[MLIR-HLO] Missing legalization for mhlo.scatter to standard MLIR

[dime10]

Click to expand!

Issue Type

Support

Have you reproduced the bug with TF nightly?

No

Source

source

Tensorflow Version

main branch

Custom Code

Yes

OS Platform and Distribution

No response

Mobile device

No response

Python version

No response

Bazel version

No response

GCC/Compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current Behaviour?

See below.

Standalone code to reproduce the issue

mhlo.scatter

Relevant log output

No response

Problem Statement

Is there a pass (sequence) that can lower the mhlo.scatter operation to standard MLIR dialects, such as linalg, tensor, arith, and/or scf?

The goal is to ultimately lower to the LLVM dialect and perform codegen with LLVM. I wasn't able to find a pass that converts the mhlo.scatter op out of the MLIR-HLO dialect domain. Most other ops can be converter via passes like --hlo-legalize-to-linalg, --mhlo-legalize-to-std, or --mhlo-legalize-control-flow.

(Duplicate of tensorflow/mlir-hlo#64 since I'm not sure that repository is monitored for issues.)