src/picongpu/include/plugins/Radiation.hpp:781:14:
  Warnung: Vergleich zwischen vorzeichenbehafteten und vorzeichenlosen
  Ganzzahlausdrücken [-Wsign-compare]

@PrometheusPi See here

List dependencies ( apt-get install ) for Debian/Ubuntu.

Write a Readme file for each example.
The Readme should contain

• a description
• tests to run (on a cluster)
• commands for post processing
• files to compare (within a separate git repository)

Progress:

documentation / tests

• [x] Readme template (@ax3l)
• [ ] Bunch (@PrometheusPi)
• [ ] LaserWakefield (@ax3l)
• [ ] SingleParticleTest (@ax3l)
• [ ] ThermalTest (@Heikman)
• [ ] KelvinHelmholtz (@ax3l)
• [ ] SingleParticleCurrent (@Heikman)
• [ ] WeibelTransverse (@ax3l)
• [ ] SingleParticleRadiationWithLaser (@PrometheusPi)

Build a plugin for a single rcf stack

• transversal 2D plane with scalar energy information with restriction to the two transversal aperture angles
• options:
  • energyMin/Max range
  • stack: normal vector, spatial bins (number and size), distance
• rcf multi: allows defining the above plugin multiple times

• Check if the problems in the LiveVisualization with Density and EnergyDensity came from the Particle2Cell algorithms or from wrong usage (Note: could be caused by #174 setValue bug)
• Check Phase Space Plugin stability

Gather the local arrays of the LinceSliceFields plugin before dumping it to a file.
Write in a parallel hdf5 file.

Btw: may we can find a more intuitive name for that plugin...

Add missing meta data:

• GRID data needs self-descriptive edge lengths, too
  • example: vacuum simulations -> no info on CELL_WIDTH, ... so far
• POLYDATA: unit for momentum is wrong
  • dimension: velocity * mass
  • unit: UNIT_SPEED * UNIT_MASS
• time step: DELTA_T
• maybe add the seven base units (see output)
  • UNIT_LENGTH, UNIT_MASS, UNIT_TIME, UNIT_CHARGE/UNIT_CURRENT, UNIT_TEMPERATURE
  • skip luminous intensity and amount of substance ?
• ...

• add git support to tool addLicense
• delete white spaces in license text before end of line

Species (Electrons and Ions) should select their participation in algorithm XYZ in a parameter file, but not by a #define.

The definition of additional attributes should be automatically triggered, e.g. by the radiation plugin, but a user should have the possibility to manually add this attribute even without an algorithm requiring it.

Available attributes will be provided by PIConGPU, plugins can select out of them. (If an attribute is missing, a plugin developer should propose a new particle attributes to PIConGPU in general, so other plugins can use it too).

Example:

• Electrons: Vay Push + Radiation + momentum_mt1(*) + localB + localE + shape:PSQ
• Ions: Boris Push + momentum_mt1+ shape:TSC

(*) implicit added by Radiation; if selected explicit, add only once

Comment:

• localB, localE could be attributes remembering the last E and B field a particle saw during its push (this could also be time averaged over N steps)

There should be only one singleton (Environment) in PIC. All previous singletons can only be accessed via Environment.

This allows to implement a stable init/deinit point and order.
We would have a stable point where to put cudaDeviceReset.

We are going to change our file I/O strategy to parallel hdf5.

Our I/O library libSplash is undergoing an update to add this feature.

Major updates:

• 1 parallel written file per timestep (now: n files for all timesteps)

Rewrite gbAnalysis tools for parallel splash? :

• Algorithmic building blocks like forEachCell (fields) and forEachParticle ( parallel c++ analysis interfaces for mpi)
• splash2txt -> move back to libSplash

Later (1.0?):

• New high level interfaces for equal spaced field data and particle data

The configure script, called with a invalid parameter like configure ../pathTypo/, should either give a warning about a non-existent path or abort.

Pretending the path in $1 is valid and compiling the default all-disabled example is not a good idea.

PIC does NOT compile with gcc 4.7.1 (tested on machine joker).
Compile error is:
[ 83%] Built target cuda_memtest /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/ExchangeMapping.hpp: In instantiation of ‘PMacc::DataSpace<DIM> PMacc::ExchangeMapping<areaType, baseClass<DIM, SuperCellSize_> >::getGridDim() [with unsigned int areaType = 4u; baseClass = PMacc::MappingDescription; unsigned int DIM = 3u; SuperCellSize_ = PMacc::TVec<8u, 8u, 4u>]’: /home.local/others/fschmitt/pic/picongpu/src/picongpu/include/fields/FieldJ.tpp:120:31: required from here /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/ExchangeMapping.hpp:97:87: error: ‘reduce’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/ExchangeMapping.hpp:97:87: note: declarations in dependent base ‘PMacc::CudaGridDimRestrictions<3u>’ are not found by unqualified lookup /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/ExchangeMapping.hpp:97:87: note: use ‘this->reduce’ instead /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/AreaMapping.hpp: In instantiation of ‘PMacc::DataSpace<DIM> PMacc::AreaMapping<areaType, baseClass<DIM, SuperCellSize_> >::getGridDim() [with unsigned int areaType = 3u; baseClass = PMacc::MappingDescription; unsigned int DIM = 3u; SuperCellSize_ = PMacc::TVec<8u, 8u, 4u>]’: /home.local/others/fschmitt/pic/picongpu/src/picongpu/include/plugins/SumCurrents.hpp:190:246: required from here /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/AreaMapping.hpp:72:98: error: ‘reduce’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/AreaMapping.hpp:72:98: note: declarations in dependent base ‘PMacc::CudaGridDimRestrictions<3u>’ are not found by unqualified lookup /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/AreaMapping.hpp:72:98: note: use ‘this->reduce’ instead /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/StrideMapping.hpp: In instantiation of ‘PMacc::DataSpace<DIM> PMacc::StrideMapping<areaType, stride, baseClass<DIM, SuperCellSize_> >::getGridDim() [with unsigned int areaType = 3u; unsigned int stride = 3u; baseClass = PMacc::MappingDescription; unsigned int DIM = 3u; SuperCellSize_ = PMacc::TVec<8u, 8u, 4u>]’: /home.local/others/fschmitt/pic/picongpu/src/picongpu/include/fields/FieldJ.tpp:212:125: required from ‘void picongpu::FieldJ::computeCurrent(ParticlesClass&, uint32_t) [with unsigned int AREA = 3u; ParticlesClass = picongpu::Particles<boost::mpl::vector2<picongpu::ElectronMethods<>, picongpu::ParticlesData<> > >; uint32_t = unsigned int]’ /home.local/others/fschmitt/pic/picongpu/src/picongpu/include/simulationControl/MySimulation.hpp:372:131: required from here /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/StrideMapping.hpp:75:114: error: ‘reduce’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/StrideMapping.hpp:75:114: note: declarations in dependent base ‘PMacc::CudaGridDimRestrictions<3u>’ are not found by unqualified lookup /home.local/others/fschmitt/pic/picongpu/src/libPMacc/include/mappings/kernel/StrideMapping.hpp:75:114: note: use ‘this->reduce’ instead

Add new memory controller.

Tasks:

• Patch bugs in mem controller
• report fixed version back to the original developers (they will be happy about that)
• select the memory controller to use in memory.param (preferred) or by macro
  • CUDA native new (required by GPUOcelot and good for comparisons)
  • new mem controller (good for NVidia GPUs)
  • old ring buffer (if still possible)
  • ...

The correct spot seems to be
pmacc/inlcude/eventSystem/streams/StreamController.hpp:81 (end of destructor).

cudaDeviceReset is necessary to support the CUDA 5.5 CUPTI interface.

Add Posix signal support as described in our doc/SIGNALS.md

We will probably need a more general Simulation Control class to get the communication between the signal handler (thread?), plugins (signal reduce/sync) and the simulation itself clean.

Add a travis-ci.org like status checker for all commits and pull requests.

• Merge test -> done by GitHub itself
• Compile tests -> should be done automatically by our Compile Suite and should set a status [v3 API]
• Run time tests -> preparation see #7

All files that may be submitted in any form to the batch system must be ASCII to work properly on all machines. Currently, they are UTF-8 due to a special character in René. We might have to change this to Rene to get this working.

While using methods like writeAttribute with DCollector::Dimensions objects, one should use getPointer() instead of &:

&sim_size -> sim_size.getPointer()

DCollector::Dimensions sim_size( 1, 2, 3);
domainCollector.writeAttribute( currentStep,
                                DCollector::ColTypeDim(),
                                dataSetName.str().c_str(),
                                "sim_size",
                                sim_size.getPointer() );

It should be clear in the submit/.cfg files which variables are used and which are just examples and need to be added to e.g. TBG_programParams to be actually used.

example: TBG_movingWindow in 0016gpus.cfg.

Follow up to #40, in order of priority

• Restart from ADIOS files (currently working on: #679 @ax3l)
• ADIOS multi-species support (broken?) ok
• ADIOS 2D support (currently working on: no one #679)
• ADIOS particle dumps
• Add an ADIOS version of splash2txt (adios2txt)

Member inside a class must ordered by it size. First big member than smaller. If you don't do this you get a undefined behavior with sm_13 and slow byte by byte copies inside a kernel in higher architectures.
This can be the reason of compiler bug #76. IMO both bugs point to the same code lines.

ptx example with bad sm_35 code:

///home/widera/workspace/dev/src/libPMacc/include/lambda/Expression.hpp:63
        .loc 75 63 1
        ld.u8   %rc1, [%rd2];
        ld.u8   %rc2, [%rd2+1];
        ld.u8   %rc3, [%rd2+2];
        ld.u8   %rc4, [%rd2+3];
        ld.u8   %rc5, [%rd2+4];
        ld.u8   %rc6, [%rd2+5];
        ld.u8   %rc7, [%rd2+6];
        st.u8   [%rd1+6], %rc7;
        st.u8   [%rd1+5], %rc6;
        st.u8   [%rd1+4], %rc5;
        st.u8   [%rd1+3], %rc4;
        st.u8   [%rd1+2], %rc3;
        st.u8   [%rd1+1], %rc2;
        st.u8   [%rd1], %rc1;

All attribute writes with DCollector::ColTypeDim() attributes are broken.

One should use e.g.

dataCollector->writeAttribute(
    params->currentStep,
    DCollector::ColTypeDim(),
    str.str().c_str(),
    "sim_size",
    sim_size.getPointer());

instead of

dataCollector->writeAttribute(
    params->currentStep,
    DCollector::ColTypeDim(),
    str.str().c_str(),
    "sim_size",
    &sim_size);

See related libSplash bug 33.

• Translate, update and push old file doc/physical_inputs.txt -> wiki (describe .param structure as in beta-rc6 changelog)
• describe multi-species initalization workflow
• Fix SI:: vs _SI redundancy...
• "dot" file with dependency diagram

Output time-averaged fields, too.

• Reduce (add) selected field in given interval [nextDump - Navg ; nextDump] on host-side
• Should be possible for any field wie dump already (E,B,J,Tmp)
• Add to dump as new field (and divide value by number of timesteps in the interval)

Occurs with precision64Bit, e.g. with examples/LaserWakefield cmakePreset #5

En:

The ABI for passing parameters with 32-byte alignment has changed in GCC 4.6

De:

src/libPMacc/include/eventSystem/tasks/TaskSetValue.hpp:111:1:
Anmerkung: Das ABI der Parameterübergabe mit 32-Byte-Ausrichtung hat sich in GCC 4.6 geändert

related to alignment in src/libPMAcc/include/types.h __optimal_align__ for host-side data

References:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44948

We need to add " " here to prevent this error when compiling with vtc++:

"STREQUAL" "PGI"

Unknown arguments specified

Use hierarchical groups to better distinguish particles, fields, ghost cells, indices

Current idea:
fileset 1 (simulation data):

/data/timestamp/...

fields/FieldB/x
fields/FieldB/y

particles/electrons/momentum
particles/electrons/positions
particles/ions/

fileset 2 (simulation restart data):

as above + 
fields/FieldB/_ghosts/top/x
fields/FieldB/_ghosts/bottom/x

I just noticed that during a restart with moving window enabled, some of the first y-cells are not loaded in the X-Z plane.

This picture was taken some steps after the restart. At the time of the restart, the virtualWindow was exactly at the positon where the "gap" came from.

Size of the gap: 7px, scale to cellsize: true, dx/dy = 5.3125 -> ~37 cells or 4-5 super cells? (Standard super cell: 8x8x4).

Can someone reproduce that?

7000 (restart point):

7500:

Restart because of a crash at 7200 with:

terminate called after throwing an instance of '
std::runtime_error'
 what():  [CUDA] Error: unspecified launch failure` `

FieldJ.tpp:
if you set workerMultiplier to 1 everything is fine.

Plugins should describe their generated output in a common interface.

For example a plugin calculates:

• a simple number: unit, type of a single datum, dimension=0, no axis, name
• a 2D matrix (e.g. a slice of density data): unit, type of a single datum, optional axis for x, y, colorbar, dimension=2, name/title
• a histogram (e.g. for particle energies): unit, type of the bins, optional axis for x, y, dimension=1, name/title

Axis information should be optional, deciding either to specify:

• min/max range + lin/log scale (preferred if possible)
• n discrete numbers/ticks on that axis/"color"bar

Advantages:

• one can implement general outputs to txt, hdf5, anything, ... for n-dimensional data
• one can forward that data to an other plugin by that common interface, e.g. a live visualization plugin

• integrate ADIOS API in a new plugin for writing field data
• add an ADIOS xml description for a typical 3D field-only scenario

The PseudoBuffer in plugins/TotalDivJ.tpp:84 causes a warning when compiled with

• gcc/4.6.2
• openmpi/1.6.3
• cuda/5.0
• boost/1.49.0

on hypnos:

src/libPMacc/include/cuSTL/container/CartBuffer.tpp:
  In member function ‘virtual void picongpu::TotalDivJ::notify(uint32_t)’:
src/libPMacc/include/cuSTL/container/CartBuffer.tpp:154:1:
  warning: ‘*((void*)& fieldJ +8)’ may be used uninitialized in this
  function [-Wuninitialized]
src/picongpu/include/plugins/TotalDivJ.tpp:84:69:
  note: ‘*((void*)& fieldJ +8)’ was declared here

Pls fix me :)

Create a my first run example.

• Link to Install.md
• close to last section of the install.md but for a single desktop GPU?
• create, configure, make install
• tbg bash mpiexec for 1 GPU
• Add Run an Example Video
• add splash2txt example

Check out http://catalyst.paraview.org and what we can do with it.

Wiki: http://www.paraview.org/Wiki/ParaView/Catalyst/Overview
Webinar: http://vimeo.com/75793492
Examples: https://github.com/acbauer/CatalystExampleCode

Add to doc/PARTICIPATE.md.

• close namespaces with a comment: } // namespace itsName
• ...

Please add your drafts here ( do not push your drafts to the PARTICIPATE.md pls!)

Extend to multiple species (arbitrary number [0, N]). Depends on #109

it should either be possible to compile on a sm_13 system for sm_20+ or this should be prevented and a error message should be given to the user.

Plugins must be able to restart themselves.
The SimRestartInitializer is crap since this should be handled by the hdf5-plugin, method restart (another init). This restart does not only reload the state of the plugin itself, but may also affect the state of PIConGPU.

One may have to trigger a checkpoint like a notify to save the internal state, too.

An other trivial example:

Plugins that write to txt files should remove post-restart related data during restart and append to the already created files. (Similar how our hdf5 plugin works)

Add our LiveVisServer for 2D live previews.

I tried to use math::pow( float_X value, int) but did not find an implementation (on the device).

Could you please add it?
We will need at least the following C++98 overloads for optimal performance

float_X pow (float_X base, float_X exponent);
float_X pow (float_X base, int exponent);

Especially, it is not desired to cast the exponents from ints to floats since the standard algorithms for integer exponents can be performed much faster than for floating point exponents.

Thanks :)

Plugins that calculate

• total Energy
• energy histograms
• images
• data output
• slices
• phase space
• ...

should operate on the volume given by the VirtualWindow only.

Affects: plugins for moving window simulations.

Possible solution: #1288

If one enables -Wunused-parameter for GCC, one gets a warning for cuSTL/algorithm/kernel/Foreach.hpp during its usage with lambda parameters, for example in cuSTL/container/assigner/DeviceMemAssigner.hpp:62:

cuSTL/algorithm/kernel/Foreach.hpp:
In Instanziierung von »void PMacc::algorithm::kernel::detail::__wrapper__device_stub_kernelForeach(Mapper&, C0&, Functor&) [with Mapper = PMacc::algorithm::kernel::detail::SphericMapper<2, PMacc::math::CT::Int<1, 1, 1> >, C0 = PMacc::cursor::BufferCursor<int, 2>, Functor = PMacc::lambda::ExprFunctor<PMacc::lambda::Expression<PMacc::lambda::exprTypes::assign, boost::mpl::vector<PMacc::lambda::Expression<PMacc::lambda::exprTypes::terminal, boost::mpl::vector<PMacc::lambda::placeholder<0> > >, PMacc::lambda::Expression<PMacc::lambda::exprTypes::terminal, ... > >]«:

cuSTL/algorithm/kernel/Foreach.hpp:65:109:
instanziiert von »void PMacc::algorithm::kernel::detail::kernelForeach(Mapper, C0, Functor) [with Mapper = PMacc::algorithm::kernel::detail::SphericMapper<2, PMacc::math::CT::Int<1, 1, 1> >, C0 = PMacc::cursor::BufferCursor<int, 2>, Functor = PMacc::lambda::ExprFunctor<PMacc::lambda::Expression<PMacc::lambda::exprTypes::assign, boost::mpl::vector<PMacc::lambda::Expression<PMacc::lambda::exprTypes::terminal, boost::mpl::vector<PMacc::lambda::placeholder<0> > >, PMacc::lambda::Expression<PMacc::lambda::exprTypes::terminal, ... > >]«

cuSTL/algorithm/kernel/Foreach.hpp:91:414:
instanziiert von »void PMacc::algorithm::kernel::Foreach::operator()(const Zone&, C0, const Functor&) [with Zone = PMacc::zone::SphericZone<2>, C0 = PMacc::cursor::BufferCursor<int, 2>, Functor = PMacc::lambda::Expression<PMacc::lambda::exprTypes::assign, boost::mpl::vector<PMacc::lambda::Expression<PMacc::lambda::exprTypes::terminal, boost::mpl::vector<PMacc::lambda::placeholder<0> > >, PMacc::lambda::Expression<PMacc::lambda::exprTypes::terminal, ... >, BlockDim = PMacc::math::CT::Int<1, 1, 1>]«

cuSTL/container/assigner/DeviceMemAssigner.hpp:62:1:
instanziiert von »static void PMacc::assigner::DeviceMemAssigner<_dim>::assign(Type*, const PMacc::math::Size_t<(dim - 1)>&, const Type&, const PMacc::math::Size_t&) [with Type = int, int _dim = 2]«
cuSTL/container/CartBuffer.tpp:215:1: instanziiert von »void PMacc::container::CartBuffer<Type, _dim, Allocator, Copier, Assigner>::assign(const Type&) [with Type = int, int _dim = 2, Allocator = PMacc::allocator::DeviceMemAllocator<int, 2>, Copier = PMacc::copier::D2DCopier<2>, Assigner = PMacc::assigner::DeviceMemAssigner<2>]«

src/picongpu/include/plugins/ParticleDensity.tpp:128:1:
instanziiert von »void picongpu::heiko::ParticleDensity::notify(uint32_t) [with ParticlesType = picongpu::Particlesboost::mpl::vector2<picongpu::ElectronMethods<, picongpu::ParticlesData<> > >, uint32_t = unsigned int]«

/tmp/tmpxft_000043cb_00000000-3_main.cudafe1.stub.c:270:334:
instanziiert von hier

cuSTL/algorithm/kernel/Foreach.hpp:65:62: Warnung: unbenutzter Parameter »mapper« [-Wunused-parameter]
cuSTL/algorithm/kernel/Foreach.hpp:65:62: Warnung: unbenutzter Parameter »c0« [-Wunused-parameter]
cuSTL/algorithm/kernel/Foreach.hpp:65:62: Warnung: unbenutzter Parameter »functor« [-Wunused-parameter]

otherwise the output is malformed

Many batch systems support a flag if a job can be aborted and/or restarted. We may want to use that sometimes.

List of flags and their meaning

qsub/torque: Rerunable, fault_tolerant
...

assign does not work for at least the following buffer configuration:

container::DeviceBuffer<float_X, 2> dBuffer;
dBuffer = new container::DeviceBuffer<float_X, 2>( 128, 1024 );

icc 12.1 on taurus (module load intel/12.1 boost/1.54.0-intel12.1) fails since #42 was merged in eef746a

[ 16%] Building NVCC (Device) object build_picongpu/CMakeFiles/picongpu.dir//./picongpu_generated_main.cu.o
/sw/taurus/libraries/boost/1.54.0-intel12.1/include/boost/mpl/vector/aux_/preprocessed/plain/vector10.hpp(153): error: class "boost::mpl::vector<PMacc::lam
bda::placeholder<0>, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_
::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na>" has no member "item2"
          detected during:
            instantiation of class "boost::mpl::v_at<V, 2L> [with V=boost::mpl::vector<PMacc::lambda::placeholder<0>, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na>]" 
/sw/taurus/libraries/boost/1.54.0-intel12.1/include/boost/mpl/vector/aux_/at.hpp(69): here
            instantiation of class "boost::mpl::at_impl<boost::mpl::aux::vector_tag<n_>>::apply<Vector, N> [with n_=1L, Vector=boost::mpl::vector<PMacc::lambda::placeholder<0>, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na>, N=mpl_::long_<2L>]" 
/sw/taurus/libraries/boost/1.54.0-intel12.1/include/boost/mpl/at.hpp(43): here
            instantiation of class "boost::mpl::at_c<Sequence, N> [with Sequence=boost::mpl::vector<PMacc::lambda::placeholder<0>, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na>, N=2L]" 
/home/s4343183/src/picongpu/src/libPMacc/include/lambda/Expression.hpp(121): here
            instantiation of class "PMacc::lambda::Expression<_ExprType, _Childs> [with _ExprType=PMacc::lambda::exprTypes::terminal, _Childs=boost::mpl::vector<PMacc::lambda::placeholder<0>, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na, mpl_::na>]" 
/home/s4343183/src/picongpu/src/libPMacc/include/lambda/Expression.hpp(215): here
...

Replace all sys/stat.h calls to mkdir with

http://www.boost.org/doc/libs/1_54_0/libs/filesystem/doc/reference.html#create_directory

-> check how to set permissions to 0755 on *nix systems.