[hamradio-commits] [gnss-sdr] 233/251: First working version of the GPU GPS tracking block (it requires NVIDIA CUDA 3.0 GPU hardware)
Carles Fernandez
carles_fernandez-guest at moszumanska.debian.org
Wed Sep 2 00:22:56 UTC 2015
This is an automated email from the git hooks/post-receive script.
carles_fernandez-guest pushed a commit to branch master
in repository gnss-sdr.
commit 26cf90cdd488620b3b63ddc2002fe1be35abcd84
Author: Javier Arribas <javiarribas at gmail.com>
Date: Fri Jul 24 17:21:25 2015 +0200
First working version of the GPU GPS tracking block (it requires NVIDIA
CUDA
3.0 GPU hardware)
---
conf/gnss-sdr_GPS_L1_gr_complex_gpu.conf | 305 +++++++++++++++++++++
src/algorithms/tracking/adapters/CMakeLists.txt | 5 +
.../adapters/gps_l1_ca_dll_pll_tracking_gpu.cc | 158 +++++++++++
.../adapters/gps_l1_ca_dll_pll_tracking_gpu.h | 113 ++++++++
.../tracking/gnuradio_blocks/CMakeLists.txt | 3 +-
.../gps_l1_ca_dll_pll_tracking_gpu_cc.cc | 20 +-
src/algorithms/tracking/libs/CMakeLists.txt | 8 +-
.../tracking/libs/cuda_multicorrelator.cu | 31 ++-
.../tracking/libs/cuda_multicorrelator.h | 9 +-
src/core/receiver/CMakeLists.txt | 7 +-
src/core/receiver/gnss_block_factory.cc | 7 +
src/main/CMakeLists.txt | 1 +
12 files changed, 631 insertions(+), 36 deletions(-)
diff --git a/conf/gnss-sdr_GPS_L1_gr_complex_gpu.conf b/conf/gnss-sdr_GPS_L1_gr_complex_gpu.conf
new file mode 100644
index 0000000..a8e576d
--- /dev/null
+++ b/conf/gnss-sdr_GPS_L1_gr_complex_gpu.conf
@@ -0,0 +1,305 @@
+; Default configuration file
+; You can define your own receiver and invoke it by doing
+; gnss-sdr --config_file=my_GNSS_SDR_configuration.conf
+;
+
+[GNSS-SDR]
+
+;######### GLOBAL OPTIONS ##################
+;internal_fs_hz: Internal signal sampling frequency after the signal conditioning stage [Hz].
+GNSS-SDR.internal_fs_hz=4000000
+
+;######### CONTROL_THREAD CONFIG ############
+ControlThread.wait_for_flowgraph=false
+
+;######### SIGNAL_SOURCE CONFIG ############
+;#implementation: Use [File_Signal_Source] or [UHD_Signal_Source] or [GN3S_Signal_Source] (experimental)
+SignalSource.implementation=File_Signal_Source
+
+;#filename: path to file with the captured GNSS signal samples to be processed
+SignalSource.filename=/home/javier/signals/4msps.dat
+
+;#item_type: Type and resolution for each of the signal samples. Use only gr_complex in this version.
+SignalSource.item_type=gr_complex
+
+;#sampling_frequency: Original Signal sampling frequency in [Hz]
+SignalSource.sampling_frequency=4000000
+
+;#freq: RF front-end center frequency in [Hz]
+SignalSource.freq=1575420000
+
+;#gain: Front-end Gain in [dB]
+SignalSource.gain=60
+
+;#subdevice: UHD subdevice specification (for USRP1 use A:0 or B:0)
+SignalSource.subdevice=B:0
+
+;#samples: Number of samples to be processed. Notice that 0 indicates the entire file.
+SignalSource.samples=0
+
+;#repeat: Repeat the processing file. Disable this option in this version
+SignalSource.repeat=false
+
+;#dump: Dump the Signal source data to a file. Disable this option in this version
+SignalSource.dump=false
+
+SignalSource.dump_filename=../data/signal_source.dat
+
+
+;#enable_throttle_control: Enabling this option tells the signal source to keep the delay between samples in post processing.
+; it helps to not overload the CPU, but the processing time will be longer.
+SignalSource.enable_throttle_control=false
+
+
+;######### SIGNAL_CONDITIONER CONFIG ############
+;## It holds blocks to change data type, filter and resample input data.
+
+;#implementation: Use [Pass_Through] or [Signal_Conditioner]
+;#[Pass_Through] disables this block and the [DataTypeAdapter], [InputFilter] and [Resampler] blocks
+;#[Signal_Conditioner] enables this block. Then you have to configure [DataTypeAdapter], [InputFilter] and [Resampler] blocks
+;SignalConditioner.implementation=Signal_Conditioner
+SignalConditioner.implementation=Pass_Through
+
+;######### DATA_TYPE_ADAPTER CONFIG ############
+;## Changes the type of input data. Please disable it in this version.
+;#implementation: [Pass_Through] disables this block
+DataTypeAdapter.implementation=Pass_Through
+
+;######### INPUT_FILTER CONFIG ############
+;## Filter the input data. Can be combined with frequency translation for IF signals
+
+;#implementation: Use [Pass_Through] or [Fir_Filter] or [Freq_Xlating_Fir_Filter]
+;#[Pass_Through] disables this block
+;#[Fir_Filter] enables a FIR Filter
+;#[Freq_Xlating_Fir_Filter] enables FIR filter and a composite frequency translation that shifts IF down to zero Hz.
+
+;InputFilter.implementation=Fir_Filter
+;InputFilter.implementation=Freq_Xlating_Fir_Filter
+InputFilter.implementation=Pass_Through
+
+;#dump: Dump the filtered data to a file.
+InputFilter.dump=false
+
+;#dump_filename: Log path and filename.
+InputFilter.dump_filename=../data/input_filter.dat
+
+;#The following options are used in the filter design of Fir_Filter and Freq_Xlating_Fir_Filter implementation.
+;#These options are based on parameters of gnuradio's function: gr_remez.
+;#These function calculates the optimal (in the Chebyshev/minimax sense) FIR filter inpulse reponse given a set of band edges, the desired reponse on those bands, and the weight given to the error in those bands.
+
+;#input_item_type: Type and resolution for input signal samples. Use only gr_complex in this version.
+InputFilter.input_item_type=gr_complex
+
+;#outut_item_type: Type and resolution for output filtered signal samples. Use only gr_complex in this version.
+InputFilter.output_item_type=gr_complex
+
+;#taps_item_type: Type and resolution for the taps of the filter. Use only float in this version.
+InputFilter.taps_item_type=float
+
+;#number_of_taps: Number of taps in the filter. Increasing this parameter increases the processing time
+InputFilter.number_of_taps=5
+
+;#number_of _bands: Number of frequency bands in the filter.
+InputFilter.number_of_bands=2
+
+;#bands: frequency at the band edges [ b1 e1 b2 e2 b3 e3 ...].
+;#Frequency is in the range [0, 1], with 1 being the Nyquist frequency (Fs/2)
+;#The number of band_begin and band_end elements must match the number of bands
+
+InputFilter.band1_begin=0.0
+InputFilter.band1_end=0.45
+InputFilter.band2_begin=0.55
+InputFilter.band2_end=1.0
+
+;#ampl: desired amplitude at the band edges [ a(b1) a(e1) a(b2) a(e2) ...].
+;#The number of ampl_begin and ampl_end elements must match the number of bands
+
+InputFilter.ampl1_begin=1.0
+InputFilter.ampl1_end=1.0
+InputFilter.ampl2_begin=0.0
+InputFilter.ampl2_end=0.0
+
+;#band_error: weighting applied to each band (usually 1).
+;#The number of band_error elements must match the number of bands
+InputFilter.band1_error=1.0
+InputFilter.band2_error=1.0
+
+;#filter_type: one of "bandpass", "hilbert" or "differentiator"
+InputFilter.filter_type=bandpass
+
+;#grid_density: determines how accurately the filter will be constructed.
+;The minimum value is 16; higher values are slower to compute the filter.
+InputFilter.grid_density=16
+
+;#The following options are used only in Freq_Xlating_Fir_Filter implementation.
+;#InputFilter.IF is the intermediate frequency (in Hz) shifted down to zero Hz
+
+InputFilter.sampling_frequency=4000000
+InputFilter.IF=0
+
+
+
+;######### RESAMPLER CONFIG ############
+;## Resamples the input data.
+
+;#implementation: Use [Pass_Through] or [Direct_Resampler]
+;#[Pass_Through] disables this block
+;#[Direct_Resampler] enables a resampler that implements a nearest neigbourhood interpolation
+;Resampler.implementation=Direct_Resampler
+Resampler.implementation=Pass_Through
+
+;#dump: Dump the resamplered data to a file.
+Resampler.dump=false
+;#dump_filename: Log path and filename.
+Resampler.dump_filename=../data/resampler.dat
+
+;#item_type: Type and resolution for each of the signal samples. Use only gr_complex in this version.
+Resampler.item_type=gr_complex
+
+;#sample_freq_in: the sample frequency of the input signal
+Resampler.sample_freq_in=8000000
+
+;#sample_freq_out: the desired sample frequency of the output signal
+Resampler.sample_freq_out=4000000
+
+
+;######### CHANNELS GLOBAL CONFIG ############
+;#count: Number of available GPS satellite channels.
+Channels_GPS.count=8
+;#count: Number of available Galileo satellite channels.
+Channels_Galileo.count=0
+;#in_acquisition: Number of channels simultaneously acquiring for the whole receiver
+Channels.in_acquisition=1
+;#system: GPS, GLONASS, GALILEO, SBAS or COMPASS
+;#if the option is disabled by default is assigned GPS
+Channel.system=GPS
+
+;#if the option is disabled by default is assigned "1C" GPS L1 C/A
+Channel.signal=1C
+
+
+;######### SPECIFIC CHANNELS CONFIG ######
+;#The following options are specific to each channel and overwrite the generic options
+
+;######### CHANNEL 0 CONFIG ############
+
+;Channel0.system=GPS
+;Channel0.signal=1C
+
+;#satellite: Satellite PRN ID for this channel. Disable this option to random search
+;Channel0.satellite=11
+
+;######### CHANNEL 1 CONFIG ############
+
+;Channel1.system=GPS
+;Channel1.signal=1C
+;Channel1.satellite=18
+
+;######### ACQUISITION GLOBAL CONFIG ############
+
+;#dump: Enable or disable the acquisition internal data file logging [true] or [false]
+Acquisition_GPS.dump=false
+;#filename: Log path and filename
+Acquisition_GPS.dump_filename=./acq_dump.dat
+;#item_type: Type and resolution for each of the signal samples. Use only gr_complex in this version.
+Acquisition_GPS.item_type=gr_complex
+;#if: Signal intermediate frequency in [Hz]
+Acquisition_GPS.if=0
+;#sampled_ms: Signal block duration for the acquisition signal detection [ms]
+Acquisition_GPS.sampled_ms=1
+;#implementation: Acquisition algorithm selection for this channel: [GPS_L1_CA_PCPS_Acquisition] or [Galileo_E1_PCPS_Ambiguous_Acquisition]
+Acquisition_GPS.implementation=GPS_L1_CA_PCPS_Acquisition
+;#threshold: Acquisition threshold
+Acquisition_GPS.threshold=0.005
+;#pfa: Acquisition false alarm probability. This option overrides the threshold option. Only use with implementations: [GPS_L1_CA_PCPS_Acquisition] or [Galileo_E1_PCPS_Ambiguous_Acquisition]
+;Acquisition_GPS.pfa=0.01
+;#doppler_max: Maximum expected Doppler shift [Hz]
+Acquisition_GPS.doppler_max=10000
+;#doppler_max: Doppler step in the grid search [Hz]
+Acquisition_GPS.doppler_step=500
+
+;######### TRACKING GLOBAL CONFIG ############
+
+;#implementation: Selected tracking algorithm: [GPS_L1_CA_DLL_PLL_Tracking] or [GPS_L1_CA_DLL_FLL_PLL_Tracking] or [GPS_L1_CA_TCP_CONNECTOR_Tracking] or [Galileo_E1_DLL_PLL_VEML_Tracking]
+Tracking_GPS.implementation=GPS_L1_CA_DLL_PLL_Tracking_GPU
+;#item_type: Type and resolution for each of the signal samples. Use only [gr_complex] in this version.
+Tracking_GPS.item_type=gr_complex
+
+;#sampling_frequency: Signal Intermediate Frequency in [Hz]
+Tracking_GPS.if=0
+
+;#dump: Enable or disable the Tracking internal binary data file logging [true] or [false]
+Tracking_GPS.dump=false
+
+;#dump_filename: Log path and filename. Notice that the tracking channel will add "x.dat" where x is the channel number.
+Tracking_GPS.dump_filename=../data/epl_tracking_ch_
+
+;#pll_bw_hz: PLL loop filter bandwidth [Hz]
+Tracking_GPS.pll_bw_hz=45.0;
+
+;#dll_bw_hz: DLL loop filter bandwidth [Hz]
+Tracking_GPS.dll_bw_hz=2.0;
+
+;#fll_bw_hz: FLL loop filter bandwidth [Hz]
+Tracking_GPS.fll_bw_hz=10.0;
+
+;#order: PLL/DLL loop filter order [2] or [3]
+Tracking_GPS.order=3;
+
+;######### TELEMETRY DECODER GPS CONFIG ############
+;#implementation: Use [GPS_L1_CA_Telemetry_Decoder] for GPS L1 C/A
+TelemetryDecoder_GPS.implementation=GPS_L1_CA_Telemetry_Decoder
+TelemetryDecoder_GPS.dump=false
+;#decimation factor
+TelemetryDecoder_GPS.decimation_factor=1;
+
+;######### OBSERVABLES CONFIG ############
+;#implementation: Use [GPS_L1_CA_Observables] for GPS L1 C/A.
+Observables.implementation=GPS_L1_CA_Observables
+
+;#dump: Enable or disable the Observables internal binary data file logging [true] or [false]
+Observables.dump=false
+
+;#dump_filename: Log path and filename.
+Observables.dump_filename=./observables.dat
+
+
+;######### PVT CONFIG ############
+;#implementation: Position Velocity and Time (PVT) implementation algorithm: Use [GPS_L1_CA_PVT] in this version.
+PVT.implementation=GPS_L1_CA_PVT
+
+;#averaging_depth: Number of PVT observations in the moving average algorithm
+PVT.averaging_depth=100
+
+;#flag_average: Enables the PVT averaging between output intervals (arithmetic mean) [true] or [false]
+PVT.flag_averaging=false
+
+;#output_rate_ms: Period between two PVT outputs. Notice that the minimum period is equal to the tracking integration time (for GPS CA L1 is 1ms) [ms]
+PVT.output_rate_ms=10
+
+;#display_rate_ms: Position console print (std::out) interval [ms]. Notice that output_rate_ms<=display_rate_ms.
+PVT.display_rate_ms=500
+
+;# RINEX, KML, and NMEA output configuration
+
+;#dump_filename: Log path and filename without extension. Notice that PVT will add ".dat" to the binary dump and ".kml" to GoogleEarth dump.
+PVT.dump_filename=./PVT
+
+;#nmea_dump_filename: NMEA log path and filename
+PVT.nmea_dump_filename=./gnss_sdr_pvt.nmea;
+
+;#flag_nmea_tty_port: Enable or disable the NMEA log to a serial TTY port (Can be used with real hardware or virtual one)
+PVT.flag_nmea_tty_port=false;
+
+;#nmea_dump_devname: serial device descriptor for NMEA logging
+PVT.nmea_dump_devname=/dev/pts/4
+
+
+;#dump: Enable or disable the PVT internal binary data file logging [true] or [false]
+PVT.dump=false
+
+;######### OUTPUT_FILTER CONFIG ############
+;# Receiver output filter: Leave this block disabled in this version
+OutputFilter.implementation=Null_Sink_Output_Filter
+OutputFilter.filename=data/gnss-sdr.dat
+OutputFilter.item_type=gr_complex
diff --git a/src/algorithms/tracking/adapters/CMakeLists.txt b/src/algorithms/tracking/adapters/CMakeLists.txt
index fda0724..3e92ac6 100644
--- a/src/algorithms/tracking/adapters/CMakeLists.txt
+++ b/src/algorithms/tracking/adapters/CMakeLists.txt
@@ -16,6 +16,9 @@
# along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
#
+if(ENABLE_CUDA)
+ FIND_PACKAGE(CUDA REQUIRED)
+endif(ENABLE_CUDA)
set(TRACKING_ADAPTER_SOURCES
galileo_e1_dll_pll_veml_tracking.cc
@@ -27,6 +30,7 @@ set(TRACKING_ADAPTER_SOURCES
gps_l1_ca_tcp_connector_tracking.cc
galileo_e5a_dll_pll_tracking.cc
gps_l2_m_dll_pll_tracking.cc
+ gps_l1_ca_dll_pll_tracking_gpu.cc
)
include_directories(
@@ -40,6 +44,7 @@ include_directories(
${GLOG_INCLUDE_DIRS}
${GFlags_INCLUDE_DIRS}
${GNURADIO_RUNTIME_INCLUDE_DIRS}
+ ${CUDA_INCLUDE_DIRS}
)
file(GLOB TRACKING_ADAPTER_HEADERS "*.h")
diff --git a/src/algorithms/tracking/adapters/gps_l1_ca_dll_pll_tracking_gpu.cc b/src/algorithms/tracking/adapters/gps_l1_ca_dll_pll_tracking_gpu.cc
new file mode 100644
index 0000000..4ae8864
--- /dev/null
+++ b/src/algorithms/tracking/adapters/gps_l1_ca_dll_pll_tracking_gpu.cc
@@ -0,0 +1,158 @@
+/*!
+ * \file gps_l1_ca_dll_pll_tracking_gpu.cc
+ * \brief Implementation of an adapter of a DLL+PLL tracking loop block using GPU accelerated functions
+ * for GPS L1 C/A to a TrackingInterface
+ * \author Javier Arribas, 2015. jarribas(at)cttc.es
+ *
+ * Code DLL + carrier PLL according to the algorithms described in:
+ * K.Borre, D.M.Akos, N.Bertelsen, P.Rinder, and S.H.Jensen,
+ * A Software-Defined GPS and Galileo Receiver. A Single-Frequency
+ * Approach, Birkhauser, 2007
+ *
+ * -------------------------------------------------------------------------
+ *
+ * Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
+ *
+ * GNSS-SDR is a software defined Global Navigation
+ * Satellite Systems receiver
+ *
+ * This file is part of GNSS-SDR.
+ *
+ * GNSS-SDR 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 of the License, or
+ * (at your option) any later version.
+ *
+ * GNSS-SDR 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 GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
+ *
+ * -------------------------------------------------------------------------
+ */
+
+
+#include "gps_l1_ca_dll_pll_tracking_gpu.h"
+#include <glog/logging.h>
+#include "GPS_L1_CA.h"
+#include "configuration_interface.h"
+
+
+using google::LogMessage;
+
+GpsL1CaDllPllTrackingGPU::GpsL1CaDllPllTrackingGPU(
+ ConfigurationInterface* configuration, std::string role,
+ unsigned int in_streams, unsigned int out_streams,
+ boost::shared_ptr<gr::msg_queue> queue) :
+ role_(role), in_streams_(in_streams), out_streams_(out_streams),
+ queue_(queue)
+{
+ DLOG(INFO) << "role " << role;
+ //################# CONFIGURATION PARAMETERS ########################
+ int fs_in;
+ int vector_length;
+ int f_if;
+ bool dump;
+ std::string dump_filename;
+ std::string item_type;
+ std::string default_item_type = "gr_complex";
+ float pll_bw_hz;
+ float dll_bw_hz;
+ float early_late_space_chips;
+ item_type = configuration->property(role + ".item_type", default_item_type);
+ //vector_length = configuration->property(role + ".vector_length", 2048);
+ fs_in = configuration->property("GNSS-SDR.internal_fs_hz", 2048000);
+ f_if = configuration->property(role + ".if", 0);
+ dump = configuration->property(role + ".dump", false);
+ pll_bw_hz = configuration->property(role + ".pll_bw_hz", 50.0);
+ dll_bw_hz = configuration->property(role + ".dll_bw_hz", 2.0);
+ early_late_space_chips = configuration->property(role + ".early_late_space_chips", 0.5);
+ std::string default_dump_filename = "./track_ch";
+ dump_filename = configuration->property(role + ".dump_filename",
+ default_dump_filename); //unused!
+ vector_length = std::round(fs_in / (GPS_L1_CA_CODE_RATE_HZ / GPS_L1_CA_CODE_LENGTH_CHIPS));
+
+ //################# MAKE TRACKING GNURadio object ###################
+ if (item_type.compare("gr_complex") == 0)
+ {
+ item_size_ = sizeof(gr_complex);
+ tracking_ = gps_l1_ca_dll_pll_make_tracking_gpu_cc(
+ f_if,
+ fs_in,
+ vector_length,
+ queue_,
+ dump,
+ dump_filename,
+ pll_bw_hz,
+ dll_bw_hz,
+ early_late_space_chips);
+ }
+ else
+ {
+ item_size_ = sizeof(gr_complex);
+ LOG(WARNING) << item_type << " unknown tracking item type.";
+ }
+ channel_ = 0;
+ channel_internal_queue_ = 0;
+ DLOG(INFO) << "tracking(" << tracking_->unique_id() << ")";
+}
+
+
+GpsL1CaDllPllTrackingGPU::~GpsL1CaDllPllTrackingGPU()
+{}
+
+
+void GpsL1CaDllPllTrackingGPU::start_tracking()
+{
+ tracking_->start_tracking();
+}
+
+/*
+ * Set tracking channel unique ID
+ */
+void GpsL1CaDllPllTrackingGPU::set_channel(unsigned int channel)
+{
+ channel_ = channel;
+ tracking_->set_channel(channel);
+}
+
+/*
+ * Set tracking channel internal queue
+ */
+void GpsL1CaDllPllTrackingGPU::set_channel_queue(
+ concurrent_queue<int> *channel_internal_queue)
+{
+ channel_internal_queue_ = channel_internal_queue;
+ tracking_->set_channel_queue(channel_internal_queue_);
+}
+
+void GpsL1CaDllPllTrackingGPU::set_gnss_synchro(Gnss_Synchro* p_gnss_synchro)
+{
+ tracking_->set_gnss_synchro(p_gnss_synchro);
+}
+
+void GpsL1CaDllPllTrackingGPU::connect(gr::top_block_sptr top_block)
+{
+ if(top_block) { /* top_block is not null */};
+ //nothing to connect, now the tracking uses gr_sync_decimator
+}
+
+void GpsL1CaDllPllTrackingGPU::disconnect(gr::top_block_sptr top_block)
+{
+ if(top_block) { /* top_block is not null */};
+ //nothing to disconnect, now the tracking uses gr_sync_decimator
+}
+
+gr::basic_block_sptr GpsL1CaDllPllTrackingGPU::get_left_block()
+{
+ return tracking_;
+}
+
+gr::basic_block_sptr GpsL1CaDllPllTrackingGPU::get_right_block()
+{
+ return tracking_;
+}
+
diff --git a/src/algorithms/tracking/adapters/gps_l1_ca_dll_pll_tracking_gpu.h b/src/algorithms/tracking/adapters/gps_l1_ca_dll_pll_tracking_gpu.h
new file mode 100644
index 0000000..22bbe5d
--- /dev/null
+++ b/src/algorithms/tracking/adapters/gps_l1_ca_dll_pll_tracking_gpu.h
@@ -0,0 +1,113 @@
+/*!
+ * \file gps_l1_ca_dll_pll_tracking_gpu.h
+ * \brief Implementation of an adapter of a DLL+PLL tracking loop block using GPU accelerated functions
+ * for GPS L1 C/A to a TrackingInterface
+ * \author Javier Arribas, 2015. jarribas(at)cttc.es
+ *
+ * Code DLL + carrier PLL according to the algorithms described in:
+ * K.Borre, D.M.Akos, N.Bertelsen, P.Rinder, and S.H.Jensen,
+ * A Software-Defined GPS and Galileo Receiver. A Single-Frequency
+ * Approach, Birkha user, 2007
+ *
+ * -------------------------------------------------------------------------
+ *
+ * Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
+ *
+ * GNSS-SDR is a software defined Global Navigation
+ * Satellite Systems receiver
+ *
+ * This file is part of GNSS-SDR.
+ *
+ * GNSS-SDR 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 of the License, or
+ * (at your option) any later version.
+ *
+ * GNSS-SDR 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 GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
+ *
+ * -------------------------------------------------------------------------
+ */
+
+#ifndef GNSS_SDR_GPS_L1_CA_DLL_PLL_TRACKING_GPU_H_
+#define GNSS_SDR_GPS_L1_CA_DLL_PLL_TRACKING_GPU_H_
+
+#include <string>
+#include <gnuradio/msg_queue.h>
+#include "tracking_interface.h"
+#include "gps_l1_ca_dll_pll_tracking_gpu_cc.h"
+
+
+class ConfigurationInterface;
+
+/*!
+ * \brief This class implements a code DLL + carrier PLL tracking loop using GPU accelerated functions
+ */
+class GpsL1CaDllPllTrackingGPU : public TrackingInterface
+{
+public:
+
+ GpsL1CaDllPllTrackingGPU(ConfigurationInterface* configuration,
+ std::string role,
+ unsigned int in_streams,
+ unsigned int out_streams,
+ boost::shared_ptr<gr::msg_queue> queue);
+
+ virtual ~GpsL1CaDllPllTrackingGPU();
+
+ std::string role()
+ {
+ return role_;
+ }
+
+ //! Returns "GPS_L1_CA_DLL_PLL_Tracking"
+ std::string implementation()
+ {
+ return "GPS_L1_CA_DLL_PLL_Tracking_GPU";
+ }
+ size_t item_size()
+ {
+ return item_size_;
+ }
+
+ void connect(gr::top_block_sptr top_block);
+ void disconnect(gr::top_block_sptr top_block);
+ gr::basic_block_sptr get_left_block();
+ gr::basic_block_sptr get_right_block();
+
+
+ /*!
+ * \brief Set tracking channel unique ID
+ */
+ void set_channel(unsigned int channel);
+
+ /*!
+ * \brief Set acquisition/tracking common Gnss_Synchro object pointer
+ * to efficiently exchange synchronization data between acquisition and tracking blocks
+ */
+ void set_gnss_synchro(Gnss_Synchro* p_gnss_synchro);
+
+ /*!
+ * \brief Set tracking channel internal queue
+ */
+ void set_channel_queue(concurrent_queue<int> *channel_internal_queue);
+
+ void start_tracking();
+
+private:
+ gps_l1_ca_dll_pll_tracking_gpu_cc_sptr tracking_;
+ size_t item_size_;
+ unsigned int channel_;
+ std::string role_;
+ unsigned int in_streams_;
+ unsigned int out_streams_;
+ boost::shared_ptr<gr::msg_queue> queue_;
+ concurrent_queue<int> *channel_internal_queue_;
+};
+
+#endif // GNSS_SDR_GPS_L1_CA_DLL_PLL_TRACKING_GPU_H_
diff --git a/src/algorithms/tracking/gnuradio_blocks/CMakeLists.txt b/src/algorithms/tracking/gnuradio_blocks/CMakeLists.txt
index d300a9e..8589832 100644
--- a/src/algorithms/tracking/gnuradio_blocks/CMakeLists.txt
+++ b/src/algorithms/tracking/gnuradio_blocks/CMakeLists.txt
@@ -57,7 +57,8 @@ endif(ENABLE_GENERIC_ARCH)
file(GLOB TRACKING_GR_BLOCKS_HEADERS "*.h")
add_library(tracking_gr_blocks ${TRACKING_GR_BLOCKS_SOURCES} ${TRACKING_GR_BLOCKS_HEADERS})
source_group(Headers FILES ${TRACKING_GR_BLOCKS_HEADERS})
-target_link_libraries(tracking_gr_blocks tracking_lib ${GNURADIO_RUNTIME_LIBRARIES} gnss_sp_libs ${Boost_LIBRARIES} ${VOLK_GNSSSDR_LIBRARIES} ${ORC_LIBRARIES} )
+
+target_link_libraries(tracking_gr_blocks tracking_lib ${GNURADIO_RUNTIME_LIBRARIES} gnss_sp_libs ${Boost_LIBRARIES} ${VOLK_GNSSSDR_LIBRARIES} ${ORC_LIBRARIES} ${CUDA_LIBRARIES})
if(NOT VOLK_GNSSSDR_FOUND)
add_dependencies(tracking_gr_blocks volk_gnsssdr_module)
endif(NOT VOLK_GNSSSDR_FOUND)
diff --git a/src/algorithms/tracking/gnuradio_blocks/gps_l1_ca_dll_pll_tracking_gpu_cc.cc b/src/algorithms/tracking/gnuradio_blocks/gps_l1_ca_dll_pll_tracking_gpu_cc.cc
index b443aeb..3de6111 100644
--- a/src/algorithms/tracking/gnuradio_blocks/gps_l1_ca_dll_pll_tracking_gpu_cc.cc
+++ b/src/algorithms/tracking/gnuradio_blocks/gps_l1_ca_dll_pll_tracking_gpu_cc.cc
@@ -48,17 +48,11 @@
#include "GPS_L1_CA.h"
#include "control_message_factory.h"
#include <volk/volk.h> //volk_alignement
-
-#include <cuda.h>
-// CUDA runtime
-#include <cuda_runtime.h>
// includes
#include <cuda_profiler_api.h>
#include <helper_functions.h> // helper for shared functions common to CUDA Samples
#include <helper_cuda.h> // helper functions for CUDA error checking and initialization
-
-
/*!
* \todo Include in definition header file
*/
@@ -130,8 +124,9 @@ Gps_L1_Ca_Dll_Pll_Tracking_GPU_cc::Gps_L1_Ca_Dll_Pll_Tracking_GPU_cc(
multicorrelator_gpu = new cuda_multicorrelator();
- // Get space for the resampled early / prompt / late local replicas
int N_CORRELATORS=3;
+ multicorrelator_gpu->init_cuda(0, NULL, 2 * d_vector_length , 2 * d_vector_length , N_CORRELATORS);
+ // Get space for the resampled early / prompt / late local replicas
checkCudaErrors(cudaHostAlloc((void**)&d_local_code_shift_samples, N_CORRELATORS * sizeof(int), cudaHostAllocMapped ));
@@ -323,9 +318,6 @@ Gps_L1_Ca_Dll_Pll_Tracking_GPU_cc::~Gps_L1_Ca_Dll_Pll_Tracking_GPU_cc()
multicorrelator_gpu->free_cuda();
delete(multicorrelator_gpu);
- volk_free(d_Early);
- volk_free(d_Prompt);
- volk_free(d_Late);
volk_free(d_ca_code);
delete[] d_Prompt_buffer;
@@ -381,18 +373,20 @@ int Gps_L1_Ca_Dll_Pll_Tracking_GPU_cc::general_work (int noutput_items, gr_vecto
// UPDATE NCO COMMAND
float phase_step_rad = static_cast<float>(GPS_TWO_PI) * d_carrier_doppler_hz / static_cast<float>(d_fs_in);
-
+ //std::cout<<"d_current_prn_length_samples="<<d_current_prn_length_samples<<std::endl;
// perform carrier wipe-off and compute Early, Prompt and Late correlation
+ cudaProfilerStart();
multicorrelator_gpu->Carrier_wipeoff_multicorrelator_cuda(
d_corr_outs_gpu,
- in,//in_gpu,
+ in,
d_local_codes_gpu,
d_rem_carr_phase_rad,
phase_step_rad,
d_local_code_shift_samples,
d_current_prn_length_samples,
3);
-
+ cudaProfilerStop();
+ //std::cout<<"d_Prompt="<<*d_Prompt<<"d_Early="<<*d_Early<<"d_Late="<<*d_Late<<std::endl;
// check for samples consistency (this should be done before in the receiver / here only if the source is a file)
if (std::isnan((*d_Prompt).real()) == true or std::isnan((*d_Prompt).imag()) == true ) // or std::isinf(in[i].real())==true or std::isinf(in[i].imag())==true)
{
diff --git a/src/algorithms/tracking/libs/CMakeLists.txt b/src/algorithms/tracking/libs/CMakeLists.txt
index bb1a36c..175646a 100644
--- a/src/algorithms/tracking/libs/CMakeLists.txt
+++ b/src/algorithms/tracking/libs/CMakeLists.txt
@@ -21,9 +21,9 @@ if(ENABLE_CUDA)
FIND_PACKAGE(CUDA REQUIRED)
# Append current NVCC flags by something, eg comput capability
- # set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --gpu-architecture sm_30 --default-stream-per-thread)
+ # set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --gpu-architecture sm_30)
- list(APPEND CUDA_NVCC_FLAGS "-gencode arch=compute_30,code=sm_30; -std=c++11;-O3; -use_fast_math")
+ list(APPEND CUDA_NVCC_FLAGS "-gencode arch=compute_30,code=sm_30; -std=c++11;-O3; -use_fast_math -default-stream per-thread")
SET(CUDA_PROPAGATE_HOST_FLAGS OFF)
CUDA_INCLUDE_DIRECTORIES(
@@ -33,8 +33,6 @@ if(ENABLE_CUDA)
SET(LIB_TYPE STATIC) #set the lib type
CUDA_ADD_LIBRARY(CUDA_CORRELATOR_LIB ${LIB_TYPE} cuda_multicorrelator.h cuda_multicorrelator.cu)
-
- set(OPT_LIBRARIES ${OPT_LIBRARIES} ${CUDA_CORRELATOR_LIB})
endif(ENABLE_CUDA)
@@ -71,4 +69,4 @@ endif(SSE3_AVAILABLE)
file(GLOB TRACKING_LIB_HEADERS "*.h")
add_library(tracking_lib ${TRACKING_LIB_SOURCES} ${TRACKING_LIB_HEADERS})
source_group(Headers FILES ${TRACKING_LIB_HEADERS})
-target_link_libraries(tracking_lib ${VOLK_LIBRARIES} ${GNURADIO_RUNTIME_LIBRARIES} ${OPT_LIBRARIES})
\ No newline at end of file
+target_link_libraries(tracking_lib CUDA_CORRELATOR_LIB ${VOLK_LIBRARIES} ${GNURADIO_RUNTIME_LIBRARIES})
\ No newline at end of file
diff --git a/src/algorithms/tracking/libs/cuda_multicorrelator.cu b/src/algorithms/tracking/libs/cuda_multicorrelator.cu
index 1909e13..3f027cb 100644
--- a/src/algorithms/tracking/libs/cuda_multicorrelator.cu
+++ b/src/algorithms/tracking/libs/cuda_multicorrelator.cu
@@ -261,7 +261,7 @@ CUDA_32fc_x2_add_32fc( GPU_Complex *A, GPU_Complex *B, GPU_Complex *C, int
}
-bool cuda_multicorrelator::init_cuda(const int argc, const char **argv, int signal_length_samples, int *shifts_samples, int n_correlators)
+bool cuda_multicorrelator::init_cuda(const int argc, const char **argv, int signal_length_samples, int local_codes_length_samples, int n_correlators)
{
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
findCudaDevice(argc, (const char **)argv);
@@ -298,7 +298,8 @@ bool cuda_multicorrelator::init_cuda(const int argc, const char **argv, int sign
// new version: only one vector with extra samples to shift the local code for the correlator set
// Required: The last correlator tap in d_shifts_samples has the largest sample shift
- checkCudaErrors(cudaMalloc((void **)&d_local_codes_in, size+sizeof(GPU_Complex)*shifts_samples[n_correlators-1]));
+ size_t size_local_code_bytes = local_codes_length_samples * sizeof(GPU_Complex);
+ checkCudaErrors(cudaMalloc((void **)&d_local_codes_in, size_local_code_bytes));
checkCudaErrors(cudaMalloc((void **)&d_shifts_samples, size+sizeof(int)*n_correlators));
//scalars
@@ -308,6 +309,8 @@ bool cuda_multicorrelator::init_cuda(const int argc, const char **argv, int sign
threadsPerBlock = 256;
blocksPerGrid =(int)(signal_length_samples+threadsPerBlock-1)/threadsPerBlock;
+ cudaStreamCreate (&stream1) ;
+ cudaStreamCreate (&stream2) ;
return true;
}
@@ -323,14 +326,10 @@ bool cuda_multicorrelator::Carrier_wipeoff_multicorrelator_cuda(
int n_correlators)
{
- cudaStream_t stream1;
- cudaStream_t stream2;
- cudaStreamCreate ( &stream1) ;
- cudaStreamCreate ( &stream2) ;
-
size_t memSize = signal_length_samples * sizeof(std::complex<float>);
// input signal CPU -> GPU copy memory
+
checkCudaErrors(cudaMemcpyAsync(d_sig_in, sig_in, memSize,
cudaMemcpyHostToDevice, stream1));
@@ -362,7 +361,9 @@ bool cuda_multicorrelator::Carrier_wipeoff_multicorrelator_cuda(
//printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid, threadsPerBlock);
//wait for Doppler wipeoff end...
- checkCudaErrors(cudaDeviceSynchronize());
+ checkCudaErrors(cudaStreamSynchronize(stream1));
+ checkCudaErrors(cudaStreamSynchronize(stream2));
+ //checkCudaErrors(cudaDeviceSynchronize());
//old
// scalarProdGPUCPXxN<<<blocksPerGrid, threadsPerBlock,0 ,stream2>>>(
@@ -385,16 +386,13 @@ bool cuda_multicorrelator::Carrier_wipeoff_multicorrelator_cuda(
);
checkCudaErrors(cudaGetLastError());
//wait for correlators end...
- checkCudaErrors(cudaDeviceSynchronize());
+ checkCudaErrors(cudaStreamSynchronize(stream2));
// Copy the device result vector in device memory to the host result vector
// in host memory.
//scalar products (correlators outputs)
- checkCudaErrors(cudaMemcpyAsync(corr_out, d_corr_out, sizeof(std::complex<float>)*n_correlators,
- cudaMemcpyDeviceToHost, 0));
-
- cudaStreamDestroy(stream1) ;
- cudaStreamDestroy(stream2) ;
+ checkCudaErrors(cudaMemcpy(corr_out, d_corr_out, sizeof(std::complex<float>)*n_correlators,
+ cudaMemcpyDeviceToHost));
return true;
}
@@ -406,13 +404,16 @@ bool cuda_multicorrelator::free_cuda()
cudaFree(d_local_codes_in);
cudaFree(d_corr_out);
+ cudaStreamDestroy(stream1) ;
+ cudaStreamDestroy(stream2) ;
+
// Reset the device and exit
// cudaDeviceReset causes the driver to clean up all state. While
// not mandatory in normal operation, it is good practice. It is also
// needed to ensure correct operation when the application is being
// profiled. Calling cudaDeviceReset causes all profile data to be
// flushed before the application exits
- checkCudaErrors(cudaDeviceReset());
+ //checkCudaErrors(cudaDeviceReset());
return true;
}
diff --git a/src/algorithms/tracking/libs/cuda_multicorrelator.h b/src/algorithms/tracking/libs/cuda_multicorrelator.h
index 8819edf..e29cba5 100644
--- a/src/algorithms/tracking/libs/cuda_multicorrelator.h
+++ b/src/algorithms/tracking/libs/cuda_multicorrelator.h
@@ -46,6 +46,10 @@
#include <complex>
+#include <cuda.h>
+// CUDA runtime
+#include <cuda_runtime.h>
+
// GPU new internal data types for complex numbers
struct GPU_Complex {
@@ -109,7 +113,7 @@ struct GPU_Complex_Short {
class cuda_multicorrelator
{
public:
- bool init_cuda(const int argc, const char **argv, int signal_length_samples, int *shifts_samples, int n_correlators);
+ bool init_cuda(const int argc, const char **argv, int signal_length_samples, int local_codes_length_samples, int n_correlators);
bool free_cuda();
bool Carrier_wipeoff_multicorrelator_cuda(
@@ -132,6 +136,9 @@ private:
int threadsPerBlock;
int blocksPerGrid;
+ cudaStream_t stream1;
+ cudaStream_t stream2;
+
};
diff --git a/src/core/receiver/CMakeLists.txt b/src/core/receiver/CMakeLists.txt
index 677efd7..cf4d183 100644
--- a/src/core/receiver/CMakeLists.txt
+++ b/src/core/receiver/CMakeLists.txt
@@ -16,6 +16,10 @@
# along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
#
+if(ENABLE_CUDA)
+ FIND_PACKAGE(CUDA REQUIRED)
+endif(ENABLE_CUDA)
+
set(GNSS_RECEIVER_SOURCES
control_thread.cc
control_message_factory.cc
@@ -64,8 +68,9 @@ include_directories(
${GFlags_INCLUDE_DIRS}
${Boost_INCLUDE_DIRS}
${GNURADIO_RUNTIME_INCLUDE_DIRS}
+ ${CUDA_INCLUDE_DIRS}
)
-
+
if(Boost_VERSION LESS 105000)
add_definitions(-DOLD_BOOST=1)
endif(Boost_VERSION LESS 105000)
diff --git a/src/core/receiver/gnss_block_factory.cc b/src/core/receiver/gnss_block_factory.cc
index 770c1a6..35e4c83 100644
--- a/src/core/receiver/gnss_block_factory.cc
+++ b/src/core/receiver/gnss_block_factory.cc
@@ -77,6 +77,7 @@
#include "galileo_e1_pcps_quicksync_ambiguous_acquisition.h"
#include "galileo_e5a_noncoherent_iq_acquisition_caf.h"
#include "gps_l1_ca_dll_pll_tracking.h"
+#include "gps_l1_ca_dll_pll_tracking_gpu.h"
#include "gps_l1_ca_dll_pll_optim_tracking.h"
#include "gps_l1_ca_dll_fll_pll_tracking.h"
#include "gps_l1_ca_tcp_connector_tracking.h"
@@ -1609,6 +1610,12 @@ std::unique_ptr<TrackingInterface> GNSSBlockFactory::GetTrkBlock(
out_streams, queue));
block = std::move(block_);
}
+ else if (implementation.compare("GPS_L1_CA_DLL_PLL_Tracking_GPU") == 0)
+ {
+ std::unique_ptr<TrackingInterface> block_(new GpsL1CaDllPllTrackingGPU(configuration.get(), role, in_streams,
+ out_streams, queue));
+ block = std::move(block_);
+ }
else
{
// Log fatal. This causes execution to stop.
diff --git a/src/main/CMakeLists.txt b/src/main/CMakeLists.txt
index af8e7bc..0275810 100644
--- a/src/main/CMakeLists.txt
+++ b/src/main/CMakeLists.txt
@@ -16,6 +16,7 @@
# along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
#
+
set(GNSS_SDR_OPTIONAL_LIBS "")
set(GNSS_SDR_OPTIONAL_HEADERS "")
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/pkg-hamradio/gnss-sdr.git
More information about the pkg-hamradio-commits
mailing list