| commit | 2c88e49d184b621c763a5cfb4af693f8dcbc6a07 | [log] [tgz] |
|---|---|---|
| author | Stanley Winata <68087699+raikonenfnu@users.noreply.github.com> | Tue Apr 02 15:34:31 2024 -0700 |
| committer | GitHub <noreply@github.com> | Tue Apr 02 15:34:31 2024 -0700 |
| tree | 44dd446a22bcfb92236ca914805a74bb6053a5dc | |
| parent | d1eef77a7c2a324b4cfa7dc004deb10bc9bd6c67 [diff] |
[LLVMGPU] Wmma layout for LLVMGPU vector distribute pipeline (#16928) This PR introduces WMMA layout on to LLVMGPU pipeline. The main changes are actually not too big, surrounding introduction of the WMMA layouts, data-duplicate to express duplication of data through modification of the thread basis, and emitting WMMA intrinsic. Large portions of the changes are generalizing the names of classes and variables to represent that we are doing MMA in general and not mfma specific things since most parts of the mfma layout work that we have done is reusable. A todo that I plan to handle after this patch is to get layout for 16x16x16 with an FP16 accumulator since it has a weird requirement to further interleave the output/C-matrix data since it still only does 8 elements per wmma instruction on C-Matrix, however we want to represent it as 16 elements where index=0,2,..,14 holds the real value.
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.