Module plan 

Bench plans (plan.v1): everything the runner needs, self-contained — candidate configs, PTX paths, launch geometry, and the argument layout matching cuda-oxide’s PTX parameter lowering (each slice becomes a ptr, len pair of .param slots; scalars are single slots).

A kernel’s [bench] section in kernel.toml declares the workload; sizes and grid shapes are arithmetic expressions over the kernel’s dimensions plus the elements variable.

Structs

BenchPlan

A plan.v1 document: every candidate to measure, and how.

BenchSpec

The [bench] section, parsed once and used to build every candidate.

Candidate

One configuration, ready to launch: everything the runner needs and nothing it has to recompute.

Enums

ArgSpec

One PTX kernel argument slot group, in PTX parameter order.

PlanError

What can go wrong turning a spec into a benchmark plan.