5 comments

[ 4.3 ms ] story [ 13.5 ms ] thread
interesting work.
This is super cool. Any comparison with Weld? https://weld-project.github.io/
as far as i know weld is optimized for data analytics workload so far(dataframe) while tvm is optimized for tensor deep learning workloads with great cpu, gpu and other supports
(comment deleted)
A collection about depthwise conv:

Optimize Deep Learning GPU Operators with TVM: A Depthwise Convolution Example http://tvmlang.org/2017/08/22/Optimize-Deep-Learning-GPU-Ope...

MobileNet and Depthwise Separable Convolution · Issue #70 · changtimwu/changtimwu.github.com https://github.com/changtimwu/changtimwu.github.com/issues/7...

yonghenglh6/DepthwiseConvolution: A personal mobile convolution layer implementation on caffe by liuhao.(only GPU) https://github.com/yonghenglh6/DepthwiseConvolution