skills.rest
mindspore-ai
mindspore-aiOFFICIAL
Explore all skills by this author →
WebsiteGitHub

tilelang-cuda-synchronization

Standardize T.sync_threads usage for safe TileLang CUDA kernel generation.

258|48|Updated Jun 22, 2020
Search Tags:#synchronization#cuda#deadlocks#thread-safety#shared-memory#tilelang
One-click install
npx skills add https://github.com/mindspore-ai/akg --skill tilelang-cuda-synchronization-mindspore-ai
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: tilelang-cuda-synchronization
Source: https://github.com/mindspore-ai/akg/tree/main/akg_agents/python/akg_agents/op/resources/skills/tilelang-cuda/guides/tilelang-cuda-synchronization
Command: npx skills add https://github.com/mindspore-ai/akg --skill tilelang-cuda-synchronization-mindspore-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

TileLang CUDA synchronization ensures correct and deadlock-free kernel generation by standardizing the use of T.sync_threads(), preventing common thread-safety pitfalls in shared-memory CUDA kernels.

Core Features & Use Cases

  • Strict rules for T.sync_threads() usage to avoid deadlocks and race conditions.
  • Guidance on when and how to synchronize around shared memory allocations and reads.
  • Real-world scenarios for multi-thread collaboration and kernel correctness with TileLang DSL.

Quick Start

Provide a safe TileLang CUDA kernel example that uses T.sync_threads() after all writes to shared memory.

Frequently Asked Questions about tilelang-cuda-synchronization

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent deadlocks when using T.sync_threads in TileLang CUDA kernels?▼

To prevent deadlocks in TileLang CUDA kernels, use T.sync_threads safely by avoiding conditional synchronization and ensuring all threads participate in the synchronization barrier.

What is the correct way to synchronize shared memory reads in TileLang?▼

Synchronizing shared memory in TileLang requires calling T.sync_threads after all threads complete their writes to shared memory, preventing race conditions during multi-thread collaboration.

Why does my TileLang CUDA kernel hang with conditional synchronization?▼

Your TileLang CUDA kernel hangs because conditional synchronization causes deadlocks; all threads must reach T.sync_threads uniformly, so avoid placing synchronization inside divergent control flow.

Do I need to use built-in reductions for thread-safety in TileLang CUDA?▼

Yes, using built-in reductions in TileLang CUDA is required for thread-safety, as they internally manage synchronization and prevent race conditions during multi-thread collaboration.

Does TileLang support standardizing thread synchronization across different CUDA backends?▼

TileLang supports standardizing thread synchronization across CUDA backends by enforcing strict T.sync_threads usage rules, ensuring deadlock-free kernel generation and shared memory safety.

Related Skills

Explore More Software Engineering Skills→
mindspore-aimindspore-ai
258

tilelang-cuda-patterns

Generate TileLang CUDA kernels from elementwise, reduction, matmul, and GEMV templates.

Official
Intermediate
xchang1121xchang1121
6

tilelang-cuda-synchronization

Enforce safe synchronization in TileLang CUDA kernels using T.sync_threads().

Community
Advanced
mindspore-aimindspore-ai
258

tilelang-cuda-basics

Write GPU kernels with TileLang CUDA using T.Kernel and built-in primitives.

Official
Intermediate
mindspore-aimindspore-ai
258

cuda-basics

Explain CUDA programming fundamentals covering memory models, thread hierarchy, and optimization techniques.

Official
Basic