High Performance Zero-Memory Overhead Direct Convolutions

HighPerformanceZero-MemoryOverheadDirectConvolutionsJiyuanZhang1FranzFranchetti1TzeMengLow1AbstractPerformancenormalizedtoOpenBLASGEMMonAMDPileDriverThecomputationofconvolutionlayersindeep4.0GHz,4/...