Tensor Core
作者：互联网
参考：
https://forums.developer.nvidia.com/t/how-to-use-wmma-efficiently/157619/2
https://github.com/BigNerd95/CUDASamples/tree/master/samples/0_Simple/cudaTensorCoreGemm
（配置WARPS_PER_BLOCK为4，即可达到接近100TFLOS，一般100TFLOPS性能已经比较好。80%peak ）
标签：Core,Tensor,WARPS,forums,PER,https,com 来源： https://www.cnblogs.com/simpleminds/p/16313068.html