Skip to content

GitLab

Commit b41f9709 authored by Sebastian Eibl's avatar Sebastian Eibl
Browse files

extended checks in DynamicParMetis

parent 1b218e73
Hide whitespace changes
Inline Side-by-side
src/blockforest/loadbalancing/DynamicParMetis.cpp
View file @ b41f9709
......@@ -37,10 +37,13 @@
#include <boost/algorithm/string/case_conv.hpp>
#include <boost/algorithm/string/trim.hpp>
#include <array>
#include <vector>
namespace walberla {
namespace blockforest {
std::pair<uint_t, uint_t> getBlockSequenceRange( const PhantomBlockForest & phantomForest, MPI_Comm comm )
std::pair<uint_t, uint_t> getBlockSequenceRange( uint_t numLocalBlocks, MPI_Comm comm )
{
const uint_t rank = uint_c(mpi::translateRank(mpi::MPIManager::instance()->comm(), comm, MPIManager::instance()->rank()));
WALBERLA_DEBUG_SECTION()
......@@ -50,8 +53,6 @@ std::pair<uint_t, uint_t> getBlockSequenceRange( const PhantomBlockForest & phan
WALBERLA_ASSERT_EQUAL(rank, rankRaw);
}
uint_t numLocalBlocks = phantomForest.getNumberOfBlocks();
size_t sequenceStartOnProcess = 0;
MPI_Exscan( &numLocalBlocks, &sequenceStartOnProcess, 1, MPITrait<uint_t>::type(), MPI_SUM, comm );
if( rank == 0 )
......@@ -60,14 +61,13 @@ std::pair<uint_t, uint_t> getBlockSequenceRange( const PhantomBlockForest & phan
return std::make_pair( sequenceStartOnProcess, sequenceStartOnProcess + numLocalBlocks );
}
std::map< blockforest::BlockID, uint_t > getBlockIdToSequenceMapping( const PhantomBlockForest & phantomForest, const std::pair<uint_t, uint_t> & blockSequenceRange, MPI_Comm comm )
std::map< blockforest::BlockID, uint_t > getBlockIdToSequenceMapping( const PhantomBlockForest& phantomForest, const std::vector< std::pair< const PhantomBlock *, uint_t > > & targetProcess, const std::pair<uint_t, uint_t> & blockSequenceRange, MPI_Comm comm )
{
std::map< blockforest::BlockID, uint_t > mapping;
const auto & blockMap = phantomForest.getBlockMap();
uint_t sequenceId = blockSequenceRange.first;
for( auto it = blockMap.begin(); it != blockMap.end(); ++it )
mapping.insert( std::make_pair( it->first, sequenceId++ ) );
for( auto it = targetProcess.begin(); it != targetProcess.end(); ++it )
mapping.insert( std::make_pair( it->first->getId(), sequenceId++ ) );
WALBERLA_ASSERT_EQUAL( sequenceId, blockSequenceRange.second );
const std::vector<uint_t> neighborProcesses = phantomForest.getNeighboringProcesses();
......@@ -124,11 +124,11 @@ bool DynamicParMetis::operator()( std::vector< std::pair< const PhantomBlock *,
globalTimer.start();
//create new communicator which excludes processes which do not have blocks
MPI_Comm subComm = MPIManager::instance()->comm();
MPI_Comm subComm = MPI_COMM_NULL;
MPI_Group allGroup, subGroup;
MPI_Comm_group( MPIManager::instance()->comm(), &allGroup );
std::vector<int> ranks;
if (phantomForest.getNumberOfBlocks() > 0)
if (targetProcess.size() > 0)
ranks.push_back( MPIManager::instance()->rank() );
ranks = mpi::allGatherv( ranks, MPIManager::instance()->comm() );
auto numSubProcesses = ranks.size();
......@@ -136,22 +136,51 @@ bool DynamicParMetis::operator()( std::vector< std::pair< const PhantomBlock *,
MPI_Group_incl(allGroup, int_c(ranks.size()), &ranks[0], &subGroup);
MPI_Comm_create( MPIManager::instance()->comm(), subGroup, &subComm);
if ( targetProcess.size() != 0)
{
int subRank;
int subSize;
MPI_Comm_rank(subComm, &subRank);
MPI_Comm_size(subComm, &subSize);
WALBERLA_CHECK_GREATER_EQUAL(subRank, 0);
WALBERLA_CHECK_LESS(subRank, subSize);
} else
{
int subRank;
MPI_Comm_rank(subComm, &subRank);
WALBERLA_CHECK_EQUAL(subRank, MPI_UNDEFINED);
}
int64_t edgecut = 0;
std::vector<int64_t> part( phantomForest.getNumberOfBlocks(), int64_c( MPIManager::instance()->rank() ) );
WALBERLA_CHECK_EQUAL( phantomForest.getNumberOfBlocks(), targetProcess.size() );
std::vector<int64_t> part( targetProcess.size(), int64_c( MPIManager::instance()->rank() ) );
if (subComm != MPI_COMM_NULL)
{
const std::pair<uint_t, uint_t> blockSequenceRange = getBlockSequenceRange( phantomForest, subComm );
const std::map< blockforest::BlockID, uint_t > mapping = getBlockIdToSequenceMapping( phantomForest, blockSequenceRange, subComm ); //blockid to vertex id
int subRank;
int subSize;
MPI_Comm_rank(subComm, &subRank);
MPI_Comm_size(subComm, &subSize);
WALBERLA_CHECK_UNEQUAL(targetProcess.size(), 0);
const std::pair<uint_t, uint_t> blockSequenceRange = getBlockSequenceRange( targetProcess.size(), subComm );
const std::map< blockforest::BlockID, uint_t > mapping = getBlockIdToSequenceMapping( phantomForest, targetProcess, blockSequenceRange, subComm ); //blockid to vertex id
std::vector<int64_t> vtxdist = mpi::allGather( int64_c( blockSequenceRange.second ), subComm );
vtxdist.insert( vtxdist.begin(), uint_t( 0 ) );
WALBERLA_CHECK_EQUAL( vtxdist.size(), subSize + 1 );
for (size_t i = 1; i < vtxdist.size(); ++i)
{
WALBERLA_ASSERT_LESS( vtxdist[i-1], vtxdist[i] );
}
std::vector<int64_t> adjncy, xadj, vsize, vwgt, adjwgt;
std::vector<double> xyz;
for( auto it = targetProcess.begin(); it != targetProcess.end(); ++it )
uint_t blockIndex = 0;
for( auto it = targetProcess.begin(); it != targetProcess.end(); ++it, ++blockIndex )
{
WALBERLA_CHECK_EQUAL(blockIndex, mapping.find(it->first->getId())->second - blockSequenceRange.first);
xadj.push_back( int64_c( adjncy.size() ) );
const PhantomBlock & block = *( it->first );
auto bi = block.getData< DynamicParMetisBlockInfo >();
......@@ -163,9 +192,12 @@ bool DynamicParMetis::operator()( std::vector< std::pair< const PhantomBlock *,
{
auto mapIt = mapping.find( nit->getId() );
WALBERLA_ASSERT_UNEQUAL( mapIt, mapping.end(), "BlockId of neighbor is not contained in sequence mapping!" );
WALBERLA_CHECK_GREATER_EQUAL( mapIt->second, 0 );
WALBERLA_CHECK_LESS( mapIt->second, vtxdist.back() );
adjncy.push_back( int64_c( mapIt->second ) );
auto edgeWeightIt = bi.getEdgeWeights().find( nit->getId() );
adjwgt.push_back( edgeWeightIt == bi.getEdgeWeights().end() ? int64_t( 0 ) : edgeWeightIt->second );
//WALBERLA_CHECK_UNEQUAL( edgeWeightIt->second, 0 ); // perhaps WARNING
adjwgt.push_back( edgeWeightIt == bi.getEdgeWeights().end() ? int64_t( 1 ) : edgeWeightIt->second );
}
break;
case PARMETIS_EDGES_FROM_EDGE_WEIGHTS:
......@@ -173,10 +205,15 @@ bool DynamicParMetis::operator()( std::vector< std::pair< const PhantomBlock *,
{
auto mapIt = mapping.find( edgeIt->first );
WALBERLA_ASSERT_UNEQUAL( mapIt, mapping.end(), "BlockId of neighbor is not contained in sequence mapping!" );
WALBERLA_CHECK_GREATER_EQUAL( mapIt->second, 0 );
WALBERLA_CHECK_LESS( mapIt->second, vtxdist.back() );
adjncy.push_back( int64_c( mapIt->second ) );
//WALBERLA_CHECK_UNEQUAL( edgeIt->second, 0 ); // perhaps WARNING
adjwgt.push_back( edgeIt->second );
}
break;
}
WALBERLA_CHECK_UNEQUAL( bi.getVertexWeight(), 0 );
vwgt.push_back( bi.getVertexWeight() );
vsize.push_back( bi.getVertexSize() );
xyz.push_back( bi.getVertexCoords()[0] );
......@@ -185,46 +222,58 @@ bool DynamicParMetis::operator()( std::vector< std::pair< const PhantomBlock *,
}
xadj.push_back( int64_c( adjncy.size() ) );
WALBERLA_ASSERT_EQUAL( vtxdist.size(), numSubProcesses + uint_t( 1 ) );
WALBERLA_ASSERT_EQUAL( xadj.size(), phantomForest.getNumberOfBlocks() + 1 );
WALBERLA_ASSERT_EQUAL( vwgt.size(), phantomForest.getNumberOfBlocks() );
WALBERLA_ASSERT_EQUAL( vsize.size(), phantomForest.getNumberOfBlocks() );
WALBERLA_ASSERT_EQUAL( adjncy.size(), adjwgt.size() );
WALBERLA_ASSERT_EQUAL( adjwgt.size(), xadj.back() );
int64_t wgtflag = weightsToUse_;
int64_t numflag = 0; // C-style ordering
int64_t ncon = 1; // Number of constraints
int64_t ndims = 3; // Number of dimensions
double ubvec[] = { real_t( 1.05 ) }; // imbalance tolerance
int64_t nparts = int64_c( MPIManager::instance()->numProcesses() ); // number of subdomains
auto ipc2redist = ipc2redist_;
MPI_Comm comm = subComm; //MPIManager::instance()->comm();
WALBERLA_CHECK_EQUAL( vtxdist.size(), numSubProcesses + uint_t( 1 ) );
WALBERLA_CHECK_EQUAL( xadj.size(), targetProcess.size() + 1 );
WALBERLA_CHECK_EQUAL( xadj.front(), 0);
WALBERLA_CHECK_EQUAL( xadj.back(), adjncy.size() );
for (size_t i = 1; i < xadj.size(); ++i)
{
WALBERLA_ASSERT_LESS( xadj[i-1], xadj[i] );
}
WALBERLA_CHECK_EQUAL( vwgt.size(), targetProcess.size() );
WALBERLA_CHECK_EQUAL( vsize.size(), targetProcess.size() );
WALBERLA_CHECK_EQUAL( xyz.size(), targetProcess.size() * 3 );
WALBERLA_CHECK_EQUAL( adjncy.size(), adjwgt.size() );
WALBERLA_CHECK_EQUAL( adjwgt.size(), xadj.back() );
int64_t wgtflag = weightsToUse_;
int64_t numflag = 0; // C-style ordering
int64_t ncon = 1; // Number of constraints
int64_t ndims = 3; // Number of dimensions
std::vector<double> ubvec( uint_c(ncon), double_c( 1.05 ) ); // imbalance tolerance
int64_t nparts = int64_c( MPIManager::instance()->numProcesses() ); // number of subdomains
double ipc2redist = double_c(ipc2redist_);
MPI_Comm comm = subComm; //MPIManager::instance()->comm();
std::vector<double> tpwgts( uint_c(nparts * ncon), 1.0 / double_c( nparts ) ); // vertex weight fraction that is stored in a subdomain
int64_t options[] = { int64_t( 1 ), int64_t( 0 ), int64_t( 23 ), int64_t( 1 ) };
std::vector<int64_t> options = { int64_t( 1 ), int64_t( 0 ), int64_t( 23 ), int64_t( 1 ) };
int metisResult = core::METIS_OK;
switch( algorithm_ )
{
case PARMETIS_PART_GEOM:
parmetisTimer.start();
metisResult = core::ParMETIS_V3_PartGeom( ptr( vtxdist ), &ndims, ptr( xyz ), ptr( part ), &comm );
parmetisTimer.end();
break;
case PARMETIS_PART_GEOM_KWAY:
parmetisTimer.start();
metisResult = core::ParMETIS_V3_PartGeomKway( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( adjwgt ), &wgtflag, &numflag, &ndims, ptr( xyz ), &ncon, &nparts, ptr( tpwgts ), ubvec, options, &edgecut, ptr( part ), &comm );
metisResult = core::ParMETIS_V3_PartGeomKway( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( adjwgt ), &wgtflag, &numflag, &ndims, ptr( xyz ), &ncon, &nparts, ptr( tpwgts ), ptr( ubvec ), ptr( options ), &edgecut, ptr( part ), &comm );
parmetisTimer.end();
break;
case PARMETIS_PART_KWAY:
parmetisTimer.start();
metisResult = core::ParMETIS_V3_PartKway( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( adjwgt ), &wgtflag, &numflag, &ncon, &nparts, ptr( tpwgts ), ubvec, options, &edgecut, ptr( part ), &comm );
metisResult = core::ParMETIS_V3_PartKway( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( adjwgt ), &wgtflag, &numflag, &ncon, &nparts, ptr( tpwgts ), ptr(ubvec), ptr(options), &edgecut, ptr( part ), &comm );
parmetisTimer.end();
break;
case PARMETIS_ADAPTIVE_REPART:
parmetisTimer.start();
metisResult = core::ParMETIS_V3_AdaptiveRepart( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( vsize ), ptr( adjwgt ), &wgtflag, &numflag, &ncon, &nparts, ptr( tpwgts ), ubvec, &ipc2redist, options, &edgecut, ptr( part ), &comm );
metisResult = core::ParMETIS_V3_AdaptiveRepart( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( vsize ), ptr( adjwgt ), &wgtflag, &numflag, &ncon, &nparts, ptr( tpwgts ), ptr(ubvec), &ipc2redist, ptr(options), &edgecut, ptr( part ), &comm );
parmetisTimer.end();
break;
case PARMETIS_REFINE_KWAY:
parmetisTimer.start();
metisResult = core::ParMETIS_V3_RefineKway( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( adjwgt ), &wgtflag, &numflag, &ncon, &nparts, ptr( tpwgts ), ubvec, options, &edgecut, ptr( part ), &comm );
metisResult = core::ParMETIS_V3_RefineKway( ptr( vtxdist ), ptr( xadj ), ptr( adjncy ), ptr( vwgt ), ptr( adjwgt ), &wgtflag, &numflag, &ncon, &nparts, ptr( tpwgts ), ptr(ubvec), ptr(options), &edgecut, ptr( part ), &comm );
parmetisTimer.end();
break;
}
......@@ -279,6 +328,8 @@ DynamicParMetis::Algorithm DynamicParMetis::stringToAlgorithm( std::string s )
if( s == "PART_GEOM_KWAY" )
return PARMETIS_PART_GEOM_KWAY;
else if( s == "PART_GEOM" )
return PARMETIS_PART_GEOM;
else if( s == "PART_KWAY" )
return PARMETIS_PART_KWAY;
else if( s == "PART_ADAPTIVE_REPART" )
......@@ -328,6 +379,8 @@ std::string DynamicParMetis::algorithmToString( ) const
{
case walberla::blockforest::DynamicParMetis::PARMETIS_PART_GEOM_KWAY:
return "PART_GEOM_KWAY";
case walberla::blockforest::DynamicParMetis::PARMETIS_PART_GEOM:
return "PART_GEOM";
case walberla::blockforest::DynamicParMetis::PARMETIS_PART_KWAY:
return "PART_KWAY";
case walberla::blockforest::DynamicParMetis::PARMETIS_ADAPTIVE_REPART:
......
src/blockforest/loadbalancing/DynamicParMetis.h
View file @ b41f9709
......@@ -42,7 +42,7 @@ std::map< blockforest::BlockID, uint_t > getBlockIdToSequenceMapping( const Phan
class DynamicParMetis
{
public:
enum Algorithm { PARMETIS_PART_GEOM_KWAY, PARMETIS_PART_KWAY, PARMETIS_ADAPTIVE_REPART, PARMETIS_REFINE_KWAY };
enum Algorithm { PARMETIS_PART_GEOM, PARMETIS_PART_GEOM_KWAY, PARMETIS_PART_KWAY, PARMETIS_ADAPTIVE_REPART, PARMETIS_REFINE_KWAY };
enum WeightsToUse { PARMETIS_NO_WEIGHTS = 0, PARMETIS_EDGE_WEIGHTS = 1, PARMETIS_VERTEX_WEIGHTS = 2, PARMETIS_BOTH_WEIGHTS = 3 };
enum EdgeSource { PARMETIS_EDGES_FROM_FOREST, PARMETIS_EDGES_FROM_EDGE_WEIGHTS };
......
src/core/mpi/MPIHelper.cpp
View file @ b41f9709
......@@ -34,6 +34,11 @@ namespace mpi {
//!
int translateRank(const MPI_Comm srcComm, const MPI_Comm destComm, const int srcRank)
{
if (srcComm == destComm)
{
return srcRank;
}
int destRank = -1;
MPI_Group srcGroup, destGroup;
MPI_Comm_group(srcComm, &srcGroup);
......@@ -41,11 +46,9 @@ int translateRank(const MPI_Comm srcComm, const MPI_Comm destComm, const int src
MPI_Group_translate_ranks(srcGroup, 1, const_cast<int*>(&srcRank), destGroup, &destRank);
int size;
MPI_Comm_size(destComm, &size);
if (destRank < 0 || destRank >= size)
{
WALBERLA_CHECK_EQUAL( destRank, MPI_UNDEFINED );
destRank = -1;
}
if (destRank == MPI_UNDEFINED) destRank = -1;
WALBERLA_CHECK_GREATER_EQUAL(destRank, -1);
WALBERLA_CHECK_LESS(destRank, size);
MPI_Group_free(&srcGroup);
MPI_Group_free(&destGroup);
return destRank;
......@@ -60,6 +63,11 @@ int translateRank(const MPI_Comm srcComm, const MPI_Comm destComm, const int src
//!
std::vector<int> translateRank(const MPI_Comm srcComm, const MPI_Comm destComm, const std::vector<int>& srcRank)
{
if (srcComm == destComm)
{
return srcRank;
}
std::vector<int> destRank(srcRank.size(), -1);
MPI_Group srcGroup, destGroup;
MPI_Comm_group(srcComm, &srcGroup);
......@@ -69,11 +77,9 @@ std::vector<int> translateRank(const MPI_Comm srcComm, const MPI_Comm destComm,
MPI_Comm_size(destComm, &size);
for (auto& dstRnk : destRank)
{
if (dstRnk < 0 || dstRnk >= size)
{
WALBERLA_CHECK_EQUAL( dstRnk, MPI_UNDEFINED );
dstRnk = -1;
}
if (dstRnk == MPI_UNDEFINED) dstRnk = -1;
WALBERLA_CHECK_GREATER_EQUAL(dstRnk, -1);
WALBERLA_CHECK_LESS(dstRnk, size);
}
MPI_Group_free(&srcGroup);
MPI_Group_free(&destGroup);
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment