| commit | fccf80910b22875bea7aa533b8cfbb1dd06c62ae | [log] [tgz] |
|---|---|---|
| author | Stella Laurenzo <stellaraccident@gmail.com> | Tue Aug 16 08:50:12 2022 -0700 |
| committer | GitHub <noreply@github.com> | Tue Aug 16 08:50:12 2022 -0700 |
| tree | 77c78be7050bbf435fdfbd909973a96da9ea45c7 | |
| parent | 34d9339c6961fe6e5d79dce52ca9890d25f87142 [diff] |
Add an e2e CUDA transform dialect example. (#10066) * Add and e2e CUDA transform dialect example. This shows an example of e2e codegen driven by fine-grained transform dialect strategies. The target is controlled warp distribution of a reduction op. This exhibits the need to have a place to put such transform dialect IR such that we can from a centralized place and with the same tooling: 1. write an iree-opt test that checks expectes IR without having to write HAL abstractions. 2. write an e2e execution test to check results w/ accuracy etc. 3. selectively import as an expert strategy and use without surprises in IREE, when the time comes. A first tentative location for this is in `compiler/transform_dialect/`. The GPU-specific transforms are relaxed to allow targeting either hal.executable or hal.executable.variant which lets them apply with either an iree-run-mlir or iree-opt flow. Enable e2e testing for transform_dialect full bits. I think I got the tags right for this to work. You can run with: `ctest -R transform_dialect/cuda` (as an example) * Fix buildifier * Remove spurious data files * Fix test command * Rename spec files so they are more meaningful * Actually fix the test * Update test CLI * Add transform dialect spec files as data * Apply iree_cmake_extra_content change suggested by ScottTodd@ * Add driver=cuda to BUILD / CMake * Double dashes instead of single dash Co-authored-by: Nicolas Vasilache <nicolas.vasilache@gmail.com>
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.