Peter Hawkins
Peter Hawkins
> I'm currently working on upgrading to nanobind v2.4.0, but it will take a few more days. Out of curiosity, how do you guys at JAX handle generated stubs for...
> The tsan log does not make much sense to me. Can't say I'm much happier with helgrind though, as it flags a couple of atomic operations in blas_server.c and...
> [@hawkinsp](https://github.com/hawkinsp) indeed I appear to have overlooked the ATOMIC_RELAXED qualifier - or trusted the submitter to know what he was doing - when I merged the relevant PR, which...
> [@hawkinsp](https://github.com/hawkinsp) indeed I appear to have overlooked the ATOMIC_RELAXED qualifier - or trusted the submitter to know what he was doing - when I merged the relevant PR, which...
@Saladino93 This is an issue in the Metal plugin, so it can only be fixed by Apple. The right people to take a look are already assigned to the bug.
Thanks @rajasekharporeddy . I think we can now redesignate this as an "enhancement", not a "bug". It's a bug if the plugin crashes or produces wrong output. It's now just...
You'll need to sign the CLA.
It looks like this is breaking some of our tests that rely on forward compatibility (i.e., we don't emit stablehlo that's too new for some consumers, e.g., TF, to understand)....
`jax.numpy.tan` call adds `chlo.tan` call to exported StableHLO; `stablehlo.tan` should be preferred
I suspect that `stablehlo.tan` was added after that code was written. Do you want to send a PR switching it to stablehlo? It should be a tiny change.
Sorry for the slow review, I had missed that this PR got updated. One additional change: can you please squash your commits? Thanks!