RISC-V Summit

Apache TVM is an open source machine learning compiler framework which supports several CPUs and GPUs. To optimize diversified machine learning applications on various hardware platforms, TVM recently introduced AutoScheduler that explores many optimization strategies and then identify the best one using evolutionary algorithm. In this talk, the compilation flow of TVM and AutoScheduler is introduced first. Then, several enhancements used to enable AutoScheduler on a platform with Andes vector processor NX27V with RISC-V RVV Extension are also explained. To demonstrate the power of AutoScheduler on Andes NX27V platform, we will go through some experiments with AutoScheduler and compare their results with those from hand-optimized code.