Documentation

Detailed documentation for Raster's subsystems. For a quick overview, see the main README.

Guides

Document	Description
Compiler Pipeline	Nanopass architecture, pass sequence, compile-aot, diagnostics
Automatic Differentiation	Forward-mode, reverse-mode, rrules, sensitivity analysis
GPU Computing	Parallel primitives, session API, backends, SoA layout
Deep Learning	Layers, loss, optimizers, compiled training, einsum

Interactive Notebooks

The notebooks/raster/ directory contains Kindly/Clay notebooks that walk through features with runnable code:

Notebook	Topics
Getting Started	deftm, typed dispatch, value types
Automatic Differentiation	Forward-mode, reverse-mode, value+grad
ODE Solvers	Lorenz attractor, adaptive solvers, events
Linear Algebra	Vec/Mat types, LU, Cholesky, SVD
Optimization	L-BFGS, Nelder-Mead, Newton's method
Deep Learning	MLP training with compiled AD

Architecture

The design/ directory contains architecture documents for ongoing and planned work:

Document	Topic
Compiler	Compiler architecture overview
Type System	Typed multiple dispatch design
Autodiff	AD design and implementation
SOAC Fusion	Loop fusion design
Functional Combinators	Pure parallel form design
Compiler North Star	Verified GPU IR, scheduling, artifacts, autotuning, and distributed execution