diff options
| -rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorBase.h | 17 | ||||
| -rw-r--r-- | unsupported/test/cxx11_tensor_cuda.cu | 38 |
2 files changed, 55 insertions, 0 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h b/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h index 0729455fb..69d1802d5 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h @@ -400,6 +400,23 @@ class TensorBase<Derived, ReadOnlyAccessors> return operator!=(constant(threshold)); } + // Checks + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE const TensorCwiseUnaryOp<internal::scalar_isnan_op<Scalar>, const Derived> + (isnan)() const { + return unaryExpr(internal::scalar_isnan_op<Scalar>()); + } + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE const TensorCwiseUnaryOp<internal::scalar_isinf_op<Scalar>, const Derived> + (isinf)() const { + return unaryExpr(internal::scalar_isinf_op<Scalar>()); + } + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE const TensorCwiseUnaryOp<internal::scalar_isfinite_op<Scalar>, const Derived> + (isfinite)() const { + return unaryExpr(internal::scalar_isfinite_op<Scalar>()); + } + // Coefficient-wise ternary operators. template<typename ThenDerived, typename ElseDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const TensorSelectOp<const Derived, const ThenDerived, const ElseDerived> diff --git a/unsupported/test/cxx11_tensor_cuda.cu b/unsupported/test/cxx11_tensor_cuda.cu index 33796690d..5f548ff0c 100644 --- a/unsupported/test/cxx11_tensor_cuda.cu +++ b/unsupported/test/cxx11_tensor_cuda.cu @@ -122,6 +122,43 @@ void test_cuda_elementwise() cudaFree(d_out); } +void test_cuda_props() { + Tensor<float, 1> in1(200); + Tensor<bool, 1> out(200); + in1.setRandom(); + + std::size_t in1_bytes = in1.size() * sizeof(float); + std::size_t out_bytes = out.size() * sizeof(bool); + + float* d_in1; + bool* d_out; + cudaMalloc((void**)(&d_in1), in1_bytes); + cudaMalloc((void**)(&d_out), out_bytes); + + cudaMemcpy(d_in1, in1.data(), in1_bytes, cudaMemcpyHostToDevice); + + Eigen::CudaStreamDevice stream; + Eigen::GpuDevice gpu_device(&stream); + + Eigen::TensorMap<Eigen::Tensor<float, 1>, Eigen::Aligned> gpu_in1( + d_in1, 200); + Eigen::TensorMap<Eigen::Tensor<bool, 1>, Eigen::Aligned> gpu_out( + d_out, 200); + + gpu_out.device(gpu_device) = (gpu_in1.isnan)(); + + assert(cudaMemcpyAsync(out.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, + gpu_device.stream()) == cudaSuccess); + assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess); + + for (int i = 0; i < 200; ++i) { + VERIFY_IS_EQUAL(out(i), (std::isinf)(in1(i))); + } + + cudaFree(d_in1); + cudaFree(d_out); +} + void test_cuda_reduction() { Tensor<float, 4> in1(72,53,97,113); @@ -964,6 +1001,7 @@ void test_cxx11_tensor_cuda() { CALL_SUBTEST_1(test_cuda_elementwise_small()); CALL_SUBTEST_1(test_cuda_elementwise()); + CALL_SUBTEST_1(test_cuda_props()); CALL_SUBTEST_1(test_cuda_reduction()); CALL_SUBTEST_2(test_cuda_contraction<ColMajor>()); CALL_SUBTEST_2(test_cuda_contraction<RowMajor>()); |