| commit | 38ad88efe0f1c4589f2c0b6faea4b03e87a1474c | [log] [tgz] |
|---|---|---|
| author | Ben Vanik <ben.vanik@gmail.com> | Tue Feb 21 16:56:57 2023 -0800 |
| committer | GitHub <noreply@github.com> | Tue Feb 21 16:56:57 2023 -0800 |
| tree | 8ea1af42325331902176fded8546741ec0aeb1b8 | |
| parent | c889a4cb3515b53c60166f302ff01f3c43e29a8a [diff] |
Adding support for linking bitcode files in LLVM CPU/GPU targets. (#12265) Object files specified with .bc extensions are linked into our LLVM IR prior to performing code generation. This allows for target specialization by way of our own __nvvm_reflect-like mechanisms. On the CPU side with our libdevice we can have libraries use calls into that we specialize based on target configuration. As part of testing the CPU/CUDA target has gained support for the `--iree-hal-dump-executable-intermediates-to=` flag by dumping the bitcode produced by codegen, the linked bitcode, and the final optimized bitcode prior to so/o/PTX emission. Future changes will demonstrate using libdevice on the CPU to query target features and using both CPU and CUDA bitcode files to provide microkernel implementations in the custom_dispatch samples. Progress on https://github.com/iree-org/iree/issues/12222 (externalized CUDA bitcode linking not yet done). benchmarks: cuda, x86_64, comp-stats
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.