resolve some compiler warnings with intel c/c++ compiler
[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_callocN(x*tot*2*sizeof(float), "lamp jitter tab");
637         
638         /* if 1 sample, we leave table to be zero's */
639         if(tot>1) {
640                 
641                 /* set per-lamp fixed seed */
642                 BLI_srandom(tot);
643                 
644                 /* fill table with random locations, area_size large */
645                 for(x=0; x<tot; x++, fp+=2) {
646                         fp[0]= (BLI_frand()-0.5)*lar->area_size;
647                         fp[1]= (BLI_frand()-0.5)*lar->area_sizey;
648                 }
649                 
650                 while(iter--) {
651                         fp= lar->jitter;
652                         for(x=tot; x>0; x--, fp+=2) {
653                                 DP_energy(lar->jitter, fp, tot, lar->area_size, lar->area_sizey);
654                         }
655                 }
656         }       
657         /* create the dithered tables (could just check lamp type!) */
658         jitter_plane_offset(lar->jitter, lar->jitter+2*tot, tot, lar->area_size, lar->area_sizey, 0.5f, 0.0f);
659         jitter_plane_offset(lar->jitter, lar->jitter+4*tot, tot, lar->area_size, lar->area_sizey, 0.5f, 0.5f);
660         jitter_plane_offset(lar->jitter, lar->jitter+6*tot, tot, lar->area_size, lar->area_sizey, 0.0f, 0.5f);
661 }
662
663 /* table around origin, -0.5*size to 0.5*size */
664 static float *give_jitter_plane(LampRen *lar, int thread, int xs, int ys)
665 {
666         int tot;
667         
668         tot= lar->ray_totsamp;
669                         
670         if(lar->ray_samp_type & LA_SAMP_JITTER) {
671                 /* made it threadsafe */
672                 
673                 if(lar->xold[thread]!=xs || lar->yold[thread]!=ys) {
674                         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));
675                         lar->xold[thread]= xs; 
676                         lar->yold[thread]= ys;
677                 }
678                 return lar->jitter+2*(thread+1)*tot;
679         }
680         if(lar->ray_samp_type & LA_SAMP_DITHER) {
681                 return lar->jitter + 2*tot*((xs & 1)+2*(ys & 1));
682         }
683         
684         return lar->jitter;
685 }
686
687
688 /* **************** QMC sampling *************** */
689
690 static void halton_sample(double *ht_invprimes, double *ht_nums, double *v)
691 {
692         // incremental halton sequence generator, from:
693         // "Instant Radiosity", Keller A.
694         unsigned int i;
695         
696         for (i = 0; i < 2; i++)
697         {
698                 double r = fabs((1.0 - ht_nums[i]) - 1e-10);
699                 
700                 if (ht_invprimes[i] >= r)
701                 {
702                         double lasth;
703                         double h = ht_invprimes[i];
704                         
705                         do {
706                                 lasth = h;
707                                 h *= ht_invprimes[i];
708                         } while (h >= r);
709                         
710                         ht_nums[i] += ((lasth + h) - 1.0);
711                 }
712                 else
713                         ht_nums[i] += ht_invprimes[i];
714                 
715                 v[i] = (float)ht_nums[i];
716         }
717 }
718
719 /* Generate Hammersley points in [0,1)^2
720  * From Lucille renderer */
721 static void hammersley_create(double *out, int n)
722 {
723         double p, t;
724         int k, kk;
725
726         for (k = 0; k < n; k++) {
727                 t = 0;
728                 for (p = 0.5, kk = k; kk; p *= 0.5, kk >>= 1) {
729                         if (kk & 1) {           /* kk mod 2 = 1         */
730                                 t += p;
731                         }
732                 }
733         
734                 out[2 * k + 0] = (double)k / (double)n;
735                 out[2 * k + 1] = t;
736         }
737 }
738
739 static struct QMCSampler *QMC_initSampler(int type, int tot)
740 {       
741         QMCSampler *qsa = MEM_callocN(sizeof(QMCSampler), "qmc sampler");
742         qsa->samp2d = MEM_callocN(2*sizeof(double)*tot, "qmc sample table");
743
744         qsa->tot = tot;
745         qsa->type = type;
746         
747         if (qsa->type==SAMP_TYPE_HAMMERSLEY) 
748                 hammersley_create(qsa->samp2d, qsa->tot);
749                 
750         return qsa;
751 }
752
753 static void QMC_initPixel(QMCSampler *qsa, int thread)
754 {
755         if (qsa->type==SAMP_TYPE_HAMMERSLEY)
756         {
757                 /* hammersley sequence is fixed, already created in QMCSampler init.
758                  * per pixel, gets a random offset. We create separate offsets per thread, for write-safety */
759                 qsa->offs[thread][0] = 0.5 * BLI_thread_frand(thread);
760                 qsa->offs[thread][1] = 0.5 * BLI_thread_frand(thread);
761         }
762         else {  /* SAMP_TYPE_HALTON */
763                 
764                 /* generate a new randomised halton sequence per pixel
765                  * to alleviate qmc artifacts and make it reproducable 
766                  * between threads/frames */
767                 double ht_invprimes[2], ht_nums[2];
768                 double r[2];
769                 int i;
770         
771                 ht_nums[0] = BLI_thread_frand(thread);
772                 ht_nums[1] = BLI_thread_frand(thread);
773                 ht_invprimes[0] = 0.5;
774                 ht_invprimes[1] = 1.0/3.0;
775                 
776                 for (i=0; i< qsa->tot; i++) {
777                         halton_sample(ht_invprimes, ht_nums, r);
778                         qsa->samp2d[2*i+0] = r[0];
779                         qsa->samp2d[2*i+1] = r[1];
780                 }
781         }
782 }
783
784 static void QMC_freeSampler(QMCSampler *qsa)
785 {
786         MEM_freeN(qsa->samp2d);
787         MEM_freeN(qsa);
788 }
789
790 static void QMC_getSample(double *s, QMCSampler *qsa, int thread, int num)
791 {
792         if (qsa->type == SAMP_TYPE_HAMMERSLEY) {
793                 s[0] = fmod(qsa->samp2d[2*num+0] + qsa->offs[thread][0], 1.0f);
794                 s[1] = fmod(qsa->samp2d[2*num+1] + qsa->offs[thread][1], 1.0f);
795         }
796         else { /* SAMP_TYPE_HALTON */
797                 s[0] = qsa->samp2d[2*num+0];
798                 s[1] = qsa->samp2d[2*num+1];
799         }
800 }
801
802 /* phong weighted disc using 'blur' for exponent, centred on 0,0 */
803 static void QMC_samplePhong(float *vec, QMCSampler *qsa, int thread, int num, float blur)
804 {
805         double s[2];
806         float phi, pz, sqr;
807         
808         QMC_getSample(s, qsa, thread, num);
809
810         phi = s[0]*2*M_PI;
811         pz = pow(s[1], blur);
812         sqr = sqrt(1.0f-pz*pz);
813
814         vec[0] = cos(phi)*sqr;
815         vec[1] = sin(phi)*sqr;
816         vec[2] = 0.0f;
817 }
818
819 /* rect of edge lengths sizex, sizey, centred on 0.0,0.0 i.e. ranging from -sizex/2 to +sizey/2 */
820 static void QMC_sampleRect(float *vec, QMCSampler *qsa, int thread, int num, float sizex, float sizey)
821 {
822         double s[2];
823
824         QMC_getSample(s, qsa, thread, num);
825                 
826         vec[0] = (s[0] - 0.5) * sizex;
827         vec[1] = (s[1] - 0.5) * sizey;
828         vec[2] = 0.0f;
829 }
830
831 /* disc of radius 'radius', centred on 0,0 */
832 static void QMC_sampleDisc(float *vec, QMCSampler *qsa, int thread, int num, float radius)
833 {
834         double s[2];
835         float phi, sqr;
836         
837         QMC_getSample(s, qsa, thread, num);
838         
839         phi = s[0]*2*M_PI;
840         sqr = sqrt(s[1]);
841
842         vec[0] = cos(phi)*sqr* radius/2.0;
843         vec[1] = sin(phi)*sqr* radius/2.0;
844         vec[2] = 0.0f;
845 }
846
847 /* uniform hemisphere sampling */
848 static void QMC_sampleHemi(float *vec, QMCSampler *qsa, int thread, int num)
849 {
850         double s[2];
851         float phi, sqr;
852         
853         QMC_getSample(s, qsa, thread, num);
854         
855         phi = s[0]*2.f*M_PI;    
856         sqr = sqrt(s[1]);
857
858         vec[0] = cos(phi)*sqr;
859         vec[1] = sin(phi)*sqr;
860         vec[2] = 1.f - s[1]*s[1];
861 }
862
863 #if 0 /* currently not used */
864 /* cosine weighted hemisphere sampling */
865 static void QMC_sampleHemiCosine(float *vec, QMCSampler *qsa, int thread, int num)
866 {
867         double s[2];
868         float phi, sqr;
869         
870         QMC_getSample(s, qsa, thread, num);
871         
872         phi = s[0]*2.f*M_PI;    
873         sqr = s[1]*sqrt(2-s[1]*s[1]);
874
875         vec[0] = cos(phi)*sqr;
876         vec[1] = sin(phi)*sqr;
877         vec[2] = 1.f - s[1]*s[1];
878
879 }
880 #endif
881
882 /* called from convertBlenderScene.c */
883 void init_render_qmcsampler(Render *re)
884 {
885         re->qmcsamplers= MEM_callocN(sizeof(ListBase)*BLENDER_MAX_THREADS, "QMCListBase");
886 }
887
888 static QMCSampler *get_thread_qmcsampler(Render *re, int thread, int type, int tot)
889 {
890         QMCSampler *qsa;
891
892         /* create qmc samplers as needed, since recursion makes it hard to
893          * predict how many are needed */
894
895         for(qsa=re->qmcsamplers[thread].first; qsa; qsa=qsa->next) {
896                 if(qsa->type == type && qsa->tot == tot && !qsa->used) {
897                         qsa->used= 1;
898                         return qsa;
899                 }
900         }
901
902         qsa= QMC_initSampler(type, tot);
903         qsa->used= 1;
904         BLI_addtail(&re->qmcsamplers[thread], qsa);
905
906         return qsa;
907 }
908
909 static void release_thread_qmcsampler(Render *re, int thread, QMCSampler *qsa)
910 {
911         qsa->used= 0;
912 }
913
914 void free_render_qmcsampler(Render *re)
915 {
916         QMCSampler *qsa, *next;
917         int a;
918
919         if(re->qmcsamplers) {
920                 for(a=0; a<BLENDER_MAX_THREADS; a++) {
921                         for(qsa=re->qmcsamplers[a].first; qsa; qsa=next) {
922                                 next= qsa->next;
923                                 QMC_freeSampler(qsa);
924                         }
925
926                         re->qmcsamplers[a].first= re->qmcsamplers[a].last= NULL;
927                 }
928
929                 MEM_freeN(re->qmcsamplers);
930                 re->qmcsamplers= NULL;
931         }
932 }
933
934 static int adaptive_sample_variance(int samples, float *col, float *colsq, float thresh)
935 {
936         float var[3], mean[3];
937
938         /* scale threshold just to give a bit more precision in input rather than dealing with 
939          * tiny tiny numbers in the UI */
940         thresh /= 2;
941         
942         mean[0] = col[0] / (float)samples;
943         mean[1] = col[1] / (float)samples;
944         mean[2] = col[2] / (float)samples;
945
946         var[0] = (colsq[0] / (float)samples) - (mean[0]*mean[0]);
947         var[1] = (colsq[1] / (float)samples) - (mean[1]*mean[1]);
948         var[2] = (colsq[2] / (float)samples) - (mean[2]*mean[2]);
949         
950         if ((var[0] * 0.4 < thresh) && (var[1] * 0.3 < thresh) && (var[2] * 0.6 < thresh))
951                 return 1;
952         else
953                 return 0;
954 }
955
956 static int adaptive_sample_contrast_val(int samples, float prev, float val, float thresh)
957 {
958         /* if the last sample's contribution to the total value was below a small threshold
959          * (i.e. the samples taken are very similar), then taking more samples that are probably 
960          * going to be the same is wasting effort */
961         if (fabs( prev/(float)(samples-1) - val/(float)samples ) < thresh) {
962                 return 1;
963         } else
964                 return 0;
965 }
966
967 static float get_avg_speed(ShadeInput *shi)
968 {
969         float pre_x, pre_y, post_x, post_y, speedavg;
970         
971         pre_x = (shi->winspeed[0] == PASS_VECTOR_MAX)?0.0:shi->winspeed[0];
972         pre_y = (shi->winspeed[1] == PASS_VECTOR_MAX)?0.0:shi->winspeed[1];
973         post_x = (shi->winspeed[2] == PASS_VECTOR_MAX)?0.0:shi->winspeed[2];
974         post_y = (shi->winspeed[3] == PASS_VECTOR_MAX)?0.0:shi->winspeed[3];
975         
976         speedavg = (sqrt(pre_x*pre_x + pre_y*pre_y) + sqrt(post_x*post_x + post_y*post_y)) / 2.0;
977         
978         return speedavg;
979 }
980
981 /* ***************** main calls ************** */
982
983
984 static void trace_refract(float *col, ShadeInput *shi, ShadeResult *shr)
985 {
986         QMCSampler *qsa=NULL;
987         int samp_type;
988         
989         float samp3d[3], orthx[3], orthy[3];
990         float v_refract[3], v_refract_new[3];
991         float sampcol[4], colsq[4];
992         
993         float blur = pow(1.0 - shi->mat->gloss_tra, 3);
994         short max_samples = shi->mat->samp_gloss_tra;
995         float adapt_thresh = shi->mat->adapt_thresh_tra;
996         
997         int samples=0;
998         
999         colsq[0] = colsq[1] = colsq[2] = 0.0;
1000         col[0] = col[1] = col[2] = 0.0;
1001         col[3]= shr->alpha;
1002         
1003         if (blur > 0.0) {
1004                 if (adapt_thresh != 0.0) samp_type = SAMP_TYPE_HALTON;
1005                 else samp_type = SAMP_TYPE_HAMMERSLEY;
1006                         
1007                 /* all samples are generated per pixel */
1008                 qsa = get_thread_qmcsampler(&R, shi->thread, samp_type, max_samples);
1009                 QMC_initPixel(qsa, shi->thread);
1010         } else 
1011                 max_samples = 1;
1012         
1013
1014         while (samples < max_samples) {         
1015                 refraction(v_refract, shi->vn, shi->view, shi->ang);
1016                 
1017                 if (max_samples > 1) {
1018                         /* get a quasi-random vector from a phong-weighted disc */
1019                         QMC_samplePhong(samp3d, qsa, shi->thread, samples, blur);
1020                                                 
1021                         VecOrthoBasisf(v_refract, orthx, orthy);
1022                         VecMulf(orthx, samp3d[0]);
1023                         VecMulf(orthy, samp3d[1]);
1024                                 
1025                         /* and perturb the refraction vector in it */
1026                         VecAddf(v_refract_new, v_refract, orthx);
1027                         VecAddf(v_refract_new, v_refract_new, orthy);
1028                         
1029                         Normalize(v_refract_new);
1030                 } else {
1031                         /* no blurriness, use the original normal */
1032                         VECCOPY(v_refract_new, v_refract);
1033                 }
1034         
1035                 traceray(shi, shr, shi->mat->ray_depth_tra, shi->co, v_refract_new, sampcol, shi->obi, shi->vlr, RAY_TRA|RAY_TRAFLIP);
1036         
1037                 col[0] += sampcol[0];
1038                 col[1] += sampcol[1];
1039                 col[2] += sampcol[2];
1040                 col[3] += sampcol[3];
1041                 
1042                 /* for variance calc */
1043                 colsq[0] += sampcol[0]*sampcol[0];
1044                 colsq[1] += sampcol[1]*sampcol[1];
1045                 colsq[2] += sampcol[2]*sampcol[2];
1046                 
1047                 samples++;
1048                 
1049                 /* adaptive sampling */
1050                 if (adapt_thresh < 1.0 && samples > max_samples/2) 
1051                 {
1052                         if (adaptive_sample_variance(samples, col, colsq, adapt_thresh))
1053                                 break;
1054                         
1055                         /* if the pixel so far is very dark, we can get away with less samples */
1056                         if ( (col[0] + col[1] + col[2])/3.0/(float)samples < 0.01 )
1057                                 max_samples--;
1058                 }
1059         }
1060         
1061         col[0] /= (float)samples;
1062         col[1] /= (float)samples;
1063         col[2] /= (float)samples;
1064         col[3] /= (float)samples;
1065         
1066         if (qsa)
1067                 release_thread_qmcsampler(&R, shi->thread, qsa);
1068 }
1069
1070 static void trace_reflect(float *col, ShadeInput *shi, ShadeResult *shr, float fresnelfac)
1071 {
1072         QMCSampler *qsa=NULL;
1073         int samp_type;
1074         
1075         float samp3d[3], orthx[3], orthy[3];
1076         float v_nor_new[3], v_reflect[3];
1077         float sampcol[4], colsq[4];
1078                 
1079         float blur = pow(1.0 - shi->mat->gloss_mir, 3);
1080         short max_samples = shi->mat->samp_gloss_mir;
1081         float adapt_thresh = shi->mat->adapt_thresh_mir;
1082         float aniso = 1.0 - shi->mat->aniso_gloss_mir;
1083         
1084         int samples=0;
1085         
1086         col[0] = col[1] = col[2] = 0.0;
1087         colsq[0] = colsq[1] = colsq[2] = 0.0;
1088         
1089         if (blur > 0.0) {
1090                 if (adapt_thresh != 0.0) samp_type = SAMP_TYPE_HALTON;
1091                 else samp_type = SAMP_TYPE_HAMMERSLEY;
1092                         
1093                 /* all samples are generated per pixel */
1094                 qsa = get_thread_qmcsampler(&R, shi->thread, samp_type, max_samples);
1095                 QMC_initPixel(qsa, shi->thread);
1096         } else 
1097                 max_samples = 1;
1098         
1099         while (samples < max_samples) {
1100                                 
1101                 if (max_samples > 1) {
1102                         /* get a quasi-random vector from a phong-weighted disc */
1103                         QMC_samplePhong(samp3d, qsa, shi->thread, samples, blur);
1104
1105                         /* find the normal's perpendicular plane, blurring along tangents
1106                          * if tangent shading enabled */
1107                         if (shi->mat->mode & (MA_TANGENT_V)) {
1108                                 Crossf(orthx, shi->vn, shi->tang);      // bitangent
1109                                 VECCOPY(orthy, shi->tang);
1110                                 VecMulf(orthx, samp3d[0]);
1111                                 VecMulf(orthy, samp3d[1]*aniso);
1112                         } else {
1113                                 VecOrthoBasisf(shi->vn, orthx, orthy);
1114                                 VecMulf(orthx, samp3d[0]);
1115                                 VecMulf(orthy, samp3d[1]);
1116                         }
1117
1118                         /* and perturb the normal in it */
1119                         VecAddf(v_nor_new, shi->vn, orthx);
1120                         VecAddf(v_nor_new, v_nor_new, orthy);
1121                         Normalize(v_nor_new);
1122                 } else {
1123                         /* no blurriness, use the original normal */
1124                         VECCOPY(v_nor_new, shi->vn);
1125                 }
1126                 
1127                 if((shi->vlr->flag & R_SMOOTH)) 
1128                         reflection(v_reflect, v_nor_new, shi->view, shi->facenor);
1129                 else
1130                         reflection(v_reflect, v_nor_new, shi->view, NULL);
1131                 
1132                 traceray(shi, shr, shi->mat->ray_depth, shi->co, v_reflect, sampcol, shi->obi, shi->vlr, 0);
1133
1134                 
1135                 col[0] += sampcol[0];
1136                 col[1] += sampcol[1];
1137                 col[2] += sampcol[2];
1138         
1139                 /* for variance calc */
1140                 colsq[0] += sampcol[0]*sampcol[0];
1141                 colsq[1] += sampcol[1]*sampcol[1];
1142                 colsq[2] += sampcol[2]*sampcol[2];
1143                 
1144                 samples++;
1145
1146                 /* adaptive sampling */
1147                 if (adapt_thresh > 0.0 && samples > max_samples/3) 
1148                 {
1149                         if (adaptive_sample_variance(samples, col, colsq, adapt_thresh))
1150                                 break;
1151                         
1152                         /* if the pixel so far is very dark, we can get away with less samples */
1153                         if ( (col[0] + col[1] + col[2])/3.0/(float)samples < 0.01 )
1154                                 max_samples--;
1155                 
1156                         /* reduce samples when reflection is dim due to low ray mirror blend value or fresnel factor
1157                          * and when reflection is blurry */
1158                         if (fresnelfac < 0.1 * (blur+1)) {
1159                                 max_samples--;
1160                                 
1161                                 /* even more for very dim */
1162                                 if (fresnelfac < 0.05 * (blur+1)) 
1163                                         max_samples--;
1164                         }
1165                 }
1166         }
1167         
1168         col[0] /= (float)samples;
1169         col[1] /= (float)samples;
1170         col[2] /= (float)samples;
1171         
1172         if (qsa)
1173                 release_thread_qmcsampler(&R, shi->thread, qsa);
1174 }
1175
1176 /* extern call from render loop */
1177 void ray_trace(ShadeInput *shi, ShadeResult *shr)
1178 {
1179         float i, f, f1, fr, fg, fb;
1180         float mircol[4], tracol[4];
1181         float diff[3];
1182         int do_tra, do_mir;
1183         
1184         do_tra= ((shi->mat->mode & (MA_RAYTRANSP)) && shr->alpha!=1.0f);
1185         do_mir= ((shi->mat->mode & MA_RAYMIRROR) && shi->ray_mirror!=0.0f);
1186
1187         
1188         /* raytrace mirror amd refract like to separate the spec color */
1189         if(shi->combinedflag & SCE_PASS_SPEC)
1190                 VECSUB(diff, shr->combined, shr->spec) /* no ; */
1191         else
1192                 VECCOPY(diff, shr->combined);
1193         
1194         if(do_tra) {
1195                 float olddiff[3];
1196                 
1197                 trace_refract(tracol, shi, shr);
1198                 
1199                 f= shr->alpha; f1= 1.0f-f;
1200                 fr= 1.0f+ shi->mat->filter*(shi->r-1.0f);
1201                 fg= 1.0f+ shi->mat->filter*(shi->g-1.0f);
1202                 fb= 1.0f+ shi->mat->filter*(shi->b-1.0f);
1203                 
1204                 /* for refract pass */
1205                 VECCOPY(olddiff, diff);
1206                 
1207                 diff[0]= f*diff[0] + f1*fr*tracol[0];
1208                 diff[1]= f*diff[1] + f1*fg*tracol[1];
1209                 diff[2]= f*diff[2] + f1*fb*tracol[2];
1210                 
1211                 if(shi->passflag & SCE_PASS_REFRACT)
1212                         VECSUB(shr->refr, diff, olddiff);
1213                 
1214                 if(!(shi->combinedflag & SCE_PASS_REFRACT))
1215                         VECSUB(diff, diff, shr->refr);
1216                 
1217                 shr->alpha= tracol[3];
1218         }
1219         
1220         if(do_mir) {
1221         
1222                 i= shi->ray_mirror*fresnel_fac(shi->view, shi->vn, shi->mat->fresnel_mir_i, shi->mat->fresnel_mir);
1223                 if(i!=0.0f) {
1224                 
1225                         trace_reflect(mircol, shi, shr, i);
1226                         
1227                         fr= i*shi->mirr;
1228                         fg= i*shi->mirg;
1229                         fb= i*shi->mirb;
1230
1231                         if(shi->passflag & SCE_PASS_REFLECT) {
1232                                 /* mirror pass is not blocked out with spec */
1233                                 shr->refl[0]= fr*mircol[0] - fr*diff[0];
1234                                 shr->refl[1]= fg*mircol[1] - fg*diff[1];
1235                                 shr->refl[2]= fb*mircol[2] - fb*diff[2];
1236                         }
1237                         
1238                         if(shi->combinedflag & SCE_PASS_REFLECT) {
1239                                 
1240                                 f= fr*(1.0f-shr->spec[0]);      f1= 1.0f-i;
1241                                 diff[0]= f*mircol[0] + f1*diff[0];
1242                                 
1243                                 f= fg*(1.0f-shr->spec[1]);      f1= 1.0f-i;
1244                                 diff[1]= f*mircol[1] + f1*diff[1];
1245                                 
1246                                 f= fb*(1.0f-shr->spec[2]);      f1= 1.0f-i;
1247                                 diff[2]= f*mircol[2] + f1*diff[2];
1248                         }
1249                 }
1250         }
1251         /* put back together */
1252         if(shi->combinedflag & SCE_PASS_SPEC)
1253                 VECADD(shr->combined, diff, shr->spec) /* no ; */
1254         else
1255                 VECCOPY(shr->combined, diff);
1256 }
1257
1258 /* color 'shadfac' passes through 'col' with alpha and filter */
1259 /* filter is only applied on alpha defined transparent part */
1260 static void addAlphaLight(float *shadfac, float *col, float alpha, float filter)
1261 {
1262         float fr, fg, fb;
1263         
1264         fr= 1.0f+ filter*(col[0]-1.0f);
1265         fg= 1.0f+ filter*(col[1]-1.0f);
1266         fb= 1.0f+ filter*(col[2]-1.0f);
1267         
1268         shadfac[0]= alpha*col[0] + fr*(1.0f-alpha)*shadfac[0];
1269         shadfac[1]= alpha*col[1] + fg*(1.0f-alpha)*shadfac[1];
1270         shadfac[2]= alpha*col[2] + fb*(1.0f-alpha)*shadfac[2];
1271         
1272         shadfac[3]= (1.0f-alpha)*shadfac[3];
1273 }
1274
1275 static void ray_trace_shadow_tra(Isect *is, int depth, int traflag)
1276 {
1277         /* ray to lamp, find first face that intersects, check alpha properties,
1278            if it has col[3]>0.0f  continue. so exit when alpha is full */
1279         ShadeInput shi;
1280         ShadeResult shr;
1281
1282         if(RE_ray_tree_intersect(R.raytree, is)) {
1283                 float d= 1.0f;
1284                 /* we got a face */
1285                 
1286                 /* Warning, This is not that nice, and possibly a bit slow for every ray,
1287                 however some variables were not initialized properly in, unless using shade_input_initialize(...), we need to do a memset */
1288                 memset(&shi, 0, sizeof(ShadeInput)); 
1289                 /* end warning! - Campbell */
1290                 
1291                 shi.depth= 1;                                   /* only used to indicate tracing */
1292                 shi.mask= 1;
1293                 
1294                 /*shi.osatex= 0;
1295                 shi.thread= shi.sample= 0;
1296                 shi.lay= 0;
1297                 shi.passflag= 0;
1298                 shi.combinedflag= 0;
1299                 shi.do_preview= 0;
1300                 shi.light_override= NULL;
1301                 shi.mat_override= NULL;*/
1302                 
1303                 shade_ray(is, &shi, &shr);
1304                 if (traflag & RAY_TRA)
1305                         d= shade_by_transmission(is, &shi, &shr);
1306                 
1307                 /* mix colors based on shadfac (rgb + amount of light factor) */
1308                 addAlphaLight(is->col, shr.diff, shr.alpha, d*shi.mat->filter);
1309                 
1310                 if(depth>0 && is->col[3]>0.0f) {
1311                         
1312                         /* adapt isect struct */
1313                         VECCOPY(is->start, shi.co);
1314                         is->oborig= RAY_OBJECT_SET(&R, shi.obi);
1315                         is->faceorig= (RayFace*)shi.vlr;
1316
1317                         ray_trace_shadow_tra(is, depth-1, traflag | RAY_TRA);
1318                 }
1319         }
1320 }
1321
1322 /* not used, test function for ambient occlusion (yaf: pathlight) */
1323 /* main problem; has to be called within shading loop, giving unwanted recursion */
1324 int ray_trace_shadow_rad(ShadeInput *ship, ShadeResult *shr)
1325 {
1326         static int counter=0, only_one= 0;
1327         extern float hashvectf[];
1328         Isect isec;
1329         ShadeInput shi;
1330         ShadeResult shr_t;
1331         float vec[3], accum[3], div= 0.0f, maxsize= RE_ray_tree_max_size(R.raytree);
1332         int a;
1333         
1334         if(only_one) {
1335                 return 0;
1336         }
1337         only_one= 1;
1338         
1339         accum[0]= accum[1]= accum[2]= 0.0f;
1340         isec.mode= RE_RAY_MIRROR;
1341         isec.faceorig= (RayFace*)ship->vlr;
1342         isec.oborig= RAY_OBJECT_SET(&R, ship->obi);
1343         
1344         for(a=0; a<8*8; a++) {
1345                 
1346                 counter+=3;
1347                 counter %= 768;
1348                 VECCOPY(vec, hashvectf+counter);
1349                 if(ship->vn[0]*vec[0]+ship->vn[1]*vec[1]+ship->vn[2]*vec[2]>0.0f) {
1350                         vec[0]-= vec[0];
1351                         vec[1]-= vec[1];
1352                         vec[2]-= vec[2];
1353                 }
1354                 VECCOPY(isec.start, ship->co);
1355                 isec.end[0]= isec.start[0] + maxsize*vec[0];
1356                 isec.end[1]= isec.start[1] + maxsize*vec[1];
1357                 isec.end[2]= isec.start[2] + maxsize*vec[2];
1358                 
1359                 if(RE_ray_tree_intersect(R.raytree, &isec)) {
1360                         float fac;
1361                         
1362                         /* Warning, This is not that nice, and possibly a bit slow for every ray,
1363                         however some variables were not initialized properly in, unless using shade_input_initialize(...), we need to do a memset */
1364                         memset(&shi, 0, sizeof(ShadeInput)); 
1365                         /* end warning! - Campbell */
1366                         
1367                         shade_ray(&isec, &shi, &shr_t);
1368                         fac= isec.labda*isec.labda;
1369                         fac= 1.0f;
1370                         accum[0]+= fac*(shr_t.diff[0]+shr_t.spec[0]);
1371                         accum[1]+= fac*(shr_t.diff[1]+shr_t.spec[1]);
1372                         accum[2]+= fac*(shr_t.diff[2]+shr_t.spec[2]);
1373                         div+= fac;
1374                 }
1375                 else div+= 1.0f;
1376         }
1377         
1378         if(div!=0.0f) {
1379                 shr->diff[0]+= accum[0]/div;
1380                 shr->diff[1]+= accum[1]/div;
1381                 shr->diff[2]+= accum[2]/div;
1382         }
1383         shr->alpha= 1.0f;
1384         
1385         only_one= 0;
1386         return 1;
1387 }
1388
1389 /* aolight: function to create random unit sphere vectors for total random sampling */
1390 static void RandomSpherical(float *v)
1391 {
1392         float r;
1393         v[2] = 2.f*BLI_frand()-1.f;
1394         if ((r = 1.f - v[2]*v[2])>0.f) {
1395                 float a = 6.283185307f*BLI_frand();
1396                 r = sqrt(r);
1397                 v[0] = r * cos(a);
1398                 v[1] = r * sin(a);
1399         }
1400         else v[2] = 1.f;
1401 }
1402
1403 /* calc distributed spherical energy */
1404 static void DS_energy(float *sphere, int tot, float *vec)
1405 {
1406         float *fp, fac, force[3], res[3];
1407         int a;
1408         
1409         res[0]= res[1]= res[2]= 0.0f;
1410         
1411         for(a=0, fp=sphere; a<tot; a++, fp+=3) {
1412                 VecSubf(force, vec, fp);
1413                 fac= force[0]*force[0] + force[1]*force[1] + force[2]*force[2];
1414                 if(fac!=0.0f) {
1415                         fac= 1.0f/fac;
1416                         res[0]+= fac*force[0];
1417                         res[1]+= fac*force[1];
1418                         res[2]+= fac*force[2];
1419                 }
1420         }
1421
1422         VecMulf(res, 0.5);
1423         VecAddf(vec, vec, res);
1424         Normalize(vec);
1425         
1426 }
1427
1428 /* called from convertBlenderScene.c */
1429 /* creates an equally distributed spherical sample pattern */
1430 /* and allocates threadsafe memory */
1431 void init_ao_sphere(World *wrld)
1432 {
1433         float *fp;
1434         int a, tot, iter= 16;
1435
1436         /* we make twice the amount of samples, because only a hemisphere is used */
1437         tot= 2*wrld->aosamp*wrld->aosamp;
1438         
1439         wrld->aosphere= MEM_mallocN(3*tot*sizeof(float), "AO sphere");
1440         
1441         /* fixed random */
1442         BLI_srandom(tot);
1443         
1444         /* init */
1445         fp= wrld->aosphere;
1446         for(a=0; a<tot; a++, fp+= 3) {
1447                 RandomSpherical(fp);
1448         }
1449         
1450         while(iter--) {
1451                 for(a=0, fp= wrld->aosphere; a<tot; a++, fp+= 3) {
1452                         DS_energy(wrld->aosphere, tot, fp);
1453                 }
1454         }
1455         
1456         /* tables */
1457         wrld->aotables= MEM_mallocN(BLENDER_MAX_THREADS*3*tot*sizeof(float), "AO tables");
1458 }
1459
1460 /* give per thread a table, we have to compare xs ys because of way OSA works... */
1461 static float *threadsafe_table_sphere(int test, int thread, int xs, int ys, int tot)
1462 {
1463         static int xso[BLENDER_MAX_THREADS], yso[BLENDER_MAX_THREADS];
1464         static int firsttime= 1;
1465         
1466         if(firsttime) {
1467                 memset(xso, 255, sizeof(xso));
1468                 memset(yso, 255, sizeof(yso));
1469                 firsttime= 0;
1470         }
1471         
1472         if(xs==xso[thread] && ys==yso[thread]) return R.wrld.aotables+ thread*tot*3;
1473         if(test) return NULL;
1474         xso[thread]= xs; yso[thread]= ys;
1475         return R.wrld.aotables+ thread*tot*3;
1476 }
1477
1478 static float *sphere_sampler(int type, int resol, int thread, int xs, int ys)
1479 {
1480         int tot;
1481         float *vec;
1482         
1483         tot= 2*resol*resol;
1484
1485         if (type & WO_AORNDSMP) {
1486                 float *sphere;
1487                 int a;
1488                 
1489                 // always returns table
1490                 sphere= threadsafe_table_sphere(0, thread, xs, ys, tot);
1491
1492                 /* total random sampling. NOT THREADSAFE! (should be removed, is not useful) */
1493                 vec= sphere;
1494                 for (a=0; a<tot; a++, vec+=3) {
1495                         RandomSpherical(vec);
1496                 }
1497                 
1498                 return sphere;
1499         } 
1500         else {
1501                 float *sphere;
1502                 float cosfi, sinfi, cost, sint;
1503                 float ang, *vec1;
1504                 int a;
1505                 
1506                 // returns table if xs and ys were equal to last call
1507                 sphere= threadsafe_table_sphere(1, thread, xs, ys, tot);
1508                 if(sphere==NULL) {
1509                         sphere= threadsafe_table_sphere(0, thread, xs, ys, tot);
1510                         
1511                         // random rotation
1512                         ang= BLI_thread_frand(thread);
1513                         sinfi= sin(ang); cosfi= cos(ang);
1514                         ang= BLI_thread_frand(thread);
1515                         sint= sin(ang); cost= cos(ang);
1516                         
1517                         vec= R.wrld.aosphere;
1518                         vec1= sphere;
1519                         for (a=0; a<tot; a++, vec+=3, vec1+=3) {
1520                                 vec1[0]= cost*cosfi*vec[0] - sinfi*vec[1] + sint*cosfi*vec[2];
1521                                 vec1[1]= cost*sinfi*vec[0] + cosfi*vec[1] + sint*sinfi*vec[2];
1522                                 vec1[2]= -sint*vec[0] + cost*vec[2];                    
1523                         }
1524                 }
1525                 return sphere;
1526         }
1527 }
1528
1529 static void ray_ao_qmc(ShadeInput *shi, float *shadfac)
1530 {
1531         Isect isec;
1532         QMCSampler *qsa=NULL;
1533         float samp3d[3];
1534         float up[3], side[3], dir[3], nrm[3];
1535         
1536         float maxdist = R.wrld.aodist;
1537         float fac=0.0f, prev=0.0f;
1538         float adapt_thresh = G.scene->world->ao_adapt_thresh;
1539         float adapt_speed_fac = G.scene->world->ao_adapt_speed_fac;
1540         
1541         int samples=0;
1542         int max_samples = R.wrld.aosamp*R.wrld.aosamp;
1543         
1544         float dxyview[3], skyadded=0, div;
1545         int aocolor;
1546         
1547         isec.faceorig= (RayFace*)shi->vlr;
1548         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
1549         isec.face_last= NULL;
1550         isec.ob_last= 0;
1551         isec.mode= (R.wrld.aomode & WO_AODIST)?RE_RAY_SHADOW_TRA:RE_RAY_SHADOW;
1552         isec.lay= -1;
1553         
1554         shadfac[0]= shadfac[1]= shadfac[2]= 0.0f;
1555         
1556         /* prevent sky colors to be added for only shadow (shadow becomes alpha) */
1557         aocolor= R.wrld.aocolor;
1558         if(shi->mat->mode & MA_ONLYSHADOW)
1559                 aocolor= WO_AOPLAIN;
1560         
1561         if(aocolor == WO_AOSKYTEX) {
1562                 dxyview[0]= 1.0f/(float)R.wrld.aosamp;
1563                 dxyview[1]= 1.0f/(float)R.wrld.aosamp;
1564                 dxyview[2]= 0.0f;
1565         }
1566         
1567         if(shi->vlr->flag & R_SMOOTH) {
1568                 VECCOPY(nrm, shi->vn);
1569         }
1570         else {
1571                 VECCOPY(nrm, shi->facenor);
1572         }
1573         
1574         VecOrthoBasisf(nrm, up, side);
1575         
1576         /* sampling init */
1577         if (R.wrld.ao_samp_method==WO_AOSAMP_HALTON) {
1578                 float speedfac;
1579                 
1580                 speedfac = get_avg_speed(shi) * adapt_speed_fac;
1581                 CLAMP(speedfac, 1.0, 1000.0);
1582                 max_samples /= speedfac;
1583                 if (max_samples < 5) max_samples = 5;
1584                 
1585                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HALTON, max_samples);
1586         } else if (R.wrld.ao_samp_method==WO_AOSAMP_HAMMERSLEY)
1587                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HAMMERSLEY, max_samples);
1588
1589         QMC_initPixel(qsa, shi->thread);
1590         
1591         while (samples < max_samples) {
1592
1593                 /* sampling, returns quasi-random vector in unit hemisphere */
1594                 QMC_sampleHemi(samp3d, qsa, shi->thread, samples);
1595
1596                 dir[0] = (samp3d[0]*up[0] + samp3d[1]*side[0] + samp3d[2]*nrm[0]);
1597                 dir[1] = (samp3d[0]*up[1] + samp3d[1]*side[1] + samp3d[2]*nrm[1]);
1598                 dir[2] = (samp3d[0]*up[2] + samp3d[1]*side[2] + samp3d[2]*nrm[2]);
1599                 
1600                 Normalize(dir);
1601                         
1602                 VECCOPY(isec.start, shi->co);
1603                 isec.end[0] = shi->co[0] - maxdist*dir[0];
1604                 isec.end[1] = shi->co[1] - maxdist*dir[1];
1605                 isec.end[2] = shi->co[2] - maxdist*dir[2];
1606                 
1607                 prev = fac;
1608                 
1609                 if(RE_ray_tree_intersect(R.raytree, &isec)) {
1610                         if (R.wrld.aomode & WO_AODIST) fac+= exp(-isec.labda*R.wrld.aodistfac); 
1611                         else fac+= 1.0f;
1612                 }
1613                 else if(aocolor!=WO_AOPLAIN) {
1614                         float skycol[4];
1615                         float skyfac, view[3];
1616                         
1617                         view[0]= -dir[0];
1618                         view[1]= -dir[1];
1619                         view[2]= -dir[2];
1620                         Normalize(view);
1621                         
1622                         if(aocolor==WO_AOSKYCOL) {
1623                                 skyfac= 0.5*(1.0f+view[0]*R.grvec[0]+ view[1]*R.grvec[1]+ view[2]*R.grvec[2]);
1624                                 shadfac[0]+= (1.0f-skyfac)*R.wrld.horr + skyfac*R.wrld.zenr;
1625                                 shadfac[1]+= (1.0f-skyfac)*R.wrld.horg + skyfac*R.wrld.zeng;
1626                                 shadfac[2]+= (1.0f-skyfac)*R.wrld.horb + skyfac*R.wrld.zenb;
1627                         }
1628                         else {  /* WO_AOSKYTEX */
1629                                 shadeSkyView(skycol, isec.start, view, dxyview);
1630                                 shadfac[0]+= skycol[0];
1631                                 shadfac[1]+= skycol[1];
1632                                 shadfac[2]+= skycol[2];
1633                         }
1634                         skyadded++;
1635                 }
1636                 
1637                 samples++;
1638                 
1639                 if (qsa->type == SAMP_TYPE_HALTON) {
1640                         /* adaptive sampling - consider samples below threshold as in shadow (or vice versa) and exit early */          
1641                         if (adapt_thresh > 0.0 && (samples > max_samples/2) ) {
1642                                 
1643                                 if (adaptive_sample_contrast_val(samples, prev, fac, adapt_thresh)) {
1644                                         break;
1645                                 }
1646                         }
1647                 }
1648         }
1649         
1650         if(aocolor!=WO_AOPLAIN && skyadded) {
1651                 div= (1.0f - fac/(float)samples)/((float)skyadded);
1652                 
1653                 shadfac[0]*= div;       // average color times distances/hits formula
1654                 shadfac[1]*= div;       // average color times distances/hits formula
1655                 shadfac[2]*= div;       // average color times distances/hits formula
1656         } else {
1657                 shadfac[0]= shadfac[1]= shadfac[2]= 1.0f - fac/(float)samples;
1658         }
1659         
1660         if (qsa)
1661                 release_thread_qmcsampler(&R, shi->thread, qsa);
1662 }
1663
1664 /* extern call from shade_lamp_loop, ambient occlusion calculus */
1665 static void ray_ao_spheresamp(ShadeInput *shi, float *shadfac)
1666 {
1667         Isect isec;
1668         float *vec, *nrm, div, bias, sh=0.0f;
1669         float maxdist = R.wrld.aodist;
1670         float dxyview[3];
1671         int j= -1, tot, actual=0, skyadded=0, aocolor, resol= R.wrld.aosamp;
1672         
1673         isec.faceorig= (RayFace*)shi->vlr;
1674         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
1675         isec.face_last= NULL;
1676         isec.ob_last= 0;
1677         isec.mode= (R.wrld.aomode & WO_AODIST)?RE_RAY_SHADOW_TRA:RE_RAY_SHADOW;
1678         isec.lay= -1;
1679
1680
1681         shadfac[0]= shadfac[1]= shadfac[2]= 0.0f;
1682
1683         /* bias prevents smoothed faces to appear flat */
1684         if(shi->vlr->flag & R_SMOOTH) {
1685                 bias= G.scene->world->aobias;
1686                 nrm= shi->vn;
1687         }
1688         else {
1689                 bias= 0.0f;
1690                 nrm= shi->facenor;
1691         }
1692
1693         /* prevent sky colors to be added for only shadow (shadow becomes alpha) */
1694         aocolor= R.wrld.aocolor;
1695         if(shi->mat->mode & MA_ONLYSHADOW)
1696                 aocolor= WO_AOPLAIN;
1697         
1698         if(resol>32) resol= 32;
1699         
1700         vec= sphere_sampler(R.wrld.aomode, resol, shi->thread, shi->xs, shi->ys);
1701         
1702         // warning: since we use full sphere now, and dotproduct is below, we do twice as much
1703         tot= 2*resol*resol;
1704
1705         if(aocolor == WO_AOSKYTEX) {
1706                 dxyview[0]= 1.0f/(float)resol;
1707                 dxyview[1]= 1.0f/(float)resol;
1708                 dxyview[2]= 0.0f;
1709         }
1710         
1711         while(tot--) {
1712                 
1713                 if ((vec[0]*nrm[0] + vec[1]*nrm[1] + vec[2]*nrm[2]) > bias) {
1714                         /* only ao samples for mask */
1715                         if(R.r.mode & R_OSA) {
1716                                 j++;
1717                                 if(j==R.osa) j= 0;
1718                                 if(!(shi->mask & (1<<j))) {
1719                                         vec+=3;
1720                                         continue;
1721                                 }
1722                         }
1723                         
1724                         actual++;
1725                         
1726                         /* always set start/end, RE_ray_tree_intersect clips it */
1727                         VECCOPY(isec.start, shi->co);
1728                         isec.end[0] = shi->co[0] - maxdist*vec[0];
1729                         isec.end[1] = shi->co[1] - maxdist*vec[1];
1730                         isec.end[2] = shi->co[2] - maxdist*vec[2];
1731                         
1732                         /* do the trace */
1733                         if(RE_ray_tree_intersect(R.raytree, &isec)) {
1734                                 if (R.wrld.aomode & WO_AODIST) sh+= exp(-isec.labda*R.wrld.aodistfac); 
1735                                 else sh+= 1.0f;
1736                         }
1737                         else if(aocolor!=WO_AOPLAIN) {
1738                                 float skycol[4];
1739                                 float fac, view[3];
1740                                 
1741                                 view[0]= -vec[0];
1742                                 view[1]= -vec[1];
1743                                 view[2]= -vec[2];
1744                                 Normalize(view);
1745                                 
1746                                 if(aocolor==WO_AOSKYCOL) {
1747                                         fac= 0.5*(1.0f+view[0]*R.grvec[0]+ view[1]*R.grvec[1]+ view[2]*R.grvec[2]);
1748                                         shadfac[0]+= (1.0f-fac)*R.wrld.horr + fac*R.wrld.zenr;
1749                                         shadfac[1]+= (1.0f-fac)*R.wrld.horg + fac*R.wrld.zeng;
1750                                         shadfac[2]+= (1.0f-fac)*R.wrld.horb + fac*R.wrld.zenb;
1751                                 }
1752                                 else {  /* WO_AOSKYTEX */
1753                                         shadeSkyView(skycol, isec.start, view, dxyview);
1754                                         shadfac[0]+= skycol[0];
1755                                         shadfac[1]+= skycol[1];
1756                                         shadfac[2]+= skycol[2];
1757                                 }
1758                                 skyadded++;
1759                         }
1760                 }
1761                 // samples
1762                 vec+= 3;
1763         }
1764         
1765         if(actual==0) sh= 1.0f;
1766         else sh = 1.0f - sh/((float)actual);
1767         
1768         if(aocolor!=WO_AOPLAIN && skyadded) {
1769                 div= sh/((float)skyadded);
1770                 
1771                 shadfac[0]*= div;       // average color times distances/hits formula
1772                 shadfac[1]*= div;       // average color times distances/hits formula
1773                 shadfac[2]*= div;       // average color times distances/hits formula
1774         }
1775         else {
1776                 shadfac[0]= shadfac[1]= shadfac[2]= sh;
1777         }
1778 }
1779
1780 void ray_ao(ShadeInput *shi, float *shadfac)
1781 {
1782         /* Unfortunately, the unusual way that the sphere sampler calculates roughly twice as many
1783          * samples as are actually traced, and skips them based on bias and OSA settings makes it very difficult
1784          * to reuse code between these two functions. This is the easiest way I can think of to do it
1785          * --broken */
1786         if (ELEM(R.wrld.ao_samp_method, WO_AOSAMP_HAMMERSLEY, WO_AOSAMP_HALTON))
1787                 ray_ao_qmc(shi, shadfac);
1788         else if (R.wrld.ao_samp_method == WO_AOSAMP_CONSTANT)
1789                 ray_ao_spheresamp(shi, shadfac);
1790 }
1791
1792
1793 static void ray_shadow_qmc(ShadeInput *shi, LampRen *lar, float *lampco, float *shadfac, Isect *isec)
1794 {
1795         QMCSampler *qsa=NULL;
1796         QMCSampler *qsa_jit=NULL;
1797         int samples=0;
1798         float samp3d[3], jit[3], jitbias= 0.0f;
1799
1800         float fac=0.0f, vec[3];
1801         float colsq[4];
1802         float adapt_thresh = lar->adapt_thresh;
1803         int max_samples = lar->ray_totsamp;
1804         float pos[3];
1805         int do_soft=1, full_osa=0;
1806
1807         colsq[0] = colsq[1] = colsq[2] = 0.0;
1808         if(isec->mode==RE_RAY_SHADOW_TRA) {
1809                 shadfac[0]= shadfac[1]= shadfac[2]= shadfac[3]= 0.0f;
1810         } else
1811                 shadfac[3]= 1.0f;
1812         
1813         if (lar->ray_totsamp < 2) do_soft = 0;
1814         if ((R.r.mode & R_OSA) && (R.osa > 0) && (shi->vlr->flag & R_FULL_OSA)) full_osa = 1;
1815         
1816         if (full_osa) {
1817                 if (do_soft) max_samples  = max_samples/R.osa + 1;
1818                 else max_samples = 1;
1819         } else {
1820                 if (do_soft) max_samples = lar->ray_totsamp;
1821                 else max_samples = (R.osa > 4)?R.osa:5;
1822         }
1823
1824         if(shi->vlr && ((shi->vlr->flag & R_FULL_OSA) == 0))
1825                 jitbias= 0.5f*(VecLength(shi->dxco) + VecLength(shi->dyco));
1826
1827         /* sampling init */
1828         if (lar->ray_samp_method==LA_SAMP_HALTON) {
1829                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HALTON, max_samples);
1830                 qsa_jit = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HALTON, max_samples);
1831         } else if (lar->ray_samp_method==LA_SAMP_HAMMERSLEY) {
1832                 qsa = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HAMMERSLEY, max_samples);
1833                 qsa_jit = get_thread_qmcsampler(&R, shi->thread, SAMP_TYPE_HAMMERSLEY, max_samples);
1834         }
1835         
1836         QMC_initPixel(qsa, shi->thread);
1837         QMC_initPixel(qsa_jit, shi->thread);
1838         
1839         VECCOPY(vec, lampco);
1840         
1841         
1842         while (samples < max_samples) {
1843                 isec->faceorig= (RayFace*)shi->vlr;
1844                 isec->oborig= RAY_OBJECT_SET(&R, shi->obi);
1845                 
1846                 /* manually jitter the start shading co-ord per sample
1847                  * based on the pre-generated OSA texture sampling offsets, 
1848                  * for anti-aliasing sharp shadow edges. */
1849                 VECCOPY(pos, shi->co);
1850                 if (shi->vlr && !full_osa) {
1851                         QMC_sampleRect(jit, qsa_jit, shi->thread, samples, 1.0, 1.0);
1852                         
1853                         pos[0] += shi->dxco[0]*jit[0] + shi->dyco[0]*jit[1];
1854                         pos[1] += shi->dxco[1]*jit[0] + shi->dyco[1]*jit[1];
1855                         pos[2] += shi->dxco[2]*jit[0] + shi->dyco[2]*jit[1];
1856                 }
1857
1858                 if (do_soft) {
1859                         /* sphere shadow source */
1860                         if (lar->type == LA_LOCAL) {
1861                                 float ru[3], rv[3], v[3], s[3];
1862                                 
1863                                 /* calc tangent plane vectors */
1864                                 v[0] = pos[0] - lampco[0];
1865                                 v[1] = pos[1] - lampco[1];
1866                                 v[2] = pos[2] - lampco[2];
1867                                 Normalize(v);
1868                                 VecOrthoBasisf(v, ru, rv);
1869                                 
1870                                 /* sampling, returns quasi-random vector in area_size disc */
1871                                 QMC_sampleDisc(samp3d, qsa, shi->thread, samples,lar->area_size);
1872
1873                                 /* distribute disc samples across the tangent plane */
1874                                 s[0] = samp3d[0]*ru[0] + samp3d[1]*rv[0];
1875                                 s[1] = samp3d[0]*ru[1] + samp3d[1]*rv[1];
1876                                 s[2] = samp3d[0]*ru[2] + samp3d[1]*rv[2];
1877                                 
1878                                 VECCOPY(samp3d, s);
1879
1880                                 if(jitbias != 0.0f) {
1881                                         /* bias away somewhat to avoid self intersection */
1882                                         pos[0] -= jitbias*v[0];
1883                                         pos[1] -= jitbias*v[1];
1884                                         pos[2] -= jitbias*v[2];
1885                                 }
1886                         }
1887                         else {
1888                                 /* sampling, returns quasi-random vector in [sizex,sizey]^2 plane */
1889                                 QMC_sampleRect(samp3d, qsa, shi->thread, samples, lar->area_size, lar->area_sizey);
1890                                                                 
1891                                 /* align samples to lamp vector */
1892                                 Mat3MulVecfl(lar->mat, samp3d);
1893                         }
1894                         isec->end[0]= vec[0]+samp3d[0];
1895                         isec->end[1]= vec[1]+samp3d[1];
1896                         isec->end[2]= vec[2]+samp3d[2];
1897                 } else {
1898                         VECCOPY(isec->end, vec);
1899                 }
1900
1901                 if(jitbias != 0.0f && !(do_soft && lar->type==LA_LOCAL)) {
1902                         /* bias away somewhat to avoid self intersection */
1903                         float v[3];
1904
1905                         VECSUB(v, pos, isec->end);
1906                         Normalize(v);
1907
1908                         pos[0] -= jitbias*v[0];
1909                         pos[1] -= jitbias*v[1];
1910                         pos[2] -= jitbias*v[2];
1911                 }
1912
1913                 VECCOPY(isec->start, pos);
1914                 
1915                 
1916                 /* trace the ray */
1917                 if(isec->mode==RE_RAY_SHADOW_TRA) {
1918                         isec->col[0]= isec->col[1]= isec->col[2]=  1.0f;
1919                         isec->col[3]= 1.0f;
1920                         
1921                         ray_trace_shadow_tra(isec, DEPTH_SHADOW_TRA, 0);
1922                         shadfac[0] += isec->col[0];
1923                         shadfac[1] += isec->col[1];
1924                         shadfac[2] += isec->col[2];
1925                         shadfac[3] += isec->col[3];
1926                         
1927                         /* for variance calc */
1928                         colsq[0] += isec->col[0]*isec->col[0];
1929                         colsq[1] += isec->col[1]*isec->col[1];
1930                         colsq[2] += isec->col[2]*isec->col[2];
1931                 }
1932                 else {
1933                         if( RE_ray_tree_intersect(R.raytree, isec) ) fac+= 1.0f;
1934                 }
1935                 
1936                 samples++;
1937                 
1938                 if ((lar->ray_samp_method == LA_SAMP_HALTON)) {
1939                 
1940                         /* adaptive sampling - consider samples below threshold as in shadow (or vice versa) and exit early */
1941                         if ((max_samples > 4) && (adapt_thresh > 0.0) && (samples > max_samples / 3)) {
1942                                 if (isec->mode==RE_RAY_SHADOW_TRA) {
1943                                         if ((shadfac[3] / samples > (1.0-adapt_thresh)) || (shadfac[3] / samples < adapt_thresh))
1944                                                 break;
1945                                         else if (adaptive_sample_variance(samples, shadfac, colsq, adapt_thresh))
1946                                                 break;
1947                                 } else {
1948                                         if ((fac / samples > (1.0-adapt_thresh)) || (fac / samples < adapt_thresh))
1949                                                 break;
1950                                 }
1951                         }
1952                 }
1953         }
1954         
1955         if(isec->mode==RE_RAY_SHADOW_TRA) {
1956                 shadfac[0] /= samples;
1957                 shadfac[1] /= samples;
1958                 shadfac[2] /= samples;
1959                 shadfac[3] /= samples;
1960         } else
1961                 shadfac[3]= 1.0f-fac/samples;
1962
1963         if (qsa_jit)
1964                 release_thread_qmcsampler(&R, shi->thread, qsa_jit);
1965         if (qsa)
1966                 release_thread_qmcsampler(&R, shi->thread, qsa);
1967 }
1968
1969 static void ray_shadow_jitter(ShadeInput *shi, LampRen *lar, float *lampco, float *shadfac, Isect *isec)
1970 {
1971         /* area soft shadow */
1972         float *jitlamp;
1973         float fac=0.0f, div=0.0f, vec[3];
1974         int a, j= -1, mask;
1975         
1976         if(isec->mode==RE_RAY_SHADOW_TRA) {
1977                 shadfac[0]= shadfac[1]= shadfac[2]= shadfac[3]= 0.0f;
1978         }
1979         else shadfac[3]= 1.0f;
1980         
1981         fac= 0.0f;
1982         jitlamp= give_jitter_plane(lar, shi->thread, shi->xs, shi->ys);
1983
1984         a= lar->ray_totsamp;
1985         
1986         /* this correction to make sure we always take at least 1 sample */
1987         mask= shi->mask;
1988         if(a==4) mask |= (mask>>4)|(mask>>8);
1989         else if(a==9) mask |= (mask>>9);
1990         
1991         while(a--) {
1992                 
1993                 if(R.r.mode & R_OSA) {
1994                         j++;
1995                         if(j>=R.osa) j= 0;
1996                         if(!(mask & (1<<j))) {
1997                                 jitlamp+= 2;
1998                                 continue;
1999                         }
2000                 }
2001                 
2002                 isec->faceorig= (RayFace*)shi->vlr;
2003                 isec->oborig= RAY_OBJECT_SET(&R, shi->obi);
2004                 
2005                 vec[0]= jitlamp[0];
2006                 vec[1]= jitlamp[1];
2007                 vec[2]= 0.0f;
2008                 Mat3MulVecfl(lar->mat, vec);
2009                 
2010                 /* set start and end, RE_ray_tree_intersect clips it */
2011                 VECCOPY(isec->start, shi->co);
2012                 isec->end[0]= lampco[0]+vec[0];
2013                 isec->end[1]= lampco[1]+vec[1];
2014                 isec->end[2]= lampco[2]+vec[2];
2015                 
2016                 if(isec->mode==RE_RAY_SHADOW_TRA) {
2017                         /* isec.col is like shadfac, so defines amount of light (0.0 is full shadow) */
2018                         isec->col[0]= isec->col[1]= isec->col[2]=  1.0f;
2019                         isec->col[3]= 1.0f;
2020                         
2021                         ray_trace_shadow_tra(isec, DEPTH_SHADOW_TRA, 0);
2022                         shadfac[0] += isec->col[0];
2023                         shadfac[1] += isec->col[1];
2024                         shadfac[2] += isec->col[2];
2025                         shadfac[3] += isec->col[3];
2026                 }
2027                 else if( RE_ray_tree_intersect(R.raytree, isec) ) fac+= 1.0f;
2028                 
2029                 div+= 1.0f;
2030                 jitlamp+= 2;
2031         }
2032         
2033         if(isec->mode==RE_RAY_SHADOW_TRA) {
2034                 shadfac[0] /= div;
2035                 shadfac[1] /= div;
2036                 shadfac[2] /= div;
2037                 shadfac[3] /= div;
2038         }
2039         else {
2040                 // sqrt makes nice umbra effect
2041                 if(lar->ray_samp_type & LA_SAMP_UMBRA)
2042                         shadfac[3]= sqrt(1.0f-fac/div);
2043                 else
2044                         shadfac[3]= 1.0f-fac/div;
2045         }
2046 }
2047 /* extern call from shade_lamp_loop */
2048 void ray_shadow(ShadeInput *shi, LampRen *lar, float *shadfac)
2049 {
2050         Isect isec;
2051         float lampco[3], maxsize;
2052
2053         /* setup isec */
2054         if(shi->mat->mode & MA_SHADOW_TRA) isec.mode= RE_RAY_SHADOW_TRA;
2055         else isec.mode= RE_RAY_SHADOW;
2056         
2057         if(lar->mode & (LA_LAYER|LA_LAYER_SHADOW))
2058                 isec.lay= lar->lay;
2059         else
2060                 isec.lay= -1;
2061
2062         /* only when not mir tracing, first hit optimm */
2063         if(shi->depth==0) {
2064                 isec.face_last= (RayFace*)lar->vlr_last[shi->thread];
2065                 isec.ob_last= RAY_OBJECT_SET(&R, lar->obi_last[shi->thread]);
2066         }
2067         else {
2068                 isec.face_last= NULL;
2069                 isec.ob_last= 0;
2070         }
2071         
2072         if(lar->type==LA_SUN || lar->type==LA_HEMI) {
2073                 maxsize= RE_ray_tree_max_size(R.raytree);
2074                 lampco[0]= shi->co[0] - maxsize*lar->vec[0];
2075                 lampco[1]= shi->co[1] - maxsize*lar->vec[1];
2076                 lampco[2]= shi->co[2] - maxsize*lar->vec[2];
2077         }
2078         else {
2079                 VECCOPY(lampco, lar->co);
2080         }
2081         
2082         if (ELEM(lar->ray_samp_method, LA_SAMP_HALTON, LA_SAMP_HAMMERSLEY)) {
2083                 
2084                 ray_shadow_qmc(shi, lar, lampco, shadfac, &isec);
2085                 
2086         } else {
2087                 if(lar->ray_totsamp<2) {
2088                         
2089                         isec.faceorig= (RayFace*)shi->vlr;
2090                         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
2091                         shadfac[3]= 1.0f; // 1.0=full light
2092                         
2093                         /* set up isec vec */
2094                         VECCOPY(isec.start, shi->co);
2095                         VECCOPY(isec.end, lampco);
2096
2097                         if(isec.mode==RE_RAY_SHADOW_TRA) {
2098                                 /* isec.col is like shadfac, so defines amount of light (0.0 is full shadow) */
2099                                 isec.col[0]= isec.col[1]= isec.col[2]=  1.0f;
2100                                 isec.col[3]= 1.0f;
2101
2102                                 ray_trace_shadow_tra(&isec, DEPTH_SHADOW_TRA, 0);
2103                                 QUATCOPY(shadfac, isec.col);
2104                         }
2105                         else if(RE_ray_tree_intersect(R.raytree, &isec)) shadfac[3]= 0.0f;
2106                 }
2107                 else {
2108                         ray_shadow_jitter(shi, lar, lampco, shadfac, &isec);
2109                 }
2110         }
2111                 
2112         /* for first hit optim, set last interesected shadow face */
2113         if(shi->depth==0) {
2114                 lar->vlr_last[shi->thread]= (VlakRen*)isec.face_last;
2115                 lar->obi_last[shi->thread]= RAY_OBJECT_GET(&R, isec.ob_last);
2116         }
2117
2118 }
2119
2120 /* only when face points away from lamp, in direction of lamp, trace ray and find first exit point */
2121 static void ray_translucent(ShadeInput *shi, LampRen *lar, float *distfac, float *co)
2122 {
2123         Isect isec;
2124         float lampco[3], maxsize;
2125         
2126         /* setup isec */
2127         isec.mode= RE_RAY_SHADOW_TRA;
2128         
2129         if(lar->mode & LA_LAYER) isec.lay= lar->lay; else isec.lay= -1;
2130         
2131         if(lar->type==LA_SUN || lar->type==LA_HEMI) {
2132                 maxsize= RE_ray_tree_max_size(R.raytree);
2133                 lampco[0]= shi->co[0] - maxsize*lar->vec[0];
2134                 lampco[1]= shi->co[1] - maxsize*lar->vec[1];
2135                 lampco[2]= shi->co[2] - maxsize*lar->vec[2];
2136         }
2137         else {
2138                 VECCOPY(lampco, lar->co);
2139         }
2140         
2141         isec.faceorig= (RayFace*)shi->vlr;
2142         isec.oborig= RAY_OBJECT_SET(&R, shi->obi);
2143         
2144         /* set up isec vec */
2145         VECCOPY(isec.start, shi->co);
2146         VECCOPY(isec.end, lampco);
2147         
2148         if(RE_ray_tree_intersect(R.raytree, &isec)) {
2149                 /* we got a face */
2150                 
2151                 /* render co */
2152                 co[0]= isec.start[0]+isec.labda*(isec.vec[0]);
2153                 co[1]= isec.start[1]+isec.labda*(isec.vec[1]);
2154                 co[2]= isec.start[2]+isec.labda*(isec.vec[2]);
2155                 
2156                 *distfac= VecLength(isec.vec);
2157         }
2158         else
2159                 *distfac= 0.0f;
2160 }
2161
2162