microsoft / tensorflow-directml-plugin

DirectML PluggableDevice plugin for TensorFlow 2
Apache License 2.0
179 stars 23 forks source link

Add missing kernel profiling #302

Closed PatriceVignola closed 1 year ago

PatriceVignola commented 1 year ago

We were previously only profiling compute for kernels wrapped by the DmlKernelWrapperBase class, but not all kernels do. Since some of the kernels inherit directly from OpKernel, the compute logging was moved up into OpKernel::Compute as well.