volk_32fc_magnitude_32f.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_32fc_magnitude_32f_u_H
#define INCLUDED_volk_32fc_magnitude_32f_u_H
 
#include <inttypes.h>
#include <math.h>
#include <stdio.h>
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_32fc_magnitude_32f_generic(float* magnitudeVector,
                                                   const lv_32fc_t* complexVector,
                                                   unsigned int num_points)
{
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
    unsigned int number = 0;
    for (number = 0; number < num_points; number++) {
        const float real = *complexVectorPtr++;
        const float imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((real * real) + (imag * imag));
    }
}
#endif /* LV_HAVE_GENERIC */
 
#ifdef LV_HAVE_AVX512F
#include <immintrin.h>
 
static inline void volk_32fc_magnitude_32f_u_avx512f(float* magnitudeVector,
                                                     const lv_32fc_t* complexVector,
                                                     unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int eighthPoints = num_points / 8;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m512 cplxValue;
    __m512 iValues, qValues;
    __m512 result;
 
    for (; number < eighthPoints; number++) {
        // Load 8 complex numbers (16 floats): I0,Q0,I1,Q1,...,I7,Q7
        cplxValue = _mm512_loadu_ps(complexVectorPtr);
 
        // Separate I and Q values using permute
        // Extract I values (even indices: 0,2,4,6,8,10,12,14)
        iValues = _mm512_permutexvar_ps(
            _mm512_setr_epi32(0, 2, 4, 6, 8, 10, 12, 14, 0, 0, 0, 0, 0, 0, 0, 0),
            cplxValue);
        // Extract Q values (odd indices: 1,3,5,7,9,11,13,15)
        qValues = _mm512_permutexvar_ps(
            _mm512_setr_epi32(1, 3, 5, 7, 9, 11, 13, 15, 0, 0, 0, 0, 0, 0, 0, 0),
            cplxValue);
 
        // Square and add: I^2 + Q^2
        result = _mm512_fmadd_ps(iValues, iValues, _mm512_mul_ps(qValues, qValues));
 
        // Square root
        result = _mm512_sqrt_ps(result);
 
        // Store only first 8 results
        _mm256_storeu_ps(magnitudeVectorPtr, _mm512_castps512_ps256(result));
 
        complexVectorPtr += 16;
        magnitudeVectorPtr += 8;
    }
 
    number = eighthPoints * 8;
    volk_32fc_magnitude_32f_generic(
        magnitudeVectorPtr, (const lv_32fc_t*)complexVectorPtr, num_points - number);
}
#endif /* LV_HAVE_AVX512F */
 
#ifdef LV_HAVE_AVX
#include <immintrin.h>
#include <volk/volk_avx_intrinsics.h>
 
static inline void volk_32fc_magnitude_32f_u_avx(float* magnitudeVector,
                                                 const lv_32fc_t* complexVector,
                                                 unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int eighthPoints = num_points / 8;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m256 cplxValue1, cplxValue2, result;
 
    for (; number < eighthPoints; number++) {
        cplxValue1 = _mm256_loadu_ps(complexVectorPtr);
        cplxValue2 = _mm256_loadu_ps(complexVectorPtr + 8);
        result = _mm256_magnitude_ps(cplxValue1, cplxValue2);
        _mm256_storeu_ps(magnitudeVectorPtr, result);
 
        complexVectorPtr += 16;
        magnitudeVectorPtr += 8;
    }
 
    number = eighthPoints * 8;
    for (; number < num_points; number++) {
        float val1Real = *complexVectorPtr++;
        float val1Imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
    }
}
#endif /* LV_HAVE_AVX */
 
#ifdef LV_HAVE_SSE3
#include <pmmintrin.h>
#include <volk/volk_sse3_intrinsics.h>
 
static inline void volk_32fc_magnitude_32f_u_sse3(float* magnitudeVector,
                                                  const lv_32fc_t* complexVector,
                                                  unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int quarterPoints = num_points / 4;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m128 cplxValue1, cplxValue2, result;
    for (; number < quarterPoints; number++) {
        cplxValue1 = _mm_loadu_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        cplxValue2 = _mm_loadu_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        result = _mm_magnitude_ps_sse3(cplxValue1, cplxValue2);
 
        _mm_storeu_ps(magnitudeVectorPtr, result);
        magnitudeVectorPtr += 4;
    }
 
    number = quarterPoints * 4;
    for (; number < num_points; number++) {
        float val1Real = *complexVectorPtr++;
        float val1Imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
    }
}
#endif /* LV_HAVE_SSE3 */
 
 
#ifdef LV_HAVE_SSE
#include <volk/volk_sse_intrinsics.h>
#include <xmmintrin.h>
 
static inline void volk_32fc_magnitude_32f_u_sse(float* magnitudeVector,
                                                 const lv_32fc_t* complexVector,
                                                 unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int quarterPoints = num_points / 4;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m128 cplxValue1, cplxValue2, result;
 
    for (; number < quarterPoints; number++) {
        cplxValue1 = _mm_loadu_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        cplxValue2 = _mm_loadu_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        result = _mm_magnitude_ps(cplxValue1, cplxValue2);
        _mm_storeu_ps(magnitudeVectorPtr, result);
        magnitudeVectorPtr += 4;
    }
 
    number = quarterPoints * 4;
    for (; number < num_points; number++) {
        float val1Real = *complexVectorPtr++;
        float val1Imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
    }
}
#endif /* LV_HAVE_SSE */
 
#endif /* INCLUDED_volk_32fc_magnitude_32f_u_H */
#ifndef INCLUDED_volk_32fc_magnitude_32f_a_H
#define INCLUDED_volk_32fc_magnitude_32f_a_H
 
#include <inttypes.h>
#include <math.h>
#include <stdio.h>
 
#ifdef LV_HAVE_AVX512F
#include <immintrin.h>
 
static inline void volk_32fc_magnitude_32f_a_avx512f(float* magnitudeVector,
                                                     const lv_32fc_t* complexVector,
                                                     unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int eighthPoints = num_points / 8;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m512 cplxValue;
    __m512 iValues, qValues;
    __m512 result;
 
    for (; number < eighthPoints; number++) {
        // Load 8 complex numbers (16 floats): I0,Q0,I1,Q1,...,I7,Q7
        cplxValue = _mm512_load_ps(complexVectorPtr);
 
        // Separate I and Q values using permute
        iValues = _mm512_permutexvar_ps(
            _mm512_setr_epi32(0, 2, 4, 6, 8, 10, 12, 14, 0, 0, 0, 0, 0, 0, 0, 0),
            cplxValue);
        qValues = _mm512_permutexvar_ps(
            _mm512_setr_epi32(1, 3, 5, 7, 9, 11, 13, 15, 0, 0, 0, 0, 0, 0, 0, 0),
            cplxValue);
 
        // Square and add: I^2 + Q^2
        result = _mm512_fmadd_ps(iValues, iValues, _mm512_mul_ps(qValues, qValues));
 
        // Square root
        result = _mm512_sqrt_ps(result);
 
        // Store only first 8 results
        _mm256_store_ps(magnitudeVectorPtr, _mm512_castps512_ps256(result));
 
        complexVectorPtr += 16;
        magnitudeVectorPtr += 8;
    }
 
    number = eighthPoints * 8;
    volk_32fc_magnitude_32f_generic(
        magnitudeVectorPtr, (const lv_32fc_t*)complexVectorPtr, num_points - number);
}
#endif /* LV_HAVE_AVX512F */
 
#ifdef LV_HAVE_AVX
#include <immintrin.h>
#include <volk/volk_avx_intrinsics.h>
 
static inline void volk_32fc_magnitude_32f_a_avx(float* magnitudeVector,
                                                 const lv_32fc_t* complexVector,
                                                 unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int eighthPoints = num_points / 8;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m256 cplxValue1, cplxValue2, result;
    for (; number < eighthPoints; number++) {
        cplxValue1 = _mm256_load_ps(complexVectorPtr);
        complexVectorPtr += 8;
 
        cplxValue2 = _mm256_load_ps(complexVectorPtr);
        complexVectorPtr += 8;
 
        result = _mm256_magnitude_ps(cplxValue1, cplxValue2);
        _mm256_store_ps(magnitudeVectorPtr, result);
        magnitudeVectorPtr += 8;
    }
 
    number = eighthPoints * 8;
    for (; number < num_points; number++) {
        float val1Real = *complexVectorPtr++;
        float val1Imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
    }
}
#endif /* LV_HAVE_AVX */
 
#ifdef LV_HAVE_SSE3
#include <pmmintrin.h>
#include <volk/volk_sse3_intrinsics.h>
 
static inline void volk_32fc_magnitude_32f_a_sse3(float* magnitudeVector,
                                                  const lv_32fc_t* complexVector,
                                                  unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int quarterPoints = num_points / 4;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m128 cplxValue1, cplxValue2, result;
    for (; number < quarterPoints; number++) {
        cplxValue1 = _mm_load_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        cplxValue2 = _mm_load_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        result = _mm_magnitude_ps_sse3(cplxValue1, cplxValue2);
        _mm_store_ps(magnitudeVectorPtr, result);
        magnitudeVectorPtr += 4;
    }
 
    number = quarterPoints * 4;
    for (; number < num_points; number++) {
        float val1Real = *complexVectorPtr++;
        float val1Imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
    }
}
#endif /* LV_HAVE_SSE3 */
 
#ifdef LV_HAVE_SSE
#include <volk/volk_sse_intrinsics.h>
#include <xmmintrin.h>
 
static inline void volk_32fc_magnitude_32f_a_sse(float* magnitudeVector,
                                                 const lv_32fc_t* complexVector,
                                                 unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int quarterPoints = num_points / 4;
 
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    __m128 cplxValue1, cplxValue2, result;
    for (; number < quarterPoints; number++) {
        cplxValue1 = _mm_load_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        cplxValue2 = _mm_load_ps(complexVectorPtr);
        complexVectorPtr += 4;
 
        result = _mm_magnitude_ps(cplxValue1, cplxValue2);
        _mm_store_ps(magnitudeVectorPtr, result);
        magnitudeVectorPtr += 4;
    }
 
    number = quarterPoints * 4;
    for (; number < num_points; number++) {
        float val1Real = *complexVectorPtr++;
        float val1Imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
    }
}
#endif /* LV_HAVE_SSE */
 
 
#ifdef LV_HAVE_NEON
#include <arm_neon.h>
 
static inline void volk_32fc_magnitude_32f_neon(float* magnitudeVector,
                                                const lv_32fc_t* complexVector,
                                                unsigned int num_points)
{
    unsigned int number;
    unsigned int quarter_points = num_points / 4;
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    float32x4x2_t complex_vec;
    float32x4_t magnitude_vec;
    for (number = 0; number < quarter_points; number++) {
        complex_vec = vld2q_f32(complexVectorPtr);
        complex_vec.val[0] = vmulq_f32(complex_vec.val[0], complex_vec.val[0]);
        magnitude_vec =
            vmlaq_f32(complex_vec.val[0], complex_vec.val[1], complex_vec.val[1]);
        magnitude_vec = vrsqrteq_f32(magnitude_vec);
        magnitude_vec = vrecpeq_f32(magnitude_vec); // no plain ol' sqrt
        vst1q_f32(magnitudeVectorPtr, magnitude_vec);
 
        complexVectorPtr += 8;
        magnitudeVectorPtr += 4;
    }
 
    for (number = quarter_points * 4; number < num_points; number++) {
        const float real = *complexVectorPtr++;
        const float imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((real * real) + (imag * imag));
    }
}
#endif /* LV_HAVE_NEON */
 
 
#ifdef LV_HAVE_NEONV8
#include <arm_neon.h>
 
static inline void volk_32fc_magnitude_32f_neonv8(float* magnitudeVector,
                                                  const lv_32fc_t* complexVector,
                                                  unsigned int num_points)
{
    unsigned int number;
    unsigned int quarter_points = num_points / 4;
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    float32x4x2_t complex_vec;
    float32x4_t magnitude_sq, magnitude_vec;
    for (number = 0; number < quarter_points; number++) {
        complex_vec = vld2q_f32(complexVectorPtr);
        __VOLK_PREFETCH(complexVectorPtr + 8);
        // Use FMA for magnitude squared: real^2 + imag^2
        magnitude_sq = vfmaq_f32(vmulq_f32(complex_vec.val[0], complex_vec.val[0]),
                                 complex_vec.val[1],
                                 complex_vec.val[1]);
        // ARMv8 native sqrt
        magnitude_vec = vsqrtq_f32(magnitude_sq);
        vst1q_f32(magnitudeVectorPtr, magnitude_vec);
 
        complexVectorPtr += 8;
        magnitudeVectorPtr += 4;
    }
 
    for (number = quarter_points * 4; number < num_points; number++) {
        const float real = *complexVectorPtr++;
        const float imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((real * real) + (imag * imag));
    }
}
#endif /* LV_HAVE_NEONV8 */
 
 
#ifdef LV_HAVE_NEON
static inline void volk_32fc_magnitude_32f_neon_fancy_sweet(
    float* magnitudeVector, const lv_32fc_t* complexVector, unsigned int num_points)
{
    unsigned int number;
    unsigned int quarter_points = num_points / 4;
    const float* complexVectorPtr = (float*)complexVector;
    float* magnitudeVectorPtr = magnitudeVector;
 
    const float threshold = 0.4142135;
 
    float32x4_t a_vec, b_vec, a_high, a_low, b_high, b_low;
    a_high = vdupq_n_f32(0.84);
    b_high = vdupq_n_f32(0.561);
    a_low = vdupq_n_f32(0.99);
    b_low = vdupq_n_f32(0.197);
 
    uint32x4_t comp0, comp1;
 
    float32x4x2_t complex_vec;
    float32x4_t min_vec, max_vec, magnitude_vec;
    float32x4_t real_abs, imag_abs;
    for (number = 0; number < quarter_points; number++) {
        complex_vec = vld2q_f32(complexVectorPtr);
 
        real_abs = vabsq_f32(complex_vec.val[0]);
        imag_abs = vabsq_f32(complex_vec.val[1]);
 
        min_vec = vminq_f32(real_abs, imag_abs);
        max_vec = vmaxq_f32(real_abs, imag_abs);
 
        // effective branch to choose coefficient pair.
        comp0 = vcgtq_f32(min_vec, vmulq_n_f32(max_vec, threshold));
        comp1 = vcleq_f32(min_vec, vmulq_n_f32(max_vec, threshold));
 
        // and 0s or 1s with coefficients from previous effective branch
        a_vec = (float32x4_t)vaddq_s32(vandq_s32((int32x4_t)comp0, (int32x4_t)a_high),
                                       vandq_s32((int32x4_t)comp1, (int32x4_t)a_low));
        b_vec = (float32x4_t)vaddq_s32(vandq_s32((int32x4_t)comp0, (int32x4_t)b_high),
                                       vandq_s32((int32x4_t)comp1, (int32x4_t)b_low));
 
        // coefficients chosen, do the weighted sum
        min_vec = vmulq_f32(min_vec, b_vec);
        max_vec = vmulq_f32(max_vec, a_vec);
 
        magnitude_vec = vaddq_f32(min_vec, max_vec);
        vst1q_f32(magnitudeVectorPtr, magnitude_vec);
 
        complexVectorPtr += 8;
        magnitudeVectorPtr += 4;
    }
 
    for (number = quarter_points * 4; number < num_points; number++) {
        const float real = *complexVectorPtr++;
        const float imag = *complexVectorPtr++;
        *magnitudeVectorPtr++ = sqrtf((real * real) + (imag * imag));
    }
}
#endif /* LV_HAVE_NEON */
 
#ifdef LV_HAVE_RVV
#include <riscv_vector.h>
 
static inline void volk_32fc_magnitude_32f_rvv(float* magnitudeVector,
                                               const lv_32fc_t* complexVector,
                                               unsigned int num_points)
{
    size_t n = num_points;
    for (size_t vl; n > 0; n -= vl, complexVector += vl, magnitudeVector += vl) {
        vl = __riscv_vsetvl_e32m4(n);
        vuint64m8_t vc = __riscv_vle64_v_u64m8((const uint64_t*)complexVector, vl);
        vfloat32m4_t vr = __riscv_vreinterpret_f32m4(__riscv_vnsrl(vc, 0, vl));
        vfloat32m4_t vi = __riscv_vreinterpret_f32m4(__riscv_vnsrl(vc, 32, vl));
        vfloat32m4_t v = __riscv_vfmacc(__riscv_vfmul(vi, vi, vl), vr, vr, vl);
        __riscv_vse32(magnitudeVector, __riscv_vfsqrt(v, vl), vl);
    }
}
#endif /*LV_HAVE_RVV*/
 
#ifdef LV_HAVE_RVVSEG
#include <riscv_vector.h>
 
static inline void volk_32fc_magnitude_32f_rvvseg(float* magnitudeVector,
                                                  const lv_32fc_t* complexVector,
                                                  unsigned int num_points)
{
    size_t n = num_points;
    for (size_t vl; n > 0; n -= vl, complexVector += vl, magnitudeVector += vl) {
        vl = __riscv_vsetvl_e32m4(n);
        vfloat32m4x2_t vc = __riscv_vlseg2e32_v_f32m4x2((const float*)complexVector, vl);
        vfloat32m4_t vr = __riscv_vget_f32m4(vc, 0);
        vfloat32m4_t vi = __riscv_vget_f32m4(vc, 1);
        vfloat32m4_t v = __riscv_vfmacc(__riscv_vfmul(vi, vi, vl), vr, vr, vl);
        __riscv_vse32(magnitudeVector, __riscv_vfsqrt(v, vl), vl);
    }
}
#endif /*LV_HAVE_RVVSEG*/
 
#endif /* INCLUDED_volk_32fc_magnitude_32f_a_H */