volk_32fc_x2_conjugate_dot_prod_32fc.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_x2_conjugate_dot_prod_32fc_u_H
#define INCLUDED_volk_32fc_x2_conjugate_dot_prod_32fc_u_H
 
 
#include <volk/volk_complex.h>
 
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_generic(lv_32fc_t* result,
                                                                const lv_32fc_t* input,
                                                                const lv_32fc_t* taps,
                                                                unsigned int num_points)
{
    lv_32fc_t res = lv_cmake(0.f, 0.f);
    for (unsigned int i = 0; i < num_points; ++i) {
        res += (*input++) * lv_conj((*taps++));
    }
    *result = res;
}
 
#endif /*LV_HAVE_GENERIC*/
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_block(lv_32fc_t* result,
                                                              const lv_32fc_t* input,
                                                              const lv_32fc_t* taps,
                                                              unsigned int num_points)
{
 
    const unsigned int num_bytes = num_points * 8;
 
    float* res = (float*)result;
    float* in = (float*)input;
    float* tp = (float*)taps;
    unsigned int n_2_ccomplex_blocks = num_bytes >> 4;
 
    float sum0[2] = { 0, 0 };
    float sum1[2] = { 0, 0 };
    unsigned int i = 0;
 
    for (i = 0; i < n_2_ccomplex_blocks; ++i) {
        sum0[0] += in[0] * tp[0] + in[1] * tp[1];
        sum0[1] += (-in[0] * tp[1]) + in[1] * tp[0];
        sum1[0] += in[2] * tp[2] + in[3] * tp[3];
        sum1[1] += (-in[2] * tp[3]) + in[3] * tp[2];
 
        in += 4;
        tp += 4;
    }
 
    res[0] = sum0[0] + sum1[0];
    res[1] = sum0[1] + sum1[1];
 
    if (num_bytes >> 3 & 1) {
        *result += input[(num_bytes >> 3) - 1] * lv_conj(taps[(num_bytes >> 3) - 1]);
    }
}
 
#endif /*LV_HAVE_GENERIC*/
 
#ifdef LV_HAVE_SSE3
 
#include <pmmintrin.h>
#include <xmmintrin.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_u_sse3(lv_32fc_t* result,
                                                               const lv_32fc_t* input,
                                                               const lv_32fc_t* taps,
                                                               unsigned int num_points)
{
    // Partial sums for indices i and i+1.
    __m128 sum_a_mult_b_real = _mm_setzero_ps();
    __m128 sum_a_mult_b_imag = _mm_setzero_ps();
 
    for (long unsigned i = 0; i < (num_points & ~1u); i += 2) {
        /* Two complex elements a time are processed.
         * (ar + j⋅ai)*conj(br + j⋅bi) =
         * ar⋅br + ai⋅bi + j⋅(ai⋅br − ar⋅bi)
         */
 
        /* Load input and taps, split and duplicate real und imaginary parts of taps.
         * a: | ai,i+1 | ar,i+1 | ai,i+0 | ar,i+0 |
         * b: | bi,i+1 | br,i+1 | bi,i+0 | br,i+0 |
         * b_real: | br,i+1 | br,i+1 | br,i+0 | br,i+0 |
         * b_imag: | bi,i+1 | bi,i+1 | bi,i+0 | bi,i+0 |
         */
        __m128 a = _mm_loadu_ps((const float*)&input[i]);
        __m128 b = _mm_loadu_ps((const float*)&taps[i]);
        __m128 b_real = _mm_moveldup_ps(b);
        __m128 b_imag = _mm_movehdup_ps(b);
 
        // Add | ai⋅br,i+1 | ar⋅br,i+1 | ai⋅br,i+0 | ar⋅br,i+0 | to partial sum.
        sum_a_mult_b_real = _mm_add_ps(sum_a_mult_b_real, _mm_mul_ps(a, b_real));
        // Add | ai⋅bi,i+1 | −ar⋅bi,i+1 | ai⋅bi,i+0 | −ar⋅bi,i+0 | to partial sum.
        sum_a_mult_b_imag = _mm_addsub_ps(sum_a_mult_b_imag, _mm_mul_ps(a, b_imag));
    }
 
    // Swap position of −ar⋅bi and ai⋅bi.
    sum_a_mult_b_imag =
        _mm_shuffle_ps(sum_a_mult_b_imag, sum_a_mult_b_imag, _MM_SHUFFLE(2, 3, 0, 1));
    // | ai⋅br + ai⋅bi | ai⋅br − ar⋅bi |, sum contains two such partial sums.
    __m128 sum = _mm_add_ps(sum_a_mult_b_real, sum_a_mult_b_imag);
    // Sum the two partial sums.
    sum = _mm_add_ps(sum, _mm_shuffle_ps(sum, sum, _MM_SHUFFLE(1, 0, 3, 2)));
    // Store result.
    _mm_storel_pi((__m64*)result, sum);
 
    // Handle the last element if num_points mod 2 is 1.
    if (num_points & 1u) {
        *result += lv_cmake(
            lv_creal(input[num_points - 1]) * lv_creal(taps[num_points - 1]) +
                lv_cimag(input[num_points - 1]) * lv_cimag(taps[num_points - 1]),
            lv_cimag(input[num_points - 1]) * lv_creal(taps[num_points - 1]) -
                lv_creal(input[num_points - 1]) * lv_cimag(taps[num_points - 1]));
    }
}
 
#endif /*LV_HAVE_SSE3*/
 
#ifdef LV_HAVE_SSE3
 
#include <pmmintrin.h>
#include <xmmintrin.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_a_sse3(lv_32fc_t* result,
                                                               const lv_32fc_t* input,
                                                               const lv_32fc_t* taps,
                                                               unsigned int num_points)
{
    // Partial sums for indices i and i+1.
    __m128 sum_a_mult_b_real = _mm_setzero_ps();
    __m128 sum_a_mult_b_imag = _mm_setzero_ps();
 
    for (long unsigned i = 0; i < (num_points & ~1u); i += 2) {
        /* Two complex elements a time are processed.
         * (ar + j⋅ai)*conj(br + j⋅bi) =
         * ar⋅br + ai⋅bi + j⋅(ai⋅br − ar⋅bi)
         */
 
        /* Load input and taps, split and duplicate real und imaginary parts of taps.
         * a: | ai,i+1 | ar,i+1 | ai,i+0 | ar,i+0 |
         * b: | bi,i+1 | br,i+1 | bi,i+0 | br,i+0 |
         * b_real: | br,i+1 | br,i+1 | br,i+0 | br,i+0 |
         * b_imag: | bi,i+1 | bi,i+1 | bi,i+0 | bi,i+0 |
         */
        __m128 a = _mm_load_ps((const float*)&input[i]);
        __m128 b = _mm_load_ps((const float*)&taps[i]);
        __m128 b_real = _mm_moveldup_ps(b);
        __m128 b_imag = _mm_movehdup_ps(b);
 
        // Add | ai⋅br,i+1 | ar⋅br,i+1 | ai⋅br,i+0 | ar⋅br,i+0 | to partial sum.
        sum_a_mult_b_real = _mm_add_ps(sum_a_mult_b_real, _mm_mul_ps(a, b_real));
        // Add | ai⋅bi,i+1 | −ar⋅bi,i+1 | ai⋅bi,i+0 | −ar⋅bi,i+0 | to partial sum.
        sum_a_mult_b_imag = _mm_addsub_ps(sum_a_mult_b_imag, _mm_mul_ps(a, b_imag));
    }
 
    // Swap position of −ar⋅bi and ai⋅bi.
    sum_a_mult_b_imag =
        _mm_shuffle_ps(sum_a_mult_b_imag, sum_a_mult_b_imag, _MM_SHUFFLE(2, 3, 0, 1));
    // | ai⋅br + ai⋅bi | ai⋅br − ar⋅bi |, sum contains two such partial sums.
    __m128 sum = _mm_add_ps(sum_a_mult_b_real, sum_a_mult_b_imag);
    // Sum the two partial sums.
    sum = _mm_add_ps(sum, _mm_shuffle_ps(sum, sum, _MM_SHUFFLE(1, 0, 3, 2)));
    // Store result.
    _mm_storel_pi((__m64*)result, sum);
 
    // Handle the last element if num_points mod 2 is 1.
    if (num_points & 1u) {
        *result += lv_cmake(
            lv_creal(input[num_points - 1]) * lv_creal(taps[num_points - 1]) +
                lv_cimag(input[num_points - 1]) * lv_cimag(taps[num_points - 1]),
            lv_cimag(input[num_points - 1]) * lv_creal(taps[num_points - 1]) -
                lv_creal(input[num_points - 1]) * lv_cimag(taps[num_points - 1]));
    }
}
 
#endif /*LV_HAVE_SSE3*/
 
#ifdef LV_HAVE_AVX
 
#include <immintrin.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_u_avx(lv_32fc_t* result,
                                                              const lv_32fc_t* input,
                                                              const lv_32fc_t* taps,
                                                              unsigned int num_points)
{
    // Partial sums for indices i, i+1, i+2 and i+3.
    __m256 sum_a_mult_b_real = _mm256_setzero_ps();
    __m256 sum_a_mult_b_imag = _mm256_setzero_ps();
 
    for (long unsigned i = 0; i < (num_points & ~3u); i += 4) {
        /* Four complex elements a time are processed.
         * (ar + j⋅ai)*conj(br + j⋅bi) =
         * ar⋅br + ai⋅bi + j⋅(ai⋅br − ar⋅bi)
         */
 
        /* Load input and taps, split and duplicate real und imaginary parts of taps.
         * a: | ai,i+3 | ar,i+3 | … | ai,i+1 | ar,i+1 | ai,i+0 | ar,i+0 |
         * b: | bi,i+3 | br,i+3 | … | bi,i+1 | br,i+1 | bi,i+0 | br,i+0 |
         * b_real: | br,i+3 | br,i+3 | … | br,i+1 | br,i+1 | br,i+0 | br,i+0 |
         * b_imag: | bi,i+3 | bi,i+3 | … | bi,i+1 | bi,i+1 | bi,i+0 | bi,i+0 |
         */
        __m256 a = _mm256_loadu_ps((const float*)&input[i]);
        __m256 b = _mm256_loadu_ps((const float*)&taps[i]);
        __m256 b_real = _mm256_moveldup_ps(b);
        __m256 b_imag = _mm256_movehdup_ps(b);
 
        // Add | ai⋅br,i+3 | ar⋅br,i+3 | … | ai⋅br,i+0 | ar⋅br,i+0 | to partial sum.
        sum_a_mult_b_real = _mm256_add_ps(sum_a_mult_b_real, _mm256_mul_ps(a, b_real));
        // Add | ai⋅bi,i+3 | −ar⋅bi,i+3 | … | ai⋅bi,i+0 | −ar⋅bi,i+0 | to partial sum.
        sum_a_mult_b_imag = _mm256_addsub_ps(sum_a_mult_b_imag, _mm256_mul_ps(a, b_imag));
    }
 
    // Swap position of −ar⋅bi and ai⋅bi.
    sum_a_mult_b_imag = _mm256_permute_ps(sum_a_mult_b_imag, _MM_SHUFFLE(2, 3, 0, 1));
    // | ai⋅br + ai⋅bi | ai⋅br − ar⋅bi |, sum contains four such partial sums.
    __m256 sum = _mm256_add_ps(sum_a_mult_b_real, sum_a_mult_b_imag);
    /* Sum the four partial sums: Add high half of vector sum to the low one, i.e.
     * s1 + s3 and s0 + s2 …
     */
    sum = _mm256_add_ps(sum, _mm256_permute2f128_ps(sum, sum, 0x01));
    // … and now (s0 + s2) + (s1 + s3)
    sum = _mm256_add_ps(sum, _mm256_permute_ps(sum, _MM_SHUFFLE(1, 0, 3, 2)));
    // Store result.
    __m128 lower = _mm256_extractf128_ps(sum, 0);
    _mm_storel_pi((__m64*)result, lower);
 
    // Handle the last elements if num_points mod 4 is bigger than 0.
    for (long unsigned i = num_points & ~3u; i < num_points; ++i) {
        *result += lv_cmake(lv_creal(input[i]) * lv_creal(taps[i]) +
                                lv_cimag(input[i]) * lv_cimag(taps[i]),
                            lv_cimag(input[i]) * lv_creal(taps[i]) -
                                lv_creal(input[i]) * lv_cimag(taps[i]));
    }
}
 
#endif /* LV_HAVE_AVX */
 
#ifdef LV_HAVE_AVX
#include <immintrin.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_a_avx(lv_32fc_t* result,
                                                              const lv_32fc_t* input,
                                                              const lv_32fc_t* taps,
                                                              unsigned int num_points)
{
    // Partial sums for indices i, i+1, i+2 and i+3.
    __m256 sum_a_mult_b_real = _mm256_setzero_ps();
    __m256 sum_a_mult_b_imag = _mm256_setzero_ps();
 
    for (long unsigned i = 0; i < (num_points & ~3u); i += 4) {
        /* Four complex elements a time are processed.
         * (ar + j⋅ai)*conj(br + j⋅bi) =
         * ar⋅br + ai⋅bi + j⋅(ai⋅br − ar⋅bi)
         */
 
        /* Load input and taps, split and duplicate real und imaginary parts of taps.
         * a: | ai,i+3 | ar,i+3 | … | ai,i+1 | ar,i+1 | ai,i+0 | ar,i+0 |
         * b: | bi,i+3 | br,i+3 | … | bi,i+1 | br,i+1 | bi,i+0 | br,i+0 |
         * b_real: | br,i+3 | br,i+3 | … | br,i+1 | br,i+1 | br,i+0 | br,i+0 |
         * b_imag: | bi,i+3 | bi,i+3 | … | bi,i+1 | bi,i+1 | bi,i+0 | bi,i+0 |
         */
        __m256 a = _mm256_load_ps((const float*)&input[i]);
        __m256 b = _mm256_load_ps((const float*)&taps[i]);
        __m256 b_real = _mm256_moveldup_ps(b);
        __m256 b_imag = _mm256_movehdup_ps(b);
 
        // Add | ai⋅br,i+3 | ar⋅br,i+3 | … | ai⋅br,i+0 | ar⋅br,i+0 | to partial sum.
        sum_a_mult_b_real = _mm256_add_ps(sum_a_mult_b_real, _mm256_mul_ps(a, b_real));
        // Add | ai⋅bi,i+3 | −ar⋅bi,i+3 | … | ai⋅bi,i+0 | −ar⋅bi,i+0 | to partial sum.
        sum_a_mult_b_imag = _mm256_addsub_ps(sum_a_mult_b_imag, _mm256_mul_ps(a, b_imag));
    }
 
    // Swap position of −ar⋅bi and ai⋅bi.
    sum_a_mult_b_imag = _mm256_permute_ps(sum_a_mult_b_imag, _MM_SHUFFLE(2, 3, 0, 1));
    // | ai⋅br + ai⋅bi | ai⋅br − ar⋅bi |, sum contains four such partial sums.
    __m256 sum = _mm256_add_ps(sum_a_mult_b_real, sum_a_mult_b_imag);
    /* Sum the four partial sums: Add high half of vector sum to the low one, i.e.
     * s1 + s3 and s0 + s2 …
     */
    sum = _mm256_add_ps(sum, _mm256_permute2f128_ps(sum, sum, 0x01));
    // … and now (s0 + s2) + (s1 + s3)
    sum = _mm256_add_ps(sum, _mm256_permute_ps(sum, _MM_SHUFFLE(1, 0, 3, 2)));
    // Store result.
    __m128 lower = _mm256_extractf128_ps(sum, 0);
    _mm_storel_pi((__m64*)result, lower);
 
    // Handle the last elements if num_points mod 4 is bigger than 0.
    for (long unsigned i = num_points & ~3u; i < num_points; ++i) {
        *result += lv_cmake(lv_creal(input[i]) * lv_creal(taps[i]) +
                                lv_cimag(input[i]) * lv_cimag(taps[i]),
                            lv_cimag(input[i]) * lv_creal(taps[i]) -
                                lv_creal(input[i]) * lv_cimag(taps[i]));
    }
}
#endif /* LV_HAVE_AVX */
 
#if LV_HAVE_AVX512F && LV_HAVE_AVX512DQ
 
#include <immintrin.h>
 
static inline void
volk_32fc_x2_conjugate_dot_prod_32fc_u_avx512dq(lv_32fc_t* result,
                                                const lv_32fc_t* input,
                                                const lv_32fc_t* taps,
                                                unsigned int num_points)
{
    // Partial sums for indices i through i+7 (8 complex elements).
    __m512 sum_a_mult_b_real = _mm512_setzero_ps();
    __m512 sum_a_mult_b_imag = _mm512_setzero_ps();
 
    // Sign mask for emulating addsub: negate even indices (real parts)
    const __m512 sign_mask = _mm512_castsi512_ps(_mm512_set_epi32(0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000));
 
    for (long unsigned i = 0; i < (num_points & ~7u); i += 8) {
        /* Eight complex elements a time are processed.
         * (ar + j⋅ai)*conj(br + j⋅bi) =
         * ar⋅br + ai⋅bi + j⋅(ai⋅br − ar⋅bi)
         */
 
        // Load input and taps.
        __m512 a = _mm512_loadu_ps((const float*)&input[i]);
        __m512 b = _mm512_loadu_ps((const float*)&taps[i]);
 
        // Duplicate real and imaginary parts of taps
        __m512 b_real = _mm512_moveldup_ps(b); // duplicate even elements (real)
        __m512 b_imag = _mm512_movehdup_ps(b); // duplicate odd elements (imag)
 
        // Accumulate ar⋅br and ai⋅br
        sum_a_mult_b_real = _mm512_fmadd_ps(a, b_real, sum_a_mult_b_real);
 
        // Emulate addsub: compute a*b_imag then negate even indices
        __m512 mult_imag = _mm512_mul_ps(a, b_imag);
        mult_imag = _mm512_xor_ps(mult_imag, sign_mask); // flip sign of even indices
        sum_a_mult_b_imag = _mm512_add_ps(sum_a_mult_b_imag, mult_imag);
    }
 
    // Swap position of −ar⋅bi and ai⋅bi, then add
    sum_a_mult_b_imag = _mm512_permute_ps(sum_a_mult_b_imag, 0xB1); // swap pairs
    __m512 sum = _mm512_add_ps(sum_a_mult_b_real, sum_a_mult_b_imag);
 
    // Horizontal sum: reduce 8 complex numbers to 1
    // First reduce to 4 complex (256 bits)
    __m256 sum_high = _mm512_extractf32x8_ps(sum, 1);
    __m256 sum_low = _mm512_castps512_ps256(sum);
    __m256 sum256 = _mm256_add_ps(sum_high, sum_low);
 
    // Reduce to 2 complex (128 bits)
    __m128 sum128_high = _mm256_extractf128_ps(sum256, 1);
    __m128 sum128_low = _mm256_castps256_ps128(sum256);
    __m128 sum128 = _mm_add_ps(sum128_high, sum128_low);
 
    // Final reduction to 1 complex number
    sum128 = _mm_add_ps(sum128, _mm_shuffle_ps(sum128, sum128, _MM_SHUFFLE(1, 0, 3, 2)));
 
    // Store result
    _mm_storel_pi((__m64*)result, sum128);
 
    // Handle remaining elements
    for (long unsigned i = num_points & ~7u; i < num_points; ++i) {
        *result += lv_cmake(lv_creal(input[i]) * lv_creal(taps[i]) +
                                lv_cimag(input[i]) * lv_cimag(taps[i]),
                            lv_cimag(input[i]) * lv_creal(taps[i]) -
                                lv_creal(input[i]) * lv_cimag(taps[i]));
    }
}
 
#endif /* LV_HAVE_AVX512F && LV_HAVE_AVX512DQ */
 
#if LV_HAVE_AVX512F && LV_HAVE_AVX512DQ
 
#include <immintrin.h>
 
static inline void
volk_32fc_x2_conjugate_dot_prod_32fc_a_avx512dq(lv_32fc_t* result,
                                                const lv_32fc_t* input,
                                                const lv_32fc_t* taps,
                                                unsigned int num_points)
{
    // Partial sums for indices i through i+7 (8 complex elements).
    __m512 sum_a_mult_b_real = _mm512_setzero_ps();
    __m512 sum_a_mult_b_imag = _mm512_setzero_ps();
 
    // Sign mask for emulating addsub: negate even indices (real parts)
    const __m512 sign_mask = _mm512_castsi512_ps(_mm512_set_epi32(0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000,
                                                                  0,
                                                                  0x80000000));
 
    for (long unsigned i = 0; i < (num_points & ~7u); i += 8) {
        /* Eight complex elements a time are processed.
         * (ar + j⋅ai)*conj(br + j⋅bi) =
         * ar⋅br + ai⋅bi + j⋅(ai⋅br − ar⋅bi)
         */
 
        // Load input and taps (aligned).
        __m512 a = _mm512_load_ps((const float*)&input[i]);
        __m512 b = _mm512_load_ps((const float*)&taps[i]);
 
        // Duplicate real and imaginary parts of taps
        __m512 b_real = _mm512_moveldup_ps(b); // duplicate even elements (real)
        __m512 b_imag = _mm512_movehdup_ps(b); // duplicate odd elements (imag)
 
        // Accumulate ar⋅br and ai⋅br
        sum_a_mult_b_real = _mm512_fmadd_ps(a, b_real, sum_a_mult_b_real);
 
        // Emulate addsub: compute a*b_imag then negate even indices
        __m512 mult_imag = _mm512_mul_ps(a, b_imag);
        mult_imag = _mm512_xor_ps(mult_imag, sign_mask); // flip sign of even indices
        sum_a_mult_b_imag = _mm512_add_ps(sum_a_mult_b_imag, mult_imag);
    }
 
    // Swap position of −ar⋅bi and ai⋅bi, then add
    sum_a_mult_b_imag = _mm512_permute_ps(sum_a_mult_b_imag, 0xB1);
    __m512 sum = _mm512_add_ps(sum_a_mult_b_real, sum_a_mult_b_imag);
 
    // Horizontal sum: reduce 8 complex numbers to 1
    __m256 sum_high = _mm512_extractf32x8_ps(sum, 1);
    __m256 sum_low = _mm512_castps512_ps256(sum);
    __m256 sum256 = _mm256_add_ps(sum_high, sum_low);
 
    __m128 sum128_high = _mm256_extractf128_ps(sum256, 1);
    __m128 sum128_low = _mm256_castps256_ps128(sum256);
    __m128 sum128 = _mm_add_ps(sum128_high, sum128_low);
 
    sum128 = _mm_add_ps(sum128, _mm_shuffle_ps(sum128, sum128, _MM_SHUFFLE(1, 0, 3, 2)));
 
    _mm_storel_pi((__m64*)result, sum128);
 
    // Handle remaining elements
    for (long unsigned i = num_points & ~7u; i < num_points; ++i) {
        *result += lv_cmake(lv_creal(input[i]) * lv_creal(taps[i]) +
                                lv_cimag(input[i]) * lv_cimag(taps[i]),
                            lv_cimag(input[i]) * lv_creal(taps[i]) -
                                lv_creal(input[i]) * lv_cimag(taps[i]));
    }
}
#endif /* LV_HAVE_AVX512F && LV_HAVE_AVX512DQ */
 
#ifdef LV_HAVE_NEON
#include <arm_neon.h>
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_neon(lv_32fc_t* result,
                                                             const lv_32fc_t* input,
                                                             const lv_32fc_t* taps,
                                                             unsigned int num_points)
{
 
    unsigned int quarter_points = num_points / 4;
    unsigned int number;
 
    lv_32fc_t* a_ptr = (lv_32fc_t*)taps;
    lv_32fc_t* b_ptr = (lv_32fc_t*)input;
    // for 2-lane vectors, 1st lane holds the real part,
    // 2nd lane holds the imaginary part
    float32x4x2_t a_val, b_val, accumulator;
    float32x4x2_t tmp_imag;
    accumulator.val[0] = vdupq_n_f32(0);
    accumulator.val[1] = vdupq_n_f32(0);
 
    for (number = 0; number < quarter_points; ++number) {
        a_val = vld2q_f32((float*)a_ptr); // a0r|a1r|a2r|a3r || a0i|a1i|a2i|a3i
        b_val = vld2q_f32((float*)b_ptr); // b0r|b1r|b2r|b3r || b0i|b1i|b2i|b3i
        __VOLK_PREFETCH(a_ptr + 8);
        __VOLK_PREFETCH(b_ptr + 8);
 
        // do the first multiply
        tmp_imag.val[1] = vmulq_f32(a_val.val[1], b_val.val[0]);
        tmp_imag.val[0] = vmulq_f32(a_val.val[0], b_val.val[0]);
 
        // use multiply accumulate/subtract to get result
        tmp_imag.val[1] = vmlsq_f32(tmp_imag.val[1], a_val.val[0], b_val.val[1]);
        tmp_imag.val[0] = vmlaq_f32(tmp_imag.val[0], a_val.val[1], b_val.val[1]);
 
        accumulator.val[0] = vaddq_f32(accumulator.val[0], tmp_imag.val[0]);
        accumulator.val[1] = vaddq_f32(accumulator.val[1], tmp_imag.val[1]);
 
        // increment pointers
        a_ptr += 4;
        b_ptr += 4;
    }
    lv_32fc_t accum_result[4];
    vst2q_f32((float*)accum_result, accumulator);
    *result = accum_result[0] + accum_result[1] + accum_result[2] + accum_result[3];
 
    // tail case
    for (number = quarter_points * 4; number < num_points; ++number) {
        *result += (*a_ptr++) * lv_conj(*b_ptr++);
    }
    *result = lv_conj(*result);
}
#endif /*LV_HAVE_NEON*/
 
 
#ifdef LV_HAVE_NEONV8
#include <arm_neon.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_neonv8(lv_32fc_t* result,
                                                               const lv_32fc_t* input,
                                                               const lv_32fc_t* taps,
                                                               unsigned int num_points)
{
    unsigned int n = num_points;
    const lv_32fc_t* a = input; /* input */
    const lv_32fc_t* b = taps;  /* taps (will be conjugated) */
 
    /* Use 4 accumulators to break data dependencies */
    float32x4_t acc0_r = vdupq_n_f32(0);
    float32x4_t acc0_i = vdupq_n_f32(0);
    float32x4_t acc1_r = vdupq_n_f32(0);
    float32x4_t acc1_i = vdupq_n_f32(0);
 
    /* Process 8 complex numbers per iteration (2x unroll) */
    while (n >= 8) {
        float32x4x2_t a0 = vld2q_f32((const float*)a);
        float32x4x2_t b0 = vld2q_f32((const float*)b);
        float32x4x2_t a1 = vld2q_f32((const float*)(a + 4));
        float32x4x2_t b1 = vld2q_f32((const float*)(b + 4));
        __VOLK_PREFETCH(a + 8);
        __VOLK_PREFETCH(b + 8);
 
        /* Conjugate dot product: input * conj(taps)
         * (ar + j*ai) * (br - j*bi) = ar*br + ai*bi + j*(ai*br - ar*bi)
         * real += ar*br + ai*bi
         * imag += ai*br - ar*bi
         */
        acc0_r = vfmaq_f32(acc0_r, a0.val[0], b0.val[0]); /* ar*br */
        acc0_r = vfmaq_f32(acc0_r, a0.val[1], b0.val[1]); /* + ai*bi */
        acc0_i = vfmaq_f32(acc0_i, a0.val[1], b0.val[0]); /* ai*br */
        acc0_i = vfmsq_f32(acc0_i, a0.val[0], b0.val[1]); /* - ar*bi */
 
        acc1_r = vfmaq_f32(acc1_r, a1.val[0], b1.val[0]);
        acc1_r = vfmaq_f32(acc1_r, a1.val[1], b1.val[1]);
        acc1_i = vfmaq_f32(acc1_i, a1.val[1], b1.val[0]);
        acc1_i = vfmsq_f32(acc1_i, a1.val[0], b1.val[1]);
 
        a += 8;
        b += 8;
        n -= 8;
    }
 
    /* Process remaining 4 */
    if (n >= 4) {
        float32x4x2_t a0 = vld2q_f32((const float*)a);
        float32x4x2_t b0 = vld2q_f32((const float*)b);
 
        acc0_r = vfmaq_f32(acc0_r, a0.val[0], b0.val[0]);
        acc0_r = vfmaq_f32(acc0_r, a0.val[1], b0.val[1]);
        acc0_i = vfmaq_f32(acc0_i, a0.val[1], b0.val[0]);
        acc0_i = vfmsq_f32(acc0_i, a0.val[0], b0.val[1]);
 
        a += 4;
        b += 4;
        n -= 4;
    }
 
    /* Combine accumulators */
    acc0_r = vaddq_f32(acc0_r, acc1_r);
    acc0_i = vaddq_f32(acc0_i, acc1_i);
 
    /* Horizontal sum using pairwise add */
    float32x2_t sum_r = vadd_f32(vget_low_f32(acc0_r), vget_high_f32(acc0_r));
    float32x2_t sum_i = vadd_f32(vget_low_f32(acc0_i), vget_high_f32(acc0_i));
    sum_r = vpadd_f32(sum_r, sum_r);
    sum_i = vpadd_f32(sum_i, sum_i);
 
    float res_r = vget_lane_f32(sum_r, 0);
    float res_i = vget_lane_f32(sum_i, 0);
 
    /* Scalar tail */
    while (n > 0) {
        res_r += lv_creal(*a) * lv_creal(*b) + lv_cimag(*a) * lv_cimag(*b);
        res_i += lv_cimag(*a) * lv_creal(*b) - lv_creal(*a) * lv_cimag(*b);
        a++;
        b++;
        n--;
    }
 
    *result = lv_cmake(res_r, res_i);
}
 
#endif /* LV_HAVE_NEONV8 */
 
 
#ifdef LV_HAVE_RVV
#include <riscv_vector.h>
#include <volk/volk_rvv_intrinsics.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_rvv(lv_32fc_t* result,
                                                            const lv_32fc_t* input,
                                                            const lv_32fc_t* taps,
                                                            unsigned int num_points)
{
    vfloat32m2_t vsumr = __riscv_vfmv_v_f_f32m2(0, __riscv_vsetvlmax_e32m2());
    vfloat32m2_t vsumi = vsumr;
    size_t n = num_points;
    for (size_t vl; n > 0; n -= vl, input += vl, taps += vl) {
        vl = __riscv_vsetvl_e32m2(n);
        vuint64m4_t va = __riscv_vle64_v_u64m4((const uint64_t*)input, vl);
        vuint64m4_t vb = __riscv_vle64_v_u64m4((const uint64_t*)taps, vl);
        vfloat32m2_t var = __riscv_vreinterpret_f32m2(__riscv_vnsrl(va, 0, vl));
        vfloat32m2_t vbr = __riscv_vreinterpret_f32m2(__riscv_vnsrl(vb, 0, vl));
        vfloat32m2_t vai = __riscv_vreinterpret_f32m2(__riscv_vnsrl(va, 32, vl));
        vfloat32m2_t vbi = __riscv_vreinterpret_f32m2(__riscv_vnsrl(vb, 32, vl));
        vbi = __riscv_vfneg(vbi, vl);
        vfloat32m2_t vr = __riscv_vfnmsac(__riscv_vfmul(var, vbr, vl), vai, vbi, vl);
        vfloat32m2_t vi = __riscv_vfmacc(__riscv_vfmul(var, vbi, vl), vai, vbr, vl);
        vsumr = __riscv_vfadd_tu(vsumr, vsumr, vr, vl);
        vsumi = __riscv_vfadd_tu(vsumi, vsumi, vi, vl);
    }
    size_t vl = __riscv_vsetvlmax_e32m1();
    vfloat32m1_t vr = RISCV_SHRINK2(vfadd, f, 32, vsumr);
    vfloat32m1_t vi = RISCV_SHRINK2(vfadd, f, 32, vsumi);
    vfloat32m1_t z = __riscv_vfmv_s_f_f32m1(0, vl);
    *result = lv_cmake(__riscv_vfmv_f(__riscv_vfredusum(vr, z, vl)),
                       __riscv_vfmv_f(__riscv_vfredusum(vi, z, vl)));
}
#endif /*LV_HAVE_RVV*/
 
#ifdef LV_HAVE_RVVSEG
#include <riscv_vector.h>
#include <volk/volk_rvv_intrinsics.h>
 
static inline void volk_32fc_x2_conjugate_dot_prod_32fc_rvvseg(lv_32fc_t* result,
                                                               const lv_32fc_t* input,
                                                               const lv_32fc_t* taps,
                                                               unsigned int num_points)
{
    vfloat32m2_t vsumr = __riscv_vfmv_v_f_f32m2(0, __riscv_vsetvlmax_e32m2());
    vfloat32m2_t vsumi = vsumr;
    size_t n = num_points;
    for (size_t vl; n > 0; n -= vl, input += vl, taps += vl) {
        vl = __riscv_vsetvl_e32m2(n);
        vfloat32m2x2_t va = __riscv_vlseg2e32_v_f32m2x2((const float*)input, vl);
        vfloat32m2x2_t vb = __riscv_vlseg2e32_v_f32m2x2((const float*)taps, vl);
        vfloat32m2_t var = __riscv_vget_f32m2(va, 0), vai = __riscv_vget_f32m2(va, 1);
        vfloat32m2_t vbr = __riscv_vget_f32m2(vb, 0), vbi = __riscv_vget_f32m2(vb, 1);
        vbi = __riscv_vfneg(vbi, vl);
        vfloat32m2_t vr = __riscv_vfnmsac(__riscv_vfmul(var, vbr, vl), vai, vbi, vl);
        vfloat32m2_t vi = __riscv_vfmacc(__riscv_vfmul(var, vbi, vl), vai, vbr, vl);
        vsumr = __riscv_vfadd_tu(vsumr, vsumr, vr, vl);
        vsumi = __riscv_vfadd_tu(vsumi, vsumi, vi, vl);
    }
    size_t vl = __riscv_vsetvlmax_e32m1();
    vfloat32m1_t vr = RISCV_SHRINK2(vfadd, f, 32, vsumr);
    vfloat32m1_t vi = RISCV_SHRINK2(vfadd, f, 32, vsumi);
    vfloat32m1_t z = __riscv_vfmv_s_f_f32m1(0, vl);
    *result = lv_cmake(__riscv_vfmv_f(__riscv_vfredusum(vr, z, vl)),
                       __riscv_vfmv_f(__riscv_vfredusum(vi, z, vl)));
}
#endif /*LV_HAVE_RVVSEG*/
 
#endif /*INCLUDED_volk_32fc_x2_conjugate_dot_prod_32fc_u_H*/
 
#ifndef INCLUDED_volk_32fc_x2_conjugate_dot_prod_32fc_a_H
#define INCLUDED_volk_32fc_x2_conjugate_dot_prod_32fc_a_H
 
#include <stdio.h>
#include <volk/volk_common.h>
#include <volk/volk_complex.h>
 
#endif /*INCLUDED_volk_32fc_x2_conjugate_dot_prod_32fc_a_H*/