`@fast` macro

[mattsignorelli]

At the time of this issue, the function gofix looks like

function gofix(xy::DAMap, order=1)
  desc = getdesc(xy)
  nv = numvars(desc)

  # 1: v = map-identity in harmonic planes, identity in spin
  v = zero(xy)
  for i=1:nv
    @inbounds v.x[i] =  xy.x[i] - mono(i,use=desc)
  end
  v.Q.q[1] = 1

  # 2: map is cut to order 2 or above
  cut!(v,v,order+1)

  # 3: map is inverted at least to order 1:
  inv!(v,v)

  # 4: a map x is created with dimension nv
  # x is zero except for the parameters and delta if coasting
  x = zero(v) # default identity in parameters
  x.Q.q[1] = 1 # identity in spin
  compose!(v,v,x)

  # 5: add back in identity
  for i=1:nv
    @inbounds v.x[i] += mono(i,use=desc)
  end

  return v
end

This is pretty fast (could be made faster by proper use of work temporaries accessible by user of inv and compose) , however this is not the easiest to read. gofix could instead be written as

function gofix(xy::DAMap, order=1)
  m = cut(xy, order+1)
  return (m-I)^-1∘zero(m)+I
end

This is much easier to understand and read, however much slower. For the best of both worlds, a macro must be implemented to convert the above easy-to-read function to its faster less-easy-to-read function. The approach could be the same as that taken for @FastGTPSA in GTPSA.jl

[mattsignorelli]

The way that I have restructured the code significantly from this comment, it seems that at least in the nonlinear normalizing map loop, there is basically negligible performance gain by using preallocated maps as opposed to using out of place operators. Therefore, for code readability, this issue is a much lower priority now

[mattsignorelli]

no longer will be considered due to performance improvements already made