[CI] Run examples in CI

[oraluben]

Run examples in CI (H100 only, as it's ~2min faster than A100, and additional example takes 30s and doesn't depend on the device).

[jlebar]

Let's see how long it takes to run the examples in CI, I'm not sure it will be fast.

[oraluben]

Let's see how long it takes to run the examples in CI, I'm not sure it will be fast.

It looks simple and on my dev env it takes ~10s. Should be similar in CI.

[jlebar]

It looks like running the examples takes 30 seconds, 9m57s here on A100 vs 9m25s in https://github.com/openai/triton/actions/runs/8253494425/job/22575490486?pr=3353. (The H100 build is failing for reasons unrelated to this PR, the VM is unhappy.)

I'm not sure running the examples is worth increasing CI time by 5% for all PRs. We could make it part of a nightly run, although in truth people don't look at those very carefully.

As a compromise I guess we could run the examples only on H100, since the H100 builds are a few minutes faster than A100.

[jlebar]

Thanks, this LGTM but I'll let Thomas approve and decide on the question of merging those two op definitions. I'm grateful that you're cleaning things up based on the bugs you've found and fixed, thank you.

[ptillet]

I don't think it is the right place to run examples. In fact, I don't even think the examples directory should exist anymore. All educational material should live in tutorials, which should be built nightly.

[oraluben]

All educational material should live in tutorials, which should be built nightly.

Fair enough, this indeed sounds better.

Thanks for reviewing this! @jlebar