GraalVM: Run Programs Faster Anywhere • Hacker News Board

Re: GraalVM: Run Programs Faster Anywhere

#41

Post by grashalm » Tue, Apr 17, 2018, 9:38 PM UTC

chc4 wrote: post #39

grashalm wrote: post #37

Earlier quoted context omitted.

Which one was it? Can you remember? One of these? https://github.com/oracle/graal/blob/master/truffle/docs/Lan... If not. Please let me know.

It's not listed there, and didn't reach full feature parity (and seems to be dead now). https://github.com/frodwith/jaque is the project though, which is an implementation of the Nock VM from Urbit

Ho wow, never heard of it. Will put it on the experiments list. Thanks a lot!

Re: GraalVM: Run Programs Faster Anywhere

#42

Post by kllrnohj » Tue, Apr 17, 2018, 9:39 PM UTC

zackbloom wrote: post #30

How is this better than compiling everything to WebAssembly and using V8 as your interpreter?

Because WebAssembly is extremely limited? Unless you need to ship on the web, using WebAssembly is almost certainly a bad decision as it makes highly compromising tradeoffs that you don't need.

Re: GraalVM: Run Programs Faster Anywhere

#43

Post by thinkloop » Tue, Apr 17, 2018, 9:40 PM UTC

grashalm wrote: post #36

maltalex wrote: post #26

Can someone please explain this? > Are you working on a Java application? Do you want a faster runtime? Enhance your code with JavaScript! First, the JVM already has the Nashorn JS engine. Second, if I'm "working on a Java app", that means that it's already running in a JVM. Where does the faster runtime come from? Is executing JS from Java using Graal faster than running actual Java code? Is it just faster than Nash…

The Graal compiler is compiling Java code to faster machine code on top of the JVM. Also GraalVM's JavaScript implementation uses an entirely different approach by using Truffle for its implementation making it significantly faster. Also GraalVM ships with Node.js support, tooling and other languages. Here is a full list of reasons: http://www.graalvm.org/docs/why-graal/ Here are some (maybe a bit outdated, but indep…

From the provided link:

- graal has superior compiler technology based off of a research paper [1] that:

-- has superior ability to remove costly object allocations in many scenarios

-- has better inlining and more aggressive speculative optimizations

-- has smarter jit compilation

[1] http://www.ssw.uni-linz.ac.at/Research/Papers/Stadler14/Stad... (pdf)

Re: GraalVM: Run Programs Faster Anywhere

#44

Post by discreteevent » Tue, Apr 17, 2018, 9:42 PM UTC

zackbloom wrote: post #30

How is this better than compiling everything to WebAssembly and using V8 as your interpreter?

We'll one reason would be that it looks like it actually works with a lot of languages right now. Do we have any idea when ruby will work in V8?

BTW. In my experience so far, when it comes to software - a bird in the hand is worth ten in the bush

Re: GraalVM: Run Programs Faster Anywhere

#45

Post by jashmatthews » Tue, Apr 17, 2018, 9:50 PM UTC

zackbloom wrote: post #30

How is this better than compiling everything to WebAssembly and using V8 as your interpreter?

Many optimizations need to be performed at a higher level, using language details. By compiling something to WebAssembly you lose all the ability to do that.

For dynamic languages like JS you can’t actually make these optimizations AOT, which is why we JIT JS in the first place.

Re: GraalVM: Run Programs Faster Anywhere

#46

Post by jahlqvis » Tue, Apr 17, 2018, 9:50 PM UTC

Does this mean I can convert my C++ win32 desktop app to run on virtually all platforms?

Re: GraalVM: Run Programs Faster Anywhere

#47

Post by royjacobs » Tue, Apr 17, 2018, 9:52 PM UTC

This is wonderful. I'm looking forward to playing around with JavaFX. Being able to compile that to a native image is huge! It would effectively invalidate the Electron approach of building "native" applications.

Re: GraalVM: Run Programs Faster Anywhere

#48

Post by gcommer » Tue, Apr 17, 2018, 9:55 PM UTC

polskibus wrote: post #40

To what extent is this compatible with all existing libraries for allegedly supported languages? I'm also curious if .net standard/c# support is being planned.

I tested it out with some fairly dependency-heavy node JS projects and it worked great (albeit with a human-noticeable perf difference vs node). R projects suffer the usual R-on-JVM issues, see eg https://github.com/oracle/fastr#status-and-limitations . I could, however, use ggplot2 and dplyr, which I don't recall working the last time I tried R-on-JVM. So that's awesome.

Being able to reasonably use R stuff from clojure (and other languages better for general purpose programming) is my data science dream. Though I can't say I'm sad with my current unix-y solutions.

Re: GraalVM: Run Programs Faster Anywhere

#49

Post by earenndil » Tue, Apr 17, 2018, 10:04 PM UTC

jahlqvis wrote: post #46

Does this mean I can convert my C++ win32 desktop app to run on virtually all platforms?

Probably not. Win32 itself is a windows-only API -- although wine is a thing.

Re: GraalVM: Run Programs Faster Anywhere

#50

Post by aidenn0 » Tue, Apr 17, 2018, 10:06 PM UTC

grashalm wrote: post #16

Graal dev here. If you want to checkout all the sources/licenses of GraalVM CE. They are here: Graal: https://github.com/oracle/graal (GPLv2 with CPE) JavaScript with Node Integration: https://github.com/graalvm/graaljs (UPL - BSD license) Ruby: https://github.com/oracle/truffleruby (EPL, GPLv2, LGPL) R: https://github.com/oracle/fastr (GPLv2) Python: https://github.com/graalvm/graalpython (UPL - BSD license) LLVM/Su…

Can GraalVM target the browser?