Write CUDA GEMM Kernel

What problem does it solve?

This Skill guides engineers through designing and implementing a correct, performance-aware CUDA GEMM kernel (C = alpha * A * B + beta * C) with decisions about tiling strategy, memory hierarchy usage, tensor core eligibility, and when to defer to cuBLAS or CUTLASS.

Core Features & Use Cases

• Guided design of a custom GEMM kernel including tiling, shared memory layout, and epilogue fusion decisions.
• Decision framework for when to use cuBLAS or CUTLASS and how to handle batched or nonstandard layouts.
• Applicable in constrained hardware scenarios or research contexts where maximum control over memory and compute is required.

Quick Start

Provide the problem specifications (M, N, K; dtypes; layouts; transpositions; target SM) and initiate the kernel design workflow for a CUDA GEMM.