Made sample type agnostic and CUDA compatible

src/mps.jl

@@ -655,6 +655,8 @@ function sample(rng::AbstractRNG, m::MPS)
     error("sample: MPS is not normalized, norm=$(norm(m[1]))")
   end
 
+  ElT = promote_itensor_eltype(m)
+
   result = zeros(Int, N)
   A = m[1]
 
@@ -664,16 +666,16 @@ function sample(rng::AbstractRNG, m::MPS)
     # Compute the probability of each state
     # one-by-one and stop when the random
     # number r is below the total prob so far
-    pdisc = 0.0
+    pdisc = zero(real(ElT))
     r = rand(rng)
     # Will need n,An, and pn below
     n = 1
     An = ITensor()
-    pn = 0.0
+    pn = zero(real(ElT))
     while n <= d
       projn = ITensor(s)
-      projn[s => n] = 1.0
-      An = A * dag(projn)
+      projn[s => n] = one(ElT)
+      An = A * dag(adapt(datatype(A), projn))
       pn = real(scalar(dag(An) * An))
       pdisc += pn
       (r < pdisc) && break
@@ -682,7 +684,7 @@ function sample(rng::AbstractRNG, m::MPS)
     result[j] = n
     if j < N
       A = m[j + 1] * An
-      A *= (1.0 / sqrt(pn))
+      A *= (one(ElT) / sqrt(pn))
     end
   end
   return result