volk_64f_x2_add_64f.h

Go to the documentation of this file.

/* -*- c++ -*- */
/*
 * Copyright 2018 Free Software Foundation, Inc.
 *
 * This file is part of VOLK
 *
 * SPDX-License-Identifier: LGPL-3.0-or-later
 */
 
#ifndef INCLUDED_volk_64f_x2_add_64f_H
#define INCLUDED_volk_64f_x2_add_64f_H
 
#include <inttypes.h>
 
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_64f_x2_add_64f_generic(double* cVector,
                                               const double* aVector,
                                               const double* bVector,
                                               unsigned int num_points)
{
    double* cPtr = cVector;
    const double* aPtr = aVector;
    const double* bPtr = bVector;
    unsigned int number = 0;
 
    for (number = 0; number < num_points; number++) {
        *cPtr++ = (*aPtr++) + (*bPtr++);
    }
}
 
#endif /* LV_HAVE_GENERIC */
 
/*
 * Unaligned versions
 */
 
#ifdef LV_HAVE_SSE2
 
#include <emmintrin.h>
 
static inline void volk_64f_x2_add_64f_u_sse2(double* cVector,
                                              const double* aVector,
                                              const double* bVector,
                                              unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int half_points = num_points / 2;
 
    double* cPtr = cVector;
    const double* aPtr = aVector;
    const double* bPtr = bVector;
 
    __m128d aVal, bVal, cVal;
    for (; number < half_points; number++) {
        aVal = _mm_loadu_pd(aPtr);
        bVal = _mm_loadu_pd(bPtr);
 
        cVal = _mm_add_pd(aVal, bVal);
 
        _mm_storeu_pd(cPtr, cVal); // Store the results back into the C container
 
        aPtr += 2;
        bPtr += 2;
        cPtr += 2;
    }
 
    number = half_points * 2;
    for (; number < num_points; number++) {
        *cPtr++ = (*aPtr++) + (*bPtr++);
    }
}
 
#endif /* LV_HAVE_SSE2 */
 
 
#ifdef LV_HAVE_AVX
 
#include <immintrin.h>
 
static inline void volk_64f_x2_add_64f_u_avx(double* cVector,
                                             const double* aVector,
                                             const double* bVector,
                                             unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int quarter_points = num_points / 4;
 
    double* cPtr = cVector;
    const double* aPtr = aVector;
    const double* bPtr = bVector;
 
    __m256d aVal, bVal, cVal;
    for (; number < quarter_points; number++) {
 
        aVal = _mm256_loadu_pd(aPtr);
        bVal = _mm256_loadu_pd(bPtr);
 
        cVal = _mm256_add_pd(aVal, bVal);
 
        _mm256_storeu_pd(cPtr, cVal); // Store the results back into the C container
 
        aPtr += 4;
        bPtr += 4;
        cPtr += 4;
    }
 
    number = quarter_points * 4;
    for (; number < num_points; number++) {
        *cPtr++ = (*aPtr++) + (*bPtr++);
    }
}
 
#endif /* LV_HAVE_AVX */
 
/*
 * Aligned versions
 */
 
#ifdef LV_HAVE_SSE2
 
#include <emmintrin.h>
 
static inline void volk_64f_x2_add_64f_a_sse2(double* cVector,
                                              const double* aVector,
                                              const double* bVector,
                                              unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int half_points = num_points / 2;
 
    double* cPtr = cVector;
    const double* aPtr = aVector;
    const double* bPtr = bVector;
 
    __m128d aVal, bVal, cVal;
    for (; number < half_points; number++) {
        aVal = _mm_load_pd(aPtr);
        bVal = _mm_load_pd(bPtr);
 
        cVal = _mm_add_pd(aVal, bVal);
 
        _mm_store_pd(cPtr, cVal); // Store the results back into the C container
 
        aPtr += 2;
        bPtr += 2;
        cPtr += 2;
    }
 
    number = half_points * 2;
    for (; number < num_points; number++) {
        *cPtr++ = (*aPtr++) + (*bPtr++);
    }
}
 
#endif /* LV_HAVE_SSE2 */
 
 
#ifdef LV_HAVE_AVX
 
#include <immintrin.h>
 
static inline void volk_64f_x2_add_64f_a_avx(double* cVector,
                                             const double* aVector,
                                             const double* bVector,
                                             unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int quarter_points = num_points / 4;
 
    double* cPtr = cVector;
    const double* aPtr = aVector;
    const double* bPtr = bVector;
 
    __m256d aVal, bVal, cVal;
    for (; number < quarter_points; number++) {
 
        aVal = _mm256_load_pd(aPtr);
        bVal = _mm256_load_pd(bPtr);
 
        cVal = _mm256_add_pd(aVal, bVal);
 
        _mm256_store_pd(cPtr, cVal); // Store the results back into the C container
 
        aPtr += 4;
        bPtr += 4;
        cPtr += 4;
    }
 
    number = quarter_points * 4;
    for (; number < num_points; number++) {
        *cPtr++ = (*aPtr++) + (*bPtr++);
    }
}
 
#endif /* LV_HAVE_AVX */
 
#endif /* INCLUDED_volk_64f_x2_add_64f_u_H */