MasterJH5574 opened a new pull request, #13663: URL: https://github.com/apache/tvm/pull/13663
This PR brings the support for specifying axis dtype when creating `tir::Layout`. Prior to this PR, the construction of `tir::Layout` uses int32 for every axis. With this specification, the output of `BijectiveLayout::ForwardShape/BackwardShape(...)` will always be int32 shape, no matter if the input shape is in int32 or int64. As we will be gradually moving to int64 default dtype for tensor shapes, there are increasingly more cases where we need to handle int64 shapes. Thus, `tir::Layout` need some updates for int64 shapes. Since the creation of an `tir::Layout` instance is agnostic to the dtype of the shapes that the instance will handle later, one way of the such updates is to add a dtype parameter to the constructor of `tir::Layout`, to specify the dtype explicitly. This PR enhances `tir::Layout` in this way, with appropriate unit tests for dtype check. cc @tqchen -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
