search

soumith / cuda-convnet2.torch

Torch7 bindings for cuda-convnet2 kernels!
Apache License 2.0
40 stars 16 forks source link

add partialSum for SpatialConvolution #26

Closed soumith closed 9 years ago

soumith commented 9 years ago

just noticed that there's this partialSum thing in cuda-convnet2 that is (a) undocumented, (b) is much faster to do accGradParameters. I'm just noticing it now, fml! Have to add it.

soumith commented 9 years ago

added partialSum. with this mode (for example setting partialSum = 4), accGradParameters of SpatialConvolution is WAAAY faster! Earlier, I was just using the default partialSum suggestion that alex wrote in the docs (oH * oH). https://github.com/soumith/cuda-convnet2.torch/commit/f1fcecede9c0573aa04cfdf1d1b4b825dccdc789