Indices for the float constant map should be multiplied by 4 because
we're loading 4 component float vectors, not because the size of a
float is 4.