llvm/torch-mlir

Add support for `AtenDivTensorOp` `AtenDivScalarOp` with integer operands

Open

#456 opened on Dec 2, 2021

View on GitHub
 (1 comment) (0 reactions) (0 assignees)C++ (692 forks)github user discovery
help wanted

Repository metrics

Stars
 (1,840 stars)
PR merge metrics
 (Avg merge 11d 19h) (18 merged PRs in 30d)

Description

The type promotion rule for div ops are different when both operands are integers (integer operands are promoted to default float type) see https://pytorch.org/docs/stable/generated/torch.div.html.

To support this, AtenDivTensorOp and AtenDivScalarOp need to have some customized logic in RefineTypes rather than using visitBinaryBroadcastingOp or visitBinaryTensorScalarOp directly as it is now https://github.com/llvm/mlir-npcomp/blob/5475d02a0d03b2bda7ef9cc240773041d22a926b/lib/Dialect/Torch/Transforms/RefineTypes.cpp#L305-L308.

Once the changes in RefineTypes are made, the TorchToLinalg lowering code can be updated as well.

Contributor guide

Research direction
Study the type promotion rules for division in PyTorch and understand how RefineTypes works in the Torch dialect. Then implement custom logic for AtenDivTensorOp and AtenDivScalarOp in RefineTypes.cpp and update the TorchToLinalg lowering.
Tech stack
cpp
Domain
compilersmachine learning
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
C++MLIRPyTorch
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
35