commit | 587092d3bf4d0fea32d08327d72d1b551e3d4c97 | [log] [tgz] |
---|---|---|
author | Ahmed S. Taei <asaadaldien@users.noreply.github.com> | Thu Nov 04 14:47:39 2021 -0700 |
committer | GitHub <noreply@github.com> | Thu Nov 04 14:47:39 2021 -0700 |
tree | 14a961a4880a18369ca9c47a50316ed95c3e1a5a | |
parent | 2ef9d40b2e7f6e33e9fdee16b71ca5e0ad06431f [diff] |
- Move linalg.mmt4d to the tensor -> vector pipeline (#7307) - Move linalg.mmt4d tile and vectorzation to tensor -> vector pipeline. - Add vector transform patterns needed to get better performance. Running `iree/test/microbenchmarks/linalg_mmt4d.mlir` on Pixel4 core-7 Baseline: ``` -------------------------------------------------------------------------------------------- Benchmark Time CPU Iterations -------------------------------------------------------------------------------------------- BM_matmul_384x384x512/process_time/real_time 6.64 ms 6.63 ms 106 BM_mmt4d_384x384x512_4x1x4/process_time/real_time 9.74 ms 9.72 ms 72 BM_mmt4d_384x384x512_8x1x8/process_time/real_time 7.10 ms 7.07 ms 100 ``` This PR: ``` -------------------------------------------------------------------------------------------- Benchmark Time CPU Iterations -------------------------------------------------------------------------------------------- BM_matmul_384x384x512/process_time/real_time 6.24 ms 6.22 ms 112 BM_mmt4d_384x384x512_4x1x4/process_time/real_time 4.41 ms 4.40 ms 159 BM_mmt4d_384x384x512_8x1x8/process_time/real_time 3.83 ms 3.82 ms 183 ```
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.