dynamic parallelism with cublas in cuda 10.1

I have some dynamic parallelism based code which has a kernel from which I call cublas functions to perform matrix operations. I recently upgraded my device to cuda 10.1 and the latest nvidia drivers. After reading the documentation, I realised that the cudadevrt option has been removed in 10.1

Are there any alteratives so that cublas can be called from inside kernels?

There aren’t any.

so does this mean dynamic parallelism is no longer available in cuda 10.1?

no, that isn’t what it means.

Dynamic parallelism is still available. (All you have to do is try it, or survey any of the dynamic parallelism sample codes that are included in the cuda 10.1 toolkit.)

What is not available is a version of the cublas library that is callable from device code.

Thank you. Is “a version of the cublas library that is callable from device code” expected to be released in any later updates, or does NVIDIA recommend any other alternative to CUBLAS?

I generally don’t make forward looking statements on these forums, and its not really the role of these forums to talk about future events.

I’m not aware of any formal recommendations by NVIDIA for alternatives to CUBLAS.