mirage-project/mirage

Track some optimization/issue for persistent kernel

Open

#281 opened on May 28, 2025

View on GitHub
 (0 comments) (0 reactions) (1 assignee)Cuda (232 forks)auto 404
good first issue

Repository metrics

Stars
 (2,381 stars)
PR merge metrics
 (PR metrics pending)

Description

  • add a float4 struct to init zero_buffer, we can get rid of the forloop, we can use this as a reference
  • split the argmax task to argmax + aggregate_argmax, reduce the iterations on single threadblock
  • Filling zero values when first loading values to shared memory, this may related to this #295 . we can remove such lines, here's the reference of how to do this
  • add performance check for unit tests

Contributor guide

Research direction
Examine the CUDA kernel files mentioned, understand the async copy (cp.async) mechanism, and implement float4 loads to optimize zero buffer initialization. Also, study the argmax split and shared memory zero filling patterns.
Tech stack
cpp
Domain
backendperformance
Issue type
Performance
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
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
GitCUDA
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20