{"id":25411353,"url":"https://github.com/aryagxr/cuda","last_synced_at":"2025-10-05T17:55:28.570Z","repository":{"id":274908790,"uuid":"924459213","full_name":"aryagxr/cuda","owner":"aryagxr","description":"100 Days of CUDA!!!","archived":false,"fork":false,"pushed_at":"2025-04-14T01:17:44.000Z","size":123,"stargazers_count":5,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-04-14T10:53:50.390Z","etag":null,"topics":["cuda","gpu-programming","kernels","parallel-programming"],"latest_commit_sha":null,"homepage":"","language":"Cuda","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/aryagxr.png","metadata":{"files":{"readme":"readme.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-01-30T03:27:43.000Z","updated_at":"2025-04-14T01:17:47.000Z","dependencies_parsed_at":"2025-03-19T02:33:38.614Z","dependency_job_id":"7404759c-c1ba-4a2c-a9f2-c8f110413ab7","html_url":"https://github.com/aryagxr/cuda","commit_stats":null,"previous_names":["aryagxr/cuda"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/aryagxr/cuda","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aryagxr%2Fcuda","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aryagxr%2Fcuda/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aryagxr%2Fcuda/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aryagxr%2Fcuda/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/aryagxr","download_url":"https://codeload.github.com/aryagxr/cuda/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aryagxr%2Fcuda/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":273213953,"owners_count":25065059,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-09-01T02:00:09.058Z","response_time":120,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cuda","gpu-programming","kernels","parallel-programming"],"created_at":"2025-02-16T10:17:14.120Z","updated_at":"2025-10-05T17:55:23.530Z","avatar_url":"https://github.com/aryagxr.png","language":"Cuda","readme":"**CUDA Progress**\n\n| **Day** | **Code Summary** |\n|------------|--------------------------------------------------------------------|\n| Day 1 | CUDA set up and kernel that prints \"Hello World\" |\n| Day 2 | CUDA kernel that adds two vectors |\n| Day 3 | Adding matrices |\n| Day 4 | Vector addition using cuBLAS |\n| Day 5 | Naive matmul |\n| Day 6 | Tiled matmul using shared memory |\n| Day 7 | Naive 1D convolution with boundary checks |\n| Day 8 | Matrix multiplication using cuBLAS |\n| Day 9 | Matrix Transpose |\n| Day 10 🥳 | Naive Softmax |\n| Day 11 | Softmax using shared memory and reductions |\n| Day 12 | Softmax using warp shuffle functions |\n| Day 13 | 1D complex-to-complex fourier transform using cuFFT |\n| Day 14 | Naive layer normalization |\n| Day 15 | Optimizing layer norm using shared memory |\n| Day 16 | Optimizing layer norm using warp shuffle functions |\n| Day 17 | Optimizing layer norm using vectorized loads |\n| Day 18 | Tiled 1D convolution and halo cells |\n| Day 19 | 1D convolution using L2 cache |\n| Day 20 🥳 | [Blog Post: Optimizing Layer Normalization with CUDA](https://aryagxr.com/blogs/cuda-optimizing-layernorm) |\n| Day 21 | Simple self attention |\n| Day 22 | Optimizing self attention |\n| Day 23 | Causal attention with masking |\n| Day 24 | Causal attention + torch binding |\n| Day 25 | Multi-head attention |\n| Day 26 | Parallel add using koggle stone algorithm |\n| Day 27 | MHA debug |\n| Day 28 | Flash Attention 1 (algorithm 1) Forward pass |\n| Day 29 | Flash Attention 1 (algorithm 1) Forward pass continued |\n| Day 30 🥳 | Flash Attention 1 (algorithm 1) Forward pass |\n| Day 31 | HGEMV matvec using fp16 |\n| Day 32 | HGEMV matvec using Bfloat16 |\n| Day 33 | Matmul using Tensor cores |\n| Day 34 | Swizzle patterns on matrix transpose |\n| Day 35 | Swizzled matrix transpose using Tensor Memory Accelerators |\n| Day 36 | Brent Kung Parallel Scan algorithm |\n| Day 37 | Matvec using integer fixed point arithmetic |\n| Day 38 | Transfered 1D array from gmem-\u003esmem-\u003egmem using TMA |\n| Day 39 | Memory Coalesced layernorm + revisited Flash attention |\n| Day 40 🥳 | revisited Flash Attention 1 |\n| Day 41 | Flash Attention 1 |\n| Day 42 | Flash Attention 1 |\n| Day 43 | ReLU Activation - FP32, FP32x4, FP16, FP16x2 vectorized |\n| Day 44 | Overlapping data transfers using CUDA Streams (Vector add) |\n| Day 45 | ReLU using CUDA Streams + benchmarked |\n| Day 46 | Packed 128 bit ReLU FP16x8 kernel |\n| Day 47 | Sparse matrix-vector mul (spMV) |\n| Day 48 | Sparse padded matrix-vector mul |\n| Day 49 | RoPE Kernel: Rotary Position Embedding naive fp32 |\n| Day 50 🥳 | Optimized RoPE using vectorized loads and half precision (18x) |\n| Day 51 | Flash Attention 2 Forward |\n| Day 52 | Flash Attention 2 Forward |\n| Day 53 | Flash Attention 2 Forward |\n| Day 54 | Gaussian Elimination |\n| Day 55 | PTX vector add kernel |\n| Day 56 | GELU activatation naive fp32 kernel |\n| Day 57 | GELU activation vectorized |\n| Day 58 | Backward pass kernel for Relu activation |\n| Day 59 | Backward pass kernel for GELU activation |\n| Day 60 🥳 | LeetGPU challenge - reduction |\n| Day 61 | Optimize + benchmarked gelu kernels |\n| Day 62 | Micrograd in CUDA |\n| Day 63 | Micrograd in CUDA |\n| Day 64 | Micrograd in CUDA |\n| Day 65 | Micrograd in CUDA |\n| Day 66 | Optimized Sigmoid activation |\n| Day 67 - Day 70 🥳 | Micrograd in CUDA |\n| Day 71 | Sigmoid with half precision |\n| Day 72 | Sigmoid with fp16 vectorized |\n| Day 73 | Swish kernel |\n| Day 74 | Swish kernel vectorized |\n| Day 75 | AMD hip kernel intro + vector add kernel |\n| Day 76 | Revisiting gemm optimizations |\n| Day 77 | Gemm coalesced |\n| Day 78 | fp16 swish |\n| Day 79 | AMD competition fp8 gemm \u0026 swish optimizations |\n| Day 80 🥳 | AMD competition fp8 gemm optimizations |\n","funding_links":[],"categories":[],"sub_categories":[],"project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Faryagxr%2Fcuda","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Faryagxr%2Fcuda","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Faryagxr%2Fcuda/lists"}