commit | 10dfd9d337aef388cdaa725514acdcf0b7f4a3ee | [log] [tgz] |
---|---|---|
author | Quinn Dawkins <quinn.dawkins@gmail.com> | Fri Jul 12 16:31:15 2024 -0400 |
committer | GitHub <noreply@github.com> | Fri Jul 12 20:31:15 2024 +0000 |
tree | 0b6dad1b0ad993fa62d0f463a9168e0f35e8f780 | |
parent | d65c6d428a998a69944664bd9205c2cac0863ecf [diff] |
[Flow] Improve dispatch name categorization around broadcast/transpose (#17890) The dispatch names are largely to tell us 1) What kind of computation it is and 2) What did fusion come up with This patch changes the way that broadcast and transpose is labeled to reflect what we want to know about each dispatch. Essentially, it tries to categorize dispatches as follows: Elementwise: Dispatches that are pure elementwise (identity) maps with potentially some minor transposed/broadcasted operands. This indicates that the core memory bound operands are pure elementwise. Transpose: Same as elementwise except either the input or output maps are permuted. This indicates that there is data movement happening. Broadcast: Cases where the input maps are all strict projections of the output maps. This should only ever appear if something in fusion went off the rails.
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
See our website for more information.
Community meeting recordings: IREE YouTube channel
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.