volk_32f_x2_s32f_interleave_16ic.h

Go to the documentation of this file.

/* -*- c++ -*- */
/*
 * Copyright 2012, 2014 Free Software Foundation, Inc.
 *
 * This file is part of VOLK
 *
 * SPDX-License-Identifier: LGPL-3.0-or-later
 */
 
#ifndef INCLUDED_volk_32f_x2_s32f_interleave_16ic_a_H
#define INCLUDED_volk_32f_x2_s32f_interleave_16ic_a_H
 
#include <inttypes.h>
#include <stdio.h>
#include <volk/volk_common.h>
 
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
 
static inline void volk_32f_x2_s32f_interleave_16ic_a_avx2(lv_16sc_t* complexVector,
                                                           const float* iBuffer,
                                                           const float* qBuffer,
                                                           const float scalar,
                                                           unsigned int num_points)
{
    unsigned int number = 0;
    const float* iBufferPtr = iBuffer;
    const float* qBufferPtr = qBuffer;
 
    __m256 vScalar = _mm256_set1_ps(scalar);
 
    const unsigned int eighthPoints = num_points / 8;
 
    __m256 iValue, qValue, cplxValue1, cplxValue2;
    __m256i intValue1, intValue2;
 
    int16_t* complexVectorPtr = (int16_t*)complexVector;
 
    for (; number < eighthPoints; number++) {
        iValue = _mm256_load_ps(iBufferPtr);
        qValue = _mm256_load_ps(qBufferPtr);
 
        // Interleaves the lower two values in the i and q variables into one buffer
        cplxValue1 = _mm256_unpacklo_ps(iValue, qValue);
        cplxValue1 = _mm256_mul_ps(cplxValue1, vScalar);
 
        // Interleaves the upper two values in the i and q variables into one buffer
        cplxValue2 = _mm256_unpackhi_ps(iValue, qValue);
        cplxValue2 = _mm256_mul_ps(cplxValue2, vScalar);
 
        intValue1 = _mm256_cvtps_epi32(cplxValue1);
        intValue2 = _mm256_cvtps_epi32(cplxValue2);
 
        intValue1 = _mm256_packs_epi32(intValue1, intValue2);
 
        _mm256_store_si256((__m256i*)complexVectorPtr, intValue1);
        complexVectorPtr += 16;
 
        iBufferPtr += 8;
        qBufferPtr += 8;
    }
 
    number = eighthPoints * 8;
    complexVectorPtr = (int16_t*)(&complexVector[number]);
    for (; number < num_points; number++) {
        *complexVectorPtr++ = (int16_t)rintf(*iBufferPtr++ * scalar);
        *complexVectorPtr++ = (int16_t)rintf(*qBufferPtr++ * scalar);
    }
}
#endif /* LV_HAVE_AVX2 */
 
 
#ifdef LV_HAVE_SSE2
#include <emmintrin.h>
 
static inline void volk_32f_x2_s32f_interleave_16ic_a_sse2(lv_16sc_t* complexVector,
                                                           const float* iBuffer,
                                                           const float* qBuffer,
                                                           const float scalar,
                                                           unsigned int num_points)
{
    unsigned int number = 0;
    const float* iBufferPtr = iBuffer;
    const float* qBufferPtr = qBuffer;
 
    __m128 vScalar = _mm_set_ps1(scalar);
 
    const unsigned int quarterPoints = num_points / 4;
 
    __m128 iValue, qValue, cplxValue1, cplxValue2;
    __m128i intValue1, intValue2;
 
    int16_t* complexVectorPtr = (int16_t*)complexVector;
 
    for (; number < quarterPoints; number++) {
        iValue = _mm_load_ps(iBufferPtr);
        qValue = _mm_load_ps(qBufferPtr);
 
        // Interleaves the lower two values in the i and q variables into one buffer
        cplxValue1 = _mm_unpacklo_ps(iValue, qValue);
        cplxValue1 = _mm_mul_ps(cplxValue1, vScalar);
 
        // Interleaves the upper two values in the i and q variables into one buffer
        cplxValue2 = _mm_unpackhi_ps(iValue, qValue);
        cplxValue2 = _mm_mul_ps(cplxValue2, vScalar);
 
        intValue1 = _mm_cvtps_epi32(cplxValue1);
        intValue2 = _mm_cvtps_epi32(cplxValue2);
 
        intValue1 = _mm_packs_epi32(intValue1, intValue2);
 
        _mm_store_si128((__m128i*)complexVectorPtr, intValue1);
        complexVectorPtr += 8;
 
        iBufferPtr += 4;
        qBufferPtr += 4;
    }
 
    number = quarterPoints * 4;
    complexVectorPtr = (int16_t*)(&complexVector[number]);
    for (; number < num_points; number++) {
        *complexVectorPtr++ = (int16_t)rintf(*iBufferPtr++ * scalar);
        *complexVectorPtr++ = (int16_t)rintf(*qBufferPtr++ * scalar);
    }
}
#endif /* LV_HAVE_SSE2 */
 
 
#ifdef LV_HAVE_SSE
#include <xmmintrin.h>
 
static inline void volk_32f_x2_s32f_interleave_16ic_a_sse(lv_16sc_t* complexVector,
                                                          const float* iBuffer,
                                                          const float* qBuffer,
                                                          const float scalar,
                                                          unsigned int num_points)
{
    unsigned int number = 0;
    const float* iBufferPtr = iBuffer;
    const float* qBufferPtr = qBuffer;
 
    __m128 vScalar = _mm_set_ps1(scalar);
 
    const unsigned int quarterPoints = num_points / 4;
 
    __m128 iValue, qValue, cplxValue;
 
    int16_t* complexVectorPtr = (int16_t*)complexVector;
 
    __VOLK_ATTR_ALIGNED(16) float floatBuffer[4];
 
    for (; number < quarterPoints; number++) {
        iValue = _mm_load_ps(iBufferPtr);
        qValue = _mm_load_ps(qBufferPtr);
 
        // Interleaves the lower two values in the i and q variables into one buffer
        cplxValue = _mm_unpacklo_ps(iValue, qValue);
        cplxValue = _mm_mul_ps(cplxValue, vScalar);
 
        _mm_store_ps(floatBuffer, cplxValue);
 
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[0]);
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[1]);
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[2]);
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[3]);
 
        // Interleaves the upper two values in the i and q variables into one buffer
        cplxValue = _mm_unpackhi_ps(iValue, qValue);
        cplxValue = _mm_mul_ps(cplxValue, vScalar);
 
        _mm_store_ps(floatBuffer, cplxValue);
 
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[0]);
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[1]);
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[2]);
        *complexVectorPtr++ = (int16_t)rintf(floatBuffer[3]);
 
        iBufferPtr += 4;
        qBufferPtr += 4;
    }
 
    number = quarterPoints * 4;
    complexVectorPtr = (int16_t*)(&complexVector[number]);
    for (; number < num_points; number++) {
        *complexVectorPtr++ = (int16_t)rintf(*iBufferPtr++ * scalar);
        *complexVectorPtr++ = (int16_t)rintf(*qBufferPtr++ * scalar);
    }
}
#endif /* LV_HAVE_SSE */
 
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_32f_x2_s32f_interleave_16ic_generic(lv_16sc_t* complexVector,
                                                            const float* iBuffer,
                                                            const float* qBuffer,
                                                            const float scalar,
                                                            unsigned int num_points)
{
    int16_t* complexVectorPtr = (int16_t*)complexVector;
    const float* iBufferPtr = iBuffer;
    const float* qBufferPtr = qBuffer;
    unsigned int number = 0;
 
    for (number = 0; number < num_points; number++) {
        *complexVectorPtr++ = (int16_t)rintf(*iBufferPtr++ * scalar);
        *complexVectorPtr++ = (int16_t)rintf(*qBufferPtr++ * scalar);
    }
}
#endif /* LV_HAVE_GENERIC */
 
 
#endif /* INCLUDED_volk_32f_x2_s32f_interleave_16ic_a_H */
 
#ifndef INCLUDED_volk_32f_x2_s32f_interleave_16ic_u_H
#define INCLUDED_volk_32f_x2_s32f_interleave_16ic_u_H
 
#include <inttypes.h>
#include <stdio.h>
#include <volk/volk_common.h>
 
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
 
static inline void volk_32f_x2_s32f_interleave_16ic_u_avx2(lv_16sc_t* complexVector,
                                                           const float* iBuffer,
                                                           const float* qBuffer,
                                                           const float scalar,
                                                           unsigned int num_points)
{
    unsigned int number = 0;
    const float* iBufferPtr = iBuffer;
    const float* qBufferPtr = qBuffer;
 
    __m256 vScalar = _mm256_set1_ps(scalar);
 
    const unsigned int eighthPoints = num_points / 8;
 
    __m256 iValue, qValue, cplxValue1, cplxValue2;
    __m256i intValue1, intValue2;
 
    int16_t* complexVectorPtr = (int16_t*)complexVector;
 
    for (; number < eighthPoints; number++) {
        iValue = _mm256_loadu_ps(iBufferPtr);
        qValue = _mm256_loadu_ps(qBufferPtr);
 
        // Interleaves the lower two values in the i and q variables into one buffer
        cplxValue1 = _mm256_unpacklo_ps(iValue, qValue);
        cplxValue1 = _mm256_mul_ps(cplxValue1, vScalar);
 
        // Interleaves the upper two values in the i and q variables into one buffer
        cplxValue2 = _mm256_unpackhi_ps(iValue, qValue);
        cplxValue2 = _mm256_mul_ps(cplxValue2, vScalar);
 
        intValue1 = _mm256_cvtps_epi32(cplxValue1);
        intValue2 = _mm256_cvtps_epi32(cplxValue2);
 
        intValue1 = _mm256_packs_epi32(intValue1, intValue2);
 
        _mm256_storeu_si256((__m256i*)complexVectorPtr, intValue1);
        complexVectorPtr += 16;
 
        iBufferPtr += 8;
        qBufferPtr += 8;
    }
 
    number = eighthPoints * 8;
    complexVectorPtr = (int16_t*)(&complexVector[number]);
    for (; number < num_points; number++) {
        *complexVectorPtr++ = (int16_t)rintf(*iBufferPtr++ * scalar);
        *complexVectorPtr++ = (int16_t)rintf(*qBufferPtr++ * scalar);
    }
}
#endif /* LV_HAVE_AVX2 */
 
 
#endif /* INCLUDED_volk_32f_x2_s32f_interleave_16ic_u_H */