Interest in GSoC Project 13 (XLA Plugin for OpenVINO)

[SuperChamp234]

Hi @rkazants, @mvafin,

I’m Zain Siddavatam, and I’m interested in contributing to Project 13: Implementing an XLA plugin for OpenVINO as part of GSoC.

I’ve already contributed to OpenVINO via PR #29019, which involved working on the CPU plugin. This gave me hands-on experience with OpenVINO’s backend plugin architecture, and I’d love to extend that understanding to XLA.

Beyond OpenVINO, I have experience in system-level C++ development, FPGA-based hardware acceleration, and compiler toolchains. Recently, I’ve been diving into MLIR and HLO dialects to better understand compiler transformations.

So far, I’ve been:

• Exploring MLIR & HLO dialects to understand the conversion pipeline, particularly StableHLO's role in OpenXLA.
• Reviewing OpenVINO’s existing plugin structures to see how the XLA plugin could integrate.
• Looking into pytorch-xla for insights on IR → XLA translation.

Could you provide guidance on any additional prerequisite contributions I should focus on before finalizing my proposal?

I’m currently drafting my GSoC proposal and would love to align my approach with your expectations.

Looking forward to your thoughts. Thanks for your time!

Best,
Zain Siddavatam

[mvafin]

Hi @SuperChamp234,

Thank you for reaching out and sharing your background. It's great to see your proactive approach.
For your GSoC proposal, consider including the following, but don't be restricted to this list:

• A high-level overview of your plan for the XLA plugin integration.
• Key steps and milestones in your implementation process.
• Code snippets or pseudocode to demonstrate your approach.
• Dependencies and tools you'll use, and why they're necessary.

While this isn't a complete list of what your application should include, these elements would be nice to have to help us assess your ability to complete the project successfully.