| commit | 890b070bd94a62dc90203ed0514b212836af46ac | [log] [tgz] |
|---|---|---|
| author | Ben Vanik <ben.vanik@gmail.com> | Thu Feb 29 10:28:37 2024 -0800 |
| committer | GitHub <noreply@github.com> | Thu Feb 29 10:28:37 2024 -0800 |
| tree | 88f433fdc59301c7b6f77555c1e9b1f073ed206d | |
| parent | b2250d8ff4c81ab51095807944fe5c6809618ae0 [diff] |
Forking off device methods from TargetBackend->TargetDevice. (#16591) This is the start of changes allowing us to reason about devices independently from the backend used to generate code for them. For now the TargetDevice is fairly coupled to the TargetBackend it was forked from but in the future the intent is to split them further such that the TargetDevice is shared across multiple backends (to start the CPU -> VMVX/LLVM-CPU backends). When asked to provide a set of default or host executable targets each plugin implementing a TargetBackend now receives the device it's generating for. This allows for example a Metal device specifying runtime device properties (Metal version, etc) to have the Metal SPIRV target backend produce executables compatible with it. TargetDevice being separate allows us to provide those independently in plugins or in the main IREE compiler (like where the CPUTargetDevice will live). This will make composition possible such as multiple plugins contributing translation pipelines for a single device to allow for specialization. Future PRs will start to deprecate the `--iree-hal-target-backends=` flag in favor of specifying devices instead and then allowing each device to have its own default backends specified if the user wants global flags (`--iree-hal-cpu-target-backends=llvm-cpu,vmvx`, etc). For now it's going to remain and there's a link back from backends to devices for automatic selection.
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.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.