/*! * \file volk_gnsssdr_8ic_s8ic_multiply_8ic.h * \brief VOLK_GNSSSDR kernel: multiplies a group of 16 bits vectors by one constant vector. * \authors * * VOLK_GNSSSDR kernel that multiplies a group of 16 bits vectors * (8 bits the real part and 8 bits the imaginary part) by one constant vector * * ----------------------------------------------------------------------------- * * GNSS-SDR is a Global Navigation Satellite System software-defined receiver. * This file is part of GNSS-SDR. * * Copyright (C) 2010-2020 (see AUTHORS file for a list of contributors) * SPDX-License-Identifier: GPL-3.0-or-later * * ----------------------------------------------------------------------------- */ /*! * \page volk_gnsssdr_8ic_s8ic_multiply_8ic * * \b Overview * * Multiplies the input vector by a scalar and stores the results in the third vector * * Dispatcher Prototype * \code * void volk_gnsssdr_8ic_s8ic_multiply_8ic(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points); * \endcode * * \b Inputs * \li aVector: The vector to be multiplied. * \li scalar The complex scalar to multiply \p aVector * \li num_points: The number of complex values in \p aVector to be multiplied by \p scalar and stored into \p cVector. * * \b Outputs * \li cVector: The vector where the results will be stored * */ #ifndef INCLUDED_volk_gnsssdr_8ic_s8ic_multiply_8ic_H #define INCLUDED_volk_gnsssdr_8ic_s8ic_multiply_8ic_H #include #ifdef LV_HAVE_SSE3 #include static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_u_sse3(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points) { unsigned int number = 0; const unsigned int sse_iters = num_points / 8; __m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc; lv_8sc_t* c = cVector; const lv_8sc_t* a = aVector; mult1 = _mm_set_epi8(0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF); y = _mm_set1_epi16(*(short*)&scalar); imagy = _mm_srli_si128(y, 1); imagy = _mm_and_si128(imagy, mult1); realy = _mm_and_si128(y, mult1); for (; number < sse_iters; number++) { x = _mm_lddqu_si128((__m128i*)a); imagx = _mm_srli_si128(x, 1); imagx = _mm_and_si128(imagx, mult1); realx = _mm_and_si128(x, mult1); realx_mult_realy = _mm_mullo_epi16(realx, realy); imagx_mult_imagy = _mm_mullo_epi16(imagx, imagy); realx_mult_imagy = _mm_mullo_epi16(realx, imagy); imagx_mult_realy = _mm_mullo_epi16(imagx, realy); realc = _mm_sub_epi16(realx_mult_realy, imagx_mult_imagy); realc = _mm_and_si128(realc, mult1); imagc = _mm_add_epi16(realx_mult_imagy, imagx_mult_realy); imagc = _mm_and_si128(imagc, mult1); imagc = _mm_slli_si128(imagc, 1); totalc = _mm_or_si128(realc, imagc); _mm_storeu_si128((__m128i*)c, totalc); a += 8; c += 8; } for (number = sse_iters * 8; number < num_points; ++number) { *c++ = (*a++) * scalar; } } #endif /* LV_HAVE_SSE3 */ #ifdef LV_HAVE_GENERIC static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_generic(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points) { /*lv_8sc_t* cPtr = cVector; const lv_8sc_t* aPtr = aVector; for (int i = 0; i= 8) { *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; *cPtr++ = (*aPtr++) * scalar; number -= 8; } // clean up any remaining while (number-- > 0) *cPtr++ = *aPtr++ * scalar; } #endif /* LV_HAVE_GENERIC */ #ifdef LV_HAVE_SSE3 #include static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_a_sse3(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points) { unsigned int number = 0; const unsigned int sse_iters = num_points / 8; __m128i x, y, mult1, realx, imagx, realy, imagy, realx_mult_realy, imagx_mult_imagy, realx_mult_imagy, imagx_mult_realy, realc, imagc, totalc; lv_8sc_t* c = cVector; const lv_8sc_t* a = aVector; mult1 = _mm_set_epi8(0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF); y = _mm_set1_epi16(*(short*)&scalar); imagy = _mm_srli_si128(y, 1); imagy = _mm_and_si128(imagy, mult1); realy = _mm_and_si128(y, mult1); for (; number < sse_iters; number++) { x = _mm_load_si128((__m128i*)a); imagx = _mm_srli_si128(x, 1); imagx = _mm_and_si128(imagx, mult1); realx = _mm_and_si128(x, mult1); realx_mult_realy = _mm_mullo_epi16(realx, realy); imagx_mult_imagy = _mm_mullo_epi16(imagx, imagy); realx_mult_imagy = _mm_mullo_epi16(realx, imagy); imagx_mult_realy = _mm_mullo_epi16(imagx, realy); realc = _mm_sub_epi16(realx_mult_realy, imagx_mult_imagy); realc = _mm_and_si128(realc, mult1); imagc = _mm_add_epi16(realx_mult_imagy, imagx_mult_realy); imagc = _mm_and_si128(imagc, mult1); imagc = _mm_slli_si128(imagc, 1); totalc = _mm_or_si128(realc, imagc); _mm_store_si128((__m128i*)c, totalc); a += 8; c += 8; } for (number = sse_iters * 8; number < num_points; ++number) { *c++ = (*a++) * scalar; } } #endif /* LV_HAVE_SSE3 */ #ifdef LV_HAVE_ORC extern void volk_gnsssdr_8ic_s8ic_multiply_8ic_a_orc_impl(lv_8sc_t* cVector, const lv_8sc_t* aVector, const char scalarreal, const char scalarimag, unsigned int num_points); static inline void volk_gnsssdr_8ic_s8ic_multiply_8ic_u_orc(lv_8sc_t* cVector, const lv_8sc_t* aVector, const lv_8sc_t scalar, unsigned int num_points) { volk_gnsssdr_8ic_s8ic_multiply_8ic_a_orc_impl(cVector, aVector, lv_creal(scalar), lv_cimag(scalar), num_points); } #endif /* LV_HAVE_ORC */ #endif /* INCLUDED_volk_gnsssdr_32fc_x2_multiply_32fc_H */