From ecc316acd3352a9e8ff33f351f8360bc5afa07ee Mon Sep 17 00:00:00 2001
From: Thomas Witkowski <thomas.witkowski@gmx.de>
Date: Wed, 1 Oct 2008 11:34:22 +0000
Subject: [PATCH] * And some more changes for correct multi-threading
---
AMDiS/src/BasisFunction.cc | 6 +++---
AMDiS/src/BoundaryManager.cc | 2 +-
AMDiS/src/DOFVector.hh | 8 ++++----
AMDiS/src/DirichletBC.cc | 4 ++--
AMDiS/src/OpenMP.h | 9 +++++++++
AMDiS/src/Operator.cc | 2 +-
AMDiS/src/Operator.hh | 6 +++---
AMDiS/src/ResidualParallelEstimator.cc | 4 ++--
AMDiS/src/SecondOrderAssembler.cc | 6 +++---
AMDiS/src/SubAssembler.cc | 2 +-
AMDiS/src/TraverseParallel.cc | 2 +-
AMDiS/src/ZeroOrderAssembler.cc | 4 ++--
12 files changed, 32 insertions(+), 23 deletions(-)
diff --git a/AMDiS/src/BasisFunction.cc b/AMDiS/src/BasisFunction.cc
index b71b642c..20d5ad09 100644
--- a/AMDiS/src/BasisFunction.cc
+++ b/AMDiS/src/BasisFunction.cc
@@ -25,10 +25,10 @@ namespace AMDiS {
nDOF = NEW DimVec<int>(dim, DEFAULT_VALUE, -1);
dow = Global::getGeo(WORLD);
- grdTmpVec1.resize(omp_get_num_procs());
- grdTmpVec2.resize(omp_get_num_procs());
+ grdTmpVec1.resize(omp_get_overall_max_threads());
+ grdTmpVec2.resize(omp_get_overall_max_threads());
- for (int i = 0; i < omp_get_num_procs(); i++) {
+ for (int i = 0; i < omp_get_overall_max_threads(); i++) {
grdTmpVec1[i] = NEW DimVec<double>(dim, DEFAULT_VALUE, 0.0);
grdTmpVec2[i] = NEW DimVec<double>(dim, DEFAULT_VALUE, 0.0);
}
diff --git a/AMDiS/src/BoundaryManager.cc b/AMDiS/src/BoundaryManager.cc
index 416c08de..091c3932 100644
--- a/AMDiS/src/BoundaryManager.cc
+++ b/AMDiS/src/BoundaryManager.cc
@@ -11,7 +11,7 @@ namespace AMDiS {
BoundaryManager::BoundaryManager(const FiniteElemSpace *feSpace)
{
- localBounds.resize(omp_get_num_procs());
+ localBounds.resize(omp_get_overall_max_threads());
allocatedMemoryLocalBounds = feSpace->getBasisFcts()->getNumber();
for (int i = 0; i < static_cast<int>(localBounds.size()); i++) {
localBounds[i] = GET_MEMORY(BoundaryType, allocatedMemoryLocalBounds);
diff --git a/AMDiS/src/DOFVector.hh b/AMDiS/src/DOFVector.hh
index 068759d7..9abc9918 100644
--- a/AMDiS/src/DOFVector.hh
+++ b/AMDiS/src/DOFVector.hh
@@ -31,11 +31,11 @@ namespace AMDiS {
nBasFcts = feSpace->getBasisFcts()->getNumber();
int dim = feSpace->getMesh()->getDim();
- localIndices.resize(omp_get_num_procs());
- grdPhis.resize(omp_get_num_procs());
- D2Phis.resize(omp_get_num_procs());
+ localIndices.resize(omp_get_overall_max_threads());
+ grdPhis.resize(omp_get_overall_max_threads());
+ D2Phis.resize(omp_get_overall_max_threads());
- for (int i = 0; i < omp_get_num_procs(); i++) {
+ for (int i = 0; i < omp_get_overall_max_threads(); i++) {
localIndices[i] = GET_MEMORY(DegreeOfFreedom, this->nBasFcts);
grdPhis[i] = NEW DimVec<double>(dim, DEFAULT_VALUE, 0.0);
D2Phis[i] = NEW DimMat<double>(dim, NO_INIT);
diff --git a/AMDiS/src/DirichletBC.cc b/AMDiS/src/DirichletBC.cc
index cd67b70d..35421b31 100644
--- a/AMDiS/src/DirichletBC.cc
+++ b/AMDiS/src/DirichletBC.cc
@@ -14,7 +14,7 @@ namespace AMDiS {
f(fct),
dofVec(NULL)
{
- worldCoords.resize(omp_get_num_procs());
+ worldCoords.resize(omp_get_overall_max_threads());
};
DirichletBC::DirichletBC(BoundaryType type,
@@ -23,7 +23,7 @@ namespace AMDiS {
f(NULL),
dofVec(vec)
{
- worldCoords.resize(omp_get_num_procs());
+ worldCoords.resize(omp_get_overall_max_threads());
}
void DirichletBC::fillBoundaryCondition(DOFMatrix* matrix,
diff --git a/AMDiS/src/OpenMP.h b/AMDiS/src/OpenMP.h
index 37d351d2..799c157d 100644
--- a/AMDiS/src/OpenMP.h
+++ b/AMDiS/src/OpenMP.h
@@ -3,10 +3,15 @@
#ifdef _OPENMP
+#include <algorithm>
#include <omp.h>
const bool amdisHaveOpenMP = true;
+inline int omp_get_overall_max_threads() {
+ return std::max(omp_get_max_threads(), omp_get_num_threads());
+}
+
#else
const bool amdisHaveOpenMP = false;
@@ -27,6 +32,10 @@ inline int omp_get_thread_num() {
return 0;
}
+inline int omp_get_overall_max_threads() {
+ return 1;
+}
+
#endif
#endif
diff --git a/AMDiS/src/Operator.cc b/AMDiS/src/Operator.cc
index cedec770..676451a4 100644
--- a/AMDiS/src/Operator.cc
+++ b/AMDiS/src/Operator.cc
@@ -180,7 +180,7 @@ namespace AMDiS {
uhOld(NULL),
optimized(true)
{
- int maxThreads = omp_get_num_procs();
+ int maxThreads = omp_get_overall_max_threads();
assembler.resize(maxThreads);
secondOrder.resize(maxThreads);
diff --git a/AMDiS/src/Operator.hh b/AMDiS/src/Operator.hh
index d0508f8c..ee745836 100644
--- a/AMDiS/src/Operator.hh
+++ b/AMDiS/src/Operator.hh
@@ -6,7 +6,7 @@ namespace AMDiS {
secondOrder[0].push_back(term);
term->operat = this;
- for (int i = 1; i < omp_get_num_procs(); i++) {
+ for (int i = 1; i < omp_get_overall_max_threads(); i++) {
T *newTerm = NEW T(static_cast<const T>(*term));
secondOrder[i].push_back(newTerm);
}
@@ -23,7 +23,7 @@ namespace AMDiS {
}
term->operat = this;
- for (int i = 1; i < omp_get_num_procs(); i++) {
+ for (int i = 1; i < omp_get_overall_max_threads(); i++) {
T *newTerm = NEW T(static_cast<const T>(*term));
if (type == GRD_PSI) {
firstOrderGrdPsi[i].push_back(newTerm);
@@ -39,7 +39,7 @@ namespace AMDiS {
zeroOrder[0].push_back(term);
term->operat = this;
- for (int i = 1; i < omp_get_num_procs(); i++) {
+ for (int i = 1; i < omp_get_overall_max_threads(); i++) {
T *newTerm = NEW T(static_cast<const T>(*term));
zeroOrder[i].push_back(newTerm);
}
diff --git a/AMDiS/src/ResidualParallelEstimator.cc b/AMDiS/src/ResidualParallelEstimator.cc
index 5176a789..6e18ed62 100644
--- a/AMDiS/src/ResidualParallelEstimator.cc
+++ b/AMDiS/src/ResidualParallelEstimator.cc
@@ -14,9 +14,9 @@ namespace AMDiS {
GET_PARAMETER(0, name + "->C3", "%f", &C3);
C3 = C3 > 1.e-25 ? sqr(C3) : 0.0;
- seqEstimators_.resize(omp_get_num_procs());
+ seqEstimators_.resize(omp_get_overall_max_threads());
- for (int i = 0; i < omp_get_num_procs(); i++) {
+ for (int i = 0; i < omp_get_overall_max_threads(); i++) {
seqEstimators_[i] = NEW ResidualEstimator(name, r);
}
}
diff --git a/AMDiS/src/SecondOrderAssembler.cc b/AMDiS/src/SecondOrderAssembler.cc
index 181ef0ca..90e0329b 100644
--- a/AMDiS/src/SecondOrderAssembler.cc
+++ b/AMDiS/src/SecondOrderAssembler.cc
@@ -88,8 +88,8 @@ namespace AMDiS {
q11 = Q11PsiPhi::provideQ11PsiPhi(owner->getRowFESpace()->getBasisFcts(),
owner->getColFESpace()->getBasisFcts(),
quadrature);
- tmpLALt.resize(omp_get_num_procs());
- for (int i = 0; i < omp_get_num_procs(); i++) {
+ tmpLALt.resize(omp_get_overall_max_threads());
+ for (int i = 0; i < omp_get_overall_max_threads(); i++) {
tmpLALt[i] = NEW DimMat<double>*;
*(tmpLALt[i]) = NEW DimMat<double>(dim, NO_INIT);
}
@@ -163,7 +163,7 @@ namespace AMDiS {
Quad2::Quad2(Operator *op, Assembler *assembler, Quadrature *quad)
: SecondOrderAssembler(op, assembler, quad, true)
{
- tmpLALt.resize(omp_get_num_procs());
+ tmpLALt.resize(omp_get_overall_max_threads());
}
Quad2::~Quad2()
diff --git a/AMDiS/src/SubAssembler.cc b/AMDiS/src/SubAssembler.cc
index 3f8e7fd7..cc9048e3 100644
--- a/AMDiS/src/SubAssembler.cc
+++ b/AMDiS/src/SubAssembler.cc
@@ -34,7 +34,7 @@ namespace AMDiS {
nRow = psi->getNumber();
nCol = phi->getNumber();
- int maxThreads = omp_get_num_procs();
+ int maxThreads = omp_get_overall_max_threads();
terms.resize(maxThreads);
switch (order) {
diff --git a/AMDiS/src/TraverseParallel.cc b/AMDiS/src/TraverseParallel.cc
index 60a627da..92afbb17 100644
--- a/AMDiS/src/TraverseParallel.cc
+++ b/AMDiS/src/TraverseParallel.cc
@@ -9,7 +9,7 @@ namespace AMDiS {
TraverseParallelStack::TraverseParallelStack(int nThreads)
{
if (nThreads == 0) {
- nThreads_ = omp_get_num_procs();
+ nThreads_ = omp_get_overall_max_threads();
} else {
nThreads_ = nThreads;
}
diff --git a/AMDiS/src/ZeroOrderAssembler.cc b/AMDiS/src/ZeroOrderAssembler.cc
index 5e2b1535..f51fe81e 100644
--- a/AMDiS/src/ZeroOrderAssembler.cc
+++ b/AMDiS/src/ZeroOrderAssembler.cc
@@ -227,12 +227,12 @@ namespace AMDiS {
FastQuadZOA::FastQuadZOA(Operator *op, Assembler *assembler, Quadrature *quad)
: ZeroOrderAssembler(op, assembler, quad, true)
{
- cPtrs.resize(omp_get_num_procs());
+ cPtrs.resize(omp_get_overall_max_threads());
}
FastQuadZOA::~FastQuadZOA()
{
- for (int i = 0; i < omp_get_num_procs(); i++) {
+ for (int i = 0; i < omp_get_overall_max_threads(); i++) {
FREE_MEMORY(cPtrs[i], double, quadrature->getNumPoints());
}
}
--
GitLab