On Thu, Mar 8, 2012 at 2:35 PM, Marcus D. Leech <mle...@ripnet.com> wrote:
> On 08/03/12 02:25 PM, Tom Rondeau wrote:
> >
> > That still might be indicative of the problem, though, that the flags
> > are being read incorrectly by the proc system. Very strange, that.
> >
> > Marcus, if you would, change the volk kernel from the aligned (_a) to
> > the unaligned one (_u) and see what that does. In the FFT, since it's
> > an FFTW buffer, it _should_ be aligned, but this will let us know.
> >
> > Tom
> >
> Ok, changed from the "_a" to the "_u" variant in
> gri_fft_filter_{ccc,fff}, and voila! It now works!
>
> --
> Principal Investigator
> Shirleys Bay Radio Astronomy Consortium
> http://www.sbrac.org
>
>
Thanks to Marcus' help, I figured out the error here was due to a very
silly assumption that I had made regarding our own array for the Fourier
taps. Apparently, this is always handled "correctly" (for Volk's definition
of correct) on 64-bit machines but not 32-bit ones.
Attached is a patch that uses fftw_malloc to create an array to store the
taps in. I created some helper functions in gri_fft to abstract the
specific use of fftw in the filter code itself like we've done with the
rest of fftw's invocation.
Please try this out and let me know. I've only verified it on a 32-bit VM,
so I just want to make sure there's nothing I'm missing (although it was
failing previously as expected, so I think this works).
There are still a couple of other 32-bit issues that I'm seeing in other
blocks. I'll look into those next.
Thanks,
Tom
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.cc b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.cc
index c894d62..1a9273a 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.cc
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.cc
@@ -45,6 +45,7 @@ gri_fft_filter_ccc_generic::~gri_fft_filter_ccc_generic ()
{
delete d_fwdfft;
delete d_invfft;
+ gri_fft_free(d_xformed_taps);
}
#if 0
@@ -115,7 +116,7 @@ gri_fft_filter_ccc_generic::compute_sizes(int ntaps)
delete d_invfft;
d_fwdfft = new gri_fft_complex(d_fftsize, true, d_nthreads);
d_invfft = new gri_fft_complex(d_fftsize, false, d_nthreads);
- d_xformed_taps.resize(d_fftsize);
+ d_xformed_taps = gri_fft_malloc_complex(d_fftsize);
}
}
@@ -152,7 +153,7 @@ gri_fft_filter_ccc_generic::filter (int nitems, const gr_complex *input, gr_comp
d_fwdfft->execute(); // compute fwd xform
gr_complex *a = d_fwdfft->get_outbuf();
- gr_complex *b = &d_xformed_taps[0];
+ gr_complex *b = d_xformed_taps;
gr_complex *c = d_invfft->get_inbuf();
volk_32fc_x2_multiply_32fc_a(c, a, b, d_fftsize);
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.h b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.h
index 217b9ab..899b59e 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.h
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.h
@@ -44,8 +44,8 @@ class GR_CORE_API gri_fft_filter_ccc_generic
gri_fft_complex *d_invfft; // inverse "plan"
int d_nthreads; // number of FFTW threads to use
std::vector<gr_complex> d_tail; // state carried between blocks for overlap-add
- std::vector<gr_complex> d_xformed_taps; // Fourier xformed taps
std::vector<gr_complex> d_new_taps;
+ gr_complex *d_xformed_taps; // Fourier xformed taps
void compute_sizes(int ntaps);
int tailsize() const { return d_ntaps - 1; }
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.cc b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.cc
index e7f66b7..0989c96 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.cc
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.cc
@@ -44,6 +44,7 @@ gri_fft_filter_fff_generic::~gri_fft_filter_fff_generic ()
{
delete d_fwdfft;
delete d_invfft;
+ gri_fft_free(d_xformed_taps);
}
/*
@@ -102,7 +103,7 @@ gri_fft_filter_fff_generic::compute_sizes(int ntaps)
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);
+ d_xformed_taps = gri_fft_malloc_complex(d_fftsize/2+1);
}
}
@@ -139,7 +140,7 @@ gri_fft_filter_fff_generic::filter (int nitems, const float *input, float *outpu
d_fwdfft->execute(); // compute fwd xform
gr_complex *a = d_fwdfft->get_outbuf();
- gr_complex *b = &d_xformed_taps[0];
+ gr_complex *b = d_xformed_taps;
gr_complex *c = d_invfft->get_inbuf();
volk_32fc_x2_multiply_32fc_a(c, a, b, d_fftsize/2+1);
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.h b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.h
index be31068..6ac30ce 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.h
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.h
@@ -41,8 +41,8 @@ class GR_CORE_API gri_fft_filter_fff_generic
gri_fft_real_rev *d_invfft; // inverse "plan"
int d_nthreads; // number of FFTW threads to use
std::vector<float> d_tail; // state carried between blocks for overlap-add
- std::vector<gr_complex> d_xformed_taps; // Fourier xformed taps
std::vector<float> d_new_taps;
+ gr_complex *d_xformed_taps; // Fourier xformed taps
void compute_sizes(int ntaps);
diff --git a/gnuradio-core/src/lib/general/gri_fft.cc b/gnuradio-core/src/lib/general/gri_fft.cc
index 63e3077..2edb5f5 100644
--- a/gnuradio-core/src/lib/general/gri_fft.cc
+++ b/gnuradio-core/src/lib/general/gri_fft.cc
@@ -47,6 +47,24 @@ static int my_fftw_read_char(void *f) { return fgetc((FILE *) f); }
#include <boost/filesystem/path.hpp>
namespace fs = boost::filesystem;
+gr_complex *
+gri_fft_malloc_complex(int size)
+{
+ return (gr_complex*)fftwf_malloc(sizeof(gr_complex)*size);
+}
+
+float *
+gri_fft_malloc_float(int size)
+{
+ return (float*)fftwf_malloc(sizeof(float)*size);
+}
+
+void
+gri_fft_free(void *b)
+{
+ fftwf_free(b);
+}
+
boost::mutex &
gri_fft_planner::mutex()
{
diff --git a/gnuradio-core/src/lib/general/gri_fft.h b/gnuradio-core/src/lib/general/gri_fft.h
index ed80bad..f77a18e 100644
--- a/gnuradio-core/src/lib/general/gri_fft.h
+++ b/gnuradio-core/src/lib/general/gri_fft.h
@@ -30,6 +30,19 @@
#include <gr_complex.h>
#include <boost/thread.hpp>
+/*! \brief Helper function for allocating complex fft buffers
+ */
+gr_complex* gri_fft_malloc_complex(int size);
+
+/*! \brief Helper function for allocating float fft buffers
+ */
+float* gri_fft_malloc_float(int size);
+
+/*! \brief Helper function for freeing fft buffers
+ */
+void gri_fft_free(void *b);
+
+
/*!
* \brief Export reference to planner mutex for those apps that
* want to use FFTW w/o using the gri_fftw* classes.
_______________________________________________
Discuss-gnuradio mailing list
Discuss-gnuradio@gnu.org
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
