Created attachment 2713 [details]
Micro-optimize VectorNormalize{,2}

The current VectorNormalize{,2} functions call sqrt unnecessarily.

vec_t VectorNormalize( vec3_t v ) {
    // NOTE: TTimo - Apple G4 altivec source uses double?
    float   length, ilength;

    length = v[0]*v[0] + v[1]*v[1] + v[2]*v[2];
    length = sqrt (length);

    if ( length ) {
        ilength = 1/length;
        v[0] *= ilength;
        v[1] *= ilength;
        v[2] *= ilength;
    }

    return length;
}

sqrt(length) == 0.0f if and only if length = 0.0f. Thus, the sqrt should be moved inside the if statement.

Additionally, gcc is unable to recognize that it may use reciprocal-sqrt when available to calculate ilength. As such, it's more efficient to do this

ilength = 1/sqrtf(length);
length *= ilength;

rather than

length = sqrt(length);
ilength = 1/length;

since the first consists of a reciprocal-sqrt and multiply and the second is a sqrt and a reciprocal or divide.

Also note, sqrt_f_ should be used instead of sqrt. Otherwise gcc wants to convert to and from double-precision which is silly. sqrtf isn't available to QVMs though, so casting (float)sqrt(length) is used since it generates identical code to sqrtf.

Attached is the final patch.