llvm/llvm-project

Implement non-unit stride dense vectorization (difficult starter)

Open

#51,660 opened on Oct 26, 2021

View on GitHub
 (4 comments) (0 reactions) (1 assignee)C++ (10,782 forks)batch import
bugzillagood first issuemlir:llvmmlir:sparsetensor

Repository metrics

Stars
 (26,378 stars)
PR merge metrics
 (Avg merge 1d 2h) (1,000 merged PRs in 30d)

Description
Bugzilla Link 52318
Version unspecified
OS Linux
CC @joker-eph

Extended Description

Implement strided load/stores on dense arrays, so that vectorization does not need to be disabled when the access pattern is non-contiguous for the dense data structures.

Start by looking at denseUnitStrides() in https://github.com/llvm/llvm-project/blob/main/mlir/lib/Dialect/SparseTensor/Transforms/Sparsification.cpp

This will require tensor loads and stores that have become non-unit stride to compute the proper strides and use this during vectorization.

This is a non-trivial starter, and requires adding vector dialect support first (for the non-unit stride load and stores) and then very careful engineering within the sparse compiler codegen).

Contributor guide

Research direction
Examine the denseUnitStrides function in Sparsification.cpp, understand how unit stride vectorization works, then design non unit stride load/store operations in the vector dialect and integrate them into the sparse compiler codegen.
Tech stack
None
Domain
backend
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++Git
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60