Tutorials

Five notebooks, read in order. Each compiles and runs a small C++ program through the same run_cpp() helper: g++ -std=c++23 -DAETHER_CPP_MODE=1 by default (no GPU or CUDA toolchain needed), and nvcc on the SAME source when a GPU is visible.

1 — One array, two machines

Build an aether::Array, read/write it through a View, and move it to the device and back with calls that are no-ops in a CPU-only build.

2 — Expressions without temporaries

Compose +/-/* into one expression tree, and see what it costs not to.

3 — Device math

Call aether::math::erf, sincos, hypot the same way from host code or a GPU kernel.

4 — Your own device function and kernel

Write AETHER_DEVICEHOST() once; dispatch it with a loop on the CPU or an AETHER_KERNEL() launch on the GPU.

5 — Adding a function to aether

The macro scaffold behind aether::math’s own dispatch, proved by adding a new function with it.