avx512.c

float sum_avx512(const float* arr, size_t n) {
    __m512 acc = _mm512_setzero_ps();
    size_t i = 0;

    for (; i + 16 <= n; i += 16) {
        __m512 v = _mm512_loadu_ps(&arr[i]);
        acc = _mm512_add_ps(acc, v);
    }

    float result = _mm512_reduce_add_ps(acc);

    // tail
    for (; i < n; i++) {
        result += arr[i];
    }

    return result;
}

float sum_avx512_masked(const float* arr, size_t n) {
    __m512 acc = _mm512_setzero_ps();
    size_t i = 0;

    for (; i + 16 <= n; i += 16) {
        acc = _mm512_add_ps(acc, _mm512_loadu_ps(&arr[i]));
    }

    if (i < n) {
        __mmask16 mask = (1U << (n - i)) - 1;
        acc = _mm512_add_ps(acc, _mm512_maskz_loadu_ps(mask, &arr[i]));
    }

    return _mm512_reduce_add_ps(acc);
}

rvv.c

#include <riscv_vector.h>

float sum_rvv(const float* arr, size_t n) {
    vfloat32m1_t acc = __riscv_vfmv_s_f_f32m1(0.0f, 1);
    
    for (size_t i = 0; i < n; ) {
        size_t vl = __riscv_vsetvl_e32m1(n - i);
        vfloat32m1_t v = __riscv_vle32_v_f32m1(&arr[i], vl);
        acc = __riscv_vfredusum_vs_f32m1_f32m1(v, acc, vl);
        i += vl;
    }

    return __riscv_vfmv_f_s_f32m1_f32(acc);
}

sve2.c

#include <arm_sve.h>

float sum_sve2(const float* arr, size_t n) {
    svfloat32_t acc = svdup_f32(0.0f);

    for (size_t i = 0; i < n; i += svcntw()) {
        svbool_t pg = svwhilelt_b32((uint64_t)i, (uint64_t)n);
        svfloat32_t v = svld1_f32(pg, &arr[i]);
        acc = svadd_f32_m(pg, acc, v);
    }

    return svaddv_f32(svptrue_b32(), acc);
}