| commit | d86ce718349d1cbc2c71af33cef887a9c20d0dca | [log] [tgz] |
|---|---|---|
| author | Ben Vanik <ben.vanik@gmail.com> | Mon Jul 11 09:54:46 2022 -0700 |
| committer | Ben Vanik <ben.vanik@gmail.com> | Thu Aug 18 12:51:24 2022 -0700 |
| tree | 24a21c9d9cc6002882befb726417a3c4b49d2cbe | |
| parent | 12b09f6276861250c8b7d46475cc321eba44ce0c [diff] |
Adding hal_inline dialect and runtime module. This lowers from the stream dialect into a much reduced form of the HAL dialect that uses a compatible type system with the HAL dialect but a restricted synchronous/local execution model. Executables translated to `vmvx-inline` are inlined directly into the host module and the only thing remaining is `!hal.buffer`/`!hal.buffer_view` management for ABI compatibility with the full HAL dialect. The tradeoff here with the full HAL dialect is that this only runs in-process and synchronously on the VM (bytecode or emitc) and is not relevant to CUDA/multithreaded CPU/etc. For a single-core embedded device with VMVX kernels, though, it should be more than enough to run all models.
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.