Sulong – A high-performance LLVM bitcode interpreter built on the GraalVM | Hacker News Reader

Sulong – A high-performance LLVM bitcode interpreter built on the GraalVM

github.com

62 pointsgfredtech17 comments

A few potential use cases:

- Compile once, run anywhere for low-level languages like C++. With limitations, of course.

- More debugging capabilities. Maybe it could be used for dynamic analysis. Theoretically it should be possible to change code at runtime.

- Something like Valgrind? Valgrind is really complex and doesn't run on Windows at all. It seems like you could make a portable substitute with this.

- Maybe GraalVM can be used to provide additional memory safety guarantees, not sure.

Kind of hard for me to say because I don't actually know exactly what GraalVM provides, so it's kind of a guess. But it does seem like there are some genuine use cases beyond the neatness of it.

I believe that one compelling use case is to avoid the overhead of cross language boundaries (ie JNI) when the native code is compiled to bitcode and run with Sulong.

Previous Graal show offs had Ruby running C extensions by implementing a Graal JIT for x86. I imagine LLVM a nicer level of abstraction to intercept at

This seems to imply only implementing a C JIT, so my memory may be wrong: http://chrisseaton.com/rubytruffle/cext

Truffleruby seems to be using Sulong now: https://github.com/oracle/truffleruby/commit/f16a52934437a96...

I wonder if that's the point. I'm not sure, I was just making guesses myself since it doesn't seem the motivation is stated.

LLVM IR usually is platform dependant. Especially so when generated by clang etc..

Obviously, you can't just run it through vanilla Clang and call it a day. However, you can do something like PNaCl where you treat the runtime environment as its own platform separate from the rest of the world with its own ABI. I'm sure some languages are more or less impacted as well, for C and C++ the problem is obvious because platforms vary so widely.

> Compiling without optimizations is not recommended with Sulong. In particular, cross-language interoperability with Java or another Truffle language will not work when the bitcode is compiled without optimizations.

This is very weird. How can (lack of) optimizations possibly interfere with things like calling conventions? Whatever they're doing, it smells of fragile.

On a related note, Sulong is what makes Nokogiri and OpenSSL possible on TruffleRuby. This only happened two weeks ago, may be it is the reason why Sulong is posted again to HN.

I am trying hard not to get overly excited, let see some real world benchmarks once Truffle is ready to run Real World Rails.

How much faster/slower is it than aot code? I'm disapointed to see no benchmarks.

It's in the papers. About 20% slower.

The main use case seems to be speed up dynamic languages (ruby, python, javascript, R), and Sulong is used to handle the C extensions, so that the optimizations can be performed across the language boundaries. Unboxing and inlining mostly in the hot paths. Makes sense. Plus more memory safety than C/C++/Fortran.

Sorry I couldn't find the paper in github page. Can you give a link?

There's a docs dir. https://github.com/graalvm/sulong/blob/master/docs/PUBLICATI...

Hm. Cool work, but I'm not sure I really see a practical use case? Ideas welcome.

For example I use it to run Ruby's C extensions on the JVM.

Polyglotting and being able to run it on any platform is pretty cool

I wonder if it would be possible to run other languages such as rust, d, or ante, that use llvm, on this? Specifically interested in how languages with gc like d or nim will fare.

Rust runs. Look at the PR queue.

Reply on news.ycombinator.com