Speeding up CPU and OpenCL convolution.
Optimizing OpenCL and interleaved execution.
Speeding up fully connected layers.
Minor full connect optimizations.
More convolution minor optimizations.
More minor convolution optimizations.
Minor optimizations and fixes.
A number of small changes: OpenCL, ResNet and Keras inspired experiment.
Adding TNNetReLU.
Adding csBackpropRnd option.