Added indices optimisation algorithm

Former-commit-id: 89086e814ca11a048949fe0fc6e20648824bbb36
This commit is contained in:
Lynix 2013-06-05 19:39:48 +02:00
parent 910a79fc84
commit ccb6f5b729
4 changed files with 539 additions and 0 deletions

View File

@ -16,6 +16,7 @@
#include <Nazara/Utility/Mesh.hpp>
NAZARA_API void NzComputeBoxIndexVertexCount(const NzVector3ui& subdivision, unsigned int* indexCount, unsigned int* vertexCount);
NAZARA_API unsigned int NzComputeCacheMissCount(NzIndexIterator indices, unsigned int indexCount);
NAZARA_API void NzComputeCubicSphereIndexVertexCount(unsigned int subdivision, unsigned int* indexCount, unsigned int* vertexCount);
NAZARA_API void NzComputeIcoSphereIndexVertexCount(unsigned int recursionLevel, unsigned int* indexCount, unsigned int* vertexCount);
NAZARA_API void NzComputePlaneIndexVertexCount(const NzVector2ui& subdivision, unsigned int* indexCount, unsigned int* vertexCount);
@ -27,6 +28,8 @@ NAZARA_API void NzGenerateIcoSphere(float size, unsigned int recursionLevel, con
NAZARA_API void NzGeneratePlane(const NzVector2ui& subdivision, const NzVector3f& position, const NzVector3f& normal, const NzVector2f& size, NzMeshVertex* vertices, NzIndexIterator indices, NzBoxf* aabb = nullptr, unsigned int indexOffset = 0);
NAZARA_API void NzGenerateUvSphere(float size, unsigned int sliceCount, unsigned int stackCount, const NzMatrix4f& matrix, NzMeshVertex* vertices, NzIndexIterator indices, NzBoxf* aabb = nullptr, unsigned int indexOffset = 0);
NAZARA_API void NzOptimizeIndices(NzIndexIterator indices, unsigned int indexCount);
inline void NzTransformVertex(NzMeshVertex* vertex, const NzMatrix4f& matrix);
inline void NzTransformVertices(NzMeshVertex* vertices, unsigned int vertexCount, const NzMatrix4f& matrix);

View File

@ -25,6 +25,8 @@ class NAZARA_API NzIndexBuffer : public NzResource
NzIndexBuffer(const NzIndexBuffer& indexBuffer);
~NzIndexBuffer();
unsigned int ComputeCacheMissCount() const;
bool Fill(const void* data, unsigned int offset, unsigned int length, bool forceDiscard = false);
NzBuffer* GetBuffer() const;
@ -41,6 +43,8 @@ class NAZARA_API NzIndexBuffer : public NzResource
void* Map(nzBufferAccess access, unsigned int offset = 0, unsigned int length = 0);
void* Map(nzBufferAccess access, unsigned int offset = 0, unsigned int length = 0) const;
void Optimize();
bool SetStorage(nzBufferStorage storage);
void Unmap() const;

View File

@ -2,8 +2,33 @@
// This file is part of the "Nazara Engine - Utility module"
// For conditions of distribution and use, see copyright notice in Config.hpp
/*
* vcacheopt.h - Vertex Cache Optimizer
* Copyright 2009 Michael Georgoulpoulos <mgeorgoulopoulos at gmail>
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include <Nazara/Utility/Algorithm.hpp>
#include <Nazara/Math/Basic.hpp>
#include <Nazara/Utility/IndexIterator.hpp>
#include <algorithm>
#include <unordered_map>
#include <Nazara/Utility/Debug.hpp>
@ -137,8 +162,481 @@ namespace
float m_size;
unsigned int m_vertexIndex;
};
// Source: https://code.google.com/p/vcacne/
// Auteur: Michael Georgoulpoulos
// Selon ce papier: http://home.comcast.net/~tom_forsyth/papers/fast_vert_cache_opt.html
// Modifié pour les besoins du moteur
///TODO: Déplacer dans un fichier à part ?
struct VertexCacheData
{
int position_in_cache = -1;
float current_score = 0.f;
int total_valence = 0; // toatl number of triangles using this vertex
int remaining_valence = 0; // number of triangles using it but not yet rendered
std::vector<int> tri_indices; // indices to the indices that use this vertex
bool calculated; // was the score calculated during this iteration?
int FindTriangle(int tri)
{
for (unsigned int i = 0; i < tri_indices.size(); ++i)
if (tri_indices[i] == tri) return i;
return -1;
}
void MoveTriangleToEnd(int tri)
{
auto it = std::find(tri_indices.begin(), tri_indices.end(), tri);
int t_ind = (it != tri_indices.end()) ? std::distance(tri_indices.begin(), it) : -1;
NazaraAssert(t_ind >= 0, "Triangle not found");
tri_indices.erase(tri_indices.begin() + t_ind, tri_indices.begin() + t_ind + 1);
tri_indices.push_back(tri);
}
};
struct TriangleCacheData
{
bool rendered = false; // has the triangle been added to the draw list yet?
float current_score = 0.f; // sum of the score of its vertices
int verts[3] = {-1, -1, -1}; // indices to the triangle's vertices
bool calculated = false; // was the score calculated during this iteration?
};
class VertexCache
{
public:
VertexCache()
{
Clear();
}
VertexCache(NzIndexIterator indices, unsigned int indexCount)
{
Clear();
for (unsigned int i = 0; i < indexCount; ++i)
AddVertex(*indices++);
}
// the vertex will be placed on top
// if the vertex didn't exist previewsly in
// the cache, then miss count is incermented
void AddVertex(unsigned int v)
{
int w = FindVertex(v);
if (w >= 0)
// remove the vertex from the cache (to reinsert it later on the top)
RemoveVertex(w);
else
// the vertex was not found in the cache - increment misses
m_misses++;
// shift all vertices down (to make room for the new top vertex)
for (int i=39; i>0; i--)
m_cache[i] = m_cache[i-1];
// add the new vertex on top
m_cache[0] = v;
}
void Clear()
{
for (int i=0; i<40; i++)
m_cache[i] = -1;
m_misses = 0;
}
int GetMissCount()
{
return m_misses;
}
int GetVertex(int which)
{
return m_cache[which];
}
private:
int FindVertex(int v)
{
for (int i = 0; i < 32; ++i)
{
if (m_cache[i] == v)
return i;
}
return -1;
}
void RemoveVertex(int stack_index)
{
for (int i=stack_index; i<38; i++)
m_cache[i] = m_cache[i+1];
}
int m_cache[40];
int m_misses; // cache miss count
};
class VertexCacheOptimizer
{
public:
enum Result
{
Success,
Fail_BadIndex,
Fail_NoVerts
};
VertexCacheOptimizer()
{
// initialize constants
m_cacheDecayPower = 1.5f;
m_lastTriScore = 0.75f;
m_valenceBoostScale = 2.0f;
m_valenceBoostPower = 0.5f;
m_bestTri = 0;
}
// stores new indices in place
Result Optimize(NzIndexIterator indices, unsigned int indexCount)
{
if (indexCount == 0)
return Fail_NoVerts;
// find vertex count
int max_vert = *std::max_element(indices, indices + indexCount);
Result res = Init(indices, indexCount, max_vert + 1);
if (res != Success)
return res;
// iterate until Iterate returns false
while (Iterate());
// rewrite optimized index list
for (int index : m_drawList)
{
*indices++ = m_triangles[index].verts[0];
*indices++ = m_triangles[index].verts[1];
*indices++ = m_triangles[index].verts[2];
}
return Success;
}
private:
float CalculateVertexScore(unsigned int vertex)
{
VertexCacheData *v = &m_vertices[vertex];
if (v->remaining_valence <= 0)
// No tri needs this vertex!
return -1.0f;
float ret = 0.0f;
if (v->position_in_cache < 0)
{
// Vertex is not in FIFO cache - no score.
}
else
{
if (v->position_in_cache < 3)
{
// This vertex was used in the last triangle,
// so it has a fixed score, whichever of the three
// it's in. Otherwise, you can get very different
// answers depending on whether you add
// the triangle 1,2,3 or 3,1,2 - which is silly.
ret = m_lastTriScore;
}
else
{
// Points for being high in the cache.
const float Scaler = 1.0f / (32 - 3);
ret = 1.0f - (v->position_in_cache - 3) * Scaler;
ret = std::pow(ret, m_cacheDecayPower);
}
}
// Bonus points for having a low number of tris still to
// use the vert, so we get rid of lone verts quickly.
float valence_boost = std::pow(static_cast<float>(v->remaining_valence), -m_valenceBoostPower);
ret += m_valenceBoostScale * valence_boost;
return ret;
}
// returns the index of the triangle with the highest score
// (or -1, if there aren't any active triangles)
int FullScoreRecalculation()
{
// calculate score for all vertices
for (unsigned int i = 0; i < m_vertices.size(); ++i)
m_vertices[i].current_score = CalculateVertexScore(i);
// calculate scores for all active triangles
float max_score;
int max_score_tri = -1;
bool first_time = true;
for (unsigned int i = 0; i < m_triangles.size(); ++i)
{
if (m_triangles[i].rendered)
continue;
// sum the score of all the triangle's vertices
float sc = m_vertices[m_triangles[i].verts[0]].current_score +
m_vertices[m_triangles[i].verts[1]].current_score +
m_vertices[m_triangles[i].verts[2]].current_score;
m_triangles[i].current_score = sc;
if (first_time || sc > max_score)
{
first_time = false;
max_score = sc;
max_score_tri = i;
}
}
return max_score_tri;
}
Result InitialPass()
{
for (unsigned int i = 0; i < m_indices.size(); ++i)
{
int index = m_indices[i];
if (index < 0 || index >= static_cast<int>(m_vertices.size()))
return Fail_BadIndex;
m_vertices[index].total_valence++;
m_vertices[index].remaining_valence++;
m_vertices[index].tri_indices.push_back(i/3);
}
m_bestTri = FullScoreRecalculation();
return Success;
}
Result Init(NzIndexIterator indices, unsigned int indexCount, int vertex_count)
{
// clear the draw list
m_drawList.clear();
// allocate and initialize vertices and triangles
m_vertices.clear(); // Pour reconstruire tous les éléments
m_vertices.resize(vertex_count);
m_triangles.clear();
for (unsigned int i = 0; i < indexCount; i += 3)
{
TriangleCacheData dat;
for (unsigned int j = 0; j < 3; ++j)
dat.verts[j] = indices[i + j];
m_triangles.push_back(dat);
}
// copy the indices
m_indices.resize(indexCount);
for (unsigned int i = 0; i < indexCount; ++i)
m_indices[i] = indices[i];
m_vertexCache.Clear();
m_bestTri = -1;
return InitialPass();
}
void AddTriangleToDrawList(int tri)
{
// reset all cache positions
for (unsigned int i = 0; i < 32; ++i)
{
int ind = m_vertexCache.GetVertex(i);
if (ind < 0)
continue;
m_vertices[ind].position_in_cache = -1;
}
TriangleCacheData* t = &m_triangles[tri];
if (t->rendered)
return; // triangle is already in the draw list
for (unsigned int i = 0; i < 3; ++i)
{
// add all triangle vertices to the cache
m_vertexCache.AddVertex(t->verts[i]);
VertexCacheData *v = &m_vertices[t->verts[i]];
// decrease remaining velence
v->remaining_valence--;
// move the added triangle to the end of the vertex's
// triangle index list, so that the first 'remaining_valence'
// triangles in the list are the active ones
v->MoveTriangleToEnd(tri);
}
m_drawList.push_back(tri);
t->rendered = true;
// update all vertex cache positions
for (unsigned int i = 0; i < 32; ++i)
{
int ind = m_vertexCache.GetVertex(i);
if (ind < 0)
continue;
m_vertices[ind].position_in_cache = i;
}
}
// Optimization: to avoid duplicate calculations durind the same iteration,
// both vertices and triangles have a 'calculated' flag. This flag
// must be cleared at the beginning of the iteration to all *active* triangles
// that have one or more of their vertices currently cached, and all their
// other vertices.
// If there aren't any active triangles in the cache, the function returns
// false and full recalculation is performed.
bool CleanCalculationFlags()
{
bool ret = false;
for (unsigned int i = 0; i < 32; ++i)
{
int vert = m_vertexCache.GetVertex(i);
if (vert < 0)
continue;
VertexCacheData *v = &m_vertices[vert];
for (int j = 0; j < v->remaining_valence; j++)
{
TriangleCacheData *t = &m_triangles[v->tri_indices[j]];
// we actually found a triangle to process
ret = true;
// clear triangle flag
t->calculated = false;
// clear vertex flags
for (unsigned int k = 0; k < 3; ++k)
m_vertices[t->verts[k]].calculated = false;
}
}
return ret;
}
void TriangleScoreRecalculation(int tri)
{
TriangleCacheData* t = &m_triangles[tri];
// calculate vertex scores
float sum = 0.0f;
for (unsigned int i = 0; i < 3; ++i)
{
VertexCacheData *v = &m_vertices[t->verts[i]];
float sc = v->current_score;
if (!v->calculated)
{
sc = CalculateVertexScore(t->verts[i]);
}
v->current_score = sc;
v->calculated = true;
sum += sc;
}
t->current_score = sum;
t->calculated = true;
}
int PartialScoreRecalculation()
{
// iterate through all the vertices of the cache
bool first_time = true;
float max_score;
int max_score_tri = -1;
for (unsigned int i = 0; i < 32; ++i)
{
int vert = m_vertexCache.GetVertex(i);
if (vert < 0)
continue;
VertexCacheData *v = &m_vertices[vert];
// iterate through all *active* triangles of this vertex
for (int j=0; j<v->remaining_valence; j++)
{
int tri = v->tri_indices[j];
TriangleCacheData *t = &m_triangles[tri];
if (!t->calculated)
// calculate triangle score
TriangleScoreRecalculation(tri);
float sc = t->current_score;
// we actually found a triangle to process
if (first_time || sc > max_score)
{
first_time = false;
max_score = sc;
max_score_tri = tri;
}
}
}
return max_score_tri;
}
// returns true while there are more steps to take
// false when optimization is complete
bool Iterate()
{
if (m_drawList.size() == m_triangles.size())
return false;
// add the selected triangle to the draw list
AddTriangleToDrawList(m_bestTri);
// recalculate vertex and triangle scores and
// select the best triangle for the next iteration
m_bestTri = (CleanCalculationFlags()) ? PartialScoreRecalculation() : FullScoreRecalculation();
return true;
}
std::vector<VertexCacheData> m_vertices;
std::vector<TriangleCacheData> m_triangles;
std::vector<int> m_indices;
int m_bestTri; // the next triangle to add to the render list
VertexCache m_vertexCache;
std::vector<int> m_drawList;
// CalculateVertexScore constants
float m_cacheDecayPower;
float m_lastTriScore;
float m_valenceBoostScale;
float m_valenceBoostPower;
};
}
/**********************************NzCompute**********************************/
void NzComputeBoxIndexVertexCount(const NzVector3ui& subdivision, unsigned int* indexCount, unsigned int* vertexCount)
{
unsigned int xIndexCount, yIndexCount, zIndexCount;
@ -155,6 +653,12 @@ void NzComputeBoxIndexVertexCount(const NzVector3ui& subdivision, unsigned int*
*vertexCount = xVertexCount*2 + yVertexCount*2 + zVertexCount*2;
}
unsigned int NzComputeCacheMissCount(NzIndexIterator indices, unsigned int indexCount)
{
VertexCache cache(indices, indexCount);
return cache.GetMissCount();
}
void NzComputeCubicSphereIndexVertexCount(unsigned int subdivision, unsigned int* indexCount, unsigned int* vertexCount)
{
// Comme tous nos plans sont identiques, on peut optimiser un peu
@ -202,6 +706,8 @@ void NzComputeUvSphereIndexVertexCount(unsigned int sliceCount, unsigned int sta
*vertexCount = sliceCount * stackCount;
}
/**********************************NzGenerate*********************************/
void NzGenerateBox(const NzVector3f& lengths, const NzVector3ui& subdivision, const NzMatrix4f& matrix, NzMeshVertex* vertices, NzIndexIterator indices, NzBoxf* aabb, unsigned int indexOffset)
{
unsigned int xIndexCount, yIndexCount, zIndexCount;
@ -394,3 +900,12 @@ void NzGenerateUvSphere(float size, unsigned int sliceCount, unsigned int stackC
aabb->Set(-totalSize, totalSize);
}
}
/************************************Autres***********************************/
void NzOptimizeIndices(NzIndexIterator indices, unsigned int indexCount)
{
VertexCacheOptimizer optimizer;
if (optimizer.Optimize(indices, indexCount) != VertexCacheOptimizer::Success)
NazaraWarning("Indices optimizer failed");
}

View File

@ -4,7 +4,10 @@
#include <Nazara/Utility/IndexBuffer.hpp>
#include <Nazara/Core/Error.hpp>
#include <Nazara/Utility/Algorithm.hpp>
#include <Nazara/Utility/Config.hpp>
#include <Nazara/Utility/IndexIterator.hpp>
#include <Nazara/Utility/IndexMapper.hpp>
#include <stdexcept>
#include <Nazara/Utility/Debug.hpp>
@ -64,6 +67,13 @@ m_startIndex(indexBuffer.m_startIndex)
NzIndexBuffer::~NzIndexBuffer() = default;
unsigned int NzIndexBuffer::ComputeCacheMissCount() const
{
NzIndexMapper mapper(this);
return NzComputeCacheMissCount(mapper.begin(), m_indexCount);
}
bool NzIndexBuffer::Fill(const void* data, unsigned int offset, unsigned int length, bool forceDiscard)
{
#if NAZARA_UTILITY_SAFE
@ -193,6 +203,13 @@ void* NzIndexBuffer::Map(nzBufferAccess access, unsigned int offset, unsigned in
return m_buffer->Map(access, m_startIndex+offset, (length) ? length : m_indexCount-offset);
}
void NzIndexBuffer::Optimize()
{
NzIndexMapper mapper(this);
NzOptimizeIndices(mapper.begin(), m_indexCount);
}
bool NzIndexBuffer::SetStorage(nzBufferStorage storage)
{
#if NAZARA_UTILITY_SAFE