Stella Laurenzo

This shows an example of e2e codegen driven by fine-grained transform dialect strategies. The target is controlled warp distribution of a reduction op. This exhibits the need to have a...

Consider this a prototype of where we need to get to. Feedback welcome. It is working but needs test fixup/extension if we want to pursue it. I got the list...

libzstd mis-detecting on MacOS Github Action runners

1

GitHub actions does install libzstd, but it is not universal so it detects from CMake and fails at build time. For the moment, just deleting it in the runner. Need...

Support device side buffers for Python API

2

Any real integration via the Python API requires the ability to pass/return opaque device side buffers. This will require some design work.

compile functions that do keyword expansion should declare all of them

2

Current python compile functions take a **kwargs that is passed to the options class. This indirection makes code complete less useful. They should be declared more appropriately for interactive use.

Slim vision models should test with various dynamic shape variants

2

Ideally: * dynamic batch size * dynamic image size This can be done in a patchy way by modifying `slim_vision_model_test.py` and none-ing out various shape dims: ``` input_shape = list(get_input_shape())...

Remove Config() constructor in favor of explicit API for creating devices

6

It is currently very easy when doing non default construction of a SystemContext to create multiple Config instances for the same device. The API needs to provide a mechanism to...

Move the XLA sanitization passes from iree-import-xla to the core pipeline.

2

Now that JAX can emit MLIR/MHLO directly, we can kill iree-xla-import. Or we could if we move a few bits into the core MHLO conversion related to sanitizing XLA inputs...

First step of adding nccl headers to the build.

13

* I ended up reworking the CUDA toolkit download because we really need to full toolkit now (or at least more headers from it), and the support that was implemented...

[BUG]: 'terminate called after throwing an instance of' in `def_buffer` method

11

### Required prerequisites - [X] Make sure you've read the [documentation](https://pybind11.readthedocs.io). Your issue may be addressed there. - [X] Search the [issue tracker](https://github.com/pybind/pybind11/issues) and [Discussions](https:/pybind/pybind11/discussions) to verify that this hasn't...