How to optimize "u[0]*v[0] + u[2]*v[2]" code line with SSE or GLSL

I have the following function (from the opensource project "recast navigation"):

/// Derives the dot product of two vectors on the xz-plane. (@p u . @p v)
///  @param[in]     u       A vector [(x, y, z)]
///  @param[in]     v       A vector [(x, y, z)]
/// @return The dot product on the xz-plane.
///
/// The vectors are projected onto the xz-plane, so the y-values are ignored.
inline float dtVdot2D(const float* u, const float* v)
{
    return u[0]*v[0] + u[2]*v[2];
}

I ran it through the VS2010 CPU performance test and it showed me that in all recast codebase codeline in this function u[0]*v[0] + u[2]*v[2] is most CPU hot .

How can I CPU optimize (via SSE or GLSL like GLM (if it is easier or faster and appropriate in such case) for example) this line?

Edit: The context in which the calls appear:

bool dtClosestHeightPointTriangle(const float* p, const float* a, const float* b, const float* c, float& h) {
    float v0[3], v1[3], v2[3];
    dtVsub(v0, c,a);
    dtVsub(v1, b,a);
    dtVsub(v2, p,a);

    const float dot00 = dtVdot2D(v0, v0);
    const float dot01 = dtVdot2D(v0, v1);
    const float dot02 = dtVdot2D(v0, v2);
    const float dot11 = dtVdot2D(v1, v1);
    const float dot12 = dtVdot2D(v1, v2);

    // Compute barycentric coordinates
    const float invDenom = 1.0f / (dot00 * dot11 - dot01 * dot01);
    const float u = (dot11 * dot02 - dot01 * dot12) * invDenom;
    const float v = (dot00 * dot12 - dot01 * dot02) * invDenom;