skills.rest
Hundo1018
Hundo1018COMMUNITY
Explore all skills by this author →
GitHub

mojo-gpu-fundamentals

Write, launch, and debug Mojo GPU kernels with TileTensor and shared memory.

7|1|Updated Apr 9, 2026
Search Tags:#gpu#kernel#mojo#memory-management#wgpu#tiletensor#cuda-migration
One-click install
npx skills add https://github.com/Hundo1018/wgpu-mojo --skill mojo-gpu-fundamentals-hundo1018
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mojo-gpu-fundamentals
Source: https://github.com/Hundo1018/wgpu-mojo/tree/main/.agents/skills/mojo-gpu-fundamentals
Command: npx skills add https://github.com/Hundo1018/wgpu-mojo --skill mojo-gpu-fundamentals-hundo1018

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill corrects common CUDA-based assumptions and provides concise, authoritative guidance for writing, launching, and managing GPU code in Mojo so developers avoid subtle correctness and performance mistakes.

Core Features & Use Cases

  • Maps familiar CUDA concepts to Mojo equivalents (kernels, launches, memory, and synchronization).
  • Explains TileTensor usage, layout creation, indexing, rebind rules, and vectorized access for correct element typing.
  • Documents buffer allocation, host-device copies, enqueue_function launch semantics, shared memory via stack_allocation, and DeviceContext workflows.
  • Use cases: porting CUDA kernels to Mojo, authoring tiled matmul and vector-add kernels, benchmarking and debugging mojo GPU code.

Quick Start

Translate this CUDA kernel into a correct Mojo kernel that uses TileTensor, stack_allocation for shared memory, and enqueue_function launch syntax.

Frequently Asked Questions about mojo-gpu-fundamentals

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

FAQPage Schema
How do I migrate a CUDA kernel to Mojo without making correctness mistakes?▼

Migrating CUDA kernels to Mojo requires correcting common CUDA assumptions by mapping concepts to Mojo equivalents, utilizing TileTensor for layouts, and applying enqueue_function for launches to ensure accurate GPU execution.

How do I launch a Mojo GPU kernel with compile-time parameters?▼

Launching Mojo GPU kernels requires using the enqueue_function launch syntax with compile-time kernel parameters, ensuring proper host-device synchronization and DeviceContext buffer creation to execute correctly.

How does TileTensor indexing and rebind work for Mojo GPU programming?▼

TileTensor in Mojo GPU programming manages layout creation and indexing, enforcing vectorized access and rebind rules to guarantee correct element typing for kernel buffers.

Can I use stack_allocation for shared memory when porting CUDA idioms to Mojo?▼

Yes, Mojo GPU programming replaces CUDA shared memory patterns by using stack_allocation to manage shared memory buffers within DeviceContext workflows for kernel execution.

What is the best way to benchmark and debug Mojo GPU code?▼

Benchmarking and debugging Mojo GPU code involves validating DeviceContext copy semantics, host-device synchronization, and TileTensor rebind usage to identify subtle performance and correctness mistakes.

Why does my vector addition kernel fail when moving from CUDA to Mojo?▼

Vector addition kernels fail during CUDA to Mojo migration due to incorrect enqueue_function launch syntax, improper TileTensor indexing, or unresolved DeviceContext buffer copy semantics.

Related Skills

Explore More Software Engineering Skills→
panicPaulpanicPaul
5

mojo-gpu-fundamentals

Translate CUDA-style GPU programming concepts into Mojo semantics.

Community
Intermediate
leonling-llleonling-ll

kernel-perf-analysis

Collect timing, hardware counters, ATT traces, and memory-hierarchy data from AMD GPU kernels.

Community
Advanced
kryptodoggkryptodogg

Oz Render Architect Skill

Optimize wgpu compute pipelines and VRAM management for DirectX 12 and Vulkan.

Community
Advanced
Harmeet10000Harmeet10000
7

mojo-gpu-fundamentals

Explain CUDA-free Mojo GPU programming with LayoutTensor and shared memory patterns.

Community
Advanced