Hi Cor, hopefully, I have a fix: Could you try
git pull https://github.com/marcusmueller/gnuradio window_fix_floating_point_math (or, alternatively, attached patch, cd gnuradio; git apply /path/to/patchfile.patch ) Thank you!! Marcus On 27.06.2017 07:09, Cor Legemaat wrote: > https://github.com/gnuradio/gnuradio/issues/1348 > > Regards: > Cor > > On Mon, 2017-06-26 at 14:11 +0200, Marcus Müller wrote: >> >> That's mightily interesting! I feel like we should be doing bug >> reports, but I'm not sure where. >> >> >> On 26.06.2017 06:42, Cor Legemaat wrote: >>> Found it: >>> >>> Created an C++ app that called that function with the same parameter >>> values as with python for this flow graph. Witch I was able to debug >>> normally. >>> >>> In window.cc line 265, with i = ntaps - 1, temp = >>> 1.0000000000000000002 that cause sqrt (0) witch return "-nan" on the >>> next line and screw all the rest of the calculations. >>> >>> This only happens when compiled with "CFLAGS=-march=native -O2", if >>> I don't specify the march it's working correctly. The function is >>> called on my system with taps=787 and beta = 7. >>> >>> Regards: >>> Cor >>> >>> On Wed, 2017-06-21 at 15:13 +0200, Cor Legemaat wrote: >>>> Hi: >>>> >>>> Sorry for the late replay... The intel pc call >>>> filter.firdes.low_pass with the same values but return 768 proper >>>> float values, not like the <nan>'s on the AMD pc. >>>> >>>> Tried to debug with "nemiver /usr/bin/python2.7 -u >>>> <path>/fm_receiver.py" and the breakpoint at firdes.cc line 100 >>>> witch get triggered and I can read the function parameters but when >>>> I try to step true the function it jumps to the assembly of >>>> pthread. If I put more breakpoints in firdes.cc I get back to the >>>> function but cant read any variables any more. Also tried exporting >>>> "export GR_SCHEDULER=STS" but the same symptoms. >>>> >>>> Don't know if Ubuntu will trigger the bug it's probably compiled >>>> more generic... >>>> >>>> Regards: >>>> Cor >>>> >>>> On Wed, 2017-06-07 at 04:26 -0400, Anon Lister wrote: >>>>> I have an AMD system with the same chip running Ubuntu 16.xx. I >>>>> can probably try to duplicate this weekend, if Cor doesn't get to >>>>> it, as another data point. >>>>> >>>>> On Jun 5, 2017 3:14 PM, "Marcus Müller" <marcus.muel...@ettus.com >>>>> <mailto:marcus.muel...@ettus.com>> wrote: >>>>> >>>>> Hi Cor, >>>>> >>>>> Excuse the language, but faaaark. Ok, looks like we have a bug >>>>> in low_pass. Or in GCC. Or SWIG (which does the >>>>> python-wrapping of the code in firdes.cc). yay. >>>>> >>>>> So, let's narrow this down: on intel and amd64, same number of >>>>> taps, right? >>>>> >>>>> Then: If I asked you to use GDB to verify the C++ low_pass >>>>> function in gr::filter::firdes::low_pass actually returned the >>>>> right float values, would you feel that, with a few hints, be >>>>> able to do that? >>>>> >>>>> Best regards, >>>>> >>>>> Marcus >>>>> >>>>> >>>>> On 01.06.2017 07:20, Cor Legemaat wrote: >>>>>> Hi: >>>>>> >>>>>> filter.firdes.low_pass get called with: >>>>>> * fractional_bw = 0.4 >>>>>> * trans_width = 0.1 >>>>>> * mid_transition_band = 0.45 >>>>>> * interpolation = 24 >>>>>> >>>>>> But return: (nan, <788 times nan>) >>>>>> >>>>>> Regards: >>>>>> Cor >>>>>> >>>>>> On Tue, 2017-05-30 at 00:06 +0200, Marcus Müller wrote: >>>>>>> Hi Cor, >>>>>>>> * When using 1 as "taps" there is output. >>>>>>> Aha!! >>>>>>> So, here's the thing: something might be going wrong in the python >>>>>>> code that sets up the taps automatically if you don't set them >>>>>>> explicitly. >>>>>>> Maybe you can figure out where things go wrong; the interesting part >>>>>>> (maybe add some `print`s here?) from [1]: >>>>>>> >>>>>>> # If we don't have user-provided taps, reduce the interp and >>>>>>> # decim values by the GCD (if there is one) and then define >>>>>>> # the taps from these new values. >>>>>>> if taps is None: >>>>>>> interpolation = interpolation // d >>>>>>> decimation = decimation // d >>>>>>> taps = design_filter(interpolation, decimation, >>>>>>> fractional_bw) >>>>>>> >>>>>>> and >>>>>>> >>>>>>> >>>>>>> def design_filter(interpolation, decimation, fractional_bw): >>>>>>> """ >>>>>>> Given the interpolation rate, decimation rate and a fractional >>>>>>> bandwidth, >>>>>>> design a set of taps. >>>>>>> >>>>>>> Args: >>>>>>> interpolation: interpolation factor (integer > 0) >>>>>>> decimation: decimation factor (integer > 0) >>>>>>> fractional_bw: fractional bandwidth in (0, 0.5) 0.4 works >>>>>>> well. (float) >>>>>>> Returns: >>>>>>> : sequence of numbers >>>>>>> """ >>>>>>> >>>>>>> if fractional_bw >= 0.5 or fractional_bw <= 0: >>>>>>> raise ValueError, "Invalid fractional_bandwidth, must be in >>>>>>> (0, 0.5)" >>>>>>> >>>>>>> beta = 7.0 >>>>>>> halfband = 0.5 >>>>>>> rate = float(interpolation)/float(decimation) >>>>>>> if(rate >= 1.0): >>>>>>> trans_width = halfband - fractional_bw >>>>>>> mid_transition_band = halfband - trans_width/2.0 >>>>>>> else: >>>>>>> trans_width = rate*(halfband - fractional_bw) >>>>>>> mid_transition_band = rate*halfband - trans_width/2.0 >>>>>>> >>>>>>> taps = filter.firdes.low_pass(interpolation, >>>>>>> # gain >>>>>>> interpolation, >>>>>>> # Fs >>>>>>> mid_transition_band, >>>>>>> # trans mid point >>>>>>> trans_width, >>>>>>> # transition width >>>>>>> filter.firdes.WIN_KAISER, >>>>>>> beta) >>>>>>> # beta >>>>>>> >>>>>>> return taps >>>>>>> >>>>>>> >>>>>>> >>>>>>> Best regards, >>>>>>> Marcus >>>>>>> >>>>>>> [1] >>>>>>> https://github.com/gnuradio/gnuradio/blob/master/gr-filter/python >>>>>>> <https://github.com/gnuradio/gnuradio/blob/master/gr-filter/python> >>>>>>> /filter/rational_resampler.py >>>>>>> >>>>>>> On 29.05.2017 19:01, Cor Legemaat wrote: >>>>>>>> Hi: >>>>>>>> >>>>>>>> * The only warning is about the thread priority but that's on >>>>>>>> both. >>>>>>>> * Type "Complex->Complex (Complex Taps)" >>>>>>>> * When using 1 as "taps" there is output. >>>>>>>> >>>>>>>> I can open it in Nemiver if I know where to put the break point... >>>>>>>> >>>>>>>> Regards: >>>>>>>> Cor >>>>>>>> >>>>>>>> On Mon, 2017-05-29 at 11:36 +0200, Marcus Müller wrote: >>>>>>>>> Hi Cor, >>>>>>>>> that's kind of surprising¹. My first bet is that your AMD system >>>>>>>>> is >>>>>>>>> missing some dependency that the intel system has, so that >>>>>>>>> something >>>>>>>>> goes wrong during build. But then again, you shouldn't be seeing >>>>>>>>> the >>>>>>>>> rational resampler block at all in that case. Let's head straight >>>>>>>>> to >>>>>>>>> debugging: >>>>>>>>> * Do you get any warning/console output during the execution of >>>>>>>>> that >>>>>>>>> flow graph? >>>>>>>>> * Which is the input/output type (float or complex, orange or >>>>>>>>> blue >>>>>>>>> connector in GRC, if using that) >>>>>>>>> * If in GRC: when explicitly using [1,] as "taps", do you get >>>>>>>>> output? >>>>>>>>> Best regards, >>>>>>>>> Marcus >>>>>>>>> >>>>>>>>> ¹ "wat?!" >>>>>>>>> >>>>>>>>> On 29.05.2017 06:35, Cor Legemaat wrote: >>>>>>>>>> Hi: >>>>>>>>>> >>>>>>>>>> I have 2 different hardware setup's with funtoo/gentoo and >>>>>>>>>> gnuradio >>>>>>>>>> installed. On the Intel system the "Rational Resampler" is >>>>>>>>>> working >>>>>>>>>> correctly but on the AMD system there is no output. This is on >>>>>>>>>> a >>>>>>>>>> flow >>>>>>>>>> graph for an basic wide band fm receiver based on the USPR >>>>>>>>>> 10min fm >>>>>>>>>> receiver tutorial. >>>>>>>>>> >>>>>>>>>> AMD system: >>>>>>>>>> * AMD FX(tm)-8150 Eight-Core Processor >>>>>>>>>> * CPU_FLAGS_X86="aes avx fma4 mmx mmxext popcnt sse sse2 sse3 >>>>>>>>>> sse4_1 sse4_2 sse4a ssse3 xop" >>>>>>>>>> >>>>>>>>>> Intel system: >>>>>>>>>> * Intel(R) Core(TM) i5-2430M CPU @ 2.40GHz >>>>>>>>>> * CPU_FLAGS_X86="aes avx mmx mmxext popcnt sse sse2 sse3 >>>>>>>>>> sse4_1 >>>>>>>>>> sse4_2 >>>>>>>>>> ssse3" >>>>>>>>>> >>>>>>>>>> Tried with different versions of GNURadio and gcc but the same >>>>>>>>>> symptoms, both systems is compiled with CFLAGS="-march=native >>>>>>>>>> -O2 >>>>>>>>>> -pipe". At the moment it is gcc:6.3.0 and net- >>>>>>>>>> wireless/gnuradio- >>>>>>>>>> 3.7.11:0/3.7.11 USE="alsa analog atsc audio channels digital >>>>>>>>>> doc >>>>>>>>>> dtv >>>>>>>>>> examples fec filter grc noaa pager performance-counters >>>>>>>>>> portaudio >>>>>>>>>> qt4 >>>>>>>>>> uhd utils vocoder wavelet wxwidgets zeromq -fcd -jack -log -oss >>>>>>>>>> -sdl {- >>>>>>>>>> test} -trellis" PYTHON_TARGETS="python2_7" >>>>>>>>>> >>>>>>>>>> Where do I start to search? >>>>>>>>>> >>>>>>>>>> Regards: >>>>>>>>>> Cor >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> _______________________________________________ >>>>>>>>>> Discuss-gnuradio mailing list >>>>>>>>>> Discuss-gnuradio@gnu.org <mailto:Discuss-gnuradio@gnu.org> >>>>>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio >>>>>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> _______________________________________________ >>>>>>>>> Discuss-gnuradio mailing list >>>>>>>>> Discuss-gnuradio@gnu.org <mailto:Discuss-gnuradio@gnu.org> >>>>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio >>>>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio> >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Discuss-gnuradio mailing list >>>>>>> Discuss-gnuradio@gnu.org <mailto:Discuss-gnuradio@gnu.org> >>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio >>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio> >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Discuss-gnuradio mailing list >>>>>>> Discuss-gnuradio@gnu.org <mailto:Discuss-gnuradio@gnu.org> >>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio >>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio> >>>>> _______________________________________________ >>>>> Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org >>>>> <mailto:Discuss-gnuradio@gnu.org> >>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio >>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio> >>>>> >>>>> _______________________________________________ >>>>> Discuss-gnuradio mailing list >>>>> Discuss-gnuradio@gnu.org <mailto:Discuss-gnuradio@gnu.org> >>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio >>>> _______________________________________________ >>>> Discuss-gnuradio mailing list >>>> Discuss-gnuradio@gnu.org <mailto:Discuss-gnuradio@gnu.org> >>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>From ef5ffe4324472e80b3c81a8323172d552a2448bd Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Marcus=20M=C3=BCller?= <marcus.muel...@ettus.com> Date: Tue, 27 Jun 2017 11:53:14 +0200 Subject: [PATCH] Fixed floating point math bug (#1348) https://github.com/gnuradio/gnuradio/issues/1348 We were doing floating point math wrong. --- gr-fft/lib/window.cc | 17 +++++++++++------ 1 file changed, 11 insertions(+), 6 deletions(-) diff --git a/gr-fft/lib/window.cc b/gr-fft/lib/window.cc index 126b28978..e80d469e9 100644 --- a/gr-fft/lib/window.cc +++ b/gr-fft/lib/window.cc @@ -30,9 +30,10 @@ namespace gr { namespace fft { -#define IzeroEPSILON 1E-21 /* Max error acceptable in Izero */ +#define BESSEL_EPSILON 1E-21 /* Max error acceptable in bessel_i_zero */ - static double Izero(double x) + /*! \brief Iterative first kind Bessel function approximation */ + static double bessel_i_zero(double x) { double sum, u, halfx, temp; int n; @@ -45,7 +46,7 @@ namespace gr { temp *= temp; u *= temp; sum += u; - } while (u >= IzeroEPSILON*sum); + } while (u >= BESSEL_EPSILON*sum); return(sum); } @@ -257,13 +258,17 @@ namespace gr { std::vector<float> taps(ntaps); - double IBeta = 1.0/Izero(beta); + double bessel_i_beta = 1.0/bessel_i_zero(beta); double inm1 = 1.0/((double)(ntaps-1)); double temp; - for(int i = 0; i < ntaps; i++) { + /* dragging taps[0] out of the loop, because ((double)(-1))*((double)(-1)) + might be 1.0 + epsilon and then we get sqrt() of something < 0, i.e. NaN. + https://github.com/gnuradio/gnuradio/issues/1348 */ + taps[0] = bessel_i_zero(0) * bessel_i_beta; + for(int i = 1; i < ntaps; i++) { temp = 2*i*inm1 - 1; - taps[i] = Izero(beta*sqrt(1.0-temp*temp)) * IBeta; + taps[i] = bessel_i_zero(beta*sqrt(1.0-temp*temp)) * bessel_i_beta; } return taps; } -- 2.13.0
_______________________________________________ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
