diff options
author | Josh Blum <josh@joshknows.com> | 2011-01-04 19:46:31 -0800 |
---|---|---|
committer | Josh Blum <josh@joshknows.com> | 2011-01-04 19:46:31 -0800 |
commit | d5aef80d35bfef42c6050d76fb9d0441a1af1cb4 (patch) | |
tree | 67f6d0e7db22a05a43c2729f80252119cd5936dd | |
parent | b5860c0fbd6f18b43b1e0578adfaca1b2ef3239c (diff) | |
download | uhd-d5aef80d35bfef42c6050d76fb9d0441a1af1cb4.tar.gz uhd-d5aef80d35bfef42c6050d76fb9d0441a1af1cb4.tar.bz2 uhd-d5aef80d35bfef42c6050d76fb9d0441a1af1cb4.zip |
uhd: removed convert types, replaced by convert
-rw-r--r-- | host/include/uhd/transport/CMakeLists.txt | 2 | ||||
-rw-r--r-- | host/include/uhd/transport/convert_types.hpp | 96 | ||||
-rw-r--r-- | host/include/uhd/transport/convert_types.ipp | 43 | ||||
-rw-r--r-- | host/lib/convert/gen_convert_impl.py | 58 | ||||
-rw-r--r-- | host/lib/transport/CMakeLists.txt | 34 | ||||
-rw-r--r-- | host/lib/transport/convert_types_impl.hpp | 345 | ||||
-rwxr-xr-x | host/lib/transport/gen_convert_types.py | 211 |
7 files changed, 17 insertions, 772 deletions
diff --git a/host/include/uhd/transport/CMakeLists.txt b/host/include/uhd/transport/CMakeLists.txt index 726306ec1..8dfd8d9f1 100644 --- a/host/include/uhd/transport/CMakeLists.txt +++ b/host/include/uhd/transport/CMakeLists.txt @@ -20,8 +20,6 @@ INSTALL(FILES bounded_buffer.hpp bounded_buffer.ipp buffer_pool.hpp - convert_types.hpp - convert_types.ipp if_addrs.hpp udp_simple.hpp udp_zero_copy.hpp diff --git a/host/include/uhd/transport/convert_types.hpp b/host/include/uhd/transport/convert_types.hpp deleted file mode 100644 index dc7fa6c1a..000000000 --- a/host/include/uhd/transport/convert_types.hpp +++ /dev/null @@ -1,96 +0,0 @@ -// -// Copyright 2010 Ettus Research LLC -// -// This program is free software: you can redistribute it and/or modify -// it under the terms of the GNU General Public License as published by -// the Free Software Foundation, either version 3 of the License, or -// (at your option) any later version. -// -// This program is distributed in the hope that it will be useful, -// but WITHOUT ANY WARRANTY; without even the implied warranty of -// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -// GNU General Public License for more details. -// -// You should have received a copy of the GNU General Public License -// along with this program. If not, see <http://www.gnu.org/licenses/>. -// - -#ifndef INCLUDED_UHD_TRANSPORT_CONVERT_TYPES_HPP -#define INCLUDED_UHD_TRANSPORT_CONVERT_TYPES_HPP - -#include <uhd/config.hpp> -#include <uhd/types/io_type.hpp> -#include <uhd/types/otw_type.hpp> -#include <vector> - -namespace uhd{ namespace transport{ - -/*! - * Convert IO samples to OWT samples. - * - * \param io_buff memory containing samples - * \param io_type the type of these samples - * \param otw_buff memory to write converted samples - * \param otw_type the type of these samples - * \param num_samps the number of samples in io_buff - */ -UHD_API void convert_io_type_to_otw_type( - const void *io_buff, const io_type_t &io_type, - void *otw_buff, const otw_type_t &otw_type, - size_t num_samps -); - -/*! - * Convert IO samples to OWT samples + interleave. - * - * \param io_buffs buffers containing samples - * \param io_type the type of these samples - * \param otw_buff memory to write converted samples - * \param otw_type the type of these samples - * \param nsamps_per_io_buff samples per io_buff - */ -UHD_API void convert_io_type_to_otw_type( - const std::vector<const void *> &io_buffs, - const io_type_t &io_type, - void *otw_buff, - const otw_type_t &otw_type, - size_t nsamps_per_io_buff -); - -/*! - * Convert OTW samples to IO samples. - * - * \param otw_buff memory containing samples - * \param otw_type the type of these samples - * \param io_buff memory to write converted samples - * \param io_type the type of these samples - * \param num_samps the number of samples in io_buff - */ -UHD_API void convert_otw_type_to_io_type( - const void *otw_buff, const otw_type_t &otw_type, - void *io_buff, const io_type_t &io_type, - size_t num_samps -); - -/*! - * Convert OTW samples to IO samples + de-interleave. - * - * \param otw_buff memory containing samples - * \param otw_type the type of these samples - * \param io_buffs buffers to write converted samples - * \param io_type the type of these samples - * \param nsamps_per_io_buff samples per io_buff - */ -UHD_API void convert_otw_type_to_io_type( - const void *otw_buff, - const otw_type_t &otw_type, - std::vector<void *> &io_buffs, - const io_type_t &io_type, - size_t nsamps_per_io_buff -); - -}} //namespace - -#include <uhd/transport/convert_types.ipp> - -#endif /* INCLUDED_UHD_TRANSPORT_CONVERT_TYPES_HPP */ diff --git a/host/include/uhd/transport/convert_types.ipp b/host/include/uhd/transport/convert_types.ipp deleted file mode 100644 index 914ca6f17..000000000 --- a/host/include/uhd/transport/convert_types.ipp +++ /dev/null @@ -1,43 +0,0 @@ -// -// Copyright 2010 Ettus Research LLC -// -// This program is free software: you can redistribute it and/or modify -// it under the terms of the GNU General Public License as published by -// the Free Software Foundation, either version 3 of the License, or -// (at your option) any later version. -// -// This program is distributed in the hope that it will be useful, -// but WITHOUT ANY WARRANTY; without even the implied warranty of -// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -// GNU General Public License for more details. -// -// You should have received a copy of the GNU General Public License -// along with this program. If not, see <http://www.gnu.org/licenses/>. -// - -#ifndef INCLUDED_UHD_TRANSPORT_CONVERT_TYPES_IPP -#define INCLUDED_UHD_TRANSPORT_CONVERT_TYPES_IPP - -UHD_INLINE void uhd::transport::convert_io_type_to_otw_type( - const void *io_buff, const io_type_t &io_type, - void *otw_buff, const otw_type_t &otw_type, - size_t num_samps -){ - std::vector<const void *> buffs(1, io_buff); - return uhd::transport::convert_io_type_to_otw_type( - buffs, io_type, otw_buff, otw_type, num_samps - ); -} - -UHD_INLINE void uhd::transport::convert_otw_type_to_io_type( - const void *otw_buff, const otw_type_t &otw_type, - void *io_buff, const io_type_t &io_type, - size_t num_samps -){ - std::vector<void *> buffs(1, io_buff); - return uhd::transport::convert_otw_type_to_io_type( - otw_buff, otw_type, buffs, io_type, num_samps - ); -} - -#endif /* INCLUDED_UHD_TRANSPORT_CONVERT_TYPES_IPP */ diff --git a/host/lib/convert/gen_convert_impl.py b/host/lib/convert/gen_convert_impl.py index 71095ab97..70d437ba2 100644 --- a/host/lib/convert/gen_convert_impl.py +++ b/host/lib/convert/gen_convert_impl.py @@ -26,6 +26,7 @@ typedef size_t pred_type; \#include <uhd/utils/algorithm.hpp> \#include <boost/lexical_cast.hpp> \#include <boost/detail/endian.hpp> +\#include <boost/cstdint.hpp> \#include <stdexcept> enum dir_type{ @@ -33,6 +34,13 @@ enum dir_type{ DIR_CPU_TO_OTW = 1 }; +struct pred_error : std::runtime_error{ + pred_error(const std::string &what) + :std::runtime_error("convert::make_pred: " + what){ + /* NOP */ + } +}; + pred_type make_pred(const std::string &markup, dir_type &dir){ pred_type pred = 0; @@ -60,10 +68,10 @@ pred_type make_pred(const std::string &markup, dir_type &dir){ if (cpu_type == "fc32") pred |= $ph.fc32_p; else if (cpu_type == "sc16") pred |= $ph.sc16_p; - else throw std::runtime_error("unhandled io type " + cpu_type); + else throw pred_error("unhandled io type " + cpu_type); if (otw_type == "item32") pred |= $ph.item32_p; - else throw std::runtime_error("unhandled otw type " + otw_type); + 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); @@ -73,16 +81,16 @@ pred_type make_pred(const std::string &markup, dir_type &dir){ case 2: pred |= $ph.chan2_p; break; case 3: pred |= $ph.chan3_p; break; case 4: pred |= $ph.chan4_p; break; - default: throw std::runtime_error("unhandled number of channels"); + 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 std::runtime_error("unhandled swap type"); + else throw pred_error("unhandled swap type"); } catch(...){ - throw std::runtime_error("convert::make_pred: could not parse markup: " + markup); + throw pred_error("could not parse markup: " + markup); } return pred; @@ -105,18 +113,18 @@ UHD_INLINE pred_type make_pred( case otw_type_t::BO_LITTLE_ENDIAN: pred |= $ph.nswap_p; break; \#endif case otw_type_t::BO_NATIVE: pred |= $ph.nswap_p; break; - default: throw std::runtime_error("unhandled otw byteorder type"); + default: throw pred_error("unhandled otw byteorder type"); } switch(otw_type.get_sample_size()){ case sizeof(boost::uint32_t): pred |= $ph.item32_p; break; - default: throw std::runtime_error("unhandled otw sample size"); + default: throw pred_error("unhandled otw sample size"); } switch(io_type.tid){ case io_type_t::COMPLEX_FLOAT32: pred |= $ph.fc32_p; break; case io_type_t::COMPLEX_INT16: pred |= $ph.sc16_p; break; - default: throw std::runtime_error("unhandled io type id"); + default: throw pred_error("unhandled io type id"); } switch(num_inputs*num_outputs){ //FIXME treated as one value @@ -124,7 +132,7 @@ UHD_INLINE pred_type make_pred( case 2: pred |= $ph.chan2_p; break; case 3: pred |= $ph.chan3_p; break; case 4: pred |= $ph.chan4_p; break; - default: throw std::runtime_error("unhandled number of channels"); + default: throw pred_error("unhandled number of channels"); } return pred; @@ -148,38 +156,6 @@ class ph: nbits = 4 #see above - @staticmethod - def has(pred, mask, flag): return (pred & mask) == flag - - @staticmethod - def get_swap_type(pred): - mask = 0b1 - if ph.has(pred, mask, ph.bswap_p): return 'bswap' - if ph.has(pred, mask, ph.nswap_p): return 'nswap' - raise NotImplementedError - - @staticmethod - def get_dev_type(pred): - mask = 0b0 - if ph.has(pred, mask, ph.item32_p): return 'item32' - raise NotImplementedError - - @staticmethod - def get_host_type(pred): - mask = 0b10 - if ph.has(pred, mask, ph.sc16_p): return 'sc16' - if ph.has(pred, mask, ph.fc32_p): return 'fc32' - raise NotImplementedError - - @staticmethod - def get_num_chans(pred): - mask = 0b1100 - if ph.has(pred, mask, ph.chan1_p): return 1 - if ph.has(pred, mask, ph.chan2_p): return 2 - if ph.has(pred, mask, ph.chan3_p): return 3 - if ph.has(pred, mask, ph.chan4_p): return 4 - raise NotImplementedError - if __name__ == '__main__': import sys, os file = os.path.basename(__file__) diff --git a/host/lib/transport/CMakeLists.txt b/host/lib/transport/CMakeLists.txt index ac92ffe6d..914ba3ba9 100644 --- a/host/lib/transport/CMakeLists.txt +++ b/host/lib/transport/CMakeLists.txt @@ -47,25 +47,6 @@ ELSE(ENABLE_USB) ENDIF(ENABLE_USB) ######################################################################## -# Check for SIMD headers -######################################################################## -MESSAGE(STATUS "") - -INCLUDE(CheckIncludeFileCXX) -CHECK_INCLUDE_FILE_CXX(emmintrin.h HAVE_EMMINTRIN_H) - -IF(HAVE_EMMINTRIN_H) - ADD_DEFINITIONS(-DHAVE_EMMINTRIN_H) -ENDIF(HAVE_EMMINTRIN_H) - -INCLUDE(CheckIncludeFileCXX) -CHECK_INCLUDE_FILE_CXX(arm_neon.h HAVE_ARM_NEON_H) - -IF(HAVE_ARM_NEON_H) - ADD_DEFINITIONS(-DHAVE_ARM_NEON_H) -ENDIF(HAVE_ARM_NEON_H) - -######################################################################## # Setup defines for interface address discovery ######################################################################## MESSAGE(STATUS "") @@ -93,21 +74,6 @@ LIBUHD_PYTHON_GEN_SOURCE( ${CMAKE_CURRENT_BINARY_DIR}/vrt_if_packet.cpp ) -LIBUHD_PYTHON_GEN_SOURCE( - ${CMAKE_CURRENT_SOURCE_DIR}/gen_convert_types.py - ${CMAKE_CURRENT_BINARY_DIR}/convert_types.cpp -) - -# append this directory to the include path so the generated convert types -# can include the implementation convert types file in the source directory -INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR}) - -# make the generated convert types depend on the implementation header -SET_SOURCE_FILES_PROPERTIES( - ${CMAKE_CURRENT_BINARY_DIR}/convert_types.cpp PROPERTIES - OBJECT_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/convert_types_impl.hpp -) - LIBUHD_APPEND_SOURCES( ${CMAKE_CURRENT_SOURCE_DIR}/buffer_pool.cpp ${CMAKE_CURRENT_SOURCE_DIR}/if_addrs.cpp diff --git a/host/lib/transport/convert_types_impl.hpp b/host/lib/transport/convert_types_impl.hpp deleted file mode 100644 index 48ff99725..000000000 --- a/host/lib/transport/convert_types_impl.hpp +++ /dev/null @@ -1,345 +0,0 @@ -// -// Copyright 2010 Ettus Research LLC -// -// This program is free software: you can redistribute it and/or modify -// it under the terms of the GNU General Public License as published by -// the Free Software Foundation, either version 3 of the License, or -// (at your option) any later version. -// -// This program is distributed in the hope that it will be useful, -// but WITHOUT ANY WARRANTY; without even the implied warranty of -// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -// GNU General Public License for more details. -// -// You should have received a copy of the GNU General Public License -// along with this program. If not, see <http://www.gnu.org/licenses/>. -// - -#ifndef INCLUDED_LIBUHD_TRANSPORT_CONVERT_TYPES_IMPL_HPP -#define INCLUDED_LIBUHD_TRANSPORT_CONVERT_TYPES_IMPL_HPP - -#include <uhd/config.hpp> -#include <uhd/utils/byteswap.hpp> -#include <boost/cstdint.hpp> -#include <cstring> -#include <complex> - -#ifdef HAVE_EMMINTRIN_H - #define USE_EMMINTRIN_H //use sse2 intrinsics -#endif - -#if defined(USE_EMMINTRIN_H) - #include <emmintrin.h> -#endif - -#ifdef HAVE_ARM_NEON_H - #define USE_ARM_NEON_H -#endif - -#if defined(USE_ARM_NEON_H) - #include <arm_neon.h> -#endif - -/*********************************************************************** - * Typedefs - **********************************************************************/ -typedef std::complex<float> fc32_t; -typedef std::complex<boost::int16_t> sc16_t; -typedef boost::uint32_t item32_t; - -/*********************************************************************** - * Convert complex short buffer to items32 - **********************************************************************/ -static UHD_INLINE item32_t sc16_to_item32(sc16_t num){ - boost::uint16_t real = num.real(); - boost::uint16_t imag = num.imag(); - return (item32_t(real) << 16) | (item32_t(imag) << 0); -} - -static UHD_INLINE void sc16_to_item32_nswap( - const sc16_t *input, item32_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = sc16_to_item32(input[i]); - } -} - -static UHD_INLINE void sc16_to_item32_bswap( - const sc16_t *input, item32_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = uhd::byteswap(sc16_to_item32(input[i])); - } -} - -/*********************************************************************** - * Convert items32 buffer to complex short - **********************************************************************/ -static UHD_INLINE sc16_t item32_to_sc16(item32_t item){ - return sc16_t( - boost::int16_t(item >> 16), - boost::int16_t(item >> 0) - ); -} - -static UHD_INLINE void item32_to_sc16_nswap( - const item32_t *input, sc16_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = item32_to_sc16(input[i]); - } -} - -static UHD_INLINE void item32_to_sc16_bswap( - const item32_t *input, sc16_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = item32_to_sc16(uhd::byteswap(input[i])); - } -} - -/*********************************************************************** - * 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); - return (item32_t(real) << 16) | (item32_t(imag) << 0); -} - -//////////////////////////////////// -// none-swap -//////////////////////////////////// -#if defined(USE_EMMINTRIN_H) -static UHD_INLINE void fc32_to_item32_nswap( - const fc32_t *input, item32_t *output, size_t nsamps -){ - __m128 scalar = _mm_set_ps1(shorts_per_float); - - //convert blocks of samples with intrinsics - size_t i = 0; for (; i < (nsamps & ~0x3); i+=4){ - //load from input - __m128 tmplo = _mm_loadu_ps(reinterpret_cast<const float *>(input+i+0)); - __m128 tmphi = _mm_loadu_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); - } - - //convert remainder - for (; i < nsamps; i++){ - output[i] = fc32_to_item32(input[i]); - } -} - -#elif defined(USE_ARM_NEON_H) -static UHD_INLINE void fc32_to_item32_nswap( - const fc32_t *input, item32_t *output, size_t nsamps) -{ - size_t i; - - float32x4_t Q0 = vdupq_n_f32(shorts_per_float); - 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]); -} - -#else -static UHD_INLINE void fc32_to_item32_nswap( - const fc32_t *input, item32_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = fc32_to_item32(input[i]); - } -} - -#endif - -//////////////////////////////////// -// byte-swap -//////////////////////////////////// -#if defined(USE_EMMINTRIN_H) -static UHD_INLINE void fc32_to_item32_bswap( - const fc32_t *input, item32_t *output, size_t nsamps -){ - __m128 scalar = _mm_set_ps1(shorts_per_float); - - //convert blocks of samples with intrinsics - size_t i = 0; for (; i < (nsamps & ~0x3); i+=4){ - //load from input - __m128 tmplo = _mm_loadu_ps(reinterpret_cast<const float *>(input+i+0)); - __m128 tmphi = _mm_loadu_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); - } - - //convert remainder - for (; i < nsamps; i++){ - output[i] = uhd::byteswap(fc32_to_item32(input[i])); - } -} - -#else -static UHD_INLINE void fc32_to_item32_bswap( - const fc32_t *input, item32_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = uhd::byteswap(fc32_to_item32(input[i])); - } -} - -#endif - -/*********************************************************************** - * 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){ - return fc32_t( - float(boost::int16_t(item >> 16)*floats_per_short), - float(boost::int16_t(item >> 0)*floats_per_short) - ); -} - -//////////////////////////////////// -// none-swap -//////////////////////////////////// -#if defined(USE_EMMINTRIN_H) -static UHD_INLINE void item32_to_fc32_nswap( - const item32_t *input, fc32_t *output, size_t nsamps -){ - __m128 scalar = _mm_set_ps1(floats_per_short/(1 << 16)); - __m128i zeroi = _mm_setzero_si128(); - - //convert blocks of samples with intrinsics - size_t i = 0; for (; i < (nsamps & ~0x3); 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_storeu_ps(reinterpret_cast<float *>(output+i+0), tmplo); - _mm_storeu_ps(reinterpret_cast<float *>(output+i+2), tmphi); - } - - //convert remainder - for (; i < nsamps; i++){ - output[i] = item32_to_fc32(input[i]); - } -} - -#elif defined(USE_ARM_NEON_H) -static UHD_INLINE void item32_to_fc32_nswap( - const item32_t *input, fc32_t *output, size_t nsamps) -{ - size_t i; - - float32x4_t Q1 = vdupq_n_f32(floats_per_short); - 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]); -} - -#else -static UHD_INLINE void item32_to_fc32_nswap( - const item32_t *input, fc32_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = item32_to_fc32(input[i]); - } -} -#endif - -//////////////////////////////////// -// byte-swap -//////////////////////////////////// -#if defined(USE_EMMINTRIN_H) -static UHD_INLINE void item32_to_fc32_bswap( - const item32_t *input, fc32_t *output, size_t nsamps -){ - __m128 scalar = _mm_set_ps1(floats_per_short/(1 << 16)); - __m128i zeroi = _mm_setzero_si128(); - - //convert blocks of samples with intrinsics - size_t i = 0; for (; i < (nsamps & ~0x3); 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_storeu_ps(reinterpret_cast<float *>(output+i+0), tmplo); - _mm_storeu_ps(reinterpret_cast<float *>(output+i+2), tmphi); - } - - //convert remainder - for (; i < nsamps; i++){ - output[i] = item32_to_fc32(uhd::byteswap(input[i])); - } -} - -#else -static UHD_INLINE void item32_to_fc32_bswap( - const item32_t *input, fc32_t *output, size_t nsamps -){ - for (size_t i = 0; i < nsamps; i++){ - output[i] = item32_to_fc32(uhd::byteswap(input[i])); - } -} - -#endif - -#endif /* INCLUDED_LIBUHD_TRANSPORT_CONVERT_TYPES_IMPL_HPP */ diff --git a/host/lib/transport/gen_convert_types.py b/host/lib/transport/gen_convert_types.py deleted file mode 100755 index f9509c81d..000000000 --- a/host/lib/transport/gen_convert_types.py +++ /dev/null @@ -1,211 +0,0 @@ -#!/usr/bin/env python -# -# Copyright 2010 Ettus Research LLC -# -# This program is free software: you can redistribute it and/or modify -# it under the terms of the GNU General Public License as published by -# the Free Software Foundation, either version 3 of the License, or -# (at your option) any later version. -# -# This program is distributed in the hope that it will be useful, -# but WITHOUT ANY WARRANTY; without even the implied warranty of -# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the -# GNU General Public License for more details. -# -# You should have received a copy of the GNU General Public License -# along with this program. If not, see <http://www.gnu.org/licenses/>. -# - -TMPL_TEXT = """ -#import time -/*********************************************************************** - * This file was generated by $file on $time.strftime("%c") - **********************************************************************/ - -\#include <uhd/config.hpp> -\#include <uhd/transport/convert_types.hpp> -\#include <boost/cstdint.hpp> -\#include <boost/detail/endian.hpp> -\#include <stdexcept> -\#include "convert_types_impl.hpp" - -using namespace uhd; - -/*********************************************************************** - * Generate predicate for jump table - **********************************************************************/ -UHD_INLINE boost::uint8_t get_pred( - const io_type_t &io_type, - const otw_type_t &otw_type, - size_t num_chans -){ - boost::uint8_t pred = 0; - - switch(otw_type.byteorder){ - \#ifdef BOOST_BIG_ENDIAN - case otw_type_t::BO_BIG_ENDIAN: pred |= $ph.nswap_p; break; - case otw_type_t::BO_LITTLE_ENDIAN: pred |= $ph.bswap_p; break; - \#else - case otw_type_t::BO_BIG_ENDIAN: pred |= $ph.bswap_p; break; - case otw_type_t::BO_LITTLE_ENDIAN: pred |= $ph.nswap_p; break; - \#endif - case otw_type_t::BO_NATIVE: pred |= $ph.nswap_p; break; - default: throw std::runtime_error("unhandled otw byteorder type"); - } - - switch(otw_type.get_sample_size()){ - case sizeof(boost::uint32_t): pred |= $ph.item32_p; break; - default: throw std::runtime_error("unhandled otw sample size"); - } - - switch(io_type.tid){ - case io_type_t::COMPLEX_FLOAT32: pred |= $ph.fc32_p; break; - case io_type_t::COMPLEX_INT16: pred |= $ph.sc16_p; break; - default: throw std::runtime_error("unhandled io type id"); - } - - switch(num_chans){ - 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 std::runtime_error("unhandled number of channels"); - } - - return pred; -} - -/*********************************************************************** - * Convert host type to device type - **********************************************************************/ -void transport::convert_io_type_to_otw_type( - const std::vector<const void *> &io_buffs, - const io_type_t &io_type, - void *otw_buff, - const otw_type_t &otw_type, - size_t nsamps_per_io_buff -){ - switch(get_pred(io_type, otw_type, io_buffs.size())){ - #for $pred in range(2**$ph.nbits) - case $pred: - #set $out_type = $ph.get_dev_type($pred) - #set $in_type = $ph.get_host_type($pred) - #set $num_chans = $ph.get_num_chans($pred) - #set $converter = '_'.join([$in_type, 'to', $out_type]) - #if $num_chans == 1 - $(converter)_$ph.get_swap_type($pred)( - reinterpret_cast<const $(in_type)_t *>(io_buffs.front()), - reinterpret_cast<$(out_type)_t *>(otw_buff), - nsamps_per_io_buff - ); - #else - for (size_t i = 0, j = 0; i < nsamps_per_io_buff; i++){ - #for $j in range($num_chans) - reinterpret_cast<$(out_type)_t *>(otw_buff)[j++] = - #if $ph.get_swap_type($pred) == 'bswap' - uhd::byteswap($(converter)(reinterpret_cast<const $(in_type)_t *>(io_buffs[$j])[i])); - #else - $(converter)(reinterpret_cast<const $(in_type)_t *>(io_buffs[$j])[i]); - #end if - #end for - } - #end if - break; - #end for - } -} - -/*********************************************************************** - * Convert device type to host type - **********************************************************************/ -void transport::convert_otw_type_to_io_type( - const void *otw_buff, - const otw_type_t &otw_type, - std::vector<void *> &io_buffs, - const io_type_t &io_type, - size_t nsamps_per_io_buff -){ - switch(get_pred(io_type, otw_type, io_buffs.size())){ - #for $pred in range(2**$ph.nbits) - case $pred: - #set $out_type = $ph.get_host_type($pred) - #set $in_type = $ph.get_dev_type($pred) - #set $num_chans = $ph.get_num_chans($pred) - #set $converter = '_'.join([$in_type, 'to', $out_type]) - #if $num_chans == 1 - $(converter)_$ph.get_swap_type($pred)( - reinterpret_cast<const $(in_type)_t *>(otw_buff), - reinterpret_cast<$(out_type)_t *>(io_buffs.front()), - nsamps_per_io_buff - ); - #else - for (size_t i = 0, j = 0; i < nsamps_per_io_buff; i++){ - #for $j in range($num_chans) - reinterpret_cast<$(out_type)_t *>(io_buffs[$j])[i] = - #if $ph.get_swap_type($pred) == 'bswap' - $(converter)(uhd::byteswap(reinterpret_cast<const $(in_type)_t *>(otw_buff)[j++])); - #else - $(converter)(reinterpret_cast<const $(in_type)_t *>(otw_buff)[j++]); - #end if - #end for - } - #end if - break; - #end for - } -} - -""" - -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 - sc16_p = 0b00010 - fc32_p = 0b00000 - chan1_p = 0b00000 - chan2_p = 0b00100 - chan3_p = 0b01000 - chan4_p = 0b01100 - - nbits = 4 #see above - - @staticmethod - def has(pred, mask, flag): return (pred & mask) == flag - - @staticmethod - def get_swap_type(pred): - mask = 0b1 - if ph.has(pred, mask, ph.bswap_p): return 'bswap' - if ph.has(pred, mask, ph.nswap_p): return 'nswap' - raise NotImplementedError - - @staticmethod - def get_dev_type(pred): - mask = 0b0 - if ph.has(pred, mask, ph.item32_p): return 'item32' - raise NotImplementedError - - @staticmethod - def get_host_type(pred): - mask = 0b10 - if ph.has(pred, mask, ph.sc16_p): return 'sc16' - if ph.has(pred, mask, ph.fc32_p): return 'fc32' - raise NotImplementedError - - @staticmethod - def get_num_chans(pred): - mask = 0b1100 - if ph.has(pred, mask, ph.chan1_p): return 1 - if ph.has(pred, mask, ph.chan2_p): return 2 - if ph.has(pred, mask, ph.chan3_p): return 3 - if ph.has(pred, mask, ph.chan4_p): return 4 - raise NotImplementedError - -if __name__ == '__main__': - import sys - open(sys.argv[1], 'w').write(parse_tmpl(TMPL_TEXT, file=__file__, ph=ph)) |