cutlass-cpp-kernel - CUTLASS and CuTe C++ Kernel Development
Write, debug, port, review, optimize CUTLASS or CuTe C++ kernels for Hopper and Blackwell GPUs
Tags
Updated: 2026-05-11Capabilities
Typical Inputs
Typical Outputs
What this skill does
- write kernel
- debug kernel
- port kernel
- review kernel
- optimize kernel
- navigate sources
- analyze templates
- analyze tiling
- analyze memory movement
- analyze kernel structure
- inspect examples
- read profiles
Inputs
- optimization guide
- CUTLASS example
- CuTe header
- kernel description
- target architecture
- GEMM schedule
- epilogue config
- pipeline config
- layout
- tile shape
- stage count
- Nsight Systems profile
- Nsight Compute profile
Outputs
- unit test
- performance comparison
- accuracy comparison
- reference pattern
- layout choice
- epilogue choice
- stage choice
Requirements
- CUDA environment
- Nsight Systems
- Nsight Compute
- optimization guides
- CUTLASS sources
- CuTe headers
- Hopper or Blackwell GPU