Fix non-cudnn reductions that were all subtly broken by 91bc16c3.

support allow_remove_{inf,nan}

Add a map of values_approx functions from cpu to gpu.

lazylinker_c: Make sure that the version number is flushed to disk

Corrected spelling mistake

Add a script to help test GPU power consumption.

Switched lambda to partial in EighGrad

Remove a template argument from batch_gemm

This might in some case where the compiler got confused and could not
infer the right type for the gemm function.

Fix versions in install doc

See gh-6543. Sometimes, np.asarray(self.data) fail with a read error …

Remove awkward language

See gh-6543. Sometimes, np.asarray(self.data) fail with a read error from the hardware. So here try some fall back that won't hide the current error being raised.

Fix function parameters order in test_bn/test_batch_normalization

It was just bugging me a little

Fix tests in DebugMode and make them less verboase

Add cuda include path for magma ops