search

nod-ai / SHARK-ModelDev

Unified compiler/runtime for interfacing with PyTorch Dynamo.
Apache License 2.0
95 stars 48 forks source link

torch.aten.max_unpool2d #764

Closed Hardcode84 closed 2 months ago

Hardcode84 commented 3 months ago

PR for max_unpool3d was merged (https://github.com/nod-ai/SHARK-Turbine/issues/765), but unfortunately, this approach won't work for max_unpool2d as corresponding aten node preserve even less info than max_unpool3d, we need a different approach, either lowering directly to (parallel) scf loops or ONNX->linalg lowering, skipping aten op. See max_unpool3d issue and PR for details.

vivekkhandelwal1 commented 3 months ago

Hi @Hardcode84, are you still working on this op?

Hardcode84 commented 3 months ago

Hi, @vivekkhandelwal1, I did max_unpool3d but I'm not working on max_unpool2d currently.

jinchen62 commented 2 months ago

Duplicate issue of https://github.com/nod-ai/SHARK-ModelDev/issues/718, so close this.