Does TVM support tensorization for OpenCL backend?
It will be grateful if there is an example.
I think it does. The tutorials is a good example.
just change it a little by combine the GPU example
I guess that it may require additional OpenCL infrastructures which can import customized OpenCL Kernels, in which way, the kernels can be built into one packed library. I think we don’t want to make the kernel intrinsic as part of TVM runtime when deploying.
1 Like