clray

annotate src/rt.cc @ 45:8047637961a2

find changesets by author, revision, files, or words in the commit message
fixed the issue of hitting maximum vertical image sizes for large kdtrees
author John Tsiombikas <nuclear@member.fsf.org>
date Sun, 29 Aug 2010 04:20:42 +0100
parents f9eec11e5acc
children 30bf84881553
rev   line source
nuclear@0 1 #include <stdio.h>
nuclear@8 2 #include <string.h>
nuclear@2 3 #include <math.h>
nuclear@0 4 #include <assert.h>
John@14 5 #include "ogl.h"
nuclear@0 6 #include "ocl.h"
nuclear@22 7 #include "scene.h"
nuclear@32 8 #include "timer.h"
nuclear@45 9 #include "common.h"
nuclear@0 10
nuclear@12 11 // kernel arguments
nuclear@12 12 enum {
nuclear@12 13 KARG_FRAMEBUFFER,
nuclear@12 14 KARG_RENDER_INFO,
nuclear@12 15 KARG_FACES,
nuclear@12 16 KARG_MATLIB,
nuclear@12 17 KARG_LIGHTS,
nuclear@12 18 KARG_PRIM_RAYS,
nuclear@12 19 KARG_XFORM,
John@14 20 KARG_INVTRANS_XFORM,
nuclear@28 21 KARG_KDTREE,
John@14 22
John@14 23 NUM_KERNEL_ARGS
nuclear@12 24 };
John@11 25
nuclear@2 26 struct RendInfo {
nuclear@22 27 float ambient[4];
nuclear@2 28 int xsz, ysz;
nuclear@9 29 int num_faces, num_lights;
nuclear@2 30 int max_iter;
nuclear@28 31 int kd_depth;
nuclear@12 32 };
nuclear@2 33
nuclear@1 34 struct Ray {
nuclear@8 35 float origin[4], dir[4];
nuclear@12 36 };
nuclear@1 37
nuclear@3 38 struct Light {
nuclear@8 39 float pos[4], color[4];
nuclear@12 40 };
nuclear@1 41
nuclear@3 42 static Ray get_primary_ray(int x, int y, int w, int h, float vfov_deg);
nuclear@43 43 static float *create_kdimage(const KDNodeGPU *kdtree, int num_nodes, int *xsz_ret, int *ysz_ret);
nuclear@3 44
nuclear@13 45 static Face *faces;
nuclear@3 46 static Ray *prim_rays;
nuclear@3 47 static CLProgram *prog;
nuclear@3 48 static int global_size;
nuclear@3 49
nuclear@4 50 static Light lightlist[] = {
nuclear@22 51 {{-8, 15, 18, 0}, {1, 1, 1, 1}}
nuclear@4 52 };
nuclear@4 53
nuclear@7 54
nuclear@4 55 static RendInfo rinf;
nuclear@4 56
nuclear@43 57 static long timing_sample_sum;
nuclear@43 58 static long num_timing_samples;
nuclear@43 59
nuclear@4 60
nuclear@39 61 bool init_renderer(int xsz, int ysz, Scene *scn, unsigned int tex)
nuclear@0 62 {
nuclear@4 63 // render info
nuclear@22 64 rinf.ambient[0] = rinf.ambient[1] = rinf.ambient[2] = 0.0;
nuclear@16 65 rinf.ambient[3] = 0.0;
nuclear@16 66
nuclear@4 67 rinf.xsz = xsz;
nuclear@4 68 rinf.ysz = ysz;
nuclear@13 69 rinf.num_faces = scn->get_num_faces();
nuclear@4 70 rinf.num_lights = sizeof lightlist / sizeof *lightlist;
nuclear@4 71 rinf.max_iter = 6;
nuclear@28 72 rinf.kd_depth = kdtree_depth(scn->kdtree);
nuclear@4 73
nuclear@3 74 /* calculate primary rays */
nuclear@3 75 prim_rays = new Ray[xsz * ysz];
nuclear@2 76
nuclear@2 77 for(int i=0; i<ysz; i++) {
nuclear@2 78 for(int j=0; j<xsz; j++) {
nuclear@2 79 prim_rays[i * xsz + j] = get_primary_ray(j, i, xsz, ysz, 45.0);
nuclear@2 80 }
nuclear@0 81 }
nuclear@0 82
nuclear@2 83 /* setup opencl */
nuclear@3 84 prog = new CLProgram("render");
nuclear@3 85 if(!prog->load("rt.cl")) {
nuclear@8 86 return false;
nuclear@0 87 }
nuclear@0 88
nuclear@24 89 if(!(faces = (Face*)scn->get_face_buffer())) {
nuclear@13 90 fprintf(stderr, "failed to create face buffer\n");
nuclear@13 91 return false;
nuclear@13 92 }
nuclear@13 93
nuclear@28 94 const KDNodeGPU *kdbuf = scn->get_kdtree_buffer();
nuclear@28 95 if(!kdbuf) {
nuclear@28 96 fprintf(stderr, "failed to create kdtree buffer\n");
nuclear@28 97 return false;
nuclear@28 98 }
nuclear@43 99
nuclear@43 100 int kdimg_xsz, kdimg_ysz;
nuclear@43 101 float *kdimg_pixels = create_kdimage(kdbuf, scn->get_num_kdnodes(), &kdimg_xsz, &kdimg_ysz);
nuclear@28 102
nuclear@3 103 /* setup argument buffers */
nuclear@41 104 #ifdef CLGL_INTEROP
nuclear@39 105 prog->set_arg_texture(KARG_FRAMEBUFFER, ARG_WR, tex);
nuclear@41 106 #else
nuclear@41 107 prog->set_arg_image(KARG_FRAMEBUFFER, ARG_WR, xsz, ysz);
nuclear@41 108 #endif
nuclear@12 109 prog->set_arg_buffer(KARG_RENDER_INFO, ARG_RD, sizeof rinf, &rinf);
John@14 110 prog->set_arg_buffer(KARG_FACES, ARG_RD, rinf.num_faces * sizeof(Face), faces);
John@14 111 prog->set_arg_buffer(KARG_MATLIB, ARG_RD, scn->get_num_materials() * sizeof(Material), scn->get_materials());
nuclear@12 112 prog->set_arg_buffer(KARG_LIGHTS, ARG_RD, sizeof lightlist, lightlist);
nuclear@12 113 prog->set_arg_buffer(KARG_PRIM_RAYS, ARG_RD, xsz * ysz * sizeof *prim_rays, prim_rays);
nuclear@12 114 prog->set_arg_buffer(KARG_XFORM, ARG_RD, 16 * sizeof(float));
nuclear@12 115 prog->set_arg_buffer(KARG_INVTRANS_XFORM, ARG_RD, 16 * sizeof(float));
nuclear@43 116 //prog->set_arg_buffer(KARG_KDTREE, ARG_RD, scn->get_num_kdnodes() * sizeof *kdbuf, kdbuf);
nuclear@43 117 prog->set_arg_image(KARG_KDTREE, ARG_RD, kdimg_xsz, kdimg_ysz, kdimg_pixels);
nuclear@43 118
nuclear@43 119 delete [] kdimg_pixels;
nuclear@43 120
nuclear@12 121
John@14 122 if(prog->get_num_args() < NUM_KERNEL_ARGS) {
John@14 123 return false;
John@14 124 }
John@14 125
nuclear@45 126 const char *opt = "-Isrc -cl-mad-enable -cl-single-precision-constant -cl-fast-relaxed-math";
nuclear@45 127 if(!prog->build(opt)) {
nuclear@16 128 return false;
nuclear@16 129 }
nuclear@16 130
nuclear@12 131 delete [] prim_rays;
nuclear@2 132
nuclear@3 133 global_size = xsz * ysz;
nuclear@3 134 return true;
nuclear@3 135 }
nuclear@3 136
nuclear@3 137 void destroy_renderer()
nuclear@3 138 {
nuclear@3 139 delete prog;
nuclear@43 140
nuclear@43 141 printf("rendertime mean: %ld msec\n", timing_sample_sum / num_timing_samples);
nuclear@3 142 }
nuclear@3 143
nuclear@3 144 bool render()
nuclear@3 145 {
nuclear@39 146 // XXX do we need to call glFinish ?
nuclear@39 147
nuclear@32 148 long tm0 = get_msec();
nuclear@32 149
nuclear@40 150 #ifdef CLGL_INTEROP
nuclear@39 151 cl_event ev;
nuclear@39 152 CLMemBuffer *texbuf = prog->get_arg_buffer(KARG_FRAMEBUFFER);
nuclear@39 153
nuclear@39 154 if(!acquire_gl_object(texbuf, &ev)) {
nuclear@39 155 return false;
nuclear@39 156 }
nuclear@39 157
nuclear@39 158 // make sure that we will wait for the acquire to finish before running
nuclear@39 159 prog->set_wait_event(ev);
nuclear@40 160 #endif
nuclear@39 161
nuclear@3 162 if(!prog->run(1, global_size)) {
nuclear@3 163 return false;
nuclear@0 164 }
John@15 165
nuclear@40 166 #ifdef CLGL_INTEROP
nuclear@39 167 if(!release_gl_object(texbuf, &ev)) {
nuclear@39 168 return false;
nuclear@39 169 }
nuclear@39 170 clWaitForEvents(1, &ev);
nuclear@40 171 #endif
nuclear@39 172
nuclear@40 173 #ifndef CLGL_INTEROP
nuclear@40 174 /* if we don't compile in CL/GL interoperability support, we need
nuclear@40 175 * to copy the output buffer to the OpenGL texture used to displaying
nuclear@40 176 * the image.
nuclear@40 177 */
nuclear@13 178 CLMemBuffer *mbuf = prog->get_arg_buffer(KARG_FRAMEBUFFER);
nuclear@12 179 void *fb = map_mem_buffer(mbuf, MAP_RD);
nuclear@13 180 if(!fb) {
nuclear@13 181 fprintf(stderr, "FAILED\n");
nuclear@13 182 return false;
nuclear@13 183 }
nuclear@13 184
nuclear@12 185 glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, rinf.xsz, rinf.ysz, GL_RGBA, GL_FLOAT, fb);
nuclear@2 186 unmap_mem_buffer(mbuf);
nuclear@40 187 #endif
nuclear@32 188
nuclear@43 189 long msec = get_msec() - tm0;
nuclear@43 190 timing_sample_sum += msec;
nuclear@43 191 num_timing_samples++;
nuclear@43 192
nuclear@43 193 printf("rendered in %ld msec\n", msec);
nuclear@3 194 return true;
nuclear@0 195 }
nuclear@2 196
nuclear@27 197 #define MIN(a, b) ((a) < (b) ? (a) : (b))
nuclear@21 198 static void dbg_set_gl_material(Material *mat)
nuclear@21 199 {
nuclear@21 200 static Material def_mat = {{0.7, 0.7, 0.7, 1}, {0, 0, 0, 0}, 0, 0, 0};
nuclear@21 201
nuclear@21 202 if(!mat) mat = &def_mat;
nuclear@21 203
nuclear@21 204 glMaterialfv(GL_FRONT_AND_BACK, GL_AMBIENT_AND_DIFFUSE, mat->kd);
nuclear@21 205 glMaterialfv(GL_FRONT_AND_BACK, GL_SPECULAR, mat->ks);
nuclear@27 206 glMaterialf(GL_FRONT_AND_BACK, GL_SHININESS, MIN(mat->spow, 128.0f));
nuclear@21 207 }
nuclear@21 208
nuclear@27 209 void dbg_render_gl(Scene *scn, bool show_tree, bool show_obj)
nuclear@8 210 {
nuclear@22 211 glPushAttrib(GL_ENABLE_BIT | GL_TRANSFORM_BIT | GL_LIGHTING_BIT);
nuclear@8 212
nuclear@21 213 for(int i=0; i<rinf.num_lights; i++) {
nuclear@21 214 float lpos[4];
nuclear@21 215
nuclear@21 216 memcpy(lpos, lightlist[i].pos, sizeof lpos);
nuclear@21 217 lpos[3] = 1.0;
nuclear@21 218
nuclear@21 219 glLightfv(GL_LIGHT0 + i, GL_POSITION, lpos);
nuclear@21 220 glLightfv(GL_LIGHT0 + i, GL_DIFFUSE, lightlist[i].color);
nuclear@22 221 glEnable(GL_LIGHT0 + i);
nuclear@21 222 }
nuclear@21 223
nuclear@12 224 glDisable(GL_TEXTURE_2D);
nuclear@12 225 glEnable(GL_DEPTH_TEST);
John@15 226 glEnable(GL_LIGHTING);
nuclear@12 227
nuclear@12 228 glMatrixMode(GL_PROJECTION);
nuclear@12 229 glPushMatrix();
nuclear@12 230 glLoadIdentity();
nuclear@12 231 gluPerspective(45.0, (float)rinf.xsz / (float)rinf.ysz, 0.5, 1000.0);
nuclear@12 232
nuclear@27 233 if(show_obj) {
nuclear@27 234 Material *materials = scn->get_materials();
John@14 235
nuclear@27 236 int num_faces = scn->get_num_faces();
nuclear@27 237 int cur_mat = -1;
nuclear@21 238
nuclear@27 239 for(int i=0; i<num_faces; i++) {
nuclear@27 240 if(faces[i].matid != cur_mat) {
nuclear@27 241 if(cur_mat != -1) {
nuclear@27 242 glEnd();
nuclear@27 243 }
nuclear@27 244 dbg_set_gl_material(materials ? materials + faces[i].matid : 0);
nuclear@27 245 cur_mat = faces[i].matid;
nuclear@27 246 glBegin(GL_TRIANGLES);
nuclear@21 247 }
nuclear@27 248
nuclear@27 249 for(int j=0; j<3; j++) {
nuclear@27 250 glNormal3fv(faces[i].v[j].normal);
nuclear@27 251 glVertex3fv(faces[i].v[j].pos);
nuclear@27 252 }
John@14 253 }
nuclear@27 254 glEnd();
nuclear@27 255 }
nuclear@12 256
nuclear@27 257 if(show_tree) {
nuclear@27 258 scn->draw_kdtree();
nuclear@12 259 }
nuclear@12 260
nuclear@12 261 glPopMatrix();
nuclear@12 262 glPopAttrib();
nuclear@22 263
nuclear@22 264 assert(glGetError() == GL_NO_ERROR);
nuclear@12 265 }
nuclear@12 266
nuclear@12 267 void set_xform(float *matrix, float *invtrans)
nuclear@12 268 {
nuclear@12 269 CLMemBuffer *mbuf_xform = prog->get_arg_buffer(KARG_XFORM);
nuclear@12 270 CLMemBuffer *mbuf_invtrans = prog->get_arg_buffer(KARG_INVTRANS_XFORM);
nuclear@12 271 assert(mbuf_xform && mbuf_invtrans);
nuclear@12 272
nuclear@12 273 float *mem = (float*)map_mem_buffer(mbuf_xform, MAP_WR);
nuclear@12 274 memcpy(mem, matrix, 16 * sizeof *mem);
nuclear@12 275 unmap_mem_buffer(mbuf_xform);
nuclear@12 276
nuclear@12 277 mem = (float*)map_mem_buffer(mbuf_invtrans, MAP_WR);
nuclear@12 278 memcpy(mem, invtrans, 16 * sizeof *mem);
nuclear@12 279 unmap_mem_buffer(mbuf_invtrans);
nuclear@8 280 }
nuclear@8 281
nuclear@3 282 static Ray get_primary_ray(int x, int y, int w, int h, float vfov_deg)
nuclear@2 283 {
nuclear@2 284 float vfov = M_PI * vfov_deg / 180.0;
nuclear@2 285 float aspect = (float)w / (float)h;
nuclear@2 286
nuclear@2 287 float ysz = 2.0;
nuclear@2 288 float xsz = aspect * ysz;
nuclear@2 289
nuclear@2 290 float px = ((float)x / (float)w) * xsz - xsz / 2.0;
nuclear@2 291 float py = 1.0 - ((float)y / (float)h) * ysz;
nuclear@2 292 float pz = 1.0 / tan(0.5 * vfov);
nuclear@2 293
nuclear@43 294 float mag = sqrt(px * px + py * py + pz * pz);
nuclear@43 295
nuclear@45 296 px = px * RAY_MAG / mag;
nuclear@45 297 py = py * RAY_MAG / mag;
nuclear@45 298 pz = pz * RAY_MAG / mag;
nuclear@2 299
nuclear@18 300 Ray ray = {{0, 0, 0, 1}, {px, py, -pz, 1}};
nuclear@2 301 return ray;
nuclear@2 302 }
nuclear@43 303
nuclear@43 304 static float *create_kdimage(const KDNodeGPU *kdtree, int num_nodes, int *xsz_ret, int *ysz_ret)
nuclear@43 305 {
nuclear@45 306 int ysz = MIN(num_nodes, KDIMG_MAX_HEIGHT);
nuclear@45 307 int columns = (num_nodes - 1) / KDIMG_MAX_HEIGHT + 1;
nuclear@45 308 int xsz = KDIMG_NODE_WIDTH * columns;
nuclear@43 309
nuclear@43 310 printf("creating kdtree image %dx%d (%d nodes)\n", xsz, ysz, num_nodes);
nuclear@43 311
nuclear@43 312 float *img = new float[4 * xsz * ysz];
nuclear@43 313 memset(img, 0, 4 * xsz * ysz * sizeof *img);
nuclear@43 314
nuclear@43 315 for(int i=0; i<num_nodes; i++) {
nuclear@45 316 int x = KDIMG_NODE_WIDTH * (i / KDIMG_MAX_HEIGHT);
nuclear@45 317 int y = i % KDIMG_MAX_HEIGHT;
nuclear@45 318
nuclear@45 319 float *ptr = img + (y * xsz + x) * 4;
nuclear@43 320
nuclear@43 321 *ptr++ = kdtree[i].aabb.min[0];
nuclear@43 322 *ptr++ = kdtree[i].aabb.min[1];
nuclear@43 323 *ptr++ = kdtree[i].aabb.min[2];
nuclear@43 324 *ptr++ = 0.0;
nuclear@43 325
nuclear@43 326 *ptr++ = kdtree[i].aabb.max[0];
nuclear@43 327 *ptr++ = kdtree[i].aabb.max[1];
nuclear@43 328 *ptr++ = kdtree[i].aabb.max[2];
nuclear@43 329 *ptr++ = 0.0;
nuclear@43 330
nuclear@43 331 for(int j=0; j<MAX_NODE_FACES; j++) {
nuclear@43 332 *ptr++ = j < kdtree[i].num_faces ? (float)kdtree[i].face_idx[j] : 0.0f;
nuclear@43 333 }
nuclear@43 334
nuclear@43 335 *ptr++ = (float)kdtree[i].num_faces;
nuclear@43 336 *ptr++ = (float)kdtree[i].left;
nuclear@43 337 *ptr++ = (float)kdtree[i].right;
nuclear@43 338 *ptr++ = 0.0;
nuclear@43 339 }
nuclear@43 340
nuclear@43 341 if(xsz_ret) *xsz_ret = xsz;
nuclear@43 342 if(ysz_ret) *ysz_ret = ysz;
nuclear@43 343 return img;
nuclear@43 344 }