| commit | e0f3c0592bad66c66d192b6c823518572d2c83b9 | [log] [tgz] |
|---|---|---|
| author | Quinn Dawkins <quinn.dawkins@gmail.com> | Mon May 20 17:26:01 2024 -0400 |
| committer | GitHub <noreply@github.com> | Mon May 20 17:26:01 2024 -0400 |
| tree | 255e84524ce07a05836d4eccbaea9590ab7ed821 | |
| parent | dc61fcc9d6fc9c237a4111d67a33fcc66b9a4ac2 [diff] |
[Codegen][GPU] Change iree_gpu.shuffle_tensor to take a region for the read (#17425) This simplifies the number of fields required for the ops and enables including reshaping of the intermediate allocation without needing to add fields to the op ad infinitum. This change has another motivation due to an issue arising from alloc reuse that naturally arises from hoisting static allocations out of loops. In short, such hoisting (and bufferization) requires a synchronization not only on the write to the allocation, but also after all reads have completed due to reusing the same allocation for each iteration of the loop. This dependency is not modeled with SSA before or after bufferization, meaning the fact that this operation represents both the write and the reads is saving us with some spooky action at a distance. This missing dependency needs more investigation in the future, but it is unclear to me at the moment how to navigate bufferization and vectorization currently. I suspect we will end up wanting a vectorization pattern for this operation, but I'm leaving that as TODO for now. This also makes the intermediate type a tensor again because we were just using `bufferization.to_memref` before to get back to a tensor and the generated IR was unnatural. Perhaps worth another look in the future as well.
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.
Community meeting recordings: IREE YouTube channel
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.