Use approx for some GPU permutes

Author: kshyatt

test/cuda/tensors.jl

@@ -304,14 +304,12 @@ for V in spacelist
                     end
                 end
 
-                CUDA.@allowscalar begin
-                    t3 = @constinferred repartition(t, $k)
-                    t3 = repartition(t, k)
-                    @test norm(t3) ≈ norm(t)
-                    t3′ = @constinferred repartition!(similar(t3), t′)
-                    @test norm(t3′) ≈ norm(t′)
-                    @test dot(t′, t) ≈ dot(t3′, t3)
-                end
+                t3 = @constinferred repartition(t, $k)
+                t3 = repartition(t, k)
+                @test norm(t3) ≈ norm(t)
+                t3′ = @constinferred repartition!(similar(t3), t′)
+                @test norm(t3′) ≈ norm(t′)
+                @test dot(t′, t) ≈ dot(t3′, t3)
             end
         end
         if BraidingStyle(I) isa SymmetricBraiding
@@ -322,9 +320,10 @@ for V in spacelist
                     for p in permutations(1:5)
                         p1 = ntuple(n -> p[n], k)
                         p2 = ntuple(n -> p[k + n], 5 - k)
-                        dt2 = CUDA.@allowscalar permute(t, (p1, p2))
                         ht2 = permute(TensorKit.to_cpu(t), (p1, p2))
-                        @test ht2 == TensorKit.to_cpu(dt2)
+                        dt2 = permute(t, (p1, p2))
+                        ht2′ = TensorKit.to_cpu(dt2)
+                        @test ht2 ≈ ht2′
                     end
 
                     dt3 = CUDA.@allowscalar repartition(t, k)
@@ -380,7 +379,7 @@ for V in spacelist
             end
             @test ta ≈ tb
         end
-        #=if BraidingStyle(I) isa Bosonic && hasfusiontensor(I)
+        if BraidingStyle(I) isa Bosonic && hasfusiontensor(I)
             @timedtestset "Tensor contraction: test via CPU" begin
                 dA1 = CUDA.randn(ComplexF64, V1' * V2', V3')
                 dA2 = CUDA.randn(ComplexF64, V3 * V4, V5)
@@ -395,7 +394,7 @@ for V in spacelist
                     TensorKit.to_cpu(dH)[s1, s2, t1, t2]
                 @test TensorKit.to_cpu(dHrA12) ≈ hHrA12
             end
-        end=# # doesn't yet work because of AdjointTensor
+        end
         @timedtestset "Index flipping: test flipping inverse" begin
             t = CUDA.rand(ComplexF64, V1 ⊗ V1' ← V1' ⊗ V1)
             for i in 1:4