fixup

src/cuda/curnn.jl

@@ -231,6 +231,9 @@ function LinearAlgebra.copy_transpose!(dst::CuArray, src::CuArray)
     dst[I...] = src[reverse(I)...]
     return
   end
+  blk, thr = cudims(dst)
+  @cuda blocks=blk threads=thr kernel(dst, src)
+  return dst
 end
 
 CuParam{T,N} = Union{CuArray{T,N},TrackedArray{T,N,CuArray{T,N}}}

test/cuda/cuda.jl

@@ -37,8 +37,7 @@ Flux.back!(sum(l))
 end
 
 if CuArrays.libcudnn != nothing
-    @info "Testing Flux/CUDNN BatchNorm"
+    @info "Testing Flux/CUDNN"
     include("cudnn.jl")
-    @info "Testing Flux/CUDNN RNN"
     include("curnn.jl")
 end