hughperkins / DeepCL

OpenCL library to train deep convolutional neural networks
Mozilla Public License 2.0
867 stars 199 forks source link

Create opencl kernels for large image sizes, using local memory #18

Closed hughperkins closed 9 years ago

hughperkins commented 9 years ago

Create opencl kernels for large image sizes, using local memory

Currently, for large images, the only working kernel is propagate1, which is generic, but doesnt use local memory. If we make a dedicated kernel, that uses local memory, eg by blocking the input images, large images should run faster (eg 128x128, this kind of size)

hughperkins commented 9 years ago

I think my brief experiment with creating issues for people to help with didnt work very well, since I got zero contributors to any of them :-P So closing this for now.