clray

annotate rt.cl @ 35:7d77ded5f890

find changesets by author, revision, files, or words in the commit message
stopped using a heap to flatten the kdtree. added explicit left/right indices
author John Tsiombikas <nuclear@member.fsf.org>
date Thu, 26 Aug 2010 20:24:07 +0100
parents 92786fc3317e
children 980bc07be868
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@29 62 struct AABBox aabb;
nuclear@28 63 int face_idx[32];
nuclear@28 64 int num_faces;
nuclear@35 65 int left, right;
nuclear@35 66 int padding;
nuclear@16 67 };
nuclear@2 68
nuclear@16 69 #define MIN_ENERGY 0.001
nuclear@21 70 #define EPSILON 1e-5
nuclear@16 71
nuclear@16 72 float4 shade(struct Ray ray, struct Scene *scn, const struct SurfPoint *sp);
nuclear@16 73 bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *sp);
nuclear@9 74 bool intersect(struct Ray ray, global const struct Face *face, struct SurfPoint *sp);
nuclear@28 75 bool intersect_aabb(struct Ray ray, struct AABBox aabb);
nuclear@16 76
nuclear@8 77 float4 reflect(float4 v, float4 n);
nuclear@8 78 float4 transform(float4 v, global const float *xform);
nuclear@16 79 void transform_ray(struct Ray *ray, global const float *xform, global const float *invtrans);
nuclear@12 80 float4 calc_bary(float4 pt, global const struct Face *face, float4 norm);
nuclear@19 81 float mean(float4 v);
nuclear@4 82
nuclear@4 83 kernel void render(global float4 *fb,
nuclear@4 84 global const struct RendInfo *rinf,
nuclear@9 85 global const struct Face *faces,
nuclear@9 86 global const struct Material *matlib,
nuclear@4 87 global const struct Light *lights,
nuclear@7 88 global const struct Ray *primrays,
nuclear@12 89 global const float *xform,
nuclear@28 90 global const float *invtrans,
nuclear@28 91 global const struct KDNode *kdtree)
nuclear@2 92 {
nuclear@2 93 int idx = get_global_id(0);
nuclear@2 94
nuclear@16 95 struct Scene scn;
nuclear@16 96 scn.ambient = rinf->ambient;
nuclear@16 97 scn.faces = faces;
nuclear@16 98 scn.num_faces = rinf->num_faces;
nuclear@16 99 scn.lights = lights;
nuclear@16 100 scn.num_lights = rinf->num_lights;
nuclear@16 101 scn.matlib = matlib;
nuclear@30 102 scn.kdtree = kdtree;
nuclear@8 103
nuclear@16 104 struct Ray ray = primrays[idx];
nuclear@16 105 transform_ray(&ray, xform, invtrans);
nuclear@4 106
nuclear@19 107 float4 pixel = (float4)(0, 0, 0, 0);
nuclear@22 108 float4 energy = (float4)(1.0, 1.0, 1.0, 0.0);
nuclear@19 109 int iter = 0;
nuclear@19 110
nuclear@19 111 while(iter++ < rinf->max_iter && mean(energy) > MIN_ENERGY) {
nuclear@19 112 struct SurfPoint sp;
nuclear@19 113 if(find_intersection(ray, &scn, &sp)) {
nuclear@19 114 pixel += shade(ray, &scn, &sp) * energy;
nuclear@19 115
nuclear@19 116 float4 refl_col = sp.mat.ks * sp.mat.kr;
nuclear@19 117
nuclear@19 118 ray.origin = sp.pos;
nuclear@19 119 ray.dir = reflect(-ray.dir, sp.norm);
nuclear@19 120
nuclear@35 121 energy *= refl_col;
nuclear@19 122 } else {
nuclear@35 123 break;
nuclear@19 124 }
nuclear@17 125 }
nuclear@19 126
nuclear@19 127 fb[idx] = pixel;
nuclear@4 128 }
nuclear@4 129
nuclear@16 130 float4 shade(struct Ray ray, struct Scene *scn, const struct SurfPoint *sp)
nuclear@16 131 {
nuclear@16 132 float4 norm = sp->norm;
nuclear@12 133 bool entering = true;
nuclear@12 134
nuclear@12 135 if(dot(ray.dir, norm) >= 0.0) {
nuclear@12 136 norm = -norm;
nuclear@12 137 entering = false;
nuclear@12 138 }
nuclear@12 139
nuclear@19 140 float4 dcol = scn->ambient * sp->mat.kd;
nuclear@8 141 float4 scol = (float4)(0, 0, 0, 0);
nuclear@5 142
nuclear@16 143 for(int i=0; i<scn->num_lights; i++) {
nuclear@16 144 float4 ldir = scn->lights[i].pos - sp->pos;
nuclear@5 145
nuclear@16 146 struct Ray shadowray;
nuclear@16 147 shadowray.origin = sp->pos;
nuclear@16 148 shadowray.dir = ldir;
nuclear@5 149
nuclear@16 150 if(!find_intersection(shadowray, scn, 0)) {
nuclear@16 151 ldir = normalize(ldir);
nuclear@16 152 float4 vdir = -normalize(ray.dir);
nuclear@16 153 float4 vref = reflect(vdir, norm);
nuclear@16 154
nuclear@16 155 float diff = fmax(dot(ldir, norm), 0.0f);
nuclear@22 156 dcol += sp->mat.kd * scn->lights[i].color * diff;
nuclear@16 157
nuclear@20 158 float spec = powr(fmax(dot(ldir, vref), 0.0f), sp->mat.spow);
nuclear@22 159 scol += sp->mat.ks * scn->lights[i].color * spec;
nuclear@16 160 }
nuclear@16 161 }
nuclear@16 162
nuclear@8 163 return dcol + scol;
nuclear@2 164 }
nuclear@2 165
nuclear@30 166 #define STACK_SIZE 64
nuclear@28 167 bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *spres)
nuclear@28 168 {
nuclear@29 169 struct SurfPoint sp0;
nuclear@29 170 sp0.t = 1.0;
nuclear@29 171 sp0.obj = 0;
nuclear@29 172
nuclear@29 173 int idxstack[STACK_SIZE];
nuclear@31 174 int top = 0; // points after the topmost element of the stack
nuclear@35 175 idxstack[top++] = 0; // root at tree[0]
nuclear@29 176
nuclear@31 177 while(top > 0) {
nuclear@31 178 int idx = idxstack[--top]; // remove this index from the stack and process it
nuclear@30 179
nuclear@31 180 global const struct KDNode *node = scn->kdtree + idx;
nuclear@29 181
nuclear@31 182 /*if(get_global_id(0) == 0) {
nuclear@31 183 for(int i=0; i<top+1; i++) {
nuclear@31 184 printf(" ");
nuclear@31 185 }
nuclear@31 186 printf("(%d) idx: %d (%p) num_faces: %d\n", top+1, idx, node, node->num_faces);
nuclear@31 187 }*/
nuclear@29 188
nuclear@29 189 if(intersect_aabb(ray, node->aabb)) {
nuclear@35 190 if(node->left == -1) {
nuclear@31 191 // leaf node... check each face in turn and update the nearest intersection as needed
nuclear@29 192 for(int i=0; i<node->num_faces; i++) {
nuclear@31 193 struct SurfPoint spt;
nuclear@29 194 int fidx = node->face_idx[i];
nuclear@29 195
nuclear@31 196 if(intersect(ray, scn->faces + fidx, &spt) && spt.t < sp0.t) {
nuclear@31 197 sp0 = spt;
nuclear@29 198 }
nuclear@29 199 }
nuclear@31 200 } else {
nuclear@31 201 // internal node... recurse to the children
nuclear@31 202 /*if(get_global_id(0) == 0) {
nuclear@35 203 printf("pushing %d's children %d and %d\n", idx, node->left, node->right);
nuclear@31 204 }*/
nuclear@35 205 idxstack[top++] = node->left;
nuclear@35 206 idxstack[top++] = node->right;
nuclear@29 207 }
nuclear@29 208 }
nuclear@29 209 }
nuclear@29 210
nuclear@29 211 if(!sp0.obj) {
nuclear@29 212 return false;
nuclear@29 213 }
nuclear@29 214
nuclear@29 215 if(spres) {
nuclear@29 216 *spres = sp0;
nuclear@29 217 spres->mat = scn->matlib[sp0.obj->matid];
nuclear@29 218 }
nuclear@29 219 return true;
nuclear@28 220 }
nuclear@16 221
nuclear@28 222 /*bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *spres)
nuclear@12 223 {
nuclear@16 224 struct SurfPoint sp, sp0;
nuclear@16 225 sp0.t = 1.0;
nuclear@16 226 sp0.obj = 0;
nuclear@16 227
nuclear@16 228 for(int i=0; i<scn->num_faces; i++) {
nuclear@16 229 if(intersect(ray, scn->faces + i, &sp) && sp.t < sp0.t) {
nuclear@16 230 sp0 = sp;
nuclear@16 231 }
nuclear@16 232 }
nuclear@16 233
nuclear@16 234 if(!sp0.obj) {
nuclear@16 235 return false;
nuclear@16 236 }
nuclear@16 237
nuclear@16 238 if(spres) {
nuclear@16 239 *spres = sp0;
nuclear@19 240 spres->mat = scn->matlib[sp0.obj->matid];
nuclear@16 241 }
nuclear@16 242 return true;
nuclear@28 243 }*/
nuclear@12 244
nuclear@16 245 bool intersect(struct Ray ray, global const struct Face *face, struct SurfPoint *sp)
nuclear@2 246 {
nuclear@12 247 float4 origin = ray.origin;
nuclear@12 248 float4 dir = ray.dir;
nuclear@12 249 float4 norm = face->normal;
nuclear@12 250
nuclear@16 251 float ndotdir = dot(dir, norm);
nuclear@12 252
nuclear@9 253 if(fabs(ndotdir) <= EPSILON) {
nuclear@9 254 return false;
nuclear@9 255 }
nuclear@2 256
nuclear@9 257 float4 pt = face->v[0].pos;
nuclear@12 258 float4 vec = pt - origin;
nuclear@2 259
nuclear@16 260 float ndotvec = dot(norm, vec);
nuclear@9 261 float t = ndotvec / ndotdir;
nuclear@2 262
nuclear@2 263 if(t < EPSILON || t > 1.0) {
nuclear@2 264 return false;
nuclear@2 265 }
nuclear@12 266 pt = origin + dir * t;
nuclear@9 267
nuclear@12 268
nuclear@12 269 float4 bc = calc_bary(pt, face, norm);
nuclear@9 270 float bc_sum = bc.x + bc.y + bc.z;
nuclear@9 271
nuclear@20 272 if(bc_sum < 1.0 - EPSILON || bc_sum > 1.0 + EPSILON) {
nuclear@9 273 return false;
nuclear@12 274 bc *= 1.2;
nuclear@9 275 }
nuclear@2 276
nuclear@2 277 sp->t = t;
nuclear@9 278 sp->pos = pt;
nuclear@21 279 sp->norm = normalize(face->v[0].normal * bc.x + face->v[1].normal * bc.y + face->v[2].normal * bc.z);
nuclear@9 280 sp->obj = face;
nuclear@12 281 sp->dbg = bc;
nuclear@2 282 return true;
nuclear@2 283 }
nuclear@5 284
nuclear@28 285 bool intersect_aabb(struct Ray ray, struct AABBox aabb)
nuclear@28 286 {
nuclear@28 287 if(ray.origin.x >= aabb.min.x && ray.origin.y >= aabb.min.y && ray.origin.z >= aabb.min.z &&
nuclear@28 288 ray.origin.x < aabb.max.x && ray.origin.y < aabb.max.y && ray.origin.z < aabb.max.z) {
nuclear@28 289 return true;
nuclear@28 290 }
nuclear@28 291
nuclear@29 292 float4 bbox[2] = {
nuclear@29 293 aabb.min.x, aabb.min.y, aabb.min.z, 0,
nuclear@29 294 aabb.max.x, aabb.max.y, aabb.max.z, 0
nuclear@29 295 };
nuclear@28 296
nuclear@28 297 int xsign = (int)(ray.dir.x < 0.0);
nuclear@28 298 float invdirx = 1.0 / ray.dir.x;
nuclear@28 299 float tmin = (bbox[xsign].x - ray.origin.x) * invdirx;
nuclear@28 300 float tmax = (bbox[1 - xsign].x - ray.origin.x) * invdirx;
nuclear@28 301
nuclear@28 302 int ysign = (int)(ray.dir.y < 0.0);
nuclear@28 303 float invdiry = 1.0 / ray.dir.y;
nuclear@28 304 float tymin = (bbox[ysign].y - ray.origin.y) * invdiry;
nuclear@28 305 float tymax = (bbox[1 - ysign].y - ray.origin.y) * invdiry;
nuclear@28 306
nuclear@28 307 if(tmin > tymax || tymin > tmax) {
nuclear@28 308 return false;
nuclear@28 309 }
nuclear@28 310
nuclear@28 311 if(tymin > tmin) tmin = tymin;
nuclear@28 312 if(tymax < tmax) tmax = tymax;
nuclear@28 313
nuclear@28 314 int zsign = (int)(ray.dir.z < 0.0);
nuclear@28 315 float invdirz = 1.0 / ray.dir.z;
nuclear@28 316 float tzmin = (bbox[zsign].z - ray.origin.z) * invdirz;
nuclear@28 317 float tzmax = (bbox[1 - zsign].z - ray.origin.z) * invdirz;
nuclear@28 318
nuclear@28 319 if(tmin > tzmax || tzmin > tmax) {
nuclear@28 320 return false;
nuclear@28 321 }
nuclear@28 322
nuclear@29 323 return tmin < 1.0 && tmax > 0.0;
nuclear@28 324 }
nuclear@28 325
nuclear@8 326 float4 reflect(float4 v, float4 n)
nuclear@5 327 {
nuclear@23 328 return 2.0f * dot(v, n) * n - v;
nuclear@5 329 }
nuclear@8 330
nuclear@8 331 float4 transform(float4 v, global const float *xform)
nuclear@8 332 {
nuclear@8 333 float4 res;
nuclear@8 334 res.x = v.x * xform[0] + v.y * xform[4] + v.z * xform[8] + xform[12];
nuclear@8 335 res.y = v.x * xform[1] + v.y * xform[5] + v.z * xform[9] + xform[13];
nuclear@8 336 res.z = v.x * xform[2] + v.y * xform[6] + v.z * xform[10] + xform[14];
nuclear@12 337 res.w = 0.0;
nuclear@8 338 return res;
nuclear@8 339 }
nuclear@8 340
nuclear@16 341 void transform_ray(struct Ray *ray, global const float *xform, global const float *invtrans)
nuclear@8 342 {
nuclear@16 343 ray->origin = transform(ray->origin, xform);
nuclear@16 344 ray->dir = transform(ray->dir, invtrans);
nuclear@8 345 }
nuclear@9 346
nuclear@12 347 float4 calc_bary(float4 pt, global const struct Face *face, float4 norm)
nuclear@9 348 {
nuclear@12 349 float4 bc = (float4)(0, 0, 0, 0);
nuclear@9 350
nuclear@12 351 // calculate area of the whole triangle
nuclear@12 352 float4 v1 = face->v[1].pos - face->v[0].pos;
nuclear@12 353 float4 v2 = face->v[2].pos - face->v[0].pos;
nuclear@12 354 float4 xv1v2 = cross(v1, v2);
nuclear@12 355
nuclear@16 356 float area = fabs(dot(xv1v2, norm)) * 0.5;
nuclear@9 357 if(area < EPSILON) {
nuclear@9 358 return bc;
nuclear@9 359 }
nuclear@9 360
nuclear@9 361 float4 pv0 = face->v[0].pos - pt;
nuclear@9 362 float4 pv1 = face->v[1].pos - pt;
nuclear@9 363 float4 pv2 = face->v[2].pos - pt;
nuclear@9 364
nuclear@12 365 // calculate the area of each sub-triangle
nuclear@12 366 float4 x12 = cross(pv1, pv2);
nuclear@12 367 float4 x20 = cross(pv2, pv0);
nuclear@12 368 float4 x01 = cross(pv0, pv1);
nuclear@12 369
nuclear@16 370 float a0 = fabs(dot(x12, norm)) * 0.5;
nuclear@16 371 float a1 = fabs(dot(x20, norm)) * 0.5;
nuclear@16 372 float a2 = fabs(dot(x01, norm)) * 0.5;
nuclear@9 373
nuclear@9 374 bc.x = a0 / area;
nuclear@9 375 bc.y = a1 / area;
nuclear@9 376 bc.z = a2 / area;
nuclear@9 377 return bc;
nuclear@9 378 }
nuclear@19 379
nuclear@19 380 float mean(float4 v)
nuclear@19 381 {
nuclear@19 382 return native_divide(v.x + v.y + v.z, 3.0);
nuclear@19 383 }