rapidsai/cudf

Use grid stride in CSV reader kernels

Open

#14,066 opened on Sep 8, 2023

View on GitHub
 (1 comment) (0 reactions) (1 assignee)C++ (6,000 stars) (735 forks)batch import
PerformancecuIOgood first issue

Description

Currently, the CSV reader parses data using a thread per row, and a separate thread is used for each row, regardless of the file size. Using a grid stride loop would allow kernels to launch with preset number of blocks even with large input.

This applies both to the parser and the data inference kernels.

Contributor guide

Tech stack
cpp
Domain
backenddataperformance
Issue type
performance
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.
half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
blocked
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
CUDA programminggrid stride loopsC++
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20
Research direction
Investigate the current CSV reader kernel implementations in the cudf repository, specifically the parser and data inference kernels. Look for where threads are assigned per row and refactor to use a grid stride loop. This will involve modifying CUDA kernel launches and loop structures in files such as csv reader.cu or similar. Refer to existing CUDA grid stride examples in the codebase for pattern consistency.