| commit | be9f097562ed9a6b0971ef8b2f09e066b412109f | [log] [tgz] |
|---|---|---|
| author | Stella Laurenzo <stellaraccident@gmail.com> | Mon Apr 01 18:12:29 2024 -0700 |
| committer | GitHub <noreply@github.com> | Mon Apr 01 18:12:29 2024 -0700 |
| tree | 3313da2c664a1f6a245e51d2649fe8f8df8af876 | |
| parent | 41e312a26cefef299976f76c8882b1d894d5734b [diff] |
[hip] Introduce options to control load of libamdhip64.so. (#16766) * Adds a multi-valued flag `--hip_dylib_path`, which if present, overrides default library search behavior: * If just a path, it is interpreted as a directory. * If prefixed with `file:` it is a specific library file and will not be expanded to anything else. * If a `--hip_dylib_path` flag is not set, then the `IREE_HIP_DYLIB_PATH` environment variable is consulted/used. * Library loading code is now much more verbose on error, adding an annotation for each combination tried/failed. Even though this adds some error message accumulation to the success path, I deem it worth it after experiencing all of the failure modes myself and needing to printf for more information: There is much more in the dlopen errors beyond whether the file could be found or not, and if we get reports from the field of issues, I want all context. * Adds a stanza to dump device info that displays the path the the OS actually resolved for the HIP runtime library. * Followups will expose this to Python and add environment variables to control it uniformly as well. This necessitated a bit of an upgrade to how we handle options, and I decided to align it with what I need next (the ability to create a driver with key/value pair options from Python) and unify the flag/option handling. I free-handed the Windows DLL handling code. Will verify on my Windows machine.
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.