Vc VS DOOM

Most high-performance math libraries perform a lot of vectorization (Eigen, etc) under the hood. And you've got stuff like Klein, Vc (which is reminiscent of std::valarray), etc. Then there's OpenMP's #pragma omp simd (assuming version 4.0 or greater).

I personally like the ideas in Parallelism v2 TS, which is available in for libstdc++ 11 onwards. The reference implementation is a library named Vc (afaik Vc is the most popular SIMD library for C++), and this has also been implemented in recent versions of HPX.

First of all about our project, I previously decided to work with VC library.https://github.com/VcDevel/Vc

I've decided to switch to something better, and after a few hours of searching, I found this repository: NSIMD https://github.com/agenium-scale/nsimd FastDifferentialCoding https://github.com/lemire/FastDifferentialCoding VS https://github.com/VcDevel/Vc XSIMD https://github.com/xtensor-stack/xsimd

> - Waiting for Cross-Platform standardized SIMD vector datatypes

which language has standardized SIMD vector datatypes ? most languages don't even have any ability to express SIMD while in C++ I can just use Vc (https://github.com/VcDevel/Vc), nsimd (https://github.com/agenium-scale/nsimd) or one of the other ton of alternatives, and have stuff that JustWorksTM on more architectures than most languages even support

- Using nonstandard extensions, libraries or home-baked solutions to run computations in parallel on many cores or on different processors than the CPU

what are the other native languages with a standardized memory model for atomics ? and, what's the problem with using libraries ? it's not like you're going to use C# or Java's built-in threadpools if you are doing any serious work, no ? Do they even have something as easy to use as https://github.com/taskflow/taskflow ?

- Debugging cross-platform code using couts, cerrs and printfs

because people never use console.log in JS or System.println in C# maybe ?

- Forced to use boost for even quite elementary operations on std::strings.

can you point to non-trivial java projects that do not use Apache Commons ? Also, the boost string algorithms are header-only so you will end up with exactly the same binaries that if it was in some std::string_algorithms namespace:

https://gcc.godbolt.org/z/43vKadbde

commercially exploit or use for any commercial purpose."

[1] https://github.com/id-Software/DOOM/commit/4eb368a960647c8cc...

The original Doom had third-party audio playback routines, so the source came with a rewritten sound server: https://github.com/id-Software/DOOM/tree/master/sndserv

    The bad news:  this code only compiles and runs on linux.  We couldn't

Here's a repo for you with no test coverage and no auto-generated DI. They using unsafe pointers all over the place, too!

https://github.com/id-Software/DOOM

Shall I prepare the postage for the letter in which you'll call John Carmack an MBA? Should we send another to Chris Sawyer? I heard he didn't even write a formal design doc for Roller Coaster Tycoon!

cURL is one of the most used C libs and is an example of good quality C code. If you follow the style used there, see e.g. https://github.com/curl/curl/blob/master/lib/dynhds.h (and associated dynhds.c) you will be good.

Looking at the source of some of the old game-engines from the era that have since been released as open-source can also be helpful, like https://github.com/id-Software/DOOM.

In both cases notice how simple and elegant a lot of the code is. There is already enough complexity inherent in the problem they are solving, and that is where the focus should be.

Any IDE with a working language server to make it easy to jump around and refactor should work fine. Limitations might be due to the C language itself?

Error handling on such a fixed platform does not need to be super-advanced. You should always be within the confines of the system so there shouldn't be much that can go wrong. If stuff goes wrong anyway just being able call a function Fatal("FooBar failed with code 34") when unexpected stuff happens and have it log somewhere to be able to dig around should be enough. You never need to be able to recover and retry.

Make sure to use https://clang.llvm.org/docs/AddressSanitizer.html or a similar tool when developing outside of the PSOne.

That said, consider statically allocating global buffers for most stuff and avoid using the heap for most stuff.

Good luck working within the confines of the PSOne! Many hackers have pulled the hair from their head on that platform ;)

Probably more easily than you'd think. DOOM is open source[1], and as GP alludes, is probably the most frequently ported game in existence, so its source code almost certainly appears multiple times in GPT-4's training set, likely alongside multiple annotated explanations.

[1] https://github.com/id-Software/DOOM

I'd define an arena as the pattern where the arena itself owns N objects. So you free the arena to free all objects.

My first job was at EA working on console games (PS2, GameCube, XBox, no OS or virtual memory on any of them), and while at the time I was too junior to touch the memory allocators themselves, we were definitely not malloc-ing and freeing all the time.

It was more like you load data for the level in one stage, which creates a ton of data structures, and then you enter a loop to draw every frame quickly. There were many global variables.

---

Wikipedia calls it a region, zone, arena, area, or memory context, and that seems about right:

https://en.wikipedia.org/wiki/Region-based_memory_management

It describes history from 1967 (before C was invented!) and has some good examples from Apache ("pools") and Postgres ("memory contexts").

I also just looked at these codebases:

https://github.com/mit-pdos/xv6-public (based on code from the 70's)

https://github.com/id-Software/DOOM (1997)

I looked at allocproc() in xv6, and gives you an object from a fixed global array. A lot of C code in the 80's and 90's was essentially "kernel code" in that it didn't have an OS underneath it. Embedded systems didn't run on full-fledges OSes.

DOOM tends to use a lot of what I would call "pools" -- arrays of objects of a fixed size, and that's basically what I remember from EA.

Though in g_game.c, there is definitely an arena of size 0x20000 called "demobuffer". It's used with a bump allocator.

---

So I'd say

- malloc / free of individual objects was NEVER what C code looked like (aside from toy code in college)

- arena allocators were used, but global vars and pools are also very common.

- arenas are more or less wash for memory safety. they help you in some ways, but hurt you in others.

The reason C programmers don't malloc/free all the time is for speed, not memory safety. Arenas are still unsafe.

When you free an arena, you have no guarantee there's nothing that points to it anymore.

Also, something that shouldn't be underestimated is that arena allocators break tools like ASAN, which use the malloc() free() interface. This was underscored to me by writing a garbage collector -- the custom allocator "broke" ASAN, and that was actually a problem:

https://www.oilshell.org/blog/2023/01/garbage-collector.html

If you want memory safety in your C code, you should be using ASAN (dynamically instrumented allocators) and good test coverage. Arenas don't help -- they can actually hurt. An arena is a trivial idea -- the problem is more if that usage pattern actually matches your application, and apps evolve over time.

Doom