4d4ca3ef3f
For HCC path concurrentKernels is set to true since all ROCR hardware supports this feature. For NVCC path concurrentKernels is obtained from CUDA's device property cudaDeviceProp::concurrentKernels.