IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler that lowers ML models to a unified IR optimized for real-time mobile/edge inference against heterogeneous hardware accelerators. IREE also provides flexible deployment solutions for the compiled ML models.
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!
- GitHub Issues: Preferred for specific technical issues and coordination on upcoming features.
- Google IREE Discord Server: The core team and collaborators coordinate daily development here; good for low-latency communication.
- Google Groups Email List: Good for general and low-priority discussion.
- MLIR topic within LLVM Discourse: IREE is enabled by and heavily relies on MLIR. IREE sometimes is referred to in certain MLIR discussions. Useful if you are also interested in MLIR evolution.
For development, IREE supports both Bazel and CMake on Windows and Linux. We are working on enabling macOS support. For deployment, IREE aims to additionally cover Android and iOS.
We also have some public talks that explain IREE’s concepts and architecture:
- 2020-03-18: Interactive HAL IR Walkthrough (Ben Vanik and core team) (recording)
- 2020-01-31: End-to-end MLIR Workflow in IREE (recording and slides)
IREE adopts a holistic approach towards ML model compilation: the IR produced contains both the scheduling logic, required to communicate data dependencies to low-level parallel pipelined hardware/API like Vulkan, and the execution logic, encoding dense computation on the hardware in the form of hardware/API-specific binaries like SPIR-V.
The architecture of IREE is best illustrated by the following picture:
Being compilation-based means IREE does not have a traditional runtime that dispatches “ops” to their fat kernel implementations. What IREE provides is a toolbox for different deployment scenarios. It scales from running generated code on a particular API (such as emitting C code calling external DSP kernels), to a HAL (Hardware Abstraction Layer) that allows the same generated code to target multiple APIs (like Vulkan and Direct3D 12), to a full VM allowing runtime model loading for flexible deployment options and heterogeneous execution.
IREE aims to
- Support advanced models on mobile/edge devices. Dynamic shapes, dynamic flow control, dynamic multi-model dispatch, streaming models, tree-based search algorithms, and other are all good examples of exciting ML evolution. We are trying to build IREE from the ground-up to enable these models and run them efficiently on modern hardware, especially on mobile/edge devices.
- Demonstrate MLIR’s ability to develop non-traditional ML compiler backends and runtimes. MLIR enables IREE’s holistic approach of focusing on the math being performed and how that math is scheduled rather than graphs of “ops”.
- Embrace standard-based ML via Vulkan. The graphics world is shifting towards favoring modern explicit APIs for performance and predictability and Vulkan is emerging as the “compatibility” layer. We would love to allow hardware vendors to be able to make ML efficient on their hardware without the need for bespoke runtimes and special access. We also would love to let developers and users utilize all the hardware available on as many platforms as possible.
IREE is in the early stages of development and not yet ready for broad adoption. Check out the long-term design roadmap to get a sense of where we’re headed.
IREE is licensed under the terms of the Apache license. See LICENSE for more information.