Compiles all mathematical functions in
comp to TensorFlow blocks.
Notice that this does not necessarily represent a strict performance
improvement. In particular, this compilation will not attempt to deduplicate
across the boundaries of communication operators, and therefore it may be
the case that compiling eagerly to TensorFlow hides the opportunity for
a dynamic cache to be used.
computation_base.Computation to compile.
computation_base.Computation representing the compiled version of