commit | b08f43071a0c7ebae505f2cfb060fc01b9eade33 | [log] [tgz] |
---|---|---|
author | Julian Walker <julianreidwalker@gmail.com> | Thu Feb 29 11:05:15 2024 -0800 |
committer | GitHub <noreply@github.com> | Thu Feb 29 11:05:15 2024 -0800 |
tree | f93934d95ff4587a9b74f113f7bbbfb1e333d2d1 | |
parent | 890b070bd94a62dc90203ed0514b212836af46ac [diff] |
Bump StableHLO to 0264c4d64c82ae74a54b85d274eec5084c2c0abf (#16561) - This update spans a change made to StableHLO to switch op attributes from `I64ElementsAttr` to `DenseI64ArrayAttr`, discussed here: https://groups.google.com/a/openxla.org/g/openxla-discuss/c/hEoA4V5DZF0/m/rdNEiM20BgAJ - Unfortunately, StableHLO generated prior to this change upstream may not be compatible with the StableHLO input plug-in in IREE after this PR. - Min version of jax/jaxlib compatible with this change: v0.4.24 - Min version of TF compatible with this change: v2.16.0 - Some of the changes here were propagated from the subset of StableHLO to LinAlg lowers that live in the StableHLO repro. To update stored SHLO bytecode, I did the following: 1. Compiled openxla/stablehlo with current IREE repo version and upgraded IREE repo versions to `old_stablehlo` and `new_stablehlo` respectively. 1. `old_stablehlo/build/bin/stablehlo-opt --mlir-print-op-generic <model>.mlirbc > <model>.mlir` 1. Manually edit <model>.mlir to bring it up to new spec 1. `new_stablehlo/build/bin/stablehlo-opt --emit-bytecode <model>.mlir > new_<model>.mlirbc` The following covered most of the edits that needed to be made to the model MLIR: ``` sed -i \ -e 's/dimensions = dense<\([0-9]*\)> : tensor<1xi64>/dimensions = array<i64: \1>/g' \ -e 's/dimensions = dense<\[\([0-9]*\), \([0-9]*\)\]> : tensor<2xi64>/dimensions = array<i64: \1, \2>/g' \ -e 's/rhs_dilation = dense<\([0-9]*\)> : tensor<2xi64>/rhs_dilation = array<i64: \1, \1>/g' \ -e 's/window_strides = dense<\([0-9]*\)> : tensor<2xi64>/window_strides = array<i64: \1, \1>/g' \ model.mlir ``` ci-extra: build_e2e_test_artifacts, test_benchmark_suites
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.