|
Packit |
67cb25 |
/* fft/hc_pass_4.c
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* Copyright (C) 1996, 1997, 1998, 1999, 2000, 2007 Brian Gough
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* This program is free software; you can redistribute it and/or modify
|
|
Packit |
67cb25 |
* it under the terms of the GNU General Public License as published by
|
|
Packit |
67cb25 |
* the Free Software Foundation; either version 3 of the License, or (at
|
|
Packit |
67cb25 |
* your option) any later version.
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* This program is distributed in the hope that it will be useful, but
|
|
Packit |
67cb25 |
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Packit |
67cb25 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Packit |
67cb25 |
* General Public License for more details.
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* You should have received a copy of the GNU General Public License
|
|
Packit |
67cb25 |
* along with this program; if not, write to the Free Software
|
|
Packit |
67cb25 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
Packit |
67cb25 |
*/
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
static void
|
|
Packit |
67cb25 |
FUNCTION(fft_halfcomplex,pass_4) (const BASE in[],
|
|
Packit |
67cb25 |
const size_t istride,
|
|
Packit |
67cb25 |
BASE out[],
|
|
Packit |
67cb25 |
const size_t ostride,
|
|
Packit |
67cb25 |
const size_t product,
|
|
Packit |
67cb25 |
const size_t n,
|
|
Packit |
67cb25 |
const TYPE(gsl_complex) twiddle1[],
|
|
Packit |
67cb25 |
const TYPE(gsl_complex) twiddle2[],
|
|
Packit |
67cb25 |
const TYPE(gsl_complex) twiddle3[])
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
size_t i, j, k, k1, jump;
|
|
Packit |
67cb25 |
size_t factor, q, m, product_1;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
i = 0;
|
|
Packit |
67cb25 |
j = 0;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
factor = 4;
|
|
Packit |
67cb25 |
m = n / factor;
|
|
Packit |
67cb25 |
q = n / product;
|
|
Packit |
67cb25 |
product_1 = product / factor;
|
|
Packit |
67cb25 |
jump = (factor - 1) * q;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (k1 = 0; k1 < product_1; k1++)
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
const size_t from0 = 4 * k1 * q;
|
|
Packit |
67cb25 |
const size_t from1 = from0 + 2 * q - 1;
|
|
Packit |
67cb25 |
const size_t from2 = from1 + 2 * q;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z0_real = VECTOR(in,istride,from0);
|
|
Packit |
67cb25 |
const ATOMIC z1_real = VECTOR(in,istride,from1);
|
|
Packit |
67cb25 |
const ATOMIC z1_imag = VECTOR(in,istride,from1 + 1);
|
|
Packit |
67cb25 |
const ATOMIC z2_real = VECTOR(in,istride,from2);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC t1_real = z0_real + z2_real;
|
|
Packit |
67cb25 |
const ATOMIC t2_real = 2 * z1_real;
|
|
Packit |
67cb25 |
const ATOMIC t3_real = z0_real - z2_real;
|
|
Packit |
67cb25 |
const ATOMIC t4_imag = 2 * z1_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const size_t to0 = q * k1;
|
|
Packit |
67cb25 |
const size_t to1 = to0 + m;
|
|
Packit |
67cb25 |
const size_t to2 = to1 + m;
|
|
Packit |
67cb25 |
const size_t to3 = to2 + m;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to0) = t1_real + t2_real;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to1) = t3_real - t4_imag;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to2) = t1_real - t2_real;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to3) = t3_real + t4_imag;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (q == 1)
|
|
Packit |
67cb25 |
return;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (k = 1; k < (q + 1) / 2; k++)
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
const ATOMIC w1_real = GSL_REAL(twiddle1[k - 1]);
|
|
Packit |
67cb25 |
const ATOMIC w1_imag = GSL_IMAG(twiddle1[k - 1]);
|
|
Packit |
67cb25 |
const ATOMIC w2_real = GSL_REAL(twiddle2[k - 1]);
|
|
Packit |
67cb25 |
const ATOMIC w2_imag = GSL_IMAG(twiddle2[k - 1]);
|
|
Packit |
67cb25 |
const ATOMIC w3_real = GSL_REAL(twiddle3[k - 1]);
|
|
Packit |
67cb25 |
const ATOMIC w3_imag = GSL_IMAG(twiddle3[k - 1]);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (k1 = 0; k1 < product_1; k1++)
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
const size_t from0 = 4 * k1 * q + 2 * k - 1;
|
|
Packit |
67cb25 |
const size_t from1 = from0 + 2 * q;
|
|
Packit |
67cb25 |
const size_t from2 = 4 * k1 * q - 2 * k + 2 * q - 1;
|
|
Packit |
67cb25 |
const size_t from3 = from2 + 2 * q;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z0_real = VECTOR(in,istride,from0);
|
|
Packit |
67cb25 |
const ATOMIC z0_imag = VECTOR(in,istride,from0 + 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z1_real = VECTOR(in,istride,from1);
|
|
Packit |
67cb25 |
const ATOMIC z1_imag = VECTOR(in,istride,from1 + 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z2_real = VECTOR(in,istride,from3);
|
|
Packit |
67cb25 |
const ATOMIC z2_imag = -VECTOR(in,istride,from3 + 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z3_real = VECTOR(in,istride,from2);
|
|
Packit |
67cb25 |
const ATOMIC z3_imag = -VECTOR(in,istride,from2 + 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* compute x = W(4) z */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* t1 = z0 + z2 */
|
|
Packit |
67cb25 |
const ATOMIC t1_real = z0_real + z2_real;
|
|
Packit |
67cb25 |
const ATOMIC t1_imag = z0_imag + z2_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* t2 = z1 + z3 */
|
|
Packit |
67cb25 |
const ATOMIC t2_real = z1_real + z3_real;
|
|
Packit |
67cb25 |
const ATOMIC t2_imag = z1_imag + z3_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* t3 = z0 - z2 */
|
|
Packit |
67cb25 |
const ATOMIC t3_real = z0_real - z2_real;
|
|
Packit |
67cb25 |
const ATOMIC t3_imag = z0_imag - z2_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* t4 = (z1 - z3) */
|
|
Packit |
67cb25 |
const ATOMIC t4_real = (z1_real - z3_real);
|
|
Packit |
67cb25 |
const ATOMIC t4_imag = (z1_imag - z3_imag);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* x0 = t1 + t2 */
|
|
Packit |
67cb25 |
const ATOMIC x0_real = t1_real + t2_real;
|
|
Packit |
67cb25 |
const ATOMIC x0_imag = t1_imag + t2_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* x1 = t3 + i t4 */
|
|
Packit |
67cb25 |
const ATOMIC x1_real = t3_real - t4_imag;
|
|
Packit |
67cb25 |
const ATOMIC x1_imag = t3_imag + t4_real;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* x2 = t1 - t2 */
|
|
Packit |
67cb25 |
const ATOMIC x2_real = t1_real - t2_real;
|
|
Packit |
67cb25 |
const ATOMIC x2_imag = t1_imag - t2_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* x3 = t3 - i t4 */
|
|
Packit |
67cb25 |
const ATOMIC x3_real = t3_real + t4_imag;
|
|
Packit |
67cb25 |
const ATOMIC x3_imag = t3_imag - t4_real;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const size_t to0 = k1 * q + 2 * k - 1;
|
|
Packit |
67cb25 |
const size_t to1 = to0 + m;
|
|
Packit |
67cb25 |
const size_t to2 = to1 + m;
|
|
Packit |
67cb25 |
const size_t to3 = to2 + m;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to0) = x0_real;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to0 + 1) = x0_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to1) = w1_real * x1_real - w1_imag * x1_imag;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to1 + 1) = w1_imag * x1_real + w1_real * x1_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to2) = w2_real * x2_real - w2_imag * x2_imag;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to2 + 1) = w2_imag * x2_real + w2_real * x2_imag;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* to3 = w3 * x3 */
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to3) = w3_real * x3_real - w3_imag * x3_imag;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to3 + 1) = w3_real * x3_imag + w3_imag * x3_real;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (q % 2 == 1)
|
|
Packit |
67cb25 |
return;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (k1 = 0; k1 < product_1; k1++)
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
const size_t from0 = 4 * k1 * q + q - 1;
|
|
Packit |
67cb25 |
const size_t from1 = from0 + 2 * q;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z0_real = VECTOR(in,istride,from0);
|
|
Packit |
67cb25 |
const ATOMIC z0_imag = VECTOR(in,istride,from0 + 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC z1_real = VECTOR(in,istride,from1);
|
|
Packit |
67cb25 |
const ATOMIC z1_imag = VECTOR(in,istride,from1 + 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC t1_real = sqrt (2.0) * (z0_imag + z1_imag);
|
|
Packit |
67cb25 |
const ATOMIC t2_real = sqrt (2.0) * (z0_real - z1_real);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const ATOMIC x0_real = 2 * (z0_real + z1_real);
|
|
Packit |
67cb25 |
const ATOMIC x1_real = t2_real - t1_real;
|
|
Packit |
67cb25 |
const ATOMIC x2_real = 2 * (z1_imag - z0_imag);
|
|
Packit |
67cb25 |
const ATOMIC x3_real = -(t2_real + t1_real);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const size_t to0 = k1 * q + q - 1;
|
|
Packit |
67cb25 |
const size_t to1 = to0 + m;
|
|
Packit |
67cb25 |
const size_t to2 = to1 + m;
|
|
Packit |
67cb25 |
const size_t to3 = to2 + m;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to0) = x0_real;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to1) = x1_real;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to2) = x2_real;
|
|
Packit |
67cb25 |
VECTOR(out,ostride,to3) = x3_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
return;
|
|
Packit |
67cb25 |
}
|