Import branch Fsword73 kernel1

hughperkins / DeepCL

OpenCL library to train deep convolutional neural networks

Mozilla Public License 2.0

865 stars 199 forks source link

Import branch Fsword73 kernel1 #131

Closed merceyz closed 6 years ago

merceyz commented 6 years ago

This PR imports the changes done by @fsword73 from the branch fsword73-kernel1 into master

Quick summary of the changes in this PR:

BackpropWeightsAuto, BackwardAuto, and ForwardAuto now uses microseconds
Fixed memory leak in BackpropWeightsAuto, BackwardAuto, and ForwardAuto
Added new kernels BackpropWeightsFsword73 and BackpropWeightsFsword73_BatchSize

Discussion and more details can be found in #66

hughperkins commented 6 years ago

Awesome! :) . Experimentally, I've granted you 'push' access to the DeepCL repo. So, once you are happy that everything looks good, please go ahead and merge the PR.

merceyz commented 6 years ago

Thank you :) I would like to rewrite the *Auto classes, would that be ok? Should probably be in a different PR though

hughperkins commented 6 years ago

I would like to rewrite the *Auto classes, would that be ok?

Sounds good to me :) . Just out of personal curiosity, what changes are you thinking of doing?

merceyz commented 6 years ago

The code is a little bit messy, can be done without the arrays, also it keeps kernels that isn't used in memory. So I thought I would change that

hughperkins commented 6 years ago

Cool, sounds good :)