Some PNaCl LLVM Patches Might Make Writing LLVM Back-Ends Easier
Written by Michael Larabel in Compiler on 5 March 2014 at 01:25 AM EST. Add A Comment
Some work from Google's Portable Native Client (PNaCl) work might be pushed upstream in LLVM as it benefits the EmScripten JavaScript back-end and other interesting use-cases.

Google's Native Client is the company's sandboxing technology for running native x86/ARM code in a sandbox from their Chrome web browser and ChromeOS. Portable Natice Client applications are compiled to a subset of the LLVM IR bytecode. While the code is available and under a BSD license, Google isn't leveraging upstream LLVM in their work. Google though is a big contributor to LLVM in general and the PNaCl team is now attempting to push some PNaCl IR passes back into upstream LLVM.

What makes these IR passes interesting and worth mentioning is that they appear to be useful simplification/lowering passes that can ease development of new LLVM back-ends, especially for interesting use-cases like EmScripten with compiling out to JavaScript for the web browser. These IR simplification passes end up simplifying LLVM IR from complex features into simpler features. With a reduced set of IR to target, it's easier for developers to make new back-ends and provides a more stable subset of LLVM IR. These PNaCl passes are modular and simplify calling conventions and module-level lowering.

Mark Seaborn of the PNaCl team at Google has proposed upstreaming this work (just these passes and not mainlining all of their PNaCl LLVM code) to help other new potential back-ends or at the very least centering the work currently done by PNaCl and EmScripten.

Alon Zakai added additional commentary about these patches in EmScripten's new back-end use-case, "It made writing the backend much easier than it otherwise would have been...I think the shared thing in these examples is that LLVM IR is very nice to work with, and there are some situations where you're using it and you have a reason to convert it into something else, and you want to do that in as _simple_ a way as possible as opposed to generating the most _optimal_ results. The PNaCl IR simplification passes are in my opinion a big help there."

Developers at Apple and Google, among other contributors, seem to be interested in this work for the potential wins. However, there is some concern amongst LLVM developers (including by Chris Lattner) over maintenance -- though Google's PNaCl team has claimed they will support it -- of these passes and not having any in-tree back-ends at the moment that utilize the passes (or any test cases right now for stressing them), but if this code pans out to be interesting and helpful, hopefully it will manage to make it in for LLVM 3.5.
About The Author
Author picture

Michael Larabel is the principal author of Phoronix.com and founded the site in 2004 with a focus on enriching the Linux hardware experience. Michael has written more than 10,000 articles covering the state of Linux hardware support, Linux performance, graphics drivers, and other topics. Michael is also the lead developer of the Phoronix Test Suite, Phoromatic, and OpenBenchmarking.org automated benchmarking software. He can be followed via Twitter or contacted via MichaelLarabel.com.

Related Compiler News
Popular News
Trending Reviews & Featured Articles