Jump to content
  • Advertisement


  • Content Count

  • Joined

  • Last visited

Community Reputation

604 Good

About zfvesoljc

  • Rank

Personal Information

  • Role
  • Interests

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. zfvesoljc

    R&D Trails

    I'm currently working on implementing a trail system and I have the basic stuff working. A moving object can specify a material, width, life time, and time based curves for evaluation of colour and width. For each trail point I generate two vertices, trail point is center, vertex A is offset "left" and vertex B is offset "right", half width each. A setting for minimum distance between two trail points determines how spread out they are. This works nice until width and turning angle are so "close" that one side of trail triangles starts overlapping and in case of additive shading causes ugly artefacts. So. I'm now playing with ideas on how to solve this: - do some vertex detection magic and check for overlapping, maybe discard overlapping vertices or move them close by - push both vertices on one side of trail, ie: A = point, B = point + width (instead of A = point + half_width, B = point - half_width), but I yet have to figure out how to detect that I need to do this And other solutions or tips? Forgot to mention, I'm doing mesh generation on cpu side.
  2. zfvesoljc

    need more randomization

    Thank you! Works perfectly, I just saved 4 bytes per particle, lost around 3% performance but gained a much nicer randomization. Good enough
  3. zfvesoljc

    need more randomization

    hi, i'm not sure i understand you correctly... could you please elaborate a little? i need to generate and keep 10 random values throughout the life of each particle, without actually storing 10 values.
  4. I have a particle system with the following layout: system / emitter / particle particle is driven by particle data, which contains a range of over lifetime properties, where some can be random between two values or even two curves. to maintain a smooth evaluation between two ranges, i randomize a "lerp offset" on particle init and use that value when evaluating curves. the issue is that i'm using that same offset value for all properties (10ish) and as a result i'm seeing some patterns, which i'd like to remove. The obvious way is to just add more storage for floats, but i'd like to avoid that. The other way is to generate a seed of some sort and a random table, and use that to generate 10 values, ie: start with short/integer, mask it, then renormalize to float 0-1. any other ideas?
  5. cpu powered, spike is 200 emitters/5k particles
  6. Well, this is a situation where I need such a feature Another way would be to create a combined particle-emitter object, which would act as both, but since particle is not really an object, I'd have to have a separate loop for those objects... I also saw the double indirection array as a possible solution and will probably try that first.
  7. I have a SOA particle buffer, that I'm accessing by an index (I build a temp class which has references to the vectors inside the arrays). When particles die, I collapse the structure, copying particles to keep a continues array of data. Now, for some cases, an emitter would need to keep a reference (basically an index) or a handle of some sort to a specific particle (parent particle, sub emitter relationship). Any ideas how to nicely/fast do that? Preferably I would like to have a bi-directional link, at the moment, particles keep a list of affected emitter(s), which works in a way, but I need to do a dirty way of checking if emitter is still valid. Any ideas would be appreciated.
  8. An approach I found very useful is to have flexible execution order at the high level (aka manager) system and then a second "bucket" based system for different kind of game objects/entities.
  9. zfvesoljc

    Custom editor undo/redo system

    i tried two version on two different projects:   - a particle editor which was using the command approach (3), I've ended up with list of following commands (new, delete, copy, deep copy, deep delete and modify property). A few quirks along the way, but it worked. wasn't too hard to make but the editor was quite specific. also, the undo/redo was not persistent and a crash meant that all unsaved data was gone (save early, save often)   - the second project was a generic editor (scene/prefab,etc..) which worked with json data. there I took a more generic way of saving json diff (2). the editor system was basically a local/remote server hosting a mongo database, to which a number of editor front ends could be connected, ie: you could write an editor as a webpage or as full standalone application -using websockets for communication. making the logic to handle json diff took some time, but in the end it magically worked. I managed to write two editor prototypes to try out the system and it was very promising. sadly, due to company changes the projects was halted... the most pretty part is that you can have a clumsy written editor, which leaks and crashes, but you don't care. the data is always safe, as each change is treated as json delta and inserted into a database automatically. the changes are persistent, logged and tracked. it does require some handling on the editor side, but once the system is up and running adding "new commands" is automatic.   credits to: http://www.insomniacgames.com/ron-pieket-a-clientserver-tools-architecture/ http://bitsquid.blogspot.si/ http://www.gamedev.net/topic/658905-remote-game-editor-web-based-or-native-code/
  10. zfvesoljc

    Attack of the Clones

  11. zfvesoljc

    3D Engine to analysis

    you can check godot engine
  12. zfvesoljc

    Evaluators (polynomial & graph)

    Optimizing polynomial evaluator seems possible only in a way to solve multiple evaluators at once, as mentioned before. Second "downer"  is that lookup value x has to be same for all evaluators. So, something as: class QuadPolynomEvaluator { private:    vec4 PolyConst[4]; // vec4 for polynomial of degree 4, array of 4 for 4 evaluators (e0, e1, e2, e3) public:    // @ either build time or load time    void SetEvalData(uint idx, float c0, float c1, float c2, float c3)    {       // quick and dirty way of filling components; or use "proper" store/splat/swizzle       float* data= (float*)&PolyConst;       data[ 0*4 + idx ] = c0;       data[ 1*4 + idx ] = c1;       data[ 2*4 + idx ] = c2;       data[ 3*4 + idx ] = c3;       // PolyConst[0] = e0.c0, e1.c0, e2.c0, e3.c0       // PolyConst[1] = e0.c1, e1.c1, e2.c1, e3.c1       // PolyConst[2] = e0.c2, e1.c2, e2.c2, e3.c2       // PolyConst[3] = e0.c3, e1.c3, e2.c3, e3.c3    }    vec4 Evaluate(float x)   {     vec4 vx = vec4(x); // result.x = (e0.c0) + (e0.c1 * x) + (e0.c2 * pow(x, 2.0f)) + (e0.c2 * pow(x, 3.0f));     vec4 result = PolyConst[0];     result = vec4add( result, vec4mul( PolyConst[1], vx ) );     result = vec4add( result, vec4mul( PolyConst[2], vec4pow( vx, vec4(2.0f) ) ) );     result = vec4add( result, vec4mul( PolyConst[3], vec4pow( vx, vec4(3.0f) ) ) );     return result;   } }; Makes sense?
  13. I'm trying to design (and hopefully implement later on) an evaluator system which would work either with polynomial based data or either with point based graph data ( list of [x,y] pairs from which interpolated values are calculated). So, for quick and dirty fade-in/out one can use simple polynomial data to quickly generate linear/exponential/... curves, while graph data points can be used for complex "animations". Evaluators are used by particle and audio system and based on profiling, they take a healthy chunk of the update time.   I already have a working point based evaluator: - supports arbitrary number of points (data ptr + count) - basic logic is that based on input relative x value, I have to find corresponding x-1 and x+1 pairs, and interpolate y between them - float points are compressed into words to reduce data size and to speed up look-up - constant / linear / step / bezier    What I don't like about it: - data layout; even though data for multiple evaluators is packed together, it is still in two places (eval object + data*) - the "find neighboring  points" is currently a for loop, but as points can be unequally spread out, I don't have any other sane way to find them and since number of points is actually low (4-16) I didn't bother with anything but linear search - logic & data layout; no sensible way to simd optimize this   Initially I had a static array of points but with a large number of evaluators the data grew fast. The data* + count solved that by a large factor. All in all, this is an ok-ish solution, I'm just wondering if there is a better way to handle it (for simd). Regarding the simd optimization, it seems that optimizing one evaluator is not really practical (or even "possible"). It seems more logical to try to run multiple evaluators at once, but to be honest I'm kinda stuck here...   The new polynomial part would be easy to add, the number of constants (degree) would be low, probably 4 or so, so a float vector would be enough to store data (c0 + c1*x + c2*xx + c3*xxx). The data would be used exclusively (its either polynomial or point list).  So, the big question is how to combine/structure both parts of data and how to make the logic simd-able?
  14. zfvesoljc

    redesigning the sse

      Sometimes you just need to rearrange the data. The benefits of simd ops do no come from a1*a2*a3*a4 but rather from: v1(a1,b1,c1,d1); v2(a2,b2,c2,d2); v3(a3,b3,c3,d3); v4(a4,b4,c4,d4); v5 = vec4mul( v1, vec4mul( v2, vec4mul( v3, v4 ) ) ); // v5( a1*a2*a3*a4, b1*b2*b3*b4, c1*c2*c3*c4, d1*d2*d3*d4 )
  • Advertisement

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

GameDev.net is your game development community. Create an account for your GameDev Portfolio and participate in the largest developer community in the games industry.

Sign me up!