Average seconds per model in each full OpenModelica Library Testing run, master versus wasm-jit, since 2025-01-01. Every run tests about 18,000–20,000 models across all libraries.
Source: the omdb tables master and wasm-jit, one point per run date. Only full runs (more than 15,000 models) are shown; partial reruns of a few libraries are left out.
Each value is the run's total time for that column divided by the number of models in the run, including models that failed before reaching the phase. Negative "not reached" markers count as 0.
Overall is exectime, the wall time of the whole per-model job: library loading, translation, build, simulation and verification.
Compilation is translation (frontend + backend + simcode + templates) plus build. For master the build is the C compiler; for wasm-jit it is the JIT compile that simulate() reports as timeCompile.
Simulation is the simulate column.
Models that simulate counts models whose finalphase is 6 (simulation finished) or 7 (also verified), divided by the models in the run. Its axis does not start at 0.
Frontend and Backend are the frontend and backend columns, two of the parts of compilation.
On 2026-09-09 the master lane moved from the Ryzen 5950X to the Ryzen 9950X that wasm-jit runs on. On the models both runs share, every column fell at once: frontend ×0.77, backend ×0.73, C compile ×0.55, simulation ×0.62. Before that date, part of the gap between the lanes is hardware. The CMake-built omc arrived in the same run.
The wasm-jit lane stopped doing the FMU export on 2026-09-11, when that work moved to the separate wasm-jit-me/-cs lanes.