Skip to content

Conversation

@ScottTodd
Copy link
Member

  • Remove Windows style line endings
  • Adjust arguments passed depending on the host platform
  • Remove unnecessary ICD manifest file

Fixes #9 (again :D)

- Remove Windows style line endings
- Adjust arguments passed depending on the host platform
- Remove unnecessary ICD manifest file

Fixes #9 (again :D)
@ScottTodd ScottTodd deleted the scotttodd-swiftshaderlinux branch October 23, 2019 01:15
stellaraccident pushed a commit that referenced this pull request Sep 24, 2023
As discussed in various forums, we are planning to componentize the XLA
SPMD partitioner and HLO optimizer. In order to maximize velocity, I
have prototyped a first draft of this in the pjrt-plugin repo since it
has the right dependencies and needs to be the first user. I'd like to
do a couple of turns on this to get it into good shape and then propose
an RFC to find it a proper home.

A few points:

* This tool is completely stablehlo in and out. Internally, it is using
the existing HLO partitioner and optimizer, but in a similar way that
IREE's compiler API supports multiple selectable pipelines, this can
give us an entry point to augment/replace this partitioner over time
while maintaining the API surface and the existing implementation for as
long as desired.
* I've set the tool up with layering to support a stable C API following
the pattern that IREE has laid out. This is all just in the tool itself
right now but is set up so it can be factored into an API+tool.
* Once the C API is in place, I'll wrap it with Python and make this
releasable.
* I need to plumb a number of flags and knobs through. Ran out of time
for the first draft.
* I have only adapted the pre-spmd and spmd sections of the XLA GPU
pipeline. This can be varied and extended over time. Some optimizations
that XLA:GPU does are not in a convenient place with respect to phase
ordering and we can look at this as we get there.
egebeysel pushed a commit to egebeysel/iree that referenced this pull request Jul 3, 2025
python bindings: add version to `__init__.py` files

Follow up from RooflineAI/roof-mlir#473 (comment)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Add SwiftShader support to the Vulkan HAL

3 participants