Alter test case to avoid use of advanced indexing

Add test case for the nonseq grad fix

[CRASH] Repair crash with cudnn r-1 following the removing of struct_id

Revert "Get rid of typeConstructor in Scan."

This reverts commit 7d70c3a4.

Output StorageMap content in raise_with_op

Use numpy's function to check array is contiguous

[ENH] Matrix exponential op and gradient (round 2!)

In particular, strides for dimensions of length 1 could be anything

Separate allow_gc flag for scan

[BUG] Fix gradient of advanced_set_subtensor1

Use gcc.cxxflags=-pipe for travis tests

I hope I got all the spots. This is insane.

Add print during unit test to help preventing timeouts

Multi fixes (pieces of the multi-gpu code)

Fix buildbot error from #2329

I'm not sure what this test is expecting.  It seems to test that the gpu will magically work and optimize the graph even though we disable InputsToGpuOptimizer (which is obviously not the case).  I wonder why it passed before.  I wonder if this is the right fix.  I also wonder what is it testing exactly.

In any case it seems to pass now.

It is suitably unique and, more importantly, shared with the rest of the methods.

Tests for fix and do the fix for AdvancedSubtesor.

Scan push out sum of dot