Use BLAS copy for small arrays and memcpy otherwise.

Threshold needs to be tuned better.
Close #121.

j/linalg_blas.j

@@ -19,6 +19,16 @@ end
 @jl_blas_copy_macro :zcopy_ Complex128
 @jl_blas_copy_macro :ccopy_ Complex64
 
+function copy_to{T<:Union(Float64,Float32,Complex128,Complex64)}(dest::Array{T}, src::Array{T})
+    n = numel(src)
+    if n < 200
+        jl_blas_copy(n, pointer(src), 1, pointer(dest), 1)
+    else
+        copy_to(pointer(dest), pointer(src), ulong(numel(src)*sizeof(T)))
+    end
+    return dest
+end
+
 # DOUBLE PRECISION FUNCTION DDOT(N,DX,INCX,DY,INCY)
 macro jl_blas_dot_macro(fname, eltype)
     quote

test/unittests.j

@@ -188,6 +188,11 @@ a = rand(n,n)
 (l,u,p) = lu(a)
 @assert sum(l[p,:]*u - a) < 1e-8
 
+# arpack
+asym = a+a'
+(d,v) = eigs(asym, 3)
+@assert sum(asym*v[:,1]-d[1]*v[:,1]) < 1e-8
+
 # hash table
 h = HashTable()
 for i=1:100