/**************************************************************************************
* Function: PostMultiplyRescale
*
* Description: post-twiddle stage of MDCT, with rescaling for extra guard bits
*
* Inputs: table index (for transform size)
* buffer of nmlt samples
* number of guard bits to remove from output
*
* Outputs: processed samples in same buffer
*
* Return: none
*
* Notes: clips output to [-2^30, 2^30 - 1], guaranteeing at least 1 guard bit
* see notes on PostMultiply(), above
**************************************************************************************/
static void PostMultiplyRescale(int tabidx, int *fft1, int es)
{
int i, nmlt, ar1, ai1, ar2, ai2, skipFactor, z, f0;
int t, cs2, sin2;
int *fft2;
const int *csptr;
nmlt = nmltTab[tabidx];
csptr = cos1sin1tab;
skipFactor = postSkip[tabidx];
fft2 = fft1 + nmlt - 1;
/* load coeffs for first pass
* cps2 = (cos+sin)/2, sin2 = sin/2, cms2 = (cos-sin)/2
*/
cs2 = *csptr++;
sin2 = *csptr;
csptr += skipFactor;
for (i = nmlt >> 2; i != 0; i--) {
ar1 = *(fft1 + 0);
ai1 = *(fft1 + 1);
ai2 = *(fft2 + 0);
/* gain 1 int bit from MULSHIFT32, and one since coeffs are stored as 0.5 * (cos+sin), 0.5*sin */
MULSHIFT32hx(sin2, ar1 + ai1,t);
z = t - MULSHIFT32(cs2, ai1);
CLIP_2N_SHIFT(z, es);
*fft2-- = z;
cs2 -= 2*sin2;
MULSHIFT32hx(cs2, ar1, f0);
z = t + f0;
CLIP_2N_SHIFT(z, es);
*fft1++ = z;
cs2 = *csptr++;
sin2 = *csptr;
csptr += skipFactor;
ar2 = *fft2;
ai2 = -ai2;
MULSHIFT32hx(sin2, ar2 + ai2,t);
z = t - MULSHIFT32(cs2, ai2);
CLIP_2N_SHIFT(z, es);
*fft2-- = z;
cs2 -= 2*sin2;
MULSHIFT32hx(cs2, ar2, f0);
z = t + f0;
CLIP_2N_SHIFT(z, es);
*fft1++ = z;
cs2 += 2*sin2;
}
/* see comments in PostMultiply() for notes on scaling */
return;
}
/**************************************************************************************
* Function: PostMultiplyRescale
*
* Description: post-twiddle stage of DCT4, with rescaling for extra guard bits
*
* Inputs: table index (for transform size)
* buffer of nmdct samples
* number of guard bits to remove from output
*
* Outputs: processed samples in same buffer
*
* Return: none
*
* Notes: clips output to [-2^30, 2^30 - 1], guaranteeing at least 1 guard bit
* see notes on PostMultiply(), above
**************************************************************************************/
static void PostMultiplyRescale(int tabidx, int *fft1, int es)
{
int i, nmdct, ar1, ai1, ar2, ai2, skipFactor, z;
int t, cs2, sin2;
int *fft2;
const int *csptr;
nmdct = nmdctTab[tabidx];
csptr = cos1sin1tab;
skipFactor = postSkip[tabidx];
fft2 = fft1 + nmdct - 1;
/* load coeffs for first pass
* cps2 = (cos+sin), sin2 = sin, cms2 = (cos-sin)
*/
cs2 = *csptr++;
sin2 = *csptr;
csptr += skipFactor;
for (i = nmdct >> 2; i != 0; i--) {
ar1 = *(fft1 + 0);
ai1 = *(fft1 + 1);
ai2 = *(fft2 + 0);
t = MULSHIFT32(sin2, ar1 + ai1);
z = t - MULSHIFT32(cs2, ai1);
CLIP_2N_SHIFT(z, es);
*fft2-- = z;
cs2 -= 2*sin2;
z = t + MULSHIFT32(cs2, ar1);
CLIP_2N_SHIFT(z, es);
*fft1++ = z;
cs2 = *csptr++;
sin2 = *csptr;
csptr += skipFactor;
ar2 = *fft2;
ai2 = -ai2;
t = MULSHIFT32(sin2, ar2 + ai2);
z = t - MULSHIFT32(cs2, ai2);
CLIP_2N_SHIFT(z, es);
*fft2-- = z;
cs2 -= 2*sin2;
z = t + MULSHIFT32(cs2, ar2);
CLIP_2N_SHIFT(z, es);
*fft1++ = z;
cs2 += 2*sin2;
}
}
