clray

annotate rt.cl @ 33:931d13b72f83

forgot to add the timers
author John Tsiombikas <nuclear@member.fsf.org>
date Tue, 24 Aug 2010 05:47:04 +0100
parents 04803c702014
children 7d77ded5f890
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@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@30 101 scn.kdtree = kdtree;
nuclear@8 102
nuclear@16 103 struct Ray ray = primrays[idx];
nuclear@16 104 transform_ray(&ray, xform, invtrans);
nuclear@4 105
nuclear@19 106 float4 pixel = (float4)(0, 0, 0, 0);
nuclear@22 107 float4 energy = (float4)(1.0, 1.0, 1.0, 0.0);
nuclear@19 108 int iter = 0;
nuclear@19 109
nuclear@19 110 while(iter++ < rinf->max_iter && mean(energy) > MIN_ENERGY) {
nuclear@19 111 struct SurfPoint sp;
nuclear@19 112 if(find_intersection(ray, &scn, &sp)) {
nuclear@19 113 pixel += shade(ray, &scn, &sp) * energy;
nuclear@19 114
nuclear@19 115 float4 refl_col = sp.mat.ks * sp.mat.kr;
nuclear@19 116
nuclear@19 117 ray.origin = sp.pos;
nuclear@19 118 ray.dir = reflect(-ray.dir, sp.norm);
nuclear@19 119
nuclear@19 120 energy *= sp.mat.ks * sp.mat.kr;
nuclear@19 121 } else {
nuclear@19 122 iter = INT_MAX - 1; // to break out of the loop
nuclear@19 123 }
nuclear@17 124 }
nuclear@19 125
nuclear@19 126 fb[idx] = pixel;
nuclear@4 127 }
nuclear@4 128
nuclear@16 129 float4 shade(struct Ray ray, struct Scene *scn, const struct SurfPoint *sp)
nuclear@16 130 {
nuclear@16 131 float4 norm = sp->norm;
nuclear@12 132 bool entering = true;
nuclear@12 133
nuclear@12 134 if(dot(ray.dir, norm) >= 0.0) {
nuclear@12 135 norm = -norm;
nuclear@12 136 entering = false;
nuclear@12 137 }
nuclear@12 138
nuclear@19 139 float4 dcol = scn->ambient * sp->mat.kd;
nuclear@8 140 float4 scol = (float4)(0, 0, 0, 0);
nuclear@5 141
nuclear@16 142 for(int i=0; i<scn->num_lights; i++) {
nuclear@16 143 float4 ldir = scn->lights[i].pos - sp->pos;
nuclear@5 144
nuclear@16 145 struct Ray shadowray;
nuclear@16 146 shadowray.origin = sp->pos;
nuclear@16 147 shadowray.dir = ldir;
nuclear@5 148
nuclear@16 149 if(!find_intersection(shadowray, scn, 0)) {
nuclear@16 150 ldir = normalize(ldir);
nuclear@16 151 float4 vdir = -normalize(ray.dir);
nuclear@16 152 float4 vref = reflect(vdir, norm);
nuclear@16 153
nuclear@16 154 float diff = fmax(dot(ldir, norm), 0.0f);
nuclear@22 155 dcol += sp->mat.kd * scn->lights[i].color * diff;
nuclear@16 156
nuclear@20 157 float spec = powr(fmax(dot(ldir, vref), 0.0f), sp->mat.spow);
nuclear@22 158 scol += sp->mat.ks * scn->lights[i].color * spec;
nuclear@16 159 }
nuclear@16 160 }
nuclear@16 161
nuclear@8 162 return dcol + scol;
nuclear@2 163 }
nuclear@2 164
nuclear@30 165 #define STACK_SIZE 64
nuclear@28 166 bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *spres)
nuclear@28 167 {
nuclear@29 168 struct SurfPoint sp0;
nuclear@29 169 sp0.t = 1.0;
nuclear@29 170 sp0.obj = 0;
nuclear@29 171
nuclear@29 172 int idxstack[STACK_SIZE];
nuclear@31 173 int top = 0; // points after the topmost element of the stack
nuclear@31 174 idxstack[top++] = 1; // root at tree[1] (heap)
nuclear@29 175
nuclear@31 176 while(top > 0) {
nuclear@31 177 int idx = idxstack[--top]; // remove this index from the stack and process it
nuclear@30 178
nuclear@31 179 global const struct KDNode *node = scn->kdtree + idx;
nuclear@29 180
nuclear@31 181 /*if(get_global_id(0) == 0) {
nuclear@31 182 for(int i=0; i<top+1; i++) {
nuclear@31 183 printf(" ");
nuclear@31 184 }
nuclear@31 185 printf("(%d) idx: %d (%p) num_faces: %d\n", top+1, idx, node, node->num_faces);
nuclear@31 186 }*/
nuclear@29 187
nuclear@29 188 if(intersect_aabb(ray, node->aabb)) {
nuclear@30 189 if(node->num_faces >= 0) {
nuclear@31 190 // leaf node... check each face in turn and update the nearest intersection as needed
nuclear@29 191 for(int i=0; i<node->num_faces; i++) {
nuclear@31 192 struct SurfPoint spt;
nuclear@29 193 int fidx = node->face_idx[i];
nuclear@29 194
nuclear@31 195 if(intersect(ray, scn->faces + fidx, &spt) && spt.t < sp0.t) {
nuclear@31 196 sp0 = spt;
nuclear@29 197 }
nuclear@29 198 }
nuclear@31 199 } else {
nuclear@31 200 // internal node... recurse to the children
nuclear@31 201 /*if(get_global_id(0) == 0) {
nuclear@31 202 printf("pushing %d's children %d and %d\n", idx, idx * 2, idx * 2 + 1);
nuclear@31 203 }*/
nuclear@31 204 idxstack[top++] = idx * 2;
nuclear@31 205 idxstack[top++] = idx * 2 + 1;
nuclear@29 206 }
nuclear@29 207 }
nuclear@29 208 }
nuclear@29 209
nuclear@29 210 if(!sp0.obj) {
nuclear@29 211 return false;
nuclear@29 212 }
nuclear@29 213
nuclear@29 214 if(spres) {
nuclear@29 215 *spres = sp0;
nuclear@29 216 spres->mat = scn->matlib[sp0.obj->matid];
nuclear@29 217 }
nuclear@29 218 return true;
nuclear@28 219 }
nuclear@16 220
nuclear@28 221 /*bool find_intersection(struct Ray ray, const struct Scene *scn, struct SurfPoint *spres)
nuclear@12 222 {
nuclear@16 223 struct SurfPoint sp, sp0;
nuclear@16 224 sp0.t = 1.0;
nuclear@16 225 sp0.obj = 0;
nuclear@16 226
nuclear@16 227 for(int i=0; i<scn->num_faces; i++) {
nuclear@16 228 if(intersect(ray, scn->faces + i, &sp) && sp.t < sp0.t) {
nuclear@16 229 sp0 = sp;
nuclear@16 230 }
nuclear@16 231 }
nuclear@16 232
nuclear@16 233 if(!sp0.obj) {
nuclear@16 234 return false;
nuclear@16 235 }
nuclear@16 236
nuclear@16 237 if(spres) {
nuclear@16 238 *spres = sp0;
nuclear@19 239 spres->mat = scn->matlib[sp0.obj->matid];
nuclear@16 240 }
nuclear@16 241 return true;
nuclear@28 242 }*/
nuclear@12 243
nuclear@16 244 bool intersect(struct Ray ray, global const struct Face *face, struct SurfPoint *sp)
nuclear@2 245 {
nuclear@12 246 float4 origin = ray.origin;
nuclear@12 247 float4 dir = ray.dir;
nuclear@12 248 float4 norm = face->normal;
nuclear@12 249
nuclear@16 250 float ndotdir = dot(dir, norm);
nuclear@12 251
nuclear@9 252 if(fabs(ndotdir) <= EPSILON) {
nuclear@9 253 return false;
nuclear@9 254 }
nuclear@2 255
nuclear@9 256 float4 pt = face->v[0].pos;
nuclear@12 257 float4 vec = pt - origin;
nuclear@2 258
nuclear@16 259 float ndotvec = dot(norm, vec);
nuclear@9 260 float t = ndotvec / ndotdir;
nuclear@2 261
nuclear@2 262 if(t < EPSILON || t > 1.0) {
nuclear@2 263 return false;
nuclear@2 264 }
nuclear@12 265 pt = origin + dir * t;
nuclear@9 266
nuclear@12 267
nuclear@12 268 float4 bc = calc_bary(pt, face, norm);
nuclear@9 269 float bc_sum = bc.x + bc.y + bc.z;
nuclear@9 270
nuclear@20 271 if(bc_sum < 1.0 - EPSILON || bc_sum > 1.0 + EPSILON) {
nuclear@9 272 return false;
nuclear@12 273 bc *= 1.2;
nuclear@9 274 }
nuclear@2 275
nuclear@2 276 sp->t = t;
nuclear@9 277 sp->pos = pt;
nuclear@21 278 sp->norm = normalize(face->v[0].normal * bc.x + face->v[1].normal * bc.y + face->v[2].normal * bc.z);
nuclear@9 279 sp->obj = face;
nuclear@12 280 sp->dbg = bc;
nuclear@2 281 return true;
nuclear@2 282 }
nuclear@5 283
nuclear@28 284 bool intersect_aabb(struct Ray ray, struct AABBox aabb)
nuclear@28 285 {
nuclear@28 286 if(ray.origin.x >= aabb.min.x && ray.origin.y >= aabb.min.y && ray.origin.z >= aabb.min.z &&
nuclear@28 287 ray.origin.x < aabb.max.x && ray.origin.y < aabb.max.y && ray.origin.z < aabb.max.z) {
nuclear@28 288 return true;
nuclear@28 289 }
nuclear@28 290
nuclear@29 291 float4 bbox[2] = {
nuclear@29 292 aabb.min.x, aabb.min.y, aabb.min.z, 0,
nuclear@29 293 aabb.max.x, aabb.max.y, aabb.max.z, 0
nuclear@29 294 };
nuclear@28 295
nuclear@28 296 int xsign = (int)(ray.dir.x < 0.0);
nuclear@28 297 float invdirx = 1.0 / ray.dir.x;
nuclear@28 298 float tmin = (bbox[xsign].x - ray.origin.x) * invdirx;
nuclear@28 299 float tmax = (bbox[1 - xsign].x - ray.origin.x) * invdirx;
nuclear@28 300
nuclear@28 301 int ysign = (int)(ray.dir.y < 0.0);
nuclear@28 302 float invdiry = 1.0 / ray.dir.y;
nuclear@28 303 float tymin = (bbox[ysign].y - ray.origin.y) * invdiry;
nuclear@28 304 float tymax = (bbox[1 - ysign].y - ray.origin.y) * invdiry;
nuclear@28 305
nuclear@28 306 if(tmin > tymax || tymin > tmax) {
nuclear@28 307 return false;
nuclear@28 308 }
nuclear@28 309
nuclear@28 310 if(tymin > tmin) tmin = tymin;
nuclear@28 311 if(tymax < tmax) tmax = tymax;
nuclear@28 312
nuclear@28 313 int zsign = (int)(ray.dir.z < 0.0);
nuclear@28 314 float invdirz = 1.0 / ray.dir.z;
nuclear@28 315 float tzmin = (bbox[zsign].z - ray.origin.z) * invdirz;
nuclear@28 316 float tzmax = (bbox[1 - zsign].z - ray.origin.z) * invdirz;
nuclear@28 317
nuclear@28 318 if(tmin > tzmax || tzmin > tmax) {
nuclear@28 319 return false;
nuclear@28 320 }
nuclear@28 321
nuclear@29 322 return tmin < 1.0 && tmax > 0.0;
nuclear@28 323 }
nuclear@28 324
nuclear@8 325 float4 reflect(float4 v, float4 n)
nuclear@5 326 {
nuclear@23 327 return 2.0f * dot(v, n) * n - v;
nuclear@5 328 }
nuclear@8 329
nuclear@8 330 float4 transform(float4 v, global const float *xform)
nuclear@8 331 {
nuclear@8 332 float4 res;
nuclear@8 333 res.x = v.x * xform[0] + v.y * xform[4] + v.z * xform[8] + xform[12];
nuclear@8 334 res.y = v.x * xform[1] + v.y * xform[5] + v.z * xform[9] + xform[13];
nuclear@8 335 res.z = v.x * xform[2] + v.y * xform[6] + v.z * xform[10] + xform[14];
nuclear@12 336 res.w = 0.0;
nuclear@8 337 return res;
nuclear@8 338 }
nuclear@8 339
nuclear@16 340 void transform_ray(struct Ray *ray, global const float *xform, global const float *invtrans)
nuclear@8 341 {
nuclear@16 342 ray->origin = transform(ray->origin, xform);
nuclear@16 343 ray->dir = transform(ray->dir, invtrans);
nuclear@8 344 }
nuclear@9 345
nuclear@12 346 float4 calc_bary(float4 pt, global const struct Face *face, float4 norm)
nuclear@9 347 {
nuclear@12 348 float4 bc = (float4)(0, 0, 0, 0);
nuclear@9 349
nuclear@12 350 // calculate area of the whole triangle
nuclear@12 351 float4 v1 = face->v[1].pos - face->v[0].pos;
nuclear@12 352 float4 v2 = face->v[2].pos - face->v[0].pos;
nuclear@12 353 float4 xv1v2 = cross(v1, v2);
nuclear@12 354
nuclear@16 355 float area = fabs(dot(xv1v2, norm)) * 0.5;
nuclear@9 356 if(area < EPSILON) {
nuclear@9 357 return bc;
nuclear@9 358 }
nuclear@9 359
nuclear@9 360 float4 pv0 = face->v[0].pos - pt;
nuclear@9 361 float4 pv1 = face->v[1].pos - pt;
nuclear@9 362 float4 pv2 = face->v[2].pos - pt;
nuclear@9 363
nuclear@12 364 // calculate the area of each sub-triangle
nuclear@12 365 float4 x12 = cross(pv1, pv2);
nuclear@12 366 float4 x20 = cross(pv2, pv0);
nuclear@12 367 float4 x01 = cross(pv0, pv1);
nuclear@12 368
nuclear@16 369 float a0 = fabs(dot(x12, norm)) * 0.5;
nuclear@16 370 float a1 = fabs(dot(x20, norm)) * 0.5;
nuclear@16 371 float a2 = fabs(dot(x01, norm)) * 0.5;
nuclear@9 372
nuclear@9 373 bc.x = a0 / area;
nuclear@9 374 bc.y = a1 / area;
nuclear@9 375 bc.z = a2 / area;
nuclear@9 376 return bc;
nuclear@9 377 }
nuclear@19 378
nuclear@19 379 float mean(float4 v)
nuclear@19 380 {
nuclear@19 381 return native_divide(v.x + v.y + v.z, 3.0);
nuclear@19 382 }