Nevermind, I figured it out. I had to disable constant folding and wrap with tensor.elemwise.
On Sunday, March 5, 2017 at 10:28:39 AM UTC+8, Adam Becker wrote: > > Hi, > > I'm writing a elemwise Op with special purpose, it's c_code should be > different when it's working with tensor objects. (Actually, it's for > gh:#5471 <https://github.com/Theano/Theano/issues/5471>) > > Current I'm only working with the CPU version. > > Here's the approach I'm taking so far, I modified this line in > tensor/elemwise.py > <https://github.com/Theano/Theano/blob/master/theano/tensor/elemwise.py#L946> > into: > > defines = "#define TENSOR_ELEMWISE" > undefs = "#undef TENSOR_ELEMWISE" > > The elemwise Op looks like: > > class MyOp(scalar.ScalarOp): > nin = 0 > # < ... > > def c_code(self, node, name, inputs, outputs, sub): > out, = outputs > op_name = type(self).__name__ > code = ''' > #ifndef TENSOR_ELEMWISE > %(out)s = 1227; > #else > %(out)s = 1337; > #endif > ''' % locals() > return code > > However, even though I build a graph with matrix input, the elemwise Op > still takes the "1227" branch. > Further investigation of generated code shows MyOp is compiled into a > separate mod.cpp than the main elemwise loop. > > I suppose this could be due to a few optimizations. However, as I'm still > unfamiliar with opt code base, I couldn't figure out where. > > Any clue how to force the c_code of MyOp into the elemwise loop, so that > the "1337" branch will be taken when dealing with tensor? > > Thanks. > > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to theano-users+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
