1 paper
Yehong Jiang, Sheng Chen, Fangwen Fu +4
New AI accelerators arrive before the kernels that make them fast, because peak kernel performance requires architecture-specific expertise in operand pipelines and data-movement t…