archive/blender-archive
Archived
1
1
Fork 0
Code Pull Requests Packages Activity
This repository has been archived on 2023-10-09. You can view files and clone it, but cannot push or open issues or pull requests.
Files
ea13ec1699bcf59960daa57fd2cd9d24e195a71b
blender-archive/source/blender/render/intern/raytrace/rayobject.cpp

554 lines
13 KiB
C++
Raw Normal View History

doxygen: prevent GPL license block from being parsed as doxygen comment.
2011-02-23 10:52:22 +00:00
/*
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
* ***** BEGIN GPL LICENSE BLOCK *****
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software Foundation,
correct fsf address
2010-02-12 13:34:04 +00:00
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
*
* The Original Code is Copyright (C) 2009 Blender Foundation.
* All rights reserved.
*
* The Original Code is: all of this file.
*
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
* Contributor(s): André Pinto.
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
*
* ***** END GPL LICENSE BLOCK *****
*/
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
doxygen: blender/render tagged.
2011-02-27 19:31:27 +00:00
/** \file blender/render/intern/raytrace/rayobject.cpp
* \ingroup render
*/
Renamed exported functions from render to have the RE_ prefix RayObject_* => RE_rayobject_*
2009-05-13 01:56:03 +00:00
#include <assert.h>
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
#include "MEM_guardedalloc.h"
Math Lib * Convert all code to use new functions. * Branch maintainers may want to skip this commit, and run this conversion script instead, if they use a lot of math functions in new code: http://www.pasteall.org/9052/python
2009-11-10 20:43:45 +00:00
#include "BLI_math.h"
split BKE_utildefines.h, now it only has blender specific defines like GS() MAKE_ID, FILE_MAXDIR, moved the generic defines to BLI_utildefines.h. no functional changes.
2011-01-07 18:36:47 +00:00
#include "BLI_utildefines.h"
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
#include "DNA_material_types.h"
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
#include "rayintersection.h"
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
#include "rayobject.h"
*Ray counters (number of BB/primitive tests/hits and other raytrace counters) can now be enabled/disabled at compile-time. #define RE_RAYCOUNTER (/source/blender/render/extern/include/RE_raytrace.h) *Some other small organization on code (will be disable as this only matters for testing and developping)
2009-09-13 20:59:25 +00:00
#include "raycounter.h"
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
#include "render_types.h"
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* RayFace
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
note we force always inline here, because compiler refuses to otherwise
because function is too long. Since this is code that is called billions
of times we really do want to inline. */
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
tag unused function args in raytrace code.
2012-01-14 10:08:47 +00:00
MALWAYS_INLINE RayObject* rayface_from_coords(RayFace *rayface, void *ob, void *face,
float *v1, float *v2, float *v3, float *v4)
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
rayface->ob = ob;
rayface->face = face;
copy_v3_v3(rayface->v1, v1);
copy_v3_v3(rayface->v2, v2);
copy_v3_v3(rayface->v3, v3);
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(v4)
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
copy_v3_v3(rayface->v4, v4);
rayface->quad = 1;
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
}
else
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
rayface->quad = 0;
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
return RE_rayobject_unalignRayFace(rayface);
}
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
MALWAYS_INLINE void rayface_from_vlak(RayFace *rayface, ObjectInstanceRen *obi, VlakRen *vlr)
{
rayface_from_coords(rayface, obi, vlr, vlr->v1->co, vlr->v2->co, vlr->v3->co, vlr->v4 ? vlr->v4->co : 0);
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(obi->transform_primitives)
{
mul_m4_v3(obi->mat, rayface->v1);
mul_m4_v3(obi->mat, rayface->v2);
mul_m4_v3(obi->mat, rayface->v3);
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(RE_rayface_isQuad(rayface))
mul_m4_v3(obi->mat, rayface->v4);
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
}
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
RayObject* RE_rayface_from_vlak(RayFace *rayface, ObjectInstanceRen *obi, VlakRen *vlr)
{
return rayface_from_coords(rayface, obi, vlr, vlr->v1->co, vlr->v2->co, vlr->v3->co, vlr->v4 ? vlr->v4->co : 0);
}
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* VlakPrimitive */
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
RayObject* RE_vlakprimitive_from_vlak(VlakPrimitive *face, struct ObjectInstanceRen *obi, struct VlakRen *vlr)
{
face->ob = obi;
face->face = vlr;
return RE_rayobject_unalignVlakPrimitive(face);
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* Checks for ignoring faces or materials */
MALWAYS_INLINE int vlr_check_intersect(Isect *is, ObjectInstanceRen *obi, VlakRen *vlr)
Bug fix (layers and some other checks were disabled) *bug found and isolated by ZanQdo (Daniel Salazar)
2009-09-07 00:58:17 +00:00
{
/* for baking selected to active non-traceable materials might still
* be in the raytree */
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(!(vlr->flag & R_TRACEBLE))
Bug fix (layers and some other checks were disabled) *bug found and isolated by ZanQdo (Daniel Salazar)
2009-09-07 00:58:17 +00:00
return 0;
/* I know... cpu cycle waste, might do smarter once */
if(is->mode==RE_RAY_MIRROR)
return !(vlr->mat->mode & MA_ONLYCAST);
else
return (is->lay & obi->lay);
}
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
tag unused function args in raytrace code.
2012-01-14 10:08:47 +00:00
MALWAYS_INLINE int vlr_check_intersect_solid(Isect *UNUSED(is), ObjectInstanceRen* UNUSED(obi), VlakRen *vlr)
svn merge -r 22571:22800 https://svn.blender.org/svnroot/bf-blender/trunk/blender svn merge -r 22800:23207 https://svn.blender.org/svnroot/bf-blender/trunk/blender Merged volumetric with new raytrace code (it compiles and rendered volume-cube.blend withouth problems) Part1: source/blender
2009-09-15 15:15:43 +00:00
{
/* solid material types only */
if (vlr->mat->material_type == MA_TYPE_SURFACE)
return 1;
else
return 0;
}
tag unused function args in raytrace code.
2012-01-14 10:08:47 +00:00
MALWAYS_INLINE int vlr_check_bake(Isect *is, ObjectInstanceRen* obi, VlakRen *UNUSED(vlr))
* Assign weight from bones in weight paint mode now respects paint face mask, also avoid making vertex groups if they will not be filled. * Add back image pin option in image editor header. * Fix deep shadow not respecting Cast Buffer Shadows option. * Tangent space normal map baking should work again now. * Fix a problem with particle duplis, due to own bugfix for #20350, the problem for that seems to be in dupliverts, not particles. * Fix external multires data link getting lost on exiting editmode. (commits 27776,27777,27830,27840,27841,27862 by Brecht from render25 branch)
2010-03-30 12:01:17 +00:00
{
return (obi->obr->ob != is->userdata);
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* Ray Triangle/Quad Intersection */
* converted raytrace visibility test on meshlaplacian.c to new raytrace API I need test scenes and test instructions to make sure this is ok, since i have no idea how to test this feature.
2009-09-17 12:56:16 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
MALWAYS_INLINE int isec_tri_quad(float start[3], float dir[3], RayFace *face, float uv[2], float *lambda)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
float co1[3], co2[3], co3[3], co4[3];
float t0[3], t1[3], x[3], r[3], m[3], u, v, divdet, det1, l;
int quad;
* Assign weight from bones in weight paint mode now respects paint face mask, also avoid making vertex groups if they will not be filled. * Add back image pin option in image editor header. * Fix deep shadow not respecting Cast Buffer Shadows option. * Tangent space normal map baking should work again now. * Fix a problem with particle duplis, due to own bugfix for #20350, the problem for that seems to be in dupliverts, not particles. * Fix external multires data link getting lost on exiting editmode. (commits 27776,27777,27830,27840,27841,27862 by Brecht from render25 branch)
2010-03-30 12:01:17 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
quad= RE_rayface_isQuad(face);
*Updated UI options and added UI options to: control whether instances are used or not control whether vertexs are stored localy or not *Removed unsused code
2009-10-01 18:30:59 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
copy_v3_v3(co1, face->v1);
copy_v3_v3(co2, face->v2);
copy_v3_v3(co3, face->v3);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
copy_v3_v3(r, dir);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* intersect triangle */
sub_v3_v3v3(t0, co3, co2);
sub_v3_v3v3(t1, co3, co1);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
cross_v3_v3v3(x, r, t1);
divdet= dot_v3v3(t0, x);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
sub_v3_v3v3(m, start, co3);
det1= dot_v3v3(m, x);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(divdet != 0.0f) {
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
divdet= 1.0f/divdet;
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
v= det1*divdet;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(v < RE_RAYTRACE_EPSILON && v > -(1.0f+RE_RAYTRACE_EPSILON)) {
float cros[3];
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
cross_v3_v3v3(cros, m, t0);
u= divdet*dot_v3v3(cros, r);
if(u < RE_RAYTRACE_EPSILON && (v + u) > -(1.0f+RE_RAYTRACE_EPSILON)) {
l= divdet*dot_v3v3(cros, t1);
/* check if intersection is within ray length */
if(l > -RE_RAYTRACE_EPSILON && l < *lambda) {
uv[0]= u;
uv[1]= v;
*lambda= l;
return 1;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
}
}
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* intersect second triangle in quad */
if(quad) {
copy_v3_v3(co4, face->v4);
sub_v3_v3v3(t0, co3, co4);
divdet= dot_v3v3(t0, x);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(divdet != 0.0f) {
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
divdet= 1.0f/divdet;
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
v = det1*divdet;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(v < RE_RAYTRACE_EPSILON && v > -(1.0f+RE_RAYTRACE_EPSILON)) {
float cros[3];
cross_v3_v3v3(cros, m, t0);
u= divdet*dot_v3v3(cros, r);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(u < RE_RAYTRACE_EPSILON && (v + u) > -(1.0f+RE_RAYTRACE_EPSILON)) {
l= divdet*dot_v3v3(cros, t1);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(l >- RE_RAYTRACE_EPSILON && l < *lambda) {
uv[0]= u;
uv[1]= -(1.0f + v + u);
*lambda= l;
return 2;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
}
}
}
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
return 0;
}
/* Simpler yes/no Ray Triangle/Quad Intersection */
MALWAYS_INLINE int isec_tri_quad_neighbour(float start[3], float dir[3], RayFace *face)
{
float co1[3], co2[3], co3[3], co4[3];
float t0[3], t1[3], x[3], r[3], m[3], u, v, divdet, det1;
int quad;
quad= RE_rayface_isQuad(face);
copy_v3_v3(co1, face->v1);
copy_v3_v3(co2, face->v2);
copy_v3_v3(co3, face->v3);
negate_v3_v3(r, dir); /* note, different than above function */
/* intersect triangle */
sub_v3_v3v3(t0, co3, co2);
sub_v3_v3v3(t1, co3, co1);
cross_v3_v3v3(x, r, t1);
divdet= dot_v3v3(t0, x);
sub_v3_v3v3(m, start, co3);
det1= dot_v3v3(m, x);
if(divdet != 0.0f) {
divdet= 1.0f/divdet;
v= det1*divdet;
if(v < RE_RAYTRACE_EPSILON && v > -(1.0f+RE_RAYTRACE_EPSILON)) {
float cros[3];
cross_v3_v3v3(cros, m, t0);
u= divdet*dot_v3v3(cros, r);
if(u < RE_RAYTRACE_EPSILON && (v + u) > -(1.0f+RE_RAYTRACE_EPSILON))
return 1;
}
}
/* intersect second triangle in quad */
if(quad) {
copy_v3_v3(co4, face->v4);
sub_v3_v3v3(t0, co3, co4);
divdet= dot_v3v3(t0, x);
if(divdet != 0.0f) {
divdet= 1.0f/divdet;
v = det1*divdet;
if(v < RE_RAYTRACE_EPSILON && v > -(1.0f+RE_RAYTRACE_EPSILON)) {
float cros[3];
cross_v3_v3v3(cros, m, t0);
u= divdet*dot_v3v3(cros, r);
if(u < RE_RAYTRACE_EPSILON && (v + u) > -(1.0f+RE_RAYTRACE_EPSILON))
return 2;
}
}
}
return 0;
}
Spelling Cleanup
2012-03-01 12:20:18 +00:00
/* RayFace intersection with checks and neighbor verifaction included,
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
Isect is modified if the face is hit. */
MALWAYS_INLINE int intersect_rayface(RayObject *hit_obj, RayFace *face, Isect *is)
{
float dist, uv[2];
int ok= 0;
/* avoid self-intersection */
if(is->orig.ob == face->ob && is->orig.face == face->face)
return 0;
/* check if we should intersect this face */
if(is->check == RE_CHECK_VLR_RENDER)
{
if(vlr_check_intersect(is, (ObjectInstanceRen*)face->ob, (VlakRen*)face->face) == 0)
return 0;
}
else if(is->check == RE_CHECK_VLR_NON_SOLID_MATERIAL)
{
if(vlr_check_intersect(is, (ObjectInstanceRen*)face->ob, (VlakRen*)face->face) == 0)
return 0;
if(vlr_check_intersect_solid(is, (ObjectInstanceRen*)face->ob, (VlakRen*)face->face) == 0)
return 0;
}
else if(is->check == RE_CHECK_VLR_BAKE) {
if(vlr_check_bake(is, (ObjectInstanceRen*)face->ob, (VlakRen*)face->face) == 0)
return 0;
}
/* ray counter */
RE_RC_COUNT(is->raycounter->faces.test);
dist= is->dist;
ok= isec_tri_quad(is->start, is->dir, face, uv, &dist);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
if(ok) {
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* when a shadow ray leaves a face, it can be little outside the edges
Spelling Cleanup
2012-03-01 12:20:18 +00:00
of it, causing intersection to be detected in its neighbor face */
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
if(is->skip & RE_SKIP_VLR_NEIGHBOUR)
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(dist < 0.1f && is->orig.ob == face->ob)
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
{
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
VlakRen * a = (VlakRen*)is->orig.face;
VlakRen * b = (VlakRen*)face->face;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* so there's a shared edge or vertex, let's intersect ray with
face itself, if that's true we can safely return 1, otherwise
we assume the intersection is invalid, 0 */
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
if(a->v1==b->v1 || a->v2==b->v1 || a->v3==b->v1 || a->v4==b->v1
|| a->v1==b->v2 || a->v2==b->v2 || a->v3==b->v2 || a->v4==b->v2
|| a->v1==b->v3 || a->v2==b->v3 || a->v3==b->v3 || a->v4==b->v3
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
|| (b->v4 && (a->v1==b->v4 || a->v2==b->v4 || a->v3==b->v4 || a->v4==b->v4))) {
/* create RayFace from original face, transformed if necessary */
RayFace origface;
ObjectInstanceRen *ob= (ObjectInstanceRen*)is->orig.ob;
rayface_from_vlak(&origface, ob, (VlakRen*)is->orig.face);
if(!isec_tri_quad_neighbour(is->start, is->dir, &origface))
{
return 0;
}
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
}
}
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
*Added output of BB/primitves test/hits *"Added" SCE_PASS_RAYHITS to visually see each pixel primitive and BB tests (not-completed for UI) *Added runtime exchange of tree structure *Removed FLOAT_EPSILON from BLI_bvhkdop BB's
2009-07-06 13:27:40 +00:00
RE_RC_COUNT(is->raycounter->faces.hit);
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
* Multiply for panorama cameras * Some cases of struct name being set where it shouldnt have been. * Spelling: wich --> which * Copy and initialize uv modifier scale, remove unneeded enum. * Ability to pin any object into the context. * Update uv window while transforming (useful when used with UVProject modifier) * Patch by Wahooney, so new template's are internal text and dont get saved over by mistake. * Fix for https://bugzilla.redhat.com/show_bug.cgi?id=572186 Bug 572186 - [abrt] crash in blender-2.49b-5.fc12: Process /usr/bin/blender.bin was killed by signal 6 (SIGABRT). Original fix submitted by Jochen Schmitt. * [#21816] bpy.data.add_image has stopped working on Windows. moved to bpy.data.images.load(), missed this call. (commits 27726,27825,27828,27831,27832,27833,27834,27836,27837,27838,27839,27858 by Campbell from render25 branch)
2010-03-30 12:15:16 +00:00
is->isect= ok; // which half of the quad
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
is->dist= dist;
is->u= uv[0]; is->v= uv[1];
Added neighbour test on detected ray hit
2009-05-11 13:56:45 +00:00
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
is->hit.ob = face->ob;
is->hit.face = face->face;
Enabled #ifdefs to test LAST_HIT, LAST_HINT Disable last_hint as it only slow downs
2009-07-08 19:39:37 +00:00
#ifdef RT_USE_LAST_HIT
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
is->last_hit = hit_obj;
Enabled #ifdefs to test LAST_HIT, LAST_HINT Disable last_hint as it only slow downs
2009-07-08 19:39:37 +00:00
#endif
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
return 1;
}
return 0;
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* Intersection */
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
int RE_rayobject_raycast(RayObject *r, Isect *isec)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
int i;
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
*Added output of BB/primitves test/hits *"Added" SCE_PASS_RAYHITS to visually see each pixel primitive and BB tests (not-completed for UI) *Added runtime exchange of tree structure *Removed FLOAT_EPSILON from BLI_bvhkdop BB's
2009-07-06 13:27:40 +00:00
RE_RC_COUNT(isec->raycounter->raycast.test);
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* setup vars used on raycast */
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
for(i=0; i<3; i++)
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
isec->idot_axis[i] = 1.0f / isec->dir[i];
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
isec->bv_index[2*i] = isec->idot_axis[i] < 0.0 ? 1 : 0;
isec->bv_index[2*i+1] = 1 - isec->bv_index[2*i];
isec->bv_index[2*i] = i+3*isec->bv_index[2*i];
isec->bv_index[2*i+1] = i+3*isec->bv_index[2*i+1];
}
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
Enabled #ifdefs to test LAST_HIT, LAST_HINT Disable last_hint as it only slow downs
2009-07-08 19:39:37 +00:00
#ifdef RT_USE_LAST_HIT
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* last hit heuristic */
Hint support (only for first hiearchic level and shadow rays, lets see if this is useful) That means each shadow ray tests: 1st: last hit face 2nd: last hit object 3rd: tree of all objects
2009-07-08 15:34:41 +00:00
if(isec->mode==RE_RAY_SHADOW && isec->last_hit)
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
{
Hint support (only for first hiearchic level and shadow rays, lets see if this is useful) That means each shadow ray tests: 1st: last hit face 2nd: last hit object 3rd: tree of all objects
2009-07-08 15:34:41 +00:00
RE_RC_COUNT(isec->raycounter->rayshadow_last_hit.test);
if(RE_rayobject_intersect(isec->last_hit, isec))
{
RE_RC_COUNT(isec->raycounter->raycast.hit);
RE_RC_COUNT(isec->raycounter->rayshadow_last_hit.hit);
return 1;
}
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
}
Enabled #ifdefs to test LAST_HIT, LAST_HINT Disable last_hint as it only slow downs
2009-07-08 19:39:37 +00:00
#endif
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Hint support (only for first hiearchic level and shadow rays, lets see if this is useful) That means each shadow ray tests: 1st: last hit face 2nd: last hit object 3rd: tree of all objects
2009-07-08 15:34:41 +00:00
#ifdef RT_USE_HINT
isec->hit_hint = 0;
#endif
*Added rayobject_bvh A bvh structure to use on the raytracer
2009-07-01 11:27:43 +00:00
if(RE_rayobject_intersect(r, isec))
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
{
*Added output of BB/primitves test/hits *"Added" SCE_PASS_RAYHITS to visually see each pixel primitive and BB tests (not-completed for UI) *Added runtime exchange of tree structure *Removed FLOAT_EPSILON from BLI_bvhkdop BB's
2009-07-06 13:27:40 +00:00
RE_RC_COUNT(isec->raycounter->raycast.hit);
Hint support (only for first hiearchic level and shadow rays, lets see if this is useful) That means each shadow ray tests: 1st: last hit face 2nd: last hit object 3rd: tree of all objects
2009-07-08 15:34:41 +00:00
#ifdef RT_USE_HINT
isec->hint = isec->hit_hint;
#endif
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
return 1;
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
*make type of acceleration structure changeable at runtime *added some counter code (test/hits) for primitives and raycasts
2009-06-29 19:48:11 +00:00
return 0;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
Renamed exported functions from render to have the RE_ prefix RayObject_* => RE_rayobject_*
2009-05-13 01:56:03 +00:00
int RE_rayobject_intersect(RayObject *r, Isect *i)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
*Changed RayObject_ calls to RE_rayobject to keep consistency on calls *Moved part of counters code to a separated file (rayobject_raycounter.c)
2009-08-25 20:26:50 +00:00
if(RE_rayobject_isRayFace(r))
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
return intersect_rayface(r, (RayFace*) RE_rayobject_align(r), i);
}
else if(RE_rayobject_isVlakPrimitive(r))
{
//TODO optimize (useless copy to RayFace to avoid duplicate code)
VlakPrimitive *face = (VlakPrimitive*) RE_rayobject_align(r);
RayFace nface;
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
rayface_from_vlak(&nface, face->ob, face->face);
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
return intersect_rayface(r, &nface, i);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
*Changed RayObject_ calls to RE_rayobject to keep consistency on calls *Moved part of counters code to a separated file (rayobject_raycounter.c)
2009-08-25 20:26:50 +00:00
else if(RE_rayobject_isRayAPI(r))
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
r = RE_rayobject_align(r);
return r->api->raycast(r, i);
}
else {
assert(0);
whitespace cleanup
2011-09-25 12:31:21 +00:00
return 0;
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* Building */
Renamed exported functions from render to have the RE_ prefix RayObject_* => RE_rayobject_*
2009-05-13 01:56:03 +00:00
void RE_rayobject_add(RayObject *r, RayObject *o)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
r = RE_rayobject_align(r);
return r->api->add(r, o);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
Renamed exported functions from render to have the RE_ prefix RayObject_* => RE_rayobject_*
2009-05-13 01:56:03 +00:00
void RE_rayobject_done(RayObject *r)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
r = RE_rayobject_align(r);
r->api->done(r);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
Renamed exported functions from render to have the RE_ prefix RayObject_* => RE_rayobject_*
2009-05-13 01:56:03 +00:00
void RE_rayobject_free(RayObject *r)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
r = RE_rayobject_align(r);
r->api->free(r);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
float RE_rayobject_cost(RayObject *r)
{
if(RE_rayobject_isRayFace(r) || RE_rayobject_isVlakPrimitive(r))
{
return 1.0f;
}
else if(RE_rayobject_isRayAPI(r))
{
r = RE_rayobject_align(r);
return r->api->cost(r);
}
else {
assert(0);
return 1.0f;
}
}
/* Bounding Boxes */
Renamed exported functions from render to have the RE_ prefix RayObject_* => RE_rayobject_*
2009-05-13 01:56:03 +00:00
void RE_rayobject_merge_bb(RayObject *r, float *min, float *max)
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
*Changed RayObject_ calls to RE_rayobject to keep consistency on calls *Moved part of counters code to a separated file (rayobject_raycounter.c)
2009-08-25 20:26:50 +00:00
if(RE_rayobject_isRayFace(r))
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
{
*Changed RayObject_ calls to RE_rayobject to keep consistency on calls *Moved part of counters code to a separated file (rayobject_raycounter.c)
2009-08-25 20:26:50 +00:00
RayFace *face = (RayFace*) RE_rayobject_align(r);
#define to store coordinates at VlakRen and not at VlakFace adds some additional cost retrieving coords during ray-primitive tests, but reduces some memory usage (4*3floats per face)
2009-08-05 15:50:57 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
DO_MINMAX(face->v1, min, max);
DO_MINMAX(face->v2, min, max);
DO_MINMAX(face->v3, min, max);
if(RE_rayface_isQuad(face)) DO_MINMAX(face->v4, min, max);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
}
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
else if(RE_rayobject_isVlakPrimitive(r))
{
VlakPrimitive *face = (VlakPrimitive*) RE_rayobject_align(r);
Merge various small changes from render branch: * Division by zero fix for TNT SVD code. * Sound fix, in case ffmpeg decode fails, don't use the samples. * Fix for incorrect bounds of transformed objects in new raytracing code. * Gave memory arena's a name used for allocations for easier memory usage debugging. * Dupligroup no_draw option was using layers but not restrict view/render setting. (not a bugfix exactly but would do display list context switching while drawing for no reason). * Fix objects instanced on hair particles not giving consistent results when the object is transformed. * New math functions: madd_v4_v4fl, len_squared_v3v3, interp_v4_v4v4v4, mul_v4_m4v4, SH and form factor functions, box_minmax_bounds_m4. * mul_m4_m4m4 and mul_m3_m3m3 now accept the same pointers for multiple arguments. * endjob callback for WM jobs system. * Geometry node uv/color layer now has search list/autocomplete. * Various small buildsystem tweaks, not strictly needed yet in trunk.
2010-04-15 10:28:32 +00:00
RayFace nface;
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
rayface_from_vlak(&nface, face->ob, face->face);
*new generic raytrace API *Adapted octree to a more generic raytrace API *ray shadow works (other untested stuff disabled atm) On the scene tested the user-cpu time got from 1:24 to 1:19/20 probably because of removed callbacks or sligtly diferente memory usage
2009-05-10 21:02:58 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
DO_MINMAX(nface.v1, min, max);
DO_MINMAX(nface.v2, min, max);
DO_MINMAX(nface.v3, min, max);
if(RE_rayface_isQuad(&nface)) DO_MINMAX(nface.v4, min, max);
*Added support for variable cost per RayObject Suposedly usefull for creating trees of objects (where objects have very diferent size-NumFaces and shape-BB) Altought the implemented costs maybe not be very correct (for now), as i didnt cared about following a specific "corrected" model
2009-07-08 20:04:40 +00:00
}
*Changed RayObject_ calls to RE_rayobject to keep consistency on calls *Moved part of counters code to a separated file (rayobject_raycounter.c)
2009-08-25 20:26:50 +00:00
else if(RE_rayobject_isRayAPI(r))
*Added support for variable cost per RayObject Suposedly usefull for creating trees of objects (where objects have very diferent size-NumFaces and shape-BB) Altought the implemented costs maybe not be very correct (for now), as i didnt cared about following a specific "corrected" model
2009-07-08 20:04:40 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
r = RE_rayobject_align(r);
r->api->bb(r, min, max);
*Added support for variable cost per RayObject Suposedly usefull for creating trees of objects (where objects have very diferent size-NumFaces and shape-BB) Altought the implemented costs maybe not be very correct (for now), as i didnt cared about following a specific "corrected" model
2009-07-08 20:04:40 +00:00
}
warning fixes.
2010-10-23 16:03:31 +00:00
else
assert(0);
*Added support for variable cost per RayObject Suposedly usefull for creating trees of objects (where objects have very diferent size-NumFaces and shape-BB) Altought the implemented costs maybe not be very correct (for now), as i didnt cared about following a specific "corrected" model
2009-07-08 20:04:40 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* Hints */
*Added support to "BB hints" (which works like a BB version of LCTS - longest common transversing subtree) It creates a tree cut after knowing that a given point will pass on a BB. This tree cut is used to accelarate the rays casted from a given BB, eliminating unnecessary BB tests from root till the tree cut.
2009-07-15 17:38:00 +00:00
void RE_rayobject_hint_bb(RayObject *r, RayHint *hint, float *min, float *max)
{
*Added VlakPrimitive (this rayobject rimitive only stores ObjectRenderInstance and VlakRen pointers) - it difers from RayFace that localy stored the vertex coordinates. - basicaly this reduces memory usage
2009-09-24 22:55:57 +00:00
if(RE_rayobject_isRayFace(r) || RE_rayobject_isVlakPrimitive(r))
*Added support to "BB hints" (which works like a BB version of LCTS - longest common transversing subtree) It creates a tree cut after knowing that a given point will pass on a BB. This tree cut is used to accelarate the rays casted from a given BB, eliminating unnecessary BB tests from root till the tree cut.
2009-07-15 17:38:00 +00:00
{
return;
}
*Changed RayObject_ calls to RE_rayobject to keep consistency on calls *Moved part of counters code to a separated file (rayobject_raycounter.c)
2009-08-25 20:26:50 +00:00
else if(RE_rayobject_isRayAPI(r))
*Added support to "BB hints" (which works like a BB version of LCTS - longest common transversing subtree) It creates a tree cut after knowing that a given point will pass on a BB. This tree cut is used to accelarate the rays casted from a given BB, eliminating unnecessary BB tests from root till the tree cut.
2009-07-15 17:38:00 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
r = RE_rayobject_align(r);
return r->api->hint_bb(r, hint, min, max);
*Added support to "BB hints" (which works like a BB version of LCTS - longest common transversing subtree) It creates a tree cut after knowing that a given point will pass on a BB. This tree cut is used to accelarate the rays casted from a given BB, eliminating unnecessary BB tests from root till the tree cut.
2009-07-15 17:38:00 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
else
assert(0);
*Added support to "BB hints" (which works like a BB version of LCTS - longest common transversing subtree) It creates a tree cut after knowing that a given point will pass on a BB. This tree cut is used to accelarate the rays casted from a given BB, eliminating unnecessary BB tests from root till the tree cut.
2009-07-15 17:38:00 +00:00
}
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
/* RayObjectControl */
Added some test_break during the build process. (Maybe later this should be done with some thread_cancel function instead of doing variable/callbacks tests)
2009-10-04 16:56:00 +00:00
int RE_rayobjectcontrol_test_break(RayObjectControl *control)
{
if(control->test_break)
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
return control->test_break(control->data);
Added some test_break during the build process. (Maybe later this should be done with some thread_cancel function instead of doing variable/callbacks tests)
2009-10-04 16:56:00 +00:00
return 0;
}
Fixed #19571 (was crashing when building a raytree from a scene withouth any geometry)
2009-10-08 11:24:06 +00:00
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
void RE_rayobject_set_control(RayObject *r, void *data, RE_rayobjectcontrol_test_break_callback test_break)
Fixed #19571 (was crashing when building a raytree from a scene withouth any geometry)
2009-10-08 11:24:06 +00:00
{
Raytrace modifications from the Render Branch. These should not have any effect on render results, except in some cases with you have overlapping faces, where the noise seems to be slightly reduced. There are some performance improvements, for simple scenes I wouldn't expect more than 5-10% to be cut off the render time, for sintel scenes we got about 50% on average, that's with millions of polygons on intel quad cores. This because memory access / cache misses were the main bottleneck for those scenes, and the optimizations improve that. Interal changes: * Remove RE_raytrace.h, raytracer is now only used by render engine again. * Split non-public parts rayobject.h into rayobject_internal.h, hopefully makes it clearer how the API is used. * Added rayintersection.h to contain some of the stuff from RE_raytrace.h * Change Isect.vec/labda to Isect.dir/dist, previously vec was sometimes normalized and sometimes not, confusing... now dir is always normalized and dist contains the distance. * Change VECCOPY and similar to BLI_math functions. * Force inlining of auxiliary functions for ray-triangle/quad intersection, helps a few percentages. * Reorganize svbvh code so all the traversal functions are in one file * Don't do test for root so that push_childs can be inlined * Make shadow a template parameter so it doesn't need to be runtime checked * Optimization in raytree building, was computing bounding boxes more often than necessary. * Leave out logf() factor in SAH, makes tree build quicker with no noticeable influence on raytracing on performance? * Set max childs to 4, simplifies traversal code a bit, but also seems to help slightly in general. * Store child pointers and child bb just as fixed arrays of size 4 in nodes, nearly all nodes have this many children, so overall it actually reduces memory usage a bit and avoids a pointer indirection.
2011-02-05 13:41:29 +00:00
if(RE_rayobject_isRayAPI(r))
{
r = RE_rayobject_align(r);
r->control.data = data;
r->control.test_break = test_break;
}
Fixed #19571 (was crashing when building a raytree from a scene withouth any geometry)
2009-10-08 11:24:06 +00:00
}
Copy Permalink