| commit | 503d81229dcf53aa3f391866c3fa93231831b7bb | [log] [tgz] |
|---|---|---|
| author | Thomas <thomasraoux@google.com> | Fri Mar 10 16:18:24 2023 -0800 |
| committer | GitHub <noreply@github.com> | Fri Mar 10 16:18:24 2023 -0800 |
| tree | 96cbf7b6b6af82bad3692be9395890c23ca38c82 | |
| parent | 087ff08a2303dda24e50cd3c390cfe6b198bd072 [diff] |
[LLVMGPU] Optimize shared memory allocation size (#12538) In order to improve performance for matmul with large tile sizes we need to be able to re-use shared memory between the allocation done for the input and output. This adds a pass to pack memory allocations if we can prove that their liveranges are disjoint. This relies on a simplisitic algorithm that will sort allocations into alias groups that are independent of each others. Then we create a single alloc that is partitioned for each independent alias group. This way we only use the amount of memory needed by the largest alias group.
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.