diff options
author | 2016-10-06 11:10:14 -0700 | |
---|---|---|
committer | 2016-10-06 11:10:14 -0700 | |
commit | bfc264abe86541632d17b146a8601a6999a0f8d6 (patch) | |
tree | 1178262d1fa577b5f350408caa862478ac3d68ae /unsupported | |
parent | e2e9cdd16970914cf0a892fea5e7c4402b3ede41 (diff) |
Add a test that GPU complex product reductions match CPU reductions.
Diffstat (limited to 'unsupported')
-rw-r--r-- | unsupported/test/cxx11_tensor_complex_cuda.cu | 38 |
1 files changed, 38 insertions, 0 deletions
diff --git a/unsupported/test/cxx11_tensor_complex_cuda.cu b/unsupported/test/cxx11_tensor_complex_cuda.cu index f895efd01..d4e111f5d 100644 --- a/unsupported/test/cxx11_tensor_complex_cuda.cu +++ b/unsupported/test/cxx11_tensor_complex_cuda.cu @@ -108,8 +108,46 @@ static void test_cuda_sum_reductions() { } +static void test_cuda_product_reductions() { + + Eigen::CudaStreamDevice stream; + Eigen::GpuDevice gpu_device(&stream); + + const int num_rows = internal::random<int>(1024, 5*1024); + const int num_cols = internal::random<int>(1024, 5*1024); + + Tensor<std::complex<float>, 2> in(num_rows, num_cols); + in.setRandom(); + + Tensor<std::complex<float>, 0> full_redux; + full_redux = in.prod(); + + std::size_t in_bytes = in.size() * sizeof(std::complex<float>); + std::size_t out_bytes = full_redux.size() * sizeof(std::complex<float>); + std::complex<float>* gpu_in_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(in_bytes)); + std::complex<float>* gpu_out_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(out_bytes)); + gpu_device.memcpyHostToDevice(gpu_in_ptr, in.data(), in_bytes); + + TensorMap<Tensor<std::complex<float>, 2> > in_gpu(gpu_in_ptr, num_rows, num_cols); + TensorMap<Tensor<std::complex<float>, 0> > out_gpu(gpu_out_ptr); + + out_gpu.device(gpu_device) = in_gpu.prod(); + + Tensor<std::complex<float>, 0> full_redux_gpu; + gpu_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_ptr, out_bytes); + gpu_device.synchronize(); + + // Check that the CPU and GPU reductions return the same result. + VERIFY_IS_APPROX(full_redux(), full_redux_gpu()); + + gpu_device.deallocate(gpu_in_ptr); + gpu_device.deallocate(gpu_out_ptr); +} + + void test_cxx11_tensor_complex() { CALL_SUBTEST(test_cuda_nullary()); CALL_SUBTEST(test_cuda_sum_reductions()); + CALL_SUBTEST(test_cuda_product_reductions()); } |