0

I am working with pymc3/theano in a pristine conda environment that I created today. I'm running the latest 'bleeding edge' install of theano & pymc3 along with Python2.7.13.

I have successfully gotten theano to compile using the MKL version of BLAS, but I am running into an Exception error that I haven't seen. I suspect it might have something to do with the gcc compiler, but I'm not sure how to fix this. My go-to software engineer is currently enjoying his High Holy Days (March Madness), so I am here pleading with the Multiverse for help.

Here's the lowest Traceback error (eg, the Exception error):

~/source/Theano/Theano/theano/gof/cmodule.pyc in compile_str(module_name, src_code, location, include_dirs, lib_dirs, libs, preargs, py_module, hide_symbols)

2336 # difficult to read.

2337 raise Exception('Compilation failed (return status=%s): %s' %

-> 2338 (status, compile_stderr.replace('\n', '. ')))

2339 elif config.cmodule.compilation_warning and compile_stderr:

2340 # Print errors just below the command line.

Exception: ('The following error happened while compiling the node', CGemv{inplace}(AllocEmpty{dtype='float64'}.0, TensorConstant{1.0}, Elemwise{mul,no_inplace}.0, p, TensorConstant{0.0}), '\n', 'Compilation failed (return status=1): g++: error: unrecognized command line option \xe2\x80\x98-Wl\xe2\x80\x99. g++: error: unrecognized command line option \xe2\x80\x98--no-as-needed\xe2\x80\x99. ', '[CGemv{inplace}(, TensorConstant{1.0}, , p, TensorConstant{0.0})]')

Any suggestions would be most welcome!

Deno Stelter
  • 36
  • 5
  • '\xe2\x80\x98' and '\xe2\x80\x99' are fancy quotation marks. So maybe somewhere there are special quotation marks instead of the normal ones. Do you have something in the ~/.theanorc? – aseyboldt Mar 16 '17 at 19:51
  • Thanks @aseyboldt! I do have a .theanorc. Here's the file contents: >[blas] >ldflags = -L/usr/lib -lopenblas >ldflags = -L${MKLROOT}/ -Wl,-lmkl_intel_ilp64 -lmkl_intel_thread ->lmkl_core -liomp5 -lpthread -lm -ldl' > >[nvcc] >fastmath = True – Deno Stelter Mar 17 '17 at 14:08
  • Urg, still learning mini-Markdown.... I spent quite some time working on this problem yesterday afternoon/night and was able to get around the problem by making a new **conda** environment with an isolated python2.7 version with the packages I needed. I ended up having to disable MKL and go with OpenBLAS, which makes me unhappy...but the code runs. It's slower than (unfermented) molasses in January. – Deno Stelter Mar 17 '17 at 14:15

0 Answers0