[saqibkh]

The cuda row in linux.yml is built with skip_tests: true (and NVIDIA_VISIBLE_DEVICES: void), and the hosted runners have no GPU, so the ctest solution_* and cli_* cases are never run on the CUDA model. This adds a separate, optional workflow .github/workflows/gpu-sim.yml that builds the CUDA model and runs them on a simulated NVIDIA H100 using PantheonSim (no GPU needed). I am the author of PantheonSim, so please weigh this accordingly.

It does not touch any source file or linux.yml, and is not meant to be a required check. A simulator cannot reproduce hardware-specific behavior or timing, so it complements real-GPU testing and does not replace it.

What it does:

• installs the CUDA 13.0 toolkit and the simulated H100
• configures with -DMODEL=cuda -DCUDA_ARCH=compute_90 (PTX only, which the simulator runs) and -DCUDA_EXTRA_FLAGS="-cudart shared" (the simulator needs a shared CUDA runtime)
• builds, then runs ctest under vgpu run
• runs on manual dispatch and on pull requests to main that touch src/cuda/, driver/, tests/, cmake/, CMakeLists.txt, Benchmarks/tea_bm_1.in or tea.problems, with a concurrency group that cancels superseded runs

Measured:

• Run on my fork: https://github.com/saqibkh/TeaLeaf/actions/runs/42746144933 . All 8 ctest tests passed (5 solution_* cases and 3 cli_* cases); the default 60 s timeout is enough.
• The job took about 16 minutes, of which about 15.5 were the simulator and toolkit setup; configure and build took about 20 seconds and the tests about a second.
• Only the small ctest decks (10x10 and 64x64 cells) are run, not the Benchmarks/ decks.