);
    fdsp->vector_fmul_scalar(odst, v1, scale, LEN);

    if (ret = compare_floats(cdst, odst, LEN, FLT_EPSILON))
        av_log(NULL, AV_LOG_ERROR, "vector_fmul_scalar failed\n");

    return ret;
}

static int test_vector_dmul_scalar(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                   const double *v1, double scale)
{
    LOCAL_ALIGNED(32, double, cdst, [LEN]);
    LOCAL_ALIGNED(32, double, odst, [LEN]);
    int ret;

    cdsp->vector_dmul_scalar(cdst, v1, scale, LEN);
    fdsp->vector_dmul_scalar(odst, v1, scale, LEN);

    if (ret = compare_doubles(cdst, odst, LEN, DBL_EPSILON))
        av_log(NULL, AV_LOG_ERROR, "vector_dmul_scalar failed\n");

    return ret;
}

#define ARBITRARY_FMUL_WINDOW_CONST 0.008
static int test_vector_fmul_window(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                   const float *v1, const float *v2, const float *v3)
{
    LOCAL_ALIGNED(32, float, cdst, [LEN]);
    LOCAL_ALIGNED(32, float, odst, [LEN]);
    int ret;

    cdsp->vector_fmul_window(cdst, v1, v2, v3, LEN / 2);
    fdsp->vector_fmul_window(odst, v1, v2, v3, LEN / 2);

    if (ret = compare_floats(cdst, odst, LEN, ARBITRARY_FMUL_WINDOW_CONST))
        av_log(NULL, AV_LOG_ERROR, "vector_fmul_window failed\n");

    return ret;
}

#define ARBITRARY_FMUL_ADD_CONST 0.005
static int test_vector_fmul_add(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                const float *v1, const float *v2, const float *v3)
{
    LOCAL_ALIGNED(32, float, cdst, [LEN]);
    LOCAL_ALIGNED(32, float, odst, [LEN]);
    int ret;

    cdsp->vector_fmul_add(cdst, v1, v2, v3, LEN);
    fdsp->vector_fmul_add(odst, v1, v2, v3, LEN);

    if (ret = compare_floats(cdst, odst, LEN, ARBITRARY_FMUL_ADD_CONST))
        av_log(NULL, AV_LOG_ERROR, "vector_fmul_add failed\n");

    return ret;
}

static int test_vector_fmul_reverse(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                    const float *v1, const float *v2)
{
    LOCAL_ALIGNED(32, float, cdst, [LEN]);
    LOCAL_ALIGNED(32, float, odst, [LEN]);
    int ret;

    cdsp->vector_fmul_reverse(cdst, v1, v2, LEN);
    fdsp->vector_fmul_reverse(odst, v1, v2, LEN);

    if (ret = compare_floats(cdst, odst, LEN, FLT_EPSILON))
        av_log(NULL, AV_LOG_ERROR, "vector_fmul_reverse failed\n");

    return ret;
}

static int test_butterflies_float(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                  const float *v1, const float *v2)
{
    LOCAL_ALIGNED(32, float, cv1, [LEN]);
    LOCAL_ALIGNED(32, float, cv2, [LEN]);
    LOCAL_ALIGNED(32, float, ov1, [LEN]);
    LOCAL_ALIGNED(32, float, ov2, [LEN]);
    int ret;

    memcpy(cv1, v1, LEN * sizeof(*v1));
    memcpy(cv2, v2, LEN * sizeof(*v2));
    memcpy(ov1, v1, LEN * sizeof(*v1));
    memcpy(ov2, v2, LEN * sizeof(*v2));

    cdsp->butterflies_float(cv1, cv2, LEN);
    fdsp->butterflies_float(ov1, ov2, LEN);

    if ((ret = compare_floats(cv1, ov1, LEN, FLT_EPSILON)) ||
        (ret = compare_floats(cv2, ov2, LEN, FLT_EPSILON)))
        av_log(NULL, AV_LOG_ERROR, "butterflies_float failed\n");

    return ret;
}

#define ARBITRARY_SCALARPRODUCT_CONST 0.2
static int test_scalarproduct_float(AVFloatDSPContext *fdsp, AVFloatDSPContext *cdsp,
                                    const float *v1, const float *v2)
{
    float cprod, oprod;
    int ret;

    cprod = cdsp->scalarproduct_float(v1, v2, LEN);
    oprod = fdsp->scalarproduct_float(v1, v2, LEN);

    if (ret = compare_floats(&cprod, &oprod, 1, ARBITRARY_SCALARPRODUCT_CONST))
        av_log(NULL, AV_LOG_ERROR, "scalarproduct_float failed\n");

    return ret;
}

int main(int argc, char **argv)
{
255
    int ret = 0, seeded = 0;
256
    uint32_t seed;
257
    AVFloatDSPContext *fdsp, *cdsp;
258 259 260 261 262 263 264 265
    AVLFG lfg;

    LOCAL_ALIGNED(32, float, src0, [LEN]);
    LOCAL_ALIGNED(32, float, src1, [LEN]);
    LOCAL_ALIGNED(32, float, src2, [LEN]);
    LOCAL_ALIGNED(32, double, dbl_src0, [LEN]);
    LOCAL_ALIGNED(32, double, dbl_src1, [LEN]);

266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
    for (;;) {
        int arg = getopt(argc, argv, "s:c:");
        if (arg == -1)
            break;
        switch (arg) {
        case 's':
            seed = strtoul(optarg, NULL, 10);
            seeded = 1;
            break;
        case 'c':
        {
            int cpuflags = av_get_cpu_flags();

            if (av_parse_cpu_caps(&cpuflags, optarg) < 0)
                return 1;

            av_force_cpu_flags(cpuflags);
            break;
        }
        }
    }
    if (!seeded)
288 289
        seed = av_get_random_seed();

290 291 292 293 294 295 296 297 298 299
    av_log(NULL, AV_LOG_INFO, "float_dsp-test: %s %u\n", seeded ? "seed" : "random seed", seed);

    fdsp = avpriv_float_dsp_alloc(1);
    av_force_cpu_flags(0);
    cdsp = avpriv_float_dsp_alloc(1);

    if (!fdsp || !cdsp) {
        ret = 1;
        goto end;
    }
300 301 302 303 304 305 306 307 308 309

    av_lfg_init(&lfg, seed);

    fill_float_array(&lfg, src0, LEN);
    fill_float_array(&lfg, src1, LEN);
    fill_float_array(&lfg, src2, LEN);

    fill_double_array(&lfg, dbl_src0, LEN);
    fill_double_array(&lfg, dbl_src1, LEN);

310
    if (test_vector_fmul(fdsp, cdsp, src0, src1))
311
        ret -= 1 << 0;
312
    if (test_vector_fmac_scalar(fdsp, cdsp, src2, src0, src1[0]))
313
        ret -= 1 << 1;
314
    if (test_vector_fmul_scalar(fdsp, cdsp, src0, src1[0]))
315
        ret -= 1 << 2;
316
    if (test_vector_fmul_window(fdsp, cdsp, src0, src1, src2))
317
        ret -= 1 << 3;
318
    if (test_vector_fmul_add(fdsp, cdsp, src0, src1, src2))
319
        ret -= 1 << 4;
320
    if (test_vector_fmul_reverse(fdsp, cdsp, src0, src1))
321
        ret -= 1 << 5;
322
    if (test_butterflies_float(fdsp, cdsp, src0, src1))
323
        ret -= 1 << 6;
324
    if (test_scalarproduct_float(fdsp, cdsp, src0, src1))
325
        ret -= 1 << 7;
326
    if (test_vector_dmul_scalar(fdsp, cdsp, dbl_src0, dbl_src1[0]))
327 328
        ret -= 1 << 8;

329 330 331
end:
    av_freep(&fdsp);
    av_freep(&cdsp);
332 333
    return ret;
}