Building NMS lib and ROIPooling operation with CUDA 10.2

[zedavid]

I'm trying to build the ROIPolling operation with TF 1.15.0, CUDA 10.2 and g++-7.5.0, but I'm getting the following errors:

`/lib/python3.6/site-packages/tensorflow_core/include/absl/strings/string_view.h(495): error: constexpr function return is non-constant

tensorflow/core/util/cuda_kernel_helper.h(52): error: class "Eigen::GpuDevice" has no member "getNumCudaMultiProcessors"

tensorflow/core/util/cuda_kernel_helper.h(52): error: class "Eigen::GpuDevice" has no member "maxCudaThreadsPerMultiProcessor"

tensorflow/core/util/cuda_kernel_helper.h(51): error: no instance of overloaded function "std::min" matches the argument list
argument types are: (, const int)

tensorflow/core/util/cuda_kernel_helper.h(54): error: class "Eigen::GpuDevice" has no member "maxCudaThreadsPerBlock"

tensorflow/core/util/cuda_kernel_helper.h(54): error: no instance of overloaded function "std::min" matches the argument list
argument types are: (int, )

tensorflow/core/util/cuda_kernel_helper.h(57): error: class "Eigen::GpuDevice" has no member "getNumCudaMultiProcessors"`

I was wondering if these sound familiar to anyone. Thanks!