commit | da3f7e3fef2f10ebff4412d2cbf1b9567a5b5bcf | [log] [tgz] |
---|---|---|
author | Lei Zhang <antiagainst@google.com> | Mon Nov 01 19:33:11 2021 -0400 |
committer | GitHub <noreply@github.com> | Mon Nov 01 19:33:11 2021 -0400 |
tree | d025b09ddd0a414f64e997d9260dde7fcbeb905d | |
parent | 456dcca4f482078fb505558ddfffdcab58ff98ca [diff] |
[spirv] Make default configuration consider all partitioned loops (#7464) This commit revises the default linalg op configuration to consider all partitioned loops when deciding workgroup size and tiling schemes. Previously we only consider the innermost partitioned loop and distribute that to 1-D workgroup. This helps to fill the GPU for the cases where we have small innermost dimensions, like 244x224x3. Under the previous configuration, we will just distribute to a workgorup of, like, 64x1x1, which starves the GPU and pushes all the work to effectively 3 threads. With the current configuration this would be distributed to a workgroup of 1x32x2.
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.