* Hybrid direct + implicit GEMM forward convolution NCHWc v5r1. Input tensor bypass LDS. Support fp32/fp16/int8
Attach a file by drag & drop or click to upload