Jonas S.

icon location Germany icon location Interested in Computer Science and Mathematics. At the moment I'm focusing on AI/DeepLearning.

Results 1 issues of Jonas S.

Possible way to implement a LoopVectorization extension for conv2d & meanpool2d & activations

21 comment

Some tests have shown that standard for-loop implementations for conv using LoopVectorization (LV) can be faster than the standard NNlib's im2col + gemm approach. There has already been some talks...

benchmark