void MeshBase::partition (const unsigned int n_parts)
{
// If we get here and we have unpartitioned elements, we need that
// fixed.
if (this->n_unpartitioned_elem() > 0)
{
libmesh_assert (partitioner().get());
libmesh_assert (this->is_serial());
partitioner()->partition (*this, n_parts);
}
// NULL partitioner means don't repartition
// Non-serial meshes may not be ready for repartitioning here.
else if(!skip_partitioning() &&
partitioner().get())
{
partitioner()->partition (*this, n_parts);
}
else
{
// Adaptive coarsening may have "orphaned" nodes on processors
// whose elements no longer share them. We need to check for
// and possibly fix that.
Partitioner::set_node_processor_ids(*this);
// Make sure locally cached partition count
this->recalculate_n_partitions();
// Make sure any other locally cached data is correct
this->update_post_partitioning();
}
}
void MeshBase::partition (const unsigned int n_parts)
{
// If we get here and we have unpartitioned elements, we need that
// fixed.
if (this->n_unpartitioned_elem() > 0)
{
libmesh_assert (partitioner().get());
libmesh_assert (this->is_serial());
partitioner()->partition (*this, n_parts);
}
// A nullptr partitioner or a skip_partitioning(true) call or a
// skip_noncritical_partitioning(true) call means don't repartition;
// skip_noncritical_partitioning() checks all these.
else if (!skip_noncritical_partitioning())
{
partitioner()->partition (*this, n_parts);
}
else
{
// Adaptive coarsening may have "orphaned" nodes on processors
// whose elements no longer share them. We need to check for
// and possibly fix that.
MeshTools::correct_node_proc_ids(*this);
// Make sure locally cached partition count is correct
this->recalculate_n_partitions();
// Make sure any other locally cached data is correct
this->update_post_partitioning();
}
}
void MeshBase::partition (const unsigned int n_parts)
{
// NULL partitioner means don't partition
// Non-serial meshes aren't ready for partitioning yet.
if(!skip_partitioning() &&
partitioner().get() &&
this->is_serial())
{
partitioner()->partition (*this, n_parts);
}
else
{
// Make sure locally cached partition count
this->recalculate_n_partitions();
// Make sure any other locally cached data is correct
this->update_post_partitioning();
}
}
void AssemblyTree::rebuild_assembly_tree()
{
// Clear the current tree.
clear();
m_assembly_instances.clear();
Statistics statistics;
// Collect all assembly instances of the scene.
AABBVector assembly_instance_bboxes;
collect_assembly_instances(assembly_instance_bboxes);
RENDERER_LOG_INFO(
"building assembly tree (%s %s)...",
pretty_int(m_assembly_instances.size()).c_str(),
plural(m_assembly_instances.size(), "assembly instance").c_str());
// Create the partitioner.
typedef bvh::SAHPartitioner<AABBVector> Partitioner;
Partitioner partitioner(
assembly_instance_bboxes,
AssemblyTreeMaxLeafSize,
AssemblyTreeInteriorNodeTraversalCost,
AssemblyTreeTriangleIntersectionCost);
// Build the assembly tree.
typedef bvh::Builder<AssemblyTree, Partitioner> Builder;
Builder builder;
builder.build<DefaultWallclockTimer>(*this, partitioner, m_assembly_instances.size(), AssemblyTreeMaxLeafSize);
statistics.insert_time("build time", builder.get_build_time());
statistics.merge(bvh::TreeStatistics<AssemblyTree>(*this, AABB3d(m_scene.compute_bbox())));
if (!m_assembly_instances.empty())
{
const vector<size_t>& ordering = partitioner.get_item_ordering();
assert(m_assembly_instances.size() == ordering.size());
// Reorder the assembly instances according to the tree ordering.
vector<const AssemblyInstance*> temp_assembly_instances(ordering.size());
small_item_reorder(
&m_assembly_instances[0],
&temp_assembly_instances[0],
&ordering[0],
ordering.size());
// Store assembly instances in the tree leaves whenever possible.
store_assembly_instances_in_leaves(statistics);
}
// Print assembly tree statistics.
RENDERER_LOG_DEBUG("%s",
StatisticsVector::make(
"assembly tree statistics",
statistics).to_string().c_str());
}
int
main(int argc, const char* const* argv)
{
// parse arguments
int argi = 1;
for (; argi < argc; argi++) {
const char* arg = argv[argi];
if (arg[0] == '-') {
if (arg[1] == '-') {
// a double '-' option
if (strcmp(arg, "--help") == 0) {
print_usage_and_exit(false);
} else
print_usage_and_exit(true);
} else {
// single '-' options
for (int i = 1; arg[i] != '\0'; i++) {
switch (arg[i]) {
case 'h':
print_usage_and_exit(false);
default:
print_usage_and_exit(true);
}
}
}
} else
break;
}
// only the device path should remain
if (argi != argc - 1)
print_usage_and_exit(true);
const char* devicePath = argv[argi];
// get the disk device
BDiskDeviceRoster roster;
BDiskDevice device;
status_t error = roster.GetDeviceForPath(devicePath, &device);
if (error != B_OK) {
fprintf(stderr, "Error: Failed to get disk device for path \"%s\": "
"%s\n", devicePath, strerror(error));
}
Partitioner partitioner(&device);
partitioner.Run();
return 0;
}
void CurveTree::build_bvh(
const ParamArray& params,
const double time,
Statistics& statistics)
{
// Collect curves for this tree.
RENDERER_LOG_INFO(
"collecting geometry for curve tree #" FMT_UNIQUE_ID " from assembly \"%s\"...",
m_arguments.m_curve_tree_uid,
m_arguments.m_assembly.get_path().c_str());
vector<GAABB3> curve_bboxes;
collect_curves(curve_bboxes);
// Print statistics about the input geometry.
RENDERER_LOG_INFO(
"building curve tree #" FMT_UNIQUE_ID " (bvh, %s %s)...",
m_arguments.m_curve_tree_uid,
pretty_uint(m_curve_keys.size()).c_str(),
plural(m_curve_keys.size(), "curve").c_str());
// Create the partitioner.
typedef bvh::SAHPartitioner<vector<GAABB3> > Partitioner;
Partitioner partitioner(
curve_bboxes,
CurveTreeDefaultMaxLeafSize,
CurveTreeDefaultInteriorNodeTraversalCost,
CurveTreeDefaultCurveIntersectionCost);
// Build the tree.
typedef bvh::Builder<CurveTree, Partitioner> Builder;
Builder builder;
builder.build<DefaultWallclockTimer>(
*this,
partitioner,
m_curves1.size() + m_curves3.size(),
CurveTreeDefaultMaxLeafSize);
statistics.merge(
bvh::TreeStatistics<CurveTree>(*this, m_arguments.m_bbox));
// Reorder the curve keys based on the nodes ordering.
if (!m_curves1.empty() || !m_curves3.empty())
{
const vector<size_t>& ordering = partitioner.get_item_ordering();
reorder_curve_keys(ordering);
reorder_curves(ordering);
reorder_curve_keys_in_leaf_nodes();
}
}
void LogMapper::send_bucket_partition_presort(
Bucket* bucket,
storage::StorageType storage_type,
storage::PartitionId partition) {
storage::Partitioner partitioner(engine_, bucket->storage_id_);
char* io_base = reinterpret_cast<char*>(io_buffer_.get_block());
presort_ouputs_.assure_capacity(sizeof(BufferPosition) * bucket->counts_);
BufferPosition* outputs = reinterpret_cast<BufferPosition*>(presort_ouputs_.get_block());
uint32_t shortest_key_length = 0xFFFF;
uint32_t longest_key_length = 0;
if (storage_type == storage::kMasstreeStorage) {
for (uint32_t i = 0; i < bucket->counts_; ++i) {
uint64_t pos = from_buffer_position(bucket->log_positions_[i]);
const log::LogHeader* header = reinterpret_cast<const log::LogHeader*>(io_base + pos);
update_key_lengthes(header, storage_type, &shortest_key_length, &longest_key_length);
}
}
LogBuffer buffer(io_base);
uint32_t count = 0;
storage::Partitioner::SortBatchArguments args = {
buffer,
bucket->log_positions_,
bucket->counts_,
shortest_key_length,
longest_key_length,
&presort_buffer_,
parent_.get_base_epoch(),
outputs,
&count};
partitioner.sort_batch(args);
ASSERT_ND(count <= bucket->counts_);
bucket->counts_ = count; // it might be compacted
// then same as usual send_bucket_partition() except we use outputs
send_bucket_partition_general(bucket, storage_type, partition, outputs);
}
void peano::kernel::regulargrid::parallel::tests::SetupPartitionerTest::test2D_400x400ForkMessages() {
#ifdef Dim2
tarch::la::Vector<DIMENSIONS,int> domain;
assignList(domain) = 401,401;
peano::kernel::regulargrid::parallel::SetupPartitioner partitioner(domain,2);
partitioner._ranks.push_back(0);
partitioner._ranks.push_back(1);
tarch::la::Vector<DIMENSIONS,int> partition;
tarch::la::Vector<DIMENSIONS,double> domainOffset;
tarch::la::Vector<DIMENSIONS,double> h;
assignList(domainOffset) = 0.0, 0.0;
assignList(h) = 1.0/400.0, 1.0/400.0;
assignList(partition) = 0,0;
peano::kernel::regulargrid::parallel::messages::ForkMessage message00 =
partitioner.getForkMessage( partition, domainOffset, h );
validateEquals( message00.getH(), h );
validateEquals( message00.getNumberOfGridPoints()(0), 201 );
validateEquals( message00.getNumberOfGridPoints()(1), 401 );
validateEquals( message00.getDomainOffset()(0), 0.0);
validateEquals( message00.getDomainOffset()(1), 0.0);
assignList(partition) = 1,0;
peano::kernel::regulargrid::parallel::messages::ForkMessage message10 =
partitioner.getForkMessage( partition, domainOffset, h );
validateEquals( message10.getH(), h );
validateEquals( message10.getNumberOfGridPoints()(0), 201 );
validateEquals( message10.getNumberOfGridPoints()(1), 401 );
validateEquals( partitioner.getOffsetOfPartition(partition)(0), 200);
validateEquals( partitioner.getOffsetOfPartition(partition)(1), 0);
validateEquals( message10.getDomainOffset()(1), 0.0);
validateEquals( message10.getDomainOffset()(0), 0.5);
validateEquals( message10.getDomainOffset()(1), 0.0);
#endif
}
void LogMapper::flush_bucket(const BucketHashList& hashlist) {
ASSERT_ND(hashlist.head_);
ASSERT_ND(hashlist.tail_);
// temporary variables to store partitioning results
BufferPosition* position_array = reinterpret_cast<BufferPosition*>(
tmp_position_array_slice_.get_block());
PartitionSortEntry* sort_array = reinterpret_cast<PartitionSortEntry*>(
tmp_sort_array_slice_.get_block());
storage::PartitionId* partition_array = reinterpret_cast<storage::PartitionId*>(
tmp_partition_array_slice_.get_block());
LogBuffer log_buffer(reinterpret_cast<char*>(io_buffer_.get_block()));
const bool multi_partitions = engine_->get_options().thread_.group_count_ > 1U;
if (!engine_->get_storage_manager()->get_storage(hashlist.storage_id_)->exists()) {
// We ignore such logs in snapshot. As DROP STORAGE immediately becomes durable,
// There is no point to collect logs for the storage.
LOG(INFO) << "These logs are sent to a dropped storage.. ignore them";
return;
}
uint64_t log_count = 0; // just for reporting
debugging::StopWatch stop_watch;
for (Bucket* bucket = hashlist.head_; bucket != nullptr; bucket = bucket->next_bucket_) {
ASSERT_ND(bucket->counts_ > 0);
ASSERT_ND(bucket->counts_ <= kBucketMaxCount);
ASSERT_ND(bucket->storage_id_ == hashlist.storage_id_);
log_count += bucket->counts_;
// if there are multiple partitions, we first partition log entries.
if (multi_partitions) {
storage::Partitioner partitioner(engine_, bucket->storage_id_);
ASSERT_ND(partitioner.is_valid());
if (partitioner.is_partitionable()) {
// calculate partitions
for (uint32_t i = 0; i < bucket->counts_; ++i) {
position_array[i] = bucket->log_positions_[i];
ASSERT_ND(log_buffer.resolve(position_array[i])->header_.storage_id_
== bucket->storage_id_);
ASSERT_ND(log_buffer.resolve(position_array[i])->header_.storage_id_
== hashlist.storage_id_);
}
storage::Partitioner::PartitionBatchArguments args = {
static_cast< storage::PartitionId >(numa_node_),
log_buffer,
position_array,
bucket->counts_,
partition_array};
partitioner.partition_batch(args);
// sort the log positions by the calculated partitions
std::memset(sort_array, 0, sizeof(PartitionSortEntry) * bucket->counts_);
for (uint32_t i = 0; i < bucket->counts_; ++i) {
sort_array[i].set(partition_array[i], bucket->log_positions_[i]);
}
std::sort(sort_array, sort_array + bucket->counts_);
// let's reuse the current bucket as a temporary memory to hold sorted entries.
// buckets are discarded after the flushing, so this doesn't cause any issue.
const uint32_t original_count = bucket->counts_;
storage::PartitionId current_partition = sort_array[0].partition_;
bucket->log_positions_[0] = sort_array[0].position_;
bucket->counts_ = 1;
for (uint32_t i = 1; i < original_count; ++i) {
if (current_partition == sort_array[i].partition_) {
bucket->log_positions_[bucket->counts_] = sort_array[i].position_;
++bucket->counts_;
ASSERT_ND(bucket->counts_ <= original_count);
} else {
// the current partition has ended.
// let's send out these log entries to this partition
send_bucket_partition(bucket, current_partition);
// this is the beginning of next partition
current_partition = sort_array[i].partition_;
bucket->log_positions_[0] = sort_array[i].position_;
bucket->counts_ = 1;
}
}
ASSERT_ND(bucket->counts_ > 0);
// send out the last partition
send_bucket_partition(bucket, current_partition);
} else {
// in this case, it's same as single partition regarding this storage.
send_bucket_partition(bucket, 0);
}
} else {
// if it's not multi-partition, we blindly send everything to partition-0 (NUMA node 0)
send_bucket_partition(bucket, 0);
}
}
stop_watch.stop();
LOG(INFO) << to_string() << " sent out " << log_count << " log entries for storage-"
<< hashlist.storage_id_ << " in " << stop_watch.elapsed_ms() << " milliseconds";
}
int main(int argc, char **argv)
{
int k, i, **fd_p_s, j, mergenow, mergenext, ***mergfd, n, num_to_wait = 0, len, in, out, count, carry, gener = 1;
char *s, key[3] = "0";
FILE *f = fopen("psort.log", "w");
freopen("out.txt", "w", stdout);
fclose(f);
setbuf(stdout, NULL);
semid = semget(IPC_PRIVATE, 1, 0);
semctl(semid, 0, SETVAL, (int)1);
k = atoi(argv[1]);// number of processes
if (argc == 3) // check if inputed key "-n"
strncpy(key, argv[2], 2);
mergfd = (int ***)malloc(1 * sizeof(int**)); //two arrays
mergfd[0] = (int**)malloc(k * sizeof(int*));
fd_p_s = (int**)malloc(k * sizeof(int*)); // fd from partitioner to sorter
s = (char*)malloc(1* sizeof(char));
for (i = 0; i < k; ++i) {
fd_p_s[i] = (int*)malloc(2 * sizeof(int));
mergfd[0][i] = (int*)malloc(2 * sizeof(int));
pipe(fd_p_s[i]);
}
if (!fork())
partitioner(fd_p_s, k);
for (i = 0; i < k; ++i)
close(fd_p_s[i][1]);
for (i = 0; i < k; ++i){
pipe(mergfd[0][i]);
if (!fork())
sorter(i, fd_p_s, k, key, mergfd[0][i][1]); //sort strings and sents it's part to mergfd[0][i][1]
}
num_to_wait = k + 1; //partitioner + k sorters
for (i = 0; i < k; ++i)
close(fd_p_s[i][0]);
//start to merge in turn,
mergenow = k;
while (mergenow != 1){
++gener;
mergenext = mergenow/2 + mergenow%2;
mergfd = (int ***)realloc(mergfd, gener * sizeof(int**));
mergfd[gener - 1] = (int**)malloc(mergenext * sizeof(int*));
for (i = 0; i < mergenext; ++i){
mergfd[gener - 1][i] = (int*)malloc(2 * sizeof(int));
pipe(mergfd[gener - 1][i]);
}
count = 0;
while (count < mergenow) {
if (count + 1 == mergenow) { // not even number of mergers on this iteration; reached last one without pair
if (!fork()){
merger1(mergfd[gener - 2][count][0], mergfd[gener - 1][count / 2][1], key, count/2, gener);
return 0;
}
}
else {
if (!fork()){
merger2(mergfd[gener - 2][count][0], mergfd[gener - 2][count + 1][0], mergfd[gener - 1][count / 2][1], key, count/2, gener);
return 0;
}
}
count +=2;
++num_to_wait;
}
for(i = 0; i < mergenow; ++i)
close(mergfd[gener - 2][i][0]);
for(i = 0; i < mergenext; ++i)
close(mergfd[gener - 1][i][1]);
mergenow = mergenext;
}
my_read(mergfd[gener - 1][0][0], &n, sizeof(int));
for (i = 0; i < n; ++i){
my_read(mergfd[gener - 1][0][0], &len, sizeof(len));
s = (char*)realloc(s, len * sizeof(char));
my_read(mergfd[gener - 1][0][0], s, len);
fputs(s, stdout);
printf("\n");
}
for (i = 0; i < num_to_wait; ++i)
wait(NULL);
free(s);
return 0;
}
int main( int argc, char *argv[] )
{
// try {
time_t programStartTime(time(NULL));
boost::filesystem::path workingDir( boost::filesystem::current_path());
// ========== PROGRAM PARAMETERS ==========
std::string progName("partitiontree");
std::string configFilename("../../config/"+progName+".cfg");
unsigned int threads(0), levelDepth(3), filterRadius(0);
bool verbose(false), niftiMode( true );
// program parameters
std::string treeFilename, outputFolder;
// Declare a group of options that will be allowed only on command line
boost::program_options::options_description genericOptions("Generic options");
genericOptions.add_options()
( "version", "Program version" )
( "help,h", "Produce extended program help message" )
( "tree,t", boost::program_options::value< std::string >(&treeFilename), "file with the tree to compute partitions from")
( "outputf,O", boost::program_options::value< std::string >(&outputFolder), "output folder where partition files will be written")
( "search-depth,d", boost::program_options::value< unsigned int >(&levelDepth)->implicit_value(3), "[opt] optimal partition search depth (default = 3)")
( "filter-radius,r", boost::program_options::value< unsigned int >(&filterRadius)->implicit_value(0), "[opt] output partition filter kernel radius (default = 0 | no filtering)")
( "hoz", "[opt] obtain horizontal cut partitions (instead of Spread-Separation ones)")
( "maxgran,m", "[opt] obtain only the maximum granularity partition")
;
// Declare a group of options that will be allowed both on command line and in config file
boost::program_options::options_description configOptions("Configuration");
configOptions.add_options()
( "verbose,v", "[opt] verbose output." )
( "vista", "[opt] use vista file format (default is nifti)." )
( "pthreads,p", boost::program_options::value< unsigned int >(&threads), "[opt] number of processing threads to run the program in parallel, default: all available")
;
// Hidden options, will be allowed both on command line and in config file, but will not be shown to the user.
boost::program_options::options_description hiddenOptions("Hidden options");
//hiddenOptions.add_options() ;
boost::program_options::options_description cmdlineOptions;
cmdlineOptions.add(genericOptions).add(configOptions).add(hiddenOptions);
boost::program_options::options_description configFileOptions;
configFileOptions.add(configOptions).add(hiddenOptions);
boost::program_options::options_description visibleOptions("Allowed options");
visibleOptions.add(genericOptions).add(configOptions);
boost::program_options::positional_options_description posOpt; //this arguments do not need to specify the option descriptor when typed in
//posOpt.add("roi-file", -1);
boost::program_options::variables_map variableMap;
store(boost::program_options::command_line_parser(argc, argv).options(cmdlineOptions).positional(posOpt).run(), variableMap);
std::ifstream ifs(configFilename.c_str());
store(parse_config_file(ifs, configFileOptions), variableMap);
notify(variableMap);
if (variableMap.count("help"))
{
std::cout << "---------------------------------------------------------------------------" << std::endl;
std::cout << std::endl;
std::cout << " Project: hClustering" << std::endl;
std::cout << std::endl;
std::cout << " Whole-Brain Connectivity-Based Hierarchical Parcellation Project" << std::endl;
std::cout << " David Moreno-Dominguez" << std::endl;
std::cout << " [email protected]" << std::endl;
std::cout << " [email protected]" << std::endl;
std::cout << " www.cbs.mpg.de/~moreno" << std::endl;
std::cout << std::endl;
std::cout << " For more reference on the underlying algorithm and research they have been used for refer to:" << std::endl;
std::cout << " - Moreno-Dominguez, D., Anwander, A., & Knösche, T. R. (2014)." << std::endl;
std::cout << " A hierarchical method for whole-brain connectivity-based parcellation." << std::endl;
std::cout << " Human Brain Mapping, 35(10), 5000-5025. doi: http://dx.doi.org/10.1002/hbm.22528" << std::endl;
std::cout << " - Moreno-Dominguez, D. (2014)." << std::endl;
std::cout << " Whole-brain cortical parcellation: A hierarchical method based on dMRI tractography." << std::endl;
std::cout << " PhD Thesis, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig." << std::endl;
std::cout << " ISBN 978-3-941504-45-5" << std::endl;
std::cout << std::endl;
std::cout << " hClustering is free software: you can redistribute it and/or modify" << std::endl;
std::cout << " it under the terms of the GNU Lesser General Public License as published by" << std::endl;
std::cout << " the Free Software Foundation, either version 3 of the License, or" << std::endl;
std::cout << " (at your option) any later version." << std::endl;
std::cout << " http://creativecommons.org/licenses/by-nc/3.0" << std::endl;
std::cout << std::endl;
std::cout << " hClustering is distributed in the hope that it will be useful," << std::endl;
std::cout << " but WITHOUT ANY WARRANTY; without even the implied warranty of" << std::endl;
std::cout << " MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the" << std::endl;
std::cout << " GNU Lesser General Public License for more details." << std::endl;
std::cout << std::endl;
std::cout << "---------------------------------------------------------------------------" << std::endl << std::endl;
std::cout << "partitiontree" << std::endl << std::endl;
std::cout << "Obtain tree partitions at all granularity levels using the Spread-Separation method (finding the the partition with highest SS index at each granularity)." << std::endl;
std::cout << " Optimal SS value for each partition is searched within a defined search-depth hierarchical levels. Final partitions can be filtered with a defined kernel size." << std::endl;
std::cout << " to keep local SS maxima within that kernel. For SS index refer to (Moreno-Dominguez, 2014)" << std::endl;
std::cout << " For an interactive 3D partition management with more options please use the Hierarchcial Clustering module developed in OpenWalnut (www.openwalnut.org)." << std::endl << std::endl;
std::cout << "* Arguments:" << std::endl << std::endl;
std::cout << " --version: Program version." << std::endl << std::endl;
std::cout << " -h --help: produce extended program help message." << std::endl << std::endl;
std::cout << " -t --tree: File with the hierarchical tree to extract partitions from." << std::endl << std::endl;
std::cout << " -O --outputf: Output folder where partition files will be written." << std::endl << std::endl;
std::cout << "[-d --search-depth]: Search optimal partition for each granularity within d hierarchical levels." << std::endl;
std::cout << " A higher value will produce more optimized partition but will increase computing time." << std::endl;
std::cout << " Default: 3. Recommendened values: 3 for good quality and fast computation, 4 for enhanced quality." << std::endl << std::endl;
std::cout << "[-r --filter-radius]: Filter output partitions to keep only local SS (partition quality) maxima" << std::endl;
std::cout << " within a r-sized kernel across the granularity dimension." << std::endl << std::endl;
std::cout << "[-h --hoz]: Write horizontal cut partitions instead of SS ones (optimal partition search is still based on SS index)." << std::endl << std::endl;
std::cout << "[-m --maxgran]: Compute and write only the maximum granularity (meta-leaves) partition." << std::endl << std::endl;
std::cout << "[-v --verbose]: verbose output (recommended)." << std::endl << std::endl;
std::cout << "[--vista]: write output tree in vista coordinates (default is nifti)." << std::endl << std::endl;
std::cout << "[-p --pthreads]: number of processing threads to run the program in parallel. Default: use all available processors." << std::endl << std::endl;
std::cout << std::endl;
std::cout << "* Usage example:" << std::endl << std::endl;
std::cout << " partitiontree -t tree_lh.txt -O results/ -d 3 -r 50 -v" << std::endl << std::endl;
std::cout << std::endl;
std::cout << "* Outputs (in output folder defined at option -O):" << std::endl << std::endl;
std::cout << " (default outputs)" << std::endl;
std::cout << " - 'allSSparts_dX.txt' - (where X is the search depth level defined at parameter -d) Contains a summary of the partition information (cut value and size) for all granularities." << std::endl;
std::cout << " - 'TREE_SSparts_dX.txt' - (where TREE is the filename of the input tree defined at parameter -t) contains a copy of the original tree file with the partitions at all granularities included in the relevant fields." << std::endl;
std::cout << " - 'partitiontree_log.txt' - A text log file containing the parameter details and in-run and completion information of the program." << std::endl;
std::cout << std::endl;
std::cout << " (additional if using option -r)" << std::endl;
std::cout << " - 'filtSSparts_dX_rY.txt' - (where Y is the filter radius defined at parameter -r) Contains a summary of the resulting filtered partitions." << std::endl;
std::cout << " - 'TREE_SSparts_dX_rY.txt' - contains a copy of the original tree file with the resulting filtered partitions included in the relevant fields." << std::endl;
std::cout << std::endl;
std::cout << " (when using --hoz option, the prefix 'SS' will be replaced by 'Hoz'')" << std::endl;
std::cout << std::endl;
std::cout << " (alternative outputs when using option --maxgran)" << std::endl;
std::cout << " - 'fmaxgranPart.txt' - Contains the size information of the resulting maximal granularity partition for that tree." << std::endl;
std::cout << " - 'TREE_maxgranPart.txt' - contains a copy of the original tree file with the resulting max granularity partition included in the relevant fields." << std::endl;
std::cout << std::endl;
exit(0);
}
if (variableMap.count("version"))
{
std::cout << progName <<", version 2.0"<<std::endl;
exit(0);
}
if (variableMap.count("verbose"))
{
std::cout << "verbose output"<<std::endl;
verbose=true;
}
if (variableMap.count("pthreads"))
{
if (threads==1)
{
std::cout <<"Using a single processor"<< std::endl;
}
else if(threads==0 || threads>=omp_get_num_procs())
{
threads = omp_get_num_procs();
std::cout <<"Using all available processors ("<< threads <<")." << std::endl;
}
else
{
std::cout <<"Using a maximum of "<< threads <<" processors "<< std::endl;
}
omp_set_num_threads( threads );
}
else
{
threads = omp_get_num_procs();
omp_set_num_threads( threads );
std::cout <<"Using all available processors ("<< threads <<")." << std::endl;
}
if ( variableMap.count( "vista" ) )
{
if( verbose )
{
std::cout << "Using vista format" << std::endl;
}
fileManagerFactory fmf;
fmf.setVista();
niftiMode = false;
}
else
{
if( verbose )
{
std::cout << "Using nifti format" << std::endl;
}
fileManagerFactory fmf;
fmf.setNifti();
niftiMode = true;
}
if (variableMap.count("tree"))
{
if(!boost::filesystem::is_regular_file(boost::filesystem::path(treeFilename)))
{
std::cerr << "ERROR: tree file \""<<treeFilename<<"\" is not a regular file"<<std::endl;
std::cerr << visibleOptions << std::endl;
exit(-1);
}
std::cout << "Roi voxels file: "<< treeFilename << std::endl;
}
else
{
std::cerr << "ERROR: no tree file stated"<<std::endl;
std::cerr << visibleOptions << std::endl;
exit(-1);
}
if (variableMap.count("outputf"))
{
if(!boost::filesystem::is_directory(boost::filesystem::path(outputFolder)))
{
std::cerr << "ERROR: output folder \""<<outputFolder<<"\" is not a directory"<<std::endl;
std::cerr << visibleOptions << std::endl;
exit(-1);
}
std::cout << "Output folder: "<< outputFolder << std::endl;
}
else
{
std::cerr << "ERROR: no output folder stated"<<std::endl;
std::cerr << visibleOptions << std::endl;
exit(-1);
}
if (variableMap.count("maxgran"))
{
std::cout<<"Obtaining only max. granularity partition..."<<std::endl;
WHtree tree(treeFilename);
std::cout<<tree.getReport( false )<<std::endl;
if( tree.testRootBaseNodes() )
{
std::vector<size_t > maxpart( tree.getRootBaseNodes() );
std::vector<std::vector<size_t > > partitionVector( 1, maxpart);
std::vector<float > partitionValues(1,0);
std::cout<<"maxgranpart size: "<<std::endl<<maxpart.size()<<std::endl;
WHtreePartition partitioner(&tree);
std::string outPartFilename( outputFolder + "/maxgranPart.txt" );
partitioner.writePartitionSet( outPartFilename, partitionValues,partitionVector);
tree.insertPartitions( partitionVector, partitionValues );
std::string outTreeFilename( outputFolder + "/" + tree.getName() + "_maxgranPart" );
outTreeFilename += ( ".txt" );
tree.writeTree( outTreeFilename, niftiMode );
return 0;
}
else
{
std::cout<<"ERROR: tree does not have a maximum granularity meta-leaf partition"<<std::endl;
return(-1);
}
}
if( levelDepth > 5 )
{
std::cout << "Level depth indicated: " << levelDepth << " is too high, setting to a maximum of 5" << std::endl;
levelDepth = 5;
}
std::cout << "Using a search depth of: " << levelDepth << std::endl;
if( filterRadius > 1000 )
{
std::cout << "filter radius indicated: " << filterRadius << " is too high (max is 1000), setting to 100" << std::endl;
filterRadius = 10;
}
if( filterRadius == 0 )
{
std::cout << "using no filtering (radius 0)" << std::endl;
}
else if( filterRadius < 0 )
{
std::cout << "filter radius indicated: " << filterRadius << " must be positive. using no filtering (radius 0)" << std::endl;
filterRadius = 0;
}
else
{
std::cout << "Using a filter radius of: " << filterRadius << std::endl;
}
/////////////////////////////////////////////////////////////////
std::string logFilename(outputFolder+"/"+progName+"_log.txt");
std::ofstream logFile(logFilename.c_str());
if(!logFile) {
std::cerr << "ERROR: unable to open log file: \""<<logFilename<<"\""<<std::endl;
exit(-1);
}
logFile <<"Start Time:\t"<< ctime(&programStartTime) <<std::endl;
logFile <<"Working directory:\t"<< workingDir.string() <<std::endl;
logFile <<"Verbose:\t"<< verbose <<std::endl;
logFile <<"Tree file:\t"<< treeFilename <<std::endl;
logFile <<"Output folder:\t"<< outputFolder <<std::endl;
logFile <<"Verbose:\t"<< verbose <<std::endl;
if( niftiMode )
{
logFile << "Using nifti file format" << std::endl;
}
else
{
logFile << "Using vista file format" << std::endl;
}
WHtree tree(treeFilename);
logFile << tree.getReport( false ) <<std::endl;
std::cout<<tree.getReport( false )<<std::endl;
std::vector< float > partitionValues;
std::vector< std::vector< size_t> > partitionVector;
WHtreePartition treePartition(&tree);
std::string prefix;
if (variableMap.count("hoz"))
{
prefix = "Hoz";
std::cout <<"getting hoz partitions at all levels..." <<std::endl;
treePartition.scanHozPartitions( &partitionValues, &partitionVector );
std::cout << partitionValues.size() << " Partitions obtained, writing to file..." <<std::endl;
logFile <<"Initial partitions:\t"<< partitionValues.size() <<std::endl;
std::string outPartFilename( outputFolder + "/all" + prefix + "parts.txt" );
treePartition.writePartitionSet( outPartFilename, partitionValues, partitionVector);
tree.insertPartitions( partitionVector, partitionValues );
std::string outTreeFilename( outputFolder + "/" + tree.getName() + "_" + prefix + "parts_d" + boost::lexical_cast<std::string>(levelDepth) );
outTreeFilename += ( ".txt" );
tree.writeTree( outTreeFilename, niftiMode );
}
else
{
prefix = "SS";
std::cout <<"getting SS partitions at all levels..." <<std::endl;
treePartition.scanOptimalPartitions( levelDepth, &partitionValues, &partitionVector );
std::cout << partitionValues.size() << " Partitions obtained, writing to file..." <<std::endl;
logFile <<"Initial partitions:\t"<< partitionValues.size() <<std::endl;
std::string outPartFilename( outputFolder + "/all" + prefix + "parts_d" + boost::lexical_cast<std::string>(levelDepth) + ".txt" );
treePartition.writePartitionSet( outPartFilename, partitionValues, partitionVector);
tree.insertPartitions( partitionVector, partitionValues );
std::string outTreeFilename( outputFolder + "/" + tree.getName() + "_" + prefix + "parts_d" + boost::lexical_cast<std::string>(levelDepth) );
outTreeFilename += ( ".txt" );
tree.writeTree( outTreeFilename, niftiMode );
}
std::vector < unsigned int > filterRadii;
//filterRadii.reserve( 6 );
// filterRadii.push_back( 1 );
// filterRadii.push_back( 2 );
// filterRadii.push_back( 5 );
// filterRadii.push_back( 10 );
// filterRadii.push_back( 15 );
// filterRadii.push_back( 20 );
filterRadii.push_back( filterRadius );
for(size_t i=0; i< filterRadii.size(); ++i)
{
if( filterRadii[i] <= 0 )
{
continue;
}
std::vector< float > filtPartValues( partitionValues );
std::vector< std::vector< size_t> > filtPartVector( partitionVector );
std::cout << "Filtering with a radius of "<< filterRadii[i] << "..." <<std::endl;
treePartition.filterMaxPartitions( filterRadii[i], &filtPartValues, &filtPartVector );
std::cout << filtPartValues.size() << " Filtered partitions obtained, writing to file..." <<std::endl;
logFile <<"Filtered partitions:\t"<< filtPartValues.size() <<std::endl;
std::string outPartFilename( outputFolder + "/filt" + prefix + "parts_d" + boost::lexical_cast<std::string>(levelDepth) );
outPartFilename += ( "_r" + boost::lexical_cast<std::string>(filterRadii[i]) + ".txt" );
treePartition.writePartitionSet(outPartFilename, filtPartValues, filtPartVector);
std::cout << "Adding filtered partitions to tree and writing..." <<std::endl;
std::string outTreeFilename( outputFolder + "/" + tree.getName() + "_" + prefix + "parts_d" + boost::lexical_cast<std::string>(levelDepth) );
outTreeFilename += ( "_r" + boost::lexical_cast<std::string>(filterRadii[i]) + ".txt" );
tree.insertPartitions( filtPartVector, filtPartValues );
tree.writeTree( outTreeFilename, niftiMode );
}
/////////////////////////////////////////////////////////////////
// save and print total time
time_t programEndTime(time(NULL));
int totalTime( difftime(programEndTime,programStartTime) );
std::cout <<"Program Finished, total time: "<< totalTime/3600 <<"h "<< (totalTime%3600)/60 <<"' "<< ((totalTime%3600)%60) <<"\" "<< std::endl;
logFile <<"-------------"<<std::endl;
logFile <<"Finish Time:\t"<< ctime(&programEndTime) <<std::endl;
logFile <<"Elapsed time : "<< totalTime/3600 <<"h "<< (totalTime%3600)/60 <<"' "<< ((totalTime%3600)%60) <<"\""<< std::endl;
// }
// catch(std::exception& e)
// {
// std::cout << e.what() << std::endl;
// return 1;
// }
return 0;
}
ErrorStack LogReducer::dump_buffer_sort_storage(
const LogBuffer &buffer,
storage::StorageId storage_id,
const std::vector<BufferPosition>& log_positions,
uint32_t* out_shortest_key_length,
uint32_t* out_longest_key_length,
uint32_t* written_count) {
// first, count how many log entries are there. this is quick as we have a statistics
// in the header.
uint64_t records = 0;
for (BufferPosition position : log_positions) {
FullBlockHeader* header = reinterpret_cast<FullBlockHeader*>(buffer.resolve(position));
if (!header->is_full_block()) {
LOG(FATAL) << to_string() << " wtf. magic word doesn't match. position=" << position
<< ", storage_id=" << storage_id << *header;
}
header->assert_key_length();
records += header->log_count_;
}
// now we need a memory for this long array. expand the memory if not sufficient.
uint64_t positions_buffer_size = records * sizeof(BufferPosition);
expand_positions_buffers_if_needed(positions_buffer_size);
BufferPosition* inputs = reinterpret_cast<BufferPosition*>(input_positions_slice_.get_block());
uint64_t cur_rec_total = 0;
// put all log positions to the array
uint32_t shortest_key_length = 0xFFFF;
uint32_t longest_key_length = 0;
for (BufferPosition position : log_positions) {
FullBlockHeader* header = reinterpret_cast<FullBlockHeader*>(buffer.resolve(position));
if (!header->is_full_block()) {
LOG(FATAL) << to_string() << " wtf. magic word doesn't match. position=" << position
<< ", storage_id=" << storage_id << *header;
}
header->assert_key_length();
shortest_key_length = std::min<uint32_t>(shortest_key_length, header->shortest_key_length_);
longest_key_length = std::max<uint32_t>(longest_key_length, header->longest_key_length_);
BufferPosition record_pos = position + to_buffer_position(sizeof(FullBlockHeader));
for (uint32_t i = 0; i < header->log_count_; ++i) {
log::RecordLogType* record = buffer.resolve(record_pos);
ASSERT_ND(record->header_.storage_id_ == storage_id);
ASSERT_ND(record->header_.log_length_ > 0);
inputs[cur_rec_total] = record_pos;
++cur_rec_total;
record_pos += to_buffer_position(record->header_.log_length_);
}
ASSERT_ND(record_pos == position + header->block_length_);
}
ASSERT_ND(cur_rec_total == records);
// Now, sort these log records by key and then ordinal. we use the partitioner object for this.
storage::Partitioner partitioner(engine_, storage_id);
BufferPosition* pos = reinterpret_cast<BufferPosition*>(output_positions_slice_.get_block());
*written_count = 0;
storage::Partitioner::SortBatchArguments args = {
buffer,
inputs,
static_cast<uint32_t>(records),
shortest_key_length,
longest_key_length,
&sort_buffer_,
parent_.get_base_epoch(),
pos,
written_count};
partitioner.sort_batch(args);
*out_shortest_key_length = shortest_key_length;
*out_longest_key_length = longest_key_length;
return kRetOk;
}
void peano::kernel::regulargrid::parallel::tests::SetupPartitionerTest::test2D_12x8ForkMessages() {
#ifdef Dim2
tarch::la::Vector<DIMENSIONS,int> domain;
assignList(domain) = 12,9;
peano::kernel::regulargrid::parallel::SetupPartitioner partitioner(domain,9);
partitioner._ranks.push_back(0);
partitioner._ranks.push_back(1);
partitioner._ranks.push_back(2);
partitioner._ranks.push_back(3);
partitioner._ranks.push_back(4);
partitioner._ranks.push_back(5);
partitioner._ranks.push_back(6);
partitioner._ranks.push_back(7);
partitioner._ranks.push_back(8);
tarch::la::Vector<DIMENSIONS,int> partition;
tarch::la::Vector<DIMENSIONS,double> domainOffset;
tarch::la::Vector<DIMENSIONS,double> h;
assignList(domainOffset) = -2.0, -3.0;
assignList(h) = 0.4, 0.5;
assignList(partition) = 0,0;
peano::kernel::regulargrid::parallel::messages::ForkMessage message00 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 1,0;
peano::kernel::regulargrid::parallel::messages::ForkMessage message10 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 2,0;
peano::kernel::regulargrid::parallel::messages::ForkMessage message20 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 0,1;
peano::kernel::regulargrid::parallel::messages::ForkMessage message01 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 1,1;
peano::kernel::regulargrid::parallel::messages::ForkMessage message11 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 2,1;
peano::kernel::regulargrid::parallel::messages::ForkMessage message21 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 0,2;
peano::kernel::regulargrid::parallel::messages::ForkMessage message02 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 1,2;
peano::kernel::regulargrid::parallel::messages::ForkMessage message12 =
partitioner.getForkMessage( partition, domainOffset, h );
assignList(partition) = 2,2;
peano::kernel::regulargrid::parallel::messages::ForkMessage message22 =
partitioner.getForkMessage( partition, domainOffset, h );
// validateEquals( message00.getNeighbourRanks(), h );
validateEquals( message00.getH(), h );
validateEquals( message00.getNumberOfGridPoints()(0), 5 );
validateEquals( message00.getNumberOfGridPoints()(1), 4 );
validateEquals( message00.getDomainOffset()(0), domainOffset(0) + 0*h(0));
validateEquals( message00.getDomainOffset()(1), domainOffset(1) + 0*h(1));
validateEquals( message10.getH(), h );
validateEquals( message10.getNumberOfGridPoints()(0), 5 );
validateEquals( message10.getNumberOfGridPoints()(1), 4 );
validateEquals( message10.getDomainOffset()(0), domainOffset(0) + 4*h(0));
validateEquals( message10.getDomainOffset()(1), domainOffset(1) + 0*h(1));
validateEquals( message20.getH(), h );
validateEquals( message20.getNumberOfGridPoints()(0), 4 );
validateEquals( message20.getNumberOfGridPoints()(1), 4 );
validateEquals( message20.getDomainOffset()(0), domainOffset(0) + 8*h(0));
validateEquals( message20.getDomainOffset()(1), domainOffset(1) + 0*h(1));
validateEquals( message01.getH(), h );
validateEquals( message01.getNumberOfGridPoints()(0), 5 );
validateEquals( message01.getNumberOfGridPoints()(1), 4 );
validateEquals( message01.getDomainOffset()(0), domainOffset(0) + 0*h(0));
validateEquals( message01.getDomainOffset()(1), domainOffset(1) + 3*h(1));
#endif
}
void FMEMultipoleKernel::quadtreeConstruction(ArrayPartition& pointPartition)
{
FMELocalContext* localContext = m_pLocalContext;
FMEGlobalContext* globalContext = m_pGlobalContext;
LinearQuadtree& tree = *globalContext->pQuadtree;
// precompute the bounding box for the quadtree points from the graph nodes
for_loop(pointPartition, min_max_x_function(localContext));
for_loop(pointPartition, min_max_y_function(localContext));
// wait until the thread's bounding box is computed
sync();
// let the main thread computed the bounding box of the bounding boxes
if (isMainThread())
{
globalContext->min_x = globalContext->pLocalContext[0]->min_x;
globalContext->min_y = globalContext->pLocalContext[0]->min_y;
globalContext->max_x = globalContext->pLocalContext[0]->max_x;
globalContext->max_y = globalContext->pLocalContext[0]->max_y;
for (__uint32 j=1; j < numThreads(); j++)
{
globalContext->min_x = min(globalContext->min_x, globalContext->pLocalContext[j]->min_x);
globalContext->min_y = min(globalContext->min_y, globalContext->pLocalContext[j]->min_y);
globalContext->max_x = max(globalContext->max_x, globalContext->pLocalContext[j]->max_x);
globalContext->max_y = max(globalContext->max_y, globalContext->pLocalContext[j]->max_y);
};
tree.init(globalContext->min_x, globalContext->min_y, globalContext->max_x, globalContext->max_y);
globalContext->coolDown *= 0.999f;
tree.clear();
};
// wait because the morton number computation needs the bounding box
sync();
// udpate morton number to prepare them for sorting
for_loop(pointPartition, LQMortonFunctor(localContext));
// wait so we can sort them by morton number
sync();
#ifdef OGDF_FME_PARALLEL_QUADTREE_SORT
// use a simple parallel sorting algorithm
LinearQuadtree::LQPoint* points = tree.pointArray();
sort_parallel(points, tree.numberOfPoints(), LQPointComparer);
#else
if (isMainThread())
{
LinearQuadtree::LQPoint* points = tree.pointArray();
sort_single(points, tree.numberOfPoints(), LQPointComparer);
};
#endif
// wait because the quadtree builder needs the sorted order
sync();
// if not a parallel run, we can do the easy way
if (isSingleThreaded())
{
LinearQuadtreeBuilder builder(tree);
// prepare the tree
builder.prepareTree();
// and link it
builder.build();
LQPartitioner partitioner( localContext );
partitioner.partition();
} else // the more difficult part
{
// snap the left point of the interval of the thread to the first in the cell
LinearQuadtree::PointID beginPoint = tree.findFirstPointInCell(pointPartition.begin);
LinearQuadtree::PointID endPoint_plus_one;
// if this thread is the last one, no snapping required for the right point
if (threadNr()==numThreads()-1)
endPoint_plus_one = tree.numberOfPoints();
else // find the left point of the next thread
endPoint_plus_one = tree.findFirstPointInCell(pointPartition.end+1);
// and calculate the number of points to prepare
__uint32 numPointsToPrepare = endPoint_plus_one - beginPoint;
// now we can prepare the snapped interval
LinearQuadtreeBuilder builder(tree);
// this function prepares the tree from begin point to endPoint_plus_one-1 (EXCLUDING endPoint_plus_one)
builder.prepareTree(beginPoint, endPoint_plus_one);
// save the start, end and count of the inner node chain in the context
localContext->firstInnerNode = builder.firstInner;
localContext->lastInnerNode = builder.lastInner;
localContext->numInnerNodes = builder.numInnerNodes;
// save the start, end and count of the leaf node chain in the context
localContext->firstLeaf = builder.firstLeaf;
localContext->lastLeaf = builder.lastLeaf;
localContext->numLeaves = builder.numLeaves;
// wait until all are finished
sync();
// now the main thread has to link the tree
if (isMainThread())
{
// with his own builder
LinearQuadtreeBuilder sbuilder(tree);
// first we need the complete chain data
sbuilder.firstInner = globalContext->pLocalContext[0]->firstInnerNode;
sbuilder.firstLeaf = globalContext->pLocalContext[0]->firstLeaf;
sbuilder.numInnerNodes = globalContext->pLocalContext[0]->numInnerNodes;
sbuilder.numLeaves = globalContext->pLocalContext[0]->numLeaves;
for (__uint32 j=1; j < numThreads(); j++)
{
sbuilder.numLeaves += globalContext->pLocalContext[j]->numLeaves;
sbuilder.numInnerNodes += globalContext->pLocalContext[j]->numInnerNodes;
};
sbuilder.lastInner = globalContext->pLocalContext[numThreads()-1]->lastInnerNode;
sbuilder.lastLeaf = globalContext->pLocalContext[numThreads()-1]->lastLeaf;
// Link the tree
sbuilder.build();
// and run the partitions
LQPartitioner partitioner(localContext);
partitioner.partition();
};
};
// wait for tree to finish
sync();
// now update the copy of the point data
for_loop(pointPartition, LQPointUpdateFunctor(localContext));
// compute the nodes coordinates and sizes
tree.forall_tree_nodes(LQCoordsFunctor(localContext), localContext->innerNodePartition.begin, localContext->innerNodePartition.numNodes)();
tree.forall_tree_nodes(LQCoordsFunctor(localContext), localContext->leafPartition.begin, localContext->leafPartition.numNodes)();
};
ErrorStack execute(
Engine* engine,
storage::StorageId id,
double* elapsed_ms,
std::vector<std::string>* papi_results) {
storage::PartitionerMetadata* metadata = storage::PartitionerMetadata::get_metadata(engine, id);
make_dummy_partitions(engine, id, metadata);
LOG(INFO) << "Allocating memories...";
debugging::StopWatch alloc_watch;
memory::AlignedMemory::AllocType kAlloc = memory::AlignedMemory::kNumaAllocOnnode;
memory::AlignedMemory work_memory(kRecords * 32ULL, 1U << 21, kAlloc, 0);
memory::AlignedMemory positions_memory(sizeof(BufferPosition) * kRecords, 1U << 12, kAlloc, 0);
memory::AlignedMemory out_memory(sizeof(BufferPosition) * kRecords, 1U << 12, kAlloc, 0);
memory::AlignedMemory partitions_memory(sizeof(uint8_t) * kRecords, 1U << 12, kAlloc, 0);
memory::AlignedMemory log_memory(kRecords * kPayloadSize * 2ULL, 1U << 21, kAlloc, 0);
alloc_watch.stop();
LOG(INFO) << "Allocated memories in " << alloc_watch.elapsed_ms() << "ms";
LOG(INFO) << "Populating logs to process...";
debugging::StopWatch log_watch;
char* log_buffer = reinterpret_cast<char*>(log_memory.get_block());
uint64_t log_size = 0;
BufferPosition* log_positions = reinterpret_cast<BufferPosition*>(positions_memory.get_block());
populate_logs(id, log_buffer, log_positions, &log_size);
log_watch.stop();
LOG(INFO) << "Populated logs to process in " << log_watch.elapsed_ms() << "ms";
if (FLAGS_profile) {
COERCE_ERROR(engine->get_debug()->start_profile("partition_experiment.prof"));
engine->get_debug()->start_papi_counters();
}
LOG(INFO) << "experiment's main part has started";
debugging::StopWatch watch;
storage::Partitioner partitioner_base(engine, id);
ASSERT_ND(partitioner_base.is_valid());
storage::hash::HashPartitioner partitioner(&partitioner_base);
LogBuffer buf(log_buffer);
if (FLAGS_run_partition) {
LOG(INFO) << "running partitioning...";
storage::Partitioner::PartitionBatchArguments partition_args = {
0,
buf,
log_positions,
kRecords,
reinterpret_cast<uint8_t*>(partitions_memory.get_block())};
partitioner.partition_batch(partition_args);
}
if (FLAGS_run_sort) {
LOG(INFO) << "running sorting...";
uint32_t written_count;
storage::Partitioner::SortBatchArguments sort_args = {
buf,
log_positions,
kRecords,
8,
8,
&work_memory,
Epoch(1),
reinterpret_cast<BufferPosition*>(out_memory.get_block()),
&written_count};
partitioner.sort_batch(sort_args);
}
watch.stop();
*elapsed_ms = watch.elapsed_ms();
LOG(INFO) << "experiment's main part has ended. Took " << *elapsed_ms << "ms";
if (FLAGS_profile) {
engine->get_debug()->stop_profile();
engine->get_debug()->stop_papi_counters();
if (FLAGS_papi) {
*papi_results = debugging::DebuggingSupports::describe_papi_counters(
engine->get_debug()->get_papi_counters());
}
}
return kRetOk;
}
