Merge branch 'next'

10 files changed, 508 insertions, 14 deletions

diff --git a/host/lib/convert/CMakeLists.txt b/host/lib/convert/CMakeLists.txt
index 98907dc29..c42a0a434 100644
--- a/host/lib/convert/CMakeLists.txt
+++ b/host/lib/convert/CMakeLists.txt
@@ -1,5 +1,5 @@
 #
-# Copyright 2011 Ettus Research LLC
+# Copyright 2011-2012 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
@@ -73,6 +73,8 @@ IF(HAVE_EMMINTRIN_H)
     SET(convert_with_sse2_sources
         ${CMAKE_CURRENT_SOURCE_DIR}/convert_fc32_with_sse2.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/convert_fc64_with_sse2.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/convert_fc32_to_sc8_with_sse2.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/convert_fc64_to_sc8_with_sse2.cpp
     )
     SET_SOURCE_FILES_PROPERTIES(
         ${convert_with_sse2_sources}
diff --git a/host/lib/convert/convert_common.hpp b/host/lib/convert/convert_common.hpp
index 699d6301b..55bc2e99d 100644
--- a/host/lib/convert/convert_common.hpp
+++ b/host/lib/convert/convert_common.hpp
@@ -1,5 +1,5 @@
 //
-// Copyright 2011 Ettus Research LLC
+// Copyright 2011-2012 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
@@ -190,4 +190,52 @@ static UHD_INLINE void item32_sc8_to_fc64(item32_t item, fc64_t &out0, fc64_t &o
     );
 }
 
+/***********************************************************************
+ * Convert complex char to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t sc8_to_item32_sc8(sc8_t in0, sc8_t in1, double){
+    return
+        (item32_t(boost::uint8_t(in0.real())) << 8) |
+        (item32_t(boost::uint8_t(in0.imag())) << 0) |
+        (item32_t(boost::uint8_t(in1.real())) << 24) |
+        (item32_t(boost::uint8_t(in1.imag())) << 16)
+    ;
+}
+
+/***********************************************************************
+ * Convert complex short to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t sc16_to_item32_sc8(sc16_t in0, sc16_t in1, double){
+    return
+        (item32_t(boost::uint8_t(in0.real())) << 8) |
+        (item32_t(boost::uint8_t(in0.imag())) << 0) |
+        (item32_t(boost::uint8_t(in1.real())) << 24) |
+        (item32_t(boost::uint8_t(in1.imag())) << 16)
+    ;
+}
+
+/***********************************************************************
+ * Convert complex float to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t fc32_to_item32_sc8(fc32_t in0, fc32_t in1, double scale_factor){
+    return
+        (item32_t(boost::uint8_t(in0.real()*float(scale_factor))) << 8) |
+        (item32_t(boost::uint8_t(in0.imag()*float(scale_factor))) << 0) |
+        (item32_t(boost::uint8_t(in1.real()*float(scale_factor))) << 24) |
+        (item32_t(boost::uint8_t(in1.imag()*float(scale_factor))) << 16)
+    ;
+}
+
+/***********************************************************************
+ * Convert complex double to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t fc64_to_item32_sc8(fc64_t in0, fc64_t in1, double scale_factor){
+    return
+        (item32_t(boost::uint8_t(in0.real()*(scale_factor))) << 8) |
+        (item32_t(boost::uint8_t(in0.imag()*(scale_factor))) << 0) |
+        (item32_t(boost::uint8_t(in1.real()*(scale_factor))) << 24) |
+        (item32_t(boost::uint8_t(in1.imag()*(scale_factor))) << 16)
+    ;
+}
+
 #endif /* INCLUDED_LIBUHD_CONVERT_COMMON_HPP */
diff --git a/host/lib/convert/convert_fc32_to_sc8_with_sse2.cpp b/host/lib/convert/convert_fc32_to_sc8_with_sse2.cpp
new file mode 100644
index 000000000..b633f487c
--- /dev/null
+++ b/host/lib/convert/convert_fc32_to_sc8_with_sse2.cpp
@@ -0,0 +1,150 @@
+//
+// Copyright 2012 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 "convert_common.hpp"
+#include <uhd/utils/byteswap.hpp>
+#include <emmintrin.h>
+
+using namespace uhd::convert;
+
+UHD_INLINE __m128i pack_sc32_4x_be(
+    const __m128 &in0, const __m128 &in1,
+    const __m128 &in2, const __m128 &in3,
+    const __m128 &scalar
+){
+    __m128i tmpi0 = _mm_cvtps_epi32(_mm_mul_ps(in0, scalar));
+    tmpi0 = _mm_shuffle_epi32(tmpi0, _MM_SHUFFLE(1, 0, 3, 2));
+    __m128i tmpi1 = _mm_cvtps_epi32(_mm_mul_ps(in1, scalar));
+    tmpi1 = _mm_shuffle_epi32(tmpi1, _MM_SHUFFLE(1, 0, 3, 2));
+    const __m128i lo = _mm_packs_epi32(tmpi0, tmpi1);
+
+    __m128i tmpi2 = _mm_cvtps_epi32(_mm_mul_ps(in2, scalar));
+    tmpi2 = _mm_shuffle_epi32(tmpi2, _MM_SHUFFLE(1, 0, 3, 2));
+    __m128i tmpi3 = _mm_cvtps_epi32(_mm_mul_ps(in3, scalar));
+    tmpi3 = _mm_shuffle_epi32(tmpi3, _MM_SHUFFLE(1, 0, 3, 2));
+    const __m128i hi = _mm_packs_epi32(tmpi2, tmpi3);
+
+    return _mm_packs_epi16(lo, hi);
+}
+
+UHD_INLINE __m128i pack_sc32_4x_le(
+    const __m128 &in0, const __m128 &in1,
+    const __m128 &in2, const __m128 &in3,
+    const __m128 &scalar
+){
+    __m128i tmpi0 = _mm_cvtps_epi32(_mm_mul_ps(in0, scalar));
+    tmpi0 = _mm_shuffle_epi32(tmpi0, _MM_SHUFFLE(2, 3, 0, 1));
+    __m128i tmpi1 = _mm_cvtps_epi32(_mm_mul_ps(in1, scalar));
+    tmpi1 = _mm_shuffle_epi32(tmpi1, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i lo = _mm_packs_epi32(tmpi0, tmpi1);
+
+    __m128i tmpi2 = _mm_cvtps_epi32(_mm_mul_ps(in2, scalar));
+    tmpi2 = _mm_shuffle_epi32(tmpi2, _MM_SHUFFLE(2, 3, 0, 1));
+    __m128i tmpi3 = _mm_cvtps_epi32(_mm_mul_ps(in3, scalar));
+    tmpi3 = _mm_shuffle_epi32(tmpi3, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i hi = _mm_packs_epi32(tmpi2, tmpi3);
+
+    return _mm_packs_epi16(lo, hi);
+}
+
+DECLARE_CONVERTER(fc32, 1, sc8_item32_be, 1, PRIORITY_SIMD){
+    const fc32_t *input = reinterpret_cast<const fc32_t *>(inputs[0]);
+    item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+    const __m128 scalar = _mm_set_ps1(float(scale_factor));
+
+    #define convert_fc32_1_to_sc8_item32_1_bswap_guts(_al_)             \
+    for (size_t j = 0; i+7 < nsamps; i+=8, j+=4){                       \
+        /* load from input */                                           \
+        __m128 tmp0 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+0)); \
+        __m128 tmp1 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+2)); \
+        __m128 tmp2 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+4)); \
+        __m128 tmp3 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+6)); \
+                                                                        \
+        /* convert */                                                   \
+        const __m128i tmpi = pack_sc32_4x_be(tmp0, tmp1, tmp2, tmp3, scalar); \
+                                                                        \
+        /* store to output */                                           \
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(output+j), tmpi);  \
+    }                                                                   \
+
+    size_t i = 0;
+
+    //dispatch according to alignment
+    if ((size_t(input) & 0xf) == 0){
+        convert_fc32_1_to_sc8_item32_1_bswap_guts(_)
+    }
+    else{
+        convert_fc32_1_to_sc8_item32_1_bswap_guts(u_)
+    }
+
+    //convert remainder
+    const size_t num_pairs = nsamps/2;
+    for (size_t j = i/2; j < num_pairs; j++, i+=2){
+        const item32_t item = fc32_to_item32_sc8(input[i], input[i+1], scale_factor);
+        output[j] = uhd::byteswap(item);
+    }
+
+    if (nsamps != num_pairs*2){
+        const item32_t item = fc32_to_item32_sc8(input[nsamps-1], 0, scale_factor);
+        output[num_pairs] = uhd::byteswap(item);
+    }
+}
+
+DECLARE_CONVERTER(fc32, 1, sc8_item32_le, 1, PRIORITY_SIMD){
+    const fc32_t *input = reinterpret_cast<const fc32_t *>(inputs[0]);
+    item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+    const __m128 scalar = _mm_set_ps1(float(scale_factor));
+
+    #define convert_fc32_1_to_sc8_item32_1_nswap_guts(_al_)             \
+    for (size_t j = 0; i+7 < nsamps; i+=8, j+=4){                       \
+        /* load from input */                                           \
+        __m128 tmp0 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+0)); \
+        __m128 tmp1 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+2)); \
+        __m128 tmp2 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+4)); \
+        __m128 tmp3 = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+6)); \
+                                                                        \
+        /* convert */                                                   \
+        const __m128i tmpi = pack_sc32_4x_le(tmp0, tmp1, tmp2, tmp3, scalar); \
+                                                                        \
+        /* store to output */                                           \
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(output+j), tmpi);  \
+    }                                                                   \
+
+    size_t i = 0;
+
+    //dispatch according to alignment
+    if ((size_t(input) & 0xf) == 0){
+        convert_fc32_1_to_sc8_item32_1_nswap_guts(_)
+    }
+    else{
+        convert_fc32_1_to_sc8_item32_1_nswap_guts(u_)
+    }
+
+    //convert remainder
+    const size_t num_pairs = nsamps/2;
+    for (size_t j = i/2; j < num_pairs; j++, i+=2){
+        const item32_t item = fc32_to_item32_sc8(input[i], input[i+1], scale_factor);
+        output[j] = (item);
+    }
+
+    if (nsamps != num_pairs*2){
+        const item32_t item = fc32_to_item32_sc8(input[nsamps-1], 0, scale_factor);
+        output[num_pairs] = (item);
+    }
+}
diff --git a/host/lib/convert/convert_fc32_with_sse2.cpp b/host/lib/convert/convert_fc32_with_sse2.cpp
index 24a939d6c..97a3e8cdc 100644
--- a/host/lib/convert/convert_fc32_with_sse2.cpp
+++ b/host/lib/convert/convert_fc32_with_sse2.cpp
@@ -28,7 +28,7 @@ DECLARE_CONVERTER(fc32, 1, sc16_item32_le, 1, PRIORITY_SIMD){
     const __m128 scalar = _mm_set_ps1(float(scale_factor));
 
     #define convert_fc32_1_to_item32_1_nswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128 tmplo = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+0)); \
         __m128 tmphi = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+2)); \
@@ -71,7 +71,7 @@ DECLARE_CONVERTER(fc32, 1, sc16_item32_be, 1, PRIORITY_SIMD){
     const __m128 scalar = _mm_set_ps1(float(scale_factor));
 
     #define convert_fc32_1_to_item32_1_bswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128 tmplo = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+0)); \
         __m128 tmphi = _mm_load ## _al_ ## ps(reinterpret_cast<const float *>(input+i+2)); \
@@ -114,7 +114,7 @@ DECLARE_CONVERTER(sc16_item32_le, 1, fc32, 1, PRIORITY_SIMD){
     const __m128i zeroi = _mm_setzero_si128();
 
     #define convert_item32_1_to_fc32_1_nswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
                                                                         \
@@ -159,7 +159,7 @@ DECLARE_CONVERTER(sc16_item32_be, 1, fc32, 1, PRIORITY_SIMD){
     const __m128i zeroi = _mm_setzero_si128();
 
     #define convert_item32_1_to_fc32_1_bswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
                                                                         \
diff --git a/host/lib/convert/convert_fc64_to_sc8_with_sse2.cpp b/host/lib/convert/convert_fc64_to_sc8_with_sse2.cpp
new file mode 100644
index 000000000..405850601
--- /dev/null
+++ b/host/lib/convert/convert_fc64_to_sc8_with_sse2.cpp
@@ -0,0 +1,156 @@
+//
+// Copyright 2012 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 "convert_common.hpp"
+#include <uhd/utils/byteswap.hpp>
+#include <emmintrin.h>
+
+using namespace uhd::convert;
+
+UHD_INLINE __m128i pack_sc8_item32_4x(
+    const __m128i &in0, const __m128i &in1,
+    const __m128i &in2, const __m128i &in3
+){
+    const __m128i lo = _mm_packs_epi32(in0, in1);
+    const __m128i hi = _mm_packs_epi32(in2, in3);
+    return _mm_packs_epi16(lo, hi);
+}
+
+UHD_INLINE __m128i pack_sc32_4x_be(
+    const __m128d &lo, const __m128d &hi,
+    const __m128d &scalar
+){
+    const __m128i tmpi_lo = _mm_cvttpd_epi32(_mm_mul_pd(hi, scalar));
+    const __m128i tmpi_hi = _mm_cvttpd_epi32(_mm_mul_pd(lo, scalar));
+    return _mm_unpacklo_epi64(tmpi_lo, tmpi_hi);
+}
+
+UHD_INLINE __m128i pack_sc32_4x_le(
+    const __m128d &lo, const __m128d &hi,
+    const __m128d &scalar
+){
+    const __m128i tmpi_lo = _mm_cvttpd_epi32(_mm_mul_pd(lo, scalar));
+    const __m128i tmpi_hi = _mm_cvttpd_epi32(_mm_mul_pd(hi, scalar));
+    const __m128i tmpi = _mm_unpacklo_epi64(tmpi_lo, tmpi_hi);
+    return _mm_shuffle_epi32(tmpi, _MM_SHUFFLE(2, 3, 0, 1));
+}
+
+DECLARE_CONVERTER(fc64, 1, sc8_item32_be, 1, PRIORITY_SIMD){
+    const fc64_t *input = reinterpret_cast<const fc64_t *>(inputs[0]);
+    item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+    const __m128d scalar = _mm_set1_pd(scale_factor);
+
+    #define convert_fc64_1_to_sc8_item32_1_bswap_guts(_al_)             \
+    for (size_t j = 0; i+7 < nsamps; i+=8, j+=4){                       \
+        /* load from input */                                           \
+        __m128d tmp0 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+0)); \
+        __m128d tmp1 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+1)); \
+        __m128d tmp2 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+2)); \
+        __m128d tmp3 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+3)); \
+        __m128d tmp4 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+4)); \
+        __m128d tmp5 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+5)); \
+        __m128d tmp6 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+6)); \
+        __m128d tmp7 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+7)); \
+                                                                        \
+        /* interleave */                                                \
+        const __m128i tmpi = pack_sc8_item32_4x(                        \
+            pack_sc32_4x_be(tmp0, tmp1, scalar),                        \
+            pack_sc32_4x_be(tmp2, tmp3, scalar),                        \
+            pack_sc32_4x_be(tmp4, tmp5, scalar),                        \
+            pack_sc32_4x_be(tmp6, tmp7, scalar)                         \
+        );                                                              \
+                                                                        \
+        /* store to output */                                           \
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(output+j), tmpi);  \
+    }                                                                   \
+
+    size_t i = 0;
+
+    //dispatch according to alignment
+    if ((size_t(input) & 0xf) == 0){
+        convert_fc64_1_to_sc8_item32_1_bswap_guts(_)
+    }
+    else{
+        convert_fc64_1_to_sc8_item32_1_bswap_guts(u_)
+    }
+
+    //convert remainder
+    const size_t num_pairs = nsamps/2;
+    for (size_t j = i/2; j < num_pairs; j++, i+=2){
+        const item32_t item = fc64_to_item32_sc8(input[i], input[i+1], scale_factor);
+        output[j] = uhd::byteswap(item);
+    }
+
+    if (nsamps != num_pairs*2){
+        const item32_t item = fc64_to_item32_sc8(input[nsamps-1], 0, scale_factor);
+        output[num_pairs] = uhd::byteswap(item);
+    }
+}
+
+DECLARE_CONVERTER(fc64, 1, sc8_item32_le, 1, PRIORITY_SIMD){
+    const fc64_t *input = reinterpret_cast<const fc64_t *>(inputs[0]);
+    item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+    const __m128d scalar = _mm_set1_pd(scale_factor);
+
+    #define convert_fc64_1_to_sc8_item32_1_nswap_guts(_al_)             \
+    for (size_t j = 0; i+7 < nsamps; i+=8, j+=4){                       \
+        /* load from input */                                           \
+        __m128d tmp0 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+0)); \
+        __m128d tmp1 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+1)); \
+        __m128d tmp2 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+2)); \
+        __m128d tmp3 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+3)); \
+        __m128d tmp4 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+4)); \
+        __m128d tmp5 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+5)); \
+        __m128d tmp6 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+6)); \
+        __m128d tmp7 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+7)); \
+                                                                        \
+        /* interleave */                                                \
+        const __m128i tmpi = pack_sc8_item32_4x(                        \
+            pack_sc32_4x_le(tmp0, tmp1, scalar),                        \
+            pack_sc32_4x_le(tmp2, tmp3, scalar),                        \
+            pack_sc32_4x_le(tmp4, tmp5, scalar),                        \
+            pack_sc32_4x_le(tmp6, tmp7, scalar)                         \
+        );                                                              \
+                                                                        \
+        /* store to output */                                           \
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(output+j), tmpi);  \
+    }                                                                   \
+
+    size_t i = 0;
+
+    //dispatch according to alignment
+    if ((size_t(input) & 0xf) == 0){
+        convert_fc64_1_to_sc8_item32_1_nswap_guts(_)
+    }
+    else{
+        convert_fc64_1_to_sc8_item32_1_nswap_guts(u_)
+    }
+
+    //convert remainder
+    const size_t num_pairs = nsamps/2;
+    for (size_t j = i/2; j < num_pairs; j++, i+=2){
+        const item32_t item = fc64_to_item32_sc8(input[i], input[i+1], scale_factor);
+        output[j] = (item);
+    }
+
+    if (nsamps != num_pairs*2){
+        const item32_t item = fc64_to_item32_sc8(input[nsamps-1], 0, scale_factor);
+        output[num_pairs] = (item);
+    }
+}
diff --git a/host/lib/convert/convert_fc64_with_sse2.cpp b/host/lib/convert/convert_fc64_with_sse2.cpp
index 837bb584e..6e097e380 100644
--- a/host/lib/convert/convert_fc64_with_sse2.cpp
+++ b/host/lib/convert/convert_fc64_with_sse2.cpp
@@ -28,7 +28,7 @@ DECLARE_CONVERTER(fc64, 1, sc16_item32_le, 1, PRIORITY_SIMD){
     const __m128d scalar = _mm_set1_pd(scale_factor);
 
     #define convert_fc64_1_to_item32_1_nswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128d tmp0 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+0)); \
         __m128d tmp1 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+1)); \
@@ -75,7 +75,7 @@ DECLARE_CONVERTER(fc64, 1, sc16_item32_be, 1, PRIORITY_SIMD){
     const __m128d scalar = _mm_set1_pd(scale_factor);
 
     #define convert_fc64_1_to_item32_1_bswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128d tmp0 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+0)); \
         __m128d tmp1 = _mm_load ## _al_ ## pd(reinterpret_cast<const double *>(input+i+1)); \
@@ -122,7 +122,7 @@ DECLARE_CONVERTER(sc16_item32_le, 1, fc64, 1, PRIORITY_SIMD){
     const __m128i zeroi = _mm_setzero_si128();
 
     #define convert_item32_1_to_fc64_1_nswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
                                                                         \
@@ -171,7 +171,7 @@ DECLARE_CONVERTER(sc16_item32_be, 1, fc64, 1, PRIORITY_SIMD){
     const __m128i zeroi = _mm_setzero_si128();
 
     #define convert_item32_1_to_fc64_1_bswap_guts(_al_)                 \
-    for (; i+4 < nsamps; i+=4){                                         \
+    for (; i+3 < nsamps; i+=4){                                         \
         /* load from input */                                           \
         __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
                                                                         \
diff --git a/host/lib/convert/convert_orc.orc b/host/lib/convert/convert_orc.orc
index 5450bf4db..f7075606e 100644
--- a/host/lib/convert/convert_orc.orc
+++ b/host/lib/convert/convert_orc.orc
@@ -61,3 +61,20 @@ x2 swapw dst, tmp
 .temp 4 tmp
 x2 swapw tmp, src
 swapl dst, tmp
+
+.function _convert_swap_byte_pairs_orc
+.source 4 src
+.dest 4 dst
+swapl dst, src
+
+.function _convert_fc32_1_to_sc8_1_nswap_orc
+.source 8 src
+.dest 2 dst
+.temp 8 tmp
+.temp 4 tmp2
+.floatparam 4 scalar
+x2 mulf tmp, src, scalar
+x2 convfl tmp, tmp
+swaplq tmp, tmp
+x2 convlw tmp2, tmp
+x2 convwb dst, tmp2
diff --git a/host/lib/convert/convert_with_orc.cpp b/host/lib/convert/convert_with_orc.cpp
index 0c46bcf1e..e44c8ca73 100644
--- a/host/lib/convert/convert_with_orc.cpp
+++ b/host/lib/convert/convert_with_orc.cpp
@@ -27,6 +27,8 @@ extern void _convert_item32_1_to_fc32_1_nswap_orc(void *, const void *, float, i
 extern void _convert_item32_1_to_fc32_1_bswap_orc(void *, const void *, float, int);
 extern void _convert_sc16_1_to_item32_1_nswap_orc(void *, const void *, float, int);
 extern void _convert_item32_1_to_sc16_1_nswap_orc(void *, const void *, float, int);
+extern void _convert_fc32_1_to_sc8_1_nswap_orc(void *, const void *, float, int);
+extern void _convert_swap_byte_pairs_orc(void *, const void *, int);
 }
 
 DECLARE_CONVERTER(fc32, 1, sc16_item32_le, 1, PRIORITY_LIBORC){
@@ -52,3 +54,12 @@ DECLARE_CONVERTER(sc16, 1, sc16_item32_le, 1, PRIORITY_LIBORC){
 DECLARE_CONVERTER(sc16_item32_le, 1, sc16, 1, PRIORITY_LIBORC){
     _convert_item32_1_to_sc16_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
 }
+
+DECLARE_CONVERTER(fc32, 1, sc8_item32_be, 1, PRIORITY_LIBORC){
+    _convert_fc32_1_to_sc8_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+    _convert_swap_byte_pairs_orc(outputs[0], outputs[0], (nsamps + 1)/2);
+}
+
+DECLARE_CONVERTER(fc32, 1, sc8_item32_le, 1, PRIORITY_LIBORC){
+    _convert_fc32_1_to_sc8_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
diff --git a/host/lib/convert/convert_with_tables.cpp b/host/lib/convert/convert_with_tables.cpp
index c45415d5d..4a3ce29b2 100644
--- a/host/lib/convert/convert_with_tables.cpp
+++ b/host/lib/convert/convert_with_tables.cpp
@@ -1,5 +1,5 @@
 //
-// Copyright 2011 Ettus Research LLC
+// Copyright 2011-2012 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
@@ -17,6 +17,7 @@
 
 #include "convert_common.hpp"
 #include <uhd/utils/byteswap.hpp>
+#include <boost/math/special_functions/round.hpp>
 #include <vector>
 
 using namespace uhd::convert;
@@ -26,6 +27,55 @@ static const size_t sc16_table_len = size_t(1 << 16);
 typedef boost::uint16_t (*tohost16_type)(boost::uint16_t);
 
 /***********************************************************************
+ * Implementation for sc16 to sc8 lookup table
+ *  - Lookup the real and imaginary parts individually
+ **********************************************************************/
+template <bool swap>
+class convert_sc16_1_to_sc8_item32_1 : public converter{
+public:
+    convert_sc16_1_to_sc8_item32_1(void): _table(sc16_table_len){}
+
+    void set_scalar(const double scalar){
+        for (size_t i = 0; i < sc16_table_len; i++){
+            const boost::int16_t val = boost::uint16_t(i);
+            _table[i] = boost::int8_t(boost::math::iround(val * scalar / 32767.));
+        }
+    }
+
+    void operator()(const input_type &inputs, const output_type &outputs, const size_t nsamps){
+        const sc16_t *input = reinterpret_cast<const sc16_t *>(inputs[0]);
+        item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+        const size_t num_pairs = nsamps/2;
+        for (size_t i = 0, j = 0; i < num_pairs; i++, j+=2){
+            output[i] = this->lookup(input[j], input[j+1]);
+        }
+
+        if (nsamps != num_pairs*2){
+            output[num_pairs] = this->lookup(input[nsamps-1], 0);;
+        }
+    }
+
+    item32_t lookup(const sc16_t &in0, const sc16_t &in1){
+        if (swap){ //hope this compiles out, its a template constant
+            return
+            (item32_t(_table[size_t(in0.real())]) << 16) |
+            (item32_t(_table[size_t(in0.imag())]) << 24) |
+            (item32_t(_table[size_t(in1.real())]) << 0) |
+            (item32_t(_table[size_t(in1.imag())]) << 8) ;
+        }
+        return
+            (item32_t(_table[size_t(in0.real())]) << 8) |
+            (item32_t(_table[size_t(in0.imag())]) << 0) |
+            (item32_t(_table[size_t(in1.real())]) << 24) |
+            (item32_t(_table[size_t(in1.imag())]) << 16) ;
+    }
+
+private:
+    std::vector<boost::uint8_t> _table;
+};
+
+/***********************************************************************
  * Implementation for sc16 lookup table
  *  - Lookup the real and imaginary parts individually
  **********************************************************************/
@@ -67,11 +117,19 @@ class convert_sc8_item32_1_to_fcxx_1 : public converter{
 public:
     convert_sc8_item32_1_to_fcxx_1(void): _table(sc16_table_len){}
 
+    //special case for sc16 type, 32767 undoes float normalization
+    static type conv(const boost::int8_t &num, const double scalar){
+        if (sizeof(type) == sizeof(s16_t)){
+            return type(boost::math::iround(num*scalar*32767));
+        }
+        return type(num*scalar);
+    }
+
     void set_scalar(const double scalar){
         for (size_t i = 0; i < sc16_table_len; i++){
             const boost::uint16_t val = tohost(boost::uint16_t(i & 0xffff));
-            const type real = type(boost::int8_t(val >> 8)*scalar);
-            const type imag = type(boost::int8_t(val >> 0)*scalar);
+            const type real = conv(boost::int8_t(val >> 8), scalar);
+            const type imag = conv(boost::int8_t(val >> 0), scalar);
             _table[i] = std::complex<type>(real, imag);
         }
     }
@@ -112,9 +170,13 @@ private:
 #ifdef BOOST_BIG_ENDIAN
 #  define SHIFT_PAIR0 16, 0
 #  define SHIFT_PAIR1 0, 16
+#  define BE_SWAP false
+#  define LE_SWAP true
 #else
 #  define SHIFT_PAIR0 0, 16
 #  define SHIFT_PAIR1 16, 0
+#  define BE_SWAP true
+#  define LE_SWAP false
 #endif
 
 static converter::sptr make_convert_sc16_item32_be_1_to_fc32_1(void){
@@ -149,6 +211,22 @@ static converter::sptr make_convert_sc8_item32_le_1_to_fc64_1(void){
     return converter::sptr(new convert_sc8_item32_1_to_fcxx_1<double, uhd::wtohx, SHIFT_PAIR0>());
 }
 
+static converter::sptr make_convert_sc8_item32_be_1_to_sc16_1(void){
+    return converter::sptr(new convert_sc8_item32_1_to_fcxx_1<s16_t, uhd::ntohx, SHIFT_PAIR1>());
+}
+
+static converter::sptr make_convert_sc8_item32_le_1_to_sc16_1(void){
+    return converter::sptr(new convert_sc8_item32_1_to_fcxx_1<s16_t, uhd::wtohx, SHIFT_PAIR0>());
+}
+
+static converter::sptr make_convert_sc16_1_to_sc8_item32_be_1(void){
+    return converter::sptr(new convert_sc16_1_to_sc8_item32_1<BE_SWAP>());
+}
+
+static converter::sptr make_convert_sc16_1_to_sc8_item32_le_1(void){
+    return converter::sptr(new convert_sc16_1_to_sc8_item32_1<LE_SWAP>());
+}
+
 UHD_STATIC_BLOCK(register_convert_sc16_item32_1_to_fcxx_1){
     uhd::convert::id_type id;
     id.num_inputs = 1;
@@ -185,4 +263,20 @@ UHD_STATIC_BLOCK(register_convert_sc16_item32_1_to_fcxx_1){
     id.output_format = "fc64";
     id.input_format = "sc8_item32_le";
     uhd::convert::register_converter(id, &make_convert_sc8_item32_le_1_to_fc64_1, PRIORITY_TABLE);
+
+    id.output_format = "sc16";
+    id.input_format = "sc8_item32_be";
+    uhd::convert::register_converter(id, &make_convert_sc8_item32_be_1_to_sc16_1, PRIORITY_TABLE);
+
+    id.output_format = "sc16";
+    id.input_format = "sc8_item32_le";
+    uhd::convert::register_converter(id, &make_convert_sc8_item32_le_1_to_sc16_1, PRIORITY_TABLE);
+
+    id.output_format = "sc16";
+    id.input_format = "sc8_item32_be";
+    uhd::convert::register_converter(id, &make_convert_sc16_1_to_sc8_item32_be_1, PRIORITY_TABLE);
+
+    id.output_format = "sc16";
+    id.input_format = "sc8_item32_le";
+    uhd::convert::register_converter(id, &make_convert_sc16_1_to_sc8_item32_le_1, PRIORITY_TABLE);
 }
diff --git a/host/lib/convert/gen_convert_general.py b/host/lib/convert/gen_convert_general.py
index 46a0c8185..364c4bd1a 100644
--- a/host/lib/convert/gen_convert_general.py
+++ b/host/lib/convert/gen_convert_general.py
@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 #
-# Copyright 2011 Ettus Research LLC
+# Copyright 2011-2012 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
@@ -92,6 +92,22 @@ DECLARE_CONVERTER(sc8_item32_$(end), 1, $(cpu_type), 1, PRIORITY_GENERAL){
         item32_sc8_to_$(cpu_type)(item_n, output[num_samps-1], dummy, scale_factor);
     }
 }
+
+DECLARE_CONVERTER($(cpu_type), 1, sc8_item32_$(end), 1, PRIORITY_GENERAL){
+    const $(cpu_type)_t *input = reinterpret_cast<const $(cpu_type)_t *>(inputs[0]);
+    item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+    const size_t num_pairs = nsamps/2;
+    for (size_t i = 0, j = 0; i < num_pairs; i++, j+=2){
+        const item32_t item = $(cpu_type)_to_item32_sc8(input[j], input[j+1], scale_factor);
+        output[i] = $(to_wire)(item);
+    }
+
+    if (nsamps != num_pairs*2){
+        const item32_t item = $(cpu_type)_to_item32_sc8(input[nsamps-1], 0, scale_factor);
+        output[num_pairs] = $(to_wire)(item);
+    }
+}
 """
 
 TMPL_CONV_USRP1_COMPLEX = """