From cfba2388ce71121bbadbf0366dbf95f726e78748 Mon Sep 17 00:00:00 2001 From: Josh Blum Date: Wed, 8 Feb 2012 23:01:46 -0800 Subject: uhd: added sse2 conversions for fc32 to sc8 --- host/lib/convert/CMakeLists.txt | 1 + host/lib/convert/convert_fc32_to_sc8_with_sse2.cpp | 150 +++++++++++++++++++++ 2 files changed, 151 insertions(+) create mode 100644 host/lib/convert/convert_fc32_to_sc8_with_sse2.cpp (limited to 'host') diff --git a/host/lib/convert/CMakeLists.txt b/host/lib/convert/CMakeLists.txt index a2a61cb9a..c42a0a434 100644 --- a/host/lib/convert/CMakeLists.txt +++ b/host/lib/convert/CMakeLists.txt @@ -73,6 +73,7 @@ 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( 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..b8545bad0 --- /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 . +// + +#include "convert_common.hpp" +#include +#include + +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(inputs[0]); + item32_t *output = reinterpret_cast(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+=2){ \ + /* load from input */ \ + __m128 tmp0 = _mm_load ## _al_ ## ps(reinterpret_cast(input+i+0)); \ + __m128 tmp1 = _mm_load ## _al_ ## ps(reinterpret_cast(input+i+2)); \ + __m128 tmp2 = _mm_load ## _al_ ## ps(reinterpret_cast(input+i+4)); \ + __m128 tmp3 = _mm_load ## _al_ ## ps(reinterpret_cast(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(inputs[0]); + item32_t *output = reinterpret_cast(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+=2){ \ + /* load from input */ \ + __m128 tmp0 = _mm_load ## _al_ ## ps(reinterpret_cast(input+i+0)); \ + __m128 tmp1 = _mm_load ## _al_ ## ps(reinterpret_cast(input+i+2)); \ + __m128 tmp2 = _mm_load ## _al_ ## ps(reinterpret_cast(input+i+4)); \ + __m128 tmp3 = _mm_load ## _al_ ## ps(reinterpret_cast(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); + } +} -- cgit v1.2.3