cuBLASDx and cuSOLVERDx for dynamic parallelism

Dear all,

I developed an accelerated ODE solver that exploited the device version of cuBLASD through dynamic parallelism. Unfortunately, such a feature has been removed from CUDA 10 (honestly, I do not understand why!). In May 2021, I asked to reintroduce such an important functionality to the new versions of CUDA and I got this reply “We are planning to release new device side libraries (i.e., cuBLASDx and cuSOLVERDx) through our Math Libraries Early Access Program”. I cannot find any cuBLASDx version in the Math Libraries Early Access Program yet, does anyone have any news?

Thank you in advance.


These libraries have not been released.