skills.rest
LancerLab
LancerLabOFFICIAL
Explore all skills by this author →
GitHub

croq-dsl-cute-cpp

Build, run, and profile CuTe/CUTLASS C++ GPU kernels with nvcc.

Updated Apr 10, 2026
Search Tags:#benchmark#gpu#kernel#performance-tuning#cuda#cutlass
One-click install
npx skills add https://github.com/LancerLab/croqtile-tuner --skill croq-dsl-cute-cpp
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: croq-dsl-cute-cpp
Source: https://github.com/LancerLab/croqtile-tuner/tree/main/.claude/skills/croq-dsl-cute-cpp
Command: npx skills add https://github.com/LancerLab/croqtile-tuner --skill croq-dsl-cute-cpp

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

This Skill provides a structured way to build and run CuTe and CUTLASS C++ template kernels, enabling performance tuning and benchmarking for GPU code.

Core Features & Use Cases

  • Build Templates: Scripts generate compiled binaries for specific kernel configurations using nvcc.
  • Run and Profile: Scripts execute the kernels and gather performance metrics like TFLOPS and execution time.
  • Use Case: A developer wants to optimize a matrix multiplication kernel for a specific GPU architecture and compare different configurations.

Quick Start

Use the build_iter.sh script to compile a kernel with a specific shape and run_iter.sh to execute it and capture performance data.

Frequently Asked Questions about croq-dsl-cute-cpp

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

FAQPage Schema
How do I benchmark and profile CuTe C++ template kernels for GPU performance tuning?▼

Benchmark CuTe kernels by using scripts like build_iter.sh to compile specific configurations with nvcc and run_iter.sh to execute them, capturing performance metrics like TFLOPS and execution time for optimization.

What is the best way to optimize CUTLASS matrix multiplication kernels for a specific GPU architecture?▼

Optimizing CUTLASS matrix multiplication kernels involves generating compiled binaries for specific kernel configurations, running them to gather detailed profiling data, and comparing configurations to identify the highest performing setup for your target GPU architecture.

Do I need nvcc to build and run CuTe GPU kernels?▼

Yes, nvcc is required to build and run CuTe GPU kernels. The Skill relies on nvcc to generate compiled binaries from C++ templates and can verify kernel execution while gathering performance metrics during the profiling workflow.

Can I compare different CUTLASS kernel configurations to measure TFLOPS?▼

Yes, you can compare different CUTLASS kernel configurations by compiling each shape into a binary, executing them, and capturing performance metrics like TFLOPS and execution time to evaluate which configuration delivers the best results.

How do I start profiling CuTe C++ templates to verify kernel execution?▼

Start profiling CuTe C++ templates by using the build_iter.sh script to compile a kernel with a specific shape, then run run_iter.sh to execute it, verify kernel execution, and capture detailed performance data for analysis.

Related Skills

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

cuda-c-examples-torch

Compile and call CUDA C kernels from PyTorch using load_inline.

Official
Advanced
michaeldubumichaeldubu

perf-analyzer

Parse and analyze CUDA, CPU, and memory benchmark outputs across CSV, JSON, and text formats.

Community
Advanced
LancerLabLancerLab

croq-dsl-cute-dsl

Tune and validate Python JIT GPU kernels using CuTe DSL.

Official
Intermediate
slowlyCslowlyC
154

cutlass-skill

Develop, debug, and optimize CUTLASS and CuTeDSL GPU kernels for NVIDIA architectures.

Community
Advanced