I need to dispatch a kernel with
AT_DISPATCH_FLOATING_TYPES then call
.exp_() on the result and then call another kernel.
- Can/should I call the
.exp_call from withing the lambda passed to
AT_DISPATCH_FLOATING_TYPES(right after kernel dispatch) and then call the second kernel
<<<>>>or is it better to call the dispatch twice, once for each kernel? IOW, What is the PyTorch-ian way?
- Also the second argument to
NAME, I intend to re-use it for both kernel calls, but was not sure if in the future you use that name to track kernels.