imdct15.c 8.42 KB
Newer Older
1 2 3
/*
 * Copyright (c) 2013-2014 Mozilla Corporation
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7 8 9 10
 * 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.
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12 13 14 15 16
 * 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
17
 * License along with FFmpeg; if not, write to the Free Software
18 19 20 21 22 23 24 25 26 27
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * Celt non-power of 2 iMDCT
 */

#include <float.h>
#include <math.h>
28 29 30
#include <stddef.h>

#include "config.h"
31 32 33 34

#include "libavutil/attributes.h"
#include "libavutil/common.h"

35
#include "avfft.h"
36
#include "imdct15.h"
37
#include "opus.h"
38 39 40

// minimal iMDCT size to make SIMD opts easier
#define CELT_MIN_IMDCT_SIZE 120
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

// complex c = a * b
#define CMUL3(cre, cim, are, aim, bre, bim)          \
do {                                                 \
    cre = are * bre - aim * bim;                     \
    cim = are * bim + aim * bre;                     \
} while (0)

#define CMUL(c, a, b) CMUL3((c).re, (c).im, (a).re, (a).im, (b).re, (b).im)

// complex c = a * b
//         d = a * conjugate(b)
#define CMUL2(c, d, a, b)                            \
do {                                                 \
    float are = (a).re;                              \
    float aim = (a).im;                              \
    float bre = (b).re;                              \
    float bim = (b).im;                              \
    float rr  = are * bre;                           \
    float ri  = are * bim;                           \
    float ir  = aim * bre;                           \
    float ii  = aim * bim;                           \
    (c).re =  rr - ii;                               \
    (c).im =  ri + ir;                               \
    (d).re =  rr + ii;                               \
    (d).im = -ri + ir;                               \
} while (0)

69
av_cold void ff_imdct15_uninit(IMDCT15Context **ps)
70
{
71
    IMDCT15Context *s = *ps;
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
    int i;

    if (!s)
        return;

    for (i = 0; i < FF_ARRAY_ELEMS(s->exptab); i++)
        av_freep(&s->exptab[i]);

    av_freep(&s->twiddle_exptab);

    av_freep(&s->tmp);

    av_freep(ps);
}

87 88
static void imdct15_half(IMDCT15Context *s, float *dst, const float *src,
                         ptrdiff_t stride, float scale);
89

90
av_cold int ff_imdct15_init(IMDCT15Context **ps, int N)
91
{
92
    IMDCT15Context *s;
93 94 95 96
    int len2 = 15 * (1 << N);
    int len  = 2 * len2;
    int i, j;

97
    if (len2 > CELT_MAX_FRAME_SIZE || len2 < CELT_MIN_IMDCT_SIZE)
98 99 100 101 102 103 104 105 106 107
        return AVERROR(EINVAL);

    s = av_mallocz(sizeof(*s));
    if (!s)
        return AVERROR(ENOMEM);

    s->fft_n = N - 1;
    s->len4 = len2 / 2;
    s->len2 = len2;

108
    s->tmp  = av_malloc_array(len, 2 * sizeof(*s->tmp));
109 110 111
    if (!s->tmp)
        goto fail;

112
    s->twiddle_exptab  = av_malloc_array(s->len4, sizeof(*s->twiddle_exptab));
113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
    if (!s->twiddle_exptab)
        goto fail;

    for (i = 0; i < s->len4; i++) {
        s->twiddle_exptab[i].re = cos(2 * M_PI * (i + 0.125 + s->len4) / len);
        s->twiddle_exptab[i].im = sin(2 * M_PI * (i + 0.125 + s->len4) / len);
    }

    for (i = 0; i < FF_ARRAY_ELEMS(s->exptab); i++) {
        int N = 15 * (1 << i);
        s->exptab[i] = av_malloc(sizeof(*s->exptab[i]) * FFMAX(N, 19));
        if (!s->exptab[i])
            goto fail;

        for (j = 0; j < N; j++) {
            s->exptab[i][j].re = cos(2 * M_PI * j / N);
            s->exptab[i][j].im = sin(2 * M_PI * j / N);
        }
    }

    // wrap around to simplify fft15
    for (j = 15; j < 19; j++)
        s->exptab[0][j] = s->exptab[0][j - 15];

137
    s->imdct_half = imdct15_half;
138 139

    if (ARCH_AARCH64)
140
        ff_imdct15_init_aarch64(s);
141

142 143 144
    *ps = s;

    return 0;
145

146
fail:
147
    ff_imdct15_uninit(&s);
148 149 150
    return AVERROR(ENOMEM);
}

151
static void fft5(FFTComplex *out, const FFTComplex *in, ptrdiff_t stride)
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
{
    // [0] = exp(2 * i * pi / 5), [1] = exp(2 * i * pi * 2 / 5)
    static const FFTComplex fact[] = { { 0.30901699437494745,  0.95105651629515353 },
                                       { -0.80901699437494734, 0.58778525229247325 } };

    FFTComplex z[4][4];

    CMUL2(z[0][0], z[0][3], in[1 * stride], fact[0]);
    CMUL2(z[0][1], z[0][2], in[1 * stride], fact[1]);
    CMUL2(z[1][0], z[1][3], in[2 * stride], fact[0]);
    CMUL2(z[1][1], z[1][2], in[2 * stride], fact[1]);
    CMUL2(z[2][0], z[2][3], in[3 * stride], fact[0]);
    CMUL2(z[2][1], z[2][2], in[3 * stride], fact[1]);
    CMUL2(z[3][0], z[3][3], in[4 * stride], fact[0]);
    CMUL2(z[3][1], z[3][2], in[4 * stride], fact[1]);

    out[0].re = in[0].re + in[stride].re + in[2 * stride].re + in[3 * stride].re + in[4 * stride].re;
    out[0].im = in[0].im + in[stride].im + in[2 * stride].im + in[3 * stride].im + in[4 * stride].im;

    out[1].re = in[0].re + z[0][0].re + z[1][1].re + z[2][2].re + z[3][3].re;
    out[1].im = in[0].im + z[0][0].im + z[1][1].im + z[2][2].im + z[3][3].im;

    out[2].re = in[0].re + z[0][1].re + z[1][3].re + z[2][0].re + z[3][2].re;
    out[2].im = in[0].im + z[0][1].im + z[1][3].im + z[2][0].im + z[3][2].im;

    out[3].re = in[0].re + z[0][2].re + z[1][0].re + z[2][3].re + z[3][1].re;
    out[3].im = in[0].im + z[0][2].im + z[1][0].im + z[2][3].im + z[3][1].im;

    out[4].re = in[0].re + z[0][3].re + z[1][2].re + z[2][1].re + z[3][0].re;
    out[4].im = in[0].im + z[0][3].im + z[1][2].im + z[2][1].im + z[3][0].im;
}

184 185
static void fft15(IMDCT15Context *s, FFTComplex *out, const FFTComplex *in,
                  ptrdiff_t stride)
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
{
    const FFTComplex *exptab = s->exptab[0];
    FFTComplex tmp[5];
    FFTComplex tmp1[5];
    FFTComplex tmp2[5];
    int k;

    fft5(tmp,  in,              stride * 3);
    fft5(tmp1, in +     stride, stride * 3);
    fft5(tmp2, in + 2 * stride, stride * 3);

    for (k = 0; k < 5; k++) {
        FFTComplex t1, t2;

        CMUL(t1, tmp1[k], exptab[k]);
        CMUL(t2, tmp2[k], exptab[2 * k]);
        out[k].re = tmp[k].re + t1.re + t2.re;
        out[k].im = tmp[k].im + t1.im + t2.im;

        CMUL(t1, tmp1[k], exptab[k + 5]);
        CMUL(t2, tmp2[k], exptab[2 * (k + 5)]);
        out[k + 5].re = tmp[k].re + t1.re + t2.re;
        out[k + 5].im = tmp[k].im + t1.im + t2.im;

        CMUL(t1, tmp1[k], exptab[k + 10]);
        CMUL(t2, tmp2[k], exptab[2 * k + 5]);
        out[k + 10].re = tmp[k].re + t1.re + t2.re;
        out[k + 10].im = tmp[k].im + t1.im + t2.im;
    }
}

/*
 * FFT of the length 15 * (2^N)
 */
220
static void fft_calc(IMDCT15Context *s, FFTComplex *out, const FFTComplex *in,
221
                     int N, ptrdiff_t stride)
222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245
{
    if (N) {
        const FFTComplex *exptab = s->exptab[N];
        const int len2 = 15 * (1 << (N - 1));
        int k;

        fft_calc(s, out,        in,          N - 1, stride * 2);
        fft_calc(s, out + len2, in + stride, N - 1, stride * 2);

        for (k = 0; k < len2; k++) {
            FFTComplex t;

            CMUL(t, out[len2 + k], exptab[k]);

            out[len2 + k].re = out[k].re - t.re;
            out[len2 + k].im = out[k].im - t.im;

            out[k].re += t.re;
            out[k].im += t.im;
        }
    } else
        fft15(s, out, in, stride);
}

246 247
static void imdct15_half(IMDCT15Context *s, float *dst, const float *src,
                         ptrdiff_t stride, float scale)
248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274
{
    FFTComplex *z = (FFTComplex *)dst;
    const int len8 = s->len4 / 2;
    const float *in1 = src;
    const float *in2 = src + (s->len2 - 1) * stride;
    int i;

    for (i = 0; i < s->len4; i++) {
        FFTComplex tmp = { *in2, *in1 };
        CMUL(s->tmp[i], tmp, s->twiddle_exptab[i]);
        in1 += 2 * stride;
        in2 -= 2 * stride;
    }

    fft_calc(s, z, s->tmp, s->fft_n, 1);

    for (i = 0; i < len8; i++) {
        float r0, i0, r1, i1;

        CMUL3(r0, i1, z[len8 - i - 1].im, z[len8 - i - 1].re,  s->twiddle_exptab[len8 - i - 1].im, s->twiddle_exptab[len8 - i - 1].re);
        CMUL3(r1, i0, z[len8 + i].im,     z[len8 + i].re,      s->twiddle_exptab[len8 + i].im,     s->twiddle_exptab[len8 + i].re);
        z[len8 - i - 1].re = scale * r0;
        z[len8 - i - 1].im = scale * i0;
        z[len8 + i].re     = scale * r1;
        z[len8 + i].im     = scale * i1;
    }
}