blender/source/gameengine/Rasterizer/RAS_MaterialBucket.cpp
Benoit Bolsee 42557f90bd 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

645 lines
16 KiB
C++

/**
* $Id$
* ***** 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,
* Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*
* 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 *****
*/
#include "RAS_MaterialBucket.h"
#ifdef WIN32
#pragma warning (disable:4786)
#include <windows.h>
#endif // WIN32
#include "RAS_Polygon.h"
#include "RAS_TexVert.h"
#include "RAS_IRasterizer.h"
#include "RAS_IRenderTools.h"
#include "RAS_MeshObject.h"
#include "RAS_Deformer.h" // __NLA
/* mesh slot */
RAS_MeshSlot::RAS_MeshSlot() : SG_QList()
{
m_clientObj = NULL;
m_pDeformer = NULL;
m_OpenGLMatrix = NULL;
m_mesh = NULL;
m_bucket = NULL;
m_bVisible = false;
m_bCulled = true;
m_bObjectColor = false;
m_RGBAcolor = MT_Vector4(0.0, 0.0, 0.0, 0.0);
m_DisplayList = NULL;
m_bDisplayList = true;
m_joinSlot = NULL;
m_pDerivedMesh = NULL;
}
RAS_MeshSlot::~RAS_MeshSlot()
{
RAS_DisplayArrayList::iterator it;
#ifdef USE_SPLIT
Split(true);
while(m_joinedSlots.size())
m_joinedSlots.front()->Split(true);
#endif
for(it=m_displayArrays.begin(); it!=m_displayArrays.end(); it++) {
(*it)->m_users--;
if((*it)->m_users == 0)
delete *it;
}
if (m_DisplayList) {
m_DisplayList->Release();
m_DisplayList = NULL;
}
}
RAS_MeshSlot::RAS_MeshSlot(const RAS_MeshSlot& slot) : SG_QList()
{
RAS_DisplayArrayList::iterator it;
m_clientObj = NULL;
m_pDeformer = NULL;
m_pDerivedMesh = NULL;
m_OpenGLMatrix = NULL;
m_mesh = slot.m_mesh;
m_bucket = slot.m_bucket;
m_bVisible = slot.m_bVisible;
m_bCulled = slot.m_bCulled;
m_bObjectColor = slot.m_bObjectColor;
m_RGBAcolor = slot.m_RGBAcolor;
m_DisplayList = NULL;
m_bDisplayList = slot.m_bDisplayList;
m_joinSlot = NULL;
m_currentArray = slot.m_currentArray;
m_displayArrays = slot.m_displayArrays;
m_joinedSlots = slot.m_joinedSlots;
m_startarray = slot.m_startarray;
m_startvertex = slot.m_startvertex;
m_startindex = slot.m_startindex;
m_endarray = slot.m_endarray;
m_endvertex = slot.m_endvertex;
m_endindex = slot.m_endindex;
for(it=m_displayArrays.begin(); it!=m_displayArrays.end(); it++) {
// don't copy display arrays for now because it breaks python
// access to vertices, but we'll need a solution if we want to
// join display arrays for reducing draw calls.
//*it = new RAS_DisplayArray(**it);
//(*it)->m_users = 1;
(*it)->m_users++;
}
}
void RAS_MeshSlot::init(RAS_MaterialBucket *bucket, int numverts)
{
m_bucket = bucket;
SetDisplayArray(numverts);
m_startarray = 0;
m_startvertex = 0;
m_startindex = 0;
m_endarray = 0;
m_endvertex = 0;
m_endindex = 0;
}
void RAS_MeshSlot::begin(RAS_MeshSlot::iterator& it)
{
int startvertex, endvertex;
int startindex, endindex;
it.array = (m_displayArrays.size() > 0)? m_displayArrays[m_startarray]: NULL;
if(it.array == NULL || it.array->m_index.size() == 0 || it.array->m_vertex.size() == 0) {
it.array = NULL;
it.vertex = NULL;
it.index = NULL;
it.startvertex = 0;
it.endvertex = 0;
it.totindex = 0;
}
else {
startvertex = m_startvertex;
endvertex = (m_startarray == m_endarray)? m_endvertex: it.array->m_vertex.size();
startindex = m_startindex;
endindex = (m_startarray == m_endarray)? m_endindex: it.array->m_index.size();
it.vertex = &it.array->m_vertex[0];
it.index = &it.array->m_index[startindex];
it.startvertex = startvertex;
it.endvertex = endvertex;
it.totindex = endindex-startindex;
it.arraynum = m_startarray;
}
}
void RAS_MeshSlot::next(RAS_MeshSlot::iterator& it)
{
int startvertex, endvertex;
int startindex, endindex;
if(it.arraynum == (size_t)m_endarray) {
it.array = NULL;
it.vertex = NULL;
it.index = NULL;
it.startvertex = 0;
it.endvertex = 0;
it.totindex = 0;
}
else {
it.arraynum++;
it.array = m_displayArrays[it.arraynum];
startindex = 0;
endindex = (it.arraynum == (size_t)m_endarray)? m_endindex: it.array->m_index.size();
startvertex = 0;
endvertex = (it.arraynum == (size_t)m_endarray)? m_endvertex: it.array->m_vertex.size();
it.vertex = &it.array->m_vertex[0];
it.index = &it.array->m_index[startindex];
it.startvertex = startvertex;
it.endvertex = endvertex;
it.totindex = endindex-startindex;
}
}
bool RAS_MeshSlot::end(RAS_MeshSlot::iterator& it)
{
return (it.array == NULL);
}
RAS_DisplayArray *RAS_MeshSlot::CurrentDisplayArray()
{
return m_currentArray;
}
void RAS_MeshSlot::SetDisplayArray(int numverts)
{
RAS_DisplayArrayList::iterator it;
RAS_DisplayArray *darray = NULL;
for(it=m_displayArrays.begin(); it!=m_displayArrays.end(); it++) {
darray = *it;
if(darray->m_type == numverts) {
if(darray->m_index.size()+numverts >= RAS_DisplayArray::BUCKET_MAX_INDEX)
darray = NULL;
else if(darray->m_vertex.size()+numverts >= RAS_DisplayArray::BUCKET_MAX_VERTEX)
darray = NULL;
else
break;
}
else
darray = NULL;
}
if(!darray) {
darray = new RAS_DisplayArray();
darray->m_users = 1;
if(numverts == 2) darray->m_type = RAS_DisplayArray::LINE;
else if(numverts == 3) darray->m_type = RAS_DisplayArray::TRIANGLE;
else darray->m_type = RAS_DisplayArray::QUAD;
m_displayArrays.push_back(darray);
if(numverts == 2)
darray->m_type = RAS_DisplayArray::LINE;
else if(numverts == 3)
darray->m_type = RAS_DisplayArray::TRIANGLE;
else if(numverts == 4)
darray->m_type = RAS_DisplayArray::QUAD;
m_endarray = m_displayArrays.size()-1;
m_endvertex = 0;
m_endindex = 0;
}
m_currentArray = darray;
}
void RAS_MeshSlot::AddPolygon(int numverts)
{
SetDisplayArray(numverts);
}
int RAS_MeshSlot::AddVertex(const RAS_TexVert& tv)
{
RAS_DisplayArray *darray;
int offset;
darray = m_currentArray;
darray->m_vertex.push_back(tv);
offset = darray->m_vertex.size()-1;
if(darray == m_displayArrays[m_endarray])
m_endvertex++;
return offset;
}
void RAS_MeshSlot::AddPolygonVertex(int offset)
{
RAS_DisplayArray *darray;
darray = m_currentArray;
darray->m_index.push_back(offset);
if(darray == m_displayArrays[m_endarray])
m_endindex++;
}
void RAS_MeshSlot::SetDeformer(RAS_Deformer* deformer)
{
if (deformer && m_pDeformer != deformer) {
// we create local copy of RAS_DisplayArray when we have a deformer:
// this way we can avoid conflict between the vertex cache of duplicates
RAS_DisplayArrayList::iterator it;
for(it=m_displayArrays.begin(); it!=m_displayArrays.end(); it++) {
if (deformer->UseVertexArray()) {
// the deformer makes use of vertex array, make sure we have our local copy
if ((*it)->m_users > 1) {
// only need to copy if there are other users
// note that this is the usual case as vertex arrays are held by the material base slot
RAS_DisplayArray *newarray = new RAS_DisplayArray(*(*it));
newarray->m_users = 1;
(*it)->m_users--;
*it = newarray;
}
} else {
// the deformer is not using vertex array (Modifier), release them
(*it)->m_users--;
if((*it)->m_users == 0)
delete *it;
}
}
if (!deformer->UseVertexArray()) {
m_displayArrays.clear();
m_startarray = 0;
m_startvertex = 0;
m_startindex = 0;
m_endarray = 0;
m_endvertex = 0;
m_endindex = 0;
}
}
m_pDeformer = deformer;
}
bool RAS_MeshSlot::Equals(RAS_MeshSlot *target)
{
if(!m_OpenGLMatrix || !target->m_OpenGLMatrix)
return false;
if(m_pDeformer || target->m_pDeformer)
return false;
if(m_bVisible != target->m_bVisible)
return false;
if(m_bObjectColor != target->m_bObjectColor)
return false;
if(m_bObjectColor && !(m_RGBAcolor == target->m_RGBAcolor))
return false;
return true;
}
bool RAS_MeshSlot::Join(RAS_MeshSlot *target, MT_Scalar distance)
{
RAS_DisplayArrayList::iterator it;
iterator mit;
size_t i;
// verify if we can join
if(m_joinSlot || m_joinedSlots.size() || target->m_joinSlot)
return false;
if(!Equals(target))
return false;
MT_Vector3 co(&m_OpenGLMatrix[12]);
MT_Vector3 targetco(&target->m_OpenGLMatrix[12]);
if((co - targetco).length() > distance)
return false;
MT_Matrix4x4 mat(m_OpenGLMatrix);
MT_Matrix4x4 targetmat(target->m_OpenGLMatrix);
targetmat.invert();
MT_Matrix4x4 transform = targetmat*mat;
// m_mesh, clientobj
m_joinSlot = target;
m_joinInvTransform = transform;
m_joinInvTransform.invert();
target->m_joinedSlots.push_back(this);
MT_Matrix4x4 ntransform = m_joinInvTransform.transposed();
ntransform[0][3]= ntransform[1][3]= ntransform[2][3]= 0.0f;
for(begin(mit); !end(mit); next(mit))
for(i=mit.startvertex; i<mit.endvertex; i++)
mit.vertex[i].Transform(transform, ntransform);
/* We know we'll need a list at least this big, reserve in advance */
target->m_displayArrays.reserve(target->m_displayArrays.size() + m_displayArrays.size());
for(it=m_displayArrays.begin(); it!=m_displayArrays.end(); it++) {
target->m_displayArrays.push_back(*it);
target->m_endarray++;
target->m_endvertex = target->m_displayArrays.back()->m_vertex.size();
target->m_endindex = target->m_displayArrays.back()->m_index.size();
}
if (m_DisplayList) {
m_DisplayList->Release();
m_DisplayList = NULL;
}
if (target->m_DisplayList) {
target->m_DisplayList->Release();
target->m_DisplayList = NULL;
}
return true;
#if 0
return false;
#endif
}
bool RAS_MeshSlot::Split(bool force)
{
list<RAS_MeshSlot*>::iterator jit;
RAS_MeshSlot *target = m_joinSlot;
RAS_DisplayArrayList::iterator it, jt;
iterator mit;
size_t i, found0 = 0, found1 = 0;
if(target && (force || !Equals(target))) {
m_joinSlot = NULL;
for(jit=target->m_joinedSlots.begin(); jit!=target->m_joinedSlots.end(); jit++) {
if(*jit == this) {
target->m_joinedSlots.erase(jit);
found0 = 1;
break;
}
}
if(!found0)
abort();
for(it=m_displayArrays.begin(); it!=m_displayArrays.end(); it++) {
found1 = 0;
for(jt=target->m_displayArrays.begin(); jt!=target->m_displayArrays.end(); jt++) {
if(*jt == *it) {
target->m_displayArrays.erase(jt);
target->m_endarray--;
found1 = 1;
break;
}
}
if(!found1)
abort();
}
if(target->m_displayArrays.size()) {
target->m_endvertex = target->m_displayArrays.back()->m_vertex.size();
target->m_endindex = target->m_displayArrays.back()->m_index.size();
}
else {
target->m_endvertex = 0;
target->m_endindex = 0;
}
MT_Matrix4x4 ntransform = m_joinInvTransform.inverse().transposed();
ntransform[0][3]= ntransform[1][3]= ntransform[2][3]= 0.0f;
for(begin(mit); !end(mit); next(mit))
for(i=mit.startvertex; i<mit.endvertex; i++)
mit.vertex[i].Transform(m_joinInvTransform, ntransform);
if (target->m_DisplayList) {
target->m_DisplayList->Release();
target->m_DisplayList = NULL;
}
return true;
}
return false;
}
#ifdef USE_SPLIT
bool RAS_MeshSlot::IsCulled()
{
if(m_joinSlot)
return true;
if(!m_bCulled)
return false;
list<RAS_MeshSlot*>::iterator it;
for(it=m_joinedSlots.begin(); it!=m_joinedSlots.end(); it++)
if(!(*it)->m_bCulled)
return false;
return true;
}
#endif
/* material bucket sorting */
struct RAS_MaterialBucket::less
{
bool operator()(const RAS_MaterialBucket* x, const RAS_MaterialBucket* y) const
{
return *x->GetPolyMaterial() < *y->GetPolyMaterial();
}
};
/* material bucket */
RAS_MaterialBucket::RAS_MaterialBucket(RAS_IPolyMaterial* mat)
{
m_material = mat;
}
RAS_MaterialBucket::~RAS_MaterialBucket()
{
}
RAS_IPolyMaterial* RAS_MaterialBucket::GetPolyMaterial() const
{
return m_material;
}
bool RAS_MaterialBucket::IsAlpha() const
{
return (m_material->IsAlpha());
}
bool RAS_MaterialBucket::IsZSort() const
{
return (m_material->IsZSort());
}
RAS_MeshSlot* RAS_MaterialBucket::AddMesh(int numverts)
{
RAS_MeshSlot *ms;
m_meshSlots.push_back(RAS_MeshSlot());
ms = &m_meshSlots.back();
ms->init(this, numverts);
return ms;
}
RAS_MeshSlot* RAS_MaterialBucket::CopyMesh(RAS_MeshSlot *ms)
{
m_meshSlots.push_back(RAS_MeshSlot(*ms));
return &m_meshSlots.back();
}
void RAS_MaterialBucket::RemoveMesh(RAS_MeshSlot* ms)
{
list<RAS_MeshSlot>::iterator it;
for(it=m_meshSlots.begin(); it!=m_meshSlots.end(); it++) {
if(&*it == ms) {
m_meshSlots.erase(it);
return;
}
}
}
list<RAS_MeshSlot>::iterator RAS_MaterialBucket::msBegin()
{
return m_meshSlots.begin();
}
list<RAS_MeshSlot>::iterator RAS_MaterialBucket::msEnd()
{
return m_meshSlots.end();
}
bool RAS_MaterialBucket::ActivateMaterial(const MT_Transform& cameratrans, RAS_IRasterizer* rasty,
RAS_IRenderTools *rendertools)
{
bool uselights;
if(!rasty->SetMaterial(*m_material))
return false;
uselights= m_material->UsesLighting(rasty);
rendertools->ProcessLighting(rasty, uselights, cameratrans);
return true;
}
void RAS_MaterialBucket::RenderMeshSlot(const MT_Transform& cameratrans, RAS_IRasterizer* rasty,
RAS_IRenderTools* rendertools, RAS_MeshSlot &ms)
{
m_material->ActivateMeshSlot(ms, rasty);
if (ms.m_pDeformer)
{
ms.m_pDeformer->Apply(m_material);
// KX_ReInstanceShapeFromMesh(ms.m_mesh); // Recompute the physics mesh. (Can't call KX_* from RAS_)
}
if(IsZSort() && rasty->GetDrawingMode() >= RAS_IRasterizer::KX_SOLID)
ms.m_mesh->SortPolygons(ms, cameratrans*MT_Transform(ms.m_OpenGLMatrix));
rendertools->PushMatrix();
if (!ms.m_pDeformer || !ms.m_pDeformer->SkipVertexTransform())
{
rendertools->applyTransform(rasty,ms.m_OpenGLMatrix,m_material->GetDrawingMode());
}
if(rasty->QueryLists())
if(ms.m_DisplayList)
ms.m_DisplayList->SetModified(ms.m_mesh->MeshModified());
// verify if we can use display list, not for deformed object, and
// also don't create a new display list when drawing shadow buffers,
// then it won't have texture coordinates for actual drawing. also
// for zsort we can't make a display list, since the polygon order
// changes all the time.
if(ms.m_pDeformer)
ms.m_bDisplayList = false;
else if(!ms.m_DisplayList && rasty->GetDrawingMode() == RAS_IRasterizer::KX_SHADOW)
ms.m_bDisplayList = false;
else if (IsZSort())
ms.m_bDisplayList = false;
else if(m_material->UsesObjectColor() && ms.m_bObjectColor)
ms.m_bDisplayList = false;
else
ms.m_bDisplayList = true;
// for text drawing using faces
if (m_material->GetDrawingMode() & RAS_IRasterizer::RAS_RENDER_3DPOLYGON_TEXT)
rasty->IndexPrimitives_3DText(ms, m_material, rendertools);
// for multitexturing
else if((m_material->GetFlag() & (RAS_MULTITEX|RAS_BLENDERGLSL)))
rasty->IndexPrimitivesMulti(ms);
// use normal IndexPrimitives
else
rasty->IndexPrimitives(ms);
if(rasty->QueryLists())
if(ms.m_DisplayList)
ms.m_mesh->SetMeshModified(false);
rendertools->PopMatrix();
}
void RAS_MaterialBucket::Optimize(MT_Scalar distance)
{
/* TODO: still have to check before this works correct:
* - lightlayer, frontface, text, billboard
* - make it work with physics */
#if 0
list<RAS_MeshSlot>::iterator it;
list<RAS_MeshSlot>::iterator jt;
// greed joining on all following buckets
for(it=m_meshSlots.begin(); it!=m_meshSlots.end(); it++)
for(jt=it, jt++; jt!=m_meshSlots.end(); jt++)
jt->Join(&*it, distance);
#endif
}