# Tongkaio/CUDA_Kernel_Samples

CUDA 算子手撕与面试指南

Repository: https://github.com/Tongkaio/CUDA_Kernel_Samples
Canonical: https://ross.abutalabs.com/products/cuda_kernel_samples
Language: Cuda
License: MIT
License Family: permissive
Last push: 2025-08-23T14:14:00+00:00

## Health v2 (maintenance only)
Score: 42/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 38, release rhythm 35, longevity 61
- inputs: {"age_days": 855, "days_push": 375, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1095, forks 117 (observed 2026-08-28T04:03:34.052780+00:00)

## What it is
A curated collection of hand-written CUDA kernel samples covering common operators like elementwise add, reduction, softmax, sgemm, and transpose, with naive and optimized implementations. It serves as an interview preparation guide for CUDA programming, pairing each operator with GPU architecture knowledge points and optimization strategies.

## Use cases
- prepare for CUDA kernel interview questions
- learn to write optimized sgemm kernels
- study reduction and softmax kernel optimization
- understand vectorized memory access with float4
- learn how to avoid bank conflicts in transpose kernels
- practice writing CUDA operators from naive to optimized versions

## When to choose
- you are preparing for GPU/CUDA programming interviews
- you want annotated, progressive examples of kernel optimization techniques
- you need a quick reference for common CUDA operator implementations

## When to avoid
- you need a production-ready CUDA operator library like cuBLAS or CUTLASS
- you are looking for a general CUDA tutorial rather than operator-focused interview prep
- you need kernels for domains beyond the covered operators (e.g., convolutions, attention)

## Facets
- artifact type: learning-resource
- maturity: active
- function: gpu-computing, developer-tools, benchmarking
- domain: gpu-computing, machine-learning, tutorials, education
- platform: cpp, cross-platform
- tags: cuda, gpu-kernels, interview-preparation, performance-optimization, sgemm, reduction, deep-learning-operators, gpu, linux

## Member repositories
- Tongkaio/CUDA_Kernel_Samples (main) score 42

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:34.052780+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:47:14.163718+00:00, confidence not recorded.
  - readme: https://github.com/Tongkaio/CUDA_Kernel_Samples (fetched 2026-08-28T04:03:34.052780+00:00, sha a3a00a116360)
- Data as of 2026-08-30T08:39:29.467469+00:00.