Force memory transfer

Author: kshyatt

src/implementations/svd.jl

@@ -229,13 +229,13 @@ end
 function svd_trunc!(A, USVᴴ::Tuple{TU, TS, TVᴴ}, alg::TruncatedAlgorithm; compute_error::Bool = true) where {TU, TS, TVᴴ}
     ϵ = similar(USVᴴ[2], compute_error)
     (U, S, Vᴴ, ϵ) = svd_trunc!(A, (USVᴴ..., ϵ), alg)
-    return compute_error ? (U, S, Vᴴ, ϵ[1]) : (U, S, Vᴴ, -one(eltype(ϵ)))
+    return compute_error ? (U, S, Vᴴ, collect(ϵ)[1]) : (U, S, Vᴴ, -one(eltype(ϵ)))
 end
 function svd_trunc!(A, USVᴴ::Nothing, alg::TruncatedAlgorithm; compute_error::Bool = true)
     Tr = real(eltype(A))
-    ϵ = compute_error ? zeros(Tr, 1) : zeros(Tr, 0)
+    ϵ = zeros(Tr, compute_error)
     U, S, Vᴴ, ϵ = svd_trunc!(A, (USVᴴ, ϵ), alg)
-    return compute_error ? (U, S, Vᴴ, ϵ[1]) : (U, S, Vᴴ, -one(Tr))
+    return compute_error ? (U, S, Vᴴ, collect(ϵ)[1]) : (U, S, Vᴴ, -one(Tr))
 end
 
 # Diagonal logic

test/cuda/svd.jl

@@ -140,7 +140,7 @@ end
                 S₀ = svd_vals(hA)
                 r = k
 
-                U1, S1, V1ᴴ, ϵ1 = @constinferred svd_trunc(A; alg, trunc = truncrank(r))
+                U1, S1, V1ᴴ, ϵ1 = @constinferred svd_trunc(A; alg, trunc = truncrank(r), compute_error=false)
                 @test length(S1.diag) == r
                 @test opnorm(A - U1 * S1 * V1ᴴ) ≈ S₀[r + 1]
                 @test norm(A - U1 * S1 * V1ᴴ) ≈ ϵ1
@@ -149,7 +149,7 @@ end
                     s = 1 + sqrt(eps(real(T)))
                     trunc2 = trunctol(; atol = s * S₀[r + 1])
 
-                    U2, S2, V2ᴴ, ϵ2 = @constinferred svd_trunc(A; alg, trunc = trunctol(; atol = s * S₀[r + 1]))
+                    U2, S2, V2ᴴ, ϵ2 = @constinferred svd_trunc(A; alg, trunc = trunctol(; atol = s * S₀[r + 1]), compute_error=false)
                     @test length(S2.diag) == r
                     @test U1 ≈ U2
                     @test parent(S1) ≈ parent(S2)