aboutsummaryrefslogtreecommitdiffstats
path: root/host/lib/convert
diff options
context:
space:
mode:
Diffstat (limited to 'host/lib/convert')
-rw-r--r--host/lib/convert/CMakeLists.txt132
-rw-r--r--host/lib/convert/convert_common.hpp107
-rw-r--r--host/lib/convert/convert_fc32_with_sse2.cpp196
-rw-r--r--host/lib/convert/convert_fc64_with_sse2.cpp212
-rw-r--r--host/lib/convert/convert_impl.cpp109
-rw-r--r--host/lib/convert/convert_orc.orc63
-rw-r--r--host/lib/convert/convert_with_neon.cpp61
-rw-r--r--host/lib/convert/convert_with_orc.cpp54
-rw-r--r--host/lib/convert/gen_convert_general.py93
-rw-r--r--host/lib/convert/gen_convert_pred.py185
10 files changed, 1212 insertions, 0 deletions
diff --git a/host/lib/convert/CMakeLists.txt b/host/lib/convert/CMakeLists.txt
new file mode 100644
index 000000000..b260cb247
--- /dev/null
+++ b/host/lib/convert/CMakeLists.txt
@@ -0,0 +1,132 @@
+#
+# Copyright 2011 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/>.
+#
+
+########################################################################
+# This file included, use CMake directory variables
+########################################################################
+INCLUDE(CheckIncludeFileCXX)
+MESSAGE(STATUS "")
+
+########################################################################
+# Look for Orc support
+########################################################################
+FIND_PACKAGE(PkgConfig)
+IF(PKG_CONFIG_FOUND)
+PKG_CHECK_MODULES(ORC "orc-0.4 > 0.4.11")
+ENDIF(PKG_CONFIG_FOUND)
+
+FIND_PROGRAM(ORCC_EXECUTABLE orcc)
+
+LIBUHD_REGISTER_COMPONENT("ORC" ENABLE_ORC ON "ENABLE_LIBUHD;ORC_FOUND;ORCC_EXECUTABLE" OFF)
+
+IF(ENABLE_ORC)
+ INCLUDE_DIRECTORIES(${ORC_INCLUDE_DIRS})
+ LINK_DIRECTORIES(${ORC_LIBRARY_DIRS})
+ ENABLE_LANGUAGE(C)
+
+ SET(orcc_src ${CMAKE_CURRENT_SOURCE_DIR}/convert_orc.orc)
+
+ GET_FILENAME_COMPONENT(orc_file_name_we ${orcc_src} NAME_WE)
+ SET(orcc_gen ${CMAKE_CURRENT_BINARY_DIR}/${orc_file_name_we}.c)
+ MESSAGE(STATUS "Orc found, enabling Orc support.")
+ ADD_CUSTOM_COMMAND(
+ COMMAND ${ORCC_EXECUTABLE} --implementation -o ${orcc_gen} ${orcc_src}
+ DEPENDS ${orcc_src} OUTPUT ${orcc_gen}
+ )
+ LIBUHD_APPEND_SOURCES(${orcc_gen})
+ LIBUHD_APPEND_SOURCES(
+ ${CMAKE_CURRENT_SOURCE_DIR}/convert_with_orc.cpp
+ )
+ LIBUHD_APPEND_LIBS(${ORC_LIBRARIES})
+ELSE(ENABLE_ORC)
+ MESSAGE(STATUS "Orc not found, disabling orc support.")
+ENDIF(ENABLE_ORC)
+
+########################################################################
+# Check for SSE2 SIMD headers
+########################################################################
+IF(CMAKE_COMPILER_IS_GNUCXX)
+ SET(EMMINTRIN_FLAGS -msse2)
+ELSEIF(MSVC)
+ SET(EMMINTRIN_FLAGS /arch:SSE2)
+ENDIF()
+
+SET(CMAKE_REQUIRED_FLAGS ${EMMINTRIN_FLAGS})
+CHECK_INCLUDE_FILE_CXX(emmintrin.h HAVE_EMMINTRIN_H)
+UNSET(CMAKE_REQUIRED_FLAGS)
+
+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
+ )
+ SET_SOURCE_FILES_PROPERTIES(
+ ${convert_with_sse2_sources}
+ PROPERTIES COMPILE_FLAGS "${EMMINTRIN_FLAGS}"
+ )
+ LIBUHD_APPEND_SOURCES(${convert_with_sse2_sources})
+ENDIF(HAVE_EMMINTRIN_H)
+
+########################################################################
+# Check for NEON SIMD headers
+########################################################################
+IF(CMAKE_COMPILER_IS_GNUCXX)
+ SET(NEON_FLAGS "-mfloat-abi=softfp -mfpu=neon")
+ SET(CMAKE_REQUIRED_FLAGS ${NEON_FLAGS})
+ CHECK_INCLUDE_FILE_CXX(arm_neon.h HAVE_ARM_NEON_H)
+ UNSET(CMAKE_REQUIRED_FLAGS)
+ENDIF(CMAKE_COMPILER_IS_GNUCXX)
+
+IF(HAVE_ARM_NEON_H AND ENABLE_ORC)
+ #prefer orc support, its faster than the current intrinsic implementations
+ MESSAGE(STATUS "Enabled conversion support with ORC.")
+ELSEIF(HAVE_ARM_NEON_H)
+ MESSAGE(STATUS "Enabled conversion support with NEON intrinsics.")
+ SET_SOURCE_FILES_PROPERTIES(
+ ${CMAKE_CURRENT_SOURCE_DIR}/convert_with_neon.cpp
+ PROPERTIES COMPILE_FLAGS "${NEON_FLAGS}"
+ )
+ LIBUHD_APPEND_SOURCES(
+ ${CMAKE_CURRENT_SOURCE_DIR}/convert_with_neon.cpp
+ )
+ENDIF()
+
+########################################################################
+# Convert types generation
+########################################################################
+INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR})
+INCLUDE_DIRECTORIES(${CMAKE_CURRENT_BINARY_DIR})
+
+LIBUHD_PYTHON_GEN_SOURCE(
+ ${CMAKE_CURRENT_SOURCE_DIR}/gen_convert_pred.py
+ ${CMAKE_CURRENT_BINARY_DIR}/convert_pred.hpp
+)
+
+INCLUDE(AddFileDependencies)
+ADD_FILE_DEPENDENCIES(
+ ${CMAKE_CURRENT_SOURCE_DIR}/convert_impl.cpp
+ ${CMAKE_CURRENT_BINARY_DIR}/convert_pred.hpp
+)
+
+LIBUHD_PYTHON_GEN_SOURCE(
+ ${CMAKE_CURRENT_SOURCE_DIR}/gen_convert_general.py
+ ${CMAKE_CURRENT_BINARY_DIR}/convert_general.cpp
+)
+
+LIBUHD_APPEND_SOURCES(
+ ${CMAKE_CURRENT_SOURCE_DIR}/convert_impl.cpp
+)
diff --git a/host/lib/convert/convert_common.hpp b/host/lib/convert/convert_common.hpp
new file mode 100644
index 000000000..7f513b124
--- /dev/null
+++ b/host/lib/convert/convert_common.hpp
@@ -0,0 +1,107 @@
+//
+// Copyright 2011 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/>.
+//
+
+#ifndef INCLUDED_LIBUHD_CONVERT_COMMON_HPP
+#define INCLUDED_LIBUHD_CONVERT_COMMON_HPP
+
+#include <uhd/convert.hpp>
+#include <uhd/utils/static.hpp>
+#include <boost/cstdint.hpp>
+#include <complex>
+
+#define DECLARE_CONVERTER(fcn, prio) \
+ static void fcn( \
+ const uhd::convert::input_type &inputs, \
+ const uhd::convert::output_type &outputs, \
+ size_t nsamps, double scale_factor \
+ ); \
+ UHD_STATIC_BLOCK(register_##fcn##_##prio){ \
+ uhd::convert::register_converter(#fcn, fcn, prio); \
+ } \
+ static void fcn( \
+ const uhd::convert::input_type &inputs, \
+ const uhd::convert::output_type &outputs, \
+ size_t nsamps, double scale_factor \
+ )
+
+/***********************************************************************
+ * Typedefs
+ **********************************************************************/
+typedef std::complex<double> fc64_t;
+typedef std::complex<float> fc32_t;
+typedef std::complex<boost::int16_t> sc16_t;
+typedef std::complex<boost::int8_t> sc8_t;
+typedef boost::uint32_t item32_t;
+
+/***********************************************************************
+ * Convert complex short buffer to items32
+ **********************************************************************/
+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);
+}
+
+/***********************************************************************
+ * Convert items32 buffer to complex short
+ **********************************************************************/
+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)
+ );
+}
+
+/***********************************************************************
+ * Convert complex float buffer to items32 (no swap)
+ **********************************************************************/
+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 UHD_INLINE fc32_t item32_to_fc32(item32_t item, float scale_factor){
+ return fc32_t(
+ 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 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 UHD_INLINE fc64_t item32_to_fc64(item32_t item, double scale_factor){
+ return fc64_t(
+ float(boost::int16_t(item >> 16)*scale_factor),
+ float(boost::int16_t(item >> 0)*scale_factor)
+ );
+}
+
+#endif /* INCLUDED_LIBUHD_CONVERT_COMMON_HPP */
diff --git a/host/lib/convert/convert_fc32_with_sse2.cpp b/host/lib/convert/convert_fc32_with_sse2.cpp
new file mode 100644
index 000000000..676e1561c
--- /dev/null
+++ b/host/lib/convert/convert_fc32_with_sse2.cpp
@@ -0,0 +1,196 @@
+//
+// Copyright 2011 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;
+
+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]);
+
+ 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){ \
+ /* 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)); \
+ \
+ /* convert and scale */ \
+ __m128i tmpilo = _mm_cvtps_epi32(_mm_mul_ps(tmplo, scalar)); \
+ __m128i tmpihi = _mm_cvtps_epi32(_mm_mul_ps(tmphi, scalar)); \
+ \
+ /* pack + swap 16-bit pairs */ \
+ __m128i tmpi = _mm_packs_epi32(tmpilo, tmpihi); \
+ tmpi = _mm_shufflelo_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ tmpi = _mm_shufflehi_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ \
+ /* store to output */ \
+ _mm_storeu_si128(reinterpret_cast<__m128i *>(output+i), tmpi); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ switch (size_t(input) & 0xf){
+ case 0x8:
+ output[i] = fc32_to_item32(input[i], float(scale_factor)); i++;
+ case 0x0:
+ convert_fc32_1_to_item32_1_nswap_guts(_)
+ break;
+ default: convert_fc32_1_to_item32_1_nswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = fc32_to_item32(input[i], float(scale_factor));
+ }
+}
+
+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]);
+
+ 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){ \
+ /* 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)); \
+ \
+ /* convert and scale */ \
+ __m128i tmpilo = _mm_cvtps_epi32(_mm_mul_ps(tmplo, scalar)); \
+ __m128i tmpihi = _mm_cvtps_epi32(_mm_mul_ps(tmphi, scalar)); \
+ \
+ /* pack + byteswap -> byteswap 16 bit words */ \
+ __m128i tmpi = _mm_packs_epi32(tmpilo, tmpihi); \
+ tmpi = _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); \
+ \
+ /* store to output */ \
+ _mm_storeu_si128(reinterpret_cast<__m128i *>(output+i), tmpi); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ switch (size_t(input) & 0xf){
+ case 0x8:
+ output[i] = uhd::byteswap(fc32_to_item32(input[i], float(scale_factor))); i++;
+ case 0x0:
+ convert_fc32_1_to_item32_1_bswap_guts(_)
+ break;
+ default: convert_fc32_1_to_item32_1_bswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = uhd::byteswap(fc32_to_item32(input[i], float(scale_factor)));
+ }
+}
+
+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]);
+
+ const __m128 scalar = _mm_set_ps1(float(scale_factor)/(1 << 16));
+ const __m128i zeroi = _mm_setzero_si128();
+
+ #define convert_item32_1_to_fc32_1_nswap_guts(_al_) \
+ for (; i+4 < nsamps; i+=4){ \
+ /* load from input */ \
+ __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
+ \
+ /* unpack + swap 16-bit pairs */ \
+ tmpi = _mm_shufflelo_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ tmpi = _mm_shufflehi_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ __m128i tmpilo = _mm_unpacklo_epi16(zeroi, tmpi); /* value in upper 16 bits */ \
+ __m128i tmpihi = _mm_unpackhi_epi16(zeroi, tmpi); \
+ \
+ /* convert and scale */ \
+ __m128 tmplo = _mm_mul_ps(_mm_cvtepi32_ps(tmpilo), scalar); \
+ __m128 tmphi = _mm_mul_ps(_mm_cvtepi32_ps(tmpihi), scalar); \
+ \
+ /* store to output */ \
+ _mm_store ## _al_ ## ps(reinterpret_cast<float *>(output+i+0), tmplo); \
+ _mm_store ## _al_ ## ps(reinterpret_cast<float *>(output+i+2), tmphi); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ switch (size_t(output) & 0xf){
+ case 0x8:
+ output[i] = item32_to_fc32(input[i], float(scale_factor)); i++;
+ case 0x0:
+ convert_item32_1_to_fc32_1_nswap_guts(_)
+ break;
+ default: convert_item32_1_to_fc32_1_nswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = item32_to_fc32(input[i], float(scale_factor));
+ }
+}
+
+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]);
+
+ const __m128 scalar = _mm_set_ps1(float(scale_factor)/(1 << 16));
+ const __m128i zeroi = _mm_setzero_si128();
+
+ #define convert_item32_1_to_fc32_1_bswap_guts(_al_) \
+ for (; i+4 < nsamps; i+=4){ \
+ /* load from input */ \
+ __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
+ \
+ /* byteswap + unpack -> byteswap 16 bit words */ \
+ tmpi = _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); \
+ __m128i tmpilo = _mm_unpacklo_epi16(zeroi, tmpi); /* value in upper 16 bits */ \
+ __m128i tmpihi = _mm_unpackhi_epi16(zeroi, tmpi); \
+ \
+ /* convert and scale */ \
+ __m128 tmplo = _mm_mul_ps(_mm_cvtepi32_ps(tmpilo), scalar); \
+ __m128 tmphi = _mm_mul_ps(_mm_cvtepi32_ps(tmpihi), scalar); \
+ \
+ /* store to output */ \
+ _mm_store ## _al_ ## ps(reinterpret_cast<float *>(output+i+0), tmplo); \
+ _mm_store ## _al_ ## ps(reinterpret_cast<float *>(output+i+2), tmphi); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ switch (size_t(output) & 0xf){
+ case 0x8:
+ output[i] = item32_to_fc32(uhd::byteswap(input[i]), float(scale_factor)); i++;
+ case 0x0:
+ convert_item32_1_to_fc32_1_bswap_guts(_)
+ break;
+ default: convert_item32_1_to_fc32_1_bswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = item32_to_fc32(uhd::byteswap(input[i]), float(scale_factor));
+ }
+}
diff --git a/host/lib/convert/convert_fc64_with_sse2.cpp b/host/lib/convert/convert_fc64_with_sse2.cpp
new file mode 100644
index 000000000..4d28396a4
--- /dev/null
+++ b/host/lib/convert/convert_fc64_with_sse2.cpp
@@ -0,0 +1,212 @@
+//
+// Copyright 2011 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;
+
+DECLARE_CONVERTER(convert_fc64_1_to_item32_1_nswap, PRIORITY_CUSTOM){
+ 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_item32_1_nswap_guts(_al_) \
+ for (; i+4 < 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)); \
+ __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)); \
+ \
+ /* convert and scale */ \
+ __m128i tmpi0 = _mm_cvttpd_epi32(_mm_mul_pd(tmp0, scalar)); \
+ __m128i tmpi1 = _mm_cvttpd_epi32(_mm_mul_pd(tmp1, scalar)); \
+ __m128i tmpilo = _mm_unpacklo_epi64(tmpi0, tmpi1); \
+ __m128i tmpi2 = _mm_cvttpd_epi32(_mm_mul_pd(tmp2, scalar)); \
+ __m128i tmpi3 = _mm_cvttpd_epi32(_mm_mul_pd(tmp3, scalar)); \
+ __m128i tmpihi = _mm_unpacklo_epi64(tmpi2, tmpi3); \
+ \
+ /* pack + swap 16-bit pairs */ \
+ __m128i tmpi = _mm_packs_epi32(tmpilo, tmpihi); \
+ tmpi = _mm_shufflelo_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ tmpi = _mm_shufflehi_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ \
+ /* store to output */ \
+ _mm_storeu_si128(reinterpret_cast<__m128i *>(output+i), tmpi); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ if ((size_t(input) & 0xf) == 0){
+ convert_fc64_1_to_item32_1_nswap_guts(_)
+ }
+ else{
+ convert_fc64_1_to_item32_1_nswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = fc64_to_item32(input[i], scale_factor);
+ }
+}
+
+DECLARE_CONVERTER(convert_fc64_1_to_item32_1_bswap, PRIORITY_CUSTOM){
+ 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_item32_1_bswap_guts(_al_) \
+ for (; i+4 < 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)); \
+ __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)); \
+ \
+ /* convert and scale */ \
+ __m128i tmpi0 = _mm_cvttpd_epi32(_mm_mul_pd(tmp0, scalar)); \
+ __m128i tmpi1 = _mm_cvttpd_epi32(_mm_mul_pd(tmp1, scalar)); \
+ __m128i tmpilo = _mm_unpacklo_epi64(tmpi0, tmpi1); \
+ __m128i tmpi2 = _mm_cvttpd_epi32(_mm_mul_pd(tmp2, scalar)); \
+ __m128i tmpi3 = _mm_cvttpd_epi32(_mm_mul_pd(tmp3, scalar)); \
+ __m128i tmpihi = _mm_unpacklo_epi64(tmpi2, tmpi3); \
+ \
+ /* pack + byteswap -> byteswap 16 bit words */ \
+ __m128i tmpi = _mm_packs_epi32(tmpilo, tmpihi); \
+ tmpi = _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); \
+ \
+ /* store to output */ \
+ _mm_storeu_si128(reinterpret_cast<__m128i *>(output+i), tmpi); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ if ((size_t(input) & 0xf) == 0){
+ convert_fc64_1_to_item32_1_bswap_guts(_)
+ }
+ else{
+ convert_fc64_1_to_item32_1_bswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = uhd::byteswap(fc64_to_item32(input[i], scale_factor));
+ }
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_fc64_1_nswap, PRIORITY_CUSTOM){
+ const item32_t *input = reinterpret_cast<const item32_t *>(inputs[0]);
+ fc64_t *output = reinterpret_cast<fc64_t *>(outputs[0]);
+
+ const __m128d scalar = _mm_set1_pd(scale_factor/(1 << 16));
+ const __m128i zeroi = _mm_setzero_si128();
+
+ #define convert_item32_1_to_fc64_1_nswap_guts(_al_) \
+ for (; i+4 < nsamps; i+=4){ \
+ /* load from input */ \
+ __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
+ \
+ /* unpack + swap 16-bit pairs */ \
+ tmpi = _mm_shufflelo_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ tmpi = _mm_shufflehi_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
+ __m128i tmpilo = _mm_unpacklo_epi16(zeroi, tmpi); /* value in upper 16 bits */ \
+ __m128i tmpihi = _mm_unpackhi_epi16(zeroi, tmpi); \
+ \
+ /* convert and scale */ \
+ __m128d tmp0 = _mm_mul_pd(_mm_cvtepi32_pd(tmpilo), scalar); \
+ tmpilo = _mm_unpackhi_epi64(tmpilo, zeroi); \
+ __m128d tmp1 = _mm_mul_pd(_mm_cvtepi32_pd(tmpilo), scalar); \
+ __m128d tmp2 = _mm_mul_pd(_mm_cvtepi32_pd(tmpihi), scalar); \
+ tmpihi = _mm_unpackhi_epi64(tmpihi, zeroi); \
+ __m128d tmp3 = _mm_mul_pd(_mm_cvtepi32_pd(tmpihi), scalar); \
+ \
+ /* store to output */ \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+0), tmp0); \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+1), tmp1); \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+2), tmp2); \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+3), tmp3); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ if ((size_t(output) & 0xf) == 0){
+ convert_item32_1_to_fc64_1_nswap_guts(_)
+ }
+ else{
+ convert_item32_1_to_fc64_1_nswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = item32_to_fc64(input[i], scale_factor);
+ }
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_fc64_1_bswap, PRIORITY_CUSTOM){
+ const item32_t *input = reinterpret_cast<const item32_t *>(inputs[0]);
+ fc64_t *output = reinterpret_cast<fc64_t *>(outputs[0]);
+
+ const __m128d scalar = _mm_set1_pd(scale_factor/(1 << 16));
+ const __m128i zeroi = _mm_setzero_si128();
+
+ #define convert_item32_1_to_fc64_1_bswap_guts(_al_) \
+ for (; i+4 < nsamps; i+=4){ \
+ /* load from input */ \
+ __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \
+ \
+ /* byteswap + unpack -> byteswap 16 bit words */ \
+ tmpi = _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); \
+ __m128i tmpilo = _mm_unpacklo_epi16(zeroi, tmpi); /* value in upper 16 bits */ \
+ __m128i tmpihi = _mm_unpackhi_epi16(zeroi, tmpi); \
+ \
+ /* convert and scale */ \
+ __m128d tmp0 = _mm_mul_pd(_mm_cvtepi32_pd(tmpilo), scalar); \
+ tmpilo = _mm_unpackhi_epi64(tmpilo, zeroi); \
+ __m128d tmp1 = _mm_mul_pd(_mm_cvtepi32_pd(tmpilo), scalar); \
+ __m128d tmp2 = _mm_mul_pd(_mm_cvtepi32_pd(tmpihi), scalar); \
+ tmpihi = _mm_unpackhi_epi64(tmpihi, zeroi); \
+ __m128d tmp3 = _mm_mul_pd(_mm_cvtepi32_pd(tmpihi), scalar); \
+ \
+ /* store to output */ \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+0), tmp0); \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+1), tmp1); \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+2), tmp2); \
+ _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+i+3), tmp3); \
+ } \
+
+ size_t i = 0;
+
+ //dispatch according to alignment
+ if ((size_t(output) & 0xf) == 0){
+ convert_item32_1_to_fc64_1_bswap_guts(_)
+ }
+ else{
+ convert_item32_1_to_fc64_1_bswap_guts(u_)
+ }
+
+ //convert remainder
+ for (; i < nsamps; i++){
+ output[i] = item32_to_fc64(uhd::byteswap(input[i]), scale_factor);
+ }
+}
diff --git a/host/lib/convert/convert_impl.cpp b/host/lib/convert/convert_impl.cpp
new file mode 100644
index 000000000..9b2cdcdc9
--- /dev/null
+++ b/host/lib/convert/convert_impl.cpp
@@ -0,0 +1,109 @@
+//
+// Copyright 2011 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/convert.hpp>
+#include <uhd/utils/log.hpp>
+#include <uhd/utils/static.hpp>
+#include <uhd/exception.hpp>
+
+using namespace uhd;
+
+#include "convert_pred.hpp"
+
+/***********************************************************************
+ * Define types for the function tables
+ **********************************************************************/
+struct fcn_table_entry_type{
+ convert::priority_type prio;
+ convert::function_type fcn;
+ fcn_table_entry_type(void)
+ : prio(convert::PRIORITY_EMPTY), fcn(NULL){
+ /* NOP */
+ }
+};
+typedef std::vector<fcn_table_entry_type> fcn_table_type;
+
+/***********************************************************************
+ * Setup the table registry
+ **********************************************************************/
+UHD_SINGLETON_FCN(fcn_table_type, get_cpu_to_otw_table);
+UHD_SINGLETON_FCN(fcn_table_type, get_otw_to_cpu_table);
+
+fcn_table_type &get_table(dir_type dir){
+ switch(dir){
+ case DIR_OTW_TO_CPU: return get_otw_to_cpu_table();
+ case DIR_CPU_TO_OTW: return get_cpu_to_otw_table();
+ }
+ UHD_THROW_INVALID_CODE_PATH();
+}
+
+/***********************************************************************
+ * The registry functions
+ **********************************************************************/
+void uhd::convert::register_converter(
+ const std::string &markup,
+ function_type fcn,
+ priority_type prio
+){
+ //extract the predicate and direction from the markup
+ dir_type dir;
+ pred_type pred = make_pred(markup, dir);
+
+ //get a reference to the function table
+ fcn_table_type &table = get_table(dir);
+
+ //resize the table so that its at least pred+1
+ if (table.size() <= pred) table.resize(pred+1);
+
+ //register the function if higher priority
+ if (table[pred].prio < prio){
+ table[pred].fcn = fcn;
+ table[pred].prio = prio;
+ }
+
+ //----------------------------------------------------------------//
+ UHD_LOGV(always) << "register_converter: " << markup << std::endl
+ << " prio: " << prio << std::endl
+ << " pred: " << pred << std::endl
+ << " dir: " << dir << std::endl
+ << std::endl
+ ;
+ //----------------------------------------------------------------//
+}
+
+/***********************************************************************
+ * The converter functions
+ **********************************************************************/
+const convert::function_type &convert::get_converter_cpu_to_otw(
+ const io_type_t &io_type,
+ const otw_type_t &otw_type,
+ size_t num_input_buffs,
+ size_t num_output_buffs
+){
+ pred_type pred = make_pred(io_type, otw_type, num_input_buffs, num_output_buffs);
+ return get_cpu_to_otw_table().at(pred).fcn;
+}
+
+const convert::function_type &convert::get_converter_otw_to_cpu(
+ const io_type_t &io_type,
+ const otw_type_t &otw_type,
+ size_t num_input_buffs,
+ size_t num_output_buffs
+){
+ pred_type pred = make_pred(io_type, otw_type, num_input_buffs, num_output_buffs);
+ return get_otw_to_cpu_table().at(pred).fcn;
+}
diff --git a/host/lib/convert/convert_orc.orc b/host/lib/convert/convert_orc.orc
new file mode 100644
index 000000000..5450bf4db
--- /dev/null
+++ b/host/lib/convert/convert_orc.orc
@@ -0,0 +1,63 @@
+.function _convert_fc32_1_to_item32_1_nswap_orc
+.source 8 src
+.dest 4 dst
+.floatparam 4 scalar
+.temp 8 scaled
+.temp 8 converted
+.temp 4 short
+x2 mulf scaled, src, scalar
+x2 convfl converted, scaled
+x2 convlw short, converted
+swapl short, short
+x2 swapw dst, short
+
+.function _convert_fc32_1_to_item32_1_bswap_orc
+.source 8 src
+.dest 4 dst
+.floatparam 4 scalar
+.temp 8 scaled
+.temp 8 converted
+.temp 4 short
+x2 mulf scaled, src, scalar
+x2 convfl converted, scaled
+x2 convlw short, converted
+x2 swapw dst, short
+
+.function _convert_item32_1_to_fc32_1_nswap_orc
+.source 4 src
+.dest 8 dst
+.floatparam 4 scalar
+.temp 4 tmp1
+.temp 8 tmp2
+x2 swapw tmp1, src
+swapl tmp1, tmp1
+x2 convswl tmp2, tmp1
+x2 convlf tmp2, tmp2
+x2 mulf dst, tmp2, scalar
+
+.function _convert_item32_1_to_fc32_1_bswap_orc
+.source 4 src
+.dest 8 dst
+.floatparam 4 scalar
+.temp 4 tmp1
+.temp 8 tmp2
+x2 swapw tmp1, src
+x2 convswl tmp2, tmp1
+x2 convlf tmp2, tmp2
+x2 mulf dst, tmp2, scalar
+
+.function _convert_sc16_1_to_item32_1_nswap_orc
+.source 4 src
+.dest 4 dst
+.temp 4 tmp
+.floatparam 4 scalar
+swapl tmp, src
+x2 swapw dst, tmp
+
+.function _convert_item32_1_to_sc16_1_nswap_orc
+.source 4 src
+.dest 4 dst
+.floatparam 4 scalar
+.temp 4 tmp
+x2 swapw tmp, src
+swapl dst, tmp
diff --git a/host/lib/convert/convert_with_neon.cpp b/host/lib/convert/convert_with_neon.cpp
new file mode 100644
index 000000000..e5f08cad9
--- /dev/null
+++ b/host/lib/convert/convert_with_neon.cpp
@@ -0,0 +1,61 @@
+//
+// Copyright 2011-2011 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 <arm_neon.h>
+
+using namespace uhd::convert;
+
+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]);
+
+ size_t i;
+
+ 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);
+ int32x4_t Q3 = vcvtq_s32_f32(Q2);
+ int16x4_t D8 = vmovn_s32(Q3);
+ int16x4_t D9 = vrev32_s16(D8);
+ vst1_s16((reinterpret_cast<int16_t *>(&output[i])), D9);
+ }
+
+ for (; i < nsamps; i++)
+ output[i] = fc32_to_item32(input[i], float(scale_factor));
+}
+
+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]);
+
+ size_t i;
+
+ 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);
+ int32x4_t Q2 = vmovl_s16(D1);
+ float32x4_t Q3 = vcvtq_f32_s32(Q2);
+ float32x4_t Q4 = vmulq_f32(Q3, Q1);
+ vst1q_f32((reinterpret_cast<float *>(&output[i])), Q4);
+ }
+
+ for (; i < nsamps; i++)
+ output[i] = item32_to_fc32(input[i], float(scale_factor));
+}
diff --git a/host/lib/convert/convert_with_orc.cpp b/host/lib/convert/convert_with_orc.cpp
new file mode 100644
index 000000000..844c2595c
--- /dev/null
+++ b/host/lib/convert/convert_with_orc.cpp
@@ -0,0 +1,54 @@
+//
+// Copyright 2011 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>
+
+using namespace uhd::convert;
+
+extern "C" {
+extern void _convert_fc32_1_to_item32_1_nswap_orc(void *, const void *, float, int);
+extern void _convert_fc32_1_to_item32_1_bswap_orc(void *, const void *, float, int);
+extern void _convert_item32_1_to_fc32_1_nswap_orc(void *, const void *, float, int);
+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);
+}
+
+DECLARE_CONVERTER(convert_fc32_1_to_item32_1_nswap, PRIORITY_LIBORC){
+ _convert_fc32_1_to_item32_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
+
+DECLARE_CONVERTER(convert_fc32_1_to_item32_1_bswap, PRIORITY_LIBORC){
+ _convert_fc32_1_to_item32_1_bswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_fc32_1_nswap, PRIORITY_LIBORC){
+ _convert_item32_1_to_fc32_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_fc32_1_bswap, PRIORITY_LIBORC){
+ _convert_item32_1_to_fc32_1_bswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
+
+DECLARE_CONVERTER(convert_sc16_1_to_item32_1_nswap, PRIORITY_LIBORC){
+ _convert_sc16_1_to_item32_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_sc16_1_nswap, PRIORITY_LIBORC){
+ _convert_item32_1_to_sc16_1_nswap_orc(outputs[0], inputs[0], scale_factor, nsamps);
+}
diff --git a/host/lib/convert/gen_convert_general.py b/host/lib/convert/gen_convert_general.py
new file mode 100644
index 000000000..8c3138bda
--- /dev/null
+++ b/host/lib/convert/gen_convert_general.py
@@ -0,0 +1,93 @@
+#!/usr/bin/env python
+#
+# Copyright 2011 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/>.
+#
+
+TMPL_HEADER = """
+#import time
+/***********************************************************************
+ * This file was generated by $file on $time.strftime("%c")
+ **********************************************************************/
+
+\#include "convert_common.hpp"
+\#include <uhd/utils/byteswap.hpp>
+
+using namespace uhd::convert;
+"""
+
+TMPL_CONV_TO_FROM_ITEM32_1 = """
+DECLARE_CONVERTER(convert_$(cpu_type)_1_to_item32_1_$(swap), PRIORITY_GENERAL){
+ const $(cpu_type)_t *input = reinterpret_cast<const $(cpu_type)_t *>(inputs[0]);
+ 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], float(scale_factor)));
+ }
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_$(cpu_type)_1_$(swap), PRIORITY_GENERAL){
+ const item32_t *input = reinterpret_cast<const item32_t *>(inputs[0]);
+ $(cpu_type)_t *output = reinterpret_cast<$(cpu_type)_t *>(outputs[0]);
+
+ for (size_t i = 0; i < nsamps; i++){
+ output[i] = item32_to_$(cpu_type)($(swap_fcn)(input[i]), float(scale_factor));
+ }
+}
+"""
+TMPL_CONV_TO_FROM_ITEM32_X = """
+DECLARE_CONVERTER(convert_$(cpu_type)_$(width)_to_item32_1_$(swap), PRIORITY_GENERAL){
+ #for $w in range($width)
+ const $(cpu_type)_t *input$(w) = reinterpret_cast<const $(cpu_type)_t *>(inputs[$(w)]);
+ #end for
+ item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+ for (size_t i = 0, j = 0; i < nsamps; i++){
+ #for $w in range($width)
+ output[j++] = $(swap_fcn)($(cpu_type)_to_item32(input$(w)[i], float(scale_factor)));
+ #end for
+ }
+}
+
+DECLARE_CONVERTER(convert_item32_1_to_$(cpu_type)_$(width)_$(swap), PRIORITY_GENERAL){
+ const item32_t *input = reinterpret_cast<const item32_t *>(inputs[0]);
+ #for $w in range($width)
+ $(cpu_type)_t *output$(w) = reinterpret_cast<$(cpu_type)_t *>(outputs[$(w)]);
+ #end for
+
+ 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++]), float(scale_factor));
+ #end for
+ }
+}
+"""
+
+def parse_tmpl(_tmpl_text, **kwargs):
+ from Cheetah.Template import Template
+ return str(Template(_tmpl_text, kwargs))
+
+if __name__ == '__main__':
+ import sys, os
+ file = os.path.basename(__file__)
+ output = parse_tmpl(TMPL_HEADER, file=file)
+ for width in 1, 2, 3, 4:
+ for swap, swap_fcn in (('nswap', ''), ('bswap', 'uhd::byteswap')):
+ for cpu_type in 'fc64', 'fc32', 'sc16':
+ output += parse_tmpl(
+ TMPL_CONV_TO_FROM_ITEM32_1 if width == 1 else TMPL_CONV_TO_FROM_ITEM32_X,
+ width=width, swap=swap, swap_fcn=swap_fcn, cpu_type=cpu_type
+ )
+ open(sys.argv[1], 'w').write(output)
diff --git a/host/lib/convert/gen_convert_pred.py b/host/lib/convert/gen_convert_pred.py
new file mode 100644
index 000000000..360fbcf44
--- /dev/null
+++ b/host/lib/convert/gen_convert_pred.py
@@ -0,0 +1,185 @@
+#!/usr/bin/env python
+#
+# Copyright 2011-2011 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/>.
+#
+
+TMPL_TEXT = """
+#import time
+/***********************************************************************
+ * This file was generated by $file on $time.strftime("%c")
+ **********************************************************************/
+\#include <uhd/exception.hpp>
+\#include <boost/tokenizer.hpp>
+\#include <boost/lexical_cast.hpp>
+\#include <boost/detail/endian.hpp>
+\#include <boost/cstdint.hpp>
+\#include <string>
+\#include <vector>
+
+typedef size_t pred_type;
+typedef std::vector<pred_type> pred_vector_type;
+
+enum dir_type{
+ DIR_OTW_TO_CPU = 0,
+ DIR_CPU_TO_OTW = 1
+};
+
+struct pred_error : uhd::value_error{
+ pred_error(const std::string &what):
+ uhd::value_error("convert::make_pred: " + what)
+ {
+ /* NOP */
+ }
+};
+
+pred_type make_pred(const std::string &markup, dir_type &dir){
+ pred_type pred = 0;
+
+ try{
+ boost::tokenizer<boost::char_separator<char> > tokenizer(markup, boost::char_separator<char>("_"));
+ std::vector<std::string> tokens(tokenizer.begin(), tokenizer.end());
+ //token 0 is <convert>
+ std::string inp_type = tokens.at(1);
+ std::string num_inps = tokens.at(2);
+ //token 3 is <to>
+ std::string out_type = tokens.at(4);
+ std::string num_outs = tokens.at(5);
+ std::string swap_type = tokens.at(6);
+
+ std::string cpu_type, otw_type;
+ if (inp_type.find("item") == std::string::npos){
+ cpu_type = inp_type;
+ otw_type = out_type;
+ dir = DIR_CPU_TO_OTW;
+ }
+ else{
+ cpu_type = out_type;
+ otw_type = inp_type;
+ dir = DIR_OTW_TO_CPU;
+ }
+
+ if (cpu_type == "fc64") pred |= $ph.fc64_p;
+ else if (cpu_type == "fc32") pred |= $ph.fc32_p;
+ else if (cpu_type == "sc16") pred |= $ph.sc16_p;
+ else if (cpu_type == "sc8") pred |= $ph.sc8_p;
+ else throw pred_error("unhandled io type " + cpu_type);
+
+ if (otw_type == "item32") pred |= $ph.item32_p;
+ else throw pred_error("unhandled otw type " + otw_type);
+
+ int num_inputs = boost::lexical_cast<int>(num_inps);
+ int num_outputs = boost::lexical_cast<int>(num_outs);
+
+ switch(num_inputs*num_outputs){ //FIXME treated as one value
+ case 1: pred |= $ph.chan1_p; break;
+ case 2: pred |= $ph.chan2_p; break;
+ case 3: pred |= $ph.chan3_p; break;
+ case 4: pred |= $ph.chan4_p; break;
+ default: throw pred_error("unhandled number of channels");
+ }
+
+ if (swap_type == "bswap") pred |= $ph.bswap_p;
+ else if (swap_type == "nswap") pred |= $ph.nswap_p;
+ else throw pred_error("unhandled swap type");
+
+ }
+ catch(...){
+ throw pred_error("could not parse markup: " + markup);
+ }
+
+ return pred;
+}
+
+#define pred_table_wildcard pred_type(~0)
+#define pred_table_max_size size_t(128)
+#define pred_table_index(e) (pred_type(e) & 0x7f)
+
+static pred_vector_type get_pred_byte_order_table(void){
+ pred_vector_type table(pred_table_max_size, pred_table_wildcard);
+ \#ifdef BOOST_BIG_ENDIAN
+ table[pred_table_index(otw_type_t::BO_BIG_ENDIAN)] = $ph.nswap_p;
+ table[pred_table_index(otw_type_t::BO_LITTLE_ENDIAN)] = $ph.bswap_p;
+ \#else
+ table[pred_table_index(otw_type_t::BO_BIG_ENDIAN)] = $ph.bswap_p;
+ table[pred_table_index(otw_type_t::BO_LITTLE_ENDIAN)] = $ph.nswap_p;
+ \#endif
+ table[pred_table_index(otw_type_t::BO_NATIVE)] = $ph.nswap_p;
+ return table;
+}
+
+static pred_vector_type get_pred_io_type_table(void){
+ pred_vector_type table(pred_table_max_size, pred_table_wildcard);
+ table[pred_table_index(io_type_t::COMPLEX_FLOAT64)] = $ph.fc64_p;
+ table[pred_table_index(io_type_t::COMPLEX_FLOAT32)] = $ph.fc32_p;
+ table[pred_table_index(io_type_t::COMPLEX_INT16)] = $ph.sc16_p;
+ return table;
+}
+
+static pred_vector_type get_pred_num_io_table(void){
+ pred_vector_type table(pred_table_max_size, pred_table_wildcard);
+ table[1] = $ph.chan1_p;
+ table[2] = $ph.chan2_p;
+ table[3] = $ph.chan3_p;
+ table[4] = $ph.chan4_p;
+ return table;
+}
+
+UHD_INLINE pred_type make_pred(
+ const io_type_t &io_type,
+ const otw_type_t &otw_type,
+ size_t num_inputs,
+ size_t num_outputs
+){
+ pred_type pred = $ph.item32_p; //only item32 supported as of now
+
+ static const pred_vector_type pred_byte_order_table(get_pred_byte_order_table());
+ pred |= pred_byte_order_table[pred_table_index(otw_type.byteorder)];
+
+ static const pred_vector_type pred_io_type_table(get_pred_io_type_table());
+ pred |= pred_io_type_table[pred_table_index(io_type.tid)];
+
+ static const pred_vector_type pred_num_io_table(get_pred_num_io_table());
+ pred |= pred_num_io_table[pred_table_index(num_inputs*num_outputs)];
+
+ if (pred == pred_table_wildcard) throw pred_error(
+ "unhanded input combination for make_pred()"
+ );
+
+ return pred;
+}
+"""
+
+def parse_tmpl(_tmpl_text, **kwargs):
+ from Cheetah.Template import Template
+ return str(Template(_tmpl_text, kwargs))
+
+class ph:
+ bswap_p = 0b00001
+ nswap_p = 0b00000
+ item32_p = 0b00000
+ sc8_p = 0b00000
+ sc16_p = 0b00010
+ fc32_p = 0b00100
+ fc64_p = 0b00110
+ chan1_p = 0b00000
+ chan2_p = 0b01000
+ chan3_p = 0b10000
+ chan4_p = 0b11000
+
+if __name__ == '__main__':
+ import sys, os
+ file = os.path.basename(__file__)
+ open(sys.argv[1], 'w').write(parse_tmpl(TMPL_TEXT, file=file, ph=ph))
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-26 06:59:30 +0000