From a4f8ad59b2abc6d4b1a6c65aa2652083486e3065 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Thu, 30 Sep 2010 11:05:33 -0700 Subject: usrp: added subdev enabled property the dboard manager will disable all subdevs at startup and shutdown setting the subdev spec will enable only the subdevs in use all dboards are currently implemented as always enabled nothing tested --- host/lib/usrp/dboard/db_basic_and_lf.cpp | 14 ++++++++++++++ host/lib/usrp/dboard/db_dbsrx.cpp | 7 +++++++ host/lib/usrp/dboard/db_rfx.cpp | 14 ++++++++++++++ host/lib/usrp/dboard/db_unknown.cpp | 14 ++++++++++++++ host/lib/usrp/dboard/db_wbx.cpp | 14 ++++++++++++++ host/lib/usrp/dboard/db_xcvr2450.cpp | 14 ++++++++++++++ host/lib/usrp/dboard_manager.cpp | 12 +++++++++++- host/lib/usrp/misc_utils.cpp | 17 +++++++++++++++++ host/lib/usrp/subdev_spec.cpp | 4 ++++ 9 files changed, 109 insertions(+), 1 deletion(-) (limited to 'host/lib') diff --git a/host/lib/usrp/dboard/db_basic_and_lf.cpp b/host/lib/usrp/dboard/db_basic_and_lf.cpp index 4b0d8bf27..41f6f8002 100644 --- a/host/lib/usrp/dboard/db_basic_and_lf.cpp +++ b/host/lib/usrp/dboard/db_basic_and_lf.cpp @@ -149,6 +149,10 @@ void basic_rx::rx_get(const wax::obj &key_, wax::obj &val){ val = sd_name_to_conn[get_subdev_name()]; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -178,6 +182,9 @@ void basic_rx::rx_set(const wax::obj &key_, const wax::obj &val){ case SUBDEV_PROP_FREQ: return; // it wont do you much good, but you can set it + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } @@ -241,6 +248,10 @@ void basic_tx::tx_get(const wax::obj &key_, wax::obj &val){ val = sd_name_to_conn[get_subdev_name()]; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -270,6 +281,9 @@ void basic_tx::tx_set(const wax::obj &key_, const wax::obj &val){ case SUBDEV_PROP_FREQ: return; // it wont do you much good, but you can set it + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } diff --git a/host/lib/usrp/dboard/db_dbsrx.cpp b/host/lib/usrp/dboard/db_dbsrx.cpp index 81434f054..23cafea53 100644 --- a/host/lib/usrp/dboard/db_dbsrx.cpp +++ b/host/lib/usrp/dboard/db_dbsrx.cpp @@ -551,6 +551,10 @@ void dbsrx::rx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -587,6 +591,9 @@ void dbsrx::rx_set(const wax::obj &key_, const wax::obj &val){ this->set_gain(val.as(), key.name); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + case SUBDEV_PROP_BANDWIDTH: this->set_bandwidth(val.as()); return; diff --git a/host/lib/usrp/dboard/db_rfx.cpp b/host/lib/usrp/dboard/db_rfx.cpp index c3ab96e59..cfc34381e 100644 --- a/host/lib/usrp/dboard/db_rfx.cpp +++ b/host/lib/usrp/dboard/db_rfx.cpp @@ -444,6 +444,10 @@ void rfx_xcvr::rx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_QI; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -474,6 +478,9 @@ void rfx_xcvr::rx_set(const wax::obj &key_, const wax::obj &val){ this->set_rx_ant(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } @@ -524,6 +531,10 @@ void rfx_xcvr::tx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = true; return; @@ -554,6 +565,9 @@ void rfx_xcvr::tx_set(const wax::obj &key_, const wax::obj &val){ this->set_tx_ant(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } diff --git a/host/lib/usrp/dboard/db_unknown.cpp b/host/lib/usrp/dboard/db_unknown.cpp index f6f4f4a61..ec7ab440b 100644 --- a/host/lib/usrp/dboard/db_unknown.cpp +++ b/host/lib/usrp/dboard/db_unknown.cpp @@ -122,6 +122,10 @@ void unknown_rx::rx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -151,6 +155,9 @@ void unknown_rx::rx_set(const wax::obj &key_, const wax::obj &val){ case SUBDEV_PROP_FREQ: return; // it wont do you much good, but you can set it + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } @@ -211,6 +218,10 @@ void unknown_tx::tx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -240,6 +251,9 @@ void unknown_tx::tx_set(const wax::obj &key_, const wax::obj &val){ case SUBDEV_PROP_FREQ: return; // it wont do you much good, but you can set it + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } diff --git a/host/lib/usrp/dboard/db_wbx.cpp b/host/lib/usrp/dboard/db_wbx.cpp index e473ce41e..907268aac 100644 --- a/host/lib/usrp/dboard/db_wbx.cpp +++ b/host/lib/usrp/dboard/db_wbx.cpp @@ -513,6 +513,10 @@ void wbx_xcvr::rx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -543,6 +547,9 @@ void wbx_xcvr::rx_set(const wax::obj &key_, const wax::obj &val){ this->set_rx_ant(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } @@ -597,6 +604,10 @@ void wbx_xcvr::tx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -627,6 +638,9 @@ void wbx_xcvr::tx_set(const wax::obj &key_, const wax::obj &val){ this->set_tx_ant(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } diff --git a/host/lib/usrp/dboard/db_xcvr2450.cpp b/host/lib/usrp/dboard/db_xcvr2450.cpp index 798ff74a3..0ee133f3d 100644 --- a/host/lib/usrp/dboard/db_xcvr2450.cpp +++ b/host/lib/usrp/dboard/db_xcvr2450.cpp @@ -484,6 +484,10 @@ void xcvr2450::rx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -518,6 +522,9 @@ void xcvr2450::rx_set(const wax::obj &key_, const wax::obj &val){ this->set_rx_ant(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } @@ -572,6 +579,10 @@ void xcvr2450::tx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_QI; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -602,6 +613,9 @@ void xcvr2450::tx_set(const wax::obj &key_, const wax::obj &val){ this->set_tx_ant(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } diff --git a/host/lib/usrp/dboard_manager.cpp b/host/lib/usrp/dboard_manager.cpp index 181f843a0..78daa1b4d 100644 --- a/host/lib/usrp/dboard_manager.cpp +++ b/host/lib/usrp/dboard_manager.cpp @@ -86,7 +86,7 @@ std::string dboard_id_t::to_pp_string(void) const{ } /*********************************************************************** - * internal helper classe + * internal helper classes **********************************************************************/ /*! * A special wax proxy object that forwards calls to a subdev. @@ -330,4 +330,14 @@ void dboard_manager_impl::set_nice_dboard_if(void){ _iface->set_pin_ctrl(unit, 0x0000); //all gpio _iface->set_clock_enabled(unit, false); //clock off } + + //disable all rx subdevices + BOOST_FOREACH(const std::string &sd_name, this->get_rx_subdev_names()){ + this->get_rx_subdev(sd_name)[SUBDEV_PROP_ENABLED] = false; + } + + //disable all tx subdevices + BOOST_FOREACH(const std::string &sd_name, this->get_tx_subdev_names()){ + this->get_tx_subdev(sd_name)[SUBDEV_PROP_ENABLED] = false; + } } diff --git a/host/lib/usrp/misc_utils.cpp b/host/lib/usrp/misc_utils.cpp index 5cfcdc8d3..05308baba 100644 --- a/host/lib/usrp/misc_utils.cpp +++ b/host/lib/usrp/misc_utils.cpp @@ -17,6 +17,7 @@ #include #include +#include #include #include #include @@ -186,6 +187,22 @@ static void verify_xx_subdev_spec( "Validate %s subdev spec failed: %s\n %s" ) % xx_type % subdev_spec.to_string() % e.what())); } + + //now use the subdev spec to enable the subdevices in use and vice-versa + BOOST_FOREACH(const std::string &db_name, mboard[dboard_names_prop].as()){ + wax::obj dboard = mboard[named_prop_t(dboard_prop, db_name)]; + BOOST_FOREACH(const std::string &sd_name, dboard[DBOARD_PROP_SUBDEV_NAMES].as()){ + try{ + bool enable = std::has(subdev_spec, subdev_spec_pair_t(db_name, sd_name)); + dboard[named_prop_t(DBOARD_PROP_SUBDEV, sd_name)][SUBDEV_PROP_ENABLED] = enable; + } + catch(const std::exception &e){ + throw std::runtime_error(str(boost::format( + "Cannot set enabled property on subdevice %s:%s\n %s" + ) % db_name % sd_name % e.what())); + } + } + } } void usrp::verify_rx_subdev_spec(subdev_spec_t &subdev_spec, wax::obj mboard){ diff --git a/host/lib/usrp/subdev_spec.cpp b/host/lib/usrp/subdev_spec.cpp index 7a3e72867..95d2cbb12 100644 --- a/host/lib/usrp/subdev_spec.cpp +++ b/host/lib/usrp/subdev_spec.cpp @@ -34,6 +34,10 @@ subdev_spec_pair_t::subdev_spec_pair_t( /* NOP */ } +bool usrp::operator==(const subdev_spec_pair_t &lhs, const subdev_spec_pair_t &rhs){ + return (lhs.db_name == rhs.db_name) and (lhs.sd_name == rhs.sd_name); +} + subdev_spec_t::subdev_spec_t(const std::string &markup){ BOOST_FOREACH(const std::string &pair, std::split_string(markup)){ if (pair == "") continue; -- cgit v1.2.3 From 00cd6018405b57a0982b0ce103ff858c646ee18c Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Fri, 1 Oct 2010 18:22:41 -0700 Subject: uhd: implemented a double timeout (in seconds) for send and recv chains converted all size_t timeout_ms to double timeout bounded and alignment buffer now take double timeout added timeout to device::send and zero_copy_if::get_send_buff --- host/include/uhd/device.hpp | 27 ++++++++------ host/include/uhd/device.ipp | 9 +++-- host/include/uhd/transport/alignment_buffer.hpp | 9 ++--- host/include/uhd/transport/alignment_buffer.ipp | 15 +++----- host/include/uhd/transport/bounded_buffer.hpp | 12 ++---- host/include/uhd/transport/bounded_buffer.ipp | 15 ++++++-- host/include/uhd/transport/zero_copy.hpp | 18 +++++---- host/lib/transport/libusb1_zero_copy.cpp | 42 ++++++++++----------- host/lib/transport/udp_zero_copy_asio.cpp | 4 +- host/lib/transport/vrt_packet_handler.hpp | 49 +++++++++++++------------ host/lib/transport/zero_copy.cpp | 6 +-- host/lib/usrp/usrp1/io_impl.cpp | 20 +++++----- host/lib/usrp/usrp1/usrp1_impl.cpp | 4 +- host/lib/usrp/usrp1/usrp1_impl.hpp | 7 ++-- host/lib/usrp/usrp2/io_impl.cpp | 32 ++++++++-------- host/lib/usrp/usrp2/usrp2_impl.hpp | 6 +-- host/test/buffer_test.cpp | 2 +- 17 files changed, 140 insertions(+), 137 deletions(-) (limited to 'host/lib') diff --git a/host/include/uhd/device.hpp b/host/include/uhd/device.hpp index 2077cae62..992276928 100644 --- a/host/include/uhd/device.hpp +++ b/host/include/uhd/device.hpp @@ -41,9 +41,6 @@ public: typedef boost::function find_t; typedef boost::function make_t; - //! A reasonable default timeout for receive - static const size_t default_recv_timeout_ms = 100; - /*! * Register a device into the discovery and factory system. * @@ -112,12 +109,15 @@ public: * * This is a blocking call and will not return until the number * of samples returned have been read out of each buffer. + * Under a timeout condition, the number of samples returned + * may be less than the number of samples specified. * * \param buffs a vector of read-only memory containing IF data * \param nsamps_per_buff the number of samples to send, per buffer * \param metadata data describing the buffer's contents * \param io_type the type of data loaded in the buffer * \param send_mode tells send how to unload the buffer + * \param timeout the timeout in seconds to wait on a packet * \return the number of samples sent */ virtual size_t send( @@ -125,7 +125,8 @@ public: size_t nsamps_per_buff, const tx_metadata_t &metadata, const io_type_t &io_type, - send_mode_t send_mode + send_mode_t send_mode, + double timeout = 0.1 ) = 0; /*! @@ -136,7 +137,8 @@ public: size_t nsamps_per_buff, const tx_metadata_t &metadata, const io_type_t &io_type, - send_mode_t send_mode + send_mode_t send_mode, + double timeout = 0.1 ); /*! @@ -154,7 +156,9 @@ public: * See the rx metadata fragment flags and offset fields for details. * * This is a blocking call and will not return until the number - * of samples returned have been written into each buffer or timeout. + * of samples returned have been written into each buffer. + * Under a timeout condition, the number of samples returned + * may be less than the number of samples specified. * * When using the full buffer recv mode, the metadata only applies * to the first packet received and written into the recv buffers. @@ -165,7 +169,7 @@ public: * \param metadata data to fill describing the buffer * \param io_type the type of data to fill into the buffer * \param recv_mode tells recv how to load the buffer - * \param timeout_ms the timeout in milliseconds to wait for a packet + * \param timeout the timeout in seconds to wait for a packet * \return the number of samples received or 0 on error */ virtual size_t recv( @@ -174,7 +178,7 @@ public: rx_metadata_t &metadata, const io_type_t &io_type, recv_mode_t recv_mode, - size_t timeout_ms = default_recv_timeout_ms + double timeout = 0.1 ) = 0; /*! @@ -186,7 +190,7 @@ public: rx_metadata_t &metadata, const io_type_t &io_type, recv_mode_t recv_mode, - size_t timeout_ms = default_recv_timeout_ms + double timeout = 0.1 ); /*! @@ -204,12 +208,11 @@ public: /*! * Receive and asynchronous message from the device. * \param async_metadata the metadata to be filled in - * \param timeout_ms the timeout in milliseconds to wait for a message + * \param timeout the timeout in seconds to wait for a message * \return true when the async_metadata is valid, false for timeout */ virtual bool recv_async_msg( - async_metadata_t &async_metadata, - size_t timeout_ms = default_recv_timeout_ms + async_metadata_t &async_metadata, double timeout = 0.1 ) = 0; }; diff --git a/host/include/uhd/device.ipp b/host/include/uhd/device.ipp index 60a3f535d..e2e51ecd0 100644 --- a/host/include/uhd/device.ipp +++ b/host/include/uhd/device.ipp @@ -25,12 +25,13 @@ namespace uhd{ size_t nsamps_per_buff, const tx_metadata_t &metadata, const io_type_t &io_type, - send_mode_t send_mode + send_mode_t send_mode, + double timeout ){ return this->send( std::vector(1, buff), nsamps_per_buff, metadata, - io_type, send_mode + io_type, send_mode, timeout ); } @@ -40,12 +41,12 @@ namespace uhd{ rx_metadata_t &metadata, const io_type_t &io_type, recv_mode_t recv_mode, - size_t timeout_ms + double timeout ){ return this->recv( std::vector(1, buff), nsamps_per_buff, metadata, - io_type, recv_mode, timeout_ms + io_type, recv_mode, timeout ); } diff --git a/host/include/uhd/transport/alignment_buffer.hpp b/host/include/uhd/transport/alignment_buffer.hpp index 29ba74efc..f44a037f8 100644 --- a/host/include/uhd/transport/alignment_buffer.hpp +++ b/host/include/uhd/transport/alignment_buffer.hpp @@ -48,20 +48,17 @@ namespace uhd{ namespace transport{ * \return true if the element fit without popping for space */ virtual bool push_with_pop_on_full( - const elem_type &elem, - const seq_type &seq, - size_t index + const elem_type &elem, const seq_type &seq, size_t index ) = 0; /*! * Pop an aligned set of elements from this alignment buffer. * \param elems a collection to store the aligned elements - * \param time the timeout time + * \param timeout the timeout in seconds * \return false when the operation times out */ virtual bool pop_elems_with_timed_wait( - std::vector &elems, - const time_duration_t &time + std::vector &elems, double timeout ) = 0; }; diff --git a/host/include/uhd/transport/alignment_buffer.ipp b/host/include/uhd/transport/alignment_buffer.ipp index 61b3b60f5..5f09de0d9 100644 --- a/host/include/uhd/transport/alignment_buffer.ipp +++ b/host/include/uhd/transport/alignment_buffer.ipp @@ -41,9 +41,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ } UHD_INLINE bool push_with_pop_on_full( - const elem_type &elem, - const seq_type &seq, - size_t index + const elem_type &elem, const seq_type &seq, size_t index ){ //clear the buffer for this index if the seqs are mis-ordered if (seq < _last_seqs[index]){ @@ -54,17 +52,16 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ } UHD_INLINE bool pop_elems_with_timed_wait( - std::vector &elems, - const time_duration_t &time + std::vector &elems, double timeout ){ - boost::system_time exit_time = boost::get_system_time() + time; + boost::system_time exit_time = boost::get_system_time() + boost::posix_time::microseconds(long(timeout*1e6)); buff_contents_type buff_contents_tmp; std::list indexes_to_do(_all_indexes); //do an initial pop to load an initial sequence id size_t index = indexes_to_do.front(); if (not _buffs[index]->pop_with_timed_wait( - buff_contents_tmp, exit_time - boost::get_system_time() + buff_contents_tmp, 1e-6*(exit_time - boost::get_system_time()).total_microseconds() )) return false; elems[index] = buff_contents_tmp.first; seq_type expected_seq_id = buff_contents_tmp.second; @@ -79,7 +76,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ indexes_to_do = _all_indexes; index = indexes_to_do.front(); if (not _buffs[index]->pop_with_timed_wait( - buff_contents_tmp, exit_time - boost::get_system_time() + buff_contents_tmp, 1e-6*(exit_time - boost::get_system_time()).total_microseconds() )) return false; elems[index] = buff_contents_tmp.first; expected_seq_id = buff_contents_tmp.second; @@ -89,7 +86,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ //pop an element off for this index index = indexes_to_do.front(); if (not _buffs[index]->pop_with_timed_wait( - buff_contents_tmp, exit_time - boost::get_system_time() + buff_contents_tmp, 1e-6*(exit_time - boost::get_system_time()).total_microseconds() )) return false; //if the sequence id matches: diff --git a/host/include/uhd/transport/bounded_buffer.hpp b/host/include/uhd/transport/bounded_buffer.hpp index d1deece96..aca93b071 100644 --- a/host/include/uhd/transport/bounded_buffer.hpp +++ b/host/include/uhd/transport/bounded_buffer.hpp @@ -20,13 +20,9 @@ #include #include -#include namespace uhd{ namespace transport{ - //! typedef for the time duration type for wait operations - typedef boost::posix_time::time_duration time_duration_t; - /*! * Implement a templated bounded buffer: * Used for passing elements between threads in a producer-consumer model. @@ -64,10 +60,10 @@ namespace uhd{ namespace transport{ * Push a new element into the bounded_buffer. * Wait until the bounded_buffer becomes non-full or timeout. * \param elem the new element to push - * \param time the timeout time + * \param timeout the timeout in seconds * \return false when the operation times out */ - virtual bool push_with_timed_wait(const elem_type &elem, const time_duration_t &time) = 0; + virtual bool push_with_timed_wait(const elem_type &elem, double timeout) = 0; /*! * Pop an element from the bounded_buffer. @@ -80,10 +76,10 @@ namespace uhd{ namespace transport{ * Pop an element from the bounded_buffer. * Wait until the bounded_buffer becomes non-empty or timeout. * \param elem the element reference pop to - * \param time the timeout time + * \param timeout the timeout in seconds * \return false when the operation times out */ - virtual bool pop_with_timed_wait(elem_type &elem, const time_duration_t &time) = 0; + virtual bool pop_with_timed_wait(elem_type &elem, double timeout) = 0; /*! * Clear all elements from the bounded_buffer. diff --git a/host/include/uhd/transport/bounded_buffer.ipp b/host/include/uhd/transport/bounded_buffer.ipp index e106e229e..71143741e 100644 --- a/host/include/uhd/transport/bounded_buffer.ipp +++ b/host/include/uhd/transport/bounded_buffer.ipp @@ -21,6 +21,7 @@ #include #include #include +#include namespace uhd{ namespace transport{ namespace{ /*anon*/ @@ -57,9 +58,12 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ _empty_cond.notify_one(); } - bool push_with_timed_wait(const elem_type &elem, const time_duration_t &time){ + bool push_with_timed_wait(const elem_type &elem, double timeout){ boost::unique_lock lock(_mutex); - if (not _full_cond.timed_wait(lock, time, boost::bind(&bounded_buffer_impl::not_full, this))) return false; + if (not _full_cond.timed_wait( + lock, boost::posix_time::microseconds(long(timeout*1e6)), + boost::bind(&bounded_buffer_impl::not_full, this) + )) return false; _buffer.push_front(elem); lock.unlock(); _empty_cond.notify_one(); @@ -74,9 +78,12 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ _full_cond.notify_one(); } - bool pop_with_timed_wait(elem_type &elem, const time_duration_t &time){ + bool pop_with_timed_wait(elem_type &elem, double timeout){ boost::unique_lock lock(_mutex); - if (not _empty_cond.timed_wait(lock, time, boost::bind(&bounded_buffer_impl::not_empty, this))) return false; + if (not _empty_cond.timed_wait( + lock, boost::posix_time::microseconds(long(timeout*1e6)), + boost::bind(&bounded_buffer_impl::not_empty, this) + )) return false; elem = _buffer.back(); _buffer.pop_back(); lock.unlock(); _full_cond.notify_one(); diff --git a/host/include/uhd/transport/zero_copy.hpp b/host/include/uhd/transport/zero_copy.hpp index 8ecafd3fb..ba19b193c 100644 --- a/host/include/uhd/transport/zero_copy.hpp +++ b/host/include/uhd/transport/zero_copy.hpp @@ -122,10 +122,10 @@ namespace uhd{ namespace transport{ /*! * Get a new receive buffer from this transport object. - * \param timeout_ms the timeout to get the buffer in ms + * \param timeout the timeout to get the buffer in seconds * \return a managed buffer, or null sptr on timeout/error */ - virtual managed_recv_buffer::sptr get_recv_buff(size_t timeout_ms) = 0; + virtual managed_recv_buffer::sptr get_recv_buff(double timeout = 0.1) = 0; /*! * Get the maximum number of receive frames: @@ -138,9 +138,10 @@ namespace uhd{ namespace transport{ /*! * Get a new send buffer from this transport object. + * \param timeout the timeout to get the buffer in seconds * \return a managed buffer, or null sptr on timeout/error */ - virtual managed_send_buffer::sptr get_send_buff(void) = 0; + virtual managed_send_buffer::sptr get_send_buff(double timeout = 0.1) = 0; /*! * Get the maximum number of send frames: @@ -172,19 +173,19 @@ namespace uhd{ namespace transport{ /*! * Get a new receive buffer from this transport object. - * \param timeout_ms the timeout to get the buffer in ms + * \param timeout the timeout to get the buffer in seconds * \return a managed buffer, or null sptr on timeout/error */ - managed_recv_buffer::sptr get_recv_buff(size_t timeout_ms); + managed_recv_buffer::sptr get_recv_buff(double timeout); private: /*! * Perform a private copying recv. * \param buff the buffer to write data into - * \param timeout_ms the timeout to get the buffer in ms + * \param timeout the timeout to get the buffer in seconds * \return the number of bytes written to buff, 0 for timeout, negative for error */ - virtual ssize_t recv(const boost::asio::mutable_buffer &buff, size_t timeout_ms) = 0; + virtual ssize_t recv(const boost::asio::mutable_buffer &buff, double timeout) = 0; UHD_PIMPL_DECL(impl) _impl; }; @@ -208,9 +209,10 @@ namespace uhd{ namespace transport{ /*! * Get a new send buffer from this transport object. + * \param timeout the timeout to get the buffer in seconds * \return a managed buffer, or null sptr on timeout/error */ - managed_send_buffer::sptr get_send_buff(void); + managed_send_buffer::sptr get_send_buff(double timeout); private: /*! diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index f9beb0b4c..7f2bc3468 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -24,12 +24,10 @@ #include #include #include -#include using namespace uhd::transport; -const int libusb_timeout = 0; - +static const double CLEANUP_TIMEOUT = 0.2; //seconds static const size_t DEFAULT_NUM_XFERS = 16; //num xfers static const size_t DEFAULT_XFER_SIZE = 32*512; //bytes @@ -84,10 +82,10 @@ public: * Get an available transfer: * For inputs, this is a just filled transfer. * For outputs, this is a just emptied transfer. - * \param timeout_ms the timeout to wait for a lut + * \param timeout the timeout to wait for a lut * \return the transfer pointer or NULL if timeout */ - libusb_transfer *get_lut_with_wait(size_t timeout_ms = 100); + libusb_transfer *get_lut_with_wait(double timeout); //Callback use only void callback_handle_transfer(libusb_transfer *lut); @@ -187,7 +185,7 @@ usb_endpoint::~usb_endpoint(void){ } //collect canceled transfers (drain the queue) - while (this->get_lut_with_wait() != NULL){}; + while (this->get_lut_with_wait(CLEANUP_TIMEOUT) != NULL){}; //free all transfers BOOST_FOREACH(libusb_transfer *lut, _all_luts){ @@ -274,12 +272,10 @@ void usb_endpoint::print_transfer_status(libusb_transfer *lut){ } } -libusb_transfer *usb_endpoint::get_lut_with_wait(size_t timeout_ms){ +libusb_transfer *usb_endpoint::get_lut_with_wait(double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw libusb_transfer *lut; - if (_completed_list->pop_with_timed_wait( - lut, boost::posix_time::milliseconds(timeout_ms) - )) return lut; + if (_completed_list->pop_with_timed_wait(lut, timeout)) return lut; return NULL; } @@ -399,8 +395,8 @@ public: size_t send_xfer_size, size_t send_num_xfers ); - managed_recv_buffer::sptr get_recv_buff(size_t timeout_ms); - managed_send_buffer::sptr get_send_buff(void); + managed_recv_buffer::sptr get_recv_buff(double); + managed_send_buffer::sptr get_send_buff(double); size_t get_num_recv_frames(void) const { return _recv_num_frames; } size_t get_num_send_frames(void) const { return _send_num_frames; } @@ -459,8 +455,8 @@ libusb_zero_copy_impl::libusb_zero_copy_impl( * Return empty pointer if no transfer is available (timeout or error). * \return pointer to a managed receive buffer */ -managed_recv_buffer::sptr libusb_zero_copy_impl::get_recv_buff(size_t timeout_ms){ - libusb_transfer *lut = _recv_ep->get_lut_with_wait(timeout_ms); +managed_recv_buffer::sptr libusb_zero_copy_impl::get_recv_buff(double timeout){ + libusb_transfer *lut = _recv_ep->get_lut_with_wait(timeout); if (lut == NULL) { return managed_recv_buffer::sptr(); } @@ -478,8 +474,8 @@ managed_recv_buffer::sptr libusb_zero_copy_impl::get_recv_buff(size_t timeout_ms * (timeout or error). * \return pointer to a managed send buffer */ -managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(void){ - libusb_transfer *lut = _send_ep->get_lut_with_wait(/* TODO timeout API */); +managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(double timeout){ + libusb_transfer *lut = _send_ep->get_lut_with_wait(timeout); if (lut == NULL) { return managed_send_buffer::sptr(); } @@ -494,18 +490,18 @@ managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(void){ * USB zero_copy make functions **********************************************************************/ usb_zero_copy::sptr usb_zero_copy::make( - usb_device_handle::sptr handle, + usb_device_handle::sptr handle, unsigned int recv_endpoint, unsigned int send_endpoint, - size_t recv_xfer_size, size_t recv_num_xfers, - size_t send_xfer_size, size_t send_num_xfers + size_t recv_xfer_size, size_t recv_num_xfers, + size_t send_xfer_size, size_t send_num_xfers ){ libusb::device_handle::sptr dev_handle(libusb::device_handle::get_cached_handle( boost::static_pointer_cast(handle)->get_device() )); return sptr(new libusb_zero_copy_impl( - dev_handle, - recv_endpoint, send_endpoint, - recv_xfer_size, recv_num_xfers, - send_xfer_size, send_num_xfers + dev_handle, + recv_endpoint, send_endpoint, + recv_xfer_size, recv_num_xfers, + send_xfer_size, send_num_xfers )); } diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 0a6c9f2af..3130830a5 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -109,11 +109,11 @@ private: boost::asio::io_service _io_service; int _sock_fd; - ssize_t recv(const boost::asio::mutable_buffer &buff, size_t timeout_ms){ + ssize_t recv(const boost::asio::mutable_buffer &buff, double timeout){ //setup timeval for timeout timeval tv; tv.tv_sec = 0; - tv.tv_usec = timeout_ms*1000; + tv.tv_usec = long(timeout*1e6); //setup rset for timeout fd_set rset; diff --git a/host/lib/transport/vrt_packet_handler.hpp b/host/lib/transport/vrt_packet_handler.hpp index b603f1371..e11afff30 100644 --- a/host/lib/transport/vrt_packet_handler.hpp +++ b/host/lib/transport/vrt_packet_handler.hpp @@ -303,18 +303,18 @@ template UHD_INLINE T get_context_code( * Pack a vrt header, copy-convert the data, and send it. * - helper function for vrt_packet_handler::send ******************************************************************/ - static UHD_INLINE void _send1( + static UHD_INLINE size_t _send1( send_state &state, const std::vector &buffs, - size_t offset_bytes, - size_t num_samps, + const size_t offset_bytes, + const size_t num_samps, uhd::transport::vrt::if_packet_info_t &if_packet_info, const uhd::io_type_t &io_type, const uhd::otw_type_t &otw_type, const vrt_packer_t &vrt_packer, const get_send_buffs_t &get_send_buffs, - size_t vrt_header_offset_words32, - size_t chans_per_otw_buff + const size_t vrt_header_offset_words32, + const size_t chans_per_otw_buff ){ //load the rest of the if_packet_info in here if_packet_info.num_payload_words32 = (num_samps*chans_per_otw_buff*otw_type.get_sample_size())/sizeof(boost::uint32_t); @@ -322,7 +322,7 @@ template UHD_INLINE T get_context_code( //get send buffers for each channel managed_send_buffs_t send_buffs(buffs.size()/chans_per_otw_buff); - UHD_ASSERT_THROW(get_send_buffs(send_buffs)); + if (not get_send_buffs(send_buffs)) return 0; std::vector io_buffs(chans_per_otw_buff); for (size_t i = 0; i < buffs.size(); i+=chans_per_otw_buff){ @@ -347,6 +347,7 @@ template UHD_INLINE T get_context_code( std::cerr << "commit to send buffer returned less than commit size" << std::endl; } } + return num_samps; } /******************************************************************* @@ -381,7 +382,6 @@ template UHD_INLINE T get_context_code( //////////////////////////////////////////////////////////////// case uhd::device::SEND_MODE_ONE_PACKET:{ //////////////////////////////////////////////////////////////// - size_t num_samps = std::min(total_num_samps, max_samples_per_packet); //fill in parts of the packet info overwrote in full buff mode if_packet_info.has_tsi = metadata.has_time_spec; @@ -389,10 +389,10 @@ template UHD_INLINE T get_context_code( if_packet_info.sob = metadata.start_of_burst; if_packet_info.eob = metadata.end_of_burst; - _send1( + return _send1( state, buffs, 0, - num_samps, + std::min(total_num_samps, max_samples_per_packet), if_packet_info, io_type, otw_type, vrt_packer, @@ -400,31 +400,32 @@ template UHD_INLINE T get_context_code( vrt_header_offset_words32, chans_per_otw_buff ); - return num_samps; } //////////////////////////////////////////////////////////////// case uhd::device::SEND_MODE_FULL_BUFF:{ //////////////////////////////////////////////////////////////// - //calculate constants for fragmentation - const size_t num_fragments = (total_num_samps+max_samples_per_packet-1)/max_samples_per_packet; - static const size_t first_fragment_index = 0; - const size_t final_fragment_index = num_fragments-1; + size_t total_num_samps_sent = 0; //loop through the following fragment indexes - for (size_t n = first_fragment_index; n <= final_fragment_index; n++){ + while(total_num_samps_sent < total_num_samps){ + + //calculate per-loop-iteration variables + const size_t total_num_samps_unsent = total_num_samps - total_num_samps_sent; + const bool first_fragment = (total_num_samps_sent == 0); + const bool final_fragment = (total_num_samps_unsent <= max_samples_per_packet); //calculate new flags for the fragments - if_packet_info.has_tsi = metadata.has_time_spec and (n == first_fragment_index); - if_packet_info.has_tsf = metadata.has_time_spec and (n == first_fragment_index); - if_packet_info.sob = metadata.start_of_burst and (n == first_fragment_index); - if_packet_info.eob = metadata.end_of_burst and (n == final_fragment_index); + if_packet_info.has_tsi = metadata.has_time_spec and first_fragment; + if_packet_info.has_tsf = if_packet_info.has_tsi; + if_packet_info.sob = metadata.start_of_burst and first_fragment; + if_packet_info.eob = metadata.end_of_burst and final_fragment; //send the fragment with the helper function - _send1( + const size_t num_samps_sent = _send1( state, - buffs, n*max_samples_per_packet*io_type.size, - (n == final_fragment_index)?(total_num_samps%max_samples_per_packet):max_samples_per_packet, + buffs, total_num_samps_sent*io_type.size, + std::min(total_num_samps_unsent, max_samples_per_packet), if_packet_info, io_type, otw_type, vrt_packer, @@ -432,8 +433,10 @@ template UHD_INLINE T get_context_code( vrt_header_offset_words32, chans_per_otw_buff ); + total_num_samps_sent += num_samps_sent; + if (num_samps_sent == 0) return total_num_samps_sent; } - return total_num_samps; + return total_num_samps_sent; } default: throw std::runtime_error("unknown send mode"); diff --git a/host/lib/transport/zero_copy.cpp b/host/lib/transport/zero_copy.cpp index 1fcf846a0..dfb65951f 100644 --- a/host/lib/transport/zero_copy.cpp +++ b/host/lib/transport/zero_copy.cpp @@ -68,12 +68,12 @@ phony_zero_copy_recv_if::~phony_zero_copy_recv_if(void){ /* NOP */ } -managed_recv_buffer::sptr phony_zero_copy_recv_if::get_recv_buff(size_t timeout_ms){ +managed_recv_buffer::sptr phony_zero_copy_recv_if::get_recv_buff(double timeout){ //allocate memory boost::uint8_t *recv_mem = new boost::uint8_t[_impl->max_buff_size]; //call recv() with timeout option - ssize_t num_bytes = this->recv(boost::asio::buffer(recv_mem, _impl->max_buff_size), timeout_ms); + ssize_t num_bytes = this->recv(boost::asio::buffer(recv_mem, _impl->max_buff_size), timeout); if (num_bytes <= 0) return managed_recv_buffer::sptr(); //NULL sptr @@ -138,6 +138,6 @@ phony_zero_copy_send_if::~phony_zero_copy_send_if(void){ delete [] _impl->send_mem; } -managed_send_buffer::sptr phony_zero_copy_send_if::get_send_buff(void){ +managed_send_buffer::sptr phony_zero_copy_send_if::get_send_buff(double){ return _impl->send_buff; //FIXME there is only ever one send buff, we assume that the caller doesnt hang onto these } diff --git a/host/lib/usrp/usrp1/io_impl.cpp b/host/lib/usrp/usrp1/io_impl.cpp index aee760a83..8d9c68961 100644 --- a/host/lib/usrp/usrp1/io_impl.cpp +++ b/host/lib/usrp/usrp1/io_impl.cpp @@ -94,12 +94,13 @@ struct usrp1_impl::io_impl{ //all of this to ensure only full buffers are committed managed_send_buffer::sptr send_buff; size_t num_bytes_committed; + double send_timeout; boost::uint8_t pseudo_buff[BYTES_PER_PACKET]; ssize_t phony_commit_pseudo_buff(size_t num_bytes); ssize_t phony_commit_send_buff(size_t num_bytes); ssize_t commit_send_buff(void); void flush_send_buff(void); - bool get_send_buffs(vrt_packet_handler::managed_send_buffs_t &); + bool get_send_buffs(vrt_packet_handler::managed_send_buffs_t &, double); //helpers to get at the send buffer + offset inline void *get_send_mem_ptr(void){ @@ -159,14 +160,15 @@ void usrp1_impl::io_impl::flush_send_buff(void){ */ ssize_t usrp1_impl::io_impl::commit_send_buff(void){ ssize_t ret = send_buff->commit(num_bytes_committed); - send_buff = data_transport->get_send_buff(); + send_buff = data_transport->get_send_buff(send_timeout); num_bytes_committed = 0; return ret; } bool usrp1_impl::io_impl::get_send_buffs( - vrt_packet_handler::managed_send_buffs_t &buffs + vrt_packet_handler::managed_send_buffs_t &buffs, double timeout ){ + send_timeout = timeout; UHD_ASSERT_THROW(buffs.size() == 1); //not enough bytes free -> use the pseudo buffer @@ -216,7 +218,7 @@ static void usrp1_bs_vrt_packer( size_t usrp1_impl::send( const std::vector &buffs, size_t num_samps, const tx_metadata_t &metadata, const io_type_t &io_type, - send_mode_t send_mode + send_mode_t send_mode, double timeout ){ size_t num_samps_sent = vrt_packet_handler::send( _io_impl->packet_handler_send_state, //last state of the send handler @@ -225,7 +227,7 @@ size_t usrp1_impl::send( io_type, _tx_otw_type, //input and output types to convert _clock_ctrl->get_master_clock_freq(), //master clock tick rate &usrp1_bs_vrt_packer, - boost::bind(&usrp1_impl::io_impl::get_send_buffs, _io_impl.get(), _1), + boost::bind(&usrp1_impl::io_impl::get_send_buffs, _io_impl.get(), _1, timeout), get_max_send_samps_per_packet(), 0, //vrt header offset _tx_subdev_spec.size() //num channels @@ -272,18 +274,18 @@ static void usrp1_bs_vrt_unpacker( } static bool get_recv_buffs( - zero_copy_if::sptr zc_if, size_t timeout_ms, + zero_copy_if::sptr zc_if, double timeout, vrt_packet_handler::managed_recv_buffs_t &buffs ){ UHD_ASSERT_THROW(buffs.size() == 1); - buffs[0] = zc_if->get_recv_buff(timeout_ms); + buffs[0] = zc_if->get_recv_buff(timeout); return buffs[0].get() != NULL; } size_t usrp1_impl::recv( const std::vector &buffs, size_t num_samps, rx_metadata_t &metadata, const io_type_t &io_type, - recv_mode_t recv_mode, size_t timeout_ms + recv_mode_t recv_mode, double timeout ){ size_t num_samps_recvd = vrt_packet_handler::recv( _io_impl->packet_handler_recv_state, //last state of the recv handler @@ -292,7 +294,7 @@ size_t usrp1_impl::recv( io_type, _rx_otw_type, //input and output types to convert _clock_ctrl->get_master_clock_freq(), //master clock tick rate &usrp1_bs_vrt_unpacker, - boost::bind(&get_recv_buffs, _data_transport, timeout_ms, _1), + boost::bind(&get_recv_buffs, _data_transport, timeout, _1), &vrt_packet_handler::handle_overflow_nop, 0, //vrt header offset _rx_subdev_spec.size() //num channels diff --git a/host/lib/usrp/usrp1/usrp1_impl.cpp b/host/lib/usrp/usrp1/usrp1_impl.cpp index 156fc119f..6ebd6bb09 100644 --- a/host/lib/usrp/usrp1/usrp1_impl.cpp +++ b/host/lib/usrp/usrp1/usrp1_impl.cpp @@ -209,9 +209,9 @@ usrp1_impl::~usrp1_impl(void){ /* NOP */ } -bool usrp1_impl::recv_async_msg(uhd::async_metadata_t &, size_t timeout_ms){ +bool usrp1_impl::recv_async_msg(uhd::async_metadata_t &, double timeout){ //dummy fill-in for the recv_async_msg - boost::this_thread::sleep(boost::posix_time::milliseconds(timeout_ms)); + boost::this_thread::sleep(boost::posix_time::microseconds(long(timeout*1e6))); return false; } diff --git a/host/lib/usrp/usrp1/usrp1_impl.hpp b/host/lib/usrp/usrp1/usrp1_impl.hpp index 20ae3c02a..f2c464610 100644 --- a/host/lib/usrp/usrp1/usrp1_impl.hpp +++ b/host/lib/usrp/usrp1/usrp1_impl.hpp @@ -83,13 +83,12 @@ public: size_t, const uhd::tx_metadata_t &, const uhd::io_type_t &, - send_mode_t); + send_mode_t, double); size_t recv(const std::vector &, size_t, uhd::rx_metadata_t &, const uhd::io_type_t &, - recv_mode_t, - size_t timeout); + recv_mode_t, double); static const size_t BYTES_PER_PACKET = 512*4; //under the transfer size @@ -101,7 +100,7 @@ public: return BYTES_PER_PACKET/_rx_otw_type.get_sample_size()/_rx_subdev_spec.size(); } - bool recv_async_msg(uhd::async_metadata_t &, size_t); + bool recv_async_msg(uhd::async_metadata_t &, double); private: /*! diff --git a/host/lib/usrp/usrp2/io_impl.cpp b/host/lib/usrp/usrp2/io_impl.cpp index 3395f94e2..c0d8ab029 100644 --- a/host/lib/usrp/usrp2/io_impl.cpp +++ b/host/lib/usrp/usrp2/io_impl.cpp @@ -33,7 +33,6 @@ using namespace uhd::transport; namespace asio = boost::asio; static const int underflow_flags = async_metadata_t::EVENT_CODE_UNDERFLOW | async_metadata_t::EVENT_CODE_UNDERFLOW_IN_PACKET; -static const double RECV_TIMEOUT_MS = 100; /*********************************************************************** * io impl details (internal to this file) @@ -59,9 +58,9 @@ struct usrp2_impl::io_impl{ recv_pirate_crew.join_all(); } - bool get_recv_buffs(vrt_packet_handler::managed_recv_buffs_t &buffs, size_t timeout_ms){ + bool get_recv_buffs(vrt_packet_handler::managed_recv_buffs_t &buffs, double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw - return recv_pirate_booty->pop_elems_with_timed_wait(buffs, boost::posix_time::milliseconds(timeout_ms)); + return recv_pirate_booty->pop_elems_with_timed_wait(buffs, timeout); } //state management for the vrt packet handler code @@ -91,7 +90,7 @@ void usrp2_impl::io_impl::recv_pirate_loop( size_t next_packet_seq = 0; while(recv_pirate_crew_raiding){ - managed_recv_buffer::sptr buff = zc_if->get_recv_buff(RECV_TIMEOUT_MS); + managed_recv_buffer::sptr buff = zc_if->get_recv_buff(); if (not buff.get()) continue; //ignore timeout/error buffers try{ @@ -151,7 +150,7 @@ void usrp2_impl::io_init(void){ std::memcpy(send_buff->cast(), &data, sizeof(data)); send_buff->commit(sizeof(data)); //drain the recv buffers (may have junk) - while (data_transport->get_recv_buff(RECV_TIMEOUT_MS).get()){}; + while (data_transport->get_recv_buff().get()){}; } //the number of recv frames is the number for the first transport @@ -179,12 +178,10 @@ void usrp2_impl::io_init(void){ * Async Data **********************************************************************/ bool usrp2_impl::recv_async_msg( - async_metadata_t &async_metadata, size_t timeout_ms + async_metadata_t &async_metadata, double timeout ){ boost::this_thread::disable_interruption di; //disable because the wait can throw - return _io_impl->async_msg_fifo->pop_with_timed_wait( - async_metadata, boost::posix_time::milliseconds(timeout_ms) - ); + return _io_impl->async_msg_fifo->pop_with_timed_wait(async_metadata, timeout); } /*********************************************************************** @@ -192,19 +189,22 @@ bool usrp2_impl::recv_async_msg( **********************************************************************/ static bool get_send_buffs( const std::vector &trans, - vrt_packet_handler::managed_send_buffs_t &buffs + vrt_packet_handler::managed_send_buffs_t &buffs, + double timeout ){ UHD_ASSERT_THROW(trans.size() == buffs.size()); + bool good = true; for (size_t i = 0; i < buffs.size(); i++){ - buffs[i] = trans[i]->get_send_buff(); + buffs[i] = trans[i]->get_send_buff(timeout); + good = good and (buffs[i].get() != NULL); } - return true; + return good; } size_t usrp2_impl::send( const std::vector &buffs, size_t num_samps, const tx_metadata_t &metadata, const io_type_t &io_type, - send_mode_t send_mode + send_mode_t send_mode, double timeout ){ return vrt_packet_handler::send( _io_impl->packet_handler_send_state, //last state of the send handler @@ -213,7 +213,7 @@ size_t usrp2_impl::send( io_type, _io_helper.get_tx_otw_type(), //input and output types to convert _mboards.front()->get_master_clock_freq(), //master clock tick rate uhd::transport::vrt::if_hdr_pack_be, - boost::bind(&get_send_buffs, _data_transports, _1), + boost::bind(&get_send_buffs, _data_transports, _1, timeout), get_max_send_samps_per_packet() ); } @@ -224,7 +224,7 @@ size_t usrp2_impl::send( size_t usrp2_impl::recv( const std::vector &buffs, size_t num_samps, rx_metadata_t &metadata, const io_type_t &io_type, - recv_mode_t recv_mode, size_t timeout_ms + recv_mode_t recv_mode, double timeout ){ return vrt_packet_handler::recv( _io_impl->packet_handler_recv_state, //last state of the recv handler @@ -233,6 +233,6 @@ size_t usrp2_impl::recv( io_type, _io_helper.get_rx_otw_type(), //input and output types to convert _mboards.front()->get_master_clock_freq(), //master clock tick rate uhd::transport::vrt::if_hdr_unpack_be, - boost::bind(&usrp2_impl::io_impl::get_recv_buffs, _io_impl.get(), _1, timeout_ms) + boost::bind(&usrp2_impl::io_impl::get_recv_buffs, _io_impl.get(), _1, timeout) ); } diff --git a/host/lib/usrp/usrp2/usrp2_impl.hpp b/host/lib/usrp/usrp2/usrp2_impl.hpp index 157d17057..e8763b284 100644 --- a/host/lib/usrp/usrp2/usrp2_impl.hpp +++ b/host/lib/usrp/usrp2/usrp2_impl.hpp @@ -234,7 +234,7 @@ public: size_t send( const std::vector &, size_t, const uhd::tx_metadata_t &, const uhd::io_type_t &, - uhd::device::send_mode_t + uhd::device::send_mode_t, double ); size_t get_max_recv_samps_per_packet(void) const{ return _io_helper.get_max_recv_samps_per_packet(); @@ -242,9 +242,9 @@ public: size_t recv( const std::vector &, size_t, uhd::rx_metadata_t &, const uhd::io_type_t &, - uhd::device::recv_mode_t, size_t + uhd::device::recv_mode_t, double ); - bool recv_async_msg(uhd::async_metadata_t &, size_t); + bool recv_async_msg(uhd::async_metadata_t &, double); private: //device properties interface diff --git a/host/test/buffer_test.cpp b/host/test/buffer_test.cpp index aadb3f951..8445412e7 100644 --- a/host/test/buffer_test.cpp +++ b/host/test/buffer_test.cpp @@ -23,7 +23,7 @@ using namespace boost::assign; using namespace uhd::transport; -static const boost::posix_time::milliseconds timeout(10); +static const double timeout = 0.01/*secs*/; BOOST_AUTO_TEST_CASE(test_bounded_buffer_with_timed_wait){ bounded_buffer::sptr bb(bounded_buffer::make(3)); -- cgit v1.2.3 From 7352c95037fa57d37dd7adc4c2ea6935006b56c8 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Sat, 2 Oct 2010 23:15:46 -0700 Subject: uhd: reworked the zero copy interface - recv buffers have a release call - safe make function for recv buffers (buff + release callback) - send buffer commits now have a void return - safe make function for send buffers (buff + commit callback) The reason for the void return from commit is that ssize_t num_bytes was never returning anything of use. That is for all of the zero copy implementations so far, commit cannot really error (being asynchronous). libusb zero copy impl was reworked to support the new interface. USRP1 io_impl with the psuedo managed buffer was replaced with safe_make. Also, usrp1 io_impl was simplified greatly due to commit returning void now. UDP zero copy asio was disabled (in this commit, until its reworked). Phony send and recv interfaces were removed completely. --- host/include/uhd/transport/zero_copy.hpp | 112 +++++++--------------- host/lib/transport/CMakeLists.txt | 4 +- host/lib/transport/libusb1_zero_copy.cpp | 149 +++++++----------------------- host/lib/transport/vrt_packet_handler.hpp | 4 +- host/lib/transport/zero_copy.cpp | 121 +++++++++--------------- host/lib/usrp/usrp1/io_impl.cpp | 65 ++++--------- 6 files changed, 130 insertions(+), 325 deletions(-) (limited to 'host/lib') diff --git a/host/include/uhd/transport/zero_copy.hpp b/host/include/uhd/transport/zero_copy.hpp index ba19b193c..9dd16280c 100644 --- a/host/include/uhd/transport/zero_copy.hpp +++ b/host/include/uhd/transport/zero_copy.hpp @@ -19,10 +19,10 @@ #define INCLUDED_UHD_TRANSPORT_ZERO_COPY_HPP #include -#include #include #include #include +#include namespace uhd{ namespace transport{ @@ -34,14 +34,27 @@ namespace uhd{ namespace transport{ class UHD_API managed_recv_buffer : boost::noncopyable{ public: typedef boost::shared_ptr sptr; + typedef boost::function release_fcn_t; + + /*! + * Make a safe managed receive buffer: + * A safe managed buffer ensures that release is called once, + * either by the user or automatically upon deconstruction. + * \param buff a reference to the constant buffer + * \param release_fcn callback to release the memory + * \return a new managed receive buffer + */ + static sptr make_safe( + const boost::asio::const_buffer &buff, + const release_fcn_t &release_fcn + ); /*! - * Managed recv buffer destructor: * Signal to the transport that we are done with the buffer. - * This should be called to release the buffer to the transport. + * This should be called to release the buffer to the transport object. * After calling, the referenced memory should be considered invalid. */ - virtual ~managed_recv_buffer(void) = 0; + virtual void release(void) = 0; /*! * Get the size of the underlying buffer. @@ -71,20 +84,34 @@ namespace uhd{ namespace transport{ /*! * A managed send buffer: * Contains a reference to transport-managed memory, - * and a method to release the memory after writing. + * and a method to commit the memory after writing. */ class UHD_API managed_send_buffer : boost::noncopyable{ public: typedef boost::shared_ptr sptr; + typedef boost::function commit_fcn_t; + + /*! + * Make a safe managed send buffer: + * A safe managed buffer ensures that commit is called once, + * either by the user or automatically upon deconstruction. + * In the later case, the deconstructor will call commit(0). + * \param buff a reference to the mutable buffer + * \param commit_fcn callback to commit the memory + * \return a new managed send buffer + */ + static sptr make_safe( + const boost::asio::mutable_buffer &buff, + const commit_fcn_t &commit_fcn + ); /*! * Signal to the transport that we are done with the buffer. * This should be called to commit the write to the transport object. * After calling, the referenced memory should be considered invalid. * \param num_bytes the number of bytes written into the buffer - * \return the number of bytes written, 0 for timeout, negative for error */ - virtual ssize_t commit(size_t num_bytes) = 0; + virtual void commit(size_t num_bytes) = 0; /*! * Get the size of the underlying buffer. @@ -154,77 +181,6 @@ namespace uhd{ namespace transport{ }; - /*! - * A phony-zero-copy interface for transport objects that - * provides a zero-copy interface on top of copying transport. - * This interface implements the get managed recv buffer, - * the base class must implement the private recv method. - */ - class UHD_API phony_zero_copy_recv_if : public virtual zero_copy_if{ - public: - /*! - * Create a phony zero copy recv interface. - * \param max_buff_size max buffer size in bytes - */ - phony_zero_copy_recv_if(size_t max_buff_size); - - //! destructor - virtual ~phony_zero_copy_recv_if(void); - - /*! - * Get a new receive buffer from this transport object. - * \param timeout the timeout to get the buffer in seconds - * \return a managed buffer, or null sptr on timeout/error - */ - managed_recv_buffer::sptr get_recv_buff(double timeout); - - private: - /*! - * Perform a private copying recv. - * \param buff the buffer to write data into - * \param timeout the timeout to get the buffer in seconds - * \return the number of bytes written to buff, 0 for timeout, negative for error - */ - virtual ssize_t recv(const boost::asio::mutable_buffer &buff, double timeout) = 0; - - UHD_PIMPL_DECL(impl) _impl; - }; - - /*! - * A phony-zero-copy interface for transport objects that - * provides a zero-copy interface on top of copying transport. - * This interface implements the get managed send buffer, - * the base class must implement the private send method. - */ - class UHD_API phony_zero_copy_send_if : public virtual zero_copy_if{ - public: - /*! - * Create a phony zero copy send interface. - * \param max_buff_size max buffer size in bytes - */ - phony_zero_copy_send_if(size_t max_buff_size); - - //! destructor - virtual ~phony_zero_copy_send_if(void); - - /*! - * Get a new send buffer from this transport object. - * \param timeout the timeout to get the buffer in seconds - * \return a managed buffer, or null sptr on timeout/error - */ - managed_send_buffer::sptr get_send_buff(double timeout); - - private: - /*! - * Perform a private copying send. - * \param buff the buffer to read data from - * \return the number of bytes read from buff, 0 for timeout, negative for error - */ - virtual ssize_t send(const boost::asio::const_buffer &buff) = 0; - - UHD_PIMPL_DECL(impl) _impl; - }; - }} //namespace #endif /* INCLUDED_UHD_TRANSPORT_ZERO_COPY_HPP */ diff --git a/host/lib/transport/CMakeLists.txt b/host/lib/transport/CMakeLists.txt index 61616d077..2be2c89ec 100644 --- a/host/lib/transport/CMakeLists.txt +++ b/host/lib/transport/CMakeLists.txt @@ -36,6 +36,8 @@ IF(LIBUSB_FOUND) INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/lib/transport/msvc) ENDIF(MSVC) SET(HAVE_USB_SUPPORT TRUE) +ELSE(LIBUSB_FOUND) + #TODO dummy usb ENDIF(LIBUSB_FOUND) IF(HAVE_USB_SUPPORT) @@ -99,7 +101,7 @@ SET_SOURCE_FILES_PROPERTIES( LIBUHD_APPEND_SOURCES( ${CMAKE_SOURCE_DIR}/lib/transport/if_addrs.cpp ${CMAKE_SOURCE_DIR}/lib/transport/udp_simple.cpp - ${CMAKE_SOURCE_DIR}/lib/transport/udp_zero_copy_asio.cpp + #${CMAKE_SOURCE_DIR}/lib/transport/udp_zero_copy_asio.cpp ${CMAKE_SOURCE_DIR}/lib/transport/vrt_packet_handler.hpp ${CMAKE_SOURCE_DIR}/lib/transport/zero_copy.cpp ) diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index 7f2bc3468..e1cc8398c 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -95,9 +95,6 @@ private: int _endpoint; bool _input; - size_t _transfer_size; - size_t _num_transfers; - //! hold a bounded buffer of completed transfers typedef bounded_buffer lut_buff_type; lut_buff_type::sptr _completed_list; @@ -154,14 +151,12 @@ usb_endpoint::usb_endpoint( ): _handle(handle), _endpoint(endpoint), - _input(input), - _transfer_size(transfer_size), - _num_transfers(num_transfers) + _input(input) { _completed_list = lut_buff_type::make(num_transfers); - for (size_t i = 0; i < _num_transfers; i++){ - _all_luts.push_back(allocate_transfer(_transfer_size)); + for (size_t i = 0; i < num_transfers; i++){ + _all_luts.push_back(allocate_transfer(transfer_size)); //input luts are immediately submitted to be filled //output luts go into the completed list as free buffers @@ -280,126 +275,44 @@ libusb_transfer *usb_endpoint::get_lut_with_wait(double timeout){ } /*********************************************************************** - * Managed buffers + * USB zero_copy device class **********************************************************************/ -/* - * Libusb managed receive buffer - * Construct a recv buffer from a libusb transfer. The memory held by - * the libusb transfer is exposed through the managed buffer interface. - * Upon destruction, the transfer and buffer are resubmitted to the - * endpoint for further use. - */ -class libusb_managed_recv_buffer_impl : public managed_recv_buffer { +class libusb_zero_copy_impl : public usb_zero_copy { public: - libusb_managed_recv_buffer_impl(libusb_transfer *lut, - usb_endpoint::sptr endpoint) - : _buff(lut->buffer, lut->length) - { - _lut = lut; - _endpoint = endpoint; - } - - ~libusb_managed_recv_buffer_impl(void){ - _endpoint->submit(_lut); - } + typedef boost::shared_ptr sptr; -private: - const boost::asio::const_buffer &get(void) const{ - return _buff; - } + libusb_zero_copy_impl( + libusb::device_handle::sptr handle, + unsigned int recv_endpoint, unsigned int send_endpoint, + size_t recv_xfer_size, size_t recv_num_xfers, + size_t send_xfer_size, size_t send_num_xfers + ); - libusb_transfer *_lut; - usb_endpoint::sptr _endpoint; - const boost::asio::const_buffer _buff; -}; + managed_recv_buffer::sptr get_recv_buff(double); + managed_send_buffer::sptr get_send_buff(double); -/* - * Libusb managed send buffer - * Construct a send buffer from a libusb transfer. The memory held by - * the libusb transfer is exposed through the managed buffer interface. - * Committing the buffer will set the data length and submit the buffer - * to the endpoint. Submitting a buffer multiple times or destroying - * the buffer before committing is an error. For the latter, the transfer - * is returned to the endpoint with no data for reuse. - */ -class libusb_managed_send_buffer_impl : public managed_send_buffer { -public: - libusb_managed_send_buffer_impl(libusb_transfer *lut, - usb_endpoint::sptr endpoint) - : _buff(lut->buffer, lut->length), _committed(false) - { - _lut = lut; - _endpoint = endpoint; - } + size_t get_num_recv_frames(void) const { return _recv_num_frames; } + size_t get_num_send_frames(void) const { return _send_num_frames; } - ~libusb_managed_send_buffer_impl(void){ - if (!_committed) { - _lut->length = 0; - _lut->actual_length = 0; - _endpoint->submit(_lut); - } +private: + void release(libusb_transfer *lut){ + _recv_ep->submit(lut); } - ssize_t commit(size_t num_bytes) - { - if (_committed) { - std::cerr << "UHD: send buffer already committed" << std::endl; - return 0; - } - - UHD_ASSERT_THROW(num_bytes <= boost::asio::buffer_size(_buff)); - - _lut->length = num_bytes; - _lut->actual_length = 0; - + void commit(libusb_transfer *lut, size_t num_bytes){ + lut->length = num_bytes; try{ - _endpoint->submit(_lut); - _committed = true; - return num_bytes; + _send_ep->submit(lut); } catch(const std::exception &e){ std::cerr << "Error in commit: " << e.what() << std::endl; - return -1; } } -private: - const boost::asio::mutable_buffer &get(void) const{ - return _buff; - } - - libusb_transfer *_lut; - usb_endpoint::sptr _endpoint; - const boost::asio::mutable_buffer _buff; - bool _committed; -}; - - -/*********************************************************************** - * USB zero_copy device class - **********************************************************************/ -class libusb_zero_copy_impl : public usb_zero_copy -{ -private: libusb::device_handle::sptr _handle; + size_t _recv_xfer_size, _send_xfer_size; size_t _recv_num_frames, _send_num_frames; usb_endpoint::sptr _recv_ep, _send_ep; - -public: - typedef boost::shared_ptr sptr; - - libusb_zero_copy_impl( - libusb::device_handle::sptr handle, - unsigned int recv_endpoint, unsigned int send_endpoint, - size_t recv_xfer_size, size_t recv_num_xfers, - size_t send_xfer_size, size_t send_num_xfers - ); - - managed_recv_buffer::sptr get_recv_buff(double); - managed_send_buffer::sptr get_send_buff(double); - - size_t get_num_recv_frames(void) const { return _recv_num_frames; } - size_t get_num_send_frames(void) const { return _send_num_frames; } }; /* @@ -426,7 +339,9 @@ libusb_zero_copy_impl::libusb_zero_copy_impl( UHD_ASSERT_THROW(send_xfer_size % 512 == 0); //store the num xfers for the num frames count + _recv_xfer_size = recv_xfer_size; _recv_num_frames = recv_num_xfers; + _send_xfer_size = send_xfer_size; _send_num_frames = send_num_xfers; _handle->claim_interface(2 /*in interface*/); @@ -461,9 +376,10 @@ managed_recv_buffer::sptr libusb_zero_copy_impl::get_recv_buff(double timeout){ return managed_recv_buffer::sptr(); } else { - return managed_recv_buffer::sptr( - new libusb_managed_recv_buffer_impl(lut, - _recv_ep)); + return managed_recv_buffer::make_safe( + boost::asio::const_buffer(lut->buffer, lut->actual_length), + boost::bind(&libusb_zero_copy_impl::release, this, lut) + ); } } @@ -480,9 +396,10 @@ managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(double timeout){ return managed_send_buffer::sptr(); } else { - return managed_send_buffer::sptr( - new libusb_managed_send_buffer_impl(lut, - _send_ep)); + return managed_send_buffer::make_safe( + boost::asio::mutable_buffer(lut->buffer, _send_xfer_size), + boost::bind(&libusb_zero_copy_impl::commit, this, lut, _1) + ); } } diff --git a/host/lib/transport/vrt_packet_handler.hpp b/host/lib/transport/vrt_packet_handler.hpp index e11afff30..939517411 100644 --- a/host/lib/transport/vrt_packet_handler.hpp +++ b/host/lib/transport/vrt_packet_handler.hpp @@ -343,9 +343,7 @@ template UHD_INLINE T get_context_code( //commit the samples to the zero-copy interface size_t num_bytes_total = (vrt_header_offset_words32+if_packet_info.num_packet_words32)*sizeof(boost::uint32_t); - if (send_buffs[i]->commit(num_bytes_total) < ssize_t(num_bytes_total)){ - std::cerr << "commit to send buffer returned less than commit size" << std::endl; - } + send_buffs[i]->commit(num_bytes_total); } return num_samps; } diff --git a/host/lib/transport/zero_copy.cpp b/host/lib/transport/zero_copy.cpp index dfb65951f..a5a864a04 100644 --- a/host/lib/transport/zero_copy.cpp +++ b/host/lib/transport/zero_copy.cpp @@ -16,32 +16,35 @@ // #include -#include -#include -#include using namespace uhd::transport; /*********************************************************************** - * The pure-virtual deconstructor needs an implementation to be happy + * Safe managed receive buffer **********************************************************************/ -managed_recv_buffer::~managed_recv_buffer(void){ +static void release_nop(void){ /* NOP */ } -/*********************************************************************** - * Phony zero-copy recv interface implementation - **********************************************************************/ - -//! phony zero-copy recv buffer implementation -class managed_recv_buffer_impl : public managed_recv_buffer{ +class safe_managed_receive_buffer : public managed_recv_buffer{ public: - managed_recv_buffer_impl(const boost::asio::const_buffer &buff) : _buff(buff){ + safe_managed_receive_buffer( + const boost::asio::const_buffer &buff, + const release_fcn_t &release_fcn + ): + _buff(buff), _release_fcn(release_fcn) + { /* NOP */ } - ~managed_recv_buffer_impl(void){ - delete [] this->cast(); + ~safe_managed_receive_buffer(void){ + _release_fcn(); + } + + void release(void){ + release_fcn_t release_fcn = _release_fcn; + _release_fcn = &release_nop; + return release_fcn(); } private: @@ -50,64 +53,42 @@ private: } const boost::asio::const_buffer _buff; + release_fcn_t _release_fcn; }; -//! phony zero-copy recv interface implementation -struct phony_zero_copy_recv_if::impl{ - impl(size_t max_buff_size) : max_buff_size(max_buff_size){ - /* NOP */ - } - size_t max_buff_size; -}; - -phony_zero_copy_recv_if::phony_zero_copy_recv_if(size_t max_buff_size){ - _impl = UHD_PIMPL_MAKE(impl, (max_buff_size)); -} - -phony_zero_copy_recv_if::~phony_zero_copy_recv_if(void){ - /* NOP */ -} - -managed_recv_buffer::sptr phony_zero_copy_recv_if::get_recv_buff(double timeout){ - //allocate memory - boost::uint8_t *recv_mem = new boost::uint8_t[_impl->max_buff_size]; - - //call recv() with timeout option - ssize_t num_bytes = this->recv(boost::asio::buffer(recv_mem, _impl->max_buff_size), timeout); - - if (num_bytes <= 0) return managed_recv_buffer::sptr(); //NULL sptr - - //create a new managed buffer to house the data - return managed_recv_buffer::sptr( - new managed_recv_buffer_impl(boost::asio::buffer(recv_mem, num_bytes)) - ); +managed_recv_buffer::sptr managed_recv_buffer::make_safe( + const boost::asio::const_buffer &buff, + const release_fcn_t &release_fcn +){ + return sptr(new safe_managed_receive_buffer(buff, release_fcn)); } /*********************************************************************** - * Phony zero-copy send interface implementation + * Safe managed send buffer **********************************************************************/ +static void commit_nop(size_t){ + /* NOP */ +} -//! phony zero-copy send buffer implementation -class managed_send_buffer_impl : public managed_send_buffer{ +class safe_managed_send_buffer : public managed_send_buffer{ public: - typedef boost::function send_fcn_t; - - managed_send_buffer_impl( + safe_managed_send_buffer( const boost::asio::mutable_buffer &buff, - const send_fcn_t &send_fcn + const commit_fcn_t &commit_fcn ): - _buff(buff), - _send_fcn(send_fcn) + _buff(buff), _commit_fcn(commit_fcn) { /* NOP */ } - ~managed_send_buffer_impl(void){ - /* NOP */ + ~safe_managed_send_buffer(void){ + _commit_fcn(0); } - ssize_t commit(size_t num_bytes){ - return _send_fcn(boost::asio::buffer(_buff, num_bytes)); + void commit(size_t num_bytes){ + commit_fcn_t commit_fcn = _commit_fcn; + _commit_fcn = &commit_nop; + return commit_fcn(num_bytes); } private: @@ -116,28 +97,12 @@ private: } const boost::asio::mutable_buffer _buff; - const send_fcn_t _send_fcn; -}; - -//! phony zero-copy send interface implementation -struct phony_zero_copy_send_if::impl{ - boost::uint8_t *send_mem; - managed_send_buffer::sptr send_buff; + commit_fcn_t _commit_fcn; }; -phony_zero_copy_send_if::phony_zero_copy_send_if(size_t max_buff_size){ - _impl = UHD_PIMPL_MAKE(impl, ()); - _impl->send_mem = new boost::uint8_t[max_buff_size]; - _impl->send_buff = managed_send_buffer::sptr(new managed_send_buffer_impl( - boost::asio::buffer(_impl->send_mem, max_buff_size), - boost::bind(&phony_zero_copy_send_if::send, this, _1) - )); -} - -phony_zero_copy_send_if::~phony_zero_copy_send_if(void){ - delete [] _impl->send_mem; -} - -managed_send_buffer::sptr phony_zero_copy_send_if::get_send_buff(double){ - return _impl->send_buff; //FIXME there is only ever one send buff, we assume that the caller doesnt hang onto these +safe_managed_send_buffer::sptr managed_send_buffer::make_safe( + const boost::asio::mutable_buffer &buff, + const commit_fcn_t &commit_fcn +){ + return sptr(new safe_managed_send_buffer(buff, commit_fcn)); } diff --git a/host/lib/usrp/usrp1/io_impl.cpp b/host/lib/usrp/usrp1/io_impl.cpp index 8d9c68961..676b1536a 100644 --- a/host/lib/usrp/usrp1/io_impl.cpp +++ b/host/lib/usrp/usrp1/io_impl.cpp @@ -33,35 +33,6 @@ using namespace uhd::usrp; using namespace uhd::transport; namespace asio = boost::asio; -/*********************************************************************** - * Pseudo send buffer implementation - **********************************************************************/ -class pseudo_managed_send_buffer : public managed_send_buffer{ -public: - - pseudo_managed_send_buffer( - const boost::asio::mutable_buffer &buff, - const boost::function &commit - ): - _buff(buff), - _commit(commit) - { - /* NOP */ - } - - ssize_t commit(size_t num_bytes){ - return _commit(num_bytes); - } - -private: - const boost::asio::mutable_buffer &get(void) const{ - return _buff; - } - - const boost::asio::mutable_buffer _buff; - const boost::function _commit; -}; - /*********************************************************************** * IO Implementation Details **********************************************************************/ @@ -96,9 +67,9 @@ struct usrp1_impl::io_impl{ size_t num_bytes_committed; double send_timeout; boost::uint8_t pseudo_buff[BYTES_PER_PACKET]; - ssize_t phony_commit_pseudo_buff(size_t num_bytes); - ssize_t phony_commit_send_buff(size_t num_bytes); - ssize_t commit_send_buff(void); + void phony_commit_pseudo_buff(size_t num_bytes); + void phony_commit_send_buff(size_t num_bytes); + void commit_send_buff(void); void flush_send_buff(void); bool get_send_buffs(vrt_packet_handler::managed_send_buffs_t &, double); @@ -119,28 +90,25 @@ struct usrp1_impl::io_impl{ * The first loop iteration will fill the remainder of the send buffer. * The second loop iteration will empty the pseudo buffer remainder. */ -ssize_t usrp1_impl::io_impl::phony_commit_pseudo_buff(size_t num_bytes){ +void usrp1_impl::io_impl::phony_commit_pseudo_buff(size_t num_bytes){ size_t bytes_to_copy = num_bytes, bytes_copied = 0; while(bytes_to_copy){ size_t bytes_copied_here = std::min(bytes_to_copy, get_send_mem_size()); std::memcpy(get_send_mem_ptr(), pseudo_buff + bytes_copied, bytes_copied_here); - ssize_t ret = phony_commit_send_buff(bytes_copied_here); - if (ret < 0) return ret; - bytes_to_copy -= ret; - bytes_copied += ret; + phony_commit_send_buff(bytes_copied_here); + bytes_to_copy -= bytes_copied_here; + bytes_copied += bytes_copied_here; } - return bytes_copied; } /*! * Accept a commit of num bytes to the send buffer. * Conditionally commit the send buffer if full. */ -ssize_t usrp1_impl::io_impl::phony_commit_send_buff(size_t num_bytes){ +void usrp1_impl::io_impl::phony_commit_send_buff(size_t num_bytes){ num_bytes_committed += num_bytes; - if (num_bytes_committed != send_buff->size()) return num_bytes; - ssize_t ret = commit_send_buff(); - return (ret < 0)? ret : num_bytes; + if (num_bytes_committed != send_buff->size()) return; + commit_send_buff(); } /*! @@ -158,11 +126,10 @@ void usrp1_impl::io_impl::flush_send_buff(void){ * Perform an actual commit on the send buffer: * Commit the contents of the send buffer and request a new buffer. */ -ssize_t usrp1_impl::io_impl::commit_send_buff(void){ - ssize_t ret = send_buff->commit(num_bytes_committed); +void usrp1_impl::io_impl::commit_send_buff(void){ + send_buff->commit(num_bytes_committed); send_buff = data_transport->get_send_buff(send_timeout); num_bytes_committed = 0; - return ret; } bool usrp1_impl::io_impl::get_send_buffs( @@ -173,17 +140,17 @@ bool usrp1_impl::io_impl::get_send_buffs( //not enough bytes free -> use the pseudo buffer if (get_send_mem_size() < BYTES_PER_PACKET){ - buffs[0] = managed_send_buffer::sptr(new pseudo_managed_send_buffer( + buffs[0] = managed_send_buffer::make_safe( boost::asio::buffer(pseudo_buff), boost::bind(&usrp1_impl::io_impl::phony_commit_pseudo_buff, this, _1) - )); + ); } //otherwise use the send buffer offset by the bytes written else{ - buffs[0] = managed_send_buffer::sptr(new pseudo_managed_send_buffer( + buffs[0] = managed_send_buffer::make_safe( boost::asio::buffer(get_send_mem_ptr(), get_send_mem_size()), boost::bind(&usrp1_impl::io_impl::phony_commit_send_buff, this, _1) - )); + ); } return buffs[0].get() != NULL; -- cgit v1.2.3 From b57c84b34bcdd6c66eb053695b83e6bd6c481774 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Sun, 3 Oct 2010 01:41:24 -0700 Subject: uhd: implemented udp zero copy asio with async calls --- host/lib/transport/CMakeLists.txt | 2 +- host/lib/transport/libusb1_zero_copy.cpp | 6 +- host/lib/transport/udp_zero_copy_asio.cpp | 186 +++++++++++++++++++++--------- 3 files changed, 133 insertions(+), 61 deletions(-) (limited to 'host/lib') diff --git a/host/lib/transport/CMakeLists.txt b/host/lib/transport/CMakeLists.txt index 2be2c89ec..bf92b0822 100644 --- a/host/lib/transport/CMakeLists.txt +++ b/host/lib/transport/CMakeLists.txt @@ -101,7 +101,7 @@ SET_SOURCE_FILES_PROPERTIES( LIBUHD_APPEND_SOURCES( ${CMAKE_SOURCE_DIR}/lib/transport/if_addrs.cpp ${CMAKE_SOURCE_DIR}/lib/transport/udp_simple.cpp - #${CMAKE_SOURCE_DIR}/lib/transport/udp_zero_copy_asio.cpp + ${CMAKE_SOURCE_DIR}/lib/transport/udp_zero_copy_asio.cpp ${CMAKE_SOURCE_DIR}/lib/transport/vrt_packet_handler.hpp ${CMAKE_SOURCE_DIR}/lib/transport/zero_copy.cpp ) diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index e1cc8398c..c819302b6 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -277,7 +277,7 @@ libusb_transfer *usb_endpoint::get_lut_with_wait(double timeout){ /*********************************************************************** * USB zero_copy device class **********************************************************************/ -class libusb_zero_copy_impl : public usb_zero_copy { +class libusb_zero_copy_impl : public usb_zero_copy, public boost::enable_shared_from_this { public: typedef boost::shared_ptr sptr; @@ -378,7 +378,7 @@ managed_recv_buffer::sptr libusb_zero_copy_impl::get_recv_buff(double timeout){ else { return managed_recv_buffer::make_safe( boost::asio::const_buffer(lut->buffer, lut->actual_length), - boost::bind(&libusb_zero_copy_impl::release, this, lut) + boost::bind(&libusb_zero_copy_impl::release, shared_from_this(), lut) ); } } @@ -398,7 +398,7 @@ managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(double timeout){ else { return managed_send_buffer::make_safe( boost::asio::mutable_buffer(lut->buffer, _send_xfer_size), - boost::bind(&libusb_zero_copy_impl::commit, this, lut, _1) + boost::bind(&libusb_zero_copy_impl::commit, shared_from_this(), lut, _1) ); } } diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 3130830a5..70e7514a1 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -17,20 +17,23 @@ #include #include //mtu +#include #include #include -#include +#include #include #include +#include #include using namespace uhd::transport; +namespace asio = boost::asio; /*********************************************************************** * Constants **********************************************************************/ //enough buffering for half a second of samples at full rate on usrp2 -static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(sizeof(boost::uint32_t) * 25e6 * 0.5); +static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(4 * 25e6 * 0.5); //Large buffers cause more underflow at high rates. //Perhaps this is due to the kernel scheduling, //but may change with host-based flow control. @@ -43,39 +46,55 @@ static const size_t MIN_SEND_SOCK_BUFF_SIZE = size_t(10e3); * However, it is not a true zero copy implementation as each * send and recv requires a copy operation to/from userspace. **********************************************************************/ -class udp_zero_copy_impl: - public phony_zero_copy_recv_if, - public phony_zero_copy_send_if, - public udp_zero_copy -{ +class udp_zero_copy_asio_impl : public udp_zero_copy, public boost::enable_shared_from_this { public: - typedef boost::shared_ptr sptr; - - udp_zero_copy_impl( - const std::string &addr, - const std::string &port - ): - phony_zero_copy_recv_if(udp_simple::mtu), - phony_zero_copy_send_if(udp_simple::mtu) - { + typedef boost::shared_ptr sptr; + + udp_zero_copy_asio_impl(const std::string &addr, const std::string &port){ //std::cout << boost::format("Creating udp transport for %s %s") % addr % port << std::endl; // resolve the address - boost::asio::ip::udp::resolver resolver(_io_service); - boost::asio::ip::udp::resolver::query query(boost::asio::ip::udp::v4(), addr, port); - boost::asio::ip::udp::endpoint receiver_endpoint = *resolver.resolve(query); + asio::ip::udp::resolver resolver(_io_service); + asio::ip::udp::resolver::query query(asio::ip::udp::v4(), addr, port); + asio::ip::udp::endpoint receiver_endpoint = *resolver.resolve(query); // create, open, and connect the socket - _socket = new boost::asio::ip::udp::socket(_io_service); - _socket->open(boost::asio::ip::udp::v4()); + _socket = new asio::ip::udp::socket(_io_service); + _socket->open(asio::ip::udp::v4()); _socket->connect(receiver_endpoint); - _sock_fd = _socket->native(); } - ~udp_zero_copy_impl(void){ + ~udp_zero_copy_asio_impl(void){ + _io_service.stop(); + _thread_group.join_all(); delete _socket; } + /*! + * Init, the second contructor: + * Allocate memory and spwan service thread. + */ + void init(void){ + //allocate all recv frames and release them to begin xfers + _pending_recv_buffs = pending_buffs_type::make(this->get_num_recv_frames()); + for (size_t i = 0; i < this->get_num_recv_frames(); i++){ + boost::shared_array buff(new char[udp_simple::mtu]); + _buffers.push_back(buff); //store a reference to this shared array + release(buff.get()); + } + + //allocate all send frames and push them into the fifo + _pending_send_buffs = pending_buffs_type::make(this->get_num_send_frames()); + for (size_t i = 0; i < this->get_num_send_frames(); i++){ + boost::shared_array buff(new char[udp_simple::mtu]); + _buffers.push_back(buff); //store a reference to this shared array + handle_send(buff.get()); + } + + //spawn the service thread that will run the io service + _thread_group.create_thread(boost::bind(&udp_zero_copy_asio_impl::service, this)); + } + //get size for internal socket buffer template size_t get_buff_size(void) const{ Opt option; @@ -90,60 +109,111 @@ public: return get_buff_size(); } - //The number of frames is approximately the buffer size divided by the max datagram size. //In reality, this is a phony zero-copy interface and the number of frames is infinite. //However, its sensible to advertise a frame count that is approximate to buffer size. //This way, the transport caller will have an idea about how much buffering to create. size_t get_num_recv_frames(void) const{ - return this->get_buff_size()/udp_simple::mtu; + return this->get_buff_size()/udp_simple::mtu; } size_t get_num_send_frames(void) const{ - return this->get_buff_size()/udp_simple::mtu; + return this->get_buff_size()/udp_simple::mtu; + } + + //! pop a filled recv buffer off of the fifo and bind with the release callback + managed_recv_buffer::sptr get_recv_buff(double timeout){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + asio::mutable_buffer buff; + if (_pending_recv_buffs->pop_with_timed_wait(buff, timeout)){ + return managed_recv_buffer::make_safe( + buff, boost::bind( + &udp_zero_copy_asio_impl::release, + shared_from_this(), + asio::buffer_cast(buff) + ) + ); + } + return managed_recv_buffer::sptr(); + } + + //! pop an empty send buffer off of the fifo and bind with the commit callback + managed_send_buffer::sptr get_send_buff(double timeout){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + asio::mutable_buffer buff; + if (_pending_send_buffs->pop_with_timed_wait(buff, timeout)){ + return managed_send_buffer::make_safe( + buff, boost::bind( + &udp_zero_copy_asio_impl::commit, + shared_from_this(), + asio::buffer_cast(buff), _1 + ) + ); + } + return managed_send_buffer::sptr(); } private: - boost::asio::ip::udp::socket *_socket; - boost::asio::io_service _io_service; - int _sock_fd; - - ssize_t recv(const boost::asio::mutable_buffer &buff, double timeout){ - //setup timeval for timeout - timeval tv; - tv.tv_sec = 0; - tv.tv_usec = long(timeout*1e6); - - //setup rset for timeout - fd_set rset; - FD_ZERO(&rset); - FD_SET(_sock_fd, &rset); - - //call select to perform timed wait - if (::select(_sock_fd+1, &rset, NULL, NULL, &tv) <= 0) return 0; - - return ::recv( - _sock_fd, - boost::asio::buffer_cast(buff), - boost::asio::buffer_size(buff), 0 + void service(void){ + _io_service.run(); + } + + /******************************************************************* + * The async send and receive callbacks + ******************************************************************/ + + //! handle a recv callback -> push the filled memory into the fifo + void handle_recv(void *mem, size_t len){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + _pending_recv_buffs->push_with_wait(boost::asio::buffer(mem, len)); + } + + //! release a recv buffer -> start an async recv on the buffer + void release(void *mem){ + _socket->async_receive( + boost::asio::buffer(mem, udp_simple::mtu), + boost::bind( + &udp_zero_copy_asio_impl::handle_recv, + shared_from_this(), mem, + asio::placeholders::bytes_transferred + ) ); } - ssize_t send(const boost::asio::const_buffer &buff){ - return ::send( - _sock_fd, - boost::asio::buffer_cast(buff), - boost::asio::buffer_size(buff), 0 + //! handle a send callback -> push the emptied memory into the fifo + void handle_send(void *mem){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + _pending_send_buffs->push_with_wait(boost::asio::buffer(mem, udp_simple::mtu)); + } + + //! commit a send buffer -> start an async send on the buffer + void commit(void *mem, size_t len){ + _socket->async_send( + boost::asio::buffer(mem, len), + boost::bind( + &udp_zero_copy_asio_impl::handle_send, + shared_from_this(), mem + ) ); } + + //asio guts -> socket and service + asio::ip::udp::socket *_socket; + asio::io_service _io_service; + + //memory management -> buffers and fifos + boost::thread_group _thread_group; + std::vector > _buffers; + typedef bounded_buffer pending_buffs_type; + pending_buffs_type::sptr _pending_recv_buffs, _pending_send_buffs; }; /*********************************************************************** * UDP zero copy make function **********************************************************************/ template static void resize_buff_helper( - udp_zero_copy_impl::sptr udp_trans, + udp_zero_copy_asio_impl::sptr udp_trans, size_t target_size, const std::string &name ){ @@ -183,11 +253,13 @@ udp_zero_copy::sptr udp_zero_copy::make( size_t recv_buff_size, size_t send_buff_size ){ - udp_zero_copy_impl::sptr udp_trans(new udp_zero_copy_impl(addr, port)); + udp_zero_copy_asio_impl::sptr udp_trans(new udp_zero_copy_asio_impl(addr, port)); //call the helper to resize send and recv buffers - resize_buff_helper(udp_trans, recv_buff_size, "recv"); - resize_buff_helper (udp_trans, send_buff_size, "send"); + resize_buff_helper(udp_trans, recv_buff_size, "recv"); + resize_buff_helper (udp_trans, send_buff_size, "send"); + + udp_trans->init(); //buffers resized -> call init() to use return udp_trans; } -- cgit v1.2.3 From c6f17f1e2f908747ae1547fa43b2c22c3c20ba50 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Sun, 3 Oct 2010 19:34:41 -0700 Subject: uhd: changed buffer allocations to be in a single chunk, udp: pass frame sizes into the impl constructor --- host/lib/transport/libusb1_zero_copy.cpp | 27 +++++----- host/lib/transport/udp_zero_copy_asio.cpp | 85 ++++++++++++++++--------------- 2 files changed, 59 insertions(+), 53 deletions(-) (limited to 'host/lib') diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index c819302b6..ab48e4fc4 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -100,13 +100,13 @@ private: lut_buff_type::sptr _completed_list; //! a list of all transfer structs we allocated - std::vector _all_luts; + std::vector _all_luts; - //! a list of shared arrays for the transfer buffers - std::vector > _buffers; + //! a block of memory for the transfer buffers + boost::shared_array _buffer; // Calls for processing asynchronous I/O - libusb_transfer *allocate_transfer(int buff_len); + libusb_transfer *allocate_transfer(void *mem, size_t len); void print_transfer_status(libusb_transfer *lut); }; @@ -154,9 +154,9 @@ usb_endpoint::usb_endpoint( _input(input) { _completed_list = lut_buff_type::make(num_transfers); - + _buffer = boost::shared_array(new char[num_transfers*transfer_size]); for (size_t i = 0; i < num_transfers; i++){ - _all_luts.push_back(allocate_transfer(transfer_size)); + _all_luts.push_back(allocate_transfer(_buffer.get() + i*transfer_size, transfer_size)); //input luts are immediately submitted to be filled //output luts go into the completed list as free buffers @@ -193,23 +193,23 @@ usb_endpoint::~usb_endpoint(void){ * Allocate a libusb transfer * The allocated transfer - and buffer it contains - is repeatedly * submitted, reaped, and reused and should not be freed until shutdown. - * \param buff_len size of the individual buffer held by each transfer + * \param mem a pointer to the buffer memory + * \param len size of the individual buffer * \return pointer to an allocated libusb_transfer */ -libusb_transfer *usb_endpoint::allocate_transfer(int buff_len){ +libusb_transfer *usb_endpoint::allocate_transfer(void *mem, size_t len){ libusb_transfer *lut = libusb_alloc_transfer(0); - - boost::shared_array buff(new boost::uint8_t[buff_len]); - _buffers.push_back(buff); //store a reference to this shared array + UHD_ASSERT_THROW(lut != NULL); unsigned int endpoint = ((_endpoint & 0x7f) | (_input ? 0x80 : 0)); + unsigned char *buff = reinterpret_cast(mem); libusb_transfer_cb_fn lut_callback = libusb_transfer_cb_fn(&callback); libusb_fill_bulk_transfer(lut, // transfer _handle->get(), // dev_handle endpoint, // endpoint - buff.get(), // buffer - buff_len, // length + buff, // buffer + len, // length lut_callback, // callback this, // user_data 0); // timeout @@ -232,6 +232,7 @@ void usb_endpoint::submit(libusb_transfer *lut){ * \param lut pointer to an libusb_transfer */ void usb_endpoint::print_transfer_status(libusb_transfer *lut){ + std::cout << "here " << lut->status << std::endl; switch (lut->status) { case LIBUSB_TRANSFER_COMPLETED: if (lut->actual_length < lut->length) { diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 70e7514a1..a1eb516fc 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -18,6 +18,7 @@ #include #include //mtu #include +#include #include #include #include @@ -26,6 +27,7 @@ #include #include +using namespace uhd; using namespace uhd::transport; namespace asio = boost::asio; @@ -34,11 +36,15 @@ namespace asio = boost::asio; **********************************************************************/ //enough buffering for half a second of samples at full rate on usrp2 static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(4 * 25e6 * 0.5); + //Large buffers cause more underflow at high rates. //Perhaps this is due to the kernel scheduling, //but may change with host-based flow control. static const size_t MIN_SEND_SOCK_BUFF_SIZE = size_t(10e3); +//the number of async frames to allocate for each send and recv +static const size_t DEFAULT_NUM_ASYNC_FRAMES = 32; + /*********************************************************************** * Zero Copy UDP implementation with ASIO: * This is the portable zero copy implementation for systems @@ -50,51 +56,52 @@ class udp_zero_copy_asio_impl : public udp_zero_copy, public boost::enable_share public: typedef boost::shared_ptr sptr; - udp_zero_copy_asio_impl(const std::string &addr, const std::string &port){ + udp_zero_copy_asio_impl( + const std::string &addr, const std::string &port, + size_t recv_frame_size, size_t num_recv_frames, + size_t send_frame_size, size_t num_send_frames + ): + _recv_frame_size(recv_frame_size), _num_recv_frames(num_recv_frames), + _send_frame_size(send_frame_size), _num_send_frames(num_send_frames) + { //std::cout << boost::format("Creating udp transport for %s %s") % addr % port << std::endl; - // resolve the address + //resolve the address asio::ip::udp::resolver resolver(_io_service); asio::ip::udp::resolver::query query(asio::ip::udp::v4(), addr, port); asio::ip::udp::endpoint receiver_endpoint = *resolver.resolve(query); - // create, open, and connect the socket + //create, open, and connect the socket _socket = new asio::ip::udp::socket(_io_service); _socket->open(asio::ip::udp::v4()); _socket->connect(receiver_endpoint); } - ~udp_zero_copy_asio_impl(void){ - _io_service.stop(); - _thread_group.join_all(); - delete _socket; - } - - /*! - * Init, the second contructor: - * Allocate memory and spwan service thread. - */ void init(void){ //allocate all recv frames and release them to begin xfers - _pending_recv_buffs = pending_buffs_type::make(this->get_num_recv_frames()); - for (size_t i = 0; i < this->get_num_recv_frames(); i++){ - boost::shared_array buff(new char[udp_simple::mtu]); - _buffers.push_back(buff); //store a reference to this shared array - release(buff.get()); + _pending_recv_buffs = pending_buffs_type::make(_num_recv_frames); + _recv_buffer = boost::shared_array(new char[_num_recv_frames*_recv_frame_size]); + for (size_t i = 0; i < _num_recv_frames; i++){ + release(_recv_buffer.get() + i*_recv_frame_size); } //allocate all send frames and push them into the fifo - _pending_send_buffs = pending_buffs_type::make(this->get_num_send_frames()); - for (size_t i = 0; i < this->get_num_send_frames(); i++){ - boost::shared_array buff(new char[udp_simple::mtu]); - _buffers.push_back(buff); //store a reference to this shared array - handle_send(buff.get()); + _pending_send_buffs = pending_buffs_type::make(_num_send_frames); + _send_buffer = boost::shared_array(new char[_num_send_frames*_send_frame_size]); + for (size_t i = 0; i < _num_send_frames; i++){ + handle_send(_send_buffer.get() + i*_send_frame_size); } //spawn the service thread that will run the io service _thread_group.create_thread(boost::bind(&udp_zero_copy_asio_impl::service, this)); } + ~udp_zero_copy_asio_impl(void){ + _io_service.stop(); + _thread_group.join_all(); + delete _socket; + } + //get size for internal socket buffer template size_t get_buff_size(void) const{ Opt option; @@ -109,19 +116,6 @@ public: return get_buff_size(); } - //The number of frames is approximately the buffer size divided by the max datagram size. - //In reality, this is a phony zero-copy interface and the number of frames is infinite. - //However, its sensible to advertise a frame count that is approximate to buffer size. - //This way, the transport caller will have an idea about how much buffering to create. - - size_t get_num_recv_frames(void) const{ - return this->get_buff_size()/udp_simple::mtu; - } - - size_t get_num_send_frames(void) const{ - return this->get_buff_size()/udp_simple::mtu; - } - //! pop a filled recv buffer off of the fifo and bind with the release callback managed_recv_buffer::sptr get_recv_buff(double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw @@ -138,6 +132,8 @@ public: return managed_recv_buffer::sptr(); } + size_t get_num_recv_frames(void) const {return _num_recv_frames;} + //! pop an empty send buffer off of the fifo and bind with the commit callback managed_send_buffer::sptr get_send_buff(double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw @@ -154,8 +150,11 @@ public: return managed_send_buffer::sptr(); } + size_t get_num_send_frames(void) const {return _num_send_frames;} + private: void service(void){ + set_thread_priority_safe(); _io_service.run(); } @@ -172,7 +171,7 @@ private: //! release a recv buffer -> start an async recv on the buffer void release(void *mem){ _socket->async_receive( - boost::asio::buffer(mem, udp_simple::mtu), + boost::asio::buffer(mem, _recv_frame_size), boost::bind( &udp_zero_copy_asio_impl::handle_recv, shared_from_this(), mem, @@ -184,7 +183,7 @@ private: //! handle a send callback -> push the emptied memory into the fifo void handle_send(void *mem){ boost::this_thread::disable_interruption di; //disable because the wait can throw - _pending_send_buffs->push_with_wait(boost::asio::buffer(mem, udp_simple::mtu)); + _pending_send_buffs->push_with_wait(boost::asio::buffer(mem, _send_frame_size)); } //! commit a send buffer -> start an async send on the buffer @@ -204,9 +203,11 @@ private: //memory management -> buffers and fifos boost::thread_group _thread_group; - std::vector > _buffers; + boost::shared_array _send_buffer, _recv_buffer; typedef bounded_buffer pending_buffs_type; pending_buffs_type::sptr _pending_recv_buffs, _pending_send_buffs; + const size_t _recv_frame_size, _num_recv_frames; + const size_t _send_frame_size, _num_send_frames; }; /*********************************************************************** @@ -253,7 +254,11 @@ udp_zero_copy::sptr udp_zero_copy::make( size_t recv_buff_size, size_t send_buff_size ){ - udp_zero_copy_asio_impl::sptr udp_trans(new udp_zero_copy_asio_impl(addr, port)); + udp_zero_copy_asio_impl::sptr udp_trans(new udp_zero_copy_asio_impl( + addr, port, + udp_simple::mtu, DEFAULT_NUM_ASYNC_FRAMES, //recv + udp_simple::mtu, DEFAULT_NUM_ASYNC_FRAMES //send + )); //call the helper to resize send and recv buffers resize_buff_helper(udp_trans, recv_buff_size, "recv"); -- cgit v1.2.3 From 57ad303e850f143573d8c6edb1a88542b7b43350 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Mon, 4 Oct 2010 09:28:26 -0700 Subject: udp: added io service work to keep service running --- host/lib/transport/udp_zero_copy_asio.cpp | 16 ++++++++++++---- 1 file changed, 12 insertions(+), 4 deletions(-) (limited to 'host/lib') diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index a1eb516fc..2cf7bde18 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -45,6 +45,9 @@ static const size_t MIN_SEND_SOCK_BUFF_SIZE = size_t(10e3); //the number of async frames to allocate for each send and recv static const size_t DEFAULT_NUM_ASYNC_FRAMES = 32; +//a single concurrent thread for io_service seems to be the fastest +static const size_t CONCURRENCY_HINT = 1; + /*********************************************************************** * Zero Copy UDP implementation with ASIO: * This is the portable zero copy implementation for systems @@ -61,6 +64,8 @@ public: size_t recv_frame_size, size_t num_recv_frames, size_t send_frame_size, size_t num_send_frames ): + _io_service(CONCURRENCY_HINT), + _work(new asio::io_service::work(_io_service)), _recv_frame_size(recv_frame_size), _num_recv_frames(num_recv_frames), _send_frame_size(send_frame_size), _num_send_frames(num_send_frames) { @@ -92,13 +97,15 @@ public: handle_send(_send_buffer.get() + i*_send_frame_size); } - //spawn the service thread that will run the io service - _thread_group.create_thread(boost::bind(&udp_zero_copy_asio_impl::service, this)); + //spawn the service threads that will run the io service + for (size_t i = 0; i < CONCURRENCY_HINT; i++) _thread_group.create_thread( + boost::bind(&udp_zero_copy_asio_impl::service, this) + ); } ~udp_zero_copy_asio_impl(void){ - _io_service.stop(); - _thread_group.join_all(); + delete _work; //allow io_service run to complete + _thread_group.join_all(); //wait for service threads to exit delete _socket; } @@ -200,6 +207,7 @@ private: //asio guts -> socket and service asio::ip::udp::socket *_socket; asio::io_service _io_service; + asio::io_service::work *_work; //memory management -> buffers and fifos boost::thread_group _thread_group; -- cgit v1.2.3 From 4fd922ba3acb8559a3d7d016b230d5e7a0a84da1 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Mon, 4 Oct 2010 09:48:25 -0700 Subject: usrp2: removed extra print-outs in init --- host/lib/usrp/usrp2/io_impl.cpp | 4 ---- 1 file changed, 4 deletions(-) (limited to 'host/lib') diff --git a/host/lib/usrp/usrp2/io_impl.cpp b/host/lib/usrp/usrp2/io_impl.cpp index c0d8ab029..33cec3cbc 100644 --- a/host/lib/usrp/usrp2/io_impl.cpp +++ b/host/lib/usrp/usrp2/io_impl.cpp @@ -168,10 +168,6 @@ void usrp2_impl::io_init(void){ _mboards.at(i), i )); } - - std::cout << "RX samples per packet: " << get_max_recv_samps_per_packet() << std::endl; - std::cout << "TX samples per packet: " << get_max_send_samps_per_packet() << std::endl; - std::cout << "Recv pirate num frames: " << num_frames << std::endl; } /*********************************************************************** -- cgit v1.2.3 From 219183453b23a572ddd6eeec81c3bc1907754560 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Mon, 4 Oct 2010 11:30:14 -0700 Subject: uhd: added include for enable_shared_from_this when used --- host/lib/transport/libusb1_zero_copy.cpp | 2 +- host/lib/transport/udp_zero_copy_asio.cpp | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) (limited to 'host/lib') diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index ab48e4fc4..c7f084f62 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -22,6 +22,7 @@ #include #include #include +#include #include #include @@ -280,7 +281,6 @@ libusb_transfer *usb_endpoint::get_lut_with_wait(double timeout){ **********************************************************************/ class libusb_zero_copy_impl : public usb_zero_copy, public boost::enable_shared_from_this { public: - typedef boost::shared_ptr sptr; libusb_zero_copy_impl( libusb::device_handle::sptr handle, diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 2cf7bde18..e9d91fe45 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -25,6 +25,7 @@ #include #include #include +#include #include using namespace uhd; -- cgit v1.2.3 From 5bd863efa05f3c6cfd672cffe87f57b33d1c32b5 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Mon, 4 Oct 2010 14:54:16 -0700 Subject: uhd: bounded buffer pop sets back element to empty to decrement references --- host/include/uhd/transport/bounded_buffer.ipp | 16 ++++++++++++++-- host/lib/transport/libusb1_zero_copy.cpp | 4 ---- 2 files changed, 14 insertions(+), 6 deletions(-) (limited to 'host/lib') diff --git a/host/include/uhd/transport/bounded_buffer.ipp b/host/include/uhd/transport/bounded_buffer.ipp index 71143741e..58f78bab4 100644 --- a/host/include/uhd/transport/bounded_buffer.ipp +++ b/host/include/uhd/transport/bounded_buffer.ipp @@ -73,7 +73,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ UHD_INLINE void pop_with_wait(elem_type &elem){ boost::unique_lock lock(_mutex); _empty_cond.wait(lock, boost::bind(&bounded_buffer_impl::not_empty, this)); - elem = _buffer.back(); _buffer.pop_back(); + this->pop_back(elem); lock.unlock(); _full_cond.notify_one(); } @@ -84,7 +84,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ lock, boost::posix_time::microseconds(long(timeout*1e6)), boost::bind(&bounded_buffer_impl::not_empty, this) )) return false; - elem = _buffer.back(); _buffer.pop_back(); + this->pop_back(elem); lock.unlock(); _full_cond.notify_one(); return true; @@ -104,6 +104,18 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ bool not_full(void) const{return not _buffer.full();} bool not_empty(void) const{return not _buffer.empty();} + + /*! + * Three part operation to pop an element: + * 1) assign elem to the back element + * 2) assign the back element to empty + * 3) pop the back to move the counter + */ + UHD_INLINE void pop_back(elem_type &elem){ + elem = _buffer.back(); + _buffer.back() = elem_type(); + _buffer.pop_back(); + } }; }}} //namespace diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index c7f084f62..819874483 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -335,10 +335,6 @@ libusb_zero_copy_impl::libusb_zero_copy_impl( if (send_xfer_size == 0) send_xfer_size = DEFAULT_XFER_SIZE; if (send_num_xfers == 0) send_num_xfers = DEFAULT_NUM_XFERS; - //sanity check the transfer sizes - UHD_ASSERT_THROW(recv_xfer_size % 512 == 0); - UHD_ASSERT_THROW(send_xfer_size % 512 == 0); - //store the num xfers for the num frames count _recv_xfer_size = recv_xfer_size; _recv_num_frames = recv_num_xfers; -- cgit v1.2.3 From 50ff7e4d3bf4e196b4c054821f67f8d1bff5203b Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Mon, 4 Oct 2010 16:34:24 -0700 Subject: usb: added dummy usb implementation for building without usb (throw, not segfault) --- host/lib/transport/CMakeLists.txt | 4 +++- host/lib/transport/usb_dummy_impl.cpp | 39 +++++++++++++++++++++++++++++++++++ 2 files changed, 42 insertions(+), 1 deletion(-) create mode 100644 host/lib/transport/usb_dummy_impl.cpp (limited to 'host/lib') diff --git a/host/lib/transport/CMakeLists.txt b/host/lib/transport/CMakeLists.txt index bf92b0822..b9ec7a6ad 100644 --- a/host/lib/transport/CMakeLists.txt +++ b/host/lib/transport/CMakeLists.txt @@ -37,7 +37,9 @@ IF(LIBUSB_FOUND) ENDIF(MSVC) SET(HAVE_USB_SUPPORT TRUE) ELSE(LIBUSB_FOUND) - #TODO dummy usb + LIBUHD_APPEND_SOURCES( + ${CMAKE_SOURCE_DIR}/lib/transport/usb_dummy_impl.cpp + ) ENDIF(LIBUSB_FOUND) IF(HAVE_USB_SUPPORT) diff --git a/host/lib/transport/usb_dummy_impl.cpp b/host/lib/transport/usb_dummy_impl.cpp new file mode 100644 index 000000000..518342aba --- /dev/null +++ b/host/lib/transport/usb_dummy_impl.cpp @@ -0,0 +1,39 @@ +// +// Copyright 2010 Ettus Research LLC +// +// This program 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. +// +// This program 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 this program. If not, see . +// + +#include +#include +#include +#include + +using namespace uhd::transport; + +std::vector usb_device_handle::get_device_list(boost::uint16_t, boost::uint16_t){ + return std::vector(); //empty list +} + +usb_control::sptr usb_control::make(usb_device_handle::sptr){ + throw std::runtime_error("no usb support -> usb_control::make not implemented"); +} + +usb_zero_copy::sptr usb_zero_copy::make( + usb_device_handle::sptr, + unsigned int, unsigned int, + size_t, size_t, size_t, size_t +){ + throw std::runtime_error("no usb support -> usb_zero_copy::make not implemented"); +} -- cgit v1.2.3 From 0cd5375b5c8a928f112a963d9c9c2556bafed108 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Tue, 5 Oct 2010 10:30:28 -0700 Subject: uhd: replaced frame params for the zero copy interfaces with a device address the device address gives a key, value pair of infinite optional capabilities added a cast option to the device address to cast string to type T added call to the zero_copy_if to get send and recv frame sizes changed the usrp2 impl to calculate recv/send spp from the data transport --- host/include/uhd/transport/udp_zero_copy.hpp | 7 ++- host/include/uhd/transport/usb_control.hpp | 2 +- host/include/uhd/transport/usb_zero_copy.hpp | 17 +++---- host/include/uhd/transport/zero_copy.hpp | 26 +++++++--- host/include/uhd/types/device_addr.hpp | 20 ++++++++ host/lib/transport/libusb1_zero_copy.cpp | 67 +++++++++++-------------- host/lib/transport/udp_zero_copy_asio.cpp | 35 +++++++------ host/lib/transport/usb_dummy_impl.cpp | 4 +- host/lib/usrp/usrp1/usrp1_impl.cpp | 48 +++++++----------- host/lib/usrp/usrp2/io_impl.cpp | 4 +- host/lib/usrp/usrp2/mboard_impl.cpp | 8 +-- host/lib/usrp/usrp2/usrp2_impl.cpp | 40 ++++++--------- host/lib/usrp/usrp2/usrp2_impl.hpp | 75 ++++++++-------------------- 13 files changed, 162 insertions(+), 191 deletions(-) (limited to 'host/lib') diff --git a/host/include/uhd/transport/udp_zero_copy.hpp b/host/include/uhd/transport/udp_zero_copy.hpp index 818709973..bbba97b21 100644 --- a/host/include/uhd/transport/udp_zero_copy.hpp +++ b/host/include/uhd/transport/udp_zero_copy.hpp @@ -20,6 +20,7 @@ #include #include +#include #include namespace uhd{ namespace transport{ @@ -50,14 +51,12 @@ public: * * \param addr a string representing the destination address * \param port a string representing the destination port - * \param recv_buff_size size in bytes for the recv buffer, 0 for automatic - * \param send_buff_size size in bytes for the send buffer, 0 for automatic + * \param hints optional parameters to pass to the underlying transport */ static sptr make( const std::string &addr, const std::string &port, - size_t recv_buff_size = 0, - size_t send_buff_size = 0 + const device_addr_t &hints = device_addr_t() ); }; diff --git a/host/include/uhd/transport/usb_control.hpp b/host/include/uhd/transport/usb_control.hpp index f9829c3ec..e6c32f78e 100644 --- a/host/include/uhd/transport/usb_control.hpp +++ b/host/include/uhd/transport/usb_control.hpp @@ -18,7 +18,7 @@ #ifndef INCLUDED_UHD_TRANSPORT_USB_CONTROL_HPP #define INCLUDED_UHD_TRANSPORT_USB_CONTROL_HPP -#include "usb_device_handle.hpp" +#include namespace uhd { namespace transport { diff --git a/host/include/uhd/transport/usb_zero_copy.hpp b/host/include/uhd/transport/usb_zero_copy.hpp index 61bf380ba..b39171fba 100644 --- a/host/include/uhd/transport/usb_zero_copy.hpp +++ b/host/include/uhd/transport/usb_zero_copy.hpp @@ -18,8 +18,9 @@ #ifndef INCLUDED_UHD_TRANSPORT_USB_ZERO_COPY_HPP #define INCLUDED_UHD_TRANSPORT_USB_ZERO_COPY_HPP -#include "usb_device_handle.hpp" +#include #include +#include namespace uhd { namespace transport { @@ -47,19 +48,13 @@ public: * \param handle a device handle that uniquely identifying the device * \param recv_endpoint an integer specifiying an IN endpoint number * \param send_endpoint an integer specifiying an OUT endpoint number - * \param recv_xfer_size the number of bytes for each receive transfer - * \param recv_num_xfers the number of simultaneous receive transfers - * \param send_xfer_size the number of bytes for each send transfer - * \param send_num_xfers the number of simultaneous send transfers + * \param hints optional parameters to pass to the underlying transport */ static sptr make( usb_device_handle::sptr handle, - unsigned int recv_endpoint, - unsigned int send_endpoint, - size_t recv_xfer_size = 0, - size_t recv_num_xfers = 0, - size_t send_xfer_size = 0, - size_t send_num_xfers = 0 + size_t recv_endpoint, + size_t send_endpoint, + const device_addr_t &hints = device_addr_t() ); }; diff --git a/host/include/uhd/transport/zero_copy.hpp b/host/include/uhd/transport/zero_copy.hpp index 9dd16280c..7d8fb4b83 100644 --- a/host/include/uhd/transport/zero_copy.hpp +++ b/host/include/uhd/transport/zero_copy.hpp @@ -155,14 +155,19 @@ namespace uhd{ namespace transport{ virtual managed_recv_buffer::sptr get_recv_buff(double timeout = 0.1) = 0; /*! - * Get the maximum number of receive frames: - * The maximum number of valid managed recv buffers, - * or the maximum number of frames in the ring buffer, - * depending upon the underlying implementation. + * Get the number of receive frames: + * The number of simultaneous receive buffers in use. * \return number of frames */ virtual size_t get_num_recv_frames(void) const = 0; + /*! + * Get the size of a receive frame: + * The maximum capacity of a single receive buffer. + * \return frame size in bytes + */ + virtual size_t get_recv_frame_size(void) const = 0; + /*! * Get a new send buffer from this transport object. * \param timeout the timeout to get the buffer in seconds @@ -171,14 +176,19 @@ namespace uhd{ namespace transport{ virtual managed_send_buffer::sptr get_send_buff(double timeout = 0.1) = 0; /*! - * Get the maximum number of send frames: - * The maximum number of valid managed send buffers, - * or the maximum number of frames in the ring buffer, - * depending upon the underlying implementation. + * Get the number of send frames: + * The number of simultaneous send buffers in use. * \return number of frames */ virtual size_t get_num_send_frames(void) const = 0; + /*! + * Get the size of a send frame: + * The maximum capacity of a single send buffer. + * \return frame size in bytes + */ + virtual size_t get_send_frame_size(void) const = 0; + }; }} //namespace diff --git a/host/include/uhd/types/device_addr.hpp b/host/include/uhd/types/device_addr.hpp index e359d9467..eb3394230 100644 --- a/host/include/uhd/types/device_addr.hpp +++ b/host/include/uhd/types/device_addr.hpp @@ -20,6 +20,8 @@ #include #include +#include +#include #include #include @@ -62,6 +64,24 @@ namespace uhd{ * \return a string with delimiter markup */ std::string to_string(void) const; + + /*! + * Lexically cast a parameter to the specified type, + * or use the default value if the key is not found. + * \param key the key as one of the address parameters + * \param def the value to use when key is not present + * \return the casted value as type T or the default + * \throw error when the parameter cannot be casted + */ + template T cast(const std::string &key, const T &def) const{ + if (not this->has_key(key)) return def; + try{ + return boost::lexical_cast((*this)[key]); + } + catch(const boost::bad_lexical_cast &){ + throw std::runtime_error("cannot cast " + key + " = " + (*this)[key]); + } + } }; //handy typedef for a vector of device addresses diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index 819874483..df6db1eb9 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -26,6 +26,7 @@ #include #include +using namespace uhd; using namespace uhd::transport; static const double CLEANUP_TIMEOUT = 0.2; //seconds @@ -284,16 +285,19 @@ public: libusb_zero_copy_impl( libusb::device_handle::sptr handle, - unsigned int recv_endpoint, unsigned int send_endpoint, - size_t recv_xfer_size, size_t recv_num_xfers, - size_t send_xfer_size, size_t send_num_xfers + size_t recv_endpoint, + size_t send_endpoint, + const device_addr_t &hints ); managed_recv_buffer::sptr get_recv_buff(double); managed_send_buffer::sptr get_send_buff(double); - size_t get_num_recv_frames(void) const { return _recv_num_frames; } - size_t get_num_send_frames(void) const { return _send_num_frames; } + size_t get_num_recv_frames(void) const { return _num_recv_frames; } + size_t get_num_send_frames(void) const { return _num_send_frames; } + + size_t get_recv_frame_size(void) const { return _recv_frame_size; } + size_t get_send_frame_size(void) const { return _send_frame_size; } private: void release(libusb_transfer *lut){ @@ -311,8 +315,8 @@ private: } libusb::device_handle::sptr _handle; - size_t _recv_xfer_size, _send_xfer_size; - size_t _recv_num_frames, _send_num_frames; + const size_t _recv_frame_size, _num_recv_frames; + const size_t _send_frame_size, _num_send_frames; usb_endpoint::sptr _recv_ep, _send_ep; }; @@ -323,24 +327,16 @@ private: */ libusb_zero_copy_impl::libusb_zero_copy_impl( libusb::device_handle::sptr handle, - unsigned int recv_endpoint, unsigned int send_endpoint, - size_t recv_xfer_size, size_t recv_num_xfers, - size_t send_xfer_size, size_t send_num_xfers -){ - _handle = handle; - - //if the sizes are left at 0 (automatic) -> use the defaults - if (recv_xfer_size == 0) recv_xfer_size = DEFAULT_XFER_SIZE; - if (recv_num_xfers == 0) recv_num_xfers = DEFAULT_NUM_XFERS; - if (send_xfer_size == 0) send_xfer_size = DEFAULT_XFER_SIZE; - if (send_num_xfers == 0) send_num_xfers = DEFAULT_NUM_XFERS; - - //store the num xfers for the num frames count - _recv_xfer_size = recv_xfer_size; - _recv_num_frames = recv_num_xfers; - _send_xfer_size = send_xfer_size; - _send_num_frames = send_num_xfers; - + size_t recv_endpoint, + size_t send_endpoint, + const device_addr_t &hints +): + _handle(handle), + _recv_frame_size(size_t(hints.cast("recv_frame_size", DEFAULT_XFER_SIZE))), + _num_recv_frames(size_t(hints.cast("num_recv_frames", DEFAULT_NUM_XFERS))), + _send_frame_size(size_t(hints.cast("send_frame_size", DEFAULT_XFER_SIZE))), + _num_send_frames(size_t(hints.cast("num_send_frames", DEFAULT_NUM_XFERS))) +{ _handle->claim_interface(2 /*in interface*/); _handle->claim_interface(1 /*out interface*/); @@ -348,16 +344,16 @@ libusb_zero_copy_impl::libusb_zero_copy_impl( _handle, // libusb device_handle recv_endpoint, // USB endpoint number true, // IN endpoint - recv_xfer_size, // buffer size per transfer - recv_num_xfers // number of libusb transfers + this->get_recv_frame_size(), // buffer size per transfer + this->get_num_recv_frames() // number of libusb transfers )); _send_ep = usb_endpoint::sptr(new usb_endpoint( _handle, // libusb device_handle send_endpoint, // USB endpoint number false, // OUT endpoint - send_xfer_size, // buffer size per transfer - send_num_xfers // number of libusb transfers + this->get_send_frame_size(), // buffer size per transfer + this->get_num_send_frames() // number of libusb transfers )); } @@ -394,7 +390,7 @@ managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(double timeout){ } else { return managed_send_buffer::make_safe( - boost::asio::mutable_buffer(lut->buffer, _send_xfer_size), + boost::asio::mutable_buffer(lut->buffer, this->get_send_frame_size()), boost::bind(&libusb_zero_copy_impl::commit, shared_from_this(), lut, _1) ); } @@ -405,17 +401,14 @@ managed_send_buffer::sptr libusb_zero_copy_impl::get_send_buff(double timeout){ **********************************************************************/ usb_zero_copy::sptr usb_zero_copy::make( usb_device_handle::sptr handle, - unsigned int recv_endpoint, unsigned int send_endpoint, - size_t recv_xfer_size, size_t recv_num_xfers, - size_t send_xfer_size, size_t send_num_xfers + size_t recv_endpoint, + size_t send_endpoint, + const device_addr_t &hints ){ libusb::device_handle::sptr dev_handle(libusb::device_handle::get_cached_handle( boost::static_pointer_cast(handle)->get_device() )); return sptr(new libusb_zero_copy_impl( - dev_handle, - recv_endpoint, send_endpoint, - recv_xfer_size, recv_num_xfers, - send_xfer_size, send_num_xfers + dev_handle, recv_endpoint, send_endpoint, hints )); } diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index e9d91fe45..ada282e07 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -44,7 +44,7 @@ static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(4 * 25e6 * 0.5); static const size_t MIN_SEND_SOCK_BUFF_SIZE = size_t(10e3); //the number of async frames to allocate for each send and recv -static const size_t DEFAULT_NUM_ASYNC_FRAMES = 32; +static const size_t DEFAULT_NUM_FRAMES = 32; //a single concurrent thread for io_service seems to be the fastest static const size_t CONCURRENCY_HINT = 1; @@ -61,14 +61,15 @@ public: typedef boost::shared_ptr sptr; udp_zero_copy_asio_impl( - const std::string &addr, const std::string &port, - size_t recv_frame_size, size_t num_recv_frames, - size_t send_frame_size, size_t num_send_frames + const std::string &addr, + const std::string &port, + const device_addr_t &hints ): - _io_service(CONCURRENCY_HINT), - _work(new asio::io_service::work(_io_service)), - _recv_frame_size(recv_frame_size), _num_recv_frames(num_recv_frames), - _send_frame_size(send_frame_size), _num_send_frames(num_send_frames) + _io_service(hints.cast("concurrency_hint", CONCURRENCY_HINT)), + _recv_frame_size(size_t(hints.cast("recv_frame_size", udp_simple::mtu))), + _num_recv_frames(size_t(hints.cast("num_recv_frames", DEFAULT_NUM_FRAMES))), + _send_frame_size(size_t(hints.cast("send_frame_size", udp_simple::mtu))), + _num_send_frames(size_t(hints.cast("num_send_frames", DEFAULT_NUM_FRAMES))) { //std::cout << boost::format("Creating udp transport for %s %s") % addr % port << std::endl; @@ -99,6 +100,7 @@ public: } //spawn the service threads that will run the io service + _work = new asio::io_service::work(_io_service); //new work to delete later for (size_t i = 0; i < CONCURRENCY_HINT; i++) _thread_group.create_thread( boost::bind(&udp_zero_copy_asio_impl::service, this) ); @@ -141,6 +143,7 @@ public: } size_t get_num_recv_frames(void) const {return _num_recv_frames;} + size_t get_recv_frame_size(void) const {return _recv_frame_size;} //! pop an empty send buffer off of the fifo and bind with the commit callback managed_send_buffer::sptr get_send_buff(double timeout){ @@ -159,6 +162,7 @@ public: } size_t get_num_send_frames(void) const {return _num_send_frames;} + size_t get_send_frame_size(void) const {return _send_frame_size;} private: void service(void){ @@ -260,14 +264,15 @@ template static void resize_buff_helper( udp_zero_copy::sptr udp_zero_copy::make( const std::string &addr, const std::string &port, - size_t recv_buff_size, - size_t send_buff_size + const device_addr_t &hints ){ - udp_zero_copy_asio_impl::sptr udp_trans(new udp_zero_copy_asio_impl( - addr, port, - udp_simple::mtu, DEFAULT_NUM_ASYNC_FRAMES, //recv - udp_simple::mtu, DEFAULT_NUM_ASYNC_FRAMES //send - )); + udp_zero_copy_asio_impl::sptr udp_trans( + new udp_zero_copy_asio_impl(addr, port, hints) + ); + + //extract buffer size hints from the device addr + size_t recv_buff_size = size_t(hints.cast("recv_buff_size", 0.0)); + size_t send_buff_size = size_t(hints.cast("send_buff_size", 0.0)); //call the helper to resize send and recv buffers resize_buff_helper(udp_trans, recv_buff_size, "recv"); diff --git a/host/lib/transport/usb_dummy_impl.cpp b/host/lib/transport/usb_dummy_impl.cpp index 518342aba..8a9772e7f 100644 --- a/host/lib/transport/usb_dummy_impl.cpp +++ b/host/lib/transport/usb_dummy_impl.cpp @@ -20,6 +20,7 @@ #include #include +using namespace uhd; using namespace uhd::transport; std::vector usb_device_handle::get_device_list(boost::uint16_t, boost::uint16_t){ @@ -32,8 +33,7 @@ usb_control::sptr usb_control::make(usb_device_handle::sptr){ usb_zero_copy::sptr usb_zero_copy::make( usb_device_handle::sptr, - unsigned int, unsigned int, - size_t, size_t, size_t, size_t + size_t, size_t, const device_addr_t & ){ throw std::runtime_error("no usb support -> usb_zero_copy::make not implemented"); } diff --git a/host/lib/usrp/usrp1/usrp1_impl.cpp b/host/lib/usrp/usrp1/usrp1_impl.cpp index 6ebd6bb09..276ca86f6 100644 --- a/host/lib/usrp/usrp1/usrp1_impl.cpp +++ b/host/lib/usrp/usrp1/usrp1_impl.cpp @@ -106,17 +106,8 @@ static device_addrs_t usrp1_find(const device_addr_t &hint) /*********************************************************************** * Make **********************************************************************/ -template static output_type cast_from_dev_addr( - const device_addr_t &device_addr, - const std::string &key, - output_type def_val -){ - return (device_addr.has_key(key))? - boost::lexical_cast(device_addr[key]) : def_val; -} +static device::sptr usrp1_make(const device_addr_t &device_addr){ -static device::sptr usrp1_make(const device_addr_t &device_addr) -{ //extract the FPGA path for the USRP1 std::string usrp1_fpga_image = find_image_path( device_addr.has_key("fpga")? device_addr["fpga"] : "usrp1_fpga.rbf" @@ -127,29 +118,26 @@ static device::sptr usrp1_make(const device_addr_t &device_addr) std::vector device_list = usb_device_handle::get_device_list(USRP1_VENDOR_ID, USRP1_PRODUCT_ID); - //create data and control transports - usb_zero_copy::sptr data_transport; - usrp_ctrl::sptr usrp_ctrl; - - - BOOST_FOREACH(usb_device_handle::sptr handle, device_list) { - if (handle->get_serial() == device_addr["serial"]) { - usb_control::sptr ctrl_transport = usb_control::make(handle); - usrp_ctrl = usrp_ctrl::make(ctrl_transport); - usrp_ctrl->usrp_load_fpga(usrp1_fpga_image); - - data_transport = usb_zero_copy::make( - handle, // identifier - 6, // IN endpoint - 2, // OUT endpoint - size_t(cast_from_dev_addr(device_addr, "recv_xfer_size", 0)), - size_t(cast_from_dev_addr(device_addr, "recv_num_xfers", 0)), - size_t(cast_from_dev_addr(device_addr, "send_xfer_size", 0)), - size_t(cast_from_dev_addr(device_addr, "send_num_xfers", 0)) - ); + //locate the matching handle in the device list + usb_device_handle::sptr handle; + BOOST_FOREACH(usb_device_handle::sptr dev_handle, device_list) { + if (dev_handle->get_serial() == device_addr["serial"]){ + handle = dev_handle; break; } } + UHD_ASSERT_THROW(handle.get() != NULL); //better be found + + //create control objects and a data transport + usb_control::sptr ctrl_transport = usb_control::make(handle); + usrp_ctrl::sptr usrp_ctrl = usrp_ctrl::make(ctrl_transport); + usrp_ctrl->usrp_load_fpga(usrp1_fpga_image); + usb_zero_copy::sptr data_transport = usb_zero_copy::make( + handle, // identifier + 6, // IN endpoint + 2, // OUT endpoint + device_addr // param hints + ); //create the usrp1 implementation guts return device::sptr(new usrp1_impl(data_transport, usrp_ctrl)); diff --git a/host/lib/usrp/usrp2/io_impl.cpp b/host/lib/usrp/usrp2/io_impl.cpp index 33cec3cbc..07eda08c3 100644 --- a/host/lib/usrp/usrp2/io_impl.cpp +++ b/host/lib/usrp/usrp2/io_impl.cpp @@ -206,7 +206,7 @@ size_t usrp2_impl::send( _io_impl->packet_handler_send_state, //last state of the send handler buffs, num_samps, //buffer to fill metadata, send_mode, //samples metadata - io_type, _io_helper.get_tx_otw_type(), //input and output types to convert + io_type, _tx_otw_type, //input and output types to convert _mboards.front()->get_master_clock_freq(), //master clock tick rate uhd::transport::vrt::if_hdr_pack_be, boost::bind(&get_send_buffs, _data_transports, _1, timeout), @@ -226,7 +226,7 @@ size_t usrp2_impl::recv( _io_impl->packet_handler_recv_state, //last state of the recv handler buffs, num_samps, //buffer to fill metadata, recv_mode, //samples metadata - io_type, _io_helper.get_rx_otw_type(), //input and output types to convert + io_type, _rx_otw_type, //input and output types to convert _mboards.front()->get_master_clock_freq(), //master clock tick rate uhd::transport::vrt::if_hdr_unpack_be, boost::bind(&usrp2_impl::io_impl::get_recv_buffs, _io_impl.get(), _1, timeout) diff --git a/host/lib/usrp/usrp2/mboard_impl.cpp b/host/lib/usrp/usrp2/mboard_impl.cpp index 0b9f8ee83..a0e6adfad 100644 --- a/host/lib/usrp/usrp2/mboard_impl.cpp +++ b/host/lib/usrp/usrp2/mboard_impl.cpp @@ -38,10 +38,10 @@ using namespace uhd::usrp; usrp2_mboard_impl::usrp2_mboard_impl( size_t index, transport::udp_simple::sptr ctrl_transport, - const usrp2_io_helper &io_helper + size_t recv_frame_size ): _index(index), - _io_helper(io_helper) + _recv_frame_size(recv_frame_size) { //make a new interface for usrp2 stuff _iface = usrp2_iface::make(ctrl_transport); @@ -75,7 +75,7 @@ usrp2_mboard_impl::usrp2_mboard_impl( this->issue_ddc_stream_cmd(stream_cmd_t::STREAM_MODE_STOP_CONTINUOUS); //init the rx control registers - _iface->poke32(U2_REG_RX_CTRL_NSAMPS_PER_PKT, _io_helper.get_max_recv_samps_per_packet()); + _iface->poke32(U2_REG_RX_CTRL_NSAMPS_PER_PKT, _recv_frame_size); _iface->poke32(U2_REG_RX_CTRL_NCHANNELS, 1); _iface->poke32(U2_REG_RX_CTRL_CLEAR_OVERRUN, 1); //reset _iface->poke32(U2_REG_RX_CTRL_VRT_HEADER, 0 @@ -178,7 +178,7 @@ void usrp2_mboard_impl::set_time_spec(const time_spec_t &time_spec, bool now){ void usrp2_mboard_impl::issue_ddc_stream_cmd(const stream_cmd_t &stream_cmd){ _iface->poke32(U2_REG_RX_CTRL_STREAM_CMD, dsp_type1::calc_stream_cmd_word( - stream_cmd, _io_helper.get_max_recv_samps_per_packet() + stream_cmd, _recv_frame_size )); _iface->poke32(U2_REG_RX_CTRL_TIME_SECS, boost::uint32_t(stream_cmd.time_spec.get_full_secs())); _iface->poke32(U2_REG_RX_CTRL_TIME_TICKS, stream_cmd.time_spec.get_tick_count(get_master_clock_freq())); diff --git a/host/lib/usrp/usrp2/usrp2_impl.cpp b/host/lib/usrp/usrp2/usrp2_impl.cpp index 568c87a22..a680708ad 100644 --- a/host/lib/usrp/usrp2/usrp2_impl.cpp +++ b/host/lib/usrp/usrp2/usrp2_impl.cpp @@ -124,26 +124,7 @@ static uhd::device_addrs_t usrp2_find(const device_addr_t &hint){ /*********************************************************************** * Make **********************************************************************/ -template -out_type lexical_cast(const in_type &in){ - try{ - return boost::lexical_cast(in); - }catch(...){ - throw std::runtime_error(str(boost::format( - "failed to cast \"%s\" to type \"%s\"" - ) % boost::lexical_cast(in) % typeid(out_type).name())); - } -} - static device::sptr usrp2_make(const device_addr_t &device_addr){ - //extract the receive and send buffer sizes - size_t recv_buff_size = 0, send_buff_size= 0 ; - if (device_addr.has_key("recv_buff_size")){ - recv_buff_size = size_t(lexical_cast(device_addr["recv_buff_size"])); - } - if (device_addr.has_key("send_buff_size")){ - send_buff_size = size_t(lexical_cast(device_addr["send_buff_size"])); - } //create a ctrl and data transport for each address std::vector ctrl_transports; @@ -154,8 +135,7 @@ static device::sptr usrp2_make(const device_addr_t &device_addr){ addr, num2str(USRP2_UDP_CTRL_PORT) )); data_transports.push_back(udp_zero_copy::make( - addr, num2str(USRP2_UDP_DATA_PORT), - recv_buff_size, send_buff_size + addr, num2str(USRP2_UDP_DATA_PORT), device_addr )); } @@ -178,11 +158,23 @@ usrp2_impl::usrp2_impl( ): _data_transports(data_transports) { + //setup rx otw type + _rx_otw_type.width = 16; + _rx_otw_type.shift = 0; + _rx_otw_type.byteorder = uhd::otw_type_t::BO_BIG_ENDIAN; + + //setup tx otw type + _tx_otw_type.width = 16; + _tx_otw_type.shift = 0; + _tx_otw_type.byteorder = uhd::otw_type_t::BO_BIG_ENDIAN; + + //!!!!! set the otw type here before continuing, its used below + //create a new mboard handler for each control transport for(size_t i = 0; i < ctrl_transports.size(); i++){ - _mboards.push_back(usrp2_mboard_impl::sptr( - new usrp2_mboard_impl(i, ctrl_transports[i], _io_helper) - )); + _mboards.push_back(usrp2_mboard_impl::sptr(new usrp2_mboard_impl( + i, ctrl_transports[i], this->get_max_recv_samps_per_packet() + ))); //use an empty name when there is only one mboard std::string name = (ctrl_transports.size() > 1)? boost::lexical_cast(i) : ""; _mboard_dict[name] = _mboards.back(); diff --git a/host/lib/usrp/usrp2/usrp2_impl.hpp b/host/lib/usrp/usrp2/usrp2_impl.hpp index e8763b284..e12c4d6d4 100644 --- a/host/lib/usrp/usrp2/usrp2_impl.hpp +++ b/host/lib/usrp/usrp2/usrp2_impl.hpp @@ -71,54 +71,6 @@ private: void set(const wax::obj &key, const wax::obj &val){return _set(key, val);} }; -/*! - * The io helper class encapculates the max packet sizes and otw types. - * The otw types are read-only for now, this will be reimplemented - * when it becomes possible to change the otw type in the usrp2. - */ -class usrp2_io_helper{ -public: - usrp2_io_helper(void){ - //setup rx otw type - _rx_otw_type.width = 16; - _rx_otw_type.shift = 0; - _rx_otw_type.byteorder = uhd::otw_type_t::BO_BIG_ENDIAN; - - //setup tx otw type - _tx_otw_type.width = 16; - _tx_otw_type.shift = 0; - _tx_otw_type.byteorder = uhd::otw_type_t::BO_BIG_ENDIAN; - } - - inline size_t get_max_send_samps_per_packet(void) const{ - return _max_tx_bytes_per_packet/_tx_otw_type.get_sample_size(); - } - - inline size_t get_max_recv_samps_per_packet(void) const{ - return _max_rx_bytes_per_packet/_rx_otw_type.get_sample_size(); - } - - inline const uhd::otw_type_t &get_rx_otw_type(void) const{ - return _rx_otw_type; - } - - inline const uhd::otw_type_t &get_tx_otw_type(void) const{ - return _tx_otw_type; - } - -private: - uhd::otw_type_t _rx_otw_type, _tx_otw_type; - static const size_t _max_rx_bytes_per_packet = uhd::transport::udp_simple::mtu - - uhd::transport::vrt::max_if_hdr_words32*sizeof(boost::uint32_t) - - sizeof(uhd::transport::vrt::if_packet_info_t().tlr) //forced to have trailer - + sizeof(uhd::transport::vrt::if_packet_info_t().cid) //no class id ever used - ; - static const size_t _max_tx_bytes_per_packet = uhd::transport::udp_simple::mtu - - uhd::transport::vrt::max_if_hdr_words32*sizeof(boost::uint32_t) - + sizeof(uhd::transport::vrt::if_packet_info_t().cid) //no class id ever used - ; -}; - /*! * USRP2 mboard implementation guts: * The implementation details are encapsulated here. @@ -129,7 +81,11 @@ public: typedef boost::shared_ptr sptr; //structors - usrp2_mboard_impl(size_t index, uhd::transport::udp_simple::sptr, const usrp2_io_helper &); + usrp2_mboard_impl( + size_t index, + uhd::transport::udp_simple::sptr, + size_t recv_frame_size + ); ~usrp2_mboard_impl(void); inline double get_master_clock_freq(void){ @@ -139,7 +95,7 @@ public: private: size_t _index; int _rev_hi, _rev_lo; - const usrp2_io_helper &_io_helper; + const size_t _recv_frame_size; //properties for this mboard void get(const wax::obj &, wax::obj &); @@ -229,7 +185,8 @@ public: //the io interface size_t get_max_send_samps_per_packet(void) const{ - return _io_helper.get_max_send_samps_per_packet(); + const size_t bytes_per_packet = _data_transports.front()->get_send_frame_size() - _max_tx_header_bytes; + return bytes_per_packet/_tx_otw_type.get_sample_size(); } size_t send( const std::vector &, size_t, @@ -237,7 +194,8 @@ public: uhd::device::send_mode_t, double ); size_t get_max_recv_samps_per_packet(void) const{ - return _io_helper.get_max_recv_samps_per_packet(); + const size_t bytes_per_packet = _data_transports.front()->get_recv_frame_size() - _max_rx_header_bytes; + return bytes_per_packet/_rx_otw_type.get_sample_size(); } size_t recv( const std::vector &, size_t, @@ -257,9 +215,20 @@ private: //io impl methods and members std::vector _data_transports; - const usrp2_io_helper _io_helper; UHD_PIMPL_DECL(io_impl) _io_impl; void io_init(void); + + //over-the-wire structs and constants + uhd::otw_type_t _rx_otw_type, _tx_otw_type; + static const size_t _max_rx_header_bytes = 0 + + uhd::transport::vrt::max_if_hdr_words32*sizeof(boost::uint32_t) + + sizeof(uhd::transport::vrt::if_packet_info_t().tlr) //forced to have trailer + - sizeof(uhd::transport::vrt::if_packet_info_t().cid) //no class id ever used + ; + static const size_t _max_tx_header_bytes = 0 + + uhd::transport::vrt::max_if_hdr_words32*sizeof(boost::uint32_t) + - sizeof(uhd::transport::vrt::if_packet_info_t().cid) //no class id ever used + ; }; #endif /* INCLUDED_USRP2_IMPL_HPP */ -- cgit v1.2.3 From d2494b0313399b141913ad332315fefbba012e94 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Tue, 5 Oct 2010 12:50:53 -0700 Subject: uhd: transport docs for UDP and USB (moved from usrp docs) --- host/docs/CMakeLists.txt | 1 + host/docs/index.rst | 9 ++-- host/docs/transport.rst | 81 +++++++++++++++++++++++++++++++ host/docs/usrp1.rst | 23 --------- host/docs/usrp2.rst | 41 ---------------- host/lib/transport/udp_zero_copy_asio.cpp | 5 +- 6 files changed, 91 insertions(+), 69 deletions(-) create mode 100644 host/docs/transport.rst (limited to 'host/lib') diff --git a/host/docs/CMakeLists.txt b/host/docs/CMakeLists.txt index bbb8812b0..65db3befc 100644 --- a/host/docs/CMakeLists.txt +++ b/host/docs/CMakeLists.txt @@ -25,6 +25,7 @@ SET(manual_sources dboards.rst general.rst images.rst + transport.rst usrp1.rst usrp2.rst ) diff --git a/host/docs/index.rst b/host/docs/index.rst index bd55edc0b..7f8129e2d 100644 --- a/host/docs/index.rst +++ b/host/docs/index.rst @@ -20,11 +20,12 @@ Building the UHD ^^^^^^^^^^^^^^^^^^^^^ Application Notes ^^^^^^^^^^^^^^^^^^^^^ -* `General App Notes <./general.html>`_ +* `General Application Notes <./general.html>`_ * `Firmware and FPGA Image Notes <./images.html>`_ -* `USRP1 App Notes <./usrp1.html>`_ -* `USRP2 App Notes <./usrp2.html>`_ -* `Daughterboard App Notes <./dboards.html>`_ +* `USRP1 Application Notes <./usrp1.html>`_ +* `USRP2 Application Notes <./usrp2.html>`_ +* `Daughterboard Application Notes <./dboards.html>`_ +* `Transport Application Notes <./transport.html>`_ ^^^^^^^^^^^^^^^^^^^^^ API Documentation diff --git a/host/docs/transport.rst b/host/docs/transport.rst new file mode 100644 index 000000000..d6a146c67 --- /dev/null +++ b/host/docs/transport.rst @@ -0,0 +1,81 @@ +======================================================================== +UHD - Transport Application Notes +======================================================================== + +.. contents:: Table of Contents + +The advanced user can pass optional parameters +into the underlying transport layer through the device address. +These optional parameters control how the transport object allocates memory, +resizes kernel buffers, spawns threads, etc. +When not spcified, the transport layer will use values for these parameters +that are known to perform well on a variety of systems. +The transport parameters are defined below for the various transports in the UHD: + +------------------------------------------------------------------------ +UDP transport (ASIO) +------------------------------------------------------------------------ +The UDP transport is implemented with Boost's ASIO library. +ASIO provides an asynchronous API for user-space sockets. +The transport implementation allocates a number of buffers +and submits asynchronous requests for send and receive. +IO service threads run in the background to process these requests. + +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Transport parameters +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +The following parameters can be used to alter the transport's default behavior: + +* **recv_frame_size:** The size of a single receive buffer in bytes +* **num_recv_frames:** The number of receive buffers to allocate +* **send_frame_size:** The size of a single send buffer in bytes +* **num_send_frames:** The number of send buffers to allocate +* **concurrency_hint:** The number of threads to run the IO service + +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Resize socket buffers +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +It may be useful increase the size of the socket buffers to +move the burden of buffering samples into the kernel, or to +buffer incoming samples faster than they can be processed. +However, if your application cannot process samples fast enough, +no amount of buffering can save you. +The following parameters can be used to alter socket's buffer sizes: + +* **recv_buff_size:** The desired size of the receive buffer in bytes +* **send_buff_size:** The desired size of the send buffer in bytes + +**Note:** Large send buffers tend to decrease transmit performance. + +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Linux specific notes +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +On linux, the maximum buffer sizes are capped by the sysctl values +**net.core.rmem_max** and **net.core.wmem_max**. +To change the maximum values, run the following commands: +:: + + sudo sysctl -w net.core.rmem_max= + sudo sysctl -w net.core.wmem_max= + +Set the values permanently by editing */etc/sysctl.conf* + +------------------------------------------------------------------------ +USB transport (libusb) +------------------------------------------------------------------------ +The USB transport is implemented with libusb. +Libusb provides an asynchronous API for USB bulk transfers. +The transport implementation allocates a number of buffers +and submits asynchronous requests through libusb. +A single thread runs in the background +and executes the libusb event handler to process these requests. + +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Transport parameters +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +The following parameters can be used to alter the transport's default behavior: + +* **recv_frame_size:** The size of a single receive transfers in bytes +* **num_recv_frames:** The number of simultaneous receive transfers +* **send_frame_size:** The size of a single send transfers in bytes +* **num_send_frames:** The number of simultaneous send transfers diff --git a/host/docs/usrp1.rst b/host/docs/usrp1.rst index 0baa93a45..3443fd871 100644 --- a/host/docs/usrp1.rst +++ b/host/docs/usrp1.rst @@ -60,29 +60,6 @@ Example device address string representations to specify non-standard firmware a fpga=usrp1_fpga_4rx.rbf, fw=usrp1_fw_custom.ihx -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -Change USB transfer parameters -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -The advanced user may manipulate parameters of the usb bulk transfers -for various reasons, such as lowering latency or increasing buffer size. -By default, the UHD will use values for these parameters -that are known to perform well on a variety of systems. -The following device address parameters can be used to manipulate USB bulk transfers: - -* **recv_xfer_size:** the size of each receive bulk transfer in bytes -* **recv_num_xfers:** the number of simultaneous receive bulk transfers -* **send_xfer_size:** the size of each send bulk transfer in bytes -* **send_num_xfers:** the number of simultaneous send bulk transfers - -Example usage, set the device address markup string to the following: -:: - - serial=12345678, recv_num_xfers=16 - - -- OR -- - - serial=12345678, recv_xfer_size=2048, recv_num_xfers=16 - ------------------------------------------------------------------------ Specifying the subdevice to use ------------------------------------------------------------------------ diff --git a/host/docs/usrp2.rst b/host/docs/usrp2.rst index 70e5ea28b..1ebab388a 100644 --- a/host/docs/usrp2.rst +++ b/host/docs/usrp2.rst @@ -165,47 +165,6 @@ The device address string representation for 2 USRP2s with IPv4 addresses 192.16 addr=192.168.10.2 192.168.20.2 ------------------------------------------------------------------------- -Resize the send and receive buffers ------------------------------------------------------------------------- -It may be useful increase the size of the socket buffers to -move the burden of buffering samples into the kernel, or to -buffer incoming samples faster than they can be processed. -However, if you application cannot process samples fast enough, -no amount of buffering can save you. - -By default, the UHD will try to resize both the send and receive buffer for optimum performance. -A warning will be printed on instantiation if the actual buffer size is insufficient. -See the OS specific notes below: - -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -OS specific notes -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -On linux, the maximum buffer sizes are capped by the sysctl values -**net.core.rmem_max** and **net.core.wmem_max**. -To change the maximum values, run the following commands: -:: - - sudo sysctl -w net.core.rmem_max= - sudo sysctl -w net.core.wmem_max= - -Set the values permanently by editing */etc/sysctl.conf* - -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -Device address params -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -To manually set the size of the buffers, -the usrp2 will accept two optional parameters in the device address. -Each parameter will accept a numeric value for the number of bytes. - -* recv_buff_size -* send_buff_size - -Example usage, set the device address markup string to the following: -:: - - addr=192.168.10.2, recv_buff_size=100e6 - ------------------------------------------------------------------------ Hardware setup notes ------------------------------------------------------------------------ diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index ada282e07..a87f17a91 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -248,7 +248,10 @@ template static void resize_buff_helper( if (actual_size < target_size) uhd::print_warning(str(boost::format( "The %s buffer is smaller than the requested size.\n" "The minimum recommended buffer size is %d bytes.\n" - "See the USRP2 application notes on buffer resizing.\n" + "See the transport application notes on buffer resizing.\n" + #if defined(UHD_PLATFORM_LINUX) + "On Linux: sudo sysctl -w net.core.rmem_max=%2%\n" + #endif /*defined(UHD_PLATFORM_LINUX)*/ ) % name % min_sock_buff_size)); } -- cgit v1.2.3 From 0fc641042ef81731fa84f796fd983a5a602e260c Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Tue, 5 Oct 2010 15:47:58 -0700 Subject: usb: moved event handler thread into the zero copy interface --- host/docs/transport.rst | 1 + host/lib/transport/libusb1_base.cpp | 18 ------------------ host/lib/transport/libusb1_zero_copy.cpp | 28 ++++++++++++++++++++++++++++ 3 files changed, 29 insertions(+), 18 deletions(-) (limited to 'host/lib') diff --git a/host/docs/transport.rst b/host/docs/transport.rst index d6a146c67..30fc1d78f 100644 --- a/host/docs/transport.rst +++ b/host/docs/transport.rst @@ -79,3 +79,4 @@ The following parameters can be used to alter the transport's default behavior: * **num_recv_frames:** The number of simultaneous receive transfers * **send_frame_size:** The size of a single send transfers in bytes * **num_send_frames:** The number of simultaneous send transfers +* **concurrency_hint:** The number of threads to run the event handler diff --git a/host/lib/transport/libusb1_base.cpp b/host/lib/transport/libusb1_base.cpp index 910b04fc8..cfa77d9ca 100644 --- a/host/lib/transport/libusb1_base.cpp +++ b/host/lib/transport/libusb1_base.cpp @@ -16,12 +16,10 @@ // #include "libusb1_base.hpp" -#include #include #include #include #include -#include #include using namespace uhd; @@ -35,12 +33,9 @@ public: libusb_session_impl(void){ UHD_ASSERT_THROW(libusb_init(&_context) == 0); libusb_set_debug(_context, debug_level); - _thread_group.create_thread(boost::bind(&libusb_session_impl::run_event_loop, this)); } ~libusb_session_impl(void){ - _running = false; - _thread_group.join_all(); libusb_exit(_context); } @@ -50,19 +45,6 @@ public: private: libusb_context *_context; - boost::thread_group _thread_group; - bool _running; - - void run_event_loop(void){ - set_thread_priority_safe(); - _running = true; - timeval tv; - while(_running){ - tv.tv_sec = 0; - tv.tv_usec = 100000; //100ms - libusb_handle_events_timeout(this->get_context(), &tv); - } - } }; libusb::session::sptr libusb::session::get_global_session(void){ diff --git a/host/lib/transport/libusb1_zero_copy.cpp b/host/lib/transport/libusb1_zero_copy.cpp index df6db1eb9..f589d7c77 100644 --- a/host/lib/transport/libusb1_zero_copy.cpp +++ b/host/lib/transport/libusb1_zero_copy.cpp @@ -18,6 +18,7 @@ #include "libusb1_base.hpp" #include #include +#include #include #include #include @@ -290,6 +291,11 @@ public: const device_addr_t &hints ); + ~libusb_zero_copy_impl(void){ + _threads_running = false; + _thread_group.join_all(); + } + managed_recv_buffer::sptr get_recv_buff(double); managed_send_buffer::sptr get_send_buff(double); @@ -318,6 +324,22 @@ private: const size_t _recv_frame_size, _num_recv_frames; const size_t _send_frame_size, _num_send_frames; usb_endpoint::sptr _recv_ep, _send_ep; + + //event handler threads + boost::thread_group _thread_group; + bool _threads_running; + + void run_event_loop(void){ + set_thread_priority_safe(); + libusb::session::sptr session = libusb::session::get_global_session(); + _threads_running = true; + while(_threads_running){ + timeval tv; + tv.tv_sec = 0; + tv.tv_usec = 100000; //100ms + libusb_handle_events_timeout(session->get_context(), &tv); + } + } }; /* @@ -355,6 +377,12 @@ libusb_zero_copy_impl::libusb_zero_copy_impl( this->get_send_frame_size(), // buffer size per transfer this->get_num_send_frames() // number of libusb transfers )); + + //spawn the event handler threads + size_t concurrency = hints.cast("concurrency_hint", 1); + for (size_t i = 0; i < concurrency; i++) _thread_group.create_thread( + boost::bind(&libusb_zero_copy_impl::run_event_loop, this) + ); } /* -- cgit v1.2.3 From 7c127b6c82db773ca2dcab372ab28b4dc4e5b347 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Wed, 6 Oct 2010 10:34:10 -0700 Subject: udp: fixed boost format syntax for warning message --- host/lib/transport/udp_zero_copy_asio.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'host/lib') diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index a87f17a91..7e28caf2d 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -246,11 +246,11 @@ template static void resize_buff_helper( "Current %s sock buff size: %d bytes" ) % name % actual_size << std::endl; if (actual_size < target_size) uhd::print_warning(str(boost::format( - "The %s buffer is smaller than the requested size.\n" - "The minimum recommended buffer size is %d bytes.\n" + "The %1% buffer is smaller than the requested size.\n" + "The minimum recommended buffer size is %2% bytes.\n" "See the transport application notes on buffer resizing.\n" #if defined(UHD_PLATFORM_LINUX) - "On Linux: sudo sysctl -w net.core.rmem_max=%2%\n" + "Please run: sudo sysctl -w net.core.rmem_max=%2%\n" #endif /*defined(UHD_PLATFORM_LINUX)*/ ) % name % min_sock_buff_size)); } -- cgit v1.2.3 From b40ace72dd1b940fc0ce6e4a5e06346439dd5625 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Wed, 6 Oct 2010 18:41:30 -0700 Subject: usrp1: use the transport frame sizes to calculate the max spp The max send spp is the frame size minus the alignment padding. This allows us to copy a remainder into a new buffer and always commit multiples of the alignment size (512 bytes). Reworked the managed send buffer implementation to handle the above. Uses only managed memory, and only mem-copied under the alignment. --- host/lib/usrp/usrp1/io_impl.cpp | 169 +++++++++++++++++++++---------------- host/lib/usrp/usrp1/usrp1_impl.hpp | 10 +-- 2 files changed, 99 insertions(+), 80 deletions(-) (limited to 'host/lib') diff --git a/host/lib/usrp/usrp1/io_impl.cpp b/host/lib/usrp/usrp1/io_impl.cpp index 676b1536a..0a16f7a43 100644 --- a/host/lib/usrp/usrp1/io_impl.cpp +++ b/host/lib/usrp/usrp1/io_impl.cpp @@ -33,6 +33,28 @@ using namespace uhd::usrp; using namespace uhd::transport; namespace asio = boost::asio; +static const size_t alignment_padding = 512; + +/*********************************************************************** + * Helper struct to associate an offset with a buffer + **********************************************************************/ +class offset_send_buffer{ +public: + typedef boost::shared_ptr sptr; + + static sptr make(managed_send_buffer::sptr buff, size_t offset = 0){ + return sptr(new offset_send_buffer(buff, offset)); + } + + //member variables + managed_send_buffer::sptr buff; + size_t offset; /* in bytes */ + +private: + offset_send_buffer(managed_send_buffer::sptr buff, size_t offset): + buff(buff), offset(offset){/* NOP */} +}; + /*********************************************************************** * IO Implementation Details **********************************************************************/ @@ -41,8 +63,7 @@ struct usrp1_impl::io_impl{ data_transport(data_transport), underflow_poll_samp_count(0), overflow_poll_samp_count(0), - send_buff(data_transport->get_send_buff()), - num_bytes_committed(0) + curr_buff(offset_send_buffer::make(data_transport->get_send_buff())) { /* NOP */ } @@ -62,98 +83,88 @@ struct usrp1_impl::io_impl{ size_t overflow_poll_samp_count; //wrapper around the actual send buffer interface - //all of this to ensure only full buffers are committed - managed_send_buffer::sptr send_buff; - size_t num_bytes_committed; - double send_timeout; - boost::uint8_t pseudo_buff[BYTES_PER_PACKET]; - void phony_commit_pseudo_buff(size_t num_bytes); - void phony_commit_send_buff(size_t num_bytes); - void commit_send_buff(void); + //all of this to ensure only aligned lengths are committed + //NOTE: you must commit before getting a new buffer + //since the vrt packet handler obeys this, we are ok + offset_send_buffer::sptr curr_buff; + void commit_send_buff(offset_send_buffer::sptr, offset_send_buffer::sptr, size_t); void flush_send_buff(void); bool get_send_buffs(vrt_packet_handler::managed_send_buffs_t &, double); - - //helpers to get at the send buffer + offset - inline void *get_send_mem_ptr(void){ - return send_buff->cast() + num_bytes_committed; - } - inline size_t get_send_mem_size(void){ - return send_buff->size() - num_bytes_committed; - } }; /*! - * Accept a commit of num bytes to the pseudo buffer. - * Memcpy the entire contents of pseudo buffer into send buffers. - * - * Under most conditions: - * The first loop iteration will fill the remainder of the send buffer. - * The second loop iteration will empty the pseudo buffer remainder. + * Perform an actual commit on the send buffer: + * Copy the remainder of alignment to the next buffer. + * Commit the current buffer at multiples of alignment. */ -void usrp1_impl::io_impl::phony_commit_pseudo_buff(size_t num_bytes){ - size_t bytes_to_copy = num_bytes, bytes_copied = 0; - while(bytes_to_copy){ - size_t bytes_copied_here = std::min(bytes_to_copy, get_send_mem_size()); - std::memcpy(get_send_mem_ptr(), pseudo_buff + bytes_copied, bytes_copied_here); - phony_commit_send_buff(bytes_copied_here); - bytes_to_copy -= bytes_copied_here; - bytes_copied += bytes_copied_here; - } -} +void usrp1_impl::io_impl::commit_send_buff( + offset_send_buffer::sptr curr, + offset_send_buffer::sptr next, + size_t num_bytes +){ + //total number of bytes now in the current buffer + size_t bytes_in_curr_buffer = curr->offset + num_bytes; -/*! - * Accept a commit of num bytes to the send buffer. - * Conditionally commit the send buffer if full. - */ -void usrp1_impl::io_impl::phony_commit_send_buff(size_t num_bytes){ - num_bytes_committed += num_bytes; - if (num_bytes_committed != send_buff->size()) return; - commit_send_buff(); + //calculate how many to commit and remainder + size_t num_bytes_remaining = bytes_in_curr_buffer % alignment_padding; + size_t num_bytes_to_commit = bytes_in_curr_buffer - num_bytes_remaining; + + //copy the remainder into the next buffer + std::memcpy( + next->buff->cast() + next->offset, + curr->buff->cast() + num_bytes_to_commit, + num_bytes_remaining + ); + + //update the offset into the next buffer + next->offset += num_bytes_remaining; + + //commit the current buffer + curr->buff->commit(num_bytes_to_commit); } /*! - * Flush the send buffer: - * Zero-pad the send buffer to the nearest 512 byte boundary and commit. + * Flush the current buffer by padding out to alignment and committing. */ void usrp1_impl::io_impl::flush_send_buff(void){ - size_t bytes_to_pad = (-1*num_bytes_committed)%512; - std::memset(get_send_mem_ptr(), 0, bytes_to_pad); - num_bytes_committed += bytes_to_pad; - commit_send_buff(); + //calculate the number of bytes to alignment + size_t bytes_to_pad = (-1*curr_buff->offset)%alignment_padding; + + //get the buffer, clear, and commit (really current buffer) + vrt_packet_handler::managed_send_buffs_t buffs(1); + if (this->get_send_buffs(buffs, 0.1)){ + std::memset(buffs[0]->cast(), 0, bytes_to_pad); + buffs[0]->commit(bytes_to_pad); + } } /*! - * Perform an actual commit on the send buffer: - * Commit the contents of the send buffer and request a new buffer. + * Get a managed send buffer with the alignment padding: + * Always grab the next send buffer so we can timeout here. */ -void usrp1_impl::io_impl::commit_send_buff(void){ - send_buff->commit(num_bytes_committed); - send_buff = data_transport->get_send_buff(send_timeout); - num_bytes_committed = 0; -} - bool usrp1_impl::io_impl::get_send_buffs( vrt_packet_handler::managed_send_buffs_t &buffs, double timeout ){ - send_timeout = timeout; UHD_ASSERT_THROW(buffs.size() == 1); - //not enough bytes free -> use the pseudo buffer - if (get_send_mem_size() < BYTES_PER_PACKET){ - buffs[0] = managed_send_buffer::make_safe( - boost::asio::buffer(pseudo_buff), - boost::bind(&usrp1_impl::io_impl::phony_commit_pseudo_buff, this, _1) - ); - } - //otherwise use the send buffer offset by the bytes written - else{ - buffs[0] = managed_send_buffer::make_safe( - boost::asio::buffer(get_send_mem_ptr(), get_send_mem_size()), - boost::bind(&usrp1_impl::io_impl::phony_commit_send_buff, this, _1) - ); - } + //try to get a new managed buffer with timeout + offset_send_buffer::sptr next_buff(offset_send_buffer::make(data_transport->get_send_buff(timeout))); + if (not next_buff->buff.get()) return false; /* propagate timeout here */ - return buffs[0].get() != NULL; + //calculate the buffer pointer and size given the offset + //references to the buffers are held in the bound function + buffs[0] = managed_send_buffer::make_safe( + boost::asio::buffer( + curr_buff->buff->cast() + curr_buff->offset, + curr_buff->buff->size() - curr_buff->offset + ), + boost::bind(&usrp1_impl::io_impl::commit_send_buff, this, curr_buff, next_buff, _1) + ); + + //store the next buffer for the next call + curr_buff = next_buff; + + return true; } /*********************************************************************** @@ -182,6 +193,13 @@ static void usrp1_bs_vrt_packer( if_packet_info.num_packet_words32 = if_packet_info.num_payload_words32; } +size_t usrp1_impl::get_max_send_samps_per_packet(void) const { + return (_data_transport->get_send_frame_size() - alignment_padding) + / _tx_otw_type.get_sample_size() + / _tx_subdev_spec.size() + ; +} + size_t usrp1_impl::send( const std::vector &buffs, size_t num_samps, const tx_metadata_t &metadata, const io_type_t &io_type, @@ -249,6 +267,13 @@ static bool get_recv_buffs( return buffs[0].get() != NULL; } +size_t usrp1_impl::get_max_recv_samps_per_packet(void) const { + return _data_transport->get_recv_frame_size() + / _rx_otw_type.get_sample_size() + / _rx_subdev_spec.size() + ; +} + size_t usrp1_impl::recv( const std::vector &buffs, size_t num_samps, rx_metadata_t &metadata, const io_type_t &io_type, diff --git a/host/lib/usrp/usrp1/usrp1_impl.hpp b/host/lib/usrp/usrp1/usrp1_impl.hpp index f2c464610..ff4d40762 100644 --- a/host/lib/usrp/usrp1/usrp1_impl.hpp +++ b/host/lib/usrp/usrp1/usrp1_impl.hpp @@ -90,15 +90,9 @@ public: const uhd::io_type_t &, recv_mode_t, double); - static const size_t BYTES_PER_PACKET = 512*4; //under the transfer size + size_t get_max_send_samps_per_packet(void) const; - size_t get_max_send_samps_per_packet(void) const { - return BYTES_PER_PACKET/_tx_otw_type.get_sample_size()/_tx_subdev_spec.size(); - } - - size_t get_max_recv_samps_per_packet(void) const { - return BYTES_PER_PACKET/_rx_otw_type.get_sample_size()/_rx_subdev_spec.size(); - } + size_t get_max_recv_samps_per_packet(void) const; bool recv_async_msg(uhd::async_metadata_t &, double); -- cgit v1.2.3 From 0059e62ed067a4844edfb720da73fcdb8e69db6a Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Thu, 7 Oct 2010 10:43:58 -0700 Subject: usrp2: moved samples per packet calculation into io_impl --- host/lib/usrp/usrp2/io_impl.cpp | 19 +++++++++++++++++++ host/lib/usrp/usrp2/usrp2_impl.hpp | 23 +++-------------------- 2 files changed, 22 insertions(+), 20 deletions(-) (limited to 'host/lib') diff --git a/host/lib/usrp/usrp2/io_impl.cpp b/host/lib/usrp/usrp2/io_impl.cpp index 07eda08c3..eba704059 100644 --- a/host/lib/usrp/usrp2/io_impl.cpp +++ b/host/lib/usrp/usrp2/io_impl.cpp @@ -197,6 +197,15 @@ static bool get_send_buffs( return good; } +size_t usrp2_impl::get_max_send_samps_per_packet(void) const{ + static const size_t hdr_size = 0 + + vrt::max_if_hdr_words32*sizeof(boost::uint32_t) + - sizeof(vrt::if_packet_info_t().cid) //no class id ever used + ; + const size_t bpp = _data_transports.front()->get_send_frame_size() - hdr_size; + return bpp/_tx_otw_type.get_sample_size(); +} + size_t usrp2_impl::send( const std::vector &buffs, size_t num_samps, const tx_metadata_t &metadata, const io_type_t &io_type, @@ -217,6 +226,16 @@ size_t usrp2_impl::send( /*********************************************************************** * Receive Data **********************************************************************/ +size_t usrp2_impl::get_max_recv_samps_per_packet(void) const{ + static const size_t hdr_size = 0 + + vrt::max_if_hdr_words32*sizeof(boost::uint32_t) + + sizeof(vrt::if_packet_info_t().tlr) //forced to have trailer + - sizeof(vrt::if_packet_info_t().cid) //no class id ever used + ; + const size_t bpp = _data_transports.front()->get_recv_frame_size() - hdr_size; + return bpp/_rx_otw_type.get_sample_size(); +} + size_t usrp2_impl::recv( const std::vector &buffs, size_t num_samps, rx_metadata_t &metadata, const io_type_t &io_type, diff --git a/host/lib/usrp/usrp2/usrp2_impl.hpp b/host/lib/usrp/usrp2/usrp2_impl.hpp index e12c4d6d4..558726a2b 100644 --- a/host/lib/usrp/usrp2/usrp2_impl.hpp +++ b/host/lib/usrp/usrp2/usrp2_impl.hpp @@ -184,24 +184,18 @@ public: ~usrp2_impl(void); //the io interface - size_t get_max_send_samps_per_packet(void) const{ - const size_t bytes_per_packet = _data_transports.front()->get_send_frame_size() - _max_tx_header_bytes; - return bytes_per_packet/_tx_otw_type.get_sample_size(); - } size_t send( const std::vector &, size_t, const uhd::tx_metadata_t &, const uhd::io_type_t &, uhd::device::send_mode_t, double ); - size_t get_max_recv_samps_per_packet(void) const{ - const size_t bytes_per_packet = _data_transports.front()->get_recv_frame_size() - _max_rx_header_bytes; - return bytes_per_packet/_rx_otw_type.get_sample_size(); - } size_t recv( const std::vector &, size_t, uhd::rx_metadata_t &, const uhd::io_type_t &, uhd::device::recv_mode_t, double ); + size_t get_max_send_samps_per_packet(void) const; + size_t get_max_recv_samps_per_packet(void) const; bool recv_async_msg(uhd::async_metadata_t &, double); private: @@ -215,20 +209,9 @@ private: //io impl methods and members std::vector _data_transports; + uhd::otw_type_t _rx_otw_type, _tx_otw_type; UHD_PIMPL_DECL(io_impl) _io_impl; void io_init(void); - - //over-the-wire structs and constants - uhd::otw_type_t _rx_otw_type, _tx_otw_type; - static const size_t _max_rx_header_bytes = 0 - + uhd::transport::vrt::max_if_hdr_words32*sizeof(boost::uint32_t) - + sizeof(uhd::transport::vrt::if_packet_info_t().tlr) //forced to have trailer - - sizeof(uhd::transport::vrt::if_packet_info_t().cid) //no class id ever used - ; - static const size_t _max_tx_header_bytes = 0 - + uhd::transport::vrt::max_if_hdr_words32*sizeof(boost::uint32_t) - - sizeof(uhd::transport::vrt::if_packet_info_t().cid) //no class id ever used - ; }; #endif /* INCLUDED_USRP2_IMPL_HPP */ -- cgit v1.2.3 From 5514e67c64551e8c1d9e706f63be487aa6912705 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Thu, 7 Oct 2010 11:06:37 -0700 Subject: usrp1: ensure that the current buffer was committed before getting a new one --- host/lib/usrp/usrp1/io_impl.cpp | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'host/lib') diff --git a/host/lib/usrp/usrp1/io_impl.cpp b/host/lib/usrp/usrp1/io_impl.cpp index 0a16f7a43..6728d9b15 100644 --- a/host/lib/usrp/usrp1/io_impl.cpp +++ b/host/lib/usrp/usrp1/io_impl.cpp @@ -63,6 +63,7 @@ struct usrp1_impl::io_impl{ data_transport(data_transport), underflow_poll_samp_count(0), overflow_poll_samp_count(0), + curr_buff_committed(true), curr_buff(offset_send_buffer::make(data_transport->get_send_buff())) { /* NOP */ @@ -86,6 +87,7 @@ struct usrp1_impl::io_impl{ //all of this to ensure only aligned lengths are committed //NOTE: you must commit before getting a new buffer //since the vrt packet handler obeys this, we are ok + bool curr_buff_committed; offset_send_buffer::sptr curr_buff; void commit_send_buff(offset_send_buffer::sptr, offset_send_buffer::sptr, size_t); void flush_send_buff(void); @@ -121,6 +123,7 @@ void usrp1_impl::io_impl::commit_send_buff( //commit the current buffer curr->buff->commit(num_bytes_to_commit); + curr_buff_committed = true; } /*! @@ -145,7 +148,7 @@ void usrp1_impl::io_impl::flush_send_buff(void){ bool usrp1_impl::io_impl::get_send_buffs( vrt_packet_handler::managed_send_buffs_t &buffs, double timeout ){ - UHD_ASSERT_THROW(buffs.size() == 1); + UHD_ASSERT_THROW(curr_buff_committed and buffs.size() == 1); //try to get a new managed buffer with timeout offset_send_buffer::sptr next_buff(offset_send_buffer::make(data_transport->get_send_buff(timeout))); @@ -163,6 +166,7 @@ bool usrp1_impl::io_impl::get_send_buffs( //store the next buffer for the next call curr_buff = next_buff; + curr_buff_committed = false; return true; } -- cgit v1.2.3 From 4c3b5212b1b91b30b44af72365448ea6af66c408 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Thu, 7 Oct 2010 12:37:42 -0700 Subject: tvrx: converted floats to doubles to avoid msvc warnings --- host/lib/usrp/dboard/db_tvrx.cpp | 110 +++++++++++++++++++-------------------- 1 file changed, 55 insertions(+), 55 deletions(-) (limited to 'host/lib') diff --git a/host/lib/usrp/dboard/db_tvrx.cpp b/host/lib/usrp/dboard/db_tvrx.cpp index 41c52fbf5..33507acaa 100644 --- a/host/lib/usrp/dboard/db_tvrx.cpp +++ b/host/lib/usrp/dboard/db_tvrx.cpp @@ -66,25 +66,25 @@ static const uhd::dict tvrx_freq_ranges = map_list_of ("UHF" , freq_range_t(454e6, 860e6)) ; -static const boost::array vhflo_gains_db = +static const boost::array vhflo_gains_db = {{-6.00000, -6.00000, -6.00000, -4.00000, 0.00000, 5.00000, 10.00000, 17.40000, 26.30000, 36.00000, 43.00000, 48.00000, 49.50000, 50.10000, 50.30000, 50.30000, 50.30000}}; - -static const boost::array vhfhi_gains_db = + +static const boost::array vhfhi_gains_db = {{13.3000, -13.3000, -13.3000, -1.0000, 7.7000, 11.0000, 14.7000, 19.3000, 26.1000, 36.0000, 42.7000, 46.0000, 47.0000, 47.8000, 48.2000, 48.2000, 48.2000}}; - -static const boost::array uhf_gains_db = + +static const boost::array uhf_gains_db = {{-8.0000, -8.0000, -7.0000, 4.0000, 10.2000, 14.5000, 17.5000, 20.0000, 24.5000, 30.8000, 37.0000, 39.8000, 40.7000, 41.6000, 42.6000, 43.2000, 43.8000}}; - -static const boost::array tvrx_if_gains_db = + +static const boost::array tvrx_if_gains_db = {{-1.50000, -1.50000, -1.50000, -1.00000, 0.20000, 2.10000, 4.30000, 6.40000, 9.00000, 12.00000, 14.80000, 18.20000, 26.10000, 32.50000, 32.50000, @@ -96,32 +96,32 @@ static const boost::array tvrx_if_gains_db = //need dang near as many coefficients as to just map it like this and interp. //these numbers are culled from the 4937DI5 datasheet and are probably totally inaccurate //but if it's better than the old linear fit i'm happy -static const uhd::dict > tvrx_rf_gains_db = map_list_of +static const uhd::dict > tvrx_rf_gains_db = map_list_of ("VHFLO", vhflo_gains_db) ("VHFHI", vhfhi_gains_db) ("UHF" , uhf_gains_db) ; //sample voltages for the above points -static const boost::array tvrx_gains_volts = +static const boost::array tvrx_gains_volts = {{0.8, 1.0, 1.2, 1.4, 1.6, 1.8, 2.0, 2.2, 2.4, 2.6, 2.8, 3.0, 3.2, 3.4, 3.6, 3.8, 4.0}}; static uhd::dict get_tvrx_gain_ranges(void) { - float rfmax = 0.0, rfmin = FLT_MAX; + double rfmax = 0.0, rfmin = FLT_MAX; BOOST_FOREACH(const std::string range, tvrx_rf_gains_db.keys()) { - float my_max = tvrx_rf_gains_db[range].back(); //we're assuming it's monotonic - float my_min = tvrx_rf_gains_db[range].front(); //if it's not this is wrong wrong wrong + double my_max = tvrx_rf_gains_db[range].back(); //we're assuming it's monotonic + double my_min = tvrx_rf_gains_db[range].front(); //if it's not this is wrong wrong wrong if(my_max > rfmax) rfmax = my_max; if(my_min < rfmin) rfmin = my_min; } - - float ifmin = tvrx_if_gains_db.front(); - float ifmax = tvrx_if_gains_db.back(); - + + double ifmin = tvrx_if_gains_db.front(); + double ifmax = tvrx_if_gains_db.back(); + return map_list_of - ("RF", gain_range_t(rfmin, rfmax, (rfmax-rfmin)/4096.0)) - ("IF", gain_range_t(ifmin, ifmax, (ifmax-ifmin)/4096.0)) - ; + ("RF", gain_range_t(float(rfmin), float(rfmax), float((rfmax-rfmin)/4096.0))) + ("IF", gain_range_t(float(ifmin), float(ifmax), float((ifmax-ifmin)/4096.0))) + ; } static const double opamp_gain = 1.22; //onboard DAC opamp gain @@ -136,7 +136,7 @@ class tvrx : public rx_dboard_base{ public: tvrx(ctor_args_t args); ~tvrx(void); - + void rx_get(const wax::obj &key, wax::obj &val); void rx_set(const wax::obj &key, const wax::obj &val); @@ -198,11 +198,11 @@ tvrx::tvrx(ctor_args_t args) : rx_dboard_base(args){ } //send initial register settings if necessary - + //set default freq _lo_freq = tvrx_freq_range.min + tvrx_if_freq; //init _lo_freq to a sane default set_freq(tvrx_freq_range.min); - + //set default gains BOOST_FOREACH(const std::string &name, get_tvrx_gain_ranges().keys()){ set_gain(get_tvrx_gain_ranges()[name].min, name); @@ -238,33 +238,33 @@ static std::string get_band(double freq) { * \return a voltage to feed the TVRX analog gain */ -static float gain_interp(float gain, boost::array db_vector, boost::array volts_vector) { - float volts; - gain = std::clip(gain, db_vector.front(), db_vector.back()); //let's not get carried away here - +static double gain_interp(double gain, boost::array db_vector, boost::array volts_vector) { + double volts; + gain = std::clip(gain, db_vector.front(), db_vector.back()); //let's not get carried away here + boost::uint8_t gain_step = 0; //find which bin we're in for(size_t i = 0; i < db_vector.size()-1; i++) { if(gain >= db_vector[i] && gain <= db_vector[i+1]) gain_step = i; } - + //find the current slope for linear interpolation - float slope = (volts_vector[gain_step + 1] - volts_vector[gain_step]) + double slope = (volts_vector[gain_step + 1] - volts_vector[gain_step]) / (db_vector[gain_step + 1] - db_vector[gain_step]); - + //the problem here is that for gains approaching the maximum, the voltage slope becomes infinite //i.e., a small change in gain requires an infinite change in voltage //to cope, we limit the slope - - if(slope == std::numeric_limits::infinity()) + + if(slope == std::numeric_limits::infinity()) return volts_vector[gain_step]; //use the volts per dB slope to find the final interpolated voltage volts = volts_vector[gain_step] + (slope * (gain - db_vector[gain_step])); - + if(tvrx_debug) std::cout << "Gain interp: gain: " << gain << ", gain_step: " << int(gain_step) << ", slope: " << slope << ", volts: " << volts << std::endl; - + return volts; } @@ -283,17 +283,17 @@ static float rf_gain_to_voltage(float gain, double lo_freq){ std::string band = get_band(lo_freq + tvrx_if_freq); //this is the voltage at the TVRX gain input - float gain_volts = gain_interp(gain, tvrx_rf_gains_db[band], tvrx_gains_volts); + double gain_volts = gain_interp(gain, tvrx_rf_gains_db[band], tvrx_gains_volts); //this is the voltage at the USRP DAC output - float dac_volts = gain_volts / opamp_gain; - - dac_volts = std::clip(dac_volts, 0.0, 3.3); + double dac_volts = gain_volts / opamp_gain; + + dac_volts = std::clip(dac_volts, 0.0, 3.3); if (tvrx_debug) std::cerr << boost::format( "tvrx RF AGC gain: %f dB, dac_volts: %f V" ) % gain % dac_volts << std::endl; - return dac_volts; + return float(dac_volts); } /*! @@ -306,17 +306,17 @@ static float rf_gain_to_voltage(float gain, double lo_freq){ static float if_gain_to_voltage(float gain){ //clip the input gain = std::clip(gain, get_tvrx_gain_ranges()["IF"].min, get_tvrx_gain_ranges()["IF"].max); - - float gain_volts = gain_interp(gain, tvrx_if_gains_db, tvrx_gains_volts); - float dac_volts = gain_volts / opamp_gain; - - dac_volts = std::clip(dac_volts, 0.0, 3.3); + + double gain_volts = gain_interp(gain, tvrx_if_gains_db, tvrx_gains_volts); + double dac_volts = gain_volts / opamp_gain; + + dac_volts = std::clip(dac_volts, 0.0, 3.3); if (tvrx_debug) std::cerr << boost::format( "tvrx IF AGC gain: %f dB, dac_volts: %f V" ) % gain % dac_volts << std::endl; - return dac_volts; + return float(dac_volts); } void tvrx::set_gain(float gain, const std::string &name){ @@ -337,27 +337,27 @@ void tvrx::set_gain(float gain, const std::string &name){ */ void tvrx::set_freq(double freq) { - freq = std::clip(freq, tvrx_freq_range.min, tvrx_freq_range.max); + freq = std::clip(freq, tvrx_freq_range.min, tvrx_freq_range.max); std::string prev_band = get_band(_lo_freq - tvrx_if_freq); std::string new_band = get_band(freq); - + double target_lo_freq = freq + tvrx_if_freq; //the desired LO freq for high-side mixing double f_ref = reference_freq / double(reference_divider); //your tuning step size - + int divisor = int((target_lo_freq + (f_ref * 4.0)) / (f_ref * 8)); //the divisor we'll use double actual_lo_freq = (f_ref * 8 * divisor); //the LO freq we'll actually get - + if((divisor & ~0x7fff)) UHD_THROW_INVALID_CODE_PATH(); - + //now we update the registers _tuner_4937di5_regs.db1 = (divisor >> 8) & 0xff; _tuner_4937di5_regs.db2 = divisor & 0xff; - + if(new_band == "VHFLO") _tuner_4937di5_regs.bandsel = tuner_4937di5_regs_t::BANDSEL_VHFLO; else if(new_band == "VHFHI") _tuner_4937di5_regs.bandsel = tuner_4937di5_regs_t::BANDSEL_VHFHI; else if(new_band == "UHF") _tuner_4937di5_regs.bandsel = tuner_4937di5_regs_t::BANDSEL_UHF; else UHD_THROW_INVALID_CODE_PATH(); - + _tuner_4937di5_regs.power = tuner_4937di5_regs_t::POWER_OFF; update_regs(); @@ -365,10 +365,10 @@ void tvrx::set_freq(double freq) { //we do this because the gains are different for different band settings //not FAR off, but we do this to be consistent if(prev_band != new_band) set_gain(_gains["RF"], "RF"); - - if(tvrx_debug) + + if(tvrx_debug) std::cout << boost::format("set_freq: target LO: %f f_ref: %f divisor: %i actual LO: %f") % target_lo_freq % f_ref % divisor % actual_lo_freq << std::endl; - + _lo_freq = actual_lo_freq; //for rx props } @@ -422,7 +422,7 @@ void tvrx::rx_get(const wax::obj &key_, wax::obj &val){ return; case SUBDEV_PROP_FREQ: - /* + /* * so here we have to do some magic. because the TVRX uses a relatively high IF, * we have to watch the sample rate to see if the IF will be aliased * or if it will fall within Nyquist. -- cgit v1.2.3 From 952db7702f52e34fa05b8795abe46fb22add137d Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Thu, 7 Oct 2010 16:31:41 -0700 Subject: udp: implementation for blocking recv w/ timeout, switch async implementation w/ #define --- host/lib/transport/udp_zero_copy_asio.cpp | 140 +++++++++++++++++++++--------- 1 file changed, 101 insertions(+), 39 deletions(-) (limited to 'host/lib') diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 7e28caf2d..798cc657d 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -35,6 +35,10 @@ namespace asio = boost::asio; /*********************************************************************** * Constants **********************************************************************/ +//Define this to the the boost async io calls to perform receive. +//Otherwise, get_recv_buff uses a blocking receive with timeout. +//#define USE_ASIO_ASYNC_RECV + //enough buffering for half a second of samples at full rate on usrp2 static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(4 * 25e6 * 0.5); @@ -43,8 +47,15 @@ static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(4 * 25e6 * 0.5); //but may change with host-based flow control. static const size_t MIN_SEND_SOCK_BUFF_SIZE = size_t(10e3); -//the number of async frames to allocate for each send and recv -static const size_t DEFAULT_NUM_FRAMES = 32; +//The number of async frames to allocate for each send and recv: +//The non-async recv can have a very large number of recv frames +//because the CPU overhead is independent of the number of frames. +#ifdef USE_ASIO_ASYNC_RECV +static const size_t DEFAULT_NUM_RECV_FRAMES = 32; +#else +static const size_t DEFAULT_NUM_RECV_FRAMES = MIN_RECV_SOCK_BUFF_SIZE/udp_simple::mtu; +#endif +static const size_t DEFAULT_NUM_SEND_FRAMES = 32; //a single concurrent thread for io_service seems to be the fastest static const size_t CONCURRENCY_HINT = 1; @@ -67,9 +78,9 @@ public: ): _io_service(hints.cast("concurrency_hint", CONCURRENCY_HINT)), _recv_frame_size(size_t(hints.cast("recv_frame_size", udp_simple::mtu))), - _num_recv_frames(size_t(hints.cast("num_recv_frames", DEFAULT_NUM_FRAMES))), + _num_recv_frames(size_t(hints.cast("num_recv_frames", DEFAULT_NUM_RECV_FRAMES))), _send_frame_size(size_t(hints.cast("send_frame_size", udp_simple::mtu))), - _num_send_frames(size_t(hints.cast("num_send_frames", DEFAULT_NUM_FRAMES))) + _num_send_frames(size_t(hints.cast("num_send_frames", DEFAULT_NUM_SEND_FRAMES))) { //std::cout << boost::format("Creating udp transport for %s %s") % addr % port << std::endl; @@ -82,6 +93,13 @@ public: _socket = new asio::ip::udp::socket(_io_service); _socket->open(asio::ip::udp::v4()); _socket->connect(receiver_endpoint); + _sock_fd = _socket->native(); + } + + ~udp_zero_copy_asio_impl(void){ + delete _work; //allow io_service run to complete + _thread_group.join_all(); //wait for service threads to exit + delete _socket; } void init(void){ @@ -106,10 +124,9 @@ public: ); } - ~udp_zero_copy_asio_impl(void){ - delete _work; //allow io_service run to complete - _thread_group.join_all(); //wait for service threads to exit - delete _socket; + void service(void){ + set_thread_priority_safe(); + _io_service.run(); } //get size for internal socket buffer @@ -126,6 +143,9 @@ public: return get_buff_size(); } + //////////////////////////////////////////////////////////////////// + #ifdef USE_ASIO_ASYNC_RECV + //////////////////////////////////////////////////////////////////// //! pop a filled recv buffer off of the fifo and bind with the release callback managed_recv_buffer::sptr get_recv_buff(double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw @@ -142,6 +162,74 @@ public: return managed_recv_buffer::sptr(); } + //! handle a recv callback -> push the filled memory into the fifo + void handle_recv(void *mem, size_t len){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + _pending_recv_buffs->push_with_wait(boost::asio::buffer(mem, len)); + } + + //! release a recv buffer -> start an async recv on the buffer + void release(void *mem){ + _socket->async_receive( + boost::asio::buffer(mem, _recv_frame_size), + boost::bind( + &udp_zero_copy_asio_impl::handle_recv, + shared_from_this(), mem, + asio::placeholders::bytes_transferred + ) + ); + } + + //////////////////////////////////////////////////////////////////// + #else /*USE_ASIO_ASYNC_RECV*/ + //////////////////////////////////////////////////////////////////// + managed_recv_buffer::sptr get_recv_buff(double timeout){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + + //setup timeval for timeout + timeval tv; + tv.tv_sec = 0; + tv.tv_usec = timeout*1e6; + + //setup rset for timeout + fd_set rset; + FD_ZERO(&rset); + FD_SET(_sock_fd, &rset); + + //call select to perform timed wait + if (::select(_sock_fd+1, &rset, NULL, NULL, &tv) <= 0) + return managed_recv_buffer::sptr(); + + //grab an available buffer + asio::mutable_buffer buff; + if (not _pending_recv_buffs->pop_with_timed_wait(buff, timeout)) + return managed_recv_buffer::sptr(); + + //receive and return the buffer + return managed_recv_buffer::make_safe( + asio::buffer( + boost::asio::buffer_cast(buff), + _socket->receive(boost::asio::buffer(buff)) + ), + boost::bind( + &udp_zero_copy_asio_impl::release, + shared_from_this(), + asio::buffer_cast(buff) + ) + ); + } + + void release(void *mem){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + _pending_recv_buffs->push_with_wait( + boost::asio::buffer(mem, this->get_recv_frame_size()) + ); + } + + //////////////////////////////////////////////////////////////////// + #endif /*USE_ASIO_ASYNC_RECV*/ + //////////////////////////////////////////////////////////////////// + size_t get_num_recv_frames(void) const {return _num_recv_frames;} size_t get_recv_frame_size(void) const {return _recv_frame_size;} @@ -161,37 +249,6 @@ public: return managed_send_buffer::sptr(); } - size_t get_num_send_frames(void) const {return _num_send_frames;} - size_t get_send_frame_size(void) const {return _send_frame_size;} - -private: - void service(void){ - set_thread_priority_safe(); - _io_service.run(); - } - - /******************************************************************* - * The async send and receive callbacks - ******************************************************************/ - - //! handle a recv callback -> push the filled memory into the fifo - void handle_recv(void *mem, size_t len){ - boost::this_thread::disable_interruption di; //disable because the wait can throw - _pending_recv_buffs->push_with_wait(boost::asio::buffer(mem, len)); - } - - //! release a recv buffer -> start an async recv on the buffer - void release(void *mem){ - _socket->async_receive( - boost::asio::buffer(mem, _recv_frame_size), - boost::bind( - &udp_zero_copy_asio_impl::handle_recv, - shared_from_this(), mem, - asio::placeholders::bytes_transferred - ) - ); - } - //! handle a send callback -> push the emptied memory into the fifo void handle_send(void *mem){ boost::this_thread::disable_interruption di; //disable because the wait can throw @@ -209,10 +266,15 @@ private: ); } + size_t get_num_send_frames(void) const {return _num_send_frames;} + size_t get_send_frame_size(void) const {return _send_frame_size;} + +private: //asio guts -> socket and service asio::ip::udp::socket *_socket; asio::io_service _io_service; asio::io_service::work *_work; + asio::ip::udp::socket::native_type _sock_fd; //memory management -> buffers and fifos boost::thread_group _thread_group; -- cgit v1.2.3 From 201af62ccb4d1065a44822f0f3ce4c81755510b5 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Thu, 7 Oct 2010 16:34:29 -0700 Subject: udp: fix msvc errors for udp transport --- host/lib/transport/udp_zero_copy_asio.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'host/lib') diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 798cc657d..40fe3fa0f 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -189,7 +189,7 @@ public: //setup timeval for timeout timeval tv; tv.tv_sec = 0; - tv.tv_usec = timeout*1e6; + tv.tv_usec = long(timeout*1e6); //setup rset for timeout fd_set rset; @@ -274,7 +274,7 @@ private: asio::ip::udp::socket *_socket; asio::io_service _io_service; asio::io_service::work *_work; - asio::ip::udp::socket::native_type _sock_fd; + int _sock_fd; //memory management -> buffers and fifos boost::thread_group _thread_group; -- cgit v1.2.3 From e8e9258acb8ce8499c40de52abaa4e6ba83d479d Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Fri, 8 Oct 2010 11:20:30 -0700 Subject: udp: worked blocking send back into udp transport, enable async with #define tweaked some code in bounded buffer --- host/include/uhd/transport/alignment_buffer.ipp | 8 +- host/include/uhd/transport/bounded_buffer.ipp | 39 +++++--- host/lib/transport/udp_zero_copy_asio.cpp | 114 +++++++++++++++--------- 3 files changed, 99 insertions(+), 62 deletions(-) (limited to 'host/lib') diff --git a/host/include/uhd/transport/alignment_buffer.ipp b/host/include/uhd/transport/alignment_buffer.ipp index 5f09de0d9..833b5d399 100644 --- a/host/include/uhd/transport/alignment_buffer.ipp +++ b/host/include/uhd/transport/alignment_buffer.ipp @@ -54,14 +54,14 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ UHD_INLINE bool pop_elems_with_timed_wait( std::vector &elems, double timeout ){ - boost::system_time exit_time = boost::get_system_time() + boost::posix_time::microseconds(long(timeout*1e6)); + boost::system_time exit_time = boost::get_system_time() + to_time_dur(timeout); buff_contents_type buff_contents_tmp; std::list indexes_to_do(_all_indexes); //do an initial pop to load an initial sequence id size_t index = indexes_to_do.front(); if (not _buffs[index]->pop_with_timed_wait( - buff_contents_tmp, 1e-6*(exit_time - boost::get_system_time()).total_microseconds() + buff_contents_tmp, from_time_dur(exit_time - boost::get_system_time()) )) return false; elems[index] = buff_contents_tmp.first; seq_type expected_seq_id = buff_contents_tmp.second; @@ -76,7 +76,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ indexes_to_do = _all_indexes; index = indexes_to_do.front(); if (not _buffs[index]->pop_with_timed_wait( - buff_contents_tmp, 1e-6*(exit_time - boost::get_system_time()).total_microseconds() + buff_contents_tmp, from_time_dur(exit_time - boost::get_system_time()) )) return false; elems[index] = buff_contents_tmp.first; expected_seq_id = buff_contents_tmp.second; @@ -86,7 +86,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ //pop an element off for this index index = indexes_to_do.front(); if (not _buffs[index]->pop_with_timed_wait( - buff_contents_tmp, 1e-6*(exit_time - boost::get_system_time()).total_microseconds() + buff_contents_tmp, from_time_dur(exit_time - boost::get_system_time()) )) return false; //if the sequence id matches: diff --git a/host/include/uhd/transport/bounded_buffer.ipp b/host/include/uhd/transport/bounded_buffer.ipp index 58f78bab4..edc7faa06 100644 --- a/host/include/uhd/transport/bounded_buffer.ipp +++ b/host/include/uhd/transport/bounded_buffer.ipp @@ -19,18 +19,28 @@ #define INCLUDED_UHD_TRANSPORT_BOUNDED_BUFFER_IPP #include +#include #include #include #include namespace uhd{ namespace transport{ namespace{ /*anon*/ + static UHD_INLINE boost::posix_time::time_duration to_time_dur(double timeout){ + return boost::posix_time::microseconds(long(timeout*1e6)); + } + + static UHD_INLINE double from_time_dur(const boost::posix_time::time_duration &time_dur){ + return 1e-6*time_dur.total_microseconds(); + } + template class bounded_buffer_impl : public bounded_buffer{ public: bounded_buffer_impl(size_t capacity) : _buffer(capacity){ - /* NOP */ + _not_full_fcn = boost::bind(&bounded_buffer_impl::not_full, this); + _not_empty_fcn = boost::bind(&bounded_buffer_impl::not_empty, this); } UHD_INLINE bool push_with_pop_on_full(const elem_type &elem){ @@ -52,17 +62,16 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ UHD_INLINE void push_with_wait(const elem_type &elem){ boost::unique_lock lock(_mutex); - _full_cond.wait(lock, boost::bind(&bounded_buffer_impl::not_full, this)); + _full_cond.wait(lock, _not_full_fcn); _buffer.push_front(elem); lock.unlock(); _empty_cond.notify_one(); } - bool push_with_timed_wait(const elem_type &elem, double timeout){ + UHD_INLINE bool push_with_timed_wait(const elem_type &elem, double timeout){ boost::unique_lock lock(_mutex); if (not _full_cond.timed_wait( - lock, boost::posix_time::microseconds(long(timeout*1e6)), - boost::bind(&bounded_buffer_impl::not_full, this) + lock, to_time_dur(timeout), _not_full_fcn )) return false; _buffer.push_front(elem); lock.unlock(); @@ -72,19 +81,18 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ UHD_INLINE void pop_with_wait(elem_type &elem){ boost::unique_lock lock(_mutex); - _empty_cond.wait(lock, boost::bind(&bounded_buffer_impl::not_empty, this)); - this->pop_back(elem); + _empty_cond.wait(lock, _not_empty_fcn); + elem = this->pop_back(); lock.unlock(); _full_cond.notify_one(); } - bool pop_with_timed_wait(elem_type &elem, double timeout){ + UHD_INLINE bool pop_with_timed_wait(elem_type &elem, double timeout){ boost::unique_lock lock(_mutex); if (not _empty_cond.timed_wait( - lock, boost::posix_time::microseconds(long(timeout*1e6)), - boost::bind(&bounded_buffer_impl::not_empty, this) + lock, to_time_dur(timeout), _not_empty_fcn )) return false; - this->pop_back(elem); + elem = this->pop_back(); lock.unlock(); _full_cond.notify_one(); return true; @@ -92,7 +100,7 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ UHD_INLINE void clear(void){ boost::unique_lock lock(_mutex); - while (not_empty()) _buffer.pop_back(); + while (not_empty()) this->pop_back(); lock.unlock(); _full_cond.notify_one(); } @@ -105,16 +113,19 @@ namespace uhd{ namespace transport{ namespace{ /*anon*/ bool not_full(void) const{return not _buffer.full();} bool not_empty(void) const{return not _buffer.empty();} + boost::function _not_full_fcn, _not_empty_fcn; + /*! * Three part operation to pop an element: * 1) assign elem to the back element * 2) assign the back element to empty * 3) pop the back to move the counter */ - UHD_INLINE void pop_back(elem_type &elem){ - elem = _buffer.back(); + UHD_INLINE elem_type pop_back(void){ + elem_type elem = _buffer.back(); _buffer.back() = elem_type(); _buffer.pop_back(); + return elem; } }; }}} //namespace diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp index 40fe3fa0f..d84aeefdd 100644 --- a/host/lib/transport/udp_zero_copy_asio.cpp +++ b/host/lib/transport/udp_zero_copy_asio.cpp @@ -39,6 +39,10 @@ namespace asio = boost::asio; //Otherwise, get_recv_buff uses a blocking receive with timeout. //#define USE_ASIO_ASYNC_RECV +//Define this to the the boost async io calls to perform send. +//Otherwise, the commit callback uses a blocking send. +//#define USE_ASIO_ASYNC_SEND + //enough buffering for half a second of samples at full rate on usrp2 static const size_t MIN_RECV_SOCK_BUFF_SIZE = size_t(4 * 25e6 * 0.5); @@ -55,7 +59,13 @@ static const size_t DEFAULT_NUM_RECV_FRAMES = 32; #else static const size_t DEFAULT_NUM_RECV_FRAMES = MIN_RECV_SOCK_BUFF_SIZE/udp_simple::mtu; #endif +//The non-async send only ever requires a single frame +//because the buffer will be committed before a new get. +#ifdef USE_ASIO_ASYNC_SEND static const size_t DEFAULT_NUM_SEND_FRAMES = 32; +#else +static const size_t DEFAULT_NUM_SEND_FRAMES = MIN_SEND_SOCK_BUFF_SIZE/udp_simple::mtu;; +#endif //a single concurrent thread for io_service seems to be the fastest static const size_t CONCURRENCY_HINT = 1; @@ -143,6 +153,12 @@ public: return get_buff_size(); } + //! handle a recv callback -> push the filled memory into the fifo + UHD_INLINE void handle_recv(void *mem, size_t len){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + _pending_recv_buffs->push_with_wait(boost::asio::buffer(mem, len)); + } + //////////////////////////////////////////////////////////////////// #ifdef USE_ASIO_ASYNC_RECV //////////////////////////////////////////////////////////////////// @@ -162,16 +178,10 @@ public: return managed_recv_buffer::sptr(); } - //! handle a recv callback -> push the filled memory into the fifo - void handle_recv(void *mem, size_t len){ - boost::this_thread::disable_interruption di; //disable because the wait can throw - _pending_recv_buffs->push_with_wait(boost::asio::buffer(mem, len)); - } - //! release a recv buffer -> start an async recv on the buffer void release(void *mem){ _socket->async_receive( - boost::asio::buffer(mem, _recv_frame_size), + boost::asio::buffer(mem, this->get_recv_frame_size()), boost::bind( &udp_zero_copy_asio_impl::handle_recv, shared_from_this(), mem, @@ -185,6 +195,7 @@ public: //////////////////////////////////////////////////////////////////// managed_recv_buffer::sptr get_recv_buff(double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw + asio::mutable_buffer buff; //setup timeval for timeout timeval tv; @@ -196,34 +207,30 @@ public: FD_ZERO(&rset); FD_SET(_sock_fd, &rset); - //call select to perform timed wait - if (::select(_sock_fd+1, &rset, NULL, NULL, &tv) <= 0) - return managed_recv_buffer::sptr(); - - //grab an available buffer - asio::mutable_buffer buff; - if (not _pending_recv_buffs->pop_with_timed_wait(buff, timeout)) - return managed_recv_buffer::sptr(); - - //receive and return the buffer - return managed_recv_buffer::make_safe( - asio::buffer( - boost::asio::buffer_cast(buff), - _socket->receive(boost::asio::buffer(buff)) - ), - boost::bind( - &udp_zero_copy_asio_impl::release, - shared_from_this(), - asio::buffer_cast(buff) - ) - ); + //call select to perform timed wait and grab an available buffer with wait + //if the condition is true, call receive and return the managed buffer + if ( + ::select(_sock_fd+1, &rset, NULL, NULL, &tv) > 0 and + _pending_recv_buffs->pop_with_timed_wait(buff, timeout) + ){ + return managed_recv_buffer::make_safe( + asio::buffer( + boost::asio::buffer_cast(buff), + _socket->receive(asio::buffer(buff)) + ), + boost::bind( + &udp_zero_copy_asio_impl::release, + shared_from_this(), + asio::buffer_cast(buff) + ) + ); + } + return managed_recv_buffer::sptr(); } void release(void *mem){ boost::this_thread::disable_interruption di; //disable because the wait can throw - _pending_recv_buffs->push_with_wait( - boost::asio::buffer(mem, this->get_recv_frame_size()) - ); + handle_recv(mem, this->get_recv_frame_size()); } //////////////////////////////////////////////////////////////////// @@ -233,6 +240,12 @@ public: size_t get_num_recv_frames(void) const {return _num_recv_frames;} size_t get_recv_frame_size(void) const {return _recv_frame_size;} + //! handle a send callback -> push the emptied memory into the fifo + UHD_INLINE void handle_send(void *mem){ + boost::this_thread::disable_interruption di; //disable because the wait can throw + _pending_send_buffs->push_with_wait(boost::asio::buffer(mem, this->get_send_frame_size())); + } + //! pop an empty send buffer off of the fifo and bind with the commit callback managed_send_buffer::sptr get_send_buff(double timeout){ boost::this_thread::disable_interruption di; //disable because the wait can throw @@ -249,12 +262,9 @@ public: return managed_send_buffer::sptr(); } - //! handle a send callback -> push the emptied memory into the fifo - void handle_send(void *mem){ - boost::this_thread::disable_interruption di; //disable because the wait can throw - _pending_send_buffs->push_with_wait(boost::asio::buffer(mem, _send_frame_size)); - } - + //////////////////////////////////////////////////////////////////// + #ifdef USE_ASIO_ASYNC_SEND + //////////////////////////////////////////////////////////////////// //! commit a send buffer -> start an async send on the buffer void commit(void *mem, size_t len){ _socket->async_send( @@ -266,6 +276,18 @@ public: ); } + //////////////////////////////////////////////////////////////////// + #else /*USE_ASIO_ASYNC_SEND*/ + //////////////////////////////////////////////////////////////////// + void commit(void *mem, size_t len){ + _socket->send(asio::buffer(mem, len)); + handle_send(mem); + } + + //////////////////////////////////////////////////////////////////// + #endif /*USE_ASIO_ASYNC_SEND*/ + //////////////////////////////////////////////////////////////////// + size_t get_num_send_frames(void) const {return _num_send_frames;} size_t get_send_frame_size(void) const {return _send_frame_size;} @@ -297,6 +319,13 @@ template static void resize_buff_helper( if (name == "recv") min_sock_buff_size = MIN_RECV_SOCK_BUFF_SIZE; if (name == "send") min_sock_buff_size = MIN_SEND_SOCK_BUFF_SIZE; + std::string help_message; + #if defined(UHD_PLATFORM_LINUX) + help_message = str(boost::format( + "Please run: sudo sysctl -w net.core.%smem_max=%d\n" + ) % ((name == "recv")?"r":"w") % min_sock_buff_size); + #endif /*defined(UHD_PLATFORM_LINUX)*/ + //resize the buffer if size was provided if (target_size > 0){ size_t actual_size = udp_trans->resize_buff(target_size); @@ -308,13 +337,10 @@ template static void resize_buff_helper( "Current %s sock buff size: %d bytes" ) % name % actual_size << std::endl; if (actual_size < target_size) uhd::print_warning(str(boost::format( - "The %1% buffer is smaller than the requested size.\n" - "The minimum recommended buffer size is %2% bytes.\n" - "See the transport application notes on buffer resizing.\n" - #if defined(UHD_PLATFORM_LINUX) - "Please run: sudo sysctl -w net.core.rmem_max=%2%\n" - #endif /*defined(UHD_PLATFORM_LINUX)*/ - ) % name % min_sock_buff_size)); + "The %s buffer is smaller than the requested size.\n" + "The minimum recommended buffer size is %d bytes.\n" + "See the transport application notes on buffer resizing.\n%s" + ) % name % min_sock_buff_size % help_message)); } //only enable on platforms that are happy with the large buffer resize -- cgit v1.2.3 From 4b5792d2a5c6d101f602b5854fb2e7cfb3acf602 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Fri, 8 Oct 2010 14:00:56 -0700 Subject: tvrx: added enabled prop to set and get --- host/lib/usrp/dboard/db_tvrx.cpp | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'host/lib') diff --git a/host/lib/usrp/dboard/db_tvrx.cpp b/host/lib/usrp/dboard/db_tvrx.cpp index 33507acaa..b1fc2ea76 100644 --- a/host/lib/usrp/dboard/db_tvrx.cpp +++ b/host/lib/usrp/dboard/db_tvrx.cpp @@ -447,6 +447,10 @@ void tvrx::rx_get(const wax::obj &key_, wax::obj &val){ val = SUBDEV_CONN_COMPLEX_IQ; return; + case SUBDEV_PROP_ENABLED: + val = true; //always enabled + return; + case SUBDEV_PROP_USE_LO_OFFSET: val = false; return; @@ -467,10 +471,14 @@ void tvrx::rx_set(const wax::obj &key_, const wax::obj &val){ case SUBDEV_PROP_GAIN: this->set_gain(val.as(), key.name); return; + case SUBDEV_PROP_FREQ: this->set_freq(val.as()); return; + case SUBDEV_PROP_ENABLED: + return; //always enabled + default: UHD_THROW_PROP_SET_ERROR(); } } -- cgit v1.2.3 From 1190bc5c1a7bd0946bc55373eca58ec60a100f63 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Fri, 8 Oct 2010 15:52:33 -0700 Subject: usrp: added calls to get subdev and mboard canonical names --- host/include/uhd/usrp/single_usrp.hpp | 16 +++++++++++++--- host/lib/usrp/single_usrp.cpp | 18 +++++++++++++++--- 2 files changed, 28 insertions(+), 6 deletions(-) (limited to 'host/lib') diff --git a/host/include/uhd/usrp/single_usrp.hpp b/host/include/uhd/usrp/single_usrp.hpp index 1b89a3620..bc6943478 100644 --- a/host/include/uhd/usrp/single_usrp.hpp +++ b/host/include/uhd/usrp/single_usrp.hpp @@ -57,15 +57,21 @@ public: */ virtual device::sptr get_device(void) = 0; + /******************************************************************* + * Mboard methods + ******************************************************************/ /*! * Get a printable name for this usrp. * \return a printable string */ virtual std::string get_pp_string(void) = 0; - /******************************************************************* - * Misc - ******************************************************************/ + /*! + * Get canonical name for this USRP motherboard. + * \return a string representing the name + */ + virtual std::string get_mboard_name(void) = 0; + /*! * Gets the current time in the usrp time registers. * \return a timespec representing current usrp time @@ -113,6 +119,8 @@ public: virtual void set_rx_subdev_spec(const uhd::usrp::subdev_spec_t &spec) = 0; virtual uhd::usrp::subdev_spec_t get_rx_subdev_spec(void) = 0; + virtual std::string get_rx_subdev_name(size_t chan = 0) = 0; + virtual void set_rx_rate(double rate) = 0; virtual double get_rx_rate(void) = 0; @@ -146,6 +154,8 @@ public: virtual void set_tx_subdev_spec(const uhd::usrp::subdev_spec_t &spec) = 0; virtual uhd::usrp::subdev_spec_t get_tx_subdev_spec(void) = 0; + virtual std::string get_tx_subdev_name(size_t chan = 0) = 0; + virtual void set_tx_rate(double rate) = 0; virtual double get_tx_rate(void) = 0; diff --git a/host/lib/usrp/single_usrp.cpp b/host/lib/usrp/single_usrp.cpp index bb4af44b8..08dc3bb1d 100644 --- a/host/lib/usrp/single_usrp.cpp +++ b/host/lib/usrp/single_usrp.cpp @@ -54,6 +54,9 @@ public: return _dev; } + /******************************************************************* + * Mboard methods + ******************************************************************/ std::string get_pp_string(void){ std::string buff = str(boost::format( "Single USRP:\n" @@ -101,9 +104,10 @@ public: return buff; } - /******************************************************************* - * Misc - ******************************************************************/ + std::string get_mboard_name(void){ + return _mboard()[MBOARD_PROP_NAME].as(); + } + time_spec_t get_time_now(void){ return _mboard()[MBOARD_PROP_TIME_NOW].as(); } @@ -135,6 +139,10 @@ public: return _mboard()[MBOARD_PROP_RX_SUBDEV_SPEC].as(); } + std::string get_rx_subdev_name(size_t chan){ + return _rx_subdev(chan)[SUBDEV_PROP_NAME].as(); + } + void set_rx_rate(double rate){ _rx_dsp()[DSP_PROP_HOST_RATE] = rate; } @@ -206,6 +214,10 @@ public: return _mboard()[MBOARD_PROP_TX_SUBDEV_SPEC].as(); } + std::string get_tx_subdev_name(size_t chan){ + return _tx_subdev(chan)[SUBDEV_PROP_NAME].as(); + } + void set_tx_rate(double rate){ _tx_dsp()[DSP_PROP_HOST_RATE] = rate; } -- cgit v1.2.3 From 787d89518d6dace0d8e6584e043eba99f5ff33e1 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Fri, 8 Oct 2010 16:04:22 -0700 Subject: uhd: better warning message for failing to set rt priority --- host/lib/utils/thread_priority.cpp | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) (limited to 'host/lib') diff --git a/host/lib/utils/thread_priority.cpp b/host/lib/utils/thread_priority.cpp index c35e5fcb1..f09d1b1d6 100644 --- a/host/lib/utils/thread_priority.cpp +++ b/host/lib/utils/thread_priority.cpp @@ -16,6 +16,8 @@ // #include +#include +#include #include #include @@ -24,7 +26,12 @@ bool uhd::set_thread_priority_safe(float priority, bool realtime){ set_thread_priority(priority, realtime); return true; }catch(const std::exception &e){ - std::cerr << "set_thread_priority: " << e.what() << std::endl; + uhd::print_warning(str(boost::format( + "%s\n" + "Failed to set thread priority %d (%s):\n" + "Performance may be negatively affected.\n" + "See the general application notes.\n" + ) % e.what() % priority % (realtime?"realtime":""))); return false; } } -- cgit v1.2.3