Is autotuning is an offline process?

[yuleigeng]

Hi, In the paper "Performance, Design, and Autotuning of Batched GEMM for GPUs" , several process such as Search Space Generation and Pruning, Autotuning output anaylsis are mention, I wonder if there process are all offline, and aim to find best kernel for given cases, and get the template selection criterion for the runtime api call? Thank a lot.