uhd: added scaling factor to conversion routines

7 files changed, 41 insertions, 49 deletions

diff --git a/host/include/uhd/convert.hpp b/host/include/uhd/convert.hpp
index 8fc2f38db..99f1860ae 100644
--- a/host/include/uhd/convert.hpp
+++ b/host/include/uhd/convert.hpp
@@ -29,7 +29,7 @@ namespace uhd{ namespace convert{
 
     typedef uhd::ref_vector<void *> output_type;
     typedef uhd::ref_vector<const void *> input_type;
-    typedef boost::function<void(const input_type&, const output_type&, size_t)> function_type;
+    typedef boost::function<void(const input_type&, const output_type&, size_t, double)> function_type;
 
     /*!
      * Describe the priority of a converter function.
diff --git a/host/lib/convert/convert_common.hpp b/host/lib/convert/convert_common.hpp
index c2ca233d9..7f513b124 100644
--- a/host/lib/convert/convert_common.hpp
+++ b/host/lib/convert/convert_common.hpp
@@ -27,7 +27,7 @@
     static void fcn( \
         const uhd::convert::input_type &inputs, \
         const uhd::convert::output_type &outputs, \
-        size_t nsamps \
+        size_t nsamps, double scale_factor \
     ); \
     UHD_STATIC_BLOCK(register_##fcn##_##prio){ \
         uhd::convert::register_converter(#fcn, fcn, prio); \
@@ -35,7 +35,7 @@
     static void fcn( \
         const uhd::convert::input_type &inputs, \
         const uhd::convert::output_type &outputs, \
-        size_t nsamps \
+        size_t nsamps, double scale_factor \
     )
 
 /***********************************************************************
@@ -50,7 +50,7 @@ typedef boost::uint32_t              item32_t;
 /***********************************************************************
  * Convert complex short buffer to items32
  **********************************************************************/
-static UHD_INLINE item32_t sc16_to_item32(sc16_t num){
+static UHD_INLINE item32_t sc16_to_item32(sc16_t num, double){
     boost::uint16_t real = num.real();
     boost::uint16_t imag = num.imag();
     return (item32_t(real) << 16) | (item32_t(imag) << 0);
@@ -59,7 +59,7 @@ static UHD_INLINE item32_t sc16_to_item32(sc16_t num){
 /***********************************************************************
  * Convert items32 buffer to complex short
  **********************************************************************/
-static UHD_INLINE sc16_t item32_to_sc16(item32_t item){
+static UHD_INLINE sc16_t item32_to_sc16(item32_t item, double){
     return sc16_t(
         boost::int16_t(item >> 16),
         boost::int16_t(item >> 0)
@@ -69,46 +69,38 @@ static UHD_INLINE sc16_t item32_to_sc16(item32_t item){
 /***********************************************************************
  * Convert complex float buffer to items32 (no swap)
  **********************************************************************/
-static const float shorts_per_float = float(32767);
-
-static UHD_INLINE item32_t fc32_to_item32(fc32_t num){
-    boost::uint16_t real = boost::int16_t(num.real()*shorts_per_float);
-    boost::uint16_t imag = boost::int16_t(num.imag()*shorts_per_float);
+static UHD_INLINE item32_t fc32_to_item32(fc32_t num, float scale_factor){
+    boost::uint16_t real = boost::int16_t(num.real()*scale_factor);
+    boost::uint16_t imag = boost::int16_t(num.imag()*scale_factor);
     return (item32_t(real) << 16) | (item32_t(imag) << 0);
 }
 
 /***********************************************************************
  * Convert items32 buffer to complex float
  **********************************************************************/
-static const float floats_per_short = float(1.0/shorts_per_float);
-
-static UHD_INLINE fc32_t item32_to_fc32(item32_t item){
+static UHD_INLINE fc32_t item32_to_fc32(item32_t item, float scale_factor){
     return fc32_t(
-        float(boost::int16_t(item >> 16)*floats_per_short),
-        float(boost::int16_t(item >> 0)*floats_per_short)
+        float(boost::int16_t(item >> 16)*scale_factor),
+        float(boost::int16_t(item >> 0)*scale_factor)
     );
 }
 
 /***********************************************************************
  * Convert complex double buffer to items32 (no swap)
  **********************************************************************/
-static const double shorts_per_double = double(32767);
-
-static UHD_INLINE item32_t fc64_to_item32(fc64_t num){
-    boost::uint16_t real = boost::int16_t(num.real()*shorts_per_double);
-    boost::uint16_t imag = boost::int16_t(num.imag()*shorts_per_double);
+static UHD_INLINE item32_t fc64_to_item32(fc64_t num, double scale_factor){
+    boost::uint16_t real = boost::int16_t(num.real()*scale_factor);
+    boost::uint16_t imag = boost::int16_t(num.imag()*scale_factor);
     return (item32_t(real) << 16) | (item32_t(imag) << 0);
 }
 
 /***********************************************************************
  * Convert items32 buffer to complex double
  **********************************************************************/
-static const double doubles_per_short = double(1.0/shorts_per_double);
-
-static UHD_INLINE fc64_t item32_to_fc64(item32_t item){
+static UHD_INLINE fc64_t item32_to_fc64(item32_t item, double scale_factor){
     return fc64_t(
-        float(boost::int16_t(item >> 16)*doubles_per_short),
-        float(boost::int16_t(item >> 0)*doubles_per_short)
+        float(boost::int16_t(item >> 16)*scale_factor),
+        float(boost::int16_t(item >> 0)*scale_factor)
     );
 }
 
diff --git a/host/lib/convert/convert_with_neon.cpp b/host/lib/convert/convert_with_neon.cpp
index 3d677db5b..e5f08cad9 100644
--- a/host/lib/convert/convert_with_neon.cpp
+++ b/host/lib/convert/convert_with_neon.cpp
@@ -26,7 +26,7 @@ DECLARE_CONVERTER(convert_fc32_1_to_item32_1_nswap, PRIORITY_CUSTOM){
 
     size_t i;
 
-    float32x4_t Q0 = vdupq_n_f32(shorts_per_float);
+    float32x4_t Q0 = vdupq_n_f32(float(scale_factor));
     for (i=0; i < (nsamps & ~0x03); i+=2) {
         float32x4_t Q1 = vld1q_f32(reinterpret_cast<const float *>(&input[i]));
         float32x4_t Q2 = vmulq_f32(Q1, Q0);
@@ -37,7 +37,7 @@ DECLARE_CONVERTER(convert_fc32_1_to_item32_1_nswap, PRIORITY_CUSTOM){
     }
 
     for (; i < nsamps; i++)
-        output[i] = fc32_to_item32(input[i]);
+        output[i] = fc32_to_item32(input[i], float(scale_factor));
 }
 
 DECLARE_CONVERTER(convert_item32_1_to_fc32_1_nswap, PRIORITY_CUSTOM){
@@ -46,7 +46,7 @@ DECLARE_CONVERTER(convert_item32_1_to_fc32_1_nswap, PRIORITY_CUSTOM){
 
     size_t i;
 
-    float32x4_t Q1 = vdupq_n_f32(floats_per_short);
+    float32x4_t Q1 = vdupq_n_f32(float(scale_factor));
     for (i=0; i < (nsamps & ~0x03); i+=2) {
         int16x4_t D0 = vld1_s16(reinterpret_cast<const int16_t *>(&input[i]));
         int16x4_t D1 = vrev32_s16(D0);
@@ -57,5 +57,5 @@ DECLARE_CONVERTER(convert_item32_1_to_fc32_1_nswap, PRIORITY_CUSTOM){
     }
 
     for (; i < nsamps; i++)
-        output[i] = item32_to_fc32(input[i]);
+        output[i] = item32_to_fc32(input[i], float(scale_factor));
 }
diff --git a/host/lib/convert/convert_with_sse2.cpp b/host/lib/convert/convert_with_sse2.cpp
index 96ee9134c..52beea24a 100644
--- a/host/lib/convert/convert_with_sse2.cpp
+++ b/host/lib/convert/convert_with_sse2.cpp
@@ -25,7 +25,7 @@ DECLARE_CONVERTER(convert_fc32_1_to_item32_1_nswap, PRIORITY_CUSTOM){
     const fc32_t *input = reinterpret_cast<const fc32_t *>(inputs[0]);
     item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
 
-    __m128 scalar = _mm_set_ps1(shorts_per_float);
+    __m128 scalar = _mm_set_ps1(float(scale_factor));
 
     //convert blocks of samples with intrinsics
     size_t i = 0; for (; i < (nsamps & ~0x3); i+=4){
@@ -48,7 +48,7 @@ DECLARE_CONVERTER(convert_fc32_1_to_item32_1_nswap, PRIORITY_CUSTOM){
 
     //convert remainder
     for (; i < nsamps; i++){
-        output[i] = fc32_to_item32(input[i]);
+        output[i] = fc32_to_item32(input[i], float(scale_factor));
     }
 }
 
@@ -56,7 +56,7 @@ DECLARE_CONVERTER(convert_fc32_1_to_item32_1_bswap, PRIORITY_CUSTOM){
     const fc32_t *input = reinterpret_cast<const fc32_t *>(inputs[0]);
     item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
 
-    __m128 scalar = _mm_set_ps1(shorts_per_float);
+    __m128 scalar = _mm_set_ps1(float(scale_factor));
 
     //convert blocks of samples with intrinsics
     size_t i = 0; for (; i < (nsamps & ~0x3); i+=4){
@@ -78,7 +78,7 @@ DECLARE_CONVERTER(convert_fc32_1_to_item32_1_bswap, PRIORITY_CUSTOM){
 
     //convert remainder
     for (; i < nsamps; i++){
-        output[i] = uhd::byteswap(fc32_to_item32(input[i]));
+        output[i] = uhd::byteswap(fc32_to_item32(input[i], float(scale_factor)));
     }
 }
 
@@ -86,7 +86,7 @@ DECLARE_CONVERTER(convert_item32_1_to_fc32_1_nswap, PRIORITY_CUSTOM){
     const item32_t *input = reinterpret_cast<const item32_t *>(inputs[0]);
     fc32_t *output = reinterpret_cast<fc32_t *>(outputs[0]);
 
-    __m128 scalar = _mm_set_ps1(floats_per_short/(1 << 16));
+    __m128 scalar = _mm_set_ps1(float(scale_factor)/(1 << 16));
     __m128i zeroi = _mm_setzero_si128();
 
     //convert blocks of samples with intrinsics
@@ -111,7 +111,7 @@ DECLARE_CONVERTER(convert_item32_1_to_fc32_1_nswap, PRIORITY_CUSTOM){
 
     //convert remainder
     for (; i < nsamps; i++){
-        output[i] = item32_to_fc32(input[i]);
+        output[i] = item32_to_fc32(input[i], float(scale_factor));
     }
 }
 
@@ -119,7 +119,7 @@ DECLARE_CONVERTER(convert_item32_1_to_fc32_1_bswap, PRIORITY_CUSTOM){
     const item32_t *input = reinterpret_cast<const item32_t *>(inputs[0]);
     fc32_t *output = reinterpret_cast<fc32_t *>(outputs[0]);
 
-    __m128 scalar = _mm_set_ps1(floats_per_short/(1 << 16));
+    __m128 scalar = _mm_set_ps1(float(scale_factor)/(1 << 16));
     __m128i zeroi = _mm_setzero_si128();
 
     //convert blocks of samples with intrinsics
@@ -143,6 +143,6 @@ DECLARE_CONVERTER(convert_item32_1_to_fc32_1_bswap, PRIORITY_CUSTOM){
 
     //convert remainder
     for (; i < nsamps; i++){
-        output[i] = item32_to_fc32(uhd::byteswap(input[i]));
+        output[i] = item32_to_fc32(uhd::byteswap(input[i]), float(scale_factor));
     }
 }
diff --git a/host/lib/convert/gen_convert_general.py b/host/lib/convert/gen_convert_general.py
index f03448047..8c3138bda 100644
--- a/host/lib/convert/gen_convert_general.py
+++ b/host/lib/convert/gen_convert_general.py
@@ -34,7 +34,7 @@ DECLARE_CONVERTER(convert_$(cpu_type)_1_to_item32_1_$(swap), PRIORITY_GENERAL){
     item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
 
     for (size_t i = 0; i < nsamps; i++){
-        output[i] = $(swap_fcn)($(cpu_type)_to_item32(input[i]));
+        output[i] = $(swap_fcn)($(cpu_type)_to_item32(input[i], float(scale_factor)));
     }
 }
 
@@ -43,7 +43,7 @@ DECLARE_CONVERTER(convert_item32_1_to_$(cpu_type)_1_$(swap), PRIORITY_GENERAL){
     $(cpu_type)_t *output = reinterpret_cast<$(cpu_type)_t *>(outputs[0]);
 
     for (size_t i = 0; i < nsamps; i++){
-        output[i] = item32_to_$(cpu_type)($(swap_fcn)(input[i]));
+        output[i] = item32_to_$(cpu_type)($(swap_fcn)(input[i]), float(scale_factor));
     }
 }
 """
@@ -56,7 +56,7 @@ DECLARE_CONVERTER(convert_$(cpu_type)_$(width)_to_item32_1_$(swap), PRIORITY_GEN
 
     for (size_t i = 0, j = 0; i < nsamps; i++){
         #for $w in range($width)
-        output[j++] = $(swap_fcn)($(cpu_type)_to_item32(input$(w)[i]));
+        output[j++] = $(swap_fcn)($(cpu_type)_to_item32(input$(w)[i], float(scale_factor)));
         #end for
     }
 }
@@ -69,7 +69,7 @@ DECLARE_CONVERTER(convert_item32_1_to_$(cpu_type)_$(width)_$(swap), PRIORITY_GEN
 
     for (size_t i = 0, j = 0; i < nsamps; i++){
         #for $w in range($width)
-        output$(w)[i] = item32_to_$(cpu_type)($(swap_fcn)(input[j++]));
+        output$(w)[i] = item32_to_$(cpu_type)($(swap_fcn)(input[j++]), float(scale_factor));
         #end for
     }
 }
diff --git a/host/lib/transport/vrt_packet_handler.hpp b/host/lib/transport/vrt_packet_handler.hpp
index d74b2c13c..26dbe1806 100644
--- a/host/lib/transport/vrt_packet_handler.hpp
+++ b/host/lib/transport/vrt_packet_handler.hpp
@@ -203,7 +203,7 @@ template <typename T> UHD_INLINE T get_context_code(
             }
 
             //copy-convert the samples from the recv buffer
-            converter(state.copy_buffs[i], state.io_buffs, nsamps_to_copy_per_io_buff);
+            converter(state.copy_buffs[i], state.io_buffs, nsamps_to_copy_per_io_buff, 1/32767.);
 
             //update the rx copy buffer to reflect the bytes copied
             state.copy_buffs[i] += bytes_to_copy;
@@ -354,7 +354,7 @@ template <typename T> UHD_INLINE T get_context_code(
             otw_mem += if_packet_info.num_header_words32;
 
             //copy-convert the samples into the send buffer
-            converter(state.io_buffs, otw_mem, num_samps);
+            converter(state.io_buffs, otw_mem, num_samps, 32767.);
 
             //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);
diff --git a/host/tests/convert_test.cpp b/host/tests/convert_test.cpp
index d1c2b7625..ff37d4a0a 100644
--- a/host/tests/convert_test.cpp
+++ b/host/tests/convert_test.cpp
@@ -55,12 +55,12 @@ template <typename Range> static void loopback(
     //convert to intermediate type
     convert::get_converter_cpu_to_otw(
         io_type, otw_type, input0.size(), output0.size()
-    )(input0, output0, nsamps);
+    )(input0, output0, nsamps, 1/32767.);
 
     //convert back to host type
     convert::get_converter_otw_to_cpu(
         io_type, otw_type, input1.size(), output1.size()
-    )(input1, output1, nsamps);
+    )(input1, output1, nsamps, 32767.);
 }
 
 /***********************************************************************
@@ -207,12 +207,12 @@ BOOST_AUTO_TEST_CASE(test_convert_types_fc32_to_sc16){
     //convert float to intermediate
     convert::get_converter_cpu_to_otw(
         io_type_in, otw_type, input0.size(), output0.size()
-    )(input0, output0, nsamps);
+    )(input0, output0, nsamps, 1/32767.);
 
     //convert intermediate to short
     convert::get_converter_otw_to_cpu(
         io_type_out, otw_type, input1.size(), output1.size()
-    )(input1, output1, nsamps);
+    )(input1, output1, nsamps, 32767.);
 
     //test that the inputs and outputs match
     for (size_t i = 0; i < nsamps; i++){
@@ -247,12 +247,12 @@ BOOST_AUTO_TEST_CASE(test_convert_types_sc16_to_fc32){
     //convert short to intermediate
     convert::get_converter_cpu_to_otw(
         io_type_in, otw_type, input0.size(), output0.size()
-    )(input0, output0, nsamps);
+    )(input0, output0, nsamps, 32767.);
 
     //convert intermediate to float
     convert::get_converter_otw_to_cpu(
         io_type_out, otw_type, input1.size(), output1.size()
-    )(input1, output1, nsamps);
+    )(input1, output1, nsamps, 1/32767.);
 
     //test that the inputs and outputs match
     for (size_t i = 0; i < nsamps; i++){