diff --git a/unit_tests/Test_TensorImpl.cpp b/unit_tests/Test_TensorImpl.cpp index cb120a970c5310f80f8c62960c029a845937ba30..c24b5b457cce602d465a1ecddefc7b7f35964794 100644 --- a/unit_tests/Test_TensorImpl.cpp +++ b/unit_tests/Test_TensorImpl.cpp @@ -37,7 +37,7 @@ TEST_CASE("CUDA test") { } // Allocate device memory - float *d_a, *d_b, *d_out; + float *d_a, *d_b, *d_out; cudaMalloc(reinterpret_cast<void**>(&d_a), sizeof(float) * N); cudaMalloc(reinterpret_cast<void**>(&d_b), sizeof(float) * N); cudaMalloc(reinterpret_cast<void**>(&d_out), sizeof(float) * N); @@ -46,9 +46,9 @@ TEST_CASE("CUDA test") { cudaMemcpy(d_a, a, sizeof(float) * N, cudaMemcpyHostToDevice); cudaMemcpy(d_b, b, sizeof(float) * N, cudaMemcpyHostToDevice); - // Executing kernel + // Executing kernel vector_add(d_out, d_a, d_b, N); - + // Transfer data back to host memory cudaMemcpy(out, d_out, sizeof(float) * N, cudaMemcpyDeviceToHost); @@ -71,7 +71,6 @@ TEST_CASE("CUDA test") { TEST_CASE("Tensor creation", "[Connector]") { SECTION("from const array") { Tensor x; - x.setBackend("cuda"); x = Array3D<int,2,2,2>{ { { @@ -83,6 +82,7 @@ TEST_CASE("Tensor creation", "[Connector]") { {7, 8} } }}; + x.setBackend("cuda"); REQUIRE(x.nbDims() == 3); REQUIRE(x.dims()[0] == 2);