we propose an automated tool flow from a domain-specific language for tensor expressions to generate massively parallel accelerators on high-bandwidth-memory-equipped FPGAs. Designers can use this ...