GLM support

[elect86]

Hi,

I am really looking forward to use joml in all my projects. Most of them are ports of C/C++ codes, where glm is intensively used.

Would it be possible to integrate in joml (some of) the glm functions?

To have an idea what I am talking about you can see this or this

[httpdigest]

Hi, fellow-Hamburger :) Sure. Please do a pull request containing the functions you need and how you envision the new methods/classes to look like. Then we could discuss based on that and align it with JOML's current design.

[elect86]

hi dear ^^

So, I cloned joml and started to moving things inside modifying them in order to match your design. No extra allocation is performed.

But I am stuck because "static import declarations are not supported in -source 1.4"

I have to import Glm.mod289() inside Vector3f.

You can take a look here and here.

[httpdigest]

Thank you for your efforts!

But I am stuck because "static import declarations are not supported in -source 1.4"

You can perfectly get away without static imports. Just write Glm.mod289 then. I mean the dependency from Vector3f to the new Glm class is there anyway, isn't it? Static imports don't change anything there. They just make the source code harder to read if one wants to understand where that dependency is coming from. :)

On another note: I wouldn't call that utility class Glm btw. I would factor the relevant responsibilities out into separate classes. Two responsibilities I see currently: "Color conversion" and "Noise".

Also there is one unnecessary method Glm.floor(float) that just delegates to the JRE method. That should be removed. Another thing is: Do we really need those Glm methods operating on arbitrary float arrays, like Glm.step and Glm.subtr? And methods seemingly operating on a 4x4 matrix, in particular Glm.yawPitchRoll should also be removed. Matrix4f already supports that with the rotateXYZ/YXZ/ZYX methods.

[elect86]

Forget it, it looks like I don't need that import anymore, probabily it was added when I pasted something.

Anyway, I will leave Glm for the moment, as a container for everything that doesn't have yet a precise destination. I created a ColorSpace class. What would you put inside Noise?

I know Glm.floor(float) is a jre method but I wanted to emulate glm.. anyway, removed.

The rest of Glm still need to be "elaborated".

[httpdigest]

What would you put inside Noise?

Simplex.

The rest of Glm still need to be "elaborated".

Okay. Tell me when you have something ready and do a pull request, so we can discuss based on it.

[httpdigest]

but I wanted to emulate glm..

That should not be the goal (at least not with JOML), in my opinion. The goal should be to provide all the functionality that GLM provides, not to do a port of GLM in JOML. The difference is that with the former way we allow to remove parts we don't want/need or have duplicates of, and we allow to reorganize the code (refactor method and class names and such) and also integrate other parts of JOML in it (for example using Matrix4f and Vector3f/4f where GLM originally only used plain float[] arrays).

This is because it could be that someone else will find another math library (let's call it "Libcmath") worth having in Java/JOML and then decide to port those functions into a Java class called CMath. There would obviously be intersections of common functionality between original JOML, GLM and CMath then.

So it would be better to extract the basic functionality and responsibilities that are needed out of those libraries in separate classes inside JOML which make their responsibilities obvious by name. But of course we can and should reference where we got/ported those functions from by adding a JavaDoc comment and say "this is that function from GLM/CMath."

[elect86]

I know, I just wanted to point out why I wrote something apparently superfluous, anyway, as I said, I removed it.

However I finished, take a look. Noise.simplex() requires 26 temp variables and it looks crap , we have a couple of options, we may:

• insert a new function where there is no tmp variable in the argument list and all of them get initialized there, so that the user may choose which one he wants to utilize
• declare the tmp as static
• pass an array of tmp variables

[httpdigest]

We also have the option to expand all vector operations and only have primitive floats. This was the way I avoided having temporary vector objects in every other JOML class. The simplex one is however the most complicated.

[elect86]

Wouldn't this be slightly slower by a performance point of view?

[httpdigest]

Nah. It would even be slightly faster in the first few runs before being JIT'ted because of no method invocations. :)

[elect86]

Ok, let me rewrite them so

[elect86]

I started with ColorSpace but now I am stuck here.

The first saturation needs a tmp mat and vec4, because it uses the mat as a partial result got from the second saturation.

[httpdigest]

"inlining" is the solution. :) Just inline the saturation method below and get rid of the matrix object.

[elect86]

I still have to finish, but I am afraid you won't accept my new changes since I am allocating new float arrays. Is this ok for you or not?

[httpdigest]

We might probably just want to look for a better original implementation of simplex noise, so as not having to manually inline and expand everything. Googling a bit for "Simplex noise Java" brought this up: http://staffwww.itn.liu.se/~stegu/simplexnoise/SimplexNoise.java We may use it and attribute to its original author (as mentioned in the JavaDocs). Having one-time statically initialized arrays there is fine, because they do not add any runtime overhead for successive invocations, nor are they being used as written shared memory, which would hinder multithreading. But even better would be to change those to instance fields and let the user instantiate a "Noise" class herself when needed and live with the per-instance allocation.

[elect86]

Sincerely I already spent way too much time on this "simple" simplex, I'd prefer to not dig another class/method if possible (also because it seems that there are different types of this method).. Since this is a method I use only once in one example.. can I write it down using static vec variables like the example you googled and we postpone the discussion about the method when more serious and complex needings will raise up in the future?

[httpdigest]

I feel with you about your time. It's just that if there is a better solution and if we save future time integrating it into JOML because it already fulfills our requirements, then I'd go with that solution. But you can of course keep your implementation. I quickly added that mentioned SimplexNoise class into a new branch noise. Maybe you can use that. Would also be interesting whether and how much the results with that method differ from the GLM one and of course to benchmark them against each other.

[pellepl]

Just a heads-up if you aren't aware of it: Perlins simplex noise has a patent from 3rd dimension and up. I am not sure this patent applies for everything and haven't digged too deep into its implications. However, it seems people choose OpenSimplex noise instead for this reason. Apparently not as efficient as Simplex noise, but still is O(n^2).

[httpdigest]

This is not Perlin's version. It is Stefan Gustavson's version, which is patent-free.

[elect86]

I totally agree, Kai, it'd be interesting profiling, benchmark and so on, but given I am using it only once in one sample (over almost 300 ones), as I said, all of this will require a lot of time, time that unfortunately at the moment I don't have (and I already abused somehow). So right now I have no other choise that to postpone it.

[httpdigest]

Yes, I'm sorry. I did not mean to imply any kind of commitment to any timeline/deadline with that statement.

[elect86]

I gave a look to the class you proposed, I think there is no way I can get the same output of the glm one.

Even supposing that I get the same output, I don't get what is the difference between Gustavson's with static variables and glm's one modified to use static variables

[httpdigest]

Could you please test whether the SimplexNoise class suits you? Thanks!

[elect86]

I don't need that anymore, thanks