// // Copyright 2018 Ettus Research, a National Instruments Company // // SPDX-License-Identifier: GPL-3.0-or-later // #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include namespace po = boost::program_options; struct traffic_counter_values { uint64_t clock_cycles; uint64_t xbar_to_shell_xfer_count; uint64_t xbar_to_shell_pkt_count; uint64_t shell_to_xbar_xfer_count; uint64_t shell_to_xbar_pkt_count; uint64_t shell_to_ce_xfer_count; uint64_t shell_to_ce_pkt_count; uint64_t ce_to_shell_xfer_count; uint64_t ce_to_shell_pkt_count; }; struct host_measurement_values { double seconds; uint64_t num_samples; uint64_t num_packets; uint64_t spp; }; struct test_results { std::vector traffic_counter; host_measurement_values host; }; struct noc_block_endpoint { std::string block_id; size_t port; }; void enable_traffic_counters( uhd::property_tree::sptr tree, uhd::fs_path noc_block_root ) { tree->access(noc_block_root/"traffic_counter/enable").set(true); } void disable_traffic_counters( uhd::property_tree::sptr tree, uhd::fs_path noc_block_root ) { tree->access(noc_block_root/"traffic_counter/enable").set(false); } traffic_counter_values read_traffic_counters( uhd::property_tree::sptr tree, uhd::fs_path noc_block_root ) { uhd::fs_path root = noc_block_root/"traffic_counter"; traffic_counter_values vals; vals.clock_cycles = tree->access(root/"bus_clock_ticks").get(); vals.xbar_to_shell_pkt_count = tree->access(root/"xbar_to_shell_pkt_count").get(); vals.xbar_to_shell_xfer_count = tree->access(root/"xbar_to_shell_xfer_count").get(); vals.shell_to_xbar_pkt_count = tree->access(root/"shell_to_xbar_pkt_count").get(); vals.shell_to_xbar_xfer_count = tree->access(root/"shell_to_xbar_xfer_count").get(); vals.shell_to_ce_pkt_count = tree->access(root/"shell_to_ce_pkt_count").get(); vals.shell_to_ce_xfer_count = tree->access(root/"shell_to_ce_xfer_count").get(); vals.ce_to_shell_pkt_count = tree->access(root/"ce_to_shell_pkt_count").get(); vals.ce_to_shell_xfer_count = tree->access(root/"ce_to_shell_xfer_count").get(); return vals; } void print_traffic_counters( const traffic_counter_values& vals ) { std::cout << "Clock cycles: " << vals.clock_cycles << std::endl; std::cout << "Xbar to shell pkt count: " << vals.xbar_to_shell_pkt_count << std::endl; std::cout << "Xbar to shell xfer count: " << vals.xbar_to_shell_xfer_count << std::endl; std::cout << "Shell to xbar pkt count: " << vals.shell_to_xbar_pkt_count << std::endl; std::cout << "Shell to xbar xfer count: " << vals.shell_to_xbar_xfer_count << std::endl; std::cout << "Shell to CE pkt count: " << vals.shell_to_ce_pkt_count << std::endl; std::cout << "Shell to CE xfer count: " << vals.shell_to_ce_xfer_count << std::endl; std::cout << "CE to shell pkt count: " << vals.ce_to_shell_pkt_count << std::endl; std::cout << "CE to shell xfer count: " << vals.ce_to_shell_xfer_count << std::endl; } void print_rx_statistics( const traffic_counter_values& vals, const double bus_clk_freq ) { const double bus_time_elapsed = vals.clock_cycles / bus_clk_freq; const uint64_t num_ce_packets_read = vals.ce_to_shell_pkt_count; const uint64_t num_ce_samples_read = (vals.ce_to_shell_xfer_count - num_ce_packets_read)*2; const uint64_t num_non_data_packets_read = vals.shell_to_xbar_pkt_count - num_ce_packets_read; const double rx_data_packet_ratio = (double)num_ce_packets_read/num_non_data_packets_read; const double calculated_throughput = num_ce_samples_read/bus_time_elapsed; std::cout << "Time elapsed: " << bus_time_elapsed << " s" << std::endl; std::cout << "Samples read: " << num_ce_samples_read << std::endl; std::cout << "Data packets read: " << num_ce_packets_read << std::endl; std::cout << "RX data packet ratio: " << rx_data_packet_ratio << " data to non-data packets" << std::endl; std::cout << "Calculated throughput: " << calculated_throughput/1e6 << " Msps" << std::endl; } void print_tx_statistics( const traffic_counter_values& vals, const double bus_clk_freq ) { const double bus_time_elapsed = vals.clock_cycles / bus_clk_freq; const uint64_t num_ce_packets_written = vals.shell_to_ce_pkt_count; const uint64_t num_ce_samples_written = (vals.shell_to_ce_xfer_count - num_ce_packets_written)*2; const uint64_t num_non_data_packets_written = vals.xbar_to_shell_pkt_count - num_ce_packets_written; const double tx_data_packet_ratio = (double)num_ce_packets_written/num_non_data_packets_written; const double calculated_throughput = num_ce_samples_written/bus_time_elapsed; std::cout << "Time elapsed: " << bus_time_elapsed << " s" << std::endl; std::cout << "Samples written: " << num_ce_samples_written << std::endl; std::cout << "Data packets written: " << num_ce_packets_written << std::endl; std::cout << "TX data packet ratio: " << tx_data_packet_ratio << " data to non-data packets" << std::endl; std::cout << "Calculated throughput: " << calculated_throughput/1e6 << " Msps" << std::endl; } void print_utilization_statistics( const traffic_counter_values& vals ) { const double rx_data_cycles = vals.ce_to_shell_xfer_count - vals.ce_to_shell_pkt_count; const double rx_idle_cycles = vals.clock_cycles - vals.shell_to_xbar_xfer_count; const double rx_data_header_cycles = vals.ce_to_shell_pkt_count; const double rx_other_cycles = vals.shell_to_xbar_xfer_count - vals.ce_to_shell_xfer_count; const double rx_data_util = rx_data_cycles / vals.clock_cycles*100; const double rx_idle_util = rx_idle_cycles / vals.clock_cycles*100; const double rx_data_header_util = rx_data_header_cycles / vals.clock_cycles * 100; const double rx_other_util = rx_other_cycles / vals.clock_cycles * 100; std::cout << "RX utilization:" << std::endl; std::cout << " data: " << rx_data_util << " %" << std::endl; std::cout << " idle: " << rx_idle_util << " %" << std::endl; std::cout << " data header: " << rx_data_header_util << " %" << std::endl; std::cout << " other: " << rx_other_util << " % (flow control, register I/O)" << std::endl; std::cout << std::endl; const double tx_data_cycles = vals.shell_to_ce_xfer_count - vals.shell_to_ce_pkt_count; const double tx_idle_cycles = vals.clock_cycles - vals.xbar_to_shell_xfer_count; const double tx_data_header_cycles = vals.shell_to_ce_pkt_count; const double tx_other_cycles = vals.xbar_to_shell_xfer_count - vals.shell_to_ce_xfer_count; const double tx_data_util = tx_data_cycles / vals.clock_cycles*100; const double tx_idle_util = tx_idle_cycles / vals.clock_cycles*100; const double tx_data_header_util = tx_data_header_cycles / vals.clock_cycles * 100; const double tx_other_util = tx_other_cycles / vals.clock_cycles * 100; std::cout << "TX utilization:" << std::endl; std::cout << " data: " << tx_data_util << " %" << std::endl; std::cout << " idle: " << tx_idle_util << " %" << std::endl; std::cout << " data header: " << tx_data_header_util << " %" << std::endl; std::cout << " other: " << tx_other_util << " % (flow control, register I/O)" << std::endl; } void print_rx_results( const test_results& results, double bus_clk_freq ) { std::cout << "------------------------------------------------------------------" << std::endl; std::cout << "------------------- Benchmarking rx stream -----------------------" << std::endl; std::cout << "------------------------------------------------------------------" << std::endl; std::cout << "RX samples per packet: " << results.host.spp << std::endl; std::cout << std::endl; for (const auto& tc : results.traffic_counter) { std::cout << "------------------ Traffic counter values ------------------------" << std::endl; print_traffic_counters(tc); std::cout << std::endl; std::cout << "------------ Values calculated from traffic counters -------------" << std::endl; print_rx_statistics(tc, bus_clk_freq); std::cout << std::endl; print_utilization_statistics(tc); std::cout << std::endl; } std::cout << "--------------------- Host measurements --------------------------" << std::endl; std::cout << "Time elapsed: " << results.host.seconds << " s" << std::endl; std::cout << "Samples read: " << results.host.num_samples << std::endl; std::cout << "Data packets read: " << results.host.num_packets << std::endl; std::cout << "Calculated throughput: " << results.host.num_samples / results.host.seconds / 1e6 << " Msps" << std::endl; } void print_tx_results( const test_results& results, const double bus_clk_freq ) { std::cout << "------------------------------------------------------------------" << std::endl; std::cout << "------------------- Benchmarking tx stream -----------------------" << std::endl; std::cout << "------------------------------------------------------------------" << std::endl; std::cout << "TX samples per packet: " << results.host.spp << std::endl; std::cout << std::endl; for (const auto& tc : results.traffic_counter) { std::cout << "------------------ Traffic counter values ------------------------" << std::endl; print_traffic_counters(tc); std::cout << std::endl; std::cout << "------------ Values calculated from traffic counters -------------" << std::endl; print_tx_statistics(tc, bus_clk_freq); std::cout << std::endl; print_utilization_statistics(tc); std::cout << std::endl; } std::cout << "--------------------- Host measurements --------------------------" << std::endl; std::cout << "Time elapsed: " << results.host.seconds << " s" << std::endl; std::cout << "Samples written: " << results.host.num_samples << std::endl; std::cout << "Data packets written: " << results.host.num_packets << std::endl; std::cout << "Calculated throughput: " << results.host.num_samples / results.host.seconds / 1e6 << " Msps" << std::endl; } void configure_ddc( uhd::device3::sptr usrp, const std::string& ddcid, double ddc_decim ) { auto ddc_ctrl = usrp->get_block_ctrl(ddcid); ddc_ctrl->set_arg("input_rate", 1, 0); ddc_ctrl->set_arg("output_rate", 1/ddc_decim, 0); double actual_rate = ddc_ctrl->get_arg("output_rate", 0); std::cout << "Actual DDC decimation: " << 1/actual_rate << std::endl; } void configure_duc( uhd::device3::sptr usrp, const std::string& ducid, double duc_interp ) { auto duc_ctrl = usrp->get_block_ctrl(ducid); duc_ctrl->set_arg("output_rate", 1, 0); duc_ctrl->set_arg("input_rate", 1/duc_interp, 0); double actual_rate = duc_ctrl->get_arg("input_rate", 0); std::cout << "Actual DUC interpolation: " << 1/actual_rate << std::endl; } uhd::rx_streamer::sptr configure_rx_streamer( uhd::device3::sptr usrp, const std::string null_id, const std::string splitter_id, const std::vector>& noc_blocks, const size_t spp, const std::string& format ) { std::cout << "Configuring rx stream with" << std::endl; std::cout << " Null ID: " << null_id << std::endl; if (not splitter_id.empty()) { std::cout << " Splitter ID: " << splitter_id << std::endl; } for (size_t i = 0; i < noc_blocks.size(); i++) { if (noc_blocks[i].size() > 0) { std::cout << " Channel " << i << std::endl; for (const auto& b : noc_blocks[i]) { std::cout << " Block ID: " << b.block_id << ", port: " << b.port << std::endl; } } } auto rx_graph = usrp->create_graph("rx_graph"); uhd::stream_args_t stream_args(format, "sc16"); std::vector channels; for (size_t i = 0; i < noc_blocks.size(); i++) { channels.push_back(i); } stream_args.channels = channels; std::vector endpoints; if (noc_blocks.size() == 1) { // No splitter required endpoints = {{null_id, 0}}; } else { // Connect to splitter rx_graph->connect(null_id, splitter_id); for (size_t i = 0; i < noc_blocks.size(); i++) { endpoints.push_back({splitter_id, i}); } } for (size_t i = 0; i < noc_blocks.size(); i++) { std::string endpoint_id = endpoints[i].block_id; size_t endpoint_port = endpoints[i].port; for (size_t j = 0; j < noc_blocks[i].size(); j++) { const auto& noc_block = noc_blocks[i][j]; rx_graph->connect( endpoint_id, endpoint_port, noc_block.block_id, noc_block.port); endpoint_id = noc_block.block_id; endpoint_port = noc_block.port; } const std::string id_str = str(boost::format("block_id%d") % i); const std::string port_str = str(boost::format("block_port%d") % i); stream_args.args[id_str] = endpoint_id; stream_args.args[port_str] = str(boost::format("%d") % endpoint_port); } if (spp != 0) { stream_args.args["spp"] = std::to_string(spp); } uhd::rx_streamer::sptr rx_stream = usrp->get_rx_stream(stream_args); // Configure null source auto null_ctrl = usrp->get_block_ctrl(null_id); const size_t otw_bytes_per_item = uhd::convert::get_bytes_per_item(stream_args.otw_format); const size_t samps_per_packet = rx_stream->get_max_num_samps(); null_ctrl->set_arg("line_rate", 0); null_ctrl->set_arg("bpp", samps_per_packet*otw_bytes_per_item); return rx_stream; } test_results benchmark_rx_streamer( uhd::device3::sptr usrp, uhd::rx_streamer::sptr rx_stream, const std::string& nullid, const double duration, const std::string& format ) { auto null_src_ctrl = usrp->get_block_ctrl(nullid); // Allocate buffer const size_t cpu_bytes_per_item = uhd::convert::get_bytes_per_item(format); const size_t samps_per_packet = rx_stream->get_max_num_samps(); const size_t num_channels = rx_stream->get_num_channels(); std::vector> buffer(num_channels); std::vector buffers; for (size_t i = 0; i < num_channels; i++) { buffer[i].resize(samps_per_packet*cpu_bytes_per_item); buffers.push_back(&buffer[i].front()); } enable_traffic_counters( usrp->get_tree(), null_src_ctrl->get_block_id().get_tree_root()); // Stream some packets uhd::stream_cmd_t stream_cmd(uhd::stream_cmd_t::STREAM_MODE_START_CONTINUOUS); stream_cmd.stream_now = true; null_src_ctrl->issue_stream_cmd(stream_cmd); const std::chrono::duration requested_duration(duration); const auto start_time = std::chrono::steady_clock::now(); auto current_time = start_time; uint64_t num_rx_samps = 0; uint64_t num_rx_packets = 0; uhd::rx_metadata_t md; while (current_time - start_time < requested_duration) { const size_t packets_per_iteration = 1000; for (size_t i = 0; i < packets_per_iteration; i++){ num_rx_samps += rx_stream->recv(buffers, samps_per_packet, md, 1.0); if (md.error_code != uhd::rx_metadata_t::ERROR_CODE_NONE) { if (md.error_code == uhd::rx_metadata_t::ERROR_CODE_OVERFLOW) { continue; } else if (md.error_code != uhd::rx_metadata_t::ERROR_CODE_TIMEOUT) { std::cout << "[ERROR] Receive timeout, aborting." << std::endl; break; } else { std::cout << std::string("[ERROR] Receiver error: ") << md.strerror() << std::endl; break; } } } num_rx_packets += packets_per_iteration; current_time = std::chrono::steady_clock::now(); } disable_traffic_counters( usrp->get_tree(), null_src_ctrl->get_block_id().get_tree_root()); null_src_ctrl->issue_stream_cmd(uhd::stream_cmd_t::STREAM_MODE_STOP_CONTINUOUS); test_results results; results.traffic_counter.push_back( read_traffic_counters( usrp->get_tree(), null_src_ctrl->get_block_id().get_tree_root())); const std::chrono::duration elapsed_time(current_time-start_time); results.host.seconds = elapsed_time.count(); results.host.num_samples = num_rx_samps; results.host.num_packets = num_rx_packets; results.host.spp = samps_per_packet; return results; } uhd::tx_streamer::sptr configure_tx_streamer( uhd::device3::sptr usrp, const std::vector> noc_blocks, const size_t spp, const std::string& format ) { std::cout << "Configuring tx stream with" << std::endl; for (size_t i = 0; i < noc_blocks.size(); i++) { std::cout << " Channel " << i << std::endl; for (const auto& b : noc_blocks[i]) { std::cout << " Block ID: " << b.block_id << ", port: " << b.port << std::endl; } } // Configure rfnoc auto tx_graph = usrp->create_graph("tx_graph"); uhd::stream_args_t stream_args(format, "sc16"); std::vector channels; for (size_t i = 0; i < noc_blocks.size(); i++) { channels.push_back(i); } stream_args.channels = channels; for (size_t i = 0; i < noc_blocks.size(); i++) { std::string endpoint_id; size_t endpoint_port; for (size_t j = 0; j < noc_blocks[i].size(); j++) { const auto& noc_block = noc_blocks[i][j]; if (j != 0) { tx_graph->connect( noc_block.block_id, noc_block.port, endpoint_id, endpoint_port); } endpoint_id = noc_block.block_id; endpoint_port = noc_block.port; } const std::string id_str = str(boost::format("block_id%d") % i); const std::string port_str = str(boost::format("block_port%d") % i); stream_args.args[id_str] = endpoint_id; stream_args.args[port_str] = str(boost::format("%d") % endpoint_port); } // Configure streamer if (spp != 0) { stream_args.args["spp"] = std::to_string(spp); } uhd::tx_streamer::sptr tx_stream = usrp->get_tx_stream(stream_args); return tx_stream; } test_results benchmark_tx_streamer( uhd::device3::sptr usrp, uhd::tx_streamer::sptr tx_stream, const std::vector& null_ids, const double duration, const std::string& format ) { std::vector> null_ctrls; for (const auto& id : null_ids) { null_ctrls.push_back( usrp->get_block_ctrl(id)); } // Allocate buffer const size_t cpu_bytes_per_item = uhd::convert::get_bytes_per_item(format); const size_t samps_per_packet = tx_stream->get_max_num_samps(); const size_t num_channels = tx_stream->get_num_channels(); std::vector> buffer(num_channels); std::vector buffers; for (size_t i = 0; i < num_channels; i++) { buffer[i].resize(samps_per_packet*cpu_bytes_per_item); buffers.push_back(&buffer[i].front()); } for (auto& null_ctrl : null_ctrls) { enable_traffic_counters( usrp->get_tree(), null_ctrl->get_block_id().get_tree_root()); } // Stream some packets uint64_t num_tx_samps = 0; uint64_t num_tx_packets = 0; uhd::tx_metadata_t md; const std::chrono::duration requested_duration(duration); const auto start_time = std::chrono::steady_clock::now(); auto current_time = start_time; while (current_time - start_time < requested_duration) { const size_t packets_per_iteration = 1000; for (size_t i = 0; i < packets_per_iteration; i++){ num_tx_samps += tx_stream->send(buffers, samps_per_packet, md); } num_tx_packets += packets_per_iteration; current_time = std::chrono::steady_clock::now(); } for (auto& null_ctrl : null_ctrls) { disable_traffic_counters( usrp->get_tree(), null_ctrl->get_block_id().get_tree_root()); } // Stop md.end_of_burst = true; tx_stream->send(buffers, 0, md); test_results results; for (auto& null_ctrl : null_ctrls) { results.traffic_counter.push_back( read_traffic_counters( usrp->get_tree(), null_ctrl->get_block_id().get_tree_root())); } const std::chrono::duration elapsed_time(current_time-start_time); results.host.seconds = elapsed_time.count(); results.host.num_samples = num_tx_samps; results.host.num_packets = num_tx_packets; results.host.spp = samps_per_packet; return results; } std::vector parse_csv( const std::string& list ) { std::vector result; std::istringstream input(list); std::string str; while (std::getline(input, str, ',')) { boost::algorithm::trim_all(str); if (not str.empty()) { result.push_back(str); } } return result; } std::deque create_noc_block_queue( const size_t num_blocks, const std::string& user_override_id_list, const std::string& prefix, const size_t num_ports ) { const std::vector overrides = parse_csv(user_override_id_list); std::deque result; for (size_t i = 0; i < num_blocks; i++) { if (i < overrides.size()) { result.push_back({overrides[i], (i % num_ports)}); } else { const std::string format_str = prefix + "_%d"; noc_block_endpoint block = { str(boost::format(format_str) % (i/num_ports)), i % num_ports}; result.push_back(block); } } return result; } int UHD_SAFE_MAIN(int argc, char *argv[]) { // Variables to be set by po bool dma_fifo, ddc, duc, tx_loopback_fifo, rx_loopback_fifo; std::string args, format; std::string null_ids, fifo_ids, ddc_ids, duc_ids, split_stream_ids; double duration; double ddc_decim, duc_interp, bus_clk_freq; size_t spp; size_t num_tx_streamers, num_rx_streamers, num_tx_channels, num_rx_channels; // Setup the program options po::options_description desc("Allowed options"); // clang-format off desc.add_options() ("help", "help message") ("args", po::value(&args)->default_value(""), "single uhd device address args") ("num_tx_streamers", po::value(&num_tx_streamers)->default_value(0), "number of tx streamers to benchmark") ("num_rx_streamers", po::value(&num_rx_streamers)->default_value(0), "number of rx streamers to benchmark") ("num_tx_channels", po::value(&num_tx_channels)->default_value(1), "number of tx channels per streamer") ("num_rx_channels", po::value(&num_rx_channels)->default_value(1), "number of rx channels per streamer") ("duration", po::value(&duration)->default_value(10.0), "duration for the test in seconds") ("spp", po::value(&spp)->default_value(0), "samples per packet (on FPGA and wire)") ("format", po::value(&format)->default_value("sc16"), "host sample type: sc16, fc32, or fc64") ("bus_clk_freq", po::value(&bus_clk_freq)->default_value(187.5e6), "bus clock frequency for throughput calculation (default: 187.5e6)") ("dma_fifo", po::bool_switch(&dma_fifo)->default_value(false), "whether to insert a DMA FIFO in the streaming path") ("tx_loopback_fifo", po::bool_switch(&tx_loopback_fifo)->default_value(false), "whether to insert a loopback FIFO in the tx streaming path") ("rx_loopback_fifo", po::bool_switch(&rx_loopback_fifo)->default_value(false), "whether to insert a loopback FIFO in the rx streaming path") ("ddc", po::bool_switch(&ddc)->default_value(false), "whether to insert a DDC in the rx streaming path") ("duc", po::bool_switch(&duc)->default_value(false), "whether to insert a DUC in the tx streaming path") ("ddc_decim", po::value(&ddc_decim)->default_value(1), "DDC decimation, between 1 and max decimation (default: 1, no decimation)") ("duc_interp", po::value(&duc_interp)->default_value(1), "DUC interpolation, between 1 and max interpolation (default: 1, no interpolation)") ("null_ids", po::value(&null_ids)->default_value(""), "optional: list of block IDs for the null sources") ("fifo_ids", po::value(&fifo_ids)->default_value(""), "optional: list of block IDs for the loopback FIFOs") ("ddc_ids", po::value(&ddc_ids)->default_value(""), "optional: list of block IDs for the DDCs") ("duc_ids", po::value(&duc_ids)->default_value(""), "optional: list of block IDs for the DUCs") ("split_stream_ids", po::value(&split_stream_ids)->default_value(""), "optional: list of block IDs for rx data splitters") ; // clang-format on po::variables_map vm; po::store(po::parse_command_line(argc, argv, desc), vm); po::notify(vm); // Print the help message const size_t num_streamers = num_rx_streamers + num_tx_streamers; if (vm.count("help") or (num_streamers == 0)) { std::cout << boost::format("UHD - Benchmark Streamer") << std::endl; std::cout << " Benchmark streamer connects a null sink/source to a streamer and\n" " measures maximum throughput. You can benchmark the operation of\n" " multiple streamers concurrently. Each streamer executes in a\n" " separate thread. The FPGA image on the device must contain a\n" " null source for each channel in the test.\n" " Benchmarks of common use-cases:\n" " Specify --num_tx_streamers=1 to test tx streamer.\n" " Specify --num_rx_streamers=1 to test rx streamer.\n" " Specify --num_tx_streamers=1 --num_tx_channels-2 to test tx\n" " streamer with two channels.\n" " Specify --num_rx_streamers=1 --num_rx_channels=2 to test rx\n" " rx streamer with two channels. This requires a split_stream\n" " RFNOC block.\n" " Specify --num_rx_streamers=1 --num_tx_streams=1 to test full\n" " duplex data transfer.\n" " Specify --num_rx_streamers=2 --num_rx_streams=2 to test full\n" " duplex data tranfser with two streamers in each direction.\n" " Benchmarks streamer allows DMA FIFOs, loopback FIFOs, DDCs, and\n" " DUCs to be added to the data path. Enable these by setting the\n" " corresponding Boolean option to true. The order of the blocks\n" " is fixed. If present, the DMA FIFO is connected to the host bus\n" " interface, followed by the loopback FIFOs, and then DUC on a tx\n" " stream or a DDC on an rx stream.\n" " Note: for full duplex tests, if a DMA FIFO is specified, it is\n" " inserted in the tx data path only.\n" " Testing multiple rx channels in a single streamer requires a\n" " split stream RFNOC block with the number of outputs equal to the\n" " number of channels. Each streamer connects to a single null\n" " source through the split stream block.\n" " In order to allow testing of blocks with different compilation\n" " parameters, such as the block FIFO size, this example provides\n" " options to override RFNOC block IDs. Block IDs can be specified\n" " as a comma-delimited list for each type of block. If the block\n" " type is used in both tx and rx streams, block IDs are assigned\n" " to tx streams first, followed by rx streams. For example, a test\n" " with two tx and two rx streams will assign the first two IDs in\n" " the null_ids list to the tx streams and the next two IDs to the\n" " rx streams.\n" << std::endl << desc << std::endl; return EXIT_SUCCESS; } uhd::set_thread_priority_safe(); std::cout << boost::format("Creating the usrp device with: %s...") % args << std::endl; uhd::device3::sptr usrp = uhd::device3::make(args); // For each block type, calculate the number of blocks needed by the test // and create block IDs, accounting for user overrides in program options. // Note that for null sources, rx only uses one NULL block per streamer // rather than one per channel, since the test uses split_stream blocks to // ensure packets on the same streamer have matching timestamps. Also, for // DMA FIFOs, if the test contains both tx and rx channels, we only insert // the FIFOs in the tx data path since that is the primary use-case. const size_t total_tx_channels = num_tx_streamers * num_tx_channels; const size_t total_rx_channels = num_rx_streamers * num_rx_channels; const size_t num_null_blocks = total_tx_channels + num_rx_streamers; const size_t num_duc_blocks = duc ? total_tx_channels : 0; const size_t num_ddc_blocks = ddc ? total_rx_channels : 0; const size_t num_tx_fifo_blocks = tx_loopback_fifo ? total_tx_channels : 0; const size_t num_rx_fifo_blocks = rx_loopback_fifo ? total_rx_channels : 0; const size_t num_fifo_blocks = num_tx_fifo_blocks + num_rx_fifo_blocks; const size_t num_splitter_blocks = num_rx_channels > 1 ? num_rx_streamers : 0; size_t num_dma_fifo_blocks = 0; bool tx_dma_fifo = false; bool rx_dma_fifo = false; if (dma_fifo) { if (total_tx_channels == 0) { num_dma_fifo_blocks = total_rx_channels; rx_dma_fifo = true; } else { num_dma_fifo_blocks = total_tx_channels; tx_dma_fifo = true; } } // Create block IDs std::deque null_blocks = create_noc_block_queue( num_null_blocks, null_ids, "0/NullSrcSink", 1); std::deque duc_blocks = create_noc_block_queue( num_duc_blocks, duc_ids, "0/DUC", 1); std::deque ddc_blocks = create_noc_block_queue( num_ddc_blocks, ddc_ids, "0/DDC", 1); std::deque fifo_blocks = create_noc_block_queue( num_fifo_blocks, fifo_ids, "0/FIFO", 1); std::deque dma_fifo_blocks = create_noc_block_queue( num_dma_fifo_blocks, "", "0/DmaFIFO", 2); std::deque splitter_blocks = create_noc_block_queue( num_splitter_blocks, split_stream_ids, "0/SplitStream", 1); // Configure all streamers usrp->clear(); std::vector tx_streamers; std::vector> tx_null_ids; for (size_t i = 0; i < num_tx_streamers; i++) { std::vector> blocks(num_tx_channels); std::vector null_ids; for (size_t ch = 0; ch < num_tx_channels; ch++) { // Store the null ids to read traffic counters later null_ids.push_back(null_blocks.front().block_id); // Add block IDs to create the graph for each channel blocks[ch].push_back(null_blocks.front()); null_blocks.pop_front(); if (duc) { configure_duc(usrp, duc_blocks.front().block_id, duc_interp); blocks[ch].push_back(duc_blocks.front()); duc_blocks.pop_front(); } if (tx_loopback_fifo) { blocks[ch].push_back(fifo_blocks.front()); fifo_blocks.pop_front(); } if (tx_dma_fifo) { blocks[ch].push_back(dma_fifo_blocks.front()); dma_fifo_blocks.pop_front(); } }; tx_streamers.push_back( configure_tx_streamer(usrp, blocks, spp, format)); tx_null_ids.push_back(null_ids); } std::vector rx_streamers; std::vector rx_null_ids; for (size_t i = 0; i < num_rx_streamers; i++) { std::vector> blocks(num_rx_channels); for (size_t ch = 0; ch < num_rx_channels; ch++) { if (ddc) { configure_ddc(usrp, ddc_blocks.front().block_id, ddc_decim); blocks[ch].push_back(ddc_blocks.front()); ddc_blocks.pop_front(); } if (rx_loopback_fifo) { blocks[ch].push_back(fifo_blocks.front()); fifo_blocks.pop_front(); } if (rx_dma_fifo) { blocks[ch].push_back(dma_fifo_blocks.front()); dma_fifo_blocks.pop_front(); } }; std::string splitter_id; if (num_rx_channels > 1) { splitter_id = splitter_blocks.front().block_id; splitter_blocks.pop_front(); } rx_streamers.push_back( configure_rx_streamer( usrp, null_blocks.front().block_id, splitter_id, blocks, spp, format)); // Store the null ids to read traffic counters later rx_null_ids.push_back(null_blocks.front().block_id); null_blocks.pop_front(); } // Start threads std::vector threads; std::vector tx_results(num_tx_streamers); for (size_t i = 0; i < num_tx_streamers; i++) { test_results& results = tx_results[i]; uhd::tx_streamer::sptr streamer = tx_streamers[i]; std::vector null_ids = tx_null_ids[i]; threads.push_back( std::thread( [&results, usrp, streamer, null_ids, duration, format]() { results = benchmark_tx_streamer( usrp, streamer, null_ids, duration, format); }) ); } std::vector rx_results(num_rx_streamers); for (size_t i = 0; i < num_rx_streamers; i++) { test_results& results = rx_results[i]; uhd::rx_streamer::sptr streamer = rx_streamers[i]; std::string null_id = rx_null_ids[i]; threads.push_back( std::thread( [&results, usrp, streamer, null_id, duration, format]() { results = benchmark_rx_streamer( usrp, streamer, null_id, duration, format); }) ); } // Join threads for (std::thread& t : threads) { t.join(); } // Print results for (const test_results& result: tx_results) { print_tx_results(result, bus_clk_freq); } for (const test_results& result: rx_results) { print_rx_results(result, bus_clk_freq); } return EXIT_SUCCESS; }