2D or not 2D (in the C++ Standard) – that is the question

Re:2D or not 2D (in the C++ Standard) – that is the question by exDM69 on Monday June 11, 2018 @04:10AM (#17282632)

What the C++ (and C) stadards could use is a native type for SIMD vectors. In other words, standardize the GCC vector extensions [0] or some variant thereof.

Just adding a new header-only library for vector arithmetic isn't something that needs to be in the language standard (library), there are lots of options for that already (such as glm[1]).

I've been using vector extensions for my 3d graphics and physics stuff for years now and it works great. There are some pain points, but it already covers most of what I need. As a bonus, it can be used in plain C and you get all the usual math operators (+, -, /, etc) for free. There are minor incompatibilities between GCC and Clang, but nothing a few #ifdefs won't solve.

Here's a quick introduction in a few lines of code:

    typedef float vec4f __attribute__((vector_size(16)));
    vec4f a = { 1, 2, 3, 4 }, b = { 5, 6, 7, 8 };
    vec4f c = a + b;
    vec4f d = (a + b) * c; // with --ffast-math, you will get fused multiply-and-add (MAD)

In my projects, I've always put together an ad-hoc collection of arithmetic routines I need (e.g. dot, cross, matrix product, inverse, etc). A proper, well tested library of basic math operations for SIMD vectors would be very useful but I haven't got the time it takes.

If anyone is interested, some of my math routines are bundled in this github repo [2]. It's incomplete, undocumented, not well tested and there are known issues I haven't gotten around to (e.g. quaternion products give wrong results).

If anyone thinks this is useful and is interested in contributing/collaborating, leave a comment.

[0] https://gcc.gnu.org/onlinedocs/gcc/Vector-Extensions.html

[1] https://glm.g-truc.net/0.9.9/index.html

[2] https://github.com/rikusalminen/threedee-simd

[ Parent | on HN ]

Re:2D or not 2D (in the C++ Standard) – that is the question by mattnewport on Monday June 11, 2018 @11:15AM (#17284938)

There are some efforts underway to standardize using SIMD but this is really a separate problem to providing a short vector math library suitable for the kind of linear algebra useful for 3D graphics and physics.

Making your vec4f type a SIMD register type is not a good way to make use of SIMD hardware. The best approach is generally an SoA (Struct of Arrays) rather than an AoS (Array of Structs) and the best CPU programming model for that is something like ispc[1].

Nathan Reed has a good summary of why "naïve" SIMD libraries after not the way to go for short vector math: http://www.reedbeta.com/blog/on-vector-math-libraries/#how-t...

On some platforms alignment issues become a big problem for naïve vector math libraries too, though this is less of an issue on current generation x64 hardware.

[1] https://ispc.github.io/ispc.html

[ Parent | on HN ]

Re:2D or not 2D (in the C++ Standard) – that is the question by phkahler on Monday June 11, 2018 @11:12AM (#17284919)

This is exactly what I'm talking about. I want these vectors as standard types. First class data types, not part of a library or typedef. The example given in the parent comment is spot on with typical use.

I like that GCC defined their own way to declare these that is independent of x86, ARM, and Altivec intrinsics so your code is portable (so long as you use GCC). Geometrically relevant vectors are common enough to be native types, let's get there.

[ Parent | on HN ]

Re:2D or not 2D (in the C++ Standard) – that is the question by exDM69 on Monday June 11, 2018 @12:45PM (#17285940)

> so your code is portable (so long as you use GCC)

Or Clang!

There are only minute differences between vector extensions in GCC vs. Clang. Little things like __builtin_shuffle vs. __builtin_shufflevector.

[ Parent | on HN ]