v4games
/
assimp
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix verbose string + refactorings

pull/5267/head
Kim Kulling 2023-10-07 14:21:59 +02:00
parent 3ce8af18e2
commit 2998ad82a1
1 changed files with 113 additions and 104 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

189
code/PostProcessing/ImproveCacheLocality.cpp
View File

@ -3,9 +3,7 @@
Open Asset Import Library (assimp) Open Asset Import Library (assimp)
--------------------------------------------------------------------------- ---------------------------------------------------------------------------
Copyright (c) 2006-2022, assimp team Copyright (c) 2006-2023, assimp team
All rights reserved. All rights reserved.
@ -59,31 +57,31 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <stdio.h> #include <stdio.h>
#include <stack> #include <stack>
using namespace Assimp; namespace Assimp {
// ------------------------------------------------------------------------------------------------ // ------------------------------------------------------------------------------------------------
// Constructor to be privately used by Importer // Constructor to be privately used by Importer
ImproveCacheLocalityProcess::ImproveCacheLocalityProcess() ImproveCacheLocalityProcess::ImproveCacheLocalityProcess() :
: mConfigCacheDepth(PP_ICL_PTCACHE_SIZE) { mConfigCacheDepth(PP_ICL_PTCACHE_SIZE) {
// empty // empty
} }
// ------------------------------------------------------------------------------------------------ // ------------------------------------------------------------------------------------------------
// Returns whether the processing step is present in the given flag field. // Returns whether the processing step is present in the given flag field.
bool ImproveCacheLocalityProcess::IsActive( unsigned int pFlags) const { bool ImproveCacheLocalityProcess::IsActive(unsigned int pFlags) const {
return (pFlags & aiProcess_ImproveCacheLocality) != 0; return (pFlags & aiProcess_ImproveCacheLocality) != 0;
} }
// ------------------------------------------------------------------------------------------------ // ------------------------------------------------------------------------------------------------
// Setup configuration // Setup configuration
void ImproveCacheLocalityProcess::SetupProperties(const Importer* pImp) { void ImproveCacheLocalityProcess::SetupProperties(const Importer *pImp) {
// AI_CONFIG_PP_ICL_PTCACHE_SIZE controls the target cache size for the optimizer // AI_CONFIG_PP_ICL_PTCACHE_SIZE controls the target cache size for the optimizer
mConfigCacheDepth = pImp->GetPropertyInteger(AI_CONFIG_PP_ICL_PTCACHE_SIZE,PP_ICL_PTCACHE_SIZE); mConfigCacheDepth = pImp->GetPropertyInteger(AI_CONFIG_PP_ICL_PTCACHE_SIZE, PP_ICL_PTCACHE_SIZE);
} }
// ------------------------------------------------------------------------------------------------ // ------------------------------------------------------------------------------------------------
// Executes the post processing step on the given imported data. // Executes the post processing step on the given imported data.
void ImproveCacheLocalityProcess::Execute( aiScene* pScene) { void ImproveCacheLocalityProcess::Execute(aiScene *pScene) {
if (!pScene->mNumMeshes) { if (!pScene->mNumMeshes) {
ASSIMP_LOG_DEBUG("ImproveCacheLocalityProcess skipped; there are no meshes"); ASSIMP_LOG_DEBUG("ImproveCacheLocalityProcess skipped; there are no meshes");
return; return;
@ -93,8 +91,8 @@ void ImproveCacheLocalityProcess::Execute( aiScene* pScene) {
float out = 0.f; float out = 0.f;
unsigned int numf = 0, numm = 0; unsigned int numf = 0, numm = 0;
for( unsigned int a = 0; a < pScene->mNumMeshes; ++a ){ for (unsigned int a = 0; a < pScene->mNumMeshes; ++a) {
const float res = ProcessMesh( pScene->mMeshes[a],a); const float res = ProcessMesh(pScene->mMeshes[a], a);
if (res) { if (res) {
numf += pScene->mMeshes[a]->mNumFaces; numf += pScene->mMeshes[a]->mNumFaces;
out += res; out += res;
@ -110,43 +108,20 @@ void ImproveCacheLocalityProcess::Execute( aiScene* pScene) {
} }
// ------------------------------------------------------------------------------------------------ // ------------------------------------------------------------------------------------------------
// Improves the cache coherency of a specific mesh static ai_real calculateInputACMR(aiMesh *pMesh, const aiFace *const pcEnd,
ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int meshNum) { unsigned int configCacheDepth, unsigned int meshNum) {
// TODO: rewrite this to use std::vector or boost::shared_array ai_real fACMR = 0.0f;
ai_assert(nullptr != pMesh); unsigned int *piFIFOStack = new unsigned int[configCacheDepth];
memset(piFIFOStack, 0xff, configCacheDepth * sizeof(unsigned int));
// Check whether the input data is valid unsigned int *piCur = piFIFOStack;
// - there must be vertices and faces const unsigned int *const piCurEnd = piFIFOStack + configCacheDepth;
// - all faces must be triangulated or we can't operate on them
if (!pMesh->HasFaces() || !pMesh->HasPositions())
return static_cast<ai_real>(0.f);
if (pMesh->mPrimitiveTypes != aiPrimitiveType_TRIANGLE) {
ASSIMP_LOG_ERROR("This algorithm works on triangle meshes only");
return static_cast<ai_real>(0.f);
}
if(pMesh->mNumVertices <= mConfigCacheDepth) {
return static_cast<ai_real>(0.f);
}
ai_real fACMR = 3.f;
const aiFace* const pcEnd = pMesh->mFaces+pMesh->mNumFaces;
// Input ACMR is for logging purposes only
if (!DefaultLogger::isNullLogger()) {
unsigned int* piFIFOStack = new unsigned int[mConfigCacheDepth];
memset(piFIFOStack,0xff,mConfigCacheDepth*sizeof(unsigned int));
unsigned int* piCur = piFIFOStack;
const unsigned int* const piCurEnd = piFIFOStack + mConfigCacheDepth;
// count the number of cache misses // count the number of cache misses
unsigned int iCacheMisses = 0; unsigned int iCacheMisses = 0;
for (const aiFace* pcFace = pMesh->mFaces;pcFace != pcEnd;++pcFace) { for (const aiFace *pcFace = pMesh->mFaces; pcFace != pcEnd; ++pcFace) {
for (unsigned int qq = 0; qq < 3;++qq) { for (unsigned int qq = 0; qq < 3; ++qq) {
bool bInCache = false; bool bInCache = false;
for (unsigned int* pp = piFIFOStack;pp < piCurEnd;++pp) { for (unsigned int *pp = piFIFOStack; pp < piCurEnd; ++pp) {
if (*pp == pcFace->mIndices[qq]) { if (*pp == pcFace->mIndices[qq]) {
// the vertex is in cache // the vertex is in cache
bInCache = true; bInCache = true;
@ -163,54 +138,88 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
} }
} }
delete[] piFIFOStack; delete[] piFIFOStack;
fACMR = (ai_real) iCacheMisses / pMesh->mNumFaces; fACMR = (ai_real)iCacheMisses / pMesh->mNumFaces;
if (3.0 == fACMR) { if (3.0 == fACMR) {
char szBuff[128]; // should be sufficiently large in every case char szBuff[128]; // should be sufficiently large in every case
// the JoinIdenticalVertices process has not been executed on this // the JoinIdenticalVertices process has not been executed on this
// mesh, otherwise this value would normally be at least minimally // mesh, otherwise this value would normally be at least minimally
// smaller than 3.0 ... // smaller than 3.0 ...
ai_snprintf(szBuff,128,"Mesh %u: Not suitable for vcache optimization",meshNum); ai_snprintf(szBuff, 128, "Mesh %u: Not suitable for vcache optimization", meshNum);
ASSIMP_LOG_WARN(szBuff); ASSIMP_LOG_WARN(szBuff);
return static_cast<ai_real>(0.f); return static_cast<ai_real>(0.f);
} }
return fACMR;
}
// ------------------------------------------------------------------------------------------------
// Improves the cache coherency of a specific mesh
ai_real ImproveCacheLocalityProcess::ProcessMesh(aiMesh *pMesh, unsigned int meshNum) {
// TODO: rewrite this to use std::vector or boost::shared_array
ai_assert(nullptr != pMesh);
// Check whether the input data is valid
// - there must be vertices and faces
// - all faces must be triangulated or we can't operate on them
if (!pMesh->HasFaces() || !pMesh->HasPositions())
return static_cast<ai_real>(0.f);
if (pMesh->mPrimitiveTypes != aiPrimitiveType_TRIANGLE) {
ASSIMP_LOG_ERROR("This algorithm works on triangle meshes only");
return static_cast<ai_real>(0.f);
}
if (pMesh->mNumVertices <= mConfigCacheDepth) {
return static_cast<ai_real>(0.f);
}
ai_real fACMR = 3.f;
const aiFace *const pcEnd = pMesh->mFaces + pMesh->mNumFaces;
// Input ACMR is for logging purposes only
if (!DefaultLogger::isNullLogger()) {
fACMR = calculateInputACMR(pMesh, pcEnd, mConfigCacheDepth, meshNum);
} }
// first we need to build a vertex-triangle adjacency list // first we need to build a vertex-triangle adjacency list
VertexTriangleAdjacency adj(pMesh->mFaces,pMesh->mNumFaces, pMesh->mNumVertices,true); VertexTriangleAdjacency adj(pMesh->mFaces, pMesh->mNumFaces, pMesh->mNumVertices, true);
// build a list to store per-vertex caching time stamps // build a list to store per-vertex caching time stamps
unsigned int* const piCachingStamps = new unsigned int[pMesh->mNumVertices]; std::vector<unsigned int> piCachingStamps;
memset(piCachingStamps,0x0,pMesh->mNumVertices*sizeof(unsigned int)); piCachingStamps.resize(pMesh->mNumVertices);
memset(&piCachingStamps[0], 0x0, pMesh->mNumVertices * sizeof(unsigned int));
// allocate an empty output index buffer. We store the output indices in one large array. // allocate an empty output index buffer. We store the output indices in one large array.
// Since the number of triangles won't change the input faces can be reused. This is how // Since the number of triangles won't change the input faces can be reused. This is how
// we save thousands of redundant mini allocations for aiFace::mIndices // we save thousands of redundant mini allocations for aiFace::mIndices
const unsigned int iIdxCnt = pMesh->mNumFaces*3; const unsigned int iIdxCnt = pMesh->mNumFaces * 3;
unsigned int* const piIBOutput = new unsigned int[iIdxCnt]; std::vector<unsigned int> piIBOutput;
unsigned int* piCSIter = piIBOutput; piIBOutput.resize(iIdxCnt);
std::vector<unsigned int>::iterator piCSIter = piIBOutput.begin();
// allocate the flag array to hold the information // allocate the flag array to hold the information
// whether a face has already been emitted or not // whether a face has already been emitted or not
std::vector<bool> abEmitted(pMesh->mNumFaces,false); std::vector<bool> abEmitted(pMesh->mNumFaces, false);
// dead-end vertex index stack // dead-end vertex index stack
std::stack<unsigned int, std::vector<unsigned int> > sDeadEndVStack; std::stack<unsigned int, std::vector<unsigned int>> sDeadEndVStack;
// create a copy of the piNumTriPtr buffer // create a copy of the piNumTriPtr buffer
unsigned int* const piNumTriPtr = adj.mLiveTriangles; unsigned int *const piNumTriPtr = adj.mLiveTriangles;
const std::vector<unsigned int> piNumTriPtrNoModify(piNumTriPtr, piNumTriPtr + pMesh->mNumVertices); const std::vector<unsigned int> piNumTriPtrNoModify(piNumTriPtr, piNumTriPtr + pMesh->mNumVertices);
// get the largest number of referenced triangles and allocate the "candidate buffer" // get the largest number of referenced triangles and allocate the "candidate buffer"
unsigned int iMaxRefTris = 0; { unsigned int iMaxRefTris = 0;
const unsigned int* piCur = adj.mLiveTriangles; {
const unsigned int* const piCurEnd = adj.mLiveTriangles+pMesh->mNumVertices; const unsigned int *piCur = adj.mLiveTriangles;
for (;piCur != piCurEnd;++piCur) { const unsigned int *const piCurEnd = adj.mLiveTriangles + pMesh->mNumVertices;
iMaxRefTris = std::max(iMaxRefTris,*piCur); for (; piCur != piCurEnd; ++piCur) {
iMaxRefTris = std::max(iMaxRefTris, *piCur);
} }
} }
ai_assert(iMaxRefTris > 0); ai_assert(iMaxRefTris > 0);
unsigned int* piCandidates = new unsigned int[iMaxRefTris*3]; std::vector<unsigned int> piCandidates;
piCandidates.resize(iMaxRefTris * 3);
unsigned int iCacheMisses = 0; unsigned int iCacheMisses = 0;
// ................................................................................... // ...................................................................................
@ -245,23 +254,23 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
int ivdx = 0; int ivdx = 0;
int ics = 1; int ics = 1;
int iStampCnt = mConfigCacheDepth+1; int iStampCnt = mConfigCacheDepth + 1;
while (ivdx >= 0) { while (ivdx >= 0) {
unsigned int icnt = piNumTriPtrNoModify[ivdx]; unsigned int icnt = piNumTriPtrNoModify[ivdx];
unsigned int* piList = adj.GetAdjacentTriangles(ivdx); unsigned int *piList = adj.GetAdjacentTriangles(ivdx);
unsigned int* piCurCandidate = piCandidates; std::vector<unsigned int>::iterator piCurCandidate = piCandidates.begin();
// get all triangles in the neighborhood // get all triangles in the neighborhood
for (unsigned int tri = 0; tri < icnt;++tri) { for (unsigned int tri = 0; tri < icnt; ++tri) {
// if they have not yet been emitted, add them to the output IB // if they have not yet been emitted, add them to the output IB
const unsigned int fidx = *piList++; const unsigned int fidx = *piList++;
if (!abEmitted[fidx]) { if (!abEmitted[fidx]) {
// so iterate through all vertices of the current triangle // so iterate through all vertices of the current triangle
const aiFace* pcFace = &pMesh->mFaces[ fidx ]; const aiFace *pcFace = &pMesh->mFaces[fidx];
unsigned nind = pcFace->mNumIndices; const unsigned nind = pcFace->mNumIndices;
for (unsigned ind = 0; ind < nind; ind++) { for (unsigned ind = 0; ind < nind; ind++) {
unsigned dp = pcFace->mIndices[ind]; unsigned dp = pcFace->mIndices[ind];
@ -281,7 +290,7 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
*piCSIter++ = dp; *piCSIter++ = dp;
// if the vertex is not yet in cache, set its cache count // if the vertex is not yet in cache, set its cache count
if (iStampCnt-piCachingStamps[dp] > mConfigCacheDepth) { if (iStampCnt - piCachingStamps[dp] > mConfigCacheDepth) {
piCachingStamps[dp] = iStampCnt++; piCachingStamps[dp] = iStampCnt++;
++iCacheMisses; ++iCacheMisses;
} }
@ -297,7 +306,7 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
// get next fanning vertex // get next fanning vertex
ivdx = -1; ivdx = -1;
int max_priority = -1; int max_priority = -1;
for (unsigned int* piCur = piCandidates;piCur != piCurCandidate;++piCur) { for (std::vector<unsigned int>::iterator piCur = piCandidates.begin(); piCur != piCurCandidate; ++piCur) {
const unsigned int dp = *piCur; const unsigned int dp = *piCur;
// must have live triangles // must have live triangles
@ -306,7 +315,7 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
// will the vertex be in cache, even after fanning occurs? // will the vertex be in cache, even after fanning occurs?
unsigned int tmp; unsigned int tmp;
if ((tmp = iStampCnt-piCachingStamps[dp]) + 2*piNumTriPtr[dp] <= mConfigCacheDepth) { if ((tmp = iStampCnt - piCachingStamps[dp]) + 2 * piNumTriPtr[dp] <= mConfigCacheDepth) {
priority = tmp; priority = tmp;
} }
@ -324,7 +333,7 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
while (!sDeadEndVStack.empty()) { while (!sDeadEndVStack.empty()) {
unsigned int iCachedIdx = sDeadEndVStack.top(); unsigned int iCachedIdx = sDeadEndVStack.top();
sDeadEndVStack.pop(); sDeadEndVStack.pop();
if (piNumTriPtr[ iCachedIdx ] > 0) { if (piNumTriPtr[iCachedIdx] > 0) {
ivdx = iCachedIdx; ivdx = iCachedIdx;
break; break;
} }
@ -345,29 +354,29 @@ ai_real ImproveCacheLocalityProcess::ProcessMesh( aiMesh* pMesh, unsigned int me
} }
ai_real fACMR2 = 0.0f; ai_real fACMR2 = 0.0f;
if (!DefaultLogger::isNullLogger()) { if (!DefaultLogger::isNullLogger()) {
fACMR2 = (float)iCacheMisses / pMesh->mNumFaces; fACMR2 = static_cast<ai_real>(iCacheMisses / pMesh->mNumFaces);
const ai_real averageACMR = ((fACMR - fACMR2) / fACMR) * 100.f;
// very intense verbose logging ... prepare for much text if there are many meshes // very intense verbose logging ... prepare for much text if there are many meshes
if ( DefaultLogger::get()->getLogSeverity() == Logger::VERBOSE) { if (DefaultLogger::get()->getLogSeverity() == Logger::VERBOSE) {
ASSIMP_LOG_VERBOSE_DEBUG("Mesh %u | ACMR in: ", meshNum, " out: ", fACMR, " | ~", fACMR2, ((fACMR - fACMR2) / fACMR) * 100.f); ASSIMP_LOG_VERBOSE_DEBUG("Mesh ", meshNum, "| ACMR in: ", fACMR, " out: ", fACMR2, " | average ACMR ", averageACMR);
} }
fACMR2 *= pMesh->mNumFaces; fACMR2 *= pMesh->mNumFaces;
} }
// sort the output index buffer back to the input array
piCSIter = piIBOutput;
for (aiFace* pcFace = pMesh->mFaces; pcFace != pcEnd;++pcFace) {
unsigned nind = pcFace->mNumIndices;
unsigned * ind = pcFace->mIndices;
if (nind > 0) ind[0] = *piCSIter++;
if (nind > 1) ind[1] = *piCSIter++;
if (nind > 2) ind[2] = *piCSIter++;
}
// delete temporary storage // sort the output index buffer back to the input array
delete[] piCachingStamps; piCSIter = piIBOutput.begin();
delete[] piIBOutput; for (aiFace *pcFace = pMesh->mFaces; pcFace != pcEnd; ++pcFace) {
delete[] piCandidates; unsigned nind = pcFace->mNumIndices;
unsigned *ind = pcFace->mIndices;
if (nind > 0)
ind[0] = *piCSIter++;
if (nind > 1)
ind[1] = *piCSIter++;
if (nind > 2)
ind[2] = *piCSIter++;
}
return fACMR2; return fACMR2;
} }
} // namespace Assimp