I've started working on a ray tracer recently, and even though I got it working pretty well, I soon ran into some trouble. I'm trying to simulate a park, and I basically have one single DistanceFromPoint(vec3 p) function that calculates the distance of a point from the nearest object, and inside that function, I call other functions like DistanceToBench() to give me distance to individual objects, whose min I use to find the minimum distance. All my individual functions like DistanceToBench first do a AABB test, and if it passes that, they proceed to do a more complete check. The part where I'm running into trouble is that if I have 1 object, it takes me nearly 70 ms or something to render one frame. If I have 2, something like 200. If I have 3, it goes up to 500 or something, and 4 objects or more simply crashes the entire scene (something like 15-20s per frame). Same thing if I try to use shadows - if I have one object, shadows increase the render time from 70 ms to something like 180 ms, but the moment I add 2 or 3 objects, I cannot render the scene with shadows anymore without crashing the whole thing. Here is part of my actual code.
float u = (pos.x - 0.5) * 2 * 1.77;
float v = (pos.y - 0.5) * 2;
vec3 ro = eye;
vec3 rd = normalize(forward * f + right * u + up * v);
int materialID;
float dist;
vec3 color = skyColor;
if (CastRay(ro, rd, dist, materialID))
{
vec3 N = NormalAtPoint(ro + rd*dist);
vec3 L = -lightDir;
vec3 V = -rd;
vec3 H = normalize(V + L);
float diffuseTerm = saturate(dot (N, L));
float aoTerm = AOAtPoint(ro + rd*dist, N);
#ifndef SHADOWS
float shadowTerm = 1;
#else
float shadowTerm = mix(0.5, 1, ShadowAtPoint(ro + rd*dist, L, materialID));
#endif
color = mix(materialDarkColor[materialID], materialLightColor[materialID], diffuseTerm) * aoTerm * shadowTerm;
}
color = ApplyFog(rd, dist, color, -lightDir);
The main function used here is CastRay, that simply moves from ro along rd until it reaches far plane, or collides with an object (which I test using the before-mentioned DistanceFromPoint()).
So my question is, is there any way to stop the program from crashing when number of objects slightly increase? I've read about partitioning the world into grids and placing objects in grids, but since my objects aren't exactly going to be in grid - and because my world will contain something like 7-8 objects, all of which will have AABB test - I decided to skip using that. Other than that, I don't know how else to speed up this thing. Any tips for handling the situation would be appreciated.
Also, I've read that shaders don't handle branching very well, since all threads must execute the same instructions at a given point of time. If that's the case, would I be better off using CPU to do the rendering instead of GPU, since there are so many branches in my code?
