SIMD dot products: ARM NEON, SSE3, SSE

dot.c

#if defined(__ARM_NEON__)
vec4 dot(vec4 a, vec4 b)
{
    vec4 prod = vmulq_f32(a, b);
    vec4 sum1 = vaddq_f32(prod, vrev64q_f32(prod));
    vec4 sum2 = vaddq_f32(sum1, vcombine_f32(vget_high_f32(sum1), vget_low_f32(sum1)));
    return sum2;
}
#else if defined(__SSE3__)
static inline vec4 vdot(vec4 x, vec4 y)
{
    vec4 prod = x * y;
    vec4 sum1 = _mm_hadd_ps(prod, prod);
    vec4 sum2 = _mm_hadd_ps(sum1, sum1);
    return sum2;
}
#else // SSE
static inline vec4 vdot(vec4 x, vec4 y)
{
    vec4 prod = x * y;
    vec4 sum1 = prod + vshuffle(prod, prod, 1, 0, 3, 2);
    vec4 sum2 = sum1 + vshuffle(sum1, sum1, 2, 2, 0, 0);
    return sum2;
}
#endif