static void auto_correlation(sbr_info *sbr, acorr_coef *ac,
qmf_t buffer[MAX_NTSRHFG][64],
uint8_t bd, uint8_t len)
{
real_t r01 = 0, r02 = 0, r11 = 0;
real_t tmp1, tmp2;
int8_t j;
uint8_t offset = sbr->tHFAdj;
const real_t rel = FRAC_CONST(0.999999); // 1 / (1 + 1e-6f);
for (j = offset; j < len + offset; j++)
{
real_t buf_j = ACDET_PRE(QMF_RE(buffer[j ][bd]));
real_t buf_j_1 = ACDET_PRE(QMF_RE(buffer[j-1][bd]));
real_t buf_j_2 = ACDET_PRE(QMF_RE(buffer[j-2][bd]));
r01 += MUL_F(buf_j , buf_j_1);
r02 += MUL_F(buf_j , buf_j_2);
r11 += MUL_F(buf_j_1, buf_j_1);
}
tmp1 = ACDET_PRE(QMF_RE(buffer[len+offset-1][bd]));
tmp2 = ACDET_PRE(QMF_RE(buffer[ offset-1][bd]));
RE(ac->r12) = r01 - MUL_F(tmp1, tmp1) + MUL_F(tmp2, tmp2);
tmp1 = ACDET_PRE(QMF_RE(buffer[len+offset-2][bd]));
tmp2 = ACDET_PRE(QMF_RE(buffer[ offset-2][bd]));
RE(ac->r22) = r11 - MUL_F(tmp1, tmp1) + MUL_F(tmp2, tmp2);
RE(ac->r01) = r01;
RE(ac->r02) = r02;
RE(ac->r11) = r11;
ac->det = MUL_F(RE(ac->r11), RE(ac->r22)) - MUL_F(MUL_F(RE(ac->r12), RE(ac->r12)), rel);
ac->det = ACDET_POST(ac->det);
}
mdct_info *faad_mdct_init(uint16_t N)
{
uint16_t k;
#ifdef FIXED_POINT
uint16_t N_idx;
real_t cangle, sangle, c, s, cold;
#endif
real_t scale;
mdct_info *mdct = (mdct_info*)faad_malloc(sizeof(mdct_info));
assert(N % 8 == 0);
mdct->N = N;
mdct->sincos = (complex_t*)faad_malloc(N/4*sizeof(complex_t));
#ifdef FIXED_POINT
N_idx = map_N_to_idx(N);
scale = const_tab[N_idx][0];
cangle = const_tab[N_idx][1];
sangle = const_tab[N_idx][2];
c = const_tab[N_idx][3];
s = const_tab[N_idx][4];
#else
scale = (real_t)sqrt(2.0 / (real_t)N);
#endif
/* (co)sine table build using recurrence relations */
/* this can also be done using static table lookup or */
/* some form of interpolation */
for (k = 0; k < N/4; k++)
{
#ifdef FIXED_POINT
RE(mdct->sincos[k]) = c; //MUL_C_C(c,scale);
IM(mdct->sincos[k]) = s; //MUL_C_C(s,scale);
cold = c;
c = MUL_F(c,cangle) - MUL_F(s,sangle);
s = MUL_F(s,cangle) + MUL_F(cold,sangle);
#else
/* no recurrence, just sines */
RE(mdct->sincos[k]) = scale*(real_t)(cos(2.0*M_PI*(k+1./8.) / (real_t)N));
IM(mdct->sincos[k]) = scale*(real_t)(sin(2.0*M_PI*(k+1./8.) / (real_t)N));
#endif
}
/* initialise fft */
mdct->cfft = cffti(N/4);
#ifdef PROFILE
mdct->cycles = 0;
mdct->fft_cycles = 0;
#endif
return mdct;
}
static void AAC_AddNMul(real_t* Dst,const real_t* a,const real_t* b,const real_t* c,int n)
{
do
{
Dst[0] = a[0]+MUL_F(-b[-1],c[0]);
Dst[1] = a[1]+MUL_F(-b[-2],c[1]);
Dst[2] = a[2]+MUL_F(-b[-3],c[2]);
Dst[3] = a[3]+MUL_F(-b[-4],c[3]);
a += 4; b -= 4; c += 4; Dst += 4; n -= 4;
} while (n>0);
}
static void AAC_MulR(real_t* Dst,const real_t* a,const real_t* b,int n)
{
do
{
Dst[0] = MUL_F(a[-1],b[-1]);
Dst[1] = MUL_F(a[-2],b[-2]);
Dst[2] = MUL_F(a[-3],b[-3]);
Dst[3] = MUL_F(a[-4],b[-4]);
a -= 4; b -= 4; Dst += 4; n -= 4;
} while (n>0);
}
void* downMix( NeAACDecHandle hDecoder, real_t **input, uint16_t frame_len, uint8_t channel)
{
real_t *a,*b,*c;
int n;
a = hDecoder->time_out[hDecoder->internal_channel[0]]; //C
b = hDecoder->time_out[hDecoder->internal_channel[channel?4:3]]; //S
c = hDecoder->time_out[hDecoder->internal_channel[channel?2:1]]; //cum
for (n=0;n<frame_len;++n)
{
real_t cum = c[n] + MUL_F(a[n],RSQRT2) + MUL_F(b[n],RSQRT2);
c[n] = MUL_F(cum, DM_MUL);
}
return c;
}
/*
* Parameter:
* coderInfo(IO)(O:coderInfo->scale_factor)
* xr(unused)
* xr_pow(IO)
* xi(O)
* xmin(I)
*/
static int32_t FixNoise(CoderInfo *coderInfo,
coef_t *xr_pow,
int32_t *xi,
real_32_t *xmin)
{
register int32_t i, sb;
int32_t start, end;
static real_32_t log_ifqstep = REAL_CONST(7.6943735514); // 1.0 / log(ifqstep)
static real_32_t realconst1 = REAL_CONST(0.6931471806); //log2
static real_32_t realconst2 = REAL_CONST(0.5);
for (sb = 0; sb < coderInfo->nr_of_sfb; sb++) {
eng_t eng = 0;
frac_t maxfixstep;
start = coderInfo->sfb_offset[sb];
end = coderInfo->sfb_offset[sb+1];
for ( i=start; i< end; i++)
eng += (int64_t)(COEF2INT(xr_pow[i]))*COEF2INT(xr_pow[i]);
if ( (eng == 0) || (xmin[sb]==0) )
maxfixstep = FRAC_ICONST(1);
else {
maxfixstep = faac_sqrt(eng/(end-start)*
MUL_R(xmin[sb],faac_sqrt(xmin[sb]))
);
if ( maxfixstep == 0 )
maxfixstep = FRAC_ICONST(1);
else
maxfixstep = ((real_t)1<<(FRAC_BITS+REAL_BITS))/maxfixstep;
}
#ifdef DUMP_MAXFIXSTEP
printf("sb = %d, maxfixstep = %.8f\n",sb, FRAC2FLOAT(maxfixstep));
#endif
for (i = start; i < end; i++) {
#ifdef DUMP_MAXFIXSTEP
printf("xr_pow[%d] = %.8f\t",i,COEF2FLOAT(xr_pow[i]));
#endif
xr_pow[i] = MUL_F(xr_pow[i],maxfixstep);
#ifdef DUMP_MAXFIXSTEP
printf("xr_pow[%d]*fix = %.8f\n",i,COEF2FLOAT(xr_pow[i]));
#endif
}
QuantizeBand(xr_pow, xi, start, end);
coderInfo->scale_factor[sb] = (int32_t)REAL2INT(MUL_R(faac_log(maxfixstep)-FRAC2REAL_BIT*realconst1,
log_ifqstep) - realconst2)+1;
#ifdef DUMP_MAXFIXSTEP
printf("scale_factor = %d\n", coderInfo->scale_factor[sb]);
#endif
}
#ifdef DUMP_MAXFIXSTEP
exit(1);
#endif
return 0;
}
static void AAC_MulNMul(real_t* Dst,const real_t* b1,const real_t* c1,const real_t* b2,const real_t* c2,int n)
{
do
{
Dst[0] = MUL_F(b1[-1],c1[-1])+MUL_F(-b2[-1],c2[0]);
Dst[1] = MUL_F(b1[-2],c1[-2])+MUL_F(-b2[-2],c2[1]);
Dst[2] = MUL_F(b1[-3],c1[-3])+MUL_F(-b2[-3],c2[2]);
Dst[3] = MUL_F(b1[-4],c1[-4])+MUL_F(-b2[-4],c2[3]);
b1 -= 4; b2 -= 4; c1 -= 4; c2 += 4; Dst += 4; n -= 4;
} while (n>0);
}
static void AAC_AddMulMul(real_t* Dst,const real_t* a,const real_t* b1,const real_t* c1,const real_t* b2,const real_t* c2,int n)
{
do
{
Dst[0] = a[0]+MUL_F(b1[0],c1[-1])+MUL_F(b2[0],c2[0]);
Dst[1] = a[1]+MUL_F(b1[1],c1[-2])+MUL_F(b2[1],c2[1]);
Dst[2] = a[2]+MUL_F(b1[2],c1[-3])+MUL_F(b2[2],c2[2]);
Dst[3] = a[3]+MUL_F(b1[3],c1[-4])+MUL_F(b2[3],c2[3]);
a += 4; b1 += 4; b2 += 4; c1 -= 4; c2 += 4; Dst += 4; n -= 4;
} while (n>0);
}
/* FIXED POINT: bwArray = COEF */
static void calc_chirp_factors(sbr_info *sbr, uint8_t ch)
{
uint8_t i;
for (i = 0; i < sbr->N_Q; i++)
{
sbr->bwArray[ch][i] = mapNewBw(sbr->bs_invf_mode[ch][i], sbr->bs_invf_mode_prev[ch][i]);
if (sbr->bwArray[ch][i] < sbr->bwArray_prev[ch][i])
sbr->bwArray[ch][i] = MUL_F(sbr->bwArray[ch][i], FRAC_CONST(0.75)) + MUL_F(sbr->bwArray_prev[ch][i], FRAC_CONST(0.25));
else
sbr->bwArray[ch][i] = MUL_F(sbr->bwArray[ch][i], FRAC_CONST(0.90625)) + MUL_F(sbr->bwArray_prev[ch][i], FRAC_CONST(0.09375));
if (sbr->bwArray[ch][i] < COEF_CONST(0.015625))
sbr->bwArray[ch][i] = COEF_CONST(0.0);
if (sbr->bwArray[ch][i] > COEF_CONST(0.99609375))
sbr->bwArray[ch][i] = COEF_CONST(0.99609375);
sbr->bwArray_prev[ch][i] = sbr->bwArray[ch][i];
sbr->bs_invf_mode_prev[ch][i] = sbr->bs_invf_mode[ch][i];
}
}
static INLINE real_t get_sample(real_t **input, uint8_t channel, uint16_t sample,
uint8_t down_matrix, uint8_t *internal_channel)
{
if (!down_matrix)
return input[internal_channel[channel]][sample];
if (channel == 0)
{
real_t C = MUL_F(input[internal_channel[0]][sample], RSQRT2);
real_t L_S = MUL_F(input[internal_channel[3]][sample], RSQRT2);
real_t cum = input[internal_channel[1]][sample] + C + L_S;
return MUL_F(cum, DM_MUL);
} else {
real_t C = MUL_F(input[internal_channel[0]][sample], RSQRT2);
real_t R_S = MUL_F(input[internal_channel[4]][sample], RSQRT2);
real_t cum = input[internal_channel[2]][sample] + C + R_S;
return MUL_F(cum, DM_MUL);
}
}
static void auto_correlation(sbr_info *sbr, acorr_coef *ac, qmf_t buffer[MAX_NTSRHFG][64],
uint8_t bd, uint8_t len)
{
real_t r01r = 0, r01i = 0, r02r = 0, r02i = 0, r11r = 0;
real_t temp1_r, temp1_i, temp2_r, temp2_i, temp3_r, temp3_i;
real_t temp4_r, temp4_i, temp5_r, temp5_i;
int8_t j;
uint8_t offset = sbr->tHFAdj;
const real_t rel = FRAC_CONST(0.999999); // 1 / (1 + 1e-6f);
temp2_r = ACDET_PRE(QMF_RE(buffer[offset-2][bd]));
temp2_i = ACDET_PRE(QMF_IM(buffer[offset-2][bd]));
temp3_r = ACDET_PRE(QMF_RE(buffer[offset-1][bd]));
temp3_i = ACDET_PRE(QMF_IM(buffer[offset-1][bd]));
// Save these because they are needed after loop
temp4_r = temp2_r;
temp4_i = temp2_i;
temp5_r = temp3_r;
temp5_i = temp3_i;
for (j = offset; j < len + offset; j++)
{
temp1_r = temp2_r;
temp1_i = temp2_i;
temp2_r = temp3_r;
temp2_i = temp3_i;
temp3_r = ACDET_PRE(QMF_RE(buffer[j][bd]));
temp3_i = ACDET_PRE(QMF_IM(buffer[j][bd]));
r01r += MUL_F(temp3_r, temp2_r) + MUL_F(temp3_i, temp2_i);
r01i += MUL_F(temp3_i, temp2_r) - MUL_F(temp3_r, temp2_i);
r02r += MUL_F(temp3_r, temp1_r) + MUL_F(temp3_i, temp1_i);
r02i += MUL_F(temp3_i, temp1_r) - MUL_F(temp3_r, temp1_i);
r11r += MUL_F(temp2_r, temp2_r) + MUL_F(temp2_i, temp2_i);
}
RE(ac->r12) = r01r - (MUL_F(temp3_r, temp2_r) + MUL_F(temp3_i, temp2_i)) +
(MUL_F(temp5_r, temp4_r) + MUL_F(temp5_i, temp4_i));
IM(ac->r12) = r01i - (MUL_F(temp3_i, temp2_r) - MUL_F(temp3_r, temp2_i)) +
(MUL_F(temp5_i, temp4_r) - MUL_F(temp5_r, temp4_i));
RE(ac->r22) = r11r - (MUL_F(temp2_r, temp2_r) + MUL_F(temp2_i, temp2_i)) +
(MUL_F(temp4_r, temp4_r) + MUL_F(temp4_i, temp4_i));
RE(ac->r01) = r01r;
IM(ac->r01) = r01i;
RE(ac->r02) = r02r;
IM(ac->r02) = r02i;
RE(ac->r11) = r11r;
ac->det = MUL_F(RE(ac->r11), RE(ac->r22)) - MUL_F((MUL_F(RE(ac->r12), RE(ac->r12)) + MUL_F(IM(ac->r12), IM(ac->r12))), rel);
ac->det = ACDET_POST(ac->det);
}
void reorder_ch_pcm(int16_t *pcmBuf, int nSamps, int nChans)
{
int i, ch, chanMap[6];
int16_t tmpBuf[6], *outbuf;
switch (nChans) {
case 3:
chanMap[0] = 1; /* L */
chanMap[1] = 2; /* R */
chanMap[2] = 0; /* C */
break;
case 4:
chanMap[0] = 1; /* L */
chanMap[1] = 2; /* R */
chanMap[2] = 0; /* C */
chanMap[3] = 3; /* S */
break;
case 5:
chanMap[0] = 1; /* L */
chanMap[1] = 2; /* R */
chanMap[2] = 0; /* C */
chanMap[3] = 3; /* LS */
chanMap[4] = 4; /* RS */
break;
case 6:
chanMap[0] = 1; /* L */
chanMap[1] = 2; /* R */
chanMap[2] = 0; /* C */
chanMap[3] = 5; /* LFE */
chanMap[4] = 3; /* LS */
chanMap[5] = 4; /* RS */
break;
default:
return;
}
if(audio_output_channels < nChans){
#if 0
int32_t C, L_S, cum;
outbuf = pcmBuf;
for (i = 0; i < nSamps; i += nChans) {
C = MUL_F(pcmBuf[0], RSQRT2);
L_S = MUL_F(pcmBuf[3], RSQRT2);
cum = pcmBuf[1] + C + L_S;
outbuf[0] = MUL_F(cum, DM_MUL);
C = MUL_F(pcmBuf[0], RSQRT2);
L_S = MUL_F(pcmBuf[4], RSQRT2);
cum = pcmBuf[2] + C + L_S;
outbuf[1] = MUL_F(cum, DM_MUL);
outbuf += 2;
pcmBuf += nChans;
}
#else
outbuf = pcmBuf;
for (i = 0; i < nSamps; i += nChans) {
outbuf[0] = pcmBuf[0]+pcmBuf[3]+pcmBuf[1];
outbuf[1] = pcmBuf[0]+pcmBuf[4]+pcmBuf[2];
outbuf += 2;
pcmBuf += nChans;
}
#endif
}else {
for (i = 0; i < nSamps; i += nChans) {
for (ch = 0; ch < nChans; ch++)
tmpBuf[ch] = pcmBuf[chanMap[ch]];
for (ch = 0; ch < nChans; ch++)
pcmBuf[ch] = tmpBuf[ch];
pcmBuf += nChans;
}
}
}
/* only works for LTP -> no overlapping, no short blocks */
void filter_bank_ltp(fb_info *fb, uint8_t window_sequence, uint8_t window_shape,
uint8_t window_shape_prev, real_t *in_data, real_t *out_mdct,
uint8_t object_type, uint16_t frame_len)
{
int16_t i;
ALIGN real_t windowed_buf[2*1024] = {0};
const real_t *window_long = NULL;
const real_t *window_long_prev = NULL;
const real_t *window_short = NULL;
const real_t *window_short_prev = NULL;
uint16_t nlong = frame_len;
uint16_t nshort = frame_len/8;
uint16_t nflat_ls = (nlong-nshort)/2;
assert(window_sequence != EIGHT_SHORT_SEQUENCE);
#ifdef LD_DEC
if (object_type == LD)
{
window_long = fb->ld_window[window_shape];
window_long_prev = fb->ld_window[window_shape_prev];
} else {
#endif
window_long = fb->long_window[window_shape];
window_long_prev = fb->long_window[window_shape_prev];
window_short = fb->short_window[window_shape];
window_short_prev = fb->short_window[window_shape_prev];
#ifdef LD_DEC
}
#endif
switch(window_sequence)
{
case ONLY_LONG_SEQUENCE:
for (i = nlong-1; i >= 0; i--)
{
windowed_buf[i] = MUL_F(in_data[i], window_long_prev[i]);
windowed_buf[i+nlong] = MUL_F(in_data[i+nlong], window_long[nlong-1-i]);
}
mdct(fb, windowed_buf, out_mdct, 2*nlong);
break;
case LONG_START_SEQUENCE:
for (i = 0; i < nlong; i++)
windowed_buf[i] = MUL_F(in_data[i], window_long_prev[i]);
for (i = 0; i < nflat_ls; i++)
windowed_buf[i+nlong] = in_data[i+nlong];
for (i = 0; i < nshort; i++)
windowed_buf[i+nlong+nflat_ls] = MUL_F(in_data[i+nlong+nflat_ls], window_short[nshort-1-i]);
for (i = 0; i < nflat_ls; i++)
windowed_buf[i+nlong+nflat_ls+nshort] = 0;
mdct(fb, windowed_buf, out_mdct, 2*nlong);
break;
case LONG_STOP_SEQUENCE:
for (i = 0; i < nflat_ls; i++)
windowed_buf[i] = 0;
for (i = 0; i < nshort; i++)
windowed_buf[i+nflat_ls] = MUL_F(in_data[i+nflat_ls], window_short_prev[i]);
for (i = 0; i < nflat_ls; i++)
windowed_buf[i+nflat_ls+nshort] = in_data[i+nflat_ls+nshort];
for (i = 0; i < nlong; i++)
windowed_buf[i+nlong] = MUL_F(in_data[i+nlong], window_long[nlong-1-i]);
mdct(fb, windowed_buf, out_mdct, 2*nlong);
break;
}
}
void ifilter_bank(fb_info *fb, uint8_t window_sequence, uint8_t window_shape,
uint8_t window_shape_prev, real_t *freq_in,
real_t *time_out, real_t *overlap,
uint8_t object_type, uint16_t frame_len)
{
int16_t i;
ALIGN real_t transf_buf[2*1024] = {0};
const real_t *window_long = NULL;
const real_t *window_long_prev = NULL;
const real_t *window_short = NULL;
const real_t *window_short_prev = NULL;
uint16_t nlong = frame_len;
uint16_t nshort = frame_len/8;
uint16_t trans = nshort/2;
uint16_t nflat_ls = (nlong-nshort)/2;
#ifdef PROFILE
int64_t count = faad_get_ts();
#endif
#ifdef LD_DEC
if (object_type == LD)
{
window_long = fb->ld_window[window_shape];
window_long_prev = fb->ld_window[window_shape_prev];
} else {
#endif
window_long = fb->long_window[window_shape];
window_long_prev = fb->long_window[window_shape_prev];
window_short = fb->short_window[window_shape];
window_short_prev = fb->short_window[window_shape_prev];
#ifdef LD_DEC
}
#endif
switch (window_sequence)
{
case ONLY_LONG_SEQUENCE:
imdct_long(fb, freq_in, transf_buf, 2*nlong);
for (i = 0; i < nlong; i+=4)
{
time_out[i] = overlap[i] + MUL_F(transf_buf[i],window_long_prev[i]);
time_out[i+1] = overlap[i+1] + MUL_F(transf_buf[i+1],window_long_prev[i+1]);
time_out[i+2] = overlap[i+2] + MUL_F(transf_buf[i+2],window_long_prev[i+2]);
time_out[i+3] = overlap[i+3] + MUL_F(transf_buf[i+3],window_long_prev[i+3]);
}
for (i = 0; i < nlong; i+=4)
{
overlap[i] = MUL_F(transf_buf[nlong+i],window_long[nlong-1-i]);
overlap[i+1] = MUL_F(transf_buf[nlong+i+1],window_long[nlong-2-i]);
overlap[i+2] = MUL_F(transf_buf[nlong+i+2],window_long[nlong-3-i]);
overlap[i+3] = MUL_F(transf_buf[nlong+i+3],window_long[nlong-4-i]);
}
break;
case LONG_START_SEQUENCE:
imdct_long(fb, freq_in, transf_buf, 2*nlong);
for (i = 0; i < nlong; i+=4)
{
time_out[i] = overlap[i] + MUL_F(transf_buf[i],window_long_prev[i]);
time_out[i+1] = overlap[i+1] + MUL_F(transf_buf[i+1],window_long_prev[i+1]);
time_out[i+2] = overlap[i+2] + MUL_F(transf_buf[i+2],window_long_prev[i+2]);
time_out[i+3] = overlap[i+3] + MUL_F(transf_buf[i+3],window_long_prev[i+3]);
}
for (i = 0; i < nflat_ls; i++)
overlap[i] = transf_buf[nlong+i];
for (i = 0; i < nshort; i++)
overlap[nflat_ls+i] = MUL_F(transf_buf[nlong+nflat_ls+i],window_short[nshort-i-1]);
for (i = 0; i < nflat_ls; i++)
overlap[nflat_ls+nshort+i] = 0;
break;
case EIGHT_SHORT_SEQUENCE:
faad_imdct(fb->mdct256, freq_in+0*nshort, transf_buf+2*nshort*0);
faad_imdct(fb->mdct256, freq_in+1*nshort, transf_buf+2*nshort*1);
faad_imdct(fb->mdct256, freq_in+2*nshort, transf_buf+2*nshort*2);
faad_imdct(fb->mdct256, freq_in+3*nshort, transf_buf+2*nshort*3);
faad_imdct(fb->mdct256, freq_in+4*nshort, transf_buf+2*nshort*4);
faad_imdct(fb->mdct256, freq_in+5*nshort, transf_buf+2*nshort*5);
faad_imdct(fb->mdct256, freq_in+6*nshort, transf_buf+2*nshort*6);
faad_imdct(fb->mdct256, freq_in+7*nshort, transf_buf+2*nshort*7);
for (i = 0; i < nflat_ls; i++)
time_out[i] = overlap[i];
for(i = 0; i < nshort; i++)
{
time_out[nflat_ls+ i] = overlap[nflat_ls+ i] + MUL_F(transf_buf[nshort*0+i],window_short_prev[i]);
time_out[nflat_ls+1*nshort+i] = overlap[nflat_ls+nshort*1+i] + MUL_F(transf_buf[nshort*1+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*2+i],window_short[i]);
time_out[nflat_ls+2*nshort+i] = overlap[nflat_ls+nshort*2+i] + MUL_F(transf_buf[nshort*3+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*4+i],window_short[i]);
time_out[nflat_ls+3*nshort+i] = overlap[nflat_ls+nshort*3+i] + MUL_F(transf_buf[nshort*5+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*6+i],window_short[i]);
if (i < trans)
time_out[nflat_ls+4*nshort+i] = overlap[nflat_ls+nshort*4+i] + MUL_F(transf_buf[nshort*7+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*8+i],window_short[i]);
}
for(i = 0; i < nshort; i++)
{
if (i >= trans)
overlap[nflat_ls+4*nshort+i-nlong] = MUL_F(transf_buf[nshort*7+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*8+i],window_short[i]);
overlap[nflat_ls+5*nshort+i-nlong] = MUL_F(transf_buf[nshort*9+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*10+i],window_short[i]);
overlap[nflat_ls+6*nshort+i-nlong] = MUL_F(transf_buf[nshort*11+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*12+i],window_short[i]);
overlap[nflat_ls+7*nshort+i-nlong] = MUL_F(transf_buf[nshort*13+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*14+i],window_short[i]);
overlap[nflat_ls+8*nshort+i-nlong] = MUL_F(transf_buf[nshort*15+i],window_short[nshort-1-i]);
}
for (i = 0; i < nflat_ls; i++)
overlap[nflat_ls+nshort+i] = 0;
break;
case LONG_STOP_SEQUENCE:
imdct_long(fb, freq_in, transf_buf, 2*nlong);
for (i = 0; i < nflat_ls; i++)
time_out[i] = overlap[i];
for (i = 0; i < nshort; i++)
time_out[nflat_ls+i] = overlap[nflat_ls+i] + MUL_F(transf_buf[nflat_ls+i],window_short_prev[i]);
for (i = 0; i < nflat_ls; i++)
time_out[nflat_ls+nshort+i] = overlap[nflat_ls+nshort+i] + transf_buf[nflat_ls+nshort+i];
for (i = 0; i < nlong; i++)
overlap[i] = MUL_F(transf_buf[nlong+i],window_long[nlong-1-i]);
break;
}
#ifdef PROFILE
count = faad_get_ts() - count;
fb->cycles += count;
#endif
}
void ifilter_bank(fb_info *fb, uint8_t window_sequence, uint8_t window_shape,
uint8_t window_shape_prev, real_t *freq_in,
real_t *time_out, real_t *overlap,
uint8_t object_type, uint16_t frame_len)
{
int16_t i;
ALIGN real_t transf_buf[2*1024] = {0};
const real_t *window_long = NULL;
const real_t *window_long_prev = NULL;
const real_t *window_short = NULL;
const real_t *window_short_prev = NULL;
uint16_t nlong = frame_len;
uint16_t nshort = frame_len/8;
uint16_t trans = nshort/2;
uint16_t nflat_ls = (nlong-nshort)/2;
#ifdef PROFILE
int64_t count = faad_get_ts();
#endif
/* select windows of current frame and previous frame (Sine or KBD) */
#ifdef LD_DEC
if (object_type == LD)
{
window_long = fb->ld_window[window_shape];
window_long_prev = fb->ld_window[window_shape_prev];
} else {
#endif
window_long = fb->long_window[window_shape];
window_long_prev = fb->long_window[window_shape_prev];
window_short = fb->short_window[window_shape];
window_short_prev = fb->short_window[window_shape_prev];
#ifdef LD_DEC
}
#endif
#if 0
for (i = 0; i < 1024; i++)
{
printf("%d\n", freq_in[i]);
}
#endif
#if 0
printf("%d %d\n", window_sequence, window_shape);
#endif
switch (window_sequence)
{
case ONLY_LONG_SEQUENCE:
/* perform iMDCT */
imdct_long(fb, freq_in, transf_buf, 2*nlong);
/* add second half output of previous frame to windowed output of current frame */
for (i = 0; i < nlong; i+=4)
{
time_out[i] = overlap[i] + MUL_F(transf_buf[i],window_long_prev[i]);
time_out[i+1] = overlap[i+1] + MUL_F(transf_buf[i+1],window_long_prev[i+1]);
time_out[i+2] = overlap[i+2] + MUL_F(transf_buf[i+2],window_long_prev[i+2]);
time_out[i+3] = overlap[i+3] + MUL_F(transf_buf[i+3],window_long_prev[i+3]);
}
/* window the second half and save as overlap for next frame */
for (i = 0; i < nlong; i+=4)
{
overlap[i] = MUL_F(transf_buf[nlong+i],window_long[nlong-1-i]);
overlap[i+1] = MUL_F(transf_buf[nlong+i+1],window_long[nlong-2-i]);
overlap[i+2] = MUL_F(transf_buf[nlong+i+2],window_long[nlong-3-i]);
overlap[i+3] = MUL_F(transf_buf[nlong+i+3],window_long[nlong-4-i]);
}
break;
case LONG_START_SEQUENCE:
/* perform iMDCT */
imdct_long(fb, freq_in, transf_buf, 2*nlong);
/* add second half output of previous frame to windowed output of current frame */
for (i = 0; i < nlong; i+=4)
{
time_out[i] = overlap[i] + MUL_F(transf_buf[i],window_long_prev[i]);
time_out[i+1] = overlap[i+1] + MUL_F(transf_buf[i+1],window_long_prev[i+1]);
time_out[i+2] = overlap[i+2] + MUL_F(transf_buf[i+2],window_long_prev[i+2]);
time_out[i+3] = overlap[i+3] + MUL_F(transf_buf[i+3],window_long_prev[i+3]);
}
/* window the second half and save as overlap for next frame */
/* construct second half window using padding with 1's and 0's */
for (i = 0; i < nflat_ls; i++)
overlap[i] = transf_buf[nlong+i];
for (i = 0; i < nshort; i++)
overlap[nflat_ls+i] = MUL_F(transf_buf[nlong+nflat_ls+i],window_short[nshort-i-1]);
for (i = 0; i < nflat_ls; i++)
overlap[nflat_ls+nshort+i] = 0;
break;
case EIGHT_SHORT_SEQUENCE:
/* perform iMDCT for each short block */
faad_imdct(fb->mdct256, freq_in+0*nshort, transf_buf+2*nshort*0);
faad_imdct(fb->mdct256, freq_in+1*nshort, transf_buf+2*nshort*1);
faad_imdct(fb->mdct256, freq_in+2*nshort, transf_buf+2*nshort*2);
faad_imdct(fb->mdct256, freq_in+3*nshort, transf_buf+2*nshort*3);
faad_imdct(fb->mdct256, freq_in+4*nshort, transf_buf+2*nshort*4);
faad_imdct(fb->mdct256, freq_in+5*nshort, transf_buf+2*nshort*5);
faad_imdct(fb->mdct256, freq_in+6*nshort, transf_buf+2*nshort*6);
faad_imdct(fb->mdct256, freq_in+7*nshort, transf_buf+2*nshort*7);
/* add second half output of previous frame to windowed output of current frame */
for (i = 0; i < nflat_ls; i++)
time_out[i] = overlap[i];
for(i = 0; i < nshort; i++)
{
time_out[nflat_ls+ i] = overlap[nflat_ls+ i] + MUL_F(transf_buf[nshort*0+i],window_short_prev[i]);
time_out[nflat_ls+1*nshort+i] = overlap[nflat_ls+nshort*1+i] + MUL_F(transf_buf[nshort*1+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*2+i],window_short[i]);
time_out[nflat_ls+2*nshort+i] = overlap[nflat_ls+nshort*2+i] + MUL_F(transf_buf[nshort*3+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*4+i],window_short[i]);
time_out[nflat_ls+3*nshort+i] = overlap[nflat_ls+nshort*3+i] + MUL_F(transf_buf[nshort*5+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*6+i],window_short[i]);
if (i < trans)
time_out[nflat_ls+4*nshort+i] = overlap[nflat_ls+nshort*4+i] + MUL_F(transf_buf[nshort*7+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*8+i],window_short[i]);
}
/* window the second half and save as overlap for next frame */
for(i = 0; i < nshort; i++)
{
if (i >= trans)
overlap[nflat_ls+4*nshort+i-nlong] = MUL_F(transf_buf[nshort*7+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*8+i],window_short[i]);
overlap[nflat_ls+5*nshort+i-nlong] = MUL_F(transf_buf[nshort*9+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*10+i],window_short[i]);
overlap[nflat_ls+6*nshort+i-nlong] = MUL_F(transf_buf[nshort*11+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*12+i],window_short[i]);
overlap[nflat_ls+7*nshort+i-nlong] = MUL_F(transf_buf[nshort*13+i],window_short[nshort-1-i]) + MUL_F(transf_buf[nshort*14+i],window_short[i]);
overlap[nflat_ls+8*nshort+i-nlong] = MUL_F(transf_buf[nshort*15+i],window_short[nshort-1-i]);
}
for (i = 0; i < nflat_ls; i++)
overlap[nflat_ls+nshort+i] = 0;
break;
case LONG_STOP_SEQUENCE:
/* perform iMDCT */
imdct_long(fb, freq_in, transf_buf, 2*nlong);
/* add second half output of previous frame to windowed output of current frame */
/* construct first half window using padding with 1's and 0's */
for (i = 0; i < nflat_ls; i++)
time_out[i] = overlap[i];
for (i = 0; i < nshort; i++)
time_out[nflat_ls+i] = overlap[nflat_ls+i] + MUL_F(transf_buf[nflat_ls+i],window_short_prev[i]);
for (i = 0; i < nflat_ls; i++)
time_out[nflat_ls+nshort+i] = overlap[nflat_ls+nshort+i] + transf_buf[nflat_ls+nshort+i];
/* window the second half and save as overlap for next frame */
for (i = 0; i < nlong; i++)
overlap[i] = MUL_F(transf_buf[nlong+i],window_long[nlong-1-i]);
break;
}
#if 0
for (i = 0; i < 1024; i++)
{
printf("%d\n", time_out[i]);
//printf("0x%.8X\n", time_out[i]);
}
#endif
#ifdef PROFILE
count = faad_get_ts() - count;
fb->cycles += count;
#endif
}
static void passf3(const uint16_t ido, const uint16_t l1, const complex_t *cc,
complex_t *ch, const complex_t *wa1, const complex_t *wa2,
const int8_t isign)
{
static real_t taur = FRAC_CONST(-0.5);
static real_t taui = FRAC_CONST(0.866025403784439);
uint16_t i, k, ac, ah;
complex_t c2, c3, d2, d3, t2;
if (ido == 1)
{
if (isign == 1)
{
for (k = 0; k < l1; k++)
{
ac = 3*k+1;
ah = k;
RE(t2) = RE(cc[ac]) + RE(cc[ac+1]);
IM(t2) = IM(cc[ac]) + IM(cc[ac+1]);
RE(c2) = RE(cc[ac-1]) + MUL_F(RE(t2),taur);
IM(c2) = IM(cc[ac-1]) + MUL_F(IM(t2),taur);
RE(ch[ah]) = RE(cc[ac-1]) + RE(t2);
IM(ch[ah]) = IM(cc[ac-1]) + IM(t2);
RE(c3) = MUL_F((RE(cc[ac]) - RE(cc[ac+1])), taui);
IM(c3) = MUL_F((IM(cc[ac]) - IM(cc[ac+1])), taui);
RE(ch[ah+l1]) = RE(c2) - IM(c3);
IM(ch[ah+l1]) = IM(c2) + RE(c3);
RE(ch[ah+2*l1]) = RE(c2) + IM(c3);
IM(ch[ah+2*l1]) = IM(c2) - RE(c3);
}
} else {
for (k = 0; k < l1; k++)
{
ac = 3*k+1;
ah = k;
RE(t2) = RE(cc[ac]) + RE(cc[ac+1]);
IM(t2) = IM(cc[ac]) + IM(cc[ac+1]);
RE(c2) = RE(cc[ac-1]) + MUL_F(RE(t2),taur);
IM(c2) = IM(cc[ac-1]) + MUL_F(IM(t2),taur);
RE(ch[ah]) = RE(cc[ac-1]) + RE(t2);
IM(ch[ah]) = IM(cc[ac-1]) + IM(t2);
RE(c3) = MUL_F((RE(cc[ac]) - RE(cc[ac+1])), taui);
IM(c3) = MUL_F((IM(cc[ac]) - IM(cc[ac+1])), taui);
RE(ch[ah+l1]) = RE(c2) + IM(c3);
IM(ch[ah+l1]) = IM(c2) - RE(c3);
RE(ch[ah+2*l1]) = RE(c2) - IM(c3);
IM(ch[ah+2*l1]) = IM(c2) + RE(c3);
}
}
} else {
if (isign == 1)
{
for (k = 0; k < l1; k++)
{
for (i = 0; i < ido; i++)
{
ac = i + (3*k+1)*ido;
ah = i + k * ido;
RE(t2) = RE(cc[ac]) + RE(cc[ac+ido]);
RE(c2) = RE(cc[ac-ido]) + MUL_F(RE(t2),taur);
IM(t2) = IM(cc[ac]) + IM(cc[ac+ido]);
IM(c2) = IM(cc[ac-ido]) + MUL_F(IM(t2),taur);
RE(ch[ah]) = RE(cc[ac-ido]) + RE(t2);
IM(ch[ah]) = IM(cc[ac-ido]) + IM(t2);
RE(c3) = MUL_F((RE(cc[ac]) - RE(cc[ac+ido])), taui);
IM(c3) = MUL_F((IM(cc[ac]) - IM(cc[ac+ido])), taui);
RE(d2) = RE(c2) - IM(c3);
IM(d3) = IM(c2) - RE(c3);
RE(d3) = RE(c2) + IM(c3);
IM(d2) = IM(c2) + RE(c3);
#if 1
ComplexMult(&IM(ch[ah+l1*ido]), &RE(ch[ah+l1*ido]),
IM(d2), RE(d2), RE(wa1[i]), IM(wa1[i]));
ComplexMult(&IM(ch[ah+2*l1*ido]), &RE(ch[ah+2*l1*ido]),
IM(d3), RE(d3), RE(wa2[i]), IM(wa2[i]));
#else
ComplexMult(&RE(ch[ah+l1*ido]), &IM(ch[ah+l1*ido]),
RE(d2), IM(d2), RE(wa1[i]), IM(wa1[i]));
ComplexMult(&RE(ch[ah+2*l1*ido]), &IM(ch[ah+2*l1*ido]),
RE(d3), IM(d3), RE(wa2[i]), IM(wa2[i]));
#endif
}
}
} else {
for (k = 0; k < l1; k++)
{
for (i = 0; i < ido; i++)
{
ac = i + (3*k+1)*ido;
ah = i + k * ido;
RE(t2) = RE(cc[ac]) + RE(cc[ac+ido]);
RE(c2) = RE(cc[ac-ido]) + MUL_F(RE(t2),taur);
IM(t2) = IM(cc[ac]) + IM(cc[ac+ido]);
IM(c2) = IM(cc[ac-ido]) + MUL_F(IM(t2),taur);
RE(ch[ah]) = RE(cc[ac-ido]) + RE(t2);
IM(ch[ah]) = IM(cc[ac-ido]) + IM(t2);
RE(c3) = MUL_F((RE(cc[ac]) - RE(cc[ac+ido])), taui);
IM(c3) = MUL_F((IM(cc[ac]) - IM(cc[ac+ido])), taui);
RE(d2) = RE(c2) + IM(c3);
IM(d3) = IM(c2) + RE(c3);
RE(d3) = RE(c2) - IM(c3);
IM(d2) = IM(c2) - RE(c3);
#if 1
ComplexMult(&RE(ch[ah+l1*ido]), &IM(ch[ah+l1*ido]),
RE(d2), IM(d2), RE(wa1[i]), IM(wa1[i]));
ComplexMult(&RE(ch[ah+2*l1*ido]), &IM(ch[ah+2*l1*ido]),
RE(d3), IM(d3), RE(wa2[i]), IM(wa2[i]));
#else
ComplexMult(&IM(ch[ah+l1*ido]), &RE(ch[ah+l1*ido]),
IM(d2), RE(d2), RE(wa1[i]), IM(wa1[i]));
ComplexMult(&IM(ch[ah+2*l1*ido]), &RE(ch[ah+2*l1*ido]),
IM(d3), RE(d3), RE(wa2[i]), IM(wa2[i]));
#endif
}
}
}
}
}
static void hf_assembly(sbr_info *sbr, sbr_hfadj_info *adj,
qmf_t Xsbr[MAX_NTSRHFG][64], uint8_t ch)
{
static real_t h_smooth[] = {
COEF_CONST(0.03183050093751), COEF_CONST(0.11516383427084),
COEF_CONST(0.21816949906249), COEF_CONST(0.30150283239582),
COEF_CONST(0.33333333333333)
};
static int8_t phi_re[] = { 1, 0, -1, 0 };
static int8_t phi_im[] = { 0, 1, 0, -1 };
uint8_t m, l, i, n;
uint16_t fIndexNoise = 0;
uint8_t fIndexSine = 0;
uint8_t assembly_reset = 0;
real_t *temp;
real_t G_filt, Q_filt;
uint8_t h_SL;
if (sbr->Reset == 1)
{
assembly_reset = 1;
fIndexNoise = 0;
} else {
fIndexNoise = sbr->index_noise_prev[ch];
}
fIndexSine = sbr->psi_is_prev[ch];
for (l = 0; l < sbr->L_E[ch]; l++)
{
uint8_t no_noise = (l == sbr->l_A[ch] || l == sbr->prevEnvIsShort[ch]) ? 1 : 0;
#ifdef SBR_LOW_POWER
h_SL = 0;
#else
h_SL = (sbr->bs_smoothing_mode == 1) ? 0 : 4;
h_SL = (no_noise ? 0 : h_SL);
#endif
if (assembly_reset)
{
for (n = 0; n < 4; n++)
{
memcpy(sbr->G_temp_prev[ch][n], adj->G_lim_boost[l], sbr->M*sizeof(real_t));
memcpy(sbr->Q_temp_prev[ch][n], adj->Q_M_lim_boost[l], sbr->M*sizeof(real_t));
}
assembly_reset = 0;
}
for (i = sbr->t_E[ch][l]; i < sbr->t_E[ch][l+1]; i++)
{
#ifdef SBR_LOW_POWER
uint8_t i_min1, i_plus1;
uint8_t sinusoids = 0;
#endif
memcpy(sbr->G_temp_prev[ch][4], adj->G_lim_boost[l], sbr->M*sizeof(real_t));
memcpy(sbr->Q_temp_prev[ch][4], adj->Q_M_lim_boost[l], sbr->M*sizeof(real_t));
for (m = 0; m < sbr->M; m++)
{
uint8_t j;
qmf_t psi;
G_filt = 0;
Q_filt = 0;
j = 0;
if (h_SL != 0)
{
for (n = 0; n <= 4; n++)
{
G_filt += MUL_C(sbr->G_temp_prev[ch][n][m], h_smooth[j]);
Q_filt += MUL_C(sbr->Q_temp_prev[ch][n][m], h_smooth[j]);
j++;
}
} else {
G_filt = sbr->G_temp_prev[ch][4][m];
Q_filt = sbr->Q_temp_prev[ch][4][m];
}
Q_filt = (adj->S_M_boost[l][m] != 0 || no_noise) ? 0 : Q_filt;
/* add noise to the output */
fIndexNoise = (fIndexNoise + 1) & 511;
/* the smoothed gain values are applied to Xsbr */
/* V is defined, not calculated */
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) = MUL_R(G_filt, QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]))
+ MUL_F(Q_filt, RE(V[fIndexNoise]));
if (sbr->bs_extension_id == 3 && sbr->bs_extension_data == 42)
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) = 16428320;
#ifndef SBR_LOW_POWER
QMF_IM(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) = MUL_R(G_filt, QMF_IM(Xsbr[i + sbr->tHFAdj][m+sbr->kx]))
+ MUL_F(Q_filt, IM(V[fIndexNoise]));
#endif
//if (adj->S_index_mapped[m][l])
{
int8_t rev = (((m + sbr->kx) & 1) ? -1 : 1);
QMF_RE(psi) = MUL_R(adj->S_M_boost[l][m], phi_re[fIndexSine]);
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) += QMF_RE(psi);
#ifndef SBR_LOW_POWER
QMF_IM(psi) = rev * MUL_R(adj->S_M_boost[l][m], phi_im[fIndexSine]);
QMF_IM(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) += QMF_IM(psi);
#else
i_min1 = (fIndexSine - 1) & 3;
i_plus1 = (fIndexSine + 1) & 3;
if (m == 0)
{
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx - 1]) -=
(-1*rev * MUL_C(MUL_R(adj->S_M_boost[l][0], phi_re[i_plus1]), COEF_CONST(0.00815)));
if(m < sbr->M - 1)
{
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) -=
(rev * MUL_C(MUL_R(adj->S_M_boost[l][1], phi_re[i_plus1]), COEF_CONST(0.00815)));
}
}
if ((m > 0) && (m < sbr->M - 1) && (sinusoids < 16))
{
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) -=
(rev * MUL_C(MUL_R(adj->S_M_boost[l][m - 1], phi_re[i_min1]), COEF_CONST(0.00815)));
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) -=
(rev * MUL_C(MUL_R(adj->S_M_boost[l][m + 1], phi_re[i_plus1]), COEF_CONST(0.00815)));
}
if ((m == sbr->M - 1) && (sinusoids < 16))
{
if (m > 0)
{
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx]) -=
(rev * MUL_C(MUL_R(adj->S_M_boost[l][m - 1], phi_re[i_min1]), COEF_CONST(0.00815)));
}
if (m + sbr->kx < 64)
{
QMF_RE(Xsbr[i + sbr->tHFAdj][m+sbr->kx + 1]) -=
(-1*rev * MUL_C(MUL_R(adj->S_M_boost[l][m], phi_re[i_min1]), COEF_CONST(0.00815)));
}
}
if (adj->S_M_boost[l][m] != 0)
sinusoids++;
#endif
}
}
fIndexSine = (fIndexSine + 1) & 3;
temp = sbr->G_temp_prev[ch][0];
for (n = 0; n < 4; n++)
sbr->G_temp_prev[ch][n] = sbr->G_temp_prev[ch][n+1];
sbr->G_temp_prev[ch][4] = temp;
temp = sbr->Q_temp_prev[ch][0];
for (n = 0; n < 4; n++)
sbr->Q_temp_prev[ch][n] = sbr->Q_temp_prev[ch][n+1];
sbr->Q_temp_prev[ch][4] = temp;
}
}
sbr->index_noise_prev[ch] = fIndexNoise;
sbr->psi_is_prev[ch] = fIndexSine;
}
