SortingNetworks VS JDK

On a side-note: I did an experiment to see whether C# could match C++ for vector-intensive computing: https://github.com/zvrba/SortingNetworks

Here's a concrete example of using pointers to access raw array memory and use SIMD intrinsics: https://github.com/zvrba/SortingNetworks

And now, for the fun of it, you can try with sorting. I've already done the hard work in C# (AVX2 intrinsics): https://github.com/zvrba/SortingNetworks

> I can't read C#

Not much different than C++...

> Do you generate the sorting network at compile time

No, except for power of two sizes up to 32. I experimented with run-time code generation (and compilation) for given sizes, but... the generated machine code has too long prologue and epilogue for that to be worth-while (though the sorting code itself is well optimized, as if directly compiled from source). That's also mentioned in "Benchmarks" section.

> What's your sorting network template?

See References.

> And probably related: how is vectorization used?

See the code. There's no template, the code is fully "dynamic" and adapts itself to array size. As for vectorization... it compares/swaps 8 ints/floats at once, with some swizzles to rearrange the elements. For sizes that are not power of 2, I use masked loads and stores and some extra logic for deciding which comparisons to skip. (I treat non-existing elements "as if" they were set to intmax or float infinity.)

This file https://github.com/zvrba/SortingNetworks/blob/master/Sorting... has it all.

> this week-end project

Sorry, can't read Rust. (Though it reminds me of days spent coding in Perl.) Most networks are not SIMD-friendly and the code as it's now is the 3rd iteration where I figured out how to best leverage SIMD to exploit the recursiveness and regularity in the network. (Not the least, no random memory accesses: only forward and backward loads and stores.)

Without SIMD, I don't think it'll be worth it, because network will also access the memory randomly (just as "standard" sort), and in addition it has worse algorithmic complexity.

So recently I posted a link with code for fast sorting of int arrays. People wondered how they'd perform for large arrays (1M elements), and I conjectured they'd be way slower because of their algorithmic complexity. Turns out I was wrong, they're 3-6x faster for arrays of length up to 1M elements. Updated code and benchmarks are now available at https://github.com/zvrba/SortingNetworks

The code (MIT license) is available here: https://github.com/zvrba/SortingNetworks

I felt out of the loop, thinking that Zero VM was some kind of new distro for OpenJDK but chasing <https://packages.debian.org/sid/openjdk-22-jre-zero#:~:text=...> to <https://sources.debian.org/src/openjdk-11/11.0.23%2B9-1/debi...> lead me to https://github.com/openjdk/jdk/tree/jdk-22-ga/src/hotspot/cp...

It seems that it's a specific CPU target for the Hotspot JIT for non-mainstream architectures (or for research purposes, as I saw mentioned once)

Completely gutted from the OpenJDK, last I checked. See here for the culprit PR: https://github.com/openjdk/jdk/pull/18688

> Yes, they're changing one aspect of signal handler use to work around this problem. They're not stopping the use of signal handlers in general. Hotspot continues to use signals for efficiency in general. See https://github.com/openjdk/jdk/blob/9059727df135dc90311bd476...

This whole thread is about SIGSEGV, and specifically their SIGSEGV handling. However, catching normal signals is not about efficiency.

Some of their exception handling is still odd: There is no reason for a program that receives SIGILL to ever attempt continuing. But others is fine, like catching SIGFPE to just forward an exception to the calling code.

(Sure, you could construct an argument to say that this is for efficiency if you considered the alternative to be implementing floating point in software so that all exceptions exist in user-space, but hardware floating point is the norm and such alternative would be wholly unreasonable.)

> The wonderful thing about choosing not to care about facts is having whatever opinions you want.

I appreciate the irony of you making such statement, proudly thinking that your opinion equals fact, and therefore any other opinion is not.

This discussion is nothing but subjective opinion vs. subjective opinion. Facts are (hopefully, as I can only speak for myself) inputs to both our opinions, but no opinion about "good" or "bad", "nasty" or not can ever be objective. Objective code quality does not exist.

I remember talking to Brendan about the PreserveFramePointer patch during my first months at Netflix in 2015. As of JDK 21, unfortunately it is no longer a general purpose solution for the JVM, because it prevents a fast path being taken for stack thawing for virtual threads: https://github.com/openjdk/jdk/blob/d32ce65781c1d7815a69ceac...