blender/source/gameengine/Rasterizer/RAS_BucketManager.cpp

373 lines
10 KiB
C++
Raw Normal View History

/*
* ***** BEGIN GPL LICENSE BLOCK *****
2002-10-12 11:37:38 +00:00
*
* 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.
2002-10-12 11:37:38 +00:00
*
* 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,
2010-02-12 13:34:04 +00:00
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
2002-10-12 11:37:38 +00:00
*
* The Original Code is Copyright (C) 2001-2002 by NaN Holding BV.
* All rights reserved.
*
* The Original Code is: all of this file.
*
* Contributor(s): none yet.
*
* ***** END GPL LICENSE BLOCK *****
2002-10-12 11:37:38 +00:00
*/
2011-02-25 13:38:24 +00:00
/** \file gameengine/Rasterizer/RAS_BucketManager.cpp
* \ingroup bgerast
*/
#if defined(WIN32) && !defined(FREE_WINDOWS)
2002-10-12 11:37:38 +00:00
// don't show these anoying STL warnings
#pragma warning (disable:4786)
#endif
#include "CTR_Map.h"
2002-10-12 11:37:38 +00:00
#include "RAS_MaterialBucket.h"
#include "STR_HashedString.h"
#include "RAS_MeshObject.h"
#include "RAS_IRasterizer.h"
#include "RAS_IRenderTools.h"
#include "RAS_BucketManager.h"
#include <algorithm>
#include <set>
/* sorting */
struct RAS_BucketManager::sortedmeshslot
2002-10-12 11:37:38 +00:00
{
public:
MT_Scalar m_z; /* depth */
RAS_MeshSlot *m_ms; /* mesh slot */
RAS_MaterialBucket *m_bucket; /* buck mesh slot came from */
2002-10-12 11:37:38 +00:00
sortedmeshslot() {}
2002-10-12 11:37:38 +00:00
void set(RAS_MeshSlot *ms, RAS_MaterialBucket *bucket, const MT_Vector3& pnorm)
{
// would be good to use the actual bounding box center instead
MT_Point3 pos(ms->m_OpenGLMatrix[12], ms->m_OpenGLMatrix[13], ms->m_OpenGLMatrix[14]);
2002-10-12 11:37:38 +00:00
m_z = MT_dot(pnorm, pos);
m_ms = ms;
m_bucket = bucket;
}
};
2002-10-12 11:37:38 +00:00
struct RAS_BucketManager::backtofront
{
bool operator()(const sortedmeshslot &a, const sortedmeshslot &b)
{
return (a.m_z < b.m_z) || (a.m_z == b.m_z && a.m_ms < b.m_ms);
}
};
struct RAS_BucketManager::fronttoback
{
bool operator()(const sortedmeshslot &a, const sortedmeshslot &b)
{
return (a.m_z > b.m_z) || (a.m_z == b.m_z && a.m_ms > b.m_ms);
}
};
/* bucket manager */
RAS_BucketManager::RAS_BucketManager()
{
}
RAS_BucketManager::~RAS_BucketManager()
{
BucketList::iterator it;
for (it = m_SolidBuckets.begin(); it != m_SolidBuckets.end(); it++)
delete (*it);
for (it = m_AlphaBuckets.begin(); it != m_AlphaBuckets.end(); it++)
delete(*it);
m_SolidBuckets.clear();
m_AlphaBuckets.clear();
}
void RAS_BucketManager::OrderBuckets(const MT_Transform& cameratrans, BucketList& buckets, vector<sortedmeshslot>& slots, bool alpha)
{
BucketList::iterator bit;
list<RAS_MeshSlot>::iterator mit;
size_t size = 0, i = 0;
/* Camera's near plane equation: pnorm.dot(point) + pval,
* but we leave out pval since it's constant anyway */
const MT_Vector3 pnorm(cameratrans.getBasis()[2]);
for (bit = buckets.begin(); bit != buckets.end(); ++bit)
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
{
SG_DList::iterator<RAS_MeshSlot> mit((*bit)->GetActiveMeshSlots());
for(mit.begin(); !mit.end(); ++mit)
size++;
}
slots.resize(size);
for (bit = buckets.begin(); bit != buckets.end(); ++bit)
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
{
RAS_MaterialBucket* bucket = *bit;
RAS_MeshSlot* ms;
// remove the mesh slot form the list, it culls them automatically for next frame
while((ms = bucket->GetNextActiveMeshSlot())) {
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
slots[i++].set(ms, bucket, pnorm);
}
}
if(alpha)
sort(slots.begin(), slots.end(), backtofront());
else
sort(slots.begin(), slots.end(), fronttoback());
}
void RAS_BucketManager::RenderAlphaBuckets(
const MT_Transform& cameratrans, RAS_IRasterizer* rasty, RAS_IRenderTools* rendertools)
{
vector<sortedmeshslot> slots;
vector<sortedmeshslot>::iterator sit;
// Having depth masks disabled/enabled gives different artifacts in
// case no sorting is done or is done inexact. For compatibility, we
// disable it.
rasty->SetDepthMask(RAS_IRasterizer::KX_DEPTHMASK_DISABLED);
OrderBuckets(cameratrans, m_AlphaBuckets, slots, true);
for(sit=slots.begin(); sit!=slots.end(); ++sit) {
rendertools->SetClientObject(rasty, sit->m_ms->m_clientObj);
while(sit->m_bucket->ActivateMaterial(cameratrans, rasty, rendertools))
sit->m_bucket->RenderMeshSlot(cameratrans, rasty, rendertools, *(sit->m_ms));
// make this mesh slot culled automatically for next frame
// it will be culled out by frustrum culling
sit->m_ms->SetCulled(true);
}
rasty->SetDepthMask(RAS_IRasterizer::KX_DEPTHMASK_ENABLED);
}
void RAS_BucketManager::RenderSolidBuckets(
const MT_Transform& cameratrans, RAS_IRasterizer* rasty, RAS_IRenderTools* rendertools)
{
BucketList::iterator bit;
rasty->SetDepthMask(RAS_IRasterizer::KX_DEPTHMASK_ENABLED);
for (bit = m_SolidBuckets.begin(); bit != m_SolidBuckets.end(); ++bit) {
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
#if 1
RAS_MaterialBucket* bucket = *bit;
RAS_MeshSlot* ms;
// remove the mesh slot form the list, it culls them automatically for next frame
while((ms = bucket->GetNextActiveMeshSlot()))
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
{
rendertools->SetClientObject(rasty, ms->m_clientObj);
while (bucket->ActivateMaterial(cameratrans, rasty, rendertools))
bucket->RenderMeshSlot(cameratrans, rasty, rendertools, *ms);
// make this mesh slot culled automatically for next frame
// it will be culled out by frustrum culling
ms->SetCulled(true);
}
#else
list<RAS_MeshSlot>::iterator mit;
for (mit = (*bit)->msBegin(); mit != (*bit)->msEnd(); ++mit) {
if (mit->IsCulled())
continue;
rendertools->SetClientObject(rasty, mit->m_clientObj);
while ((*bit)->ActivateMaterial(cameratrans, rasty, rendertools))
(*bit)->RenderMeshSlot(cameratrans, rasty, rendertools, *mit);
// make this mesh slot culled automatically for next frame
// it will be culled out by frustrum culling
mit->SetCulled(true);
}
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
#endif
}
/* this code draws meshes order front-to-back instead to reduce overdraw.
* it turned out slower due to much material state switching, a more clever
* algorithm might do better. */
#if 0
vector<sortedmeshslot> slots;
vector<sortedmeshslot>::iterator sit;
OrderBuckets(cameratrans, m_SolidBuckets, slots, false);
for(sit=slots.begin(); sit!=slots.end(); ++sit) {
rendertools->SetClientObject(rasty, sit->m_ms->m_clientObj);
while(sit->m_bucket->ActivateMaterial(cameratrans, rasty, rendertools))
sit->m_bucket->RenderMeshSlot(cameratrans, rasty, rendertools, *(sit->m_ms));
}
#endif
}
2002-10-12 11:37:38 +00:00
void RAS_BucketManager::Renderbuckets(
const MT_Transform& cameratrans, RAS_IRasterizer* rasty, RAS_IRenderTools* rendertools)
{
/* beginning each frame, clear (texture/material) caching information */
2002-10-12 11:37:38 +00:00
rasty->ClearCachingInfo();
RenderSolidBuckets(cameratrans, rasty, rendertools);
RenderAlphaBuckets(cameratrans, rasty, rendertools);
rendertools->SetClientObject(rasty, NULL);
2002-10-12 11:37:38 +00:00
}
RAS_MaterialBucket* RAS_BucketManager::FindBucket(RAS_IPolyMaterial * material, bool &bucketCreated)
2002-10-12 11:37:38 +00:00
{
BucketList::iterator it;
bucketCreated = false;
for (it = m_SolidBuckets.begin(); it != m_SolidBuckets.end(); it++)
if (*(*it)->GetPolyMaterial() == *material)
return *it;
for (it = m_AlphaBuckets.begin(); it != m_AlphaBuckets.end(); it++)
if (*(*it)->GetPolyMaterial() == *material)
return *it;
RAS_MaterialBucket *bucket = new RAS_MaterialBucket(material);
bucketCreated = true;
if (bucket->IsAlpha())
m_AlphaBuckets.push_back(bucket);
else
m_SolidBuckets.push_back(bucket);
2002-10-12 11:37:38 +00:00
return bucket;
}
void RAS_BucketManager::OptimizeBuckets(MT_Scalar distance)
2002-10-12 11:37:38 +00:00
{
BucketList::iterator bit;
2002-10-12 11:37:38 +00:00
distance = 10.0;
for (bit = m_SolidBuckets.begin(); bit != m_SolidBuckets.end(); ++bit)
(*bit)->Optimize(distance);
for (bit = m_AlphaBuckets.begin(); bit != m_AlphaBuckets.end(); ++bit)
(*bit)->Optimize(distance);
2002-10-12 11:37:38 +00:00
}
void RAS_BucketManager::ReleaseDisplayLists(RAS_IPolyMaterial *mat)
{
BucketList::iterator bit;
list<RAS_MeshSlot>::iterator mit;
for (bit = m_SolidBuckets.begin(); bit != m_SolidBuckets.end(); ++bit) {
if (mat == NULL || (mat == (*bit)->GetPolyMaterial())) {
for (mit = (*bit)->msBegin(); mit != (*bit)->msEnd(); ++mit) {
if(mit->m_DisplayList) {
mit->m_DisplayList->Release();
mit->m_DisplayList = NULL;
}
}
}
}
for (bit = m_AlphaBuckets.begin(); bit != m_AlphaBuckets.end(); ++bit) {
if (mat == NULL || (mat == (*bit)->GetPolyMaterial())) {
for (mit = (*bit)->msBegin(); mit != (*bit)->msEnd(); ++mit) {
if(mit->m_DisplayList) {
mit->m_DisplayList->Release();
mit->m_DisplayList = NULL;
}
}
}
}
}
BGE performance, 3rd round: culling and rasterizer. This commit extend the technique of dynamic linked list to the mesh slots so as to eliminate dumb scan or map lookup. It provides massive performance improvement in the culling and in the rasterizer when the majority of objects are static. Other improvements: - Compute the opengl matrix only for objects that are visible. - Simplify hash function for GEN_HasedPtr - Scan light list instead of general object list to render shadows - Remove redundant opengl calls to set specularity, shinyness and diffuse between each mesh slots. - Cache GPU material to avoid frequent call to GPU_material_from_blender - Only set once the fixed elements of mesh slot - Use more inline function The following table shows the performance increase between 2.48, 1st round and this round of improvement. The test was done with a scene containing 40000 objects, of which 1000 are in the view frustrum approximately. The object are simple textured cube to make sure the GPU is not the bottleneck. As some of the rasterizer processing time has moved under culling, I present the sum of scenegraph(includes culling)+rasterizer time Scenegraph+rasterizer(ms) 2.48 1st round 3rd round All objects static, 323.0 86.0 7.2 all visible, 1000 in the view frustrum All objects static, 219.0 49.7 N/A(*) all invisible. All objects moving, 323.0 105.6 34.7 all visible, 1000 in the view frustrum Scene destruction 40min 40min 4s (*) : this time is not representative because the frame rate was at 60fps. In that case, the GPU holds down the GE by frame sync. By design, the overhead of the rasterizer is 0 when the the objects are invisible. This table shows a global speed up between 9x and 45x compared to 2.48a for scenegraph, culling and rasterizer overhead. The speed up goes much higher when objects are invisible. An additional 2-4x speed up is possible in the scenegraph by upgrading the Moto library to use Eigen2 BLAS library instead of C++ classes but the scenegraph is already so fast that it is not a priority right now. Next speed up in logic: many things to do there...
2009-05-07 09:13:01 +00:00
void RAS_BucketManager::ReleaseMaterials(RAS_IPolyMaterial * mat)
{
BucketList::iterator bit;
list<RAS_MeshSlot>::iterator mit;
for (bit = m_SolidBuckets.begin(); bit != m_SolidBuckets.end(); ++bit) {
if (mat == NULL || (mat == (*bit)->GetPolyMaterial())) {
(*bit)->GetPolyMaterial()->ReleaseMaterial();
}
}
for (bit = m_AlphaBuckets.begin(); bit != m_AlphaBuckets.end(); ++bit) {
if (mat == NULL || (mat == (*bit)->GetPolyMaterial())) {
(*bit)->GetPolyMaterial()->ReleaseMaterial();
}
}
}
/* frees the bucket, only used when freeing scenes */
void RAS_BucketManager::RemoveMaterial(RAS_IPolyMaterial * mat)
{
BucketList::iterator bit, bitp;
list<RAS_MeshSlot>::iterator mit;
int i;
for(i=0; i<m_SolidBuckets.size(); i++) {
RAS_MaterialBucket *bucket = m_SolidBuckets[i];
if (mat == bucket->GetPolyMaterial()) {
m_SolidBuckets.erase(m_SolidBuckets.begin()+i);
delete bucket;
i--;
}
}
for(int i=0; i<m_AlphaBuckets.size(); i++) {
RAS_MaterialBucket *bucket = m_AlphaBuckets[i];
if (mat == bucket->GetPolyMaterial()) {
m_AlphaBuckets.erase(m_AlphaBuckets.begin()+i);
delete bucket;
i--;
}
}
}
//#include <stdio.h>
void RAS_BucketManager::MergeBucketManager(RAS_BucketManager *other, SCA_IScene *scene)
{
/* concatinate lists */
// printf("BEFORE %d %d\n", GetSolidBuckets().size(), GetAlphaBuckets().size());
BucketList::iterator it;
for (it = other->GetSolidBuckets().begin(); it != other->GetSolidBuckets().end(); ++it)
(*it)->GetPolyMaterial()->Replace_IScene(scene);
GetSolidBuckets().insert( GetSolidBuckets().end(), other->GetSolidBuckets().begin(), other->GetSolidBuckets().end() );
other->GetSolidBuckets().clear();
for (it = other->GetAlphaBuckets().begin(); it != other->GetAlphaBuckets().end(); ++it)
(*it)->GetPolyMaterial()->Replace_IScene(scene);
GetAlphaBuckets().insert( GetAlphaBuckets().end(), other->GetAlphaBuckets().begin(), other->GetAlphaBuckets().end() );
other->GetAlphaBuckets().clear();
//printf("AFTER %d %d\n", GetSolidBuckets().size(), GetAlphaBuckets().size());
}