3 files changed, 224 insertions, 39 deletions

diff --git a/host/lib/transport/convert_types.cpp b/host/lib/transport/convert_types.cpp
new file mode 100644
index 000000000..2a6854f50
--- /dev/null
+++ b/host/lib/transport/convert_types.cpp
@@ -0,0 +1,144 @@
+//
+// 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 <http://www.gnu.org/licenses/>.
+//
+
+#include <uhd/transport/convert_types.hpp>
+#include <uhd/utils/assert.hpp>
+#include <boost/asio.hpp> //endianness conversion
+#include <boost/cstdint.hpp>
+
+using namespace uhd;
+
+/***********************************************************************
+ * Constants
+ **********************************************************************/
+typedef std::complex<float>          fc32_t;
+
+static const float shorts_per_float = float(1 << 15);
+static const float floats_per_short = float(1.0/shorts_per_float);
+
+#define unrolled_loop(__inst, __len){ \
+    size_t __i = 0; \
+    for(; __i < (__len & ~0x3); __i+= 4){ \
+        __inst(__i+0); __inst(__i+1); \
+        __inst(__i+2); __inst(__i+3); \
+    } \
+    for(; __i < __len; __i++){ \
+        __inst(__i); \
+    } \
+}
+
+// set a boolean flag that indicates the endianess
+#ifdef HAVE_BIG_ENDIAN
+static const bool is_big_endian = true;
+#else
+static const bool is_big_endian = false;
+#endif
+
+static inline void host_floats_to_usrp2_items(
+    boost::uint32_t *usrp2_items,
+    const fc32_t *host_floats,
+    size_t num_samps
+){
+    #define host_floats_to_usrp2_items_i(i){ \
+        boost::uint16_t real = boost::int16_t(host_floats[i].real()*shorts_per_float); \
+        boost::uint16_t imag = boost::int16_t(host_floats[i].imag()*shorts_per_float); \
+        usrp2_items[i] = htonl((real << 16) | (imag << 0)); \
+    }
+    unrolled_loop(host_floats_to_usrp2_items_i, num_samps);
+}
+
+static inline void usrp2_items_to_host_floats(
+    fc32_t *host_floats,
+    const boost::uint32_t *usrp2_items,
+    size_t num_samps
+){
+    #define usrp2_items_to_host_floats_i(i){ \
+        boost::uint32_t item = ntohl(usrp2_items[i]); \
+        boost::int16_t real = boost::uint16_t(item >> 16); \
+        boost::int16_t imag = boost::uint16_t(item >> 0); \
+        host_floats[i] = fc32_t(float(real*floats_per_short), float(imag*floats_per_short)); \
+    }
+    unrolled_loop(usrp2_items_to_host_floats_i, num_samps);
+}
+
+static inline void host_items_to_usrp2_items(
+    boost::uint32_t *usrp2_items,
+    const boost::uint32_t *host_items,
+    size_t num_samps
+){
+    #define host_items_to_usrp2_items_i(i) usrp2_items[i] = htonl(host_items[i])
+    if (is_big_endian){
+        std::memcpy(usrp2_items, host_items, num_samps*sizeof(boost::uint32_t));
+    }
+    else{
+        unrolled_loop(host_items_to_usrp2_items_i, num_samps);
+    }
+}
+
+static inline void usrp2_items_to_host_items(
+    boost::uint32_t *host_items,
+    const boost::uint32_t *usrp2_items,
+    size_t num_samps
+){
+    #define usrp2_items_to_host_items_i(i) host_items[i] = ntohl(usrp2_items[i])
+    if (is_big_endian){
+        std::memcpy(host_items, usrp2_items, num_samps*sizeof(boost::uint32_t));
+    }
+    else{
+        unrolled_loop(usrp2_items_to_host_items_i, num_samps);
+    }
+}
+
+void transport::convert_io_type_to_otw_type(
+    const void *io_buff, const io_type_t &io_type,
+    void *otw_buff, const otw_type_t &otw_type,
+    size_t num_samps
+){
+    //all we handle for now:
+    ASSERT_THROW(otw_type.width == 16 and otw_type.byteorder == otw_type_t::BO_BIG_ENDIAN);
+
+    switch(io_type.tid){
+    case io_type_t::COMPLEX_FLOAT32:
+        host_floats_to_usrp2_items((boost::uint32_t *)otw_buff, (const fc32_t*)io_buff, num_samps);
+        break;
+    case io_type_t::COMPLEX_INT16:
+        host_items_to_usrp2_items((boost::uint32_t *)otw_buff, (const boost::uint32_t*)io_buff, num_samps);
+        break;
+    default:
+        throw std::runtime_error(str(boost::format("convert_types: cannot handle type \"%c\"") % io_type.tid));
+    }
+}
+
+void transport::convert_otw_type_to_io_type(
+    const void *otw_buff, const otw_type_t &otw_type,
+    void *io_buff, const io_type_t &io_type,
+    size_t num_samps
+){
+    //all we handle for now:
+    ASSERT_THROW(otw_type.width == 16 and otw_type.byteorder == otw_type_t::BO_BIG_ENDIAN);
+
+    switch(io_type.tid){
+    case io_type_t::COMPLEX_FLOAT32:
+        usrp2_items_to_host_floats((fc32_t*)io_buff, (const boost::uint32_t *)otw_buff, num_samps);
+        break;
+    case io_type_t::COMPLEX_INT16:
+        usrp2_items_to_host_items((boost::uint32_t*)io_buff, (const boost::uint32_t *)otw_buff, num_samps);
+        break;
+    default:
+        throw std::runtime_error(str(boost::format("convert_types: cannot handle type \"%c\"") % io_type.tid));
+    }
+}
diff --git a/host/lib/transport/gen_vrt.py b/host/lib/transport/gen_vrt.py
index 0f961efc2..38a394dee 100755
--- a/host/lib/transport/gen_vrt.py
+++ b/host/lib/transport/gen_vrt.py
@@ -200,10 +200,15 @@ void vrt::unpack(
 }
 """
 
+import os
 import sys
-from Cheetah import Template
-def parse_str(_tmpl_text, **kwargs): return str(Template.Template(_tmpl_text, kwargs))
+
+from Cheetah.Template import Template
+def parse_tmpl(_tmpl_text, **kwargs):
+    return str(Template(_tmpl_text, kwargs))
+def safe_makedirs(path):
+    not os.path.isdir(path) and os.makedirs(path)
 
 if __name__ == '__main__':
-    from Cheetah import Template
-    open(sys.argv[1], 'w').write(parse_str(TMPL_TEXT, file=__file__))
+    safe_makedirs(os.path.dirname(sys.argv[1]))
+    open(sys.argv[1], 'w').write(parse_tmpl(TMPL_TEXT, file=__file__))
diff --git a/host/lib/transport/udp_zero_copy_asio.cpp b/host/lib/transport/udp_zero_copy_asio.cpp
index 1fc8ce14a..7e643abad 100644
--- a/host/lib/transport/udp_zero_copy_asio.cpp
+++ b/host/lib/transport/udp_zero_copy_asio.cpp
@@ -25,39 +25,74 @@
 using namespace uhd::transport;
 
 /***********************************************************************
- * Smart buffer implementation for udp zerocopy none
- *
- * This smart buffer implemention houses a const buffer.
- * When the smart buffer is deleted, the buffer is freed.
- * The memory in the const buffer is allocated with new [],
- * and so the destructor frees the buffer with delete [].
+ * Managed receive buffer implementation for udp zero-copy asio:
+ *   Frees the memory held by the const buffer on done.
  **********************************************************************/
-class smart_buffer_impl : public smart_buffer{
+class managed_recv_buffer_impl : public managed_recv_buffer{
 public:
-    smart_buffer_impl(const boost::asio::const_buffer &buff){
-        _buff = buff;
+    managed_recv_buffer_impl(const boost::asio::const_buffer &buff) : _buff(buff){
+        _done = false;
     }
 
-    ~smart_buffer_impl(void){
+    ~managed_recv_buffer_impl(void){
+        if (not _done) this->done();
+    }
+
+    void done(void){
+        _done = true;
         delete [] boost::asio::buffer_cast<const boost::uint32_t *>(_buff);
     }
 
-    const boost::asio::const_buffer &get(void) const{
+private:
+    const boost::asio::const_buffer &get(void){
         return _buff;
     }
 
+    const boost::asio::const_buffer _buff;
+    bool _done;
+};
+
+/***********************************************************************
+ * Managed send buffer implementation for udp zero-copy asio:
+ *   Sends and frees the memory held by the mutable buffer on done.
+ **********************************************************************/
+class managed_send_buffer_impl : public managed_send_buffer{
+public:
+    managed_send_buffer_impl(
+        const boost::asio::mutable_buffer &buff,
+        boost::asio::ip::udp::socket *socket
+    ) : _buff(buff){
+        _done = false;
+        _socket = socket;
+    }
+
+    ~managed_send_buffer_impl(void){
+        if (not _done) this->done(0);
+    }
+
+    void done(size_t num_bytes){
+        _done = true;
+        boost::uint32_t *mem = boost::asio::buffer_cast<boost::uint32_t *>(_buff);
+        _socket->send(boost::asio::buffer(mem, num_bytes));
+        delete [] mem;
+    }
+
 private:
-    boost::asio::const_buffer _buff;
+    const boost::asio::mutable_buffer &get(void){
+        return _buff;
+    }
+
+    const boost::asio::mutable_buffer _buff;
+    boost::asio::ip::udp::socket      *_socket;
+    bool _done;
 };
 
 /***********************************************************************
- * UDP zero copy implementation class
- *
- * This is the portable zero copy implementation for systems
- * where a faster, platform specific solution is not available.
- *
- * It uses boost asio udp sockets and the standard recv() class,
- * and in-fact, is not actually doing a zero-copy implementation.
+ * Zero Copy UDP implementation with ASIO:
+ *   This is the portable zero copy implementation for systems
+ *   where a faster, platform specific solution is not available.
+ *   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 udp_zero_copy{
 public:
@@ -66,8 +101,8 @@ public:
     ~udp_zero_copy_impl(void);
 
     //send/recv
-    size_t send(const boost::asio::const_buffer &buff);
-    smart_buffer::sptr recv(void);
+    managed_recv_buffer::sptr get_recv_buff(void);
+    managed_send_buffer::sptr get_send_buff(void);
 
 private:
     boost::asio::ip::udp::socket   *_socket;
@@ -107,31 +142,32 @@ udp_zero_copy_impl::~udp_zero_copy_impl(void){
     delete _socket;
 }
 
-size_t udp_zero_copy_impl::send(const boost::asio::const_buffer &buff){
-    return _socket->send(boost::asio::buffer(buff));
-}
-
-smart_buffer::sptr udp_zero_copy_impl::recv(void){
-    size_t available = 0;
-
+managed_recv_buffer::sptr udp_zero_copy_impl::get_recv_buff(void){
     //implement timeout through polling and sleeping
+    size_t available = 0;
     boost::asio::deadline_timer timer(_socket->get_io_service());
     timer.expires_from_now(boost::posix_time::milliseconds(100));
     while (not ((available = _socket->available()) or timer.expires_from_now().is_negative())){
         boost::this_thread::sleep(boost::posix_time::milliseconds(1));
     }
 
-    //allocate memory and create buffer
-    boost::uint32_t *buff_mem = new boost::uint32_t[available/sizeof(boost::uint32_t)];
-    boost::asio::mutable_buffer buff(buff_mem, available);
-
     //receive only if data is available
+    boost::uint32_t *buff_mem = new boost::uint32_t[available/sizeof(boost::uint32_t)];
     if (available){
-        _socket->receive(boost::asio::buffer(buff));
+        available = _socket->receive(boost::asio::buffer(buff_mem, available));
     }
 
-    //create a new smart buffer to house the data
-    return smart_buffer::sptr(new smart_buffer_impl(buff));
+    //create a new managed buffer to house the data
+    return managed_recv_buffer::sptr(
+        new managed_recv_buffer_impl(boost::asio::buffer(buff_mem, available))
+    );
+}
+
+managed_send_buffer::sptr udp_zero_copy_impl::get_send_buff(void){
+    boost::uint32_t *buff_mem = new boost::uint32_t[2000/sizeof(boost::uint32_t)];
+    return managed_send_buffer::sptr(
+        new managed_send_buffer_impl(boost::asio::buffer(buff_mem, 2000), _socket)
+    );
 }
 
 size_t udp_zero_copy_impl::get_recv_buff_size(void){