changed check_stack_trace to first check for ops or tuple of ops before __call__ attribute, because classes have a call attribute (constructor).

added length check in check_stack_trace. This change causes 31 failures in tensor/tests/test_opt.py, most seem to be due to MergeOptimizer.

Modified test_opt.py to use check_stack_trace and modified check_stack_trace to loop over outputs for ops_to_check=last and some refactoring

test using theano tile added

Convert gpucumsum

fix issue 4303 : reshaping part

Disable python_constant_folding for GpuElemwise.

Hostname printout in do_nightly_build

Opt CpuContiguous -> GpuContiguous and add CpuContiguous.ard

Blocksparse for gpuarray

Hide the useless default nvcc output on Windows

flake8 tests pass; minor rephrasing of documentation

Update check trace to allow more argument types