[DEPRECATED] Moved to ROCm/rocm-systems repo

Online CUDA Occupancy Calculator

(Spring 2017) Assignment 2: GPU Executor

Runs a single CUDA/OpenCL kernel, taking its source from a file and arguments from the command-line

GPU Drano Static Analysis for GPU programs.

Prototype for a SPIR-V assembler and dissasembler. It provides a composable Java interface for generating SPIR-V code at runtime.

Open source skill library for AI coding agents to write, optimize, and debug high performance compute kernels across CUDA, Triton, and quantized workloads.

A self-hosted low-level functional-style programming language 🌀

Noeris — autonomous kernel fusion discovery + drop-in LLM training accelerator. Cross-op fusion, autotuning, beats cuDNN on sliding-window. pip install noeris.

High-performance GPU-accelerated C# scripting for Rhino Grasshopper, powered by ILGPU

Assignment 2: GPU Executor

Medical AI diagnostics system implementing real compiled Mojo GPU kernels with MAX Graph integration

🍭 Sweet GPU compute kernels in CUDA, wrapped via CuPy

Alternate backend for JAX and PyTorch that generates optimized kernels using AI agents

Runtime correctness checker for custom CUDA kernels. Attach a single decorator to periodically verify outputs against a reference implementation, with outlier-biased sampling and zero training graph impact.

16-step CUDA optimization of FlashAttention-2 achieving 99.2% of official performance on A100 — Ampere architecture

High-performance Triton kernels for NVIDIA H100. Implements fused FP8 LayerNorm, tiled FlashAttention, and SRAM-optimized memory primitives for Hopper architecture.

LLM primitives rebuilt in Triton — FlashAttention 2.52×, fused AdamW 3.45×, Bias+GELU 14.65× faster than PyTorch

Benchmarking hand-written CUDA C, Numba, and Triton self-attention kernels against PyTorch's SDPA - how fast can you go depending on the tool?