skills.rest
better-mojo
better-mojoOFFICIAL
Explore all skills by this author →
GitHub

mojo-gpu-fundamentals

Translate CUDA kernel definitions and launches into Mojo GPU programming constructs.

6|Updated Sep 26, 2024
Search Tags:#shared memory#mojo#gpu programming#kernel launch#TileTensor#warp primitives
One-click install
npx skills add https://github.com/better-mojo/uuid --skill mojo-gpu-fundamentals-better-mojo
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mojo-gpu-fundamentals
Source: https://github.com/better-mojo/uuid/tree/main/.pi/skills/mojo-gpu-fundamentals
Command: npx skills add https://github.com/better-mojo/uuid --skill mojo-gpu-fundamentals-better-mojo

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It prevents common, CUDA-based misconceptions when writing GPU-targeted Mojo programs by mapping CUDA concepts to Mojo’s actual GPU programming model.

Core Features & Use Cases

  • Not-CUDA concept mapping: replaces CUDA syntax and launch idioms with Mojo equivalents like ctx.enqueue_function[...], barrier(), and Atomic.fetch_add.
  • Correct kernel structure for GPUs: explains plain-function kernels, TensorLayout-parameterized TileTensor usage, and mandatory comptime assert input.flat_rank == N when subscripting TileTensor.
  • End-to-end GPU fundamentals: covers indexing (global_idx, thread_idx, block_idx), shared memory allocation (stack_allocation with AddressSpace.SHARED), warp primitives, memory management via DeviceContext/DeviceBuffer, and host vs target GPU checks (has_* vs is_*).

Quick Start

Use the mojo-gpu-fundamentals skill while writing your Mojo GPU kernels so you replace CUDA-style syntax with the Mojo ctx.enqueue_function launch pattern and std.gpu/layout primitives.

Frequently Asked Questions about mojo-gpu-fundamentals

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

FAQPage Schema
How do I write GPU kernels in Mojo without using CUDA syntax?▼

To write GPU kernels in Mojo, define plain functions and launch them using `ctx.enqueue_function[...]` instead of CUDA syntax, leveraging `std.gpu` primitives for indexing and synchronization across NVIDIA, AMD, and Apple Silicon accelerators.

What is the correct way to allocate shared memory when programming GPUs in Mojo?▼

Shared memory allocation in Mojo GPU programming uses `stack_allocation` with `AddressSpace.SHARED`, replacing CUDA shared memory syntax while providing similar on-chip memory performance for thread block communication.

How do I launch a GPU kernel function in Mojo?▼

GPU kernel launches in Mojo use the `DeviceContext` method `ctx.enqueue_function[...]` to queue plain-function kernels for execution on the target accelerator, replacing CUDA's `<<<>>>` launch configuration syntax.

Why does my Mojo GPU code fail when subscripting TileTensor types?▼

Subscripting `TileTensor` types in Mojo requires mandatory `comptime assert input.flat_rank == N` to validate the `TensorLayout` parameters, preventing compilation failures from rank mismatches.

Does Mojo GPU programming support warp primitives and thread synchronization?▼

Mojo GPU programming supports warp primitives and thread synchronization through `barrier()` and `Atomic.fetch_add` operations within `std.gpu`, providing correct synchronization mechanisms across different accelerator architectures.

When should I use `has_*` versus `is_*` architecture checks in Mojo GPU code?▼

Use `has_*` checks to verify target GPU capabilities during host code execution and `is_*` checks within device kernels to confirm the active architecture, ensuring correct cross-platform compilation.

Related Skills

Explore More Software Engineering Skills→
Harmeet10000Harmeet10000
7

mojo-gpu-fundamentals

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

Community
Advanced
huggingfacehuggingface
723

cuda-kernels

Optimizes NVIDIA GPU AI inference and training with custom CUDA kernels for HuggingFace libraries.

Official
Advanced
pedestrianlovepedestrianlove

Megatron-LM

Manage tensor parallelism, inference optimizations, and mixed-precision support for Megatron-LM.

Community
Advanced
celsowmcelsowm
1

sega-saturn-gamedev-baremetal

Compile C++ programs for Sega Saturn with a custom SH2 cross-compiler toolchain.

Community
Advanced