tensorflow
TableGen attribute name get methods on ops?
In the custom print and parse methods of ops that have a tablegen description, in several places, the names of the attributes are either hardcoded or 'get' methods for the attributes are explicitly added to the tablegen description (for eg. getPredicateAttrName, getStepAttrName, getLowerBoundAttrName). Although the 'get' methods for the attributes themselves are automatically generated, auto generating 'get' methods for the attribute names will further reduce the amount of boilerplate code and prevent hardcoding it like here: https://github.com/tensorflow/mlir/blob/master/lib/Dialect/StandardOps/Ops.cpp#L1324
It happens I've already prepared CLs for this. :) Going through reviews now.
It happens I've already prepared CLs for this. :) Going through reviews now.
Thanks very much! :) I'll wait for this before I submit my pending stuff. Please feel free to close this once that's committed.
It happens I've already prepared CLs for this. :) Going through reviews now.
@antiagainst Has this been committed? I didn't notice an update to the op def doc. Thanks.
Oh, sorry, not yet. The CL depends on another CL that handles naming style of the generated getter methods. There are discussions on that. Let me see if I can rebase to unblock.
Oh, sorry, not yet. The CL depends on another CL that handles naming style of the generated getter methods. There are discussions on that. Let me see if I can rebase to unblock.
@antiagainst Anything further on this? Thanks!
Hey @bondhugula, sorry wasn't working on this because of other priorities. Will get back to this soon.
Hey @bondhugula, sorry wasn't working on this because of other priorities. Will get back to this soon.
Okay, I couldn't wait for this :) I've submitted #225 When yours is committed, we can drop all the getAttr..Name() methods that were added to the td files.