svn merge -r 16351:16368 https://svn.blender.org/svnroot/bf-blender/trunk/blender
[blender.git] / source / blender / render / intern / source / rayshade.c
1 /**
2  * $Id$
3  *
4  * ***** BEGIN GPL LICENSE BLOCK *****
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version 2
9  * of the License, or (at your option) any later version. 
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software Foundation,
18  * Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
19  *
20  * The Original Code is Copyright (C) 1990-1998 NeoGeo BV.
21  * All rights reserved.
22  *
23  * Contributors: 2004/2005 Blender Foundation, full recode
24  *
25  * ***** END GPL LICENSE BLOCK *****
26  */
27
28 #include <math.h>
29 #include <string.h>
30 #include <stdlib.h>
31 #include <float.h>
32
33 #include "MEM_guardedalloc.h"
34
35 #include "DNA_material_types.h"
36 #include "DNA_lamp_types.h"
37
38 #include "BKE_global.h"
39 #include "BKE_node.h"
40 #include "BKE_utildefines.h"
41
42 #include "BLI_arithb.h"
43 #include "BLI_blenlib.h"
44 #include "BLI_jitter.h"
45 #include "BLI_rand.h"
46
47 #include "PIL_time.h"
48
49 #include "render_types.h"
50 #include "renderpipeline.h"
51 #include "rendercore.h"
52 #include "renderdatabase.h"
53 #include "pixelblending.h"
54 #include "pixelshading.h"
55 #include "shading.h"
56 #include "texture.h"
57
58 #include "RE_raytrace.h"
59
60 #define RAY_TRA         1
61 #define RAY_TRAFLIP     2
62
63 #define DEPTH_SHADOW_TRA  10
64
65 /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ */
66 /* defined in pipeline.c, is hardcopy of active dynamic allocated Render */
67 /* only to be used here in this file, it's for speed */
68 extern struct Render R;
69 /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ */
70
71 static void vlr_face_coords(RayFace *face, float **v1, float **v2, float **v3, float **v4)
72 {
73         VlakRen *vlr= (VlakRen*)face;
74
75         *v1 = (vlr->v1)? vlr->v1->co: NULL;
76         *v2 = (vlr->v2)? vlr->v2->co: NULL;
77         *v3 = (vlr->v3)? vlr->v3->co: NULL;
78         *v4 = (vlr->v4)? vlr->v4->co: NULL;
79 }
80
81 static int vlr_check_intersect(Isect *is, int ob, RayFace *face)
82 {
83         ObjectInstanceRen *obi= RAY_OBJECT_GET((Render*)is->userdata, ob);
84         VlakRen *vlr = (VlakRen*)face;
85
86         /* for baking selected to active non-traceable materials might still
87          * be in the raytree */
88         if(!(vlr->mat->mode & MA_TRACEBLE))
89                 return 0;
90
91         /* I know... cpu cycle waste, might do smarter once */
92         if(is->mode==RE_RAY_MIRROR)
93                 return !(vlr->mat->mode & MA_ONLYCAST);
94         else
95                 return (is->lay & obi->lay);
96 }
97
98 static float *vlr_get_transform(void *userdata, int i)
99 {
100         ObjectInstanceRen *obi= RAY_OBJECT_GET((Render*)userdata, i);
101
102         return (obi->flag & R_TRANSFORMED)? (float*)obi->mat: NULL;
103 }
104
105 void freeraytree(Render *re)
106 {
107         if(re->raytree) {
108                 RE_ray_tree_free(re->raytree);
109                 re->raytree= NULL;
110         }
111 }
112
113 void makeraytree(Render *re)
114 {
115         ObjectInstanceRen *obi;
116         ObjectRen *obr;
117         VlakRen *vlr= NULL;
118         float min[3], max[3], co1[3], co2[3], co3[3], co4[3];
119         double lasttime= PIL_check_seconds_timer();
120         int v, totv = 0, totface = 0;
121
122         INIT_MINMAX(min, max);
123
124         /* first min max raytree space */
125         for(obi=re->instancetable.first; obi; obi=obi->next) {
126                 obr= obi->obr;
127
128                 if(re->excludeob && obr->ob == re->excludeob)
129                         continue;
130
131                 for(v=0;v<obr->totvlak;v++) {
132                         if((v & 255)==0) vlr= obr->vlaknodes[v>>8].vlak;
133                         else vlr++;
134                         /* baking selected to active needs non-traceable too */
135                         if((re->flag & R_BAKE_TRACE) || (vlr->mat->mode & MA_TRACEBLE)) {       
136                                 if((vlr->mat->mode & MA_WIRE)==0) {     
137                                         VECCOPY(co1, vlr->v1->co);
138                                         VECCOPY(co2, vlr->v2->co);
139                                         VECCOPY(co3, vlr->v3->co);
140
141                                         if(obi->flag & R_TRANSFORMED) {
142                                                 Mat4MulVecfl(obi->mat, co1);
143                                                 Mat4MulVecfl(obi->mat, co2);
144                                                 Mat4MulVecfl(obi->mat, co3);
145                                         }
146
147                                         DO_MINMAX(co1, min, max);
148                                         DO_MINMAX(co2, min, max);
149                                         DO_MINMAX(co3, min, max);
150
151                                         if(vlr->v4) {
152                                                 VECCOPY(co4, vlr->v4->co);
153                                                 if(obi->flag & R_TRANSFORMED)
154                                                         Mat4MulVecfl(obi->mat, co4);
155                                                 DO_MINMAX(co4, min, max);
156                                         }
157
158                                         totface++;
159                                 }
160                         }
161                 }
162         }
163
164         re->raytree= RE_ray_tree_create(re->r.ocres, totface, min, max,
165                 vlr_face_coords, vlr_check_intersect, vlr_get_transform, re);
166
167         if(min[0] > max[0]) { /* empty raytree */
168                 RE_ray_tree_done(re->raytree);
169                 return; 
170         }
171
172         for(obi=re->instancetable.first; obi; obi=obi->next) {
173                 obr= obi->obr;
174
175                 if(re->excludeob && obr->ob == re->excludeob)
176                         continue;
177
178                 for(v=0; v<obr->totvlak; v++, totv++) {
179                         if((v & 255)==0) {
180                                 double time= PIL_check_seconds_timer();
181
182                                 vlr= obr->vlaknodes[v>>8].vlak;
183                                 if(re->test_break())
184                                         break;
185                                 if(time-lasttime>1.0f) {
186                                         char str[32];
187                                         sprintf(str, "Filling Octree: %d", totv);
188                                         re->i.infostr= str;
189                                         re->stats_draw(&re->i);
190                                         re->i.infostr= NULL;
191                                         lasttime= time;
192                                 }
193                         }
194                         else vlr++;
195                         
196                         if((re->flag & R_BAKE_TRACE) || (vlr->mat->mode & MA_TRACEBLE))
197                                 if((vlr->mat->mode & MA_WIRE)==0)
198                                         RE_ray_tree_add_face(re->raytree, RAY_OBJECT_SET(re, obi), vlr);
199                 }
200         }
201
202         RE_ray_tree_done(re->raytree);
203         
204         re->i.infostr= NULL;
205         re->stats_draw(&re->i);
206 }
207
208 static void shade_ray(Isect *is, ShadeInput *shi, ShadeResult *shr)
209 {
210         VlakRen *vlr= (VlakRen*)is->face;
211         ObjectInstanceRen *obi= RAY_OBJECT_GET(&R, is->ob);
212         int osatex= 0;
213         
214         /* set up view vector */
215         VECCOPY(shi->view, is->vec);
216
217         /* render co */
218         shi->co[0]= is->start[0]+is->labda*(shi->view[0]);
219         shi->co[1]= is->start[1]+is->labda*(shi->view[1]);
220         shi->co[2]= is->start[2]+is->labda*(shi->view[2]);
221         
222         Normalize(shi->view);
223
224         shi->obi= obi;
225         shi->obr= obi->obr;
226         shi->vlr= vlr;
227         shi->mat= vlr->mat;
228         memcpy(&shi->r, &shi->mat->r, 23*sizeof(float));        // note, keep this synced with render_types.h
229         shi->har= shi->mat->har;
230         
231         // Osa structs we leave unchanged now
232         SWAP(int, osatex, shi->osatex);
233         
234         shi->dxco[0]= shi->dxco[1]= shi->dxco[2]= 0.0f;
235         shi->dyco[0]= shi->dyco[1]= shi->dyco[2]= 0.0f;
236         
237         // but, set Osa stuff to zero where it can confuse texture code
238         if(shi->mat->texco & (TEXCO_NORM|TEXCO_REFL) ) {
239                 shi->dxno[0]= shi->dxno[1]= shi->dxno[2]= 0.0f;
240                 shi->dyno[0]= shi->dyno[1]= shi->dyno[2]= 0.0f;
241         }
242
243         if(vlr->v4) {
244                 if(is->isect==2) 
245                         shade_input_set_triangle_i(shi, obi, vlr, 2, 1, 3);
246                 else
247                         shade_input_set_triangle_i(shi, obi, vlr, 0, 1, 3);
248         }
249         else {
250                 shade_input_set_triangle_i(shi, obi, vlr, 0, 1, 2);
251         }
252
253         shi->u= is->u;
254         shi->v= is->v;
255         shi->dx_u= shi->dx_v= shi->dy_u= shi->dy_v=  0.0f;
256
257         shade_input_set_normals(shi);
258
259         /* point normals to viewing direction */
260         if(INPR(shi->facenor, shi->view) < 0.0f)
261                 shade_input_flip_normals(shi);
262
263         shade_input_set_shade_texco(shi);
264         
265         if(is->mode==RE_RAY_SHADOW_TRA) 
266                 if(shi->mat->nodetree && shi->mat->use_nodes) {
267                         ntreeShaderExecTree(shi->mat->nodetree, shi, shr);
268                         shi->mat= vlr->mat;             /* shi->mat is being set in nodetree */
269                 }
270                 else
271                         shade_color(shi, shr);
272         else {
273                 if(shi->mat->nodetree && shi->mat->use_nodes) {
274                         ntreeShaderExecTree(shi->mat->nodetree, shi, shr);
275                         shi->mat= vlr->mat;             /* shi->mat is being set in nodetree */
276                 }
277                 else
278                         shade_material_loop(shi, shr);
279                 
280                 /* raytrace likes to separate the spec color */
281                 VECSUB(shr->diff, shr->combined, shr->spec);
282         }       
283         
284         SWAP(int, osatex, shi->osatex);  // XXXXX!!!!
285
286 }
287
288 static int refraction(float *refract, float *n, float *view, float index)
289 {
290         float dot, fac;
291
292         VECCOPY(refract, view);
293         
294         dot= view[0]*n[0] + view[1]*n[1] + view[2]*n[2];
295
296         if(dot>0.0f) {
297                 index = 1.0f/index;
298                 fac= 1.0f - (1.0f - dot*dot)*index*index;
299                 if(fac<= 0.0f) return 0;
300                 fac= -dot*index + sqrt(fac);
301         }
302         else {
303                 fac= 1.0f - (1.0f - dot*dot)*index*index;
304                 if(fac<= 0.0f) return 0;
305                 fac= -dot*index - sqrt(fac);
306         }
307
308         refract[0]= index*view[0] + fac*n[0];
309         refract[1]= index*view[1] + fac*n[1];
310         refract[2]= index*view[2] + fac*n[2];
311
312         return 1;
313 }
314
315 /* orn = original face normal */
316 static void reflection(float *ref, float *n, float *view, float *orn)
317 {
318         float f1;
319         
320         f1= -2.0f*(n[0]*view[0]+ n[1]*view[1]+ n[2]*view[2]);
321         
322         ref[0]= (view[0]+f1*n[0]);
323         ref[1]= (view[1]+f1*n[1]);
324         ref[2]= (view[2]+f1*n[2]);
325
326         if(orn) {
327                 /* test phong normals, then we should prevent vector going to the back */
328                 f1= ref[0]*orn[0]+ ref[1]*orn[1]+ ref[2]*orn[2];
329                 if(f1>0.0f) {
330                         f1+= .01f;
331                         ref[0]-= f1*orn[0];
332                         ref[1]-= f1*orn[1];
333                         ref[2]-= f1*orn[2];
334                 }
335         }
336 }
337
338 #if 0
339 static void color_combine(float *result, float fac1, float fac2, float *col1, float *col2)
340 {
341         float col1t[3], col2t[3];
342         
343         col1t[0]= sqrt(col1[0]);
344         col1t[1]= sqrt(col1[1]);
345         col1t[2]= sqrt(col1[2]);
346         col2t[0]= sqrt(col2[0]);
347         col2t[1]= sqrt(col2[1]);
348         col2t[2]= sqrt(col2[2]);
349
350         result[0]= (fac1*col1t[0] + fac2*col2t[0]);
351         result[0]*= result[0];
352         result[1]= (fac1*col1t[1] + fac2*col2t[1]);
353         result[1]*= result[1];
354         result[2]= (fac1*col1t[2] + fac2*col2t[2]);
355         result[2]*= result[2];
356 }
357 #endif
358
359 static float shade_by_transmission(Isect *is, ShadeInput *shi, ShadeResult *shr)
360 {
361         float dx, dy, dz, d, p;
362
363         if (0 == (shi->mat->mode & (MA_RAYTRANSP|MA_ZTRA)))
364                 return -1;
365            
366         if (shi->mat->tx_limit <= 0.0f) {
367                 d= 1.0f;
368         } 
369         else {
370                 /* shi.co[] calculated by shade_ray() */
371                 dx= shi->co[0] - is->start[0];
372                 dy= shi->co[1] - is->start[1];
373                 dz= shi->co[2] - is->start[2];
374                 d= sqrt(dx*dx+dy*dy+dz*dz);
375                 if (d > shi->mat->tx_limit)
376                         d= shi->mat->tx_limit;
377
378                 p = shi->mat->tx_falloff;
379                 if(p < 0.0f) p= 0.0f;
380                 else if (p > 10.0f) p= 10.0f;
381
382                 shr->alpha *= pow(d, p);
383                 if (shr->alpha > 1.0f)
384                         shr->alpha= 1.0f;
385         }
386
387         return d;
388 }
389
390 static void ray_fadeout_endcolor(float *col, ShadeInput *origshi, ShadeInput *shi, ShadeResult *shr, Isect *isec, float *vec)
391 {
392         /* un-intersected rays get either rendered material color or sky color */
393         if (origshi->mat->fadeto_mir == MA_RAYMIR_FADETOMAT) {
394                 VECCOPY(col, shr->combined);
395         } else if (origshi->mat->fadeto_mir == MA_RAYMIR_FADETOSKY) {
396                 VECCOPY(shi->view, vec);
397                 Normalize(shi->view);
398                 
399                 shadeSkyView(col, isec->start, shi->view, NULL);
400         }
401 }
402
403 static void ray_fadeout(Isect *is, ShadeInput *shi, float *col, float *blendcol, float dist_mir)
404 {
405         /* if fading out, linear blend against fade color */
406         float blendfac;
407
408         blendfac = 1.0 - VecLenf(shi->co, is->start)/dist_mir;
409         
410         col[0] = col[0]*blendfac + (1.0 - blendfac)*blendcol[0];
411         col[1] = col[1]*blendfac + (1.0 - blendfac)*blendcol[1];
412         col[2] = col[2]*blendfac + (1.0 - blendfac)*blendcol[2];
413 }
414
415 /* the main recursive tracer itself */
416 static void traceray(ShadeInput *origshi, ShadeResult *origshr, short depth, float *start, float *vec, float *col, ObjectInstanceRen *obi, VlakRen *vlr, int traflag)
417 {
418         ShadeInput shi;
419         ShadeResult shr;
420         Isect isec;
421         float f, f1, fr, fg, fb;
422         float ref[3], maxsize=RE_ray_tree_max_size(R.raytree);
423         float dist_mir = origshi->mat->dist_mir;
424
425         /* Warning, This is not that nice, and possibly a bit slow for every ray,
426         however some variables were not initialized properly in, unless using shade_input_initialize(...), we need to do a memset */
427         memset(&shi, 0, sizeof(ShadeInput)); 
428         /* end warning! - Campbell */
429         
430         VECCOPY(isec.start, start);
431         if (dist_mir > 0.0) {
432                 isec.end[0]= start[0]+dist_mir*vec[0];
433                 isec.end[1]= start[1]+dist_mir*vec[1];
434                 isec.end[2]= start[2]+dist_mir*vec[2];
435         } else {
436                 isec.end[0]= start[0]+maxsize*vec[0];
437                 isec.end[1]= start[1]+maxsize*vec[1];
438                 isec.end[2]= start[2]+maxsize*vec[2];
439         }
440         isec.mode= RE_RAY_MIRROR;
441         isec.faceorig= (RayFace*)vlr;
442         isec.oborig= RAY_OBJECT_SET(&R, obi);
443
444         if(RE_ray_tree_intersect(R.raytree, &isec)) {
445                 float d= 1.0f;
446                 
447                 shi.mask= origshi->mask;
448                 shi.osatex= origshi->osatex;
449                 shi.depth= 1;                                   /* only used to indicate tracing */
450                 shi.thread= origshi->thread;
451                 //shi.sample= 0; // memset above, so dont need this
452                 shi.xs= origshi->xs;
453                 shi.ys= origshi->ys;
454                 shi.lay= origshi->lay;
455                 shi.passflag= SCE_PASS_COMBINED; /* result of tracing needs no pass info */
456                 shi.combinedflag= 0xFFFFFF;              /* ray trace does all options */
457                 //shi.do_preview= 0; // memset above, so dont need this
458                 shi.light_override= origshi->light_override;
459                 shi.mat_override= origshi->mat_override;
460                 
461                 memset(&shr, 0, sizeof(ShadeResult));
462                 
463                 shade_ray(&isec, &shi, &shr);
464                 if (traflag & RAY_TRA)
465                         d= shade_by_transmission(&isec, &shi, &shr);
466                 
467                 if(depth>0) {
468
469                         if(shi.mat->mode_l & (MA_RAYTRANSP|MA_ZTRA) && shr.alpha < 1.0f) {
470                                 float nf, f, f1, refract[3], tracol[4];
471                                 
472                                 tracol[0]= shi.r;
473                                 tracol[1]= shi.g;
474                                 tracol[2]= shi.b;
475                                 tracol[3]= col[3];      // we pass on and accumulate alpha
476                                 
477                                 if(shi.mat->mode & MA_RAYTRANSP) {
478                                         /* odd depths: use normal facing viewer, otherwise flip */
479                                         if(traflag & RAY_TRAFLIP) {
480                                                 float norm[3];
481                                                 norm[0]= - shi.vn[0];
482                                                 norm[1]= - shi.vn[1];
483                                                 norm[2]= - shi.vn[2];
484                                                 if (!refraction(refract, norm, shi.view, shi.ang))
485                                                         reflection(refract, norm, shi.view, shi.vn);
486                                         }
487                                         else {
488                                                 if (!refraction(refract, shi.vn, shi.view, shi.ang))
489                                                         reflection(refract, shi.vn, shi.view, shi.vn);
490                                         }
491                                         traflag |= RAY_TRA;
492                                         traceray(origshi, origshr, depth-1, shi.co, refract, tracol, shi.obi, shi.vlr, traflag ^ RAY_TRAFLIP);
493                                 }
494                                 else
495                                         traceray(origshi, origshr, depth-1, shi.co, shi.view, tracol, shi.obi, shi.vlr, 0);
496                                 
497                                 f= shr.alpha; f1= 1.0f-f;
498                                 nf= d * shi.mat->filter;
499                                 fr= 1.0f+ nf*(shi.r-1.0f);
500                                 fg= 1.0f+ nf*(shi.g-1.0f);
501                                 fb= 1.0f+ nf*(shi.b-1.0f);
502                                 shr.diff[0]= f*shr.diff[0] + f1*fr*tracol[0];
503                                 shr.diff[1]= f*shr.diff[1] + f1*fg*tracol[1];
504                                 shr.diff[2]= f*shr.diff[2] + f1*fb*tracol[2];
505                                 
506                                 shr.spec[0] *=f;
507                                 shr.spec[1] *=f;
508                                 shr.spec[2] *=f;
509
510                                 col[3]= f1*tracol[3] + f;
511                         }
512                         else 
513                                 col[3]= 1.0f;
514
515                         if(shi.mat->mode_l & MA_RAYMIRROR) {
516                                 f= shi.ray_mirror;
517                                 if(f!=0.0f) f*= fresnel_fac(shi.view, shi.vn, shi.mat->fresnel_mir_i, shi.mat->fresnel_mir);
518                         }
519                         else f= 0.0f;
520                         
521                         if(f!=0.0f) {
522                                 float mircol[4];
523                                 
524                                 reflection(ref, shi.vn, shi.view, NULL);                        
525                                 traceray(origshi, origshr, depth-1, shi.co, ref, mircol, shi.obi, shi.vlr, 0);
526                         
527                                 f1= 1.0f-f;
528
529                                 /* combine */
530                                 //color_combine(col, f*fr*(1.0f-shr.spec[0]), f1, col, shr.diff);
531                                 //col[0]+= shr.spec[0];
532                                 //col[1]+= shr.spec[1];
533                                 //col[2]+= shr.spec[2];
534                                 
535                                 fr= shi.mirr;
536                                 fg= shi.mirg;
537                                 fb= shi.mirb;
538                 
539                                 col[0]= f*fr*(1.0f-shr.spec[0])*mircol[0] + f1*shr.diff[0] + shr.spec[0];
540                                 col[1]= f*fg*(1.0f-shr.spec[1])*mircol[1] + f1*shr.diff[1] + shr.spec[1];
541                                 col[2]= f*fb*(1.0f-shr.spec[2])*mircol[2] + f1*shr.diff[2] + shr.spec[2];
542                         }
543                         else {
544                                 col[0]= shr.diff[0] + shr.spec[0];
545                                 col[1]= shr.diff[1] + shr.spec[1];
546                                 col[2]= shr.diff[2] + shr.spec[2];
547                         }
548                         
549                         if (dist_mir > 0.0) {
550                                 float blendcol[3];
551                                 
552                                 /* max ray distance set, but found an intersection, so fade this color
553                                  * out towards the sky/material color for a smooth transition */
554                                 ray_fadeout_endcolor(blendcol, origshi, &shi, origshr, &isec, vec);
555                                 ray_fadeout(&isec, &shi, col, blendcol, dist_mir);
556                         }
557                 }
558                 else {
559                         col[0]= shr.diff[0] + shr.spec[0];
560                         col[1]= shr.diff[1] + shr.spec[1];
561                         col[2]= shr.diff[2] + shr.spec[2];
562                 }
563                 
564         }
565         else {
566                 ray_fadeout_endcolor(col, origshi, &shi, origshr, &isec, vec);
567         }
568 }
569
570 /* **************** jitter blocks ********** */
571
572 /* calc distributed planar energy */
573
574 static void DP_energy(float *table, float *vec, int tot, float xsize, float ysize)
575 {
576         int x, y, a;
577         float *fp, force[3], result[3];
578         float dx, dy, dist, min;
579         
580         min= MIN2(xsize, ysize);
581         min*= min;
582         result[0]= result[1]= 0.0f;
583         
584         for(y= -1; y<2; y++) {
585                 dy= ysize*y;
586                 for(x= -1; x<2; x++) {
587                         dx= xsize*x;
588                         fp= table;
589                         for(a=0; a<tot; a++, fp+= 2) {
590                                 force[0]= vec[0] - fp[0]-dx;
591                                 force[1]= vec[1] - fp[1]-dy;
592                                 dist= force[0]*force[0] + force[1]*force[1];
593                                 if(dist < min && dist>0.0f) {
594                                         result[0]+= force[0]/dist;
595                                         result[1]+= force[1]/dist;
596                                 }
597                         }
598                 }
599         }
600         vec[0] += 0.1*min*result[0]/(float)tot;
601         vec[1] += 0.1*min*result[1]/(float)tot;
602         // cyclic clamping
603         vec[0]= vec[0] - xsize*floor(vec[0]/xsize + 0.5);
604         vec[1]= vec[1] - ysize*floor(vec[1]/ysize + 0.5);
605 }
606
607 // random offset of 1 in 2
608 static void jitter_plane_offset(float *jitter1, float *jitter2, int tot, float sizex, float sizey, float ofsx, float ofsy)
609 {
610         float dsizex= sizex*ofsx;
611         float dsizey= sizey*ofsy;
612         float hsizex= 0.5*sizex, hsizey= 0.5*sizey;
613         int x;
614         
615         for(x=tot; x>0; x--, jitter1+=2, jitter2+=2) {
616                 jitter2[0]= jitter1[0] + dsizex;
617                 jitter2[1]= jitter1[1] + dsizey;
618                 if(jitter2[0] > hsizex) jitter2[0]-= sizex;
619                 if(jitter2[1] > hsizey) jitter2[1]-= sizey;
620         }
621 }
622
623 /* called from convertBlenderScene.c */
624 /* we do this in advance to get consistant random, not alter the render seed, and be threadsafe */
625 void init_jitter_plane(LampRen *lar)
626 {
627         float *fp;
628         int x, iter=12, tot= lar->ray_totsamp;
629         
630         /* test if already initialized */
631         if(lar->jitter) return;
632         
633         /* at least 4, or max threads+1 tables */
634         if(BLENDER_MAX_THREADS < 4) x= 4;
635         else x= BLENDER_MAX_THREADS+1;
636         fp= lar->jitter= MEM_mallocN(x*tot*2*sizeof(float), "lamp jitter tab");
637         
638         /* set per-lamp fixed seed */
639         BLI_srandom(tot);
640         
641         /* fill table with random locations, area_size large */
642         for(x=0; x<tot; x++, fp+=2) {
643                 fp[0]= (BLI_frand()-0.5)*lar->area_size;
644                 fp[1]= (BLI_frand()-0.5)*lar->area_sizey;
645         }
646         
647         while(iter--) {
648                 fp= lar->jitter;
649                 for(x=tot; x>0; x--, fp+=2) {
650                         DP_energy(lar->jitter, fp, tot, lar->area_size, lar->area_sizey);
651                 }
652         }
653         
654         /* create the dithered tables (could just check lamp type!) */
655         jitter_plane_offset(lar->jitter, lar->jitter+2*tot, tot, lar->area_size, lar->area_sizey, 0.5f, 0.0f);
656         jitter_plane_offset(lar->jitter, lar->jitter+4*tot, tot, lar->area_size, lar->area_sizey, 0.5f, 0.5f);
657         jitter_plane_offset(lar->jitter, lar->jitter+6*tot, tot, lar->area_size, lar->area_sizey, 0.0f, 0.5f);
658 }
659
660 /* table around origin, -0.5*size to 0.5*size */
661 static float *give_jitter_plane(LampRen *lar, int thread, int xs, int ys)
662 {
663         int tot;
664         
665         tot= lar->ray_totsamp;
666                         
667         if(lar->ray_samp_type & LA_SAMP_JITTER) {
668                 /* made it threadsafe */
669                 
670                 if(lar->xold[thread]!=xs || lar->yold[thread]!=ys) {
671                         jitter_plane_offset(lar->jitter, lar->jitter+2*(thread+1)*tot, tot, lar->area_size, lar->area_sizey, BLI_thread_frand(thread), BLI_thread_frand(thread));
672                         lar->xold[thread]= xs; 
673                         lar->yold[thread]= ys;
674                 }
675                 return lar->jitter+2*(thread+1)*tot;
676         }
677         if(lar->ray_samp_type & LA_SAMP_DITHER) {
678                 return lar->jitter + 2*tot*((xs & 1)+2*(ys & 1));
679         }
680         
681         return lar->jitter;
682 }
683
684
685 /* **************** QMC sampling *************** */
686
687 static void halton_sample(double *ht_invprimes, double *ht_nums, double *v)
688 {
689         // incremental halton sequence generator, from:
690         // "Instant Radiosity", Keller A.
691         unsigned int i;
692         
693         for (i = 0; i < 2; i++)
694         {
695                 double r = fabs((1.0 - ht_nums[i]) - 1e-10);
696                 
697                 if (ht_invprimes[i] >= r)
698                 {
699                         double lasth;
700                         double h = ht_invprimes[i];
701                         
702                         do {
703                                 lasth = h;
704                                 h *= ht_invprimes[i];
705                         } while (h >= r);
706                         
707                         ht_nums[i] += ((lasth + h) - 1.0);
708                 }
709                 else
710                         ht_nums[i] += ht_invprimes[i];
711                 
712                 v[i] = (float)ht_nums[i];
713         }
714 }
715
716 /* Generate Hammersley points in [0,1)^2
717  * From Lucille renderer */
718 static void hammersley_create(double *out, int n)
719 {
720         double p, t;
721         int k, kk;
722
723         for (k = 0; k < n; k++) {
724                 t = 0;
725                 for (p = 0.5, kk = k; kk; p *= 0.5, kk >>= 1) {
726                         if (kk & 1) {           /* kk mod 2 = 1         */
727                                 t += p;
728                         }
729                 }
730         
731                 out[2 * k + 0] = (double)k / (double)n;
732                 out[2 * k + 1] = t;
733         }
734 }
735
736 struct QMCSampler *QMC_initSampler(int type, int tot)
737 {       
738         QMCSampler *qsa = MEM_callocN(sizeof(QMCSampler), "qmc sampler");
739         qsa->samp2d = MEM_callocN(2*sizeof(double)*tot, "qmc sample table");
740
741         qsa->tot = tot;
742         qsa->type = type;
743         
744         if (qsa->type==SAMP_TYPE_HAMMERSLEY) 
745                 hammersley_create(qsa->samp2d, qsa->tot);
746                 
747         return qsa;
748 }
749
750 static void QMC_initPixel(QMCSampler *qsa, int thread)
751 {
752         if (qsa->type==SAMP_TYPE_HAMMERSLEY)
753         {
754                 /* hammersley sequence is fixed, already created in QMCSampler init.
755                  * per pixel, gets a random offset. We create separate offsets per thread, for write-safety */
756                 qsa->offs[thread][0] = 0.5 * BLI_thread_frand(thread);
757                 qsa->offs[thread][1] = 0.5 * BLI_thread_frand(thread);
758         }
759         else {  /* SAMP_TYPE_HALTON */
760                 
761                 /* generate a new randomised halton sequence per pixel
762                  * to alleviate qmc artifacts and make it reproducable 
763                  * between threads/frames */
764                 double ht_invprimes[2], ht_nums[2];
765                 double r[2];
766                 int i;
767         
768                 ht_nums[0] = BLI_thread_frand(thread);
769                 ht_nums[1] = BLI_thread_frand(thread);
770                 ht_invprimes[0] = 0.5;
771                 ht_invprimes[1] = 1.0/3.0;
772                 
773                 for (i=0; i< qsa->tot; i++) {
774                         halton_sample(ht_invprimes, ht_nums, r);
775                         qsa->samp2d[2*i+0] = r[0];
776                         qsa->samp2d[2*i+1] = r[1];
777                 }
778         }
779 }
780
781 static void QMC_freeSampler(QMCSampler *qsa)
782 {
783         MEM_freeN(qsa->samp2d);
784         MEM_freeN(qsa);
785 }
786
787 static void QMC_getSample(double *s, QMCSampler *qsa, int thread, int num)
788 {
789         if (qsa->type == SAMP_TYPE_HAMMERSLEY) {
790                 s[0] = fmod(qsa->samp2d[2*num+0] + qsa->offs[thread][0], 1.0f);
791                 s[1] = fmod(qsa->samp2d[2*num+1] + qsa->offs[thread][1], 1.0f);
792         }
793         else { /* SAMP_TYPE_HALTON */
794                 s[0] = qsa->samp2d[2*num+0];
795                 s[1] = qsa->samp2d[2*num+1];
796         }
797 }
798
799 /* phong weighted disc using 'blur' for exponent, centred on 0,0 */
800 static void QMC_samplePhong(float *vec, QMCSampler *qsa, int thread, int num, float blur)
801 {
802         double s[2];
803         float phi, pz, sqr;
804         
805         QMC_getSample(s, qsa, thread, num);
806
807         phi = s[0]*2*M_PI;
808         pz = pow(s[1], blur);
809         sqr = sqrt(1.0f-pz*pz);
810
811         vec[0] = cos(phi)*sqr;
812         vec[1] = sin(phi)*sqr;
813         vec[2] = 0.0f;
814 }
815
816 /* rect of edge lengths sizex, sizey, centred on 0.0,0.0 i.e. ranging from -sizex/2 to +sizey/2 */
817 static void QMC_sampleRect(float *vec, QMCSampler *qsa, int thread, int num, float sizex, float sizey)
818 {
819         double s[2];
820
821         QMC_getSample(s, qsa, thread, num);
822                 
823         vec[0] = (s[0] - 0.5) * sizex;
824         vec[1] = (s[1] - 0.5) * sizey;
825         vec[2] = 0.0f;
826 }
827
828 /* disc of radius 'radius', centred on 0,0 */
829 static void QMC_sampleDisc(float *vec, QMCSampler *qsa, int thread, int num, float radius)
830 {
831         double s[2];
832         float phi, sqr;
833         
834         QMC_getSample(s, qsa, thread, num);
835         
836         phi = s[0]*2*M_PI;
837         sqr = sqrt(s[1]);
838
839         vec[0] = cos(phi)*sqr* radius/2.0;
840         vec[1] = sin(phi)*sqr* radius/2.0;
841         vec[2] = 0.0f;
842 }
843
844 /* uniform hemisphere sampling */
845 static void QMC_sampleHemi(float *vec, QMCSampler *qsa, int thread, int num)
846 {
847         double s[2];
848         float phi, sqr;
849         
850         QMC_getSample(s, qsa, thread, num);
851         
852         phi = s[0]*2.f*M_PI;    
853         sqr = sqrt(s[1]);
854
855         vec[0] = cos(phi)*sqr;
856         vec[1] = sin(phi)*sqr;
857         vec[2] = 1.f - s[1]*s[1];
858 }
859
860 #if 0 /* currently not used */
861 /* cosine weighted hemisphere sampling */
862 static void QMC_sampleHemiCosine(float *vec, QMCSampler *qsa, int thread, int num)
863 {
864         double s[2];
865         float phi, sqr;
866         
867         QMC_getSample(s, qsa, thread, num);
868         
869         phi = s[0]*2.f*M_PI;    
870         sqr = s[1]*sqrt(2-s[1]*s[1]);
871
872         vec[0] = cos(phi)*sqr;
873         vec[1] = sin(phi)*sqr;
874         vec[2] = 1.f - s[1]*s[1];
875
876 }
877 #endif
878
879 /* called from convertBlenderScene.c */
880 void init_render_qmcsampler(Render *re)
881 {
882         re->qmcsamplers= MEM_callocN(sizeof(ListBase)*BLENDER_MAX_THREADS, "QMCListBase");
883 }
884
885 QMCSampler *get_thread_qmcsampler(Render *re, int thread, int type, int tot)
886 {
887         QMCSampler *qsa;
888
889         /* create qmc samplers as needed, since recursion makes it hard to
890          * predict how many are needed */
891
892         for(qsa=re->qmcsamplers[thread].first; qsa; qsa=qsa->next) {
893                 if(qsa->type == type && qsa->tot == tot && !qsa->used) {
894                         qsa->used= 1;
895                         return qsa;
896                 }
897         }
898
899         qsa= QMC_initSampler(type, tot);
900         qsa->used= 1;
901         BLI_addtail(&re->qmcsamplers[thread], qsa);
902
903         return qsa;
904 }
905
906 void release_thread_qmcsampler(Render *re, int thread, QMCSampler *qsa)
907 {
908         qsa->used= 0;
909 }
910
911 void free_render_qmcsampler(Render *re)
912 {
913         QMCSampler *qsa, *next;
914         int a;
915
916         if(re->qmcsamplers) {
917                 for(a=0; a<BLENDER_MAX_THREADS; a++) {
918                         for(qsa=re->qmcsamplers[a].first; qsa; qsa=next) {
919                                 next= qsa->next;
920                                 QMC_freeSampler(qsa);
921                         }
922
923                         re->qmcsamplers[a].first= re->qmcsamplers[a].last= NULL;
924                 }
925
926                 MEM_freeN(re->qmcsamplers);
927                 re->qmcsamplers= NULL;
928         }
929 }
930
931 static int adaptive_sample_variance(int samples, float *col, float *colsq, float thresh)
932 {
933         float var[3], mean[3];
934
935         /* scale threshold just to give a bit more precision in input rather than dealing with 
936          * tiny tiny numbers in the UI */
937         thresh /= 2;
938         
939         mean[0] = col[0] / (float)samples;
940         mean[1] = col[1] / (float)samples;
941         mean[2] = col[2] / (float)samples;
942
943         var[0] = (colsq[0] / (float)samples) - (mean[0]*mean[0]);
944         var[1] = (colsq[1] / (float)samples) - (mean[1]*mean[1]);
945         var[2] = (colsq[2] / (float)samples) - (mean[2]*mean[2]);
946         
947         if ((var[0] * 0.4 < thresh) && (var[1] * 0.3 < thresh) && (var[2] * 0.6 < thresh))
948                 return 1;
949         else
950                 return 0;
951 }
952
953 static int adaptive_sample_contrast_val(int samples, float prev, float val, float thresh)
954 {
955         /* if the last sample's contribution to the total value was below a small threshold
956          * (i.e. the samples taken are very similar), then taking more samples that are probably 
957          * going to be the same is wasting effort */
958         if (fabs( prev/(float)(samples-1) - val/(float)samples ) < thresh) {
959                 return 1;
960         } else
961                 return 0;
962 }
963
964 static float get_avg_speed(ShadeInput *shi)
965 {
966         float pre_x, pre_y, post_x, post_y, speedavg;
967         
968         pre_x = (shi->winspeed[0] == PASS_VECTOR_MAX)?0.0:shi->winspeed[0];
969         pre_y = (shi->winspeed[1] == PASS_VECTOR_MAX)?0.0:shi->winspeed[1];
970         post_x = (shi->winspeed[2] == PASS_VECTOR_MAX)?0.0:shi->winspeed[2];
971         post_y = (shi->winspeed[3] == PASS_VECTOR_MAX)?0.0:shi->winspeed[3];
972         
973         speedavg = (sqrt(pre_x*pre_x + pre_y*pre_y) + sqrt(post_x*post_x + post_y*post_y)) / 2.0;
974         
975         return speedavg;
976 }
977
978 /* ***************** main calls ************** */
979
980
981 static void trace_refract(float *col, ShadeInput *shi, ShadeResult *shr)
982 {
983         QMCSampler *qsa=NULL;
984         int samp_type;
985         
986         float samp3d[3], orthx[3], orthy[3];
987         float v_refract[3], v_refract_new[3];
988         float sampcol[4], colsq[4];
989         
990         float blur = pow(1.0 - shi->mat->gloss_tra, 3);
991         short max_samples = shi->mat->samp_gloss_tra;
992         float adapt_thresh = shi->mat->adapt_thresh_tra;
993         
994         int samples=0;
995         
996         colsq[0] = colsq[1] = colsq[2] = 0.0;
997         col[0] = col[1] = col[2] = 0.0;
998         col[3]= shr->alpha;
999         
1000         if (blur > 0.0) {
1001                 if (adapt_thresh != 0.0) samp_type = SAMP_TYPE_HALTON;
1002                 else samp_type = SAMP_TYPE_HAMMERSLEY;
1003                         
1004                 /* all samples are generated per pixel */
1005                 qsa = get_thread_qmcsampler(&R, shi->thread, samp_type, max_samples);
1006                 QMC_initPixel(qsa, shi->thread);
1007         } else 
1008                 max_samples = 1;
1009         
1010
1011         while (samples < max_samples) {         
1012                 refraction(v_refract, shi->vn, shi->view, shi->ang);
1013                 
1014                 if (max_samples > 1) {
1015                         /* get a quasi-random vector from a phong-weighted disc */
1016                         QMC_samplePhong(samp3d, qsa, shi->thread, samples, blur);
1017                                                 
1018                         VecOrthoBasisf(v_refract, orthx, orthy);
1019                         VecMulf(orthx, samp3d[0]);
1020                         VecMulf(orthy, samp3d[1]);
1021                                 
1022                         /* and perturb the refraction vector in it */
1023                         VecAddf(v_refract_new, v_refract, orthx);
1024                         VecAddf(v_refract_new, v_refract_new, orthy);
1025                         
1026                         Normalize(v_refract_new);
1027                 } else {
1028                         /* no blurriness, use the original normal */
1029                         VECCOPY(v_refract_new, v_refract);
1030                 }
1031         
1032                 traceray(shi, shr, shi->mat->ray_depth_tra, shi->co, v_refract_new, sampcol, shi->obi, shi->vlr, RAY_TRA|RAY_TRAFLIP);
1033         
1034                 col[0] += sampcol[0];
1035                 col[1] += sampcol[1];
1036                 col[2] += sampcol[2];
1037                 col[3] += sampcol[3];
1038                 
1039                 /* for variance calc */
1040                 colsq[0] += sampcol[0]*sampcol[0];
1041                 colsq[1] += sampcol[1]*sampcol[1];
1042                 colsq[2] += sampcol[2]*sampcol[2];
1043                 
1044                 samples++;
1045                 
1046                 /* adaptive sampling */
1047                 if (adapt_thresh < 1.0 && samples > max_samples/2) 
1048                 {
1049                         if (adaptive_sample_variance(samples, col, colsq, adapt_thresh))
1050                                 break;
1051                         
1052                         /* if the pixel so far is very dark, we can get away with less samples */
1053                         if ( (col[0] + col[1] + col[2])/3.0/(float)samples < 0.01 )
1054                                 max_samples--;
1055                 }
1056         }
1057         
1058         col[0] /= (float)samples;
1059         col[1] /= (float)samples;
1060         col[2] /= (float)samples;
1061         col[3] /= (float)samples;
1062         
1063         if (qsa)
1064                 release_thread_qmcsampler(&R, shi->thread, qsa);
1065 }
1066
1067 static void trace_reflect(float *col, ShadeInput *shi, ShadeResult *shr, float fresnelfac)
1068 {
1069         QMCSampler *qsa=NULL;
1070         int samp_type;
1071         
1072         float samp3d[3], orthx[3], orthy[3];
1073         float v_nor_new[3], v_reflect[3];
1074         float sampcol[4], colsq[4];
1075                 
1076         float blur = pow(1.0 - shi->mat->gloss_mir, 3);
1077         short max_samples = shi->mat->samp_gloss_mir;
1078         float adapt_thresh = shi->mat->adapt_thresh_mir;
1079         float aniso = 1.0 - shi->mat->aniso_gloss_mir;
1080         
1081         int samples=0;
1082         
1083         col[0] = col[1] = col[2] = 0.0;
1084         colsq[0] = colsq[1] = colsq[2] = 0.0;
1085         
1086         if (blur > 0.0) {
1087                 if (adapt_thresh != 0.0) samp_type = SAMP_TYPE_HALTON;
1088                 else samp_type = SAMP_TYPE_HAMMERSLEY;
1089                         
1090                 /* all samples are generated per pixel */
1091                 qsa = get_thread_qmcsampler(&R, shi->thread, samp_type, max_samples);
1092                 QMC_initPixel(qsa, shi->thread);
1093         } else 
1094                 max_samples = 1;
1095         
1096         while (samples < max_samples) {
1097                                 
1098                 if (max_samples > 1) {
1099                         /* get a quasi-random vector from a phong-weighted disc */
1100                         QMC_samplePhong(samp3d, qsa, shi->thread, samples, blur);
1101
1102                         /* find the normal's perpendicular plane, blurring along tangents
1103                          * if tangent shading enabled */
1104                         if (shi->mat->mode & (MA_TANGENT_V)) {
1105                                 Crossf(orthx, shi->vn, shi->tang);      // bitangent
1106                                 VECCOPY(orthy, shi->tang);
1107                                 VecMulf(orthx, samp3d[0]);
1108                                 VecMulf(orthy, samp3d[1]*aniso);
1109                         } else {
1110                                 VecOrthoBasisf(shi->vn, orthx, orthy);
1111                                 VecMulf(orthx, samp3d[0]);
1112                                 VecMulf(orthy, samp3d[1]);
1113                         }
1114
1115                         /* and perturb the normal in it */
1116                         VecAddf(v_nor_new, shi->vn, orthx);
1117                         VecAddf(v_nor_new, v_nor_new, orthy);
1118                         Normalize(v_nor_new);
1119                 } else {
1120                         /* no blurriness, use the original normal */
1121                         VECCOPY(v_nor_new, shi->vn);
1122                 }
1123                 
1124                 if((shi->vlr->flag & R_SMOOTH)) 
1125                         reflection(v_reflect, v_nor_new, shi->view, shi->facenor);
1126                 else
1127                         reflection(v_reflect, v_nor_new, shi->view, NULL);
1128                 
1129                 traceray(shi, shr, shi->mat->ray_depth, shi->co, v_reflect, sampcol, shi->obi, shi->vlr, 0);
1130
1131                 
1132                 col[0] += sampcol[0];
1133                 col[1] += sampcol[1];
1134                 col[2] += sampcol[2];
1135         
1136                 /* for variance calc */
1137                 colsq[0] += sampcol[0]*sampcol[0];
1138                 colsq[1] += sampcol[1]*sampcol[1];
1139                 colsq[2] += sampcol[2]*sampcol[2];
1140                 
1141                 samples++;
1142
1143                 /* adaptive sampling */
1144                 if (adapt_thresh > 0.0 && samples > max_samples/3) 
1145                 {
1146                         if (adaptive_sample_variance(samples, col, colsq, adapt_thresh))
1147                                 break;
1148                         
1149                         /* if the pixel so far is very dark, we can get away with less samples */
1150                         if ( (col[0] + col[1] + col[2])/3.0/(float)samples < 0.01 )
1151                                 max_samples--;
1152                 
1153                         /* reduce samples when reflection is dim due to low ray mirror blend value or fresnel factor
1154                          * and when reflection is blurry */
1155                         if (fresnelfac < 0.1 * (blur+1)) {
1156                                 max_samples--;
1157                                 
1158                                 /* even more for very dim */
1159                                 if (fresnelfac < 0.05 * (blur+1)) 
1160                                         max_samples--;
1161                         }
1162                 }
1163         }
1164         
1165         col[0] /= (float)samples;
1166         col[1] /= (float)samples;
1167         col[2] /= (float)samples;
1168         
1169         if (qsa)
1170                 release_thread_qmcsampler(&R, shi->thread, qsa);
1171 }
1172
1173 /* extern call from render loop */
1174 void ray_trace(ShadeInput *shi, ShadeResult *shr)
1175 {
1176         VlakRen *vlr;
1177         float i, f, f1, fr, fg, fb;
1178         float mircol[4], tracol[4];
1179         float diff[3];
1180         int do_tra, do_mir;
1181         
1182         do_tra= ((shi->mat->mode & (MA_RAYTRANSP)) && shr->alpha!=1.0f);
1183         do_mir= ((shi->mat->mode & MA_RAYMIRROR) && shi->ray_mirror!=0.0f);
1184         vlr= shi->vlr;
1185         
1186         /* raytrace mirror amd refract like to separate the spec color */
1187         if(shi->combinedflag & SCE_PASS_SPEC)
1188                 VECSUB(diff, shr->combined, shr->spec) /* no ; */
1189         else
1190                 VECCOPY(diff, shr->combined);
1191         
1192         if(do_tra) {
1193                 float olddiff[3];
1194                 
1195                 trace_refract(tracol, shi, shr);
1196                 
1197                 f= shr->alpha; f1= 1.0f-f;
1198                 fr= 1.0f+ shi->mat->filter*(shi->r-1.0f);
1199                 fg= 1.0f+ shi->mat->filter*(shi->g-1.0f);
1200                 fb= 1.0f+ shi->mat->filter*(shi->b-1.0f);
1201                 
1202                 /* for refract pass */
1203                 VECCOPY(olddiff, diff);
1204                 
1205                 diff[0]= f*diff[0] + f1*fr*tracol[0];
1206                 diff[1]= f*diff[1] + f1*fg*tracol[1];
1207                 diff[2]= f*diff[2] + f1*fb*tracol[2];
1208                 
1209                 if(shi->passflag & SCE_PASS_REFRACT)
1210                         VECSUB(shr->refr, diff, olddiff);
1211                 
1212                 if(!(shi->combinedflag & SCE_PASS_REFRACT))
1213                         VECSUB(diff, diff, shr->refr);
1214                 
1215                 shr->alpha= tracol[3];
1216         }
1217         
1218         if(do_mir) {
1219         
1220                 i= shi->ray_mirror*fresnel_fac(shi->view, shi->vn, shi->mat->fresnel_mir_i, shi->mat->fresnel_mir);
1221                 if(i!=0.0f) {
1222                 
1223                         trace_reflect(mircol, shi, shr, i);
1224                         
1225                         fr= i*shi->mirr;
1226                         fg= i*shi->mirg;
1227                         fb= i*shi->mirb;
1228
1229                         if(shi->passflag & SCE_PASS_REFLECT) {
1230                                 /* mirror pass is not blocked out with spec */
1231                                 shr->refl[0]= fr*mircol[0] - fr*diff[0];
1232                                 shr->refl[1]= fg*mircol[1] - fg*diff[1];
1233                                 shr->refl[2]= fb*mircol[2] - fb*diff[2];
1234                         }
1235                         
1236                         if(shi->combinedflag & SCE_PASS_REFLECT) {
1237                                 
1238                                 f= fr*(1.0f-shr->spec[0]);      f1= 1.0f-i;
1239                                 diff[0]= f*mircol[0] + f1*diff[0];
1240                                 
1241                                 f= fg*(1.0f-shr->spec[1]);      f1= 1.0f-i;
1242                                 diff[1]= f*mircol[1] + f1*diff[1];
1243                                 
1244                                 f= fb*(1.0f-shr->spec[2]);      f1= 1.0f-i;
1245                                 diff[2]= f*mircol[2] + f1*diff[2];
1246                         }
1247                 }
1248         }
1249         /* put back together */
1250         if(shi->combinedflag & SCE_PASS_SPEC)
1251                 VECADD(shr->combined, diff, shr->spec) /* no ; */
1252         else
1253                 VECCOPY(shr->combined, diff);
1254 }
1255
1256 /* color 'shadfac' passes through 'col' with alpha and filter */
1257 /* filter is only applied on alpha defined transparent part */
1258 static void addAlphaLight(float *shadfac, float *col, float alpha, float filter)
1259 {
1260         float fr, fg, fb;
1261         
1262         fr= 1.0f+ filter*(col[0]-1.0f);
1263         fg= 1.0f+ filter*(col[1]-1.0f);
1264         fb= 1.0f+ filter*(col[2]-1.0f);
1265         
1266         shadfac[0]= alpha*col[0] + fr*(1.0f-alpha)*shadfac[0];
1267         shadfac[1]= alpha*col[1] + fg*(1.0f-alpha)*shadfac[1];
1268         shadfac[2]= alpha*col[2] + fb*(1.0f-alpha)*shadfac[2];
1269         
1270         shadfac[3]= (1.0f-alpha)*shadfac[3];
1271 }
1272
1273 static void ray_trace_shadow_tra(Isect *is, int depth, int traflag)
1274 {
1275         /* ray to lamp, find first face that intersects, check alpha properties,
1276            if it has col[3]>0.0f  continue. so exit when alpha is full */
1277         ShadeInput shi;
1278         ShadeResult shr;
1279
1280         if(RE_ray_tree_intersect(R.raytree, is)) {
1281                 float d= 1.0f;
1282                 /* we got a face */
1283                 
1284                 /* Warning, This is not that nice, and possibly a bit slow for every ray,
1285                 however some variables were not initialized properly in, unless using shade_input_initialize(...), we need to do a memset */
1286                 memset(&shi, 0, sizeof(ShadeInput)); 
1287                 /* end warning! - Campbell */
1288                 
1289                 shi.depth= 1;                                   /* only used to indicate tracing */
1290                 shi.mask= 1;
1291                 
1292                 /*shi.osatex= 0;
1293                 shi.thread= shi.sample= 0;
1294                 shi.lay= 0;
1295                 shi.passflag= 0;
1296                 shi.combinedflag= 0;
1297                 shi.do_preview= 0;
1298                 shi.light_override= NULL;
1299                 shi.mat_override= NULL;*/
1300                 
1301                 shade_ray(is, &shi, &shr);
1302                 if (traflag & RAY_TRA)
1303                         d= shade_by_transmission(is, &shi, &shr);
1304                 
1305                 /* mix colors based on shadfac (rgb + amount of light factor) */
1306                 addAlphaLight(is->col, shr.diff, shr.alpha, d*shi.mat->filter);
1307                 
1308                 if(depth>0 && is->col[3]>0.0f) {
1309                         
1310                         /* adapt isect struct */
1311                         VECCOPY(is->start, shi.co);
1312                         is->oborig= RAY_OBJECT_SET(&R, shi.obi);
1313                         is->faceorig= (RayFace*)shi.vlr;
1314
1315                         ray_trace_shadow_tra(is, depth-1, traflag | RAY_TRA);
1316                 }
1317         }
1318 }
1319
1320 /* not used, test function for ambient occlusion (yaf: pathlight) */
1321 /* main problem; has to be called within shading loop, giving unwanted recursion */
1322 int ray_trace_shadow_rad(ShadeInput *ship, ShadeResult *shr)
1323 {
1324         static int counter=0, only_one= 0;
1325         extern float hashvectf[];
1326         Isect isec;
1327         ShadeInput shi;
1328         ShadeResult shr_t;
1329         float vec[3], accum[3], div= 0.0f, maxsize= RE_ray_tree_max_size(R.raytree);
1330         int a;
1331         
1332         if(only_one) {
1333                 return 0;
1334         }
1335         only_one= 1;
1336         
1337         accum[0]= accum[1]= accum[2]= 0.0f;
1338         isec.mode= RE_RAY_MIRROR;
1339         isec.faceorig= (RayFace*)ship->vlr;
1340         isec.oborig= RAY_OBJECT_SET(&R, ship->obi);
1341         
1342         for(a=0; a<8*8; a++) {
1343                 
1344                 counter+=3;
1345                 counter %= 768;
1346                 VECCOPY(vec, hashvectf+counter);
1347                 if(ship->vn[0]*vec[0]+ship->vn[1]*vec[1]+ship->vn[2]*vec[2]>0.0f) {
1348                         vec[0]-= vec[0];
1349                         vec[1]-= vec[1];
1350                         vec[2]-= vec[2];
1351                 }
1352                 VECCOPY(isec.start, ship->co);
1353                 isec.end[0]= isec.start[0] + maxsize*vec[0];
1354                 isec.end[1]= isec.start[1] + maxsize*vec[1];
1355                 isec.end[2]= isec.start[2] + maxsize*vec[2];
1356                 
1357                 if(RE_ray_tree_intersect(R.raytree, &isec)) {
1358                         float fac;
1359                         
1360                         /* Warning, This is not that nice, and possibly a bit slow for every ray,
1361                         however some variables were not initialized properly in, unless using shade_input_initialize(...), we need to do a memset */
1362                         memset(&shi, 0, sizeof(ShadeInput)); 
1363                         /* end warning! - Campbell */
1364                         
1365                         shade_ray(&isec, &shi, &shr_t);
1366                         fac= isec.labda*isec.labda;
1367                         fac= 1.0f;
1368                         accum[0]+= fac*(shr_t.diff[0]+shr_t.spec[0]);
1369                         accum[1]+= fac*(shr_t.diff[1]+shr_t.spec[1]);
1370                         accum[2]+= fac*(shr_t.diff[2]+shr_t.spec[2]);
1371                         div+= fac;
1372                 }
1373                 else div+= 1.0f;
1374         }
1375         
1376         if(div!=0.0f) {
1377                 shr->diff[0]+= accum[0]/div;
1378                 shr->diff[1]+= accum[1]/div;
1379                 shr->diff[2]+= accum[2]/div;
1380         }
1381         shr->alpha= 1.0f;
1382         
1383         only_one= 0;
1384         return 1;
1385 }
1386
1387 /* aolight: function to create random unit sphere vectors for total random sampling */
1388 static void RandomSpherical(float *v)
1389 {
1390         float r;
1391         v[2] = 2.f*BLI_frand()-1.f;
1392         if ((r = 1.f - v[2]*v[2])>0.f) {
1393                 float a = 6.283185307f*BLI_frand();
1394                 r = sqrt(r);
1395                 v[0] = r * cos(a);
1396                 v[1] = r * sin(a);
1397         }
1398         else v[2] = 1.f;
1399 }
1400
1401 /* calc distributed spherical energy */
1402 static void DS_energy(float *sphere, int tot, float *vec)
1403 {
1404         float *fp, fac, force[3], res[3];
1405         int a;
1406         
1407         res[0]= res[1]= res[2]= 0.0f;
1408         
1409         for(a=0, fp=sphere; a<tot; a++, fp+=3) {
1410                 VecSubf(force, vec, fp);
1411                 fac= force[0]*force[0] + force[1]*force[1] + force[2]*force[2];
1412                 if(fac!=0.0f) {
1413                         fac= 1.0f/fac;
1414                         res[0]+= fac*force[0];
1415                         res[1]+= fac*force[1];
1416                         res[2]+= fac*force[2];
1417                 }
1418         }
1419
1420         VecMulf(res, 0.5);
1421         VecAddf(vec, vec, res);
1422         Normalize(vec);
1423         
1424 }
1425
1426 /* called from convertBlenderScene.c */
1427 /* creates an equally distributed spherical sample pattern */
1428 /* and allocates threadsafe memory */
1429 void init_ao_sphere(World *wrld)
1430 {
1431         float *fp;
1432         int a, tot, iter= 16;
1433
1434         /* we make twice the amount of samples, because only a hemisphere is used */
1435         tot= 2*wrld->aosamp*wrld->aosamp;
1436         
1437         wrld->aosphere= MEM_mallocN(3*tot*sizeof(float), "AO sphere");
1438         
1439         /* fixed random */
1440         BLI_srandom(tot);
1441         
1442         /* init */
1443         fp= wrld->aosphere;
1444         for(a=0; a<tot; a++, fp+= 3) {
1445                 RandomSpherical(fp);
1446         }
1447         
1448         while(iter--) {
1449                 for(a=0, fp= wrld->aosphere; a<tot; a++, fp+= 3) {
1450                         DS_energy(wrld->aosphere, tot, fp);
1451                 }
1452         }
1453         
1454         /* tables */
1455         wrld->aotables= MEM_mallocN(BLENDER_MAX_THREADS*3*tot*sizeof(float), "AO tables");
1456 }
1457
1458 /* give per thread a table, we have to compare xs ys because of way OSA works... */
1459 static float *threadsafe_table_sphere(int test, int thread, int xs, int ys, int tot)
1460 {
1461         static int xso[BLENDER_MAX_THREADS], yso[BLENDER_MAX_THREADS];
1462         static int firsttime= 1;
1463         
1464         if(firsttime) {
1465                 memset(xso, 255, sizeof(xso));
1466                 memset(yso, 255, sizeof(yso));
1467                 firsttime= 0;
1468         }
1469         
1470         if(xs==xso[thread] && ys==yso[thread]) return R.wrld.aotables+ thread*tot*3;
1471         if(test) return NULL;
1472         xso[thread]= xs; yso[thread]= ys;
1473         return R.wrld.aotables+ thread*tot*3;
1474 }
1475
1476 static float *sphere_sampler(int type, int resol, int thread, int xs, int ys)
1477 {
1478         int tot;
1479         float *vec;
1480         
1481         tot= 2*resol*resol;
1482
1483         if (type & WO_AORNDSMP) {
1484                 float *sphere;
1485                 int a;
1486                 
1487                 // always returns table
1488                 sphere= threadsafe_table_sphere(0, thread, xs, ys, tot);
1489
1490                 /* total random sampling. NOT THREADSAFE! (should be removed, is not useful) */
1491                 vec= sphere;
1492                 for (a=0; a<tot; a++, vec+=3) {
1493                         RandomSpherical(vec);
1494                 }
1495                 
1496                 return sphere;
1497         } 
1498         else {
1499                 float *sphere;
1500                 float cosfi, sinfi, cost, sint;
1501                 float ang, *vec1;
1502                 int a;
1503                 
1504                 // returns table if xs and ys were equal to last call
1505                 sphere= threadsafe_table_sphere(1, thread, xs, ys, tot);
1506                 if(sphere==NULL) {
1507                         sphere= threadsafe_table_sphere(0, thread, xs, ys, tot);
1508                         
1509                         // random rotation
1510                         ang= BLI_thread_frand(thread);
1511                         sinfi= sin(ang); cosfi= cos(ang);
1512                         ang= BLI_thread_frand(thread);
1513                         sint= sin(ang); cost= cos(ang);
1514                         
1515                         vec= R.wrld.aosphere;
1516                         vec1= sphere;
1517                         for (a=0; a<tot; a++, vec+=3, vec1+=3) {
1518                                 vec1[0]= cost*cosfi*vec[0] - sinfi*vec[1] + sint*cosfi*vec[2];
1519                                 vec1[1]= cost*sinfi*vec[0] + cosfi*vec[1] + sint*sinfi*vec[2];
1520                                 vec1[2]= -sint*vec[0] + cost*vec[2];                    
1521                         }
1522                 }
1523                 return sphere;
1524         }
1525 }
1526
1527 void ray_ao_qmc(ShadeInput *shi, float *shadfac)
1528 {
1529         Isect isec;
1530         QMCSampler *qsa=NULL;
1531         float samp3d[3];
1532         float up[3], side[3], dir[3], nrm[3];
1533         
1534         float maxdist = R.wrld.aodist;
1535         float fac=0.0f, prev=0.0f;
1536         float adapt_thresh = G.scene->world->ao_adapt_thresh;
1537         float adapt_speed_fac = G.scene->world->ao_adapt_speed_fac;
1538         float bias = G.scene->world->aobias;
1539         
1540         int samples=0;
1541         int max_samples = R.wrld.aosamp*R.wrld.aosamp;
1542         
1543         float dxyview[3], skyadded=0, div;
1544         int aocolor;
1545         
1546         isec.faceorig= (RayFace*)shi->vlr;
1547         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
1548         isec.face_last= NULL;
1549         isec.ob_last= 0;
1550         isec.mode= (R.wrld.aomode & WO_AODIST)?RE_RAY_SHADOW_TRA:RE_RAY_SHADOW;
1551         isec.lay= -1;
1552         
1553         shadfac[0]= shadfac[1]= shadfac[2]= 0.0f;
1554         
1555         /* prevent sky colors to be added for only shadow (shadow becomes alpha) */
1556         aocolor= R.wrld.aocolor;
1557         if(shi->mat->mode & MA_ONLYSHADOW)
1558                 aocolor= WO_AOPLAIN;
1559         
1560         if(aocolor == WO_AOSKYTEX) {
1561                 dxyview[0]= 1.0f/(float)R.wrld.aosamp;
1562                 dxyview[1]= 1.0f/(float)R.wrld.aosamp;
1563                 dxyview[2]= 0.0f;
1564         }
1565         
1566         /* bias prevents smoothed faces to appear flat */
1567         if(shi->vlr->flag & R_SMOOTH) {
1568                 bias= G.scene->world->aobias;
1569                 VECCOPY(nrm, shi->vn);
1570         }
1571         else {
1572                 bias= 0.0f;
1573                 VECCOPY(nrm, shi->facenor);
1574         }
1575         
1576         VecOrthoBasisf(nrm, up, side);
1577         
1578         /* sampling init */
1579         if (R.wrld.ao_samp_method==WO_AOSAMP_HALTON) {
1580                 float speedfac;
1581                 
1582                 speedfac = get_avg_speed(shi) * adapt_speed_fac;
1583                 CLAMP(speedfac, 1.0, 1000.0);
1584                 max_samples /= speedfac;
1585                 if (max_samples < 5) max_samples = 5;
1586                 
1587                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HALTON, max_samples);
1588         } else if (R.wrld.ao_samp_method==WO_AOSAMP_HAMMERSLEY)
1589                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HAMMERSLEY, max_samples);
1590
1591         QMC_initPixel(qsa, shi->thread);
1592         
1593         while (samples < max_samples) {
1594
1595                 /* sampling, returns quasi-random vector in unit hemisphere */
1596                 QMC_sampleHemi(samp3d, qsa, shi->thread, samples);
1597
1598                 dir[0] = (samp3d[0]*up[0] + samp3d[1]*side[0] + samp3d[2]*nrm[0]);
1599                 dir[1] = (samp3d[0]*up[1] + samp3d[1]*side[1] + samp3d[2]*nrm[1]);
1600                 dir[2] = (samp3d[0]*up[2] + samp3d[1]*side[2] + samp3d[2]*nrm[2]);
1601                 
1602                 Normalize(dir);
1603                         
1604                 VECCOPY(isec.start, shi->co);
1605                 isec.end[0] = shi->co[0] - maxdist*dir[0];
1606                 isec.end[1] = shi->co[1] - maxdist*dir[1];
1607                 isec.end[2] = shi->co[2] - maxdist*dir[2];
1608                 
1609                 prev = fac;
1610                 
1611                 if(RE_ray_tree_intersect(R.raytree, &isec)) {
1612                         if (R.wrld.aomode & WO_AODIST) fac+= exp(-isec.labda*R.wrld.aodistfac); 
1613                         else fac+= 1.0f;
1614                 }
1615                 else if(aocolor!=WO_AOPLAIN) {
1616                         float skycol[4];
1617                         float skyfac, view[3];
1618                         
1619                         view[0]= -dir[0];
1620                         view[1]= -dir[1];
1621                         view[2]= -dir[2];
1622                         Normalize(view);
1623                         
1624                         if(aocolor==WO_AOSKYCOL) {
1625                                 skyfac= 0.5*(1.0f+view[0]*R.grvec[0]+ view[1]*R.grvec[1]+ view[2]*R.grvec[2]);
1626                                 shadfac[0]+= (1.0f-skyfac)*R.wrld.horr + skyfac*R.wrld.zenr;
1627                                 shadfac[1]+= (1.0f-skyfac)*R.wrld.horg + skyfac*R.wrld.zeng;
1628                                 shadfac[2]+= (1.0f-skyfac)*R.wrld.horb + skyfac*R.wrld.zenb;
1629                         }
1630                         else {  /* WO_AOSKYTEX */
1631                                 shadeSkyView(skycol, isec.start, view, dxyview);
1632                                 shadfac[0]+= skycol[0];
1633                                 shadfac[1]+= skycol[1];
1634                                 shadfac[2]+= skycol[2];
1635                         }
1636                         skyadded++;
1637                 }
1638                 
1639                 samples++;
1640                 
1641                 if (qsa->type == SAMP_TYPE_HALTON) {
1642                         /* adaptive sampling - consider samples below threshold as in shadow (or vice versa) and exit early */          
1643                         if (adapt_thresh > 0.0 && (samples > max_samples/2) ) {
1644                                 
1645                                 if (adaptive_sample_contrast_val(samples, prev, fac, adapt_thresh)) {
1646                                         break;
1647                                 }
1648                         }
1649                 }
1650         }
1651         
1652         if(aocolor!=WO_AOPLAIN && skyadded) {
1653                 div= (1.0f - fac/(float)samples)/((float)skyadded);
1654                 
1655                 shadfac[0]*= div;       // average color times distances/hits formula
1656                 shadfac[1]*= div;       // average color times distances/hits formula
1657                 shadfac[2]*= div;       // average color times distances/hits formula
1658         } else {
1659                 shadfac[0]= shadfac[1]= shadfac[2]= 1.0f - fac/(float)samples;
1660         }
1661         
1662         if (qsa)
1663                 release_thread_qmcsampler(&R, shi->thread, qsa);
1664 }
1665
1666 /* extern call from shade_lamp_loop, ambient occlusion calculus */
1667 void ray_ao_spheresamp(ShadeInput *shi, float *shadfac)
1668 {
1669         Isect isec;
1670         float *vec, *nrm, div, bias, sh=0.0f;
1671         float maxdist = R.wrld.aodist;
1672         float dxyview[3];
1673         int j= -1, tot, actual=0, skyadded=0, aocolor, resol= R.wrld.aosamp;
1674         
1675         isec.faceorig= (RayFace*)shi->vlr;
1676         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
1677         isec.face_last= NULL;
1678         isec.ob_last= 0;
1679         isec.mode= (R.wrld.aomode & WO_AODIST)?RE_RAY_SHADOW_TRA:RE_RAY_SHADOW;
1680         isec.lay= -1;
1681
1682
1683         shadfac[0]= shadfac[1]= shadfac[2]= 0.0f;
1684
1685         /* bias prevents smoothed faces to appear flat */
1686         if(shi->vlr->flag & R_SMOOTH) {
1687                 bias= G.scene->world->aobias;
1688                 nrm= shi->vn;
1689         }
1690         else {
1691                 bias= 0.0f;
1692                 nrm= shi->facenor;
1693         }
1694
1695         /* prevent sky colors to be added for only shadow (shadow becomes alpha) */
1696         aocolor= R.wrld.aocolor;
1697         if(shi->mat->mode & MA_ONLYSHADOW)
1698                 aocolor= WO_AOPLAIN;
1699         
1700         if(resol>32) resol= 32;
1701         
1702         vec= sphere_sampler(R.wrld.aomode, resol, shi->thread, shi->xs, shi->ys);
1703         
1704         // warning: since we use full sphere now, and dotproduct is below, we do twice as much
1705         tot= 2*resol*resol;
1706
1707         if(aocolor == WO_AOSKYTEX) {
1708                 dxyview[0]= 1.0f/(float)resol;
1709                 dxyview[1]= 1.0f/(float)resol;
1710                 dxyview[2]= 0.0f;
1711         }
1712         
1713         while(tot--) {
1714                 
1715                 if ((vec[0]*nrm[0] + vec[1]*nrm[1] + vec[2]*nrm[2]) > bias) {
1716                         /* only ao samples for mask */
1717                         if(R.r.mode & R_OSA) {
1718                                 j++;
1719                                 if(j==R.osa) j= 0;
1720                                 if(!(shi->mask & (1<<j))) {
1721                                         vec+=3;
1722                                         continue;
1723                                 }
1724                         }
1725                         
1726                         actual++;
1727                         
1728                         /* always set start/end, RE_ray_tree_intersect clips it */
1729                         VECCOPY(isec.start, shi->co);
1730                         isec.end[0] = shi->co[0] - maxdist*vec[0];
1731                         isec.end[1] = shi->co[1] - maxdist*vec[1];
1732                         isec.end[2] = shi->co[2] - maxdist*vec[2];
1733                         
1734                         /* do the trace */
1735                         if(RE_ray_tree_intersect(R.raytree, &isec)) {
1736                                 if (R.wrld.aomode & WO_AODIST) sh+= exp(-isec.labda*R.wrld.aodistfac); 
1737                                 else sh+= 1.0f;
1738                         }
1739                         else if(aocolor!=WO_AOPLAIN) {
1740                                 float skycol[4];
1741                                 float fac, view[3];
1742                                 
1743                                 view[0]= -vec[0];
1744                                 view[1]= -vec[1];
1745                                 view[2]= -vec[2];
1746                                 Normalize(view);
1747                                 
1748                                 if(aocolor==WO_AOSKYCOL) {
1749                                         fac= 0.5*(1.0f+view[0]*R.grvec[0]+ view[1]*R.grvec[1]+ view[2]*R.grvec[2]);
1750                                         shadfac[0]+= (1.0f-fac)*R.wrld.horr + fac*R.wrld.zenr;
1751                                         shadfac[1]+= (1.0f-fac)*R.wrld.horg + fac*R.wrld.zeng;
1752                                         shadfac[2]+= (1.0f-fac)*R.wrld.horb + fac*R.wrld.zenb;
1753                                 }
1754                                 else {  /* WO_AOSKYTEX */
1755                                         shadeSkyView(skycol, isec.start, view, dxyview);
1756                                         shadfac[0]+= skycol[0];
1757                                         shadfac[1]+= skycol[1];
1758                                         shadfac[2]+= skycol[2];
1759                                 }
1760                                 skyadded++;
1761                         }
1762                 }
1763                 // samples
1764                 vec+= 3;
1765         }
1766         
1767         if(actual==0) sh= 1.0f;
1768         else sh = 1.0f - sh/((float)actual);
1769         
1770         if(aocolor!=WO_AOPLAIN && skyadded) {
1771                 div= sh/((float)skyadded);
1772                 
1773                 shadfac[0]*= div;       // average color times distances/hits formula
1774                 shadfac[1]*= div;       // average color times distances/hits formula
1775                 shadfac[2]*= div;       // average color times distances/hits formula
1776         }
1777         else {
1778                 shadfac[0]= shadfac[1]= shadfac[2]= sh;
1779         }
1780 }
1781
1782 void ray_ao(ShadeInput *shi, float *shadfac)
1783 {
1784         /* Unfortunately, the unusual way that the sphere sampler calculates roughly twice as many
1785          * samples as are actually traced, and skips them based on bias and OSA settings makes it very difficult
1786          * to reuse code between these two functions. This is the easiest way I can think of to do it
1787          * --broken */
1788         if (ELEM(R.wrld.ao_samp_method, WO_AOSAMP_HAMMERSLEY, WO_AOSAMP_HALTON))
1789                 ray_ao_qmc(shi, shadfac);
1790         else if (R.wrld.ao_samp_method == WO_AOSAMP_CONSTANT)
1791                 ray_ao_spheresamp(shi, shadfac);
1792 }
1793
1794
1795 static void ray_shadow_qmc(ShadeInput *shi, LampRen *lar, float *lampco, float *shadfac, Isect *isec)
1796 {
1797         QMCSampler *qsa=NULL;
1798         QMCSampler *qsa_jit=NULL;
1799         int samples=0;
1800         float samp3d[3], jit[3], jitbias= 0.0f;
1801
1802         float fac=0.0f, vec[3];
1803         float colsq[4];
1804         float adapt_thresh = lar->adapt_thresh;
1805         int max_samples = lar->ray_totsamp;
1806         float pos[3];
1807         int do_soft=1, full_osa=0;
1808
1809         colsq[0] = colsq[1] = colsq[2] = 0.0;
1810         if(isec->mode==RE_RAY_SHADOW_TRA) {
1811                 shadfac[0]= shadfac[1]= shadfac[2]= shadfac[3]= 0.0f;
1812         } else
1813                 shadfac[3]= 1.0f;
1814         
1815         if (lar->ray_totsamp < 2) do_soft = 0;
1816         if ((R.r.mode & R_OSA) && (R.osa > 0) && (shi->vlr->flag & R_FULL_OSA)) full_osa = 1;
1817         
1818         if (full_osa) {
1819                 if (do_soft) max_samples  = max_samples/R.osa + 1;
1820                 else max_samples = 1;
1821         } else {
1822                 if (do_soft) max_samples = lar->ray_totsamp;
1823                 else max_samples = (R.osa > 4)?R.osa:5;
1824         }
1825
1826         if(shi->vlr && ((shi->vlr->flag & R_FULL_OSA) == 0))
1827                 jitbias= 0.5f*(VecLength(shi->dxco) + VecLength(shi->dyco));
1828
1829         /* sampling init */
1830         if (lar->ray_samp_method==LA_SAMP_HALTON) {
1831                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HALTON, max_samples);
1832                 qsa_jit = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HALTON, max_samples);
1833         } else if (lar->ray_samp_method==LA_SAMP_HAMMERSLEY) {
1834                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HAMMERSLEY, max_samples);
1835                 qsa_jit = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HAMMERSLEY, max_samples);
1836         }
1837         
1838         QMC_initPixel(qsa, shi->thread);
1839         QMC_initPixel(qsa_jit, shi->thread);
1840         
1841         VECCOPY(vec, lampco);
1842         
1843         
1844         while (samples < max_samples) {
1845                 isec->faceorig= (RayFace*)shi->vlr;
1846                 isec->oborig= RAY_OBJECT_SET(&R, shi->obi);
1847                 
1848                 /* manually jitter the start shading co-ord per sample
1849                  * based on the pre-generated OSA texture sampling offsets, 
1850                  * for anti-aliasing sharp shadow edges. */
1851                 VECCOPY(pos, shi->co);
1852                 if (shi->vlr && !full_osa) {
1853                         QMC_sampleRect(jit, qsa_jit, shi->thread, samples, 1.0, 1.0);
1854                         
1855                         pos[0] += shi->dxco[0]*jit[0] + shi->dyco[0]*jit[1];
1856                         pos[1] += shi->dxco[1]*jit[0] + shi->dyco[1]*jit[1];
1857                         pos[2] += shi->dxco[2]*jit[0] + shi->dyco[2]*jit[1];
1858                 }
1859
1860                 if (do_soft) {
1861                         /* sphere shadow source */
1862                         if (lar->type == LA_LOCAL) {
1863                                 float ru[3], rv[3], v[3], s[3];
1864                                 
1865                                 /* calc tangent plane vectors */
1866                                 v[0] = pos[0] - lampco[0];
1867                                 v[1] = pos[1] - lampco[1];
1868                                 v[2] = pos[2] - lampco[2];
1869                                 Normalize(v);
1870                                 VecOrthoBasisf(v, ru, rv);
1871                                 
1872                                 /* sampling, returns quasi-random vector in area_size disc */
1873                                 QMC_sampleDisc(samp3d, qsa, shi->thread, samples,lar->area_size);
1874
1875                                 /* distribute disc samples across the tangent plane */
1876                                 s[0] = samp3d[0]*ru[0] + samp3d[1]*rv[0];
1877                                 s[1] = samp3d[0]*ru[1] + samp3d[1]*rv[1];
1878                                 s[2] = samp3d[0]*ru[2] + samp3d[1]*rv[2];
1879                                 
1880                                 VECCOPY(samp3d, s);
1881
1882                                 if(jitbias != 0.0f) {
1883                                         /* bias away somewhat to avoid self intersection */
1884                                         pos[0] -= jitbias*v[0];
1885                                         pos[1] -= jitbias*v[1];
1886                                         pos[2] -= jitbias*v[2];
1887                                 }
1888                         }
1889                         else {
1890                                 /* sampling, returns quasi-random vector in [sizex,sizey]^2 plane */
1891                                 QMC_sampleRect(samp3d, qsa, shi->thread, samples, lar->area_size, lar->area_sizey);
1892                                                                 
1893                                 /* align samples to lamp vector */
1894                                 Mat3MulVecfl(lar->mat, samp3d);
1895                         }
1896                         isec->end[0]= vec[0]+samp3d[0];
1897                         isec->end[1]= vec[1]+samp3d[1];
1898                         isec->end[2]= vec[2]+samp3d[2];
1899                 } else {
1900                         VECCOPY(isec->end, vec);
1901                 }
1902
1903                 if(jitbias != 0.0f && !(do_soft && lar->type==LA_LOCAL)) {
1904                         /* bias away somewhat to avoid self intersection */
1905                         float v[3];
1906
1907                         VECSUB(v, pos, isec->end);
1908                         Normalize(v);
1909
1910                         pos[0] -= jitbias*v[0];
1911                         pos[1] -= jitbias*v[1];
1912                         pos[2] -= jitbias*v[2];
1913                 }
1914
1915                 VECCOPY(isec->start, pos);
1916                 
1917                 
1918                 /* trace the ray */
1919                 if(isec->mode==RE_RAY_SHADOW_TRA) {
1920                         isec->col[0]= isec->col[1]= isec->col[2]=  1.0f;
1921                         isec->col[3]= 1.0f;
1922                         
1923                         ray_trace_shadow_tra(isec, DEPTH_SHADOW_TRA, 0);
1924                         shadfac[0] += isec->col[0];
1925                         shadfac[1] += isec->col[1];
1926                         shadfac[2] += isec->col[2];
1927                         shadfac[3] += isec->col[3];
1928                         
1929                         /* for variance calc */
1930                         colsq[0] += isec->col[0]*isec->col[0];
1931                         colsq[1] += isec->col[1]*isec->col[1];
1932                         colsq[2] += isec->col[2]*isec->col[2];
1933                 }
1934                 else {
1935                         if( RE_ray_tree_intersect(R.raytree, isec) ) fac+= 1.0f;
1936                 }
1937                 
1938                 samples++;
1939                 
1940                 if ((lar->ray_samp_method == LA_SAMP_HALTON)) {
1941                 
1942                         /* adaptive sampling - consider samples below threshold as in shadow (or vice versa) and exit early */
1943                         if ((max_samples > 4) && (adapt_thresh > 0.0) && (samples > max_samples / 3)) {
1944                                 if (isec->mode==RE_RAY_SHADOW_TRA) {
1945                                         if ((shadfac[3] / samples > (1.0-adapt_thresh)) || (shadfac[3] / samples < adapt_thresh))
1946                                                 break;
1947                                         else if (adaptive_sample_variance(samples, shadfac, colsq, adapt_thresh))
1948                                                 break;
1949                                 } else {
1950                                         if ((fac / samples > (1.0-adapt_thresh)) || (fac / samples < adapt_thresh))
1951                                                 break;
1952                                 }
1953                         }
1954                 }
1955         }
1956         
1957         if(isec->mode==RE_RAY_SHADOW_TRA) {
1958                 shadfac[0] /= samples;
1959                 shadfac[1] /= samples;
1960                 shadfac[2] /= samples;
1961                 shadfac[3] /= samples;
1962         } else
1963                 shadfac[3]= 1.0f-fac/samples;
1964
1965         if (qsa_jit)
1966                 release_thread_qmcsampler(&R, shi->thread, qsa_jit);
1967         if (qsa)
1968                 release_thread_qmcsampler(&R, shi->thread, qsa);
1969 }
1970
1971 static void ray_shadow_jitter(ShadeInput *shi, LampRen *lar, float *lampco, float *shadfac, Isect *isec)
1972 {
1973         /* area soft shadow */
1974         float *jitlamp;
1975         float fac=0.0f, div=0.0f, vec[3];
1976         int a, j= -1, mask;
1977         
1978         if(isec->mode==RE_RAY_SHADOW_TRA) {
1979                 shadfac[0]= shadfac[1]= shadfac[2]= shadfac[3]= 0.0f;
1980         }
1981         else shadfac[3]= 1.0f;
1982         
1983         fac= 0.0f;
1984         jitlamp= give_jitter_plane(lar, shi->thread, shi->xs, shi->ys);
1985
1986         a= lar->ray_totsamp;
1987         
1988         /* this correction to make sure we always take at least 1 sample */
1989         mask= shi->mask;
1990         if(a==4) mask |= (mask>>4)|(mask>>8);
1991         else if(a==9) mask |= (mask>>9);
1992         
1993         while(a--) {
1994                 
1995                 if(R.r.mode & R_OSA) {
1996                         j++;
1997                         if(j>=R.osa) j= 0;
1998                         if(!(mask & (1<<j))) {
1999                                 jitlamp+= 2;
2000                                 continue;
2001                         }
2002                 }
2003                 
2004                 isec->faceorig= (RayFace*)shi->vlr;
2005                 isec->oborig= RAY_OBJECT_SET(&R, shi->obi);
2006                 
2007                 vec[0]= jitlamp[0];
2008                 vec[1]= jitlamp[1];
2009                 vec[2]= 0.0f;
2010                 Mat3MulVecfl(lar->mat, vec);
2011                 
2012                 /* set start and end, RE_ray_tree_intersect clips it */
2013                 VECCOPY(isec->start, shi->co);
2014                 isec->end[0]= lampco[0]+vec[0];
2015                 isec->end[1]= lampco[1]+vec[1];
2016                 isec->end[2]= lampco[2]+vec[2];
2017                 
2018                 if(isec->mode==RE_RAY_SHADOW_TRA) {
2019                         /* isec.col is like shadfac, so defines amount of light (0.0 is full shadow) */
2020                         isec->col[0]= isec->col[1]= isec->col[2]=  1.0f;
2021                         isec->col[3]= 1.0f;
2022                         
2023                         ray_trace_shadow_tra(isec, DEPTH_SHADOW_TRA, 0);
2024                         shadfac[0] += isec->col[0];
2025                         shadfac[1] += isec->col[1];
2026                         shadfac[2] += isec->col[2];
2027                         shadfac[3] += isec->col[3];
2028                 }
2029                 else if( RE_ray_tree_intersect(R.raytree, isec) ) fac+= 1.0f;
2030                 
2031                 div+= 1.0f;
2032                 jitlamp+= 2;
2033         }
2034         
2035         if(isec->mode==RE_RAY_SHADOW_TRA) {
2036                 shadfac[0] /= div;
2037                 shadfac[1] /= div;
2038                 shadfac[2] /= div;
2039                 shadfac[3] /= div;
2040         }
2041         else {
2042                 // sqrt makes nice umbra effect
2043                 if(lar->ray_samp_type & LA_SAMP_UMBRA)
2044                         shadfac[3]= sqrt(1.0f-fac/div);
2045                 else
2046                         shadfac[3]= 1.0f-fac/div;
2047         }
2048 }
2049 /* extern call from shade_lamp_loop */
2050 void ray_shadow(ShadeInput *shi, LampRen *lar, float *shadfac)
2051 {
2052         Isect isec;
2053         float lampco[3], maxsize;
2054
2055         /* setup isec */
2056         if(shi->mat->mode & MA_SHADOW_TRA) isec.mode= RE_RAY_SHADOW_TRA;
2057         else isec.mode= RE_RAY_SHADOW;
2058         
2059         if(lar->mode & (LA_LAYER|LA_LAYER_SHADOW))
2060                 isec.lay= lar->lay;
2061         else
2062                 isec.lay= -1;
2063
2064         /* only when not mir tracing, first hit optimm */
2065         if(shi->depth==0) {
2066                 isec.face_last= (RayFace*)lar->vlr_last[shi->thread];
2067                 isec.ob_last= RAY_OBJECT_SET(&R, lar->obi_last[shi->thread]);
2068         }
2069         else {
2070                 isec.face_last= NULL;
2071                 isec.ob_last= 0;
2072         }
2073         
2074         if(lar->type==LA_SUN || lar->type==LA_HEMI) {
2075                 maxsize= RE_ray_tree_max_size(R.raytree);
2076                 lampco[0]= shi->co[0] - maxsize*lar->vec[0];
2077                 lampco[1]= shi->co[1] - maxsize*lar->vec[1];
2078                 lampco[2]= shi->co[2] - maxsize*lar->vec[2];
2079         }
2080         else {
2081                 VECCOPY(lampco, lar->co);
2082         }
2083         
2084         if (ELEM(lar->ray_samp_method, LA_SAMP_HALTON, LA_SAMP_HAMMERSLEY)) {
2085                 
2086                 ray_shadow_qmc(shi, lar, lampco, shadfac, &isec);
2087                 
2088         } else {
2089                 if(lar->ray_totsamp<2) {
2090                         
2091                         isec.faceorig= (RayFace*)shi->vlr;
2092                         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
2093                         shadfac[3]= 1.0f; // 1.0=full light
2094                         
2095                         /* set up isec vec */
2096                         VECCOPY(isec.start, shi->co);
2097                         VECCOPY(isec.end, lampco);
2098
2099                         if(isec.mode==RE_RAY_SHADOW_TRA) {
2100                                 /* isec.col is like shadfac, so defines amount of light (0.0 is full shadow) */
2101                                 isec.col[0]= isec.col[1]= isec.col[2]=  1.0f;
2102                                 isec.col[3]= 1.0f;
2103
2104                                 ray_trace_shadow_tra(&isec, DEPTH_SHADOW_TRA, 0);
2105                                 QUATCOPY(shadfac, isec.col);
2106                         }
2107                         else if(RE_ray_tree_intersect(R.raytree, &isec)) shadfac[3]= 0.0f;
2108                 }
2109                 else {
2110                         ray_shadow_jitter(shi, lar, lampco, shadfac, &isec);
2111                 }
2112         }
2113                 
2114         /* for first hit optim, set last interesected shadow face */
2115         if(shi->depth==0) {
2116                 lar->vlr_last[shi->thread]= (VlakRen*)isec.face_last;
2117                 lar->obi_last[shi->thread]= RAY_OBJECT_GET(&R, isec.ob_last);
2118         }
2119
2120 }
2121
2122 /* only when face points away from lamp, in direction of lamp, trace ray and find first exit point */
2123 void ray_translucent(ShadeInput *shi, LampRen *lar, float *distfac, float *co)
2124 {
2125         Isect isec;
2126         float lampco[3], maxsize;
2127         
2128         /* setup isec */
2129         isec.mode= RE_RAY_SHADOW_TRA;
2130         
2131         if(lar->mode & LA_LAYER) isec.lay= lar->lay; else isec.lay= -1;
2132         
2133         if(lar->type==LA_SUN || lar->type==LA_HEMI) {
2134                 maxsize= RE_ray_tree_max_size(R.raytree);
2135                 lampco[0]= shi->co[0] - maxsize*lar->vec[0];
2136                 lampco[1]= shi->co[1] - maxsize*lar->vec[1];
2137                 lampco[2]= shi->co[2] - maxsize*lar->vec[2];
2138         }
2139         else {
2140                 VECCOPY(lampco, lar->co);
2141         }
2142         
2143         isec.faceorig= (RayFace*)shi->vlr;
2144         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
2145         
2146         /* set up isec vec */
2147         VECCOPY(isec.start, shi->co);
2148         VECCOPY(isec.end, lampco);
2149         
2150         if(RE_ray_tree_intersect(R.raytree, &isec)) {
2151                 /* we got a face */
2152                 
2153                 /* render co */
2154                 co[0]= isec.start[0]+isec.labda*(isec.vec[0]);
2155                 co[1]= isec.start[1]+isec.labda*(isec.vec[1]);
2156                 co[2]= isec.start[2]+isec.labda*(isec.vec[2]);
2157                 
2158                 *distfac= VecLength(isec.vec);
2159         }
2160         else
2161                 *distfac= 0.0f;
2162 }
2163
2164