clray: 1ae68d46cfda src/rt.cc

clray

view src/rt.cc @ 49:1ae68d46cfda

- updated the visual studio project files - fixed the previously broken windows build

author	John Tsiombikas
date	Tue, 31 Aug 2010 01:28:20 +0100
parents	8047637961a2
children	d3c46803242e

line source

1 #include <stdio.h>

2 #include <string.h>

3 #include <math.h>

4 #include <assert.h>

5 #include "rt.h"

6 #include "ogl.h"

7 #include "ocl.h"

8 #include "scene.h"

9 #include "timer.h"

10 #include "common.h"

12 // kernel arguments

13 enum {

14 KARG_FRAMEBUFFER,

15 KARG_RENDER_INFO,

16 KARG_FACES,

17 KARG_MATLIB,

18 KARG_LIGHTS,

19 KARG_PRIM_RAYS,

20 KARG_XFORM,

21 KARG_INVTRANS_XFORM,

22 KARG_KDTREE,

24 NUM_KERNEL_ARGS

25 };

27 struct RendInfo {

28 float ambient[4];

29 int xsz, ysz;

30 int num_faces, num_lights;

31 int max_iter;

32 int cast_shadows;

33 };

35 struct Ray {

36 float origin[4], dir[4];

37 };

39 struct Light {

40 float pos[4], color[4];

41 };

43 static void update_render_info();

44 static Ray get_primary_ray(int x, int y, int w, int h, float vfov_deg);

45 static float *create_kdimage(const KDNodeGPU *kdtree, int num_nodes, int *xsz_ret, int *ysz_ret);

47 static Face *faces;

48 static Ray *prim_rays;

49 static CLProgram *prog;

50 static int global_size;

52 static Light lightlist[] = {

53 {{-8, 15, 18, 0}, {1, 1, 1, 1}}

54 };

57 static RendInfo rinf;

58 static int saved_iter_val;

60 static long timing_sample_sum;

61 static long num_timing_samples;

64 bool init_renderer(int xsz, int ysz, Scene *scn, unsigned int tex)

65 {

66 // render info

67 rinf.ambient[0] = rinf.ambient[1] = rinf.ambient[2] = 0.0;

68 rinf.ambient[3] = 0.0;

70 rinf.xsz = xsz;

71 rinf.ysz = ysz;

72 rinf.num_faces = scn->get_num_faces();

73 rinf.num_lights = sizeof lightlist / sizeof *lightlist;

74 rinf.max_iter = saved_iter_val = 6;

75 rinf.cast_shadows = true;

77 /* calculate primary rays */

78 prim_rays = new Ray[xsz * ysz];

80 for(int i=0; i<ysz; i++) {

81 for(int j=0; j<xsz; j++) {

82 prim_rays[i * xsz + j] = get_primary_ray(j, i, xsz, ysz, 45.0);

83 }

84 }

86 /* setup opencl */

87 prog = new CLProgram("render");

88 if(!prog->load("rt.cl")) {

89 return false;

90 }

92 if(!(faces = (Face*)scn->get_face_buffer())) {

93 fprintf(stderr, "failed to create face buffer\n");

94 return false;

95 }

97 const KDNodeGPU *kdbuf = scn->get_kdtree_buffer();

98 if(!kdbuf) {

99 fprintf(stderr, "failed to create kdtree buffer\n");

100 return false;

101 }

102

103 int kdimg_xsz, kdimg_ysz;

104 float *kdimg_pixels = create_kdimage(kdbuf, scn->get_num_kdnodes(), &kdimg_xsz, &kdimg_ysz);

105

106 /* setup argument buffers */

107 #ifdef CLGL_INTEROP

108 prog->set_arg_texture(KARG_FRAMEBUFFER, ARG_WR, tex);

109 #else

110 prog->set_arg_image(KARG_FRAMEBUFFER, ARG_WR, xsz, ysz);

111 #endif

112 prog->set_arg_buffer(KARG_RENDER_INFO, ARG_RD, sizeof rinf, &rinf);

113 prog->set_arg_buffer(KARG_FACES, ARG_RD, rinf.num_faces * sizeof(Face), faces);

114 prog->set_arg_buffer(KARG_MATLIB, ARG_RD, scn->get_num_materials() * sizeof(Material), scn->get_materials());

115 prog->set_arg_buffer(KARG_LIGHTS, ARG_RD, sizeof lightlist, lightlist);

116 prog->set_arg_buffer(KARG_PRIM_RAYS, ARG_RD, xsz * ysz * sizeof *prim_rays, prim_rays);

117 prog->set_arg_buffer(KARG_XFORM, ARG_RD, 16 * sizeof(float));

118 prog->set_arg_buffer(KARG_INVTRANS_XFORM, ARG_RD, 16 * sizeof(float));

119 //prog->set_arg_buffer(KARG_KDTREE, ARG_RD, scn->get_num_kdnodes() * sizeof *kdbuf, kdbuf);

120 prog->set_arg_image(KARG_KDTREE, ARG_RD, kdimg_xsz, kdimg_ysz, kdimg_pixels);

121

122 delete [] kdimg_pixels;

123

124

125 if(prog->get_num_args() < NUM_KERNEL_ARGS) {

126 return false;

127 }

128

129 const char *opt = "-Isrc -cl-mad-enable -cl-single-precision-constant -cl-fast-relaxed-math";

130 if(!prog->build(opt)) {

131 return false;

132 }

133

134 delete [] prim_rays;

135

136 global_size = xsz * ysz;

137 return true;

138 }

139

140 void destroy_renderer()

141 {

142 delete prog;

143

144 printf("rendertime mean: %ld msec\n", timing_sample_sum / num_timing_samples);

145 }

146

147 bool render()

148 {

149 // XXX do we need to call glFinish ?

150

151 long tm0 = get_msec();

152

153 #ifdef CLGL_INTEROP

154 cl_event ev;

155 CLMemBuffer *texbuf = prog->get_arg_buffer(KARG_FRAMEBUFFER);

156

157 if(!acquire_gl_object(texbuf, &ev)) {

158 return false;

159 }

160

161 // make sure that we will wait for the acquire to finish before running

162 prog->set_wait_event(ev);

163 #endif

164

165 if(!prog->run(1, global_size)) {

166 return false;

167 }

168

169 #ifdef CLGL_INTEROP

170 if(!release_gl_object(texbuf, &ev)) {

171 return false;

172 }

173 clWaitForEvents(1, &ev);

174 #endif

175

176 #ifndef CLGL_INTEROP

177 /* if we don't compile in CL/GL interoperability support, we need

178 * to copy the output buffer to the OpenGL texture used to displaying

179 * the image.

180 */

181 CLMemBuffer *mbuf = prog->get_arg_buffer(KARG_FRAMEBUFFER);

182 void *fb = map_mem_buffer(mbuf, MAP_RD);

183 if(!fb) {

184 fprintf(stderr, "FAILED\n");

185 return false;

186 }

187

188 glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, rinf.xsz, rinf.ysz, GL_RGBA, GL_FLOAT, fb);

189 unmap_mem_buffer(mbuf);

190 #endif

191

192 long msec = get_msec() - tm0;

193 timing_sample_sum += msec;

194 num_timing_samples++;

195

196 printf("rendered in %ld msec\n", msec);

197 return true;

198 }

199

200 #define MIN(a, b) ((a) < (b) ? (a) : (b))

201 static void dbg_set_gl_material(Material *mat)

202 {

203 static Material def_mat = {{0.7, 0.7, 0.7, 1}, {0, 0, 0, 0}, 0, 0, 0};

204

205 if(!mat) mat = &def_mat;

206

207 glMaterialfv(GL_FRONT_AND_BACK, GL_AMBIENT_AND_DIFFUSE, mat->kd);

208 glMaterialfv(GL_FRONT_AND_BACK, GL_SPECULAR, mat->ks);

209 glMaterialf(GL_FRONT_AND_BACK, GL_SHININESS, MIN(mat->spow, 128.0f));

210 }

211

212 void dbg_render_gl(Scene *scn, bool show_tree, bool show_obj)

213 {

214 glPushAttrib(GL_ENABLE_BIT | GL_TRANSFORM_BIT | GL_LIGHTING_BIT);

215

216 for(int i=0; i<rinf.num_lights; i++) {

217 float lpos[4];

218

219 memcpy(lpos, lightlist[i].pos, sizeof lpos);

220 lpos[3] = 1.0;

221

222 glLightfv(GL_LIGHT0 + i, GL_POSITION, lpos);

223 glLightfv(GL_LIGHT0 + i, GL_DIFFUSE, lightlist[i].color);

224 glEnable(GL_LIGHT0 + i);

225 }

226

227 glDisable(GL_TEXTURE_2D);

228 glEnable(GL_DEPTH_TEST);

229 glEnable(GL_LIGHTING);

230

231 glMatrixMode(GL_PROJECTION);

232 glPushMatrix();

233 glLoadIdentity();

234 gluPerspective(45.0, (float)rinf.xsz / (float)rinf.ysz, 0.5, 1000.0);

235

236 if(show_obj) {

237 Material *materials = scn->get_materials();

238

239 int num_faces = scn->get_num_faces();

240 int cur_mat = -1;

241

242 for(int i=0; i<num_faces; i++) {

243 if(faces[i].matid != cur_mat) {

244 if(cur_mat != -1) {

245 glEnd();

246 }

247 dbg_set_gl_material(materials ? materials + faces[i].matid : 0);

248 cur_mat = faces[i].matid;

249 glBegin(GL_TRIANGLES);

250 }

251

252 for(int j=0; j<3; j++) {

253 glNormal3fv(faces[i].v[j].normal);

254 glVertex3fv(faces[i].v[j].pos);

255 }

256 }

257 glEnd();

258 }

259

260 if(show_tree) {

261 scn->draw_kdtree();

262 }

263

264 glPopMatrix();

265 glPopAttrib();

266

267 assert(glGetError() == GL_NO_ERROR);

268 }

269

270 void set_xform(float *matrix, float *invtrans)

271 {

272 CLMemBuffer *mbuf_xform = prog->get_arg_buffer(KARG_XFORM);

273 CLMemBuffer *mbuf_invtrans = prog->get_arg_buffer(KARG_INVTRANS_XFORM);

274 assert(mbuf_xform && mbuf_invtrans);

275

276 float *mem = (float*)map_mem_buffer(mbuf_xform, MAP_WR);

277 memcpy(mem, matrix, 16 * sizeof *mem);

278 unmap_mem_buffer(mbuf_xform);

279

280 mem = (float*)map_mem_buffer(mbuf_invtrans, MAP_WR);

281 memcpy(mem, invtrans, 16 * sizeof *mem);

282 unmap_mem_buffer(mbuf_invtrans);

283 }

284

285 void set_render_option(int opt, bool val)

286 {

287 switch(opt) {

288 case ROPT_ITER:

289 case ROPT_REFL:

290 rinf.max_iter = val ? saved_iter_val : 0;

291 break;

292

293 case ROPT_SHAD:

294 rinf.cast_shadows = val;

295 break;

296

297 default:

298 return;

299 }

300

301 update_render_info();

302 }

303

304 void set_render_option(int opt, int val)

305 {

306 switch(opt) {

307 case ROPT_ITER:

308 rinf.max_iter = saved_iter_val = val;

309 break;

310

311 case ROPT_SHAD:

312 rinf.cast_shadows = val;

313 break;

314

315 case ROPT_REFL:

316 rinf.max_iter = val ? saved_iter_val : 0;

317 break;

318

319 default:

320 return;

321 }

322

323 update_render_info();

324 }

325

326 void set_render_option(int opt, float val)

327 {

328 set_render_option(opt, (int)val);

329 }

330

331 bool get_render_option_bool(int opt)

332 {

333 switch(opt) {

334 case ROPT_ITER:

335 return rinf.max_iter;

336 case ROPT_SHAD:

337 return rinf.cast_shadows;

338 case ROPT_REFL:

339 return rinf.max_iter == saved_iter_val;

340 default:

341 break;

342 }

343 return false;

344 }

345

346 int get_render_option_int(int opt)

347 {

348 switch(opt) {

349 case ROPT_ITER:

350 return rinf.max_iter;

351 case ROPT_SHAD:

352 return rinf.cast_shadows ? 1 : 0;

353 case ROPT_REFL:

354 return rinf.max_iter == saved_iter_val ? 1 : 0;

355 default:

356 break;

357 }

358 return -1;

359 }

360

361 float get_render_option_float(int opt)

362 {

363 return (float)get_render_option_int(opt);

364 }

365

366 static void update_render_info()

367 {

368 if(!prog) {

369 return;

370 }

371

372 CLMemBuffer *mbuf = prog->get_arg_buffer(KARG_RENDER_INFO);

373 assert(mbuf);

374

375 RendInfo *rinf_ptr = (RendInfo*)map_mem_buffer(mbuf, MAP_WR);

376 *rinf_ptr = rinf;

377 unmap_mem_buffer(mbuf);

378 }

379

380 static Ray get_primary_ray(int x, int y, int w, int h, float vfov_deg)

381 {

382 float vfov = M_PI * vfov_deg / 180.0;

383 float aspect = (float)w / (float)h;

384

385 float ysz = 2.0;

386 float xsz = aspect * ysz;

387

388 float px = ((float)x / (float)w) * xsz - xsz / 2.0;

389 float py = 1.0 - ((float)y / (float)h) * ysz;

390 float pz = 1.0 / tan(0.5 * vfov);

391

392 float mag = sqrt(px * px + py * py + pz * pz);

393

394 px = px * RAY_MAG / mag;

395 py = py * RAY_MAG / mag;

396 pz = pz * RAY_MAG / mag;

397

398 Ray ray = {{0, 0, 0, 1}, {px, py, -pz, 1}};

399 return ray;

400 }

401

402 static float *create_kdimage(const KDNodeGPU *kdtree, int num_nodes, int *xsz_ret, int *ysz_ret)

403 {

404 int ysz = MIN(num_nodes, KDIMG_MAX_HEIGHT);

405 int columns = (num_nodes - 1) / KDIMG_MAX_HEIGHT + 1;

406 int xsz = KDIMG_NODE_WIDTH * columns;

407

408 printf("creating kdtree image %dx%d (%d nodes)\n", xsz, ysz, num_nodes);

409

410 float *img = new float[4 * xsz * ysz];

411 memset(img, 0, 4 * xsz * ysz * sizeof *img);

412

413 for(int i=0; i<num_nodes; i++) {

414 int x = KDIMG_NODE_WIDTH * (i / KDIMG_MAX_HEIGHT);

415 int y = i % KDIMG_MAX_HEIGHT;

416

417 float *ptr = img + (y * xsz + x) * 4;

418

419 *ptr++ = kdtree[i].aabb.min[0];

420 *ptr++ = kdtree[i].aabb.min[1];

421 *ptr++ = kdtree[i].aabb.min[2];

422 *ptr++ = 0.0;

423

424 *ptr++ = kdtree[i].aabb.max[0];

425 *ptr++ = kdtree[i].aabb.max[1];

426 *ptr++ = kdtree[i].aabb.max[2];

427 *ptr++ = 0.0;

428

429 for(int j=0; j<MAX_NODE_FACES; j++) {

430 *ptr++ = j < kdtree[i].num_faces ? (float)kdtree[i].face_idx[j] : 0.0f;

431 }

432

433 *ptr++ = (float)kdtree[i].num_faces;

434 *ptr++ = (float)kdtree[i].left;

435 *ptr++ = (float)kdtree[i].right;

436 *ptr++ = 0.0;

437 }

438

439 if(xsz_ret) *xsz_ret = xsz;

440 if(ysz_ret) *ysz_ret = ysz;

441 return img;

442 }