static void subband_transform(DCAEncContext *c, const int32_t *input)
{
int ch, subs, i, k, j;
for (ch = 0; ch < c->fullband_channels; ch++) {
/* History is copied because it is also needed for PSY */
int32_t hist[512];
int hist_start = 0;
const int chi = c->channel_order_tab[ch];
for (i = 0; i < 512; i++)
hist[i] = c->history[i][ch];
for (subs = 0; subs < SUBBAND_SAMPLES; subs++) {
int32_t accum[64];
int32_t resp;
int band;
/* Calculate the convolutions at once */
for (i = 0; i < 64; i++)
accum[i] = 0;
for (k = 0, i = hist_start, j = 0;
i < 512; k = (k + 1) & 63, i++, j++)
accum[k] += mul32(hist[i], c->band_interpolation[j]);
for (i = 0; i < hist_start; k = (k + 1) & 63, i++, j++)
accum[k] += mul32(hist[i], c->band_interpolation[j]);
for (k = 16; k < 32; k++)
accum[k] = accum[k] - accum[31 - k];
for (k = 32; k < 48; k++)
accum[k] = accum[k] + accum[95 - k];
for (band = 0; band < 32; band++) {
resp = 0;
for (i = 16; i < 48; i++) {
int s = (2 * band + 1) * (2 * (i + 16) + 1);
resp += mul32(accum[i], cos_t(s << 3)) >> 3;
}
c->subband[subs][band][ch] = ((band + 1) & 2) ? -resp : resp;
}
/* Copy in 32 new samples from input */
for (i = 0; i < 32; i++)
hist[i + hist_start] = input[(subs * 32 + i) * c->channels + chi];
hist_start = (hist_start + 32) & 511;
}
}
}
static void fft(const int32_t in[2 * 256], cplx32 out[256])
{
cplx32 buf[256], rin[256], rout[256];
int i, j, k, l;
/* do two transforms in parallel */
for (i = 0; i < 256; i++) {
/* Apply the Hann window */
rin[i].re = mul32(in[2 * i], 0x3fffffff - (cos_t(8 * i + 2) >> 1));
rin[i].im = mul32(in[2 * i + 1], 0x3fffffff - (cos_t(8 * i + 6) >> 1));
}
/* pre-rotation */
for (i = 0; i < 256; i++) {
buf[i].re = mul32(cos_t(4 * i + 2), rin[i].re)
- mul32(sin_t(4 * i + 2), rin[i].im);
buf[i].im = mul32(cos_t(4 * i + 2), rin[i].im)
+ mul32(sin_t(4 * i + 2), rin[i].re);
}
for (j = 256, l = 1; j != 1; j >>= 1, l <<= 1) {
for (k = 0; k < 256; k += j) {
for (i = k; i < k + j / 2; i++) {
cplx32 sum, diff;
int t = 8 * l * i;
sum.re = buf[i].re + buf[i + j / 2].re;
sum.im = buf[i].im + buf[i + j / 2].im;
diff.re = buf[i].re - buf[i + j / 2].re;
diff.im = buf[i].im - buf[i + j / 2].im;
buf[i].re = half32(sum.re);
buf[i].im = half32(sum.im);
buf[i + j / 2].re = mul32(diff.re, cos_t(t))
- mul32(diff.im, sin_t(t));
buf[i + j / 2].im = mul32(diff.im, cos_t(t))
+ mul32(diff.re, sin_t(t));
}
}
}
/* post-rotation */
for (i = 0; i < 256; i++) {
int b = ff_reverse[i];
rout[i].re = mul32(buf[b].re, cos_t(4 * i))
- mul32(buf[b].im, sin_t(4 * i));
rout[i].im = mul32(buf[b].im, cos_t(4 * i))
+ mul32(buf[b].re, sin_t(4 * i));
}
for (i = 0; i < 256; i++) {
/* separate the results of the two transforms */
cplx32 o1, o2;
o1.re = rout[i].re - rout[255 - i].re;
o1.im = rout[i].im + rout[255 - i].im;
o2.re = rout[i].im - rout[255 - i].im;
o2.im = -rout[i].re - rout[255 - i].re;
/* combine them into one long transform */
out[i].re = mul32( o1.re + o2.re, cos_t(2 * i + 1))
+ mul32( o1.im - o2.im, sin_t(2 * i + 1));
out[i].im = mul32( o1.im + o2.im, cos_t(2 * i + 1))
+ mul32(-o1.re + o2.re, sin_t(2 * i + 1));
}
}
static inline int32_t sin_t(int x)
{
return cos_t(x - 512);
}