clray

annotate rt.cl @ 28:97cfd9675310

find changesets by author, revision, files, or words in the commit message
trying to pass the kdtree to the kernel
author John Tsiombikas <nuclear@member.fsf.org>
date Sat, 21 Aug 2010 03:42:49 +0100
parents 51f115e337c2
children 353d80127627
rev   line source
nuclear@12 1 /* vim: set ft=opencl:ts=4:sw=4 */
nuclear@12 2
nuclear@2 3 struct RendInfo {
nuclear@22 4 float4 ambient;
nuclear@2 5 int xsz, ysz;
nuclear@9 6 int num_faces, num_lights;
nuclear@2 7 int max_iter;
nuclear@28 8 int kd_depth;
nuclear@2 9 };
nuclear@2 10
nuclear@9 11 struct Vertex {
nuclear@2 12 float4 pos;
nuclear@9 13 float4 normal;
nuclear@12 14 float4 tex;
nuclear@12 15 float4 padding;
nuclear@9 16 };
nuclear@9 17
nuclear@9 18 struct Face {
nuclear@9 19 struct Vertex v[3];
nuclear@9 20 float4 normal;
nuclear@9 21 int matid;
nuclear@12 22 int padding[3];
nuclear@9 23 };
nuclear@9 24
nuclear@9 25 struct Material {
nuclear@5 26 float4 kd, ks;
nuclear@9 27 float kr, kt;
nuclear@9 28 float spow;
nuclear@12 29 float padding;
nuclear@2 30 };
nuclear@2 31
nuclear@3 32 struct Light {
nuclear@3 33 float4 pos, color;
nuclear@3 34 };
nuclear@3 35
nuclear@2 36 struct Ray {
nuclear@2 37 float4 origin, dir;
nuclear@2 38 };
nuclear@2 39
nuclear@2 40 struct SurfPoint {
nuclear@2 41 float t;
nuclear@12 42 float4 pos, norm, dbg;
nuclear@9 43 global const struct Face *obj;
nuclear@19 44 struct Material mat;
nuclear@2 45 };
nuclear@2 46
nuclear@16 47 struct Scene {
nuclear@16 48 float4 ambient;
nuclear@16 49 global const struct Face *faces;
nuclear@16 50 int num_faces;
nuclear@16 51 global const struct Light *lights;
nuclear@16 52 int num_lights;
nuclear@16 53 global const struct Material *matlib;
nuclear@28 54 global const struct KDNode *kdtree;
nuclear@28 55 };
nuclear@28 56
nuclear@28 57 struct AABBox {
nuclear@28 58 float4 min, max;
nuclear@28 59 };
nuclear@28 60
nuclear@28 61 struct KDNode {
nuclear@28 62 AABBox aabb;
nuclear@28 63 int face_idx[32];
nuclear@28 64 int num_faces;
nuclear@28 65 int padding[3];
nuclear@16 66 };
nuclear@2 67
nuclear@16 68 #define MIN_ENERGY 0.001
nuclear@21 69 #define EPSILON 1e-5
nuclear@16 70
nuclear@16 71 float4 shade(struct Ray ray, struct Scene *scn, const struct SurfPoint *sp);
nuclear@16 72 bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *sp);
nuclear@9 73 bool intersect(struct Ray ray, global const struct Face *face, struct SurfPoint *sp);
nuclear@28 74 bool intersect_aabb(struct Ray ray, struct AABBox aabb);
nuclear@16 75
nuclear@8 76 float4 reflect(float4 v, float4 n);
nuclear@8 77 float4 transform(float4 v, global const float *xform);
nuclear@16 78 void transform_ray(struct Ray *ray, global const float *xform, global const float *invtrans);
nuclear@12 79 float4 calc_bary(float4 pt, global const struct Face *face, float4 norm);
nuclear@19 80 float mean(float4 v);
nuclear@4 81
nuclear@4 82 kernel void render(global float4 *fb,
nuclear@4 83 global const struct RendInfo *rinf,
nuclear@9 84 global const struct Face *faces,
nuclear@9 85 global const struct Material *matlib,
nuclear@4 86 global const struct Light *lights,
nuclear@7 87 global const struct Ray *primrays,
nuclear@12 88 global const float *xform,
nuclear@28 89 global const float *invtrans,
nuclear@28 90 global const struct KDNode *kdtree)
nuclear@2 91 {
nuclear@2 92 int idx = get_global_id(0);
nuclear@2 93
nuclear@16 94 struct Scene scn;
nuclear@16 95 scn.ambient = rinf->ambient;
nuclear@16 96 scn.faces = faces;
nuclear@16 97 scn.num_faces = rinf->num_faces;
nuclear@16 98 scn.lights = lights;
nuclear@16 99 scn.num_lights = rinf->num_lights;
nuclear@16 100 scn.matlib = matlib;
nuclear@8 101
nuclear@16 102 struct Ray ray = primrays[idx];
nuclear@16 103 transform_ray(&ray, xform, invtrans);
nuclear@4 104
nuclear@19 105 float4 pixel = (float4)(0, 0, 0, 0);
nuclear@22 106 float4 energy = (float4)(1.0, 1.0, 1.0, 0.0);
nuclear@19 107 int iter = 0;
nuclear@19 108
nuclear@19 109 while(iter++ < rinf->max_iter && mean(energy) > MIN_ENERGY) {
nuclear@19 110 struct SurfPoint sp;
nuclear@19 111 if(find_intersection(ray, &scn, &sp)) {
nuclear@19 112 pixel += shade(ray, &scn, &sp) * energy;
nuclear@19 113
nuclear@19 114 float4 refl_col = sp.mat.ks * sp.mat.kr;
nuclear@19 115
nuclear@19 116 ray.origin = sp.pos;
nuclear@19 117 ray.dir = reflect(-ray.dir, sp.norm);
nuclear@19 118
nuclear@19 119 energy *= sp.mat.ks * sp.mat.kr;
nuclear@19 120 } else {
nuclear@19 121 iter = INT_MAX - 1; // to break out of the loop
nuclear@19 122 }
nuclear@17 123 }
nuclear@19 124
nuclear@19 125 fb[idx] = pixel;
nuclear@4 126 }
nuclear@4 127
nuclear@16 128 float4 shade(struct Ray ray, struct Scene *scn, const struct SurfPoint *sp)
nuclear@16 129 {
nuclear@16 130 float4 norm = sp->norm;
nuclear@12 131 bool entering = true;
nuclear@12 132
nuclear@12 133 if(dot(ray.dir, norm) >= 0.0) {
nuclear@12 134 norm = -norm;
nuclear@12 135 entering = false;
nuclear@12 136 }
nuclear@12 137
nuclear@19 138 float4 dcol = scn->ambient * sp->mat.kd;
nuclear@8 139 float4 scol = (float4)(0, 0, 0, 0);
nuclear@5 140
nuclear@16 141 for(int i=0; i<scn->num_lights; i++) {
nuclear@16 142 float4 ldir = scn->lights[i].pos - sp->pos;
nuclear@5 143
nuclear@16 144 struct Ray shadowray;
nuclear@16 145 shadowray.origin = sp->pos;
nuclear@16 146 shadowray.dir = ldir;
nuclear@5 147
nuclear@16 148 if(!find_intersection(shadowray, scn, 0)) {
nuclear@16 149 ldir = normalize(ldir);
nuclear@16 150 float4 vdir = -normalize(ray.dir);
nuclear@16 151 float4 vref = reflect(vdir, norm);
nuclear@16 152
nuclear@16 153 float diff = fmax(dot(ldir, norm), 0.0f);
nuclear@22 154 dcol += sp->mat.kd * scn->lights[i].color * diff;
nuclear@16 155
nuclear@20 156 float spec = powr(fmax(dot(ldir, vref), 0.0f), sp->mat.spow);
nuclear@22 157 scol += sp->mat.ks * scn->lights[i].color * spec;
nuclear@16 158 }
nuclear@16 159 }
nuclear@16 160
nuclear@8 161 return dcol + scol;
nuclear@2 162 }
nuclear@2 163
nuclear@28 164 bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *spres)
nuclear@28 165 {
nuclear@28 166 return false;
nuclear@28 167 }
nuclear@16 168
nuclear@28 169 /*bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *spres)
nuclear@12 170 {
nuclear@16 171 struct SurfPoint sp, sp0;
nuclear@16 172 sp0.t = 1.0;
nuclear@16 173 sp0.obj = 0;
nuclear@16 174
nuclear@16 175 for(int i=0; i<scn->num_faces; i++) {
nuclear@16 176 if(intersect(ray, scn->faces + i, &sp) && sp.t < sp0.t) {
nuclear@16 177 sp0 = sp;
nuclear@16 178 }
nuclear@16 179 }
nuclear@16 180
nuclear@16 181 if(!sp0.obj) {
nuclear@16 182 return false;
nuclear@16 183 }
nuclear@16 184
nuclear@16 185 if(spres) {
nuclear@16 186 *spres = sp0;
nuclear@19 187 spres->mat = scn->matlib[sp0.obj->matid];
nuclear@16 188 }
nuclear@16 189 return true;
nuclear@28 190 }*/
nuclear@12 191
nuclear@16 192 bool intersect(struct Ray ray, global const struct Face *face, struct SurfPoint *sp)
nuclear@2 193 {
nuclear@12 194 float4 origin = ray.origin;
nuclear@12 195 float4 dir = ray.dir;
nuclear@12 196 float4 norm = face->normal;
nuclear@12 197
nuclear@16 198 float ndotdir = dot(dir, norm);
nuclear@12 199
nuclear@9 200 if(fabs(ndotdir) <= EPSILON) {
nuclear@9 201 return false;
nuclear@9 202 }
nuclear@2 203
nuclear@9 204 float4 pt = face->v[0].pos;
nuclear@12 205 float4 vec = pt - origin;
nuclear@2 206
nuclear@16 207 float ndotvec = dot(norm, vec);
nuclear@9 208 float t = ndotvec / ndotdir;
nuclear@2 209
nuclear@2 210 if(t < EPSILON || t > 1.0) {
nuclear@2 211 return false;
nuclear@2 212 }
nuclear@12 213 pt = origin + dir * t;
nuclear@9 214
nuclear@12 215
nuclear@12 216 float4 bc = calc_bary(pt, face, norm);
nuclear@9 217 float bc_sum = bc.x + bc.y + bc.z;
nuclear@9 218
nuclear@20 219 if(bc_sum < 1.0 - EPSILON || bc_sum > 1.0 + EPSILON) {
nuclear@9 220 return false;
nuclear@12 221 bc *= 1.2;
nuclear@9 222 }
nuclear@2 223
nuclear@2 224 sp->t = t;
nuclear@9 225 sp->pos = pt;
nuclear@21 226 sp->norm = normalize(face->v[0].normal * bc.x + face->v[1].normal * bc.y + face->v[2].normal * bc.z);
nuclear@9 227 sp->obj = face;
nuclear@12 228 sp->dbg = bc;
nuclear@2 229 return true;
nuclear@2 230 }
nuclear@5 231
nuclear@28 232 bool intersect_aabb(struct Ray ray, struct AABBox aabb)
nuclear@28 233 {
nuclear@28 234 if(ray.origin.x >= aabb.min.x && ray.origin.y >= aabb.min.y && ray.origin.z >= aabb.min.z &&
nuclear@28 235 ray.origin.x < aabb.max.x && ray.origin.y < aabb.max.y && ray.origin.z < aabb.max.z) {
nuclear@28 236 return true;
nuclear@28 237 }
nuclear@28 238
nuclear@28 239 float4 bbox[2] = {aabb.min, aabb.max};
nuclear@28 240
nuclear@28 241 int xsign = (int)(ray.dir.x < 0.0);
nuclear@28 242 float invdirx = 1.0 / ray.dir.x;
nuclear@28 243 float tmin = (bbox[xsign].x - ray.origin.x) * invdirx;
nuclear@28 244 float tmax = (bbox[1 - xsign].x - ray.origin.x) * invdirx;
nuclear@28 245
nuclear@28 246 int ysign = (int)(ray.dir.y < 0.0);
nuclear@28 247 float invdiry = 1.0 / ray.dir.y;
nuclear@28 248 float tymin = (bbox[ysign].y - ray.origin.y) * invdiry;
nuclear@28 249 float tymax = (bbox[1 - ysign].y - ray.origin.y) * invdiry;
nuclear@28 250
nuclear@28 251 if(tmin > tymax || tymin > tmax) {
nuclear@28 252 return false;
nuclear@28 253 }
nuclear@28 254
nuclear@28 255 if(tymin > tmin) tmin = tymin;
nuclear@28 256 if(tymax < tmax) tmax = tymax;
nuclear@28 257
nuclear@28 258 int zsign = (int)(ray.dir.z < 0.0);
nuclear@28 259 float invdirz = 1.0 / ray.dir.z;
nuclear@28 260 float tzmin = (bbox[zsign].z - ray.origin.z) * invdirz;
nuclear@28 261 float tzmax = (bbox[1 - zsign].z - ray.origin.z) * invdirz;
nuclear@28 262
nuclear@28 263 if(tmin > tzmax || tzmin > tmax) {
nuclear@28 264 return false;
nuclear@28 265 }
nuclear@28 266
nuclear@28 267 return tmin < t1 && tmax > t0;
nuclear@28 268 }
nuclear@28 269
nuclear@8 270 float4 reflect(float4 v, float4 n)
nuclear@5 271 {
nuclear@23 272 return 2.0f * dot(v, n) * n - v;
nuclear@5 273 }
nuclear@8 274
nuclear@8 275 float4 transform(float4 v, global const float *xform)
nuclear@8 276 {
nuclear@8 277 float4 res;
nuclear@8 278 res.x = v.x * xform[0] + v.y * xform[4] + v.z * xform[8] + xform[12];
nuclear@8 279 res.y = v.x * xform[1] + v.y * xform[5] + v.z * xform[9] + xform[13];
nuclear@8 280 res.z = v.x * xform[2] + v.y * xform[6] + v.z * xform[10] + xform[14];
nuclear@12 281 res.w = 0.0;
nuclear@8 282 return res;
nuclear@8 283 }
nuclear@8 284
nuclear@16 285 void transform_ray(struct Ray *ray, global const float *xform, global const float *invtrans)
nuclear@8 286 {
nuclear@16 287 ray->origin = transform(ray->origin, xform);
nuclear@16 288 ray->dir = transform(ray->dir, invtrans);
nuclear@8 289 }
nuclear@9 290
nuclear@12 291 float4 calc_bary(float4 pt, global const struct Face *face, float4 norm)
nuclear@9 292 {
nuclear@12 293 float4 bc = (float4)(0, 0, 0, 0);
nuclear@9 294
nuclear@12 295 // calculate area of the whole triangle
nuclear@12 296 float4 v1 = face->v[1].pos - face->v[0].pos;
nuclear@12 297 float4 v2 = face->v[2].pos - face->v[0].pos;
nuclear@12 298 float4 xv1v2 = cross(v1, v2);
nuclear@12 299
nuclear@16 300 float area = fabs(dot(xv1v2, norm)) * 0.5;
nuclear@9 301 if(area < EPSILON) {
nuclear@9 302 return bc;
nuclear@9 303 }
nuclear@9 304
nuclear@9 305 float4 pv0 = face->v[0].pos - pt;
nuclear@9 306 float4 pv1 = face->v[1].pos - pt;
nuclear@9 307 float4 pv2 = face->v[2].pos - pt;
nuclear@9 308
nuclear@12 309 // calculate the area of each sub-triangle
nuclear@12 310 float4 x12 = cross(pv1, pv2);
nuclear@12 311 float4 x20 = cross(pv2, pv0);
nuclear@12 312 float4 x01 = cross(pv0, pv1);
nuclear@12 313
nuclear@16 314 float a0 = fabs(dot(x12, norm)) * 0.5;
nuclear@16 315 float a1 = fabs(dot(x20, norm)) * 0.5;
nuclear@16 316 float a2 = fabs(dot(x01, norm)) * 0.5;
nuclear@9 317
nuclear@9 318 bc.x = a0 / area;
nuclear@9 319 bc.y = a1 / area;
nuclear@9 320 bc.z = a2 / area;
nuclear@9 321 return bc;
nuclear@9 322 }
nuclear@19 323
nuclear@19 324 float mean(float4 v)
nuclear@19 325 {
nuclear@19 326 return native_divide(v.x + v.y + v.z, 3.0);
nuclear@19 327 }