Skip to content

Commit

Permalink
Test 8b (Bench Read Parallel): Support Variable encoding, Fix Bugs (o…
Browse files Browse the repository at this point in the history 
…penPMD#1131)

* added support to read variable encoding, plus fixed some bugs

* fixed style

* Update examples/8b_benchmark_read_parallel.cpp

remove commented out code

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* Update examples/8b_benchmark_read_parallel.cpp

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>

* removed commented line

* updated 8b env option

Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>
  • Loading branch information
@guj @ax3l
guj and ax3l committed Nov 3, 2021
1 parent d3f687e commit 9325daa
Show file tree
Hide file tree
Showing 2 changed files with 93 additions and 60 deletions.
8 changes: 7 additions & 1 deletion docs/source/usage/benchmarks.rst
View file
Original file line number Diff line number Diff line change
Expand Up @@ -96,7 +96,13 @@ The options are: a file prefix, and a read pattern


For example, if the files are in the format of ``/path/8a_parallel_3Db_%07T.bp``
the input will be: ``/path/8a_parallel_3Db <options>``
the input can be simply: ``/path/8a_parallel_3Db <options>``

otherwise, please use the full name of the file.

While openPMD-api supports more than one file types, this benchmark intents to read just one type.
By default, ADIOS2 file is assumed. If it is not the desired file type, one can hint with an environment variable, e.g.
``export OPENPMD_BENCHMARK_USE_BACKEND=HDF5``

The Read options intent to measure overall processing time in the following categories:

Expand Down
145 changes: 86 additions & 59 deletions examples/8b_benchmark_read_parallel.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -128,6 +128,7 @@ class Timer
//MemoryProfiler (rank, tag);
}
~Timer() {
MPI_Barrier(MPI_COMM_WORLD);
std::string tt = "~"+m_Tag;
//MemoryProfiler (m_Rank, tt.c_str());
m_End = std::chrono::system_clock::now();
Expand Down Expand Up @@ -188,9 +189,13 @@ std::vector<std::string> getBackends() {
#if openPMD_HAVE_ADIOS2
if( auxiliary::getEnvString( "OPENPMD_BP_BACKEND", "NOT_SET" ) != "ADIOS1" )
res.emplace_back(".bp");
if( auxiliary::getEnvString( "OPENPMD_BENCHMARK_USE_BACKEND", "NOT_SET" ) == "ADIOS" )
return res;
#endif

#if openPMD_HAVE_HDF5
if( auxiliary::getEnvString( "OPENPMD_BENCHMARK_USE_BACKEND", "NOT_SET" ) == "HDF5" )
res.clear();
res.emplace_back(".h5");
#endif
return res;
Expand All @@ -211,30 +216,22 @@ class TestInput
* Run the read tests
* assumes both GroupBased and fileBased series of this prefix exist.
* @ param prefix file prefix
* e.g. abc.bp (for group/variable based encoding)
* abc (for file based encoding)
*
*/
void run(const std::string& prefix)
{
{ // file based
std::ostringstream s;
s <<prefix<<"_%07T"<<m_Backend;

std::string filename = s.str();

read(filename);
}

return; // not doing group based..yet
/*
{ // group based
std::ostringstream s;
s <<prefix<<m_Backend;
std::string filename = s.str();
read(filename);
}
*/
if (prefix.find(m_Backend) == std::string::npos) {
// file based, default to %07T
std::ostringstream s;
s << prefix << "_%07T" << m_Backend;
std::string filename = s.str();
read(filename);
} else {
// group or variable based, or filebased with fullname
read(prefix);
}
} // run


Expand All @@ -245,7 +242,7 @@ class TestInput
*
*/
void
read(std::string& filename)
read(const std::string& filename)
{
try {
std::string tag = "Reading: "+filename ;
Expand All @@ -255,16 +252,22 @@ class TestInput
int numIterations = series.iterations.size();

if ( 0 == m_MPIRank )
std::cout<<"\n\t Num Iterations in " << filename<<" : " << numIterations<<std::endl;
{
std::cout << " "<<series.iterationEncoding() << std::endl;
std::cout << " Num Iterations in " << filename << " : " << numIterations << std::endl << std::endl;
}

{
int counter = 1;
for ( auto const& i : series.iterations )
for ( auto i : series.readIterations() )
{
if ( (1 == counter) || (numIterations == counter) )
readStep(series, i.first);
if ( counter % 5 == 1 )
readStep(series, i, counter - 1);
counter ++;
}
}
if ( 0 == m_MPIRank )
std::cout << " Total Num iterations read: " << (counter - 1) << std::endl << std::endl;
}
} catch (std::exception& ex)
{}
}
Expand Down Expand Up @@ -389,6 +392,7 @@ class TestInput
bool atCenter = ( (m_Pattern % 10 == 0) || (fractionOnDim == 1) );
bool atTopLeft = ( (m_Pattern % 10 == 1) && (fractionOnDim > 1) );
bool atBottomRight = ( (m_Pattern % 10 == 2) && (fractionOnDim > 1) );
bool overlay = ( (m_Pattern % 10 == 3) && (fractionOnDim > 1) );

bool rankZeroOnly = ( alongDim == 4);
bool diagnalBlocks = ( alongDim > meshExtent.size() ) && !rankZeroOnly;
Expand All @@ -404,6 +408,8 @@ class TestInput
s <<" topleft ";
else if (atBottomRight)
s <<" bottomRight ";
else if (overlay)
s << " near center ";
} else if ( diagnalBlocks )
s <<" blockStyle = diagnal";
else
Expand All @@ -423,12 +429,14 @@ class TestInput

if ( rankZeroOnly )
{
if ( atCenter )
if ( atTopLeft )
off[i] = 0; // top corner
else if ( atTopLeft )
else if ( atBottomRight )
off[i] = (meshExtent[i]-blob); // bottom corner
else if (atBottomRight)
else if (atCenter)
off[i] = (fractionOnDim/2) * blob; // middle corner
else if (overlay)
off[i] = (fractionOnDim/2) * blob - blob/3; // near middle corner
}
else
{
Expand Down Expand Up @@ -456,7 +464,7 @@ class TestInput
auto slice_data = rho.loadChunk<double>(off, ext);
series.flush();

std::cout<<"Rank: "<<m_MPIRank;
std::cout << " Rank: " << m_MPIRank;

prettyLambda(off,ext);
}
Expand Down Expand Up @@ -506,10 +514,10 @@ class TestInput
}

std::ostringstream so, sc;
so <<"Rank: "<<m_MPIRank<<" offset [ "; sc<<" count[ ";
so << " Rank: " << m_MPIRank << " offset [ "; sc << " count[ ";
for ( unsigned int i=0; i<meshExtent.size(); i++ )
{
so <<off[i]<<" ";
so << off[i] << " ";
sc <<ext[i]<<" ";
}
so <<"]"; sc<<"]";
Expand Down Expand Up @@ -573,7 +581,7 @@ class TestInput
* @param series openPMD series
*/
void
sliceField( Series& series, int ts )
sliceField(Series& series, IndexedIteration& iter)
{
if ( m_Pattern >= 100 )
return;
Expand All @@ -592,7 +600,7 @@ class TestInput
return;
whichDim -= 5;

MeshRecordComponent bx = series.iterations[ts].meshes["B"]["x"];
MeshRecordComponent bx = iter.meshes["B"]["x"];
Extent meshExtent = bx.getExtent();

if ( bx.getExtent().size() != 3) {
Expand All @@ -601,15 +609,15 @@ class TestInput
return;
}

MeshRecordComponent by = series.iterations[ts].meshes["B"]["y"];
MeshRecordComponent bz = series.iterations[ts].meshes["B"]["z"];
MeshRecordComponent by = iter.meshes["B"]["y"];
MeshRecordComponent bz = iter.meshes["B"]["z"];


Offset off(meshExtent.size(),0);
Extent ext(meshExtent.size(),1);

std::ostringstream s;
s<<" Eletrict Field slice: ";
s<<" Electric Field slice: ";
if (!getSlice(meshExtent, whichDim, rankZeroOnly, off, ext, s))
return;

Expand All @@ -626,52 +634,70 @@ class TestInput
* Read an iteration step, mesh & particles
*
* @param Series openPMD series
* @param iter iteration (actual iteration step may not equal to ts)
* @param ts timestep
*
*/
void
readStep( Series& series, int ts )
readStep( Series& series, IndexedIteration& iter, int ts )
{
std::string comp_name = openPMD::MeshRecordComponent::SCALAR;

MeshRecordComponent rho = series.iterations[ts].meshes["rho"][comp_name];

MeshRecordComponent rho = iter.meshes["rho"][comp_name];
Extent meshExtent = rho.getExtent();

if ( 0 == m_MPIRank )
{
std::cout<<"... rho meshExtent : ts="<<ts<<" [";
std::cout << "===> rho meshExtent : ts=" << ts << " [";
for (unsigned long i : meshExtent)
std::cout<<i<<" ";
std::cout<<"]"<<std::endl;
}

sliceMe(series, rho);
block(series, rho);
fullscan(series, rho);
std::vector<int> currPatterns;
if (m_Pattern > 0)
currPatterns.push_back(m_Pattern);
else
currPatterns.insert(currPatterns.end(), { 1, 5, 15, 25, 55, 65, 75, 440, 441, 442, 443, 7 });

sliceField(series, ts);
for(int i : currPatterns) {
m_Pattern = i;
sliceMe(series, rho);
block(series, rho);
fullscan(series, rho);

// read particles
if ( m_Pattern == 7 )
{
openPMD::ParticleSpecies electrons =
series.iterations[ts].particles["ion"];
RecordComponent charge = electrons["charge"][RecordComponent::SCALAR];
sliceParticles(series, charge);
}
sliceField(series, iter);

sliceParticles(series, iter);
}
if (currPatterns.size() > 1)
m_Pattern = 0;
}

/*
* Read a slice of particles
* Read a slice of id of the first particle
*
* @param series openPMD Series
* @param charge Particle record
* @param iter current iteration
*
*/
void sliceParticles(Series& series, RecordComponent& charge)
void sliceParticles(Series& series, IndexedIteration& iter)
{
Extent pExtent = charge.getExtent();
// read id of the first particle found
if ( m_Pattern != 7 )
return;

if ( 0 == iter.particles.size() )
{
if ( 0 == m_MPIRank )
std::cerr << " No Particles found. Skipping particle slicing. " << std::endl;
return;
}

openPMD::ParticleSpecies p = iter.particles.begin()->second;
RecordComponent idVal = p["id"][RecordComponent::SCALAR];

Extent pExtent = idVal.getExtent();

auto blob = pExtent[0]/(10*m_MPISize);
if (0 == blob)
Expand All @@ -689,7 +715,7 @@ class TestInput

Offset colOff = {m_MPIRank*blob};
Extent colExt = {blob};
auto col_data = charge.loadChunk<double>(colOff, colExt);
auto col_data = idVal.loadChunk<uint64_t>(colOff, colExt);
series.flush();
}

Expand Down Expand Up @@ -725,6 +751,7 @@ main( int argc, char *argv[] )
return 0;
}

{
Timer g( " Main ", input.m_MPIRank );

std::string prefix = argv[1];
Expand Down Expand Up @@ -759,7 +786,7 @@ main( int argc, char *argv[] )
input.m_Backend = which;
input.run(prefix);
}

} // Timer g
MPI_Finalize();

return 0;
Expand Down

0 comments on commit 9325daa

Please sign in to comment.