Getting started

Three pages, in order: provision an environment that produces trustworthy numbers, evaluate a volume potential, and make sure the computation ran on the device you intended.

• Installation
• A first volume potential
• Device selection

Volumential needs a working OpenCL runtime. Everything below assumes one is installed and that python -c "import pyopencl" succeeds; if it does not, start at Installation.

Before running anything, export PYOPENCL_CTX:

export PYOPENCL_CTX=portable:0   # the PoCL platform, device 0

Without it, pyopencl.create_some_context() — which laplace2d.py, laplace3d.py and poisson3d.py call — decides for you: at a terminal it stops and asks, and anywhere else (a batch job, a pipeline, CI) it picks a device in an implementation-defined manner. The first wastes your time, the second quietly makes the run unreproducible.

Every example script reads the variable. Without it, helmholtz2d.py, helmholtz3d.py and branched_flow_helmholtz2d.py do not ask create_some_context(): they enumerate the platforms themselves and prefer an fp64 GPU where there is one. Check the device the run reports rather than assuming the variable settled it. Device selection covers the rest; this one line is enough to get through A first volume potential.