Freebasic Procedural Terrain engine

[shadow008]

Since my last post in "perlin noise" thread www.basic4gl.proboards.com/thread/3538/perlin-noise , I've been working on learning freebasic. More specifically, using it to port some existing code from basic4gl to freebasic for some speed increases. Aside from that, if DJLinux would give a quick example of how he's compiling his DLL's written in freebasic, I'd really like to bring a few functions from the opengl 3.3 specs to the openGLplugin, because there's something to be said about vertex attributes

Here's the project thus far:
dl.dropboxusercontent.com/u/5319168/InfiniteTerrain.zip

Features:
-"infinite terrain"
-Basic multithreading
-Completely opengl 3.3 compliant: Uses no depreciated functions (that I'm aware of)
-Easily modifiable LOD. In fact nearly every aspect of the terrain generation is modifiable.
-Bare bones example of a GLSL 1.50 shader in use.
-EXE fits in under 300kb
-Spaghetti code.

Plan to add:
-Un-spaghettied code.
-Lighting and texturing (texturing done on the gfx card)
-Camera view frustum culling (Currently only supports camera frustum sphere culling)
-Terrain Generation Controller: Using macros to reconfigure how the noise is generated per step.
-Procedurally generated stuff like rocks and foliage (maybe multithread that because I can)
-Maybe delve more into x86 assembly and see if I can get the noise generating even faster (Because you can never be "too fast") Also note that the terrain, as it's rendered in the .exe as it sits, is pushing 41*41*64*64+overspill or about 7,000,000 points total (in just under 6 sec on my desktop).
-Skirting the heightmap. It has to happen... It's just going to be a pain with indexing everything correctly.
-Fix the glitch related to multithreading where sometimes a sections Vertex Array Object(VAO) will be generated while the vertex data is being modified, resulting in missing sections. Reloading that area fixes it, but it's still annoying. It happens when the camera is moving wayyy to fast.
-Fix the glitch where going too fast with the camera completely destroys the terrain update process. (Maybe just reset everything when that's caught... I dunno right now)
-Implement a physics engine: Been looking at DJLinux's PhysX wrapper, and it seems to be exactly what I think will work.

-More things later when I get something that looks cool.

There's also an example of 2d simplex noise that I got working, and should port to b4gl because it's more than twice as fast as standard perlin noise with the same (better?) results.

If you wanna run the test with something like FRAPS or Dxtory or something to view the framerate and report that (or blatantly obvious visual errors), that'd be appreciated.

If anyone want's in on this, I'm more than determined to keep it up.

Anyways, happy coding ＼(^^＼)(/^^)/

[matthew]

Regarding the time that the program takes to run, it took just over a minute or so before anything appeared in the window on the screen.

I could only try it once though because during execution it slowed my computer down to a crawl. :-(

I hope you can speed it up. :-D

[shadow008]

Could you by any chance give me your computer specs? I've only tested it on quad core hardware with AMD graphics cards, but it should only require a dual core.

And since that first post, I've also somehow managed to corrupt all compatibility with every other computer aside from my desktop. This compatibility thing is hard...

[matthew]

Yeah, according to CCleaner my Computer Specs are as follows... AMD C-70 APU with Radeon HD Graphics, 2.0GB RAM, AMD HD 7290 Graphics.

[shadow008]

Ah, that's probably your ram. Because the data is being stored on both the graphics memory and buffered in system memory, it's taking up quite a bit of space (I'll fix that later). But since you're using an APU, system mem and graphics mem are the same thing, and your 2GB probably overflowed and started paging memory into your hard drive... That will usually halt any computer to a crawl.

There are some parameters in code that will reduce the scale of the heightmap, if you have the FB compiler installed.

But, the shader compiled and it DID run, right? Cool!

[DJLinux]

It's works here with 1GB PC RAM and 1GB GFX RAM and is quit fast.
Only on end (after freeing some resoures) it crashed looks like you modify my OpenGL include file.

DJ

[shadow008]

Hey guys. It's spring break and I'm thinking on getting some progress done on this.

Only progress made thus far is implementation of the terrain generation controller. Which I think should instead be replaced by a decision tree...

But I'm posting mostly because I've gotten a nice implementation of worley noise (Cellular noise, voronoi noise...)

Screen shot:


Now, it's got a slight repetitive look to it, but that's because there's only 1 randomized point per computation cell.
However, that 1 point per cell allows for really fast real-time utilization of the function, and nothing to be stored in memory.

I guess people use this function for "biomes" (or whatever you want to call them), so that's worth looking into.

Source Code:
#include once "crt.bi"

#define GLOBAL_SEED 90
#define RND_MAX     32767

const NOISE_MAX as integer = 10000

const WORLEY_NOISE_GRID_SIZE as integer = 64

declare function noise(ByVal x as single, ByVal y as single) as single

'Returns a random value between 0 and NOISE_MAX
function noise(ByVal x as single, ByVal y as single) as single

   dim n as integer = (x SHL 13) * GLOBAL_SEED + y * GLOBAL_SEED
   return ((n * (n * n * 15731 + 789221) + 1376312589) AND &H7fffffff) MOD NOISE_MAX

end function

dim shared NeighborMap(8, 1) as single = {{0,0}, {1,0}, {0,1}, {-1,0}, {0,-1},_
                                         {1,1}, {-1,1},{-1,-1},{1, -1}}

function WorleyNoise2D(ByVal x as single, ByVal y as single) as single
   
   dim i as integer

   dim mainX as integer
   dim mainY as integer
   
   dim gridX as integer
   dim gridY as integer

   dim CellPoint(1) as single
   
   dim dist as single
   dim lastdist as single
   
   dim distValX as single
   dim distValY as single
   
   dim distToEdge as byte = 0
   
   mainX = (x / WORLEY_NOISE_GRID_SIZE)
   mainY = (y / WORLEY_NOISE_GRID_SIZE)
   
   for i = 0 to 8
       gridX = mainX + NeighborMap(i,0)
       gridY = mainY + NeighborMap(i,1)
       
       '1.41... ~= sqrt(2), or the ratio between a
       'squares side length and it's diagonal length.
       CellPoint(0) = (Noise(gridX, gridY) / NOISE_MAX) / 1.41 * WORLEY_NOISE_GRID_SIZE
       CellPoint(1) = (Noise(gridY, gridX) / NOISE_MAX) / 1.41 * WORLEY_NOISE_GRID_SIZE

       distValX = x - ((gridX * WORLEY_NOISE_GRID_SIZE) - CellPoint(0))
       distValY = y - ((gridY * WORLEY_NOISE_GRID_SIZE) - CellPoint(1))

       dist = sqr( distValX * distValX + distValY * distValY )

       if i = 0 then
           lastDist = dist
           
           'Optimization: If the first cell's point is closer than the
           'closest boarders, jump out here and save 7 iterations.
           
           'Start with the edges on the X axis
           if x MOD WORLEY_NOISE_GRID_SIZE > (.5 * WORLEY_NOISE_GRID_SIZE) then
               'Working with the input x point being closer to the right edge
               if distValX <= x - ((gridX + 1) * WORLEY_NOISE_GRID_SIZE) then
                   distToEdge += 1
               endif
           else
               if distValX <= x - (gridX * WORLEY_NOISE_GRID_SIZE) then
                   distToEdge += 1
               endif
           endif
           
           if y MOD WORLEY_NOISE_GRID_SIZE > (.5 * WORLEY_NOISE_GRID_SIZE) then
               if distValY <= y - ((gridY + 1) * WORLEY_NOISE_GRID_SIZE) then
                   distToEdge += 1
               endif
           else
               if distValY <= y - (gridY * WORLEY_NOISE_GRID_SIZE) then
                   distToEdge += 1
               endif
           endif
           
           if distToEdge > 1 then
               return 1 - (lastDist / WORLEY_NOISE_GRID_SIZE)
           endif
       else
           if dist < lastDist then
               lastDist = dist
           endif
       endif
       
   next

   return 1 - (lastDist / WORLEY_NOISE_GRID_SIZE)
   
end function

That's all I've got for now.

~Shadow008

PS: THEY FIXED THE CODE TAG! It's so beautiful being all tabbed and stuff...