I didn't find any discussion on the matter, but I'm wondering if there's a need for the incrementing functions to use uintptr types for index, increment, and length parameters. My main issue is with negative increments. The asm side doesn't matter, but relying on overflow mechanics on the Go side just feels wrong. It might even cause issues on some architectures (I only have access to amd64, so it's more of a nagging fear).
ex:
f64.AxpyInc(alpha float64, x,y []float64, n, incX, incY, ix, iy uintptr)
f64.ScalInc(alpha float64, x []float64, n, incX uintptr)
All of those parameters would work and make more sense as ints. I would expect most of the inputs to be stored as ints, too, so it would clean up the calling sites, as well.