AI Systems3 articles
TinyTriton
- Overview
TinyTriton: Build a GPU compiler, one step at a time
A fifteen-part series for Python programmers with no compiler background, starting with a small array program and growing toward GPU execution and optimization.
- Step 1
Step 1. From source code to instructions
Translate a small array program into instructions. Explain compile time parameters, name lookup, recursive traversal, and the complete exercise implementation.
- Step 2
Step 2. Execute the instructions on the CPU
Build a NumPy interpreter for tile IR. Connect descriptions to runtime values and explain pointer arithmetic, parameter binding, masked loads, and stores.
3D1 article
3D Gaussian Splatting
3D Gaussian Splatting from scratch with NVIDIA Warp
Follow a Gaussian from a 3D ellipsoid to a pixel, then trace image gradients back through a renderer written with NVIDIA Warp.