Right now, we only track ret values as either success or failure. A future feature could be to have histograms for different return types defined by cudaError_t for observability on why a call failed, and the frequencies of such errors.
I suggest implementing an enum in the Rust userspace code that maps one-to-one with the CUDA header definition.
Right now, we only track ret values as either success or failure. A future feature could be to have histograms for different return types defined by
cudaError_tfor observability on why a call failed, and the frequencies of such errors.I suggest implementing an enum in the Rust userspace code that maps one-to-one with the CUDA header definition.