llnl / llnl/RAJAPerf

Add MARBL Matrix Free Solve Test

Open
#223 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement
Dominant language
Jupyter Notebook
Stars
135
Forks
55
Avg merge
4d 16h
Merged PRs (30d)
5

Description

We are interested in kernels for solving the advection equation: M[du/dt] = K u with DG-FEM.

We can break this up into two parts.

y = inv(M) x
Since M is block diagonal we can look at applying a block diagonal matrix to a vector. Using hierarchical parallelism we can assign N number of blocks to a GPU thread block - N is a tunable parameter. The matrix block sizes are 27x27.

General question: Can we leverage the matrix cores for this?

  1. Applying the action of K u
    Will be similar to the other FEM_PA kernels in the suite, will push up.

Something that can come from this is guide how we develop "tunable" kernels in general with RAJA.

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 reading the existing FEM_PA kernels and the surrounding performance tests, then compare them with the issue's two proposed parts: the 27x27 block-diagonal solve and the action of K u. The scope and completion criteria need clarification before implementation, including whether the test should cover both parts and how GPU matrix-core use and tunable block counts should be evaluated.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
performance
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.