ROCm / ROCm/AMDMIGraphX

Support text export of fusion subgraphs

Open
#3,525 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
333
Forks
150
Avg merge
4d 19h
Merged PRs (30d)
54

Description

Add a command line argument to the migraphx driver (for example export_mlir), which, when set, will perform the following:
For each mlir fusion (module) that migraphx decides to perform, the corresponding IR will be exported in a text file (.mlir); The filename will ideally be reflective of what the fusion is about (e.g. via concat of the names or the operators, appended into the model name).

Those exported files will be used as follows:
Test scripts in rocMLIR will import each such file, compile the code to create a single code object, verify that this succeeds, and if it does, "verify" the correctness, and benchmark the execution time; and, finally, write all that info in a structured way, so that results from different runs can be compared.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by locating the migraphx driver and the code that decides which MLIR fusions to perform. Review how rocMLIR test scripts consume generated .mlir files and what structured results they expect. Done means an export argument produces identifiable text files for each fusion and those files can support compilation, correctness checks, benchmarking, and result comparison.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
backend, compilers, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.