aacpsdsp.c 8.99 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/*
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * FFmpeg 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 General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License along
 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
 */

#include "libavcodec/aacpsdsp.h"
20
#include "libavutil/intfloat.h"
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

#include "checkasm.h"

#define N 32
#define STRIDE 128
#define BUF_SIZE (N * STRIDE)

#define randomize(buf, len) do {                                \
    int i;                                                      \
    for (i = 0; i < len; i++) {                                 \
        const INTFLOAT f = (INTFLOAT)rnd() / UINT_MAX;          \
        (buf)[i] = f;                                           \
    }                                                           \
} while (0)

#define EPS 0.005

38 39 40 41 42 43 44 45 46 47
static void clear_less_significant_bits(INTFLOAT *buf, int len, int bits)
{
    int i;
    for (i = 0; i < len; i++) {
        union av_intfloat32 u = { .f = buf[i] };
        u.i &= (0xffffffff << bits);
        buf[i] = u.f;
    }
}

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
static void test_add_squares(void)
{
    LOCAL_ALIGNED_16(INTFLOAT, dst0, [BUF_SIZE]);
    LOCAL_ALIGNED_16(INTFLOAT, dst1, [BUF_SIZE]);
    LOCAL_ALIGNED_16(INTFLOAT, src, [BUF_SIZE], [2]);

    declare_func(void, INTFLOAT *dst,
                 const INTFLOAT (*src)[2], int n);

    randomize((INTFLOAT *)src, BUF_SIZE * 2);
    randomize(dst0, BUF_SIZE);
    memcpy(dst1, dst0, BUF_SIZE * sizeof(INTFLOAT));
    call_ref(dst0, src, BUF_SIZE);
    call_new(dst1, src, BUF_SIZE);
    if (!float_near_abs_eps_array(dst0, dst1, EPS, BUF_SIZE))
        fail();
    bench_new(dst1, src, BUF_SIZE);
}

static void test_mul_pair_single(void)
{
    LOCAL_ALIGNED_16(INTFLOAT, dst0, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, dst1, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, src0, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, src1, [BUF_SIZE]);

    declare_func(void, INTFLOAT (*dst)[2],
                       INTFLOAT (*src0)[2], INTFLOAT *src1, int n);

    randomize((INTFLOAT *)src0, BUF_SIZE * 2);
    randomize(src1, BUF_SIZE);
    call_ref(dst0, src0, src1, BUF_SIZE);
    call_new(dst1, src0, src1, BUF_SIZE);
    if (!float_near_abs_eps_array((float *)dst0, (float *)dst1, EPS, BUF_SIZE * 2))
        fail();
    bench_new(dst1, src0, src1, BUF_SIZE);
}

static void test_hybrid_analysis(void)
{
    LOCAL_ALIGNED_16(INTFLOAT, dst0, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, dst1, [BUF_SIZE], [2]);
90
    LOCAL_ALIGNED_16(INTFLOAT, in, [13], [2]);
91 92 93 94
    LOCAL_ALIGNED_16(INTFLOAT, filter, [N], [8][2]);

    declare_func(void, INTFLOAT (*out)[2], INTFLOAT (*in)[2],
                 const INTFLOAT (*filter)[8][2],
95
                 ptrdiff_t stride, int n);
96

97
    randomize((INTFLOAT *)in, 13 * 2);
98 99 100 101 102 103 104 105 106 107 108 109 110
    randomize((INTFLOAT *)filter, N * 8 * 2);

    randomize((INTFLOAT *)dst0, BUF_SIZE * 2);
    memcpy(dst1, dst0, BUF_SIZE * 2 * sizeof(INTFLOAT));

    call_ref(dst0, in, filter, STRIDE, N);
    call_new(dst1, in, filter, STRIDE, N);

    if (!float_near_abs_eps_array((float *)dst0, (float *)dst1, EPS, BUF_SIZE * 2))
        fail();
    bench_new(dst1, in, filter, STRIDE, N);
}

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
static void test_hybrid_analysis_ileave(void)
{
    LOCAL_ALIGNED_16(INTFLOAT, in,   [2], [38][64]);
    LOCAL_ALIGNED_16(INTFLOAT, out0, [91], [32][2]);
    LOCAL_ALIGNED_16(INTFLOAT, out1, [91], [32][2]);

    declare_func(void, INTFLOAT (*out)[32][2], INTFLOAT L[2][38][64],
                       int i, int len);

    randomize((INTFLOAT *)out0, 91 * 32 * 2);
    randomize((INTFLOAT *)in,    2 * 38 * 64);
    memcpy(out1, out0, 91 * 32 * 2 * sizeof(INTFLOAT));

    /* len is hardcoded to 32 as that's the only value used in
       libavcodec. asm functions are likely to be optimized
       hardcoding this value in their loops and could fail with
       anything else.
       i is hardcoded to the two values currently used by the
       aac decoder because the arm neon implementation is
       micro-optimized for them and will fail for almost every
       other value. */
    call_ref(out0, in, 3, 32);
    call_new(out1, in, 3, 32);

    /* the function just moves data around, so memcmp is enough */
    if (memcmp(out0, out1, 91 * 32 * 2 * sizeof(INTFLOAT)))
        fail();

    call_ref(out0, in, 5, 32);
    call_new(out1, in, 5, 32);

    if (memcmp(out0, out1, 91 * 32 * 2 * sizeof(INTFLOAT)))
        fail();

    bench_new(out1, in, 3, 32);
}

static void test_hybrid_synthesis_deint(void)
{
    LOCAL_ALIGNED_16(INTFLOAT, out0, [2], [38][64]);
    LOCAL_ALIGNED_16(INTFLOAT, out1, [2], [38][64]);
    LOCAL_ALIGNED_16(INTFLOAT, in,  [91], [32][2]);

    declare_func(void, INTFLOAT out[2][38][64], INTFLOAT (*in)[32][2],
                       int i, int len);

    randomize((INTFLOAT *)in,  91 * 32 * 2);
    randomize((INTFLOAT *)out0, 2 * 38 * 64);
    memcpy(out1, out0, 2 * 38 * 64 * sizeof(INTFLOAT));

    /* len is hardcoded to 32 as that's the only value used in
       libavcodec. asm functions are likely to be optimized
       hardcoding this value in their loops and could fail with
       anything else.
       i is hardcoded to the two values currently used by the
       aac decoder because the arm neon implementation is
       micro-optimized for them and will fail for almost every
       other value. */
    call_ref(out0, in, 3, 32);
    call_new(out1, in, 3, 32);

    /* the function just moves data around, so memcmp is enough */
    if (memcmp(out0, out1, 2 * 38 * 64 * sizeof(INTFLOAT)))
        fail();

    call_ref(out0, in, 5, 32);
    call_new(out1, in, 5, 32);

    if (memcmp(out0, out1, 2 * 38 * 64 * sizeof(INTFLOAT)))
        fail();

    bench_new(out1, in, 3, 32);
}

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
static void test_stereo_interpolate(PSDSPContext *psdsp)
{
    int i;
    LOCAL_ALIGNED_16(INTFLOAT, l,  [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, r,  [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, l0, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, r0, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, l1, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, r1, [BUF_SIZE], [2]);
    LOCAL_ALIGNED_16(INTFLOAT, h, [2], [4]);
    LOCAL_ALIGNED_16(INTFLOAT, h_step, [2], [4]);

    declare_func(void, INTFLOAT (*l)[2], INTFLOAT (*r)[2],
                       INTFLOAT h[2][4], INTFLOAT h_step[2][4], int len);

    randomize((INTFLOAT *)l, BUF_SIZE * 2);
    randomize((INTFLOAT *)r, BUF_SIZE * 2);

    for (i = 0; i < 2; i++) {
        if (check_func(psdsp->stereo_interpolate[i], "ps_stereo_interpolate%s", i ? "_ipdopd" : "")) {
            memcpy(l0, l, BUF_SIZE * 2 * sizeof(INTFLOAT));
            memcpy(l1, l, BUF_SIZE * 2 * sizeof(INTFLOAT));
            memcpy(r0, r, BUF_SIZE * 2 * sizeof(INTFLOAT));
            memcpy(r1, r, BUF_SIZE * 2 * sizeof(INTFLOAT));

            randomize((INTFLOAT *)h, 2 * 4);
            randomize((INTFLOAT *)h_step, 2 * 4);
212 213 214 215 216 217 218
            // Clear the least significant 14 bits of h_step, to avoid
            // divergence when accumulating h_step BUF_SIZE times into
            // a float variable which may or may not have extra intermediate
            // precision. Therefore clear roughly log2(BUF_SIZE) less
            // significant bits, to get the same result regardless of any
            // extra precision in the accumulator.
            clear_less_significant_bits((INTFLOAT *)h_step, 2 * 4, 14);
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

            call_ref(l0, r0, h, h_step, BUF_SIZE);
            call_new(l1, r1, h, h_step, BUF_SIZE);
            if (!float_near_abs_eps_array((float *)l0, (float *)l1, EPS, BUF_SIZE * 2) ||
                !float_near_abs_eps_array((float *)r0, (float *)r1, EPS, BUF_SIZE * 2))
                fail();

            memcpy(l1, l, BUF_SIZE * 2 * sizeof(INTFLOAT));
            memcpy(r1, r, BUF_SIZE * 2 * sizeof(INTFLOAT));
            bench_new(l1, r1, h, h_step, BUF_SIZE);
        }
    }
}

void checkasm_check_aacpsdsp(void)
{
    PSDSPContext psdsp;

    ff_psdsp_init(&psdsp);

    if (check_func(psdsp.add_squares, "ps_add_squares"))
        test_add_squares();
    report("add_squares");

    if (check_func(psdsp.mul_pair_single, "ps_mul_pair_single"))
        test_mul_pair_single();
    report("mul_pair_single");

    if (check_func(psdsp.hybrid_analysis, "ps_hybrid_analysis"))
        test_hybrid_analysis();
    report("hybrid_analysis");

251 252 253 254 255 256 257 258
    if (check_func(psdsp.hybrid_analysis_ileave, "ps_hybrid_analysis_ileave"))
        test_hybrid_analysis_ileave();
    report("hybrid_analysis_ileave");

    if (check_func(psdsp.hybrid_synthesis_deint, "ps_hybrid_synthesis_deint"))
        test_hybrid_synthesis_deint();
    report("hybrid_synthesis_deint");

259 260 261
    test_stereo_interpolate(&psdsp);
    report("stereo_interpolate");
}