The MCKL library provides a framework for implementing Monte Carlo algorithms. It was originally developed for the purpose of parallel implementation of SMC algorithms. But it now supports other algorithms as well. The library provides basic building blocks for implementing parallelizable Monte Carlo algorithms.
This is a header only template C++ library. To install the library just move
the contents of the include directory into a proper place, e.g.,
/usr/local/include on Unix-alike systems.
The documents are hosted by Read the Docs. To generate the documents, the following is required: Doxygen, Sphinx and the RTD theme.
cd /path/to/mckl/source
cd docs
doxygen
make html
This library requires a working BLAS implementation. Some of the library's
functionalities can only be used if a optional dependencies are present.
Notably, HDF5,
TBB, OpenMP and
MKL. One can tell the library
that these optional features are available by defining configuration macros
such as -DMCKL_HAS_HDF5=1 during compilation.
In addition to the header-only library, one can optionally build and use a runtime library. To build the library, one need a recent version of NASM assembler and the CMake program. The library is only supported on x86-64 CPUs and 64-bit Unix-alike systems.
To build and install the library,
cd /path/to/mckl/source
mkdir -p build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=/prefix/to/install/directory
make lib
make install
To use and link to the library in a user program, say prog.cpp,
c++ -DMCKL_USE_ASM_LIB=1 -o prog prog.cpp -lmckl
The runtime library provides optimized implementation of selected RNGs Performance of the header-only implementation might vary considerably depending on the compiler. Though even the worst case, the performance is still better than most alternatives, the best performance of some RNGs can provide best performance regardless of the choice of compiler.
Further, one can define a configuration macro MCKL_USE_ASM_VMF to a non-zero
value, to use an assembly implementation of selected vectorized math functions
when MKL VML is unavailable. This feature is highly experimental. These
functions are fast, but have slightly lower accuracy for some values of input
than the standard library or MKL VML in high accuracy mode.
This library makes heavy use of some template metaprogramming techniques. It requires a C++14 standard conforming compiler.
This library has been regularly tested with recent clang, GCC and Intel C++ Compiler in C++14 mode.
Other compilers might work but are not tested. Complete C++14 implementation is required with thread-local storage as an exception.
Examples are in the example subdirectory, to build them,
export CXXFLAGS="-std=c++11"
cd /path/to/mckl/source
mkdir -p build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make example
Some examples may only be built if optional dependencies are present.
The MCKL library is distributed with a 2-clause BSD license which can be found
in the LICENSE file distributed with the source.