Generate CUDA kernel scaffolds with Rust bindings and benchmarks.
npx skills add https://github.com/pilotspace/moon --skill cuda-scaffold
Please help me install this Agent Skill. Skill: cuda-scaffold Source: https://github.com/pilotspace/moon/tree/main/.claude/skills/cuda-scaffold Command: npx skills add https://github.com/pilotspace/moon --skill cuda-scaffold
Scaffold CUDA kernel development by automatically generating CUDA source, Rust bindings, CPU fallback, and benchmarks for a new kernel name and data types.
Run /cuda-scaffold vector_l2_distance f32 to scaffold a new CUDA kernel named vector_l2_distance with a f32 data type.
High-intent search queries and answers about installing and using this skill.
To scaffold a CUDA kernel with Rust, you can generate complete boilerplate including CUDA source, Rust wrappers, and CPU fallback code by running a command with your desired kernel name and data type. This enforces a structured directory layout for cargo projects.
Yes, when generating GPU kernel code with cudarc, the scaffolding process supports creating data-type variants for both f32 and f16. You simply specify the data type when running the scaffold command to produce the appropriate typed bindings.
The best way to generate boilerplate for a new CUDA kernel is using an automated scaffolding tool that creates the CUDA source, Rust integration, and optional Criterion benchmarks all at once. This eliminates manual file creation and ensures a structured workflow.
Yes, you can automatically generate Criterion benchmark scaffolding for CUDA kernels to measure performance. The scaffolding creates the necessary benchmark files alongside your kernel source and Rust bindings.
Before scaffolding a CUDA kernel project in Rust, you need a working cargo project environment. The scaffolding process integrates directly into your cargo project directory structure to generate the required source and binding files.
iamcapoteAutomate Stellar Soroban contract setup with Rust toolchains and OpenZeppelin crates.
drunkcodingIntegrate AOT CUDA/C++ kernels into sgl-kernel with CMake and tests.
89jobrienAutomate cargo bench benchmark cycles with budget checks and SQLite history tracking.
jwu1980-testAdd CUDA kernels to FlashInfer with JIT and TVM-FFI bindings.
