// create an instance of the decoder
// blocksize is fixed over the lifetime of this object for performance reasons
decoder_impl(unsigned blocksize=8192): N(blocksize), halfN(blocksize/2) {
#ifdef USE_FFTW3
// create FFTW buffers
lt = (float*)fftwf_malloc(sizeof(float)*N);
rt = (float*)fftwf_malloc(sizeof(float)*N);
dst = (float*)fftwf_malloc(sizeof(float)*N);
dftL = (fftwf_complex*)fftwf_malloc(sizeof(fftwf_complex)*N);
dftR = (fftwf_complex*)fftwf_malloc(sizeof(fftwf_complex)*N);
src = (fftwf_complex*)fftwf_malloc(sizeof(fftwf_complex)*N);
loadL = fftwf_plan_dft_r2c_1d(N, lt, dftL,FFTW_MEASURE);
loadR = fftwf_plan_dft_r2c_1d(N, rt, dftR,FFTW_MEASURE);
store = fftwf_plan_dft_c2r_1d(N, src, dst,FFTW_MEASURE);
#else
// create lavc fft buffers
lt = (float*)av_malloc(sizeof(FFTSample)*N);
rt = (float*)av_malloc(sizeof(FFTSample)*N);
dftL = (FFTComplexArray*)av_malloc(sizeof(FFTComplex)*N*2);
dftR = (FFTComplexArray*)av_malloc(sizeof(FFTComplex)*N*2);
src = (FFTComplexArray*)av_malloc(sizeof(FFTComplex)*N*2);
fftContextForward = (FFTContext*)av_malloc(sizeof(FFTContext));
memset(fftContextForward, 0, sizeof(FFTContext));
fftContextReverse = (FFTContext*)av_malloc(sizeof(FFTContext));
memset(fftContextReverse, 0, sizeof(FFTContext));
ff_fft_init(fftContextForward, 13, 0);
ff_fft_init(fftContextReverse, 13, 1);
#endif
// resize our own buffers
frontR.resize(N);
frontL.resize(N);
avg.resize(N);
surR.resize(N);
surL.resize(N);
trueavg.resize(N);
xfs.resize(N);
yfs.resize(N);
inbuf[0].resize(N);
inbuf[1].resize(N);
for (unsigned c=0;c<6;c++) {
outbuf[c].resize(N);
filter[c].resize(N);
}
sample_rate(48000);
// generate the window function (square root of hann, b/c it is applied before and after the transform)
wnd.resize(N);
for (unsigned k=0;k<N;k++)
wnd[k] = sqrt(0.5*(1-cos(2*PI*k/N))/N);
current_buf = 0;
memset(inbufs, 0, sizeof(inbufs));
memset(outbufs, 0, sizeof(outbufs));
// set the default coefficients
surround_coefficients(0.8165,0.5774);
phase_mode(0);
separation(1,1);
steering_mode(true);
}
/**
* init MDCT or IMDCT computation.
*/
int ff_mdct_init(MDCTContext *s, int nbits, int inverse)
{
int n, n4, i;
double alpha;
memset(s, 0, sizeof(*s));
n = 1 << nbits;
s->nbits = nbits;
s->n = n;
n4 = n >> 2;
s->tcos = av_malloc(n4 * sizeof(FFTSample));
if (!s->tcos)
goto fail;
s->tsin = av_malloc(n4 * sizeof(FFTSample));
if (!s->tsin)
goto fail;
for(i=0;i<n4;i++) {
alpha = 2 * M_PI * (i + 1.0 / 8.0) / n;
s->tcos[i] = -cos(alpha);
s->tsin[i] = -sin(alpha);
}
if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
goto fail;
return 0;
fail:
av_freep(&s->tcos);
av_freep(&s->tsin);
return -1;
}
static int init_cook_mlt(COOKContext *q) {
int j;
float alpha;
/* Allocate the buffers, could be replaced with a static [512]
array if needed. */
q->mlt_size = q->samples_per_channel;
q->mlt_window = av_malloc(sizeof(float)*q->mlt_size);
q->mlt_precos = av_malloc(sizeof(float)*q->mlt_size/2);
q->mlt_presin = av_malloc(sizeof(float)*q->mlt_size/2);
q->mlt_postcos = av_malloc(sizeof(float)*q->mlt_size/2);
/* Initialize the MLT window: simple sine window. */
alpha = M_PI / (2.0 * (float)q->mlt_size);
for(j=0 ; j<q->mlt_size ; j++) {
q->mlt_window[j] = sin((j + 512.0/(float)q->mlt_size) * alpha);
}
/* pre/post twiddle factors */
for (j=0 ; j<q->mlt_size/2 ; j++){
q->mlt_precos[j] = cos( ((j+0.25)*M_PI)/q->mlt_size);
q->mlt_presin[j] = sin( ((j+0.25)*M_PI)/q->mlt_size);
q->mlt_postcos[j] = (float)sqrt(2.0/(float)q->mlt_size)*cos( ((float)j*M_PI) /q->mlt_size); //sqrt(2/MLT_size) = scalefactor
}
/* Initialize the FFT. */
ff_fft_init(&q->fft_ctx, av_log2(q->mlt_size)-1, 0);
av_log(NULL,AV_LOG_DEBUG,"FFT initialized, order = %d.\n",
av_log2(q->samples_per_channel)-1);
return (int)(q->mlt_window && q->mlt_precos && q->mlt_presin && q->mlt_postcos);
}
static av_cold int imc_decode_init(AVCodecContext * avctx)
{
int i, j;
IMCContext *q = avctx->priv_data;
double r1, r2;
q->decoder_reset = 1;
for(i = 0; i < BANDS; i++)
q->old_floor[i] = 1.0;
/* Build mdct window, a simple sine window normalized with sqrt(2) */
ff_sine_window_init(q->mdct_sine_window, COEFFS);
for(i = 0; i < COEFFS; i++)
q->mdct_sine_window[i] *= sqrt(2.0);
for(i = 0; i < COEFFS/2; i++){
q->post_cos[i] = cos(i / 256.0 * M_PI);
q->post_sin[i] = sin(i / 256.0 * M_PI);
r1 = sin((i * 4.0 + 1.0) / 1024.0 * M_PI);
r2 = cos((i * 4.0 + 1.0) / 1024.0 * M_PI);
if (i & 0x1)
{
q->pre_coef1[i] = (r1 + r2) * sqrt(2.0);
q->pre_coef2[i] = -(r1 - r2) * sqrt(2.0);
}
else
{
q->pre_coef1[i] = -(r1 + r2) * sqrt(2.0);
q->pre_coef2[i] = (r1 - r2) * sqrt(2.0);
}
q->last_fft_im[i] = 0;
}
/* Generate a square root table */
for(i = 0; i < 30; i++) {
q->sqrt_tab[i] = sqrt(i);
}
/* initialize the VLC tables */
for(i = 0; i < 4 ; i++) {
for(j = 0; j < 4; j++) {
huffman_vlc[i][j].table = vlc_tables[vlc_offsets[i * 4 + j]];
huffman_vlc[i][j].table_allocated = vlc_offsets[i * 4 + j + 1] - vlc_offsets[i * 4 + j];
init_vlc(&huffman_vlc[i][j], 9, imc_huffman_sizes[i],
imc_huffman_lens[i][j], 1, 1,
imc_huffman_bits[i][j], 2, 2, INIT_VLC_USE_NEW_STATIC);
}
}
q->one_div_log2 = 1/log(2);
ff_fft_init(&q->fft, 7, 1);
dsputil_init(&q->dsp, avctx);
avctx->sample_fmt = SAMPLE_FMT_S16;
return 0;
}
static inline void fft_init(FFTContext **s, int nbits, int inverse)
{
#if AVFFT
*s = av_fft_init(nbits, inverse);
#else
ff_fft_init(*s, nbits, inverse);
#endif
}
FFTContext *av_fft_init(int nbits, int inverse)
{
FFTContext *s = av_mallocz(sizeof(*s));
if (s && ff_fft_init(s, nbits, inverse))
av_freep(&s);
return s;
}
FFTContext *av_fft_init(int nbits, int inverse)
{
FFTContext *s = av_malloc(sizeof(*s));
if (s)
ff_fft_init(s, nbits, inverse);
return s;
}
static int imc_decode_init(AVCodecContext * avctx)
{
int i, j;
IMCContext *q = avctx->priv_data;
double r1, r2;
q->decoder_reset = 1;
for(i = 0; i < BANDS; i++)
q->old_floor[i] = 1.0;
/* Build mdct window, a simple sine window normalized with sqrt(2) */
for(i = 0; i < COEFFS; i++)
q->mdct_sine_window[i] = sin((i + 0.5) / 512.0 * M_PI) * sqrt(2.0);
for(i = 0; i < COEFFS/2; i++){
q->post_cos[i] = cos(i / 256.0 * M_PI);
q->post_sin[i] = sin(i / 256.0 * M_PI);
r1 = sin((i * 4.0 + 1.0) / 1024.0 * M_PI);
r2 = cos((i * 4.0 + 1.0) / 1024.0 * M_PI);
if (i & 0x1)
{
q->pre_coef1[i] = (r1 + r2) * sqrt(2.0);
q->pre_coef2[i] = -(r1 - r2) * sqrt(2.0);
}
else
{
q->pre_coef1[i] = -(r1 + r2) * sqrt(2.0);
q->pre_coef2[i] = (r1 - r2) * sqrt(2.0);
}
q->last_fft_im[i] = 0;
}
/* Generate a square root table */
for(i = 0; i < 30; i++) {
q->sqrt_tab[i] = sqrt(i);
}
/* initialize the VLC tables */
for(i = 0; i < 4 ; i++) {
for(j = 0; j < 4; j++) {
init_vlc (&q->huffman_vlc[i][j], 9, imc_huffman_sizes[i],
imc_huffman_lens[i][j], 1, 1,
imc_huffman_bits[i][j], 2, 2, 1);
}
}
q->one_div_log2 = 1/log(2);
ff_fft_init(&q->fft, 7, 1);
dsputil_init(&q->dsp, avctx);
return 0;
}
FFTContext *av_fft_init(int nbits, int inverse)
{
FFTContext *s = (FFTContext*)malloc(sizeof(*s));
if (s && ff_fft_init(s, nbits, inverse)) {
free(s);
s = 0;
}
return s;
}
int main (void)
{
#define PRECISION 16
#define FFT_SIZE 1024
#define ftofix32(x) ((fixed32)((x) * (float)(1 << PRECISION) + ((x) < 0 ? -0.5 : 0.5)))
#define itofix32(x) ((x) << PRECISION)
#define fixtoi32(x) ((x) >> PRECISION)
int j;
const long N = FFT_SIZE;
double r[FFT_SIZE] = {0.0}, i[FFT_SIZE] = {0.0};
long n;
double t;
double amp, phase;
clock_t start, end;
double exec_time = 0;
FFTContext s;
FFTComplex z[FFT_SIZE];
memset(z, 0, 64*sizeof(FFTComplex));
/* Generate saw-tooth test data */
for (n = 0; n < FFT_SIZE; n++)
{
t = (2 * M_PI * n)/N;
/*z[n].re = 1.1 + sin( t) +
0.5 * sin(2.0 * t) +
(1.0/3.0) * sin(3.0 * t) +
0.25 * sin(4.0 * t) +
0.2 * sin(5.0 * t) +
(1.0/6.0) * sin(6.0 * t) +
(1.0/7.0) * sin(7.0 * t) ;*/
z[n].re = ftofix32(cos(2*M_PI*n/64));
//printf("z[%d] = %f\n", n, z[n].re);
//getchar();
}
ff_fft_init(&s, 10, 1);
//start = clock();
//for(n = 0; n < 1000000; n++)
ff_fft_permute_c(&s, z);
ff_fft_calc_c(&s, z);
//end = clock();
//exec_time = (((double)end-(double)start)/CLOCKS_PER_SEC);
for(j = 0; j < FFT_SIZE; j++)
{
printf("%8.4f\n", sqrt(pow(fixtof32(z[j].re),2)+ pow(fixtof32(z[j].im), 2)));
//getchar();
}
printf("muls = %d, adds = %d\n", muls, adds);
//printf(" Time elapsed = %f\n", exec_time);
//ff_fft_end(&s);
}
/**
* init MDCT or IMDCT computation.
*/
av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale)
{
int n, n4, i;
double alpha, theta;
int tstep;
memset(s, 0, sizeof(*s));
n = 1 << nbits;
s->mdct_bits = nbits;
s->mdct_size = n;
n4 = n >> 2;
s->mdct_permutation = FF_MDCT_PERM_NONE;
if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0)
goto fail;
s->tcos = av_malloc_array(n/2, sizeof(FFTSample));
if (!s->tcos)
goto fail;
switch (s->mdct_permutation) {
case FF_MDCT_PERM_NONE:
s->tsin = s->tcos + n4;
tstep = 1;
break;
case FF_MDCT_PERM_INTERLEAVE:
s->tsin = s->tcos + 1;
tstep = 2;
break;
default:
goto fail;
}
theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
scale = sqrt(fabs(scale));
for(i=0;i<n4;i++) {
alpha = 2 * M_PI * (i + theta) / n;
#if FFT_FIXED_32
s->tcos[i*tstep] = lrint(-cos(alpha) * 2147483648.0);
s->tsin[i*tstep] = lrint(-sin(alpha) * 2147483648.0);
#else
s->tcos[i*tstep] = FIX15(-cos(alpha) * scale);
s->tsin[i*tstep] = FIX15(-sin(alpha) * scale);
#endif
}
return 0;
fail:
ff_mdct_end(s);
return -1;
}
av_cold int ff_dct_init(DCTContext *s, int nbits, int inverse)
{
int n = 1 << nbits;
s->nbits = nbits;
s->inverse = inverse;
s->data = (struct FFTComplex *) av_malloc(sizeof(FFTComplex) * 2 * n);
if (!s->data)
return -1;
if (ff_fft_init(&s->fft, nbits+1, inverse) < 0)
return -1;
return 0;
}
int main(int argc, char **argv)
{
FFTComplex *tab, *tab1, *tab_ref;
FFTSample *tab2;
int it, i, c;
int do_speed = 0;
int do_mdct = 0;
int do_inverse = 0;
FFTContext s1, *s = &s1;
MDCTContext m1, *m = &m1;
int fft_nbits, fft_size;
fft_nbits = 9;
for(;;) {
c = getopt(argc, argv, "hsimn:");
if (c == -1)
break;
switch(c) {
case 'h':
help();
break;
case 's':
do_speed = 1;
break;
case 'i':
do_inverse = 1;
break;
case 'm':
do_mdct = 1;
break;
case 'n':
fft_nbits = atoi(optarg);
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 (do_mdct) {
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);
} else {
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);
fft_ref_init(fft_nbits, do_inverse);
}
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();
tab1[i].im = frandom();
}
/* checking result */
av_log(NULL, AV_LOG_INFO,"Checking...\n");
if (do_mdct) {
if (do_inverse) {
imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
ff_imdct_calc(m, tab2, (float *)tab1);
check_diff((float *)tab_ref, tab2, fft_size);
} else {
mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
ff_mdct_calc(m, tab2, (float *)tab1);
check_diff((float *)tab_ref, tab2, fft_size / 2);
}
} else {
memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
ff_fft_permute(s, tab);
ff_fft_calc(s, tab);
fft_ref(tab_ref, tab1, fft_nbits);
check_diff((float *)tab_ref, (float *)tab, fft_size * 2);
}
/* 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 = gettime();
for(it=0;it<nb_its;it++) {
if (do_mdct) {
if (do_inverse) {
ff_imdct_calc(m, (float *)tab, (float *)tab1);
} else {
ff_mdct_calc(m, (float *)tab, (float *)tab1);
}
} else {
memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
ff_fft_calc(s, tab);
}
}
duration = gettime() - 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);
}
if (do_mdct) {
ff_mdct_end(m);
} else {
ff_fft_end(s);
}
return 0;
}
int main(int argc, char **argv)
{
FFTComplex *tab, *tab1, *tab_ref;
FFTSample *tab2;
int it, i, c;
int do_speed = 0;
int err = 1;
enum tf_transform transform = TRANSFORM_FFT;
int do_inverse = 0;
FFTContext s1, *s = &s1;
FFTContext m1, *m = &m1;
RDFTContext r1, *r = &r1;
DCTContext d1, *d = &d1;
int fft_nbits, fft_size, fft_size_2;
double scale = 1.0;
AVLFG prng;
av_lfg_init(&prng, 1);
fft_nbits = 9;
for(;;) {
c = getopt(argc, argv, "hsimrdn:f:");
if (c == -1)
break;
switch(c) {
case 'h':
help();
break;
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;
}
}
fft_size = 1 << fft_nbits;
fft_size_2 = fft_size >> 1;
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));
switch (transform) {
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;
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);
fft_ref_init(fft_nbits, do_inverse);
break;
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);
fft_ref_init(fft_nbits, do_inverse);
break;
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;
}
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) {
case TRANSFORM_MDCT:
if (do_inverse) {
imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
ff_imdct_calc(m, tab2, (float *)tab1);
err = check_diff((float *)tab_ref, tab2, fft_size, scale);
} else {
mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
ff_mdct_calc(m, tab2, (float *)tab1);
err = check_diff((float *)tab_ref, tab2, fft_size / 2, scale);
}
break;
case TRANSFORM_FFT:
memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
ff_fft_permute(s, tab);
ff_fft_calc(s, tab);
fft_ref(tab_ref, tab1, fft_nbits);
err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 1.0);
break;
case TRANSFORM_RDFT:
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;
ff_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((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
} else {
for (i = 0; i < fft_size; i++) {
tab2[i] = tab1[i].re;
tab1[i].im = 0;
}
ff_rdft_calc(r, tab2);
fft_ref(tab_ref, tab1, fft_nbits);
tab_ref[0].im = tab_ref[fft_size_2].re;
err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0);
}
break;
case TRANSFORM_DCT:
memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
ff_dct_calc(d, tab);
if (do_inverse) {
idct_ref(tab_ref, tab1, fft_nbits);
} else {
dct_ref(tab_ref, tab1, fft_nbits);
}
err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
break;
}
/* 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 = gettime();
for (it = 0; it < nb_its; it++) {
switch (transform) {
case TRANSFORM_MDCT:
if (do_inverse) {
ff_imdct_calc(m, (float *)tab, (float *)tab1);
} else {
ff_mdct_calc(m, (float *)tab, (float *)tab1);
}
break;
case TRANSFORM_FFT:
memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
ff_fft_calc(s, tab);
break;
case TRANSFORM_RDFT:
memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
ff_rdft_calc(r, tab2);
break;
case TRANSFORM_DCT:
memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
ff_dct_calc(d, tab2);
break;
}
}
duration = gettime() - 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) {
case TRANSFORM_MDCT:
ff_mdct_end(m);
break;
case TRANSFORM_FFT:
ff_fft_end(s);
break;
case TRANSFORM_RDFT:
ff_rdft_end(r);
break;
case TRANSFORM_DCT:
ff_dct_end(d);
break;
}
av_free(tab);
av_free(tab1);
av_free(tab2);
av_free(tab_ref);
av_free(exptab);
return err;
}
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;
#if CONFIG_LIBFFTW3
FFTWContext fftw;
FFTWComplex *tab_fftw, *tab_fftw_copy;
#endif /* CONFIG_LIBFFTW3 */
#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, "hsitmrdn: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;
#if CONFIG_LIBFFTW3
case 't':
transform = TRANSFORM_FFTW;
break;
#endif /* CONFIG_LIBFFTW3 */
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':
{
unsigned cpuflags = av_get_cpu_flags();
if (av_parse_cpu_caps(&cpuflags, optarg) < 0)
return 1;
av_force_cpu_flags(cpuflags);
break;
}
}
}
fft_size = 1 << fft_nbits;
tab = av_malloc_array(fft_size, sizeof(FFTComplex));
tab1 = av_malloc_array(fft_size, sizeof(FFTComplex));
tab_ref = av_malloc_array(fft_size, sizeof(FFTComplex));
tab2 = av_malloc_array(fft_size, sizeof(FFTSample));
#if CONFIG_LIBFFTW3 && FFT_FLOAT
tab_fftw = av_malloc_array(fft_size, sizeof(*tab_fftw));
tab_fftw_copy = av_malloc_array(fft_size, sizeof(*tab_fftw_copy));
if (!(tab_fftw && tab_fftw_copy))
goto cleanup_fftw;
#endif /* CONFIG_LIBFFTW3 */
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 CONFIG_LIBFFTW3 && FFT_FLOAT
case TRANSFORM_FFTW:
if (do_inverse)
av_log(NULL, AV_LOG_INFO, "IFFTW");
else
av_log(NULL, AV_LOG_INFO, "FFTW");
ff_fftw_init(&fftw, fft_size, do_inverse);
if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0)
goto cleanup;
break;
#endif /* CONFIG_LIBFFTW3 */
#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);
#if CONFIG_LIBFFTW3 && FFT_FLOAT
tab_fftw[i][0] = tab1[i].re;
tab_fftw[i][1] = tab1[i].im;
#endif /* CONFIG_LIBFFTW3 */
}
#if CONFIG_LIBFFTW3 && FFT_FLOAT
memcpy(tab_fftw_copy, tab_fftw, fft_size * sizeof(*tab_fftw));
#endif
/* 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 CONFIG_LIBFFTW3 && FFT_FLOAT
case TRANSFORM_FFTW:
fftw.fft_calc(&fftw, tab_fftw);
fft_ref(tab_ref, tab1, fft_nbits);
for (i = 0; i < fft_size; i++) {
tab[i].re = tab_fftw[i][0];
tab[i].im = tab_fftw[i][1];
}
err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0);
break;
#endif /* CONFIG_LIBFFTW3 */
#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_permute(&s, tab);
s.fft_calc(&s, tab);
break;
#if CONFIG_LIBFFTW3 && FFT_FLOAT
case TRANSFORM_FFTW:
memcpy(tab_fftw, tab_fftw_copy, fft_size * sizeof(*tab_fftw));
fftw.fft_calc(&fftw, tab_fftw);
break;
#endif /* CONFIG_LIBFFTW3 */
#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: %0.2f 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 CONFIG_LIBFFTW3 && FFT_FLOAT
case TRANSFORM_FFTW:
ff_fftw_deinit(&fftw);
break;
#endif /* CONFIG_LIBFFTW3 */
#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 */
}
#if CONFIG_LIBFFTW3 && FFT_FLOAT
cleanup_fftw:
av_freep(&tab_fftw);
#endif /* CONFIG_LIBFFTW3 */
cleanup:
av_freep(&tab);
av_freep(&tab1);
av_freep(&tab2);
av_freep(&tab_ref);
av_freep(&exptab);
if (err)
printf("Error: %d.\n", err);
return !!err;
}
/**
* init MDCT or IMDCT computation.
*/
av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale)
{
int n, n4, i;
double alpha, theta;
int tstep;
memset(s, 0, sizeof(*s));
n = 1 << nbits;
s->mdct_bits = nbits;
s->mdct_size = n;
n4 = n >> 2;
s->mdct_permutation = FF_MDCT_PERM_NONE;
if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0)
goto fail;
s->imdct_calc = ff_imdct_calc_c;
s->imdct_half = ff_imdct_half_c;
s->mdct_calc = ff_mdct_calc_c;
#if FFT_FLOAT
if (ARCH_AARCH64)
ff_mdct_init_aarch64(s);
if (ARCH_ARM)
ff_mdct_init_arm(s);
if (ARCH_PPC)
ff_mdct_init_ppc(s);
if (ARCH_X86)
ff_mdct_init_x86(s);
s->mdct_calcw = s->mdct_calc;
#else
s->mdct_calcw = ff_mdct_calcw_c;
if (ARCH_ARM)
ff_mdct_fixed_init_arm(s);
#endif
s->tcos = av_malloc(n/2 * sizeof(FFTSample));
if (!s->tcos)
goto fail;
switch (s->mdct_permutation) {
case FF_MDCT_PERM_NONE:
s->tsin = s->tcos + n4;
tstep = 1;
break;
case FF_MDCT_PERM_INTERLEAVE:
s->tsin = s->tcos + 1;
tstep = 2;
break;
default:
goto fail;
}
theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
scale = sqrt(fabs(scale));
for(i=0;i<n4;i++) {
alpha = 2 * M_PI * (i + theta) / n;
s->tcos[i*tstep] = FIX15(-cos(alpha) * scale);
s->tsin[i*tstep] = FIX15(-sin(alpha) * scale);
}
return 0;
fail:
ff_mdct_end(s);
return -1;
}
av_cold int ff_mdct15_init(MDCT15Context **ps, int inverse, int N, double scale)
{
MDCT15Context *s;
double alpha, theta;
int len2 = 15 * (1 << N);
int len = 2 * len2;
int i;
/* Tested and verified to work on everything in between */
if ((N < 2) || (N > 13))
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;
s->inverse = inverse;
s->fft15 = fft15_c;
s->mdct = mdct15;
s->imdct_half = imdct15_half;
if (ff_fft_init(&s->ptwo_fft, N - 1, s->inverse) < 0)
goto fail;
if (init_pfa_reindex_tabs(s))
goto fail;
s->tmp = av_malloc_array(len, 2 * sizeof(*s->tmp));
if (!s->tmp)
goto fail;
s->twiddle_exptab = av_malloc_array(s->len4, sizeof(*s->twiddle_exptab));
if (!s->twiddle_exptab)
goto fail;
theta = 0.125f + (scale < 0 ? s->len4 : 0);
scale = sqrt(fabs(scale));
for (i = 0; i < s->len4; i++) {
alpha = 2 * M_PI * (i + theta) / len;
s->twiddle_exptab[i].re = cosf(alpha) * scale;
s->twiddle_exptab[i].im = sinf(alpha) * scale;
}
/* 15-point FFT exptab */
for (i = 0; i < 19; i++) {
if (i < 15) {
double theta = (2.0f * M_PI * i) / 15.0f;
if (!s->inverse)
theta *= -1;
s->exptab[i].re = cosf(theta);
s->exptab[i].im = sinf(theta);
} else { /* Wrap around to simplify fft15 */
s->exptab[i] = s->exptab[i - 15];
}
}
/* 5-point FFT exptab */
s->exptab[19].re = cosf(2.0f * M_PI / 5.0f);
s->exptab[19].im = sinf(2.0f * M_PI / 5.0f);
s->exptab[20].re = cosf(1.0f * M_PI / 5.0f);
s->exptab[20].im = sinf(1.0f * M_PI / 5.0f);
/* Invert the phase for an inverse transform, do nothing for a forward transform */
if (s->inverse) {
s->exptab[19].im *= -1;
s->exptab[20].im *= -1;
}
if (ARCH_X86)
ff_mdct15_init_x86(s);
*ps = s;
return 0;
fail:
ff_mdct15_uninit(&s);
return AVERROR(ENOMEM);
}
