| commit | 68e781063d350e725ace6a80d1ca78bb536fcbf3 | [log] [tgz] |
|---|---|---|
| author | Stella Laurenzo <stellaraccident@gmail.com> | Tue Jan 25 22:24:56 2022 -0800 |
| committer | GitHub <noreply@github.com> | Tue Jan 25 22:24:56 2022 -0800 |
| tree | 5ee996c5d3f47c8a01f6979928f9f5296a34eff0 | |
| parent | e22600607e4671383fa8efeaa13fbdae317e8bc2 [diff] |
Integrate llvm-project at 92c1c63daeaf0b6b7abc6561133e2d3dbda80f8c (#8174) * Integrate llvm-project at 92c1c63daeaf0b6b7abc6561133e2d3dbda80f8c * Reset third_party/llvm-project: 92c1c63daeaf0b6b7abc6561133e2d3dbda80f8c (2022-01-24 17:41:09 -0800): [mlir][sparse] integration test for sparse output operation * First step in adapting to substantial upstream changes. Roughly: * FunctionLike -> FunctionOpInterface * linalg::PadTensorOp -> tensor::PadOp * Renames and refactors of comprehensive bufferization * Block arguments now require a location * Other random upstream fallout Does not yet build: * Quite a few, likely common issues with Comprehensive Bufferization renames that were non obvious. * A couple of lingering PadOp changes that were non obvious. * VMDialect has getAsmResultNames that has mysteriously started erroring in a way that is not obvious what has changes (does not appear to be a virtual function but no recent edits). * Several bufferization/padding todos commented out to build. See changes to Codegen/... and LinalgExt/... directories. Only tested with `bazel build iree/compiler/... -k` so far. Roughly 67 failures (many common to a codegen header). Hopefully, test fallout is minimal: I haven't seen a lot of IR changes so far. * Fix some One-Shot Bufferization integration issues * Fix linter errors. * Swap mlir-hlo submodule to use iree-mhlo-fork repo This allows us to use upstream commit 9ec54804, which fixed several issues after MLIR infrastructure changes. * Update usages to fix compilation breakages * Renamed `function_like_impl` -> `function_interface_impl` * Added locations to `addArgument()` * Fixed several places that we should pass in an `ArrayRef` of locations instead of one single instance * Add a couple of missing Bazel deps. * Fix PadOp's YieldOp: now it's in tensor dialect * Deal with getAsmResultNames() being deleted from the dialect op interface. Adds the original logic to individual op implementations and implements a fallback interface for the remaining behavior. * Trivial rename of linalg.pad_tensor -> tensor.pad and remove no predecessors checks from lit tests. * Adapt to lld main signature re-order. * Fix some more linalg.pad_tensor renames and fix instances of incorrect number of locations passed to Block::addArguments(). * Buildifier. * Remove XFAIL from three integrations tests that started passing. * Fix lite tests due to `linalg.pad_tensor` -> `tensor.pad` move. * Fix another mismatched instance of Block::addArguments * Break failing tiling pad test out and XFAIL it (like the others). Co-authored-by: Matthias Springer <springerm@google.com> Co-authored-by: MaheshRavishankar <ravishankarm@google.com> Co-authored-by: Lei Zhang <antiagainst@google.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.