| commit | 9461d3b0cf01c380054f0d1c92b3e1cafa8d2a21 | [log] [tgz] |
|---|---|---|
| author | Ben Vanik <ben.vanik@gmail.com> | Tue Apr 18 16:39:25 2023 -0700 |
| committer | GitHub <noreply@github.com> | Tue Apr 18 23:39:25 2023 +0000 |
| tree | da7b3c5d84a6f9f0c75337c254a97caaa6d7c18b | |
| parent | 62153dfb9a89cddb3e72ae4620b356d18e05c661 [diff] |
Adding support for loading VM modules from dynamic libraries. (#13112) Users can now produce native shared libraries that export one or more symbols used to create VM modules allowing for multiple different modules to be packaged in the same binary. These dynamic modules receive a key-value list of parameters that they can use to customize their behavior that can be programmatically specified by whatever is loading them (python kwargs, or flag `?key=value&key=value` on the command line). On the hosting runtime side `iree_vm_dynamic_module_load_from_file` makes it easy to load and use the modules as if they were local to the binary. The command line tools have been updated to support multiple `--module=` flags and load each module into the context. This also allows for multiple vmfbs to be loaded and all kinds of linkage behavior (could have a training driver in a native module that calls functions in a compiled module loaded earlier, etc). As part of this multiple `--module=` flags can be provided now to tools for both bytecode and dynamic modules. To ensure all the tools are using the same flag helper/loader I've switched iree-benchmark-module, iree-check-module, and iree-run-module to require explicitly specifying `--module=-` to load the module from stdin. This removes the need for the "reading from stdin message" that uglied up stdout. I wasn't able to figure out Tracy support here (at least on Windows) as if tracing is globally enabled the shared libraries will get Tracy linked in and start a new tracing context. It's noted today and a warning is emitted when the sample is compiled with tracing enabled.
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.