rust-gpu-discipline

Verify Rust GPU compute paths execute on GPU without silent CPU fallbacks.

27|3|Updated Mar 5, 2026
One-click install
npx skills add https://github.com/forecast-bio/ferray --skill rust-gpu-discipline
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: rust-gpu-discipline
Source: https://github.com/forecast-bio/ferray/tree/main/.claude/skills/rust-gpu-discipline
Command: npx skills add https://github.com/forecast-bio/ferray --skill rust-gpu-discipline

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

The Rust GPU discipline skill prevents silent CPU fallbacks and ensures GPU-backed compute paths are actually executed on the GPU by enforcing pre-flight planning, mechanical verification, and honest reporting.

Core Features & Use Cases

  • Pre-flight planning before GPU work to identify required backends, kernels, and data movement.
  • Forbidden-pattern detection to block lazy CPU fallbacks and incomplete GPU implementations.
  • Mechanical self-verification and adversarial review to validate end-to-end GPU execution and correct error signaling.
  • Honest reporting that mirrors PyTorch parity for device dispatch and fallback behavior.

Quick Start

Provide a pre-flight plan and a verification checklist before writing any GPU-accelerated Rust code.

Frequently Asked Questions about rust-gpu-discipline

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

FAQPage Schema
How do I prevent silent CPU fallbacks in Rust GPU compute kernels?

You prevent silent CPU fallbacks in Rust GPU compute by enforcing a pre-flight plan, forbidden-pattern detection, and mechanical self-verification to validate end-to-end device dispatch. This ensures GPU-backed compute paths execute on the GPU.

Why does my Rust GPU kernel dispatch execute on the CPU instead of the GPU?

Your Rust GPU kernel dispatch executes on the CPU due to silent fallbacks from incomplete implementations. Applying a forbidden-pattern checklist blocks lazy CPU fallbacks and surfaces non-GPU outcomes through honest reporting.

Does this GPU verification approach work with CUDA, ROCm, Metal, and Vulkan backends?

Yes, this GPU verification approach works with CUDA, ROCm, Metal, Vulkan, and WGPU backends. It applies to Rust GPU projects involving kernel dispatch, cuBLAS, PTX, or WGSL to ensure compute paths execute on the correct device.

What is the best way to verify end-to-end GPU execution in Rust projects like burn or candle?

The best way to verify end-to-end GPU execution in Rust projects like burn or candle is through mechanical self-verification and adversarial review. This validates correct error signaling and ensures GPU work lands on GPU.

How do I plan data movement and kernel dispatch before writing Rust GPU-accelerated code?

You plan data movement and kernel dispatch before writing Rust GPU-accelerated code by creating a pre-flight plan. This identifies required backends, kernels, and data movement to prevent silent CPU fallbacks.

When do I need a forbidden-pattern checklist for Rust GPU development?

You need a forbidden-pattern checklist for Rust GPU development when implementing kernel-dispatch code to block lazy CPU fallbacks and incomplete GPU implementations. It ensures honest reporting that mirrors PyTorch parity for device dispatch.