--- /dev/null
+/* -*- c++ -*- */
+/*
+ * Copyright 2010 Free Software Foundation, Inc.
+ *
+ * This file is part of GNU Radio
+ *
+ * GNU Radio is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 3, or (at your option)
+ * any later version.
+ *
+ * GNU Radio is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with GNU Radio; see the file COPYING. If not, write to
+ * the Free Software Foundation, Inc., 51 Franklin Street,
+ * Boston, MA 02110-1301, USA.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <gri_fft_filter_fff_sse.h>
+#include <gri_fft.h>
+#include <assert.h>
+#include <stdexcept>
+#include <cstdio>
+#include <xmmintrin.h>
+#include <fftw3.h>
+
+gri_fft_filter_fff_sse::gri_fft_filter_fff_sse (int decimation,
+ const std::vector<float> &taps)
+ : d_fftsize(-1), d_decimation(decimation), d_fwdfft(0), d_invfft(0)
+{
+ d_xformed_taps = (gr_complex*)fftwf_malloc(1*sizeof(gr_complex));
+ set_taps(taps);
+}
+
+gri_fft_filter_fff_sse::~gri_fft_filter_fff_sse ()
+{
+ fftwf_free(d_xformed_taps);
+ delete d_fwdfft;
+ delete d_invfft;
+}
+
+/*
+ * determines d_ntaps, d_nsamples, d_fftsize, d_xformed_taps
+ */
+int
+gri_fft_filter_fff_sse::set_taps (const std::vector<float> &taps)
+{
+ int i = 0;
+ compute_sizes(taps.size());
+
+ d_tail.resize(tailsize());
+ for (i = 0; i < tailsize(); i++)
+ d_tail[i] = 0;
+
+ float *in = d_fwdfft->get_inbuf();
+ gr_complex *out = d_fwdfft->get_outbuf();
+
+ float scale = 1.0 / d_fftsize;
+
+ // Compute forward xform of taps.
+ // Copy taps into first ntaps slots, then pad with zeros
+ for (i = 0; i < d_ntaps; i++)
+ in[i] = taps[i] * scale;
+
+ for (; i < d_fftsize; i++)
+ in[i] = 0;
+
+ d_fwdfft->execute(); // do the xform
+
+ // now copy output to d_xformed_taps
+ for (i = 0; i < d_fftsize/2+1; i++)
+ d_xformed_taps[i] = out[i];
+
+ return d_nsamples;
+}
+
+// determine and set d_ntaps, d_nsamples, d_fftsize
+
+void
+gri_fft_filter_fff_sse::compute_sizes(int ntaps)
+{
+ int old_fftsize = d_fftsize;
+ d_ntaps = ntaps;
+ d_fftsize = (int) (2 * pow(2.0, ceil(log(ntaps) / log(2))));
+ d_nsamples = d_fftsize - d_ntaps + 1;
+
+ if (0)
+ fprintf(stderr, "gri_fft_filter_fff_sse: ntaps = %d, fftsize = %d, nsamples = %d\n",
+ d_ntaps, d_fftsize, d_nsamples);
+
+ assert(d_fftsize == d_ntaps + d_nsamples -1 );
+
+ if (d_fftsize != old_fftsize){ // compute new plans
+ delete d_fwdfft;
+ delete d_invfft;
+ d_fwdfft = new gri_fft_real_fwd(d_fftsize);
+ d_invfft = new gri_fft_real_rev(d_fftsize);
+ //d_xformed_taps.resize(d_fftsize/2+1);
+
+ fftwf_free(d_xformed_taps);
+ d_xformed_taps = (gr_complex*)fftwf_malloc((d_fftsize/2+1)*sizeof(gr_complex));
+ }
+}
+
+int
+gri_fft_filter_fff_sse::filter (int nitems, const float *input, float *output)
+{
+ int dec_ctr = 0;
+ int j = 0;
+ int ninput_items = nitems * d_decimation;
+
+ for (int i = 0; i < ninput_items; i += d_nsamples){
+
+ memcpy(d_fwdfft->get_inbuf(), &input[i], d_nsamples * sizeof(float));
+
+ for (j = d_nsamples; j < d_fftsize; j++)
+ d_fwdfft->get_inbuf()[j] = 0;
+
+ d_fwdfft->execute(); // compute fwd xform
+
+ float *a = (float*)(d_fwdfft->get_outbuf());
+ float *b = (float*)(&d_xformed_taps[0]);
+ float *c = (float*)(d_invfft->get_inbuf());
+
+ __m128 x0, x1, x2, t0, t1, m;
+ m = _mm_set_ps(-1, 1, -1, 1);
+ for (j = 0; j < d_fftsize; j+=4) { // filter in the freq domain
+ x0 = _mm_load_ps(&a[j]);
+ t0 = _mm_load_ps(&b[j]);
+
+ t1 = _mm_shuffle_ps(t0, t0, _MM_SHUFFLE(3, 3, 1, 1));
+ t0 = _mm_shuffle_ps(t0, t0, _MM_SHUFFLE(2, 2, 0, 0));
+ t1 = _mm_mul_ps(t1, m);
+
+ x1 = _mm_mul_ps(x0, t0);
+ x2 = _mm_mul_ps(x0, t1);
+
+ x2 = _mm_shuffle_ps(x2, x2, _MM_SHUFFLE(2, 3, 0, 1));
+ x2 = _mm_add_ps(x1, x2);
+
+ _mm_store_ps(&c[j], x2);
+ }
+
+ // Finish off the last one; do the complex multiply as floats
+ j = d_fftsize/2;
+ c[j] = (a[j] * b[j]) - (a[j+1] * b[j+1]);
+ c[j+1] = (a[j] * b[j+1]) + (a[j+1] * b[j]);
+
+ d_invfft->execute(); // compute inv xform
+
+ // add in the overlapping tail
+
+ for (j = 0; j < tailsize(); j++)
+ d_invfft->get_outbuf()[j] += d_tail[j];
+
+ // copy nsamples to output
+
+ //memcpy(out, d_invfft->get_outbuf(), d_nsamples * sizeof(float));
+ //out += d_nsamples;
+
+ j = dec_ctr;
+ while (j < d_nsamples) {
+ *output++ = d_invfft->get_outbuf()[j];
+ j += d_decimation;
+ }
+ dec_ctr = (j - d_nsamples);
+
+ // stash the tail
+ memcpy(&d_tail[0], d_invfft->get_outbuf() + d_nsamples,
+ tailsize() * sizeof(float));
+ }
+
+ assert(dec_ctr == 0);
+
+ return nitems;
+}
projects / debian / gnuradio / blobdiff