STAPL is a library, requiring only a C++ compiler, the Boost libraries, and established communication libraries such as MPI.
Compilers supported are:
- GCC 4.8.2 - 4.9.2
- Clang 3.6 - 3.7 (using libstdc++ for STL)
- Intel C++ 17 (using gcc compatability)
The version of Boost required is 1.63.
Compiling the STAPL library
Compiling the STAPL library requires setting the platform and stl variables. For example, on a Linux workstation with MPICH, Boost, and GCC 4.8.4 installed the library can be built by running
gmake platform=LINUX_gcc stl=./tools/libstdc++/4.8.4
The result of this command will be
Compiling STAPL programs
When compiling a STAPL program there are several flags required.
-D_STAPL: used in some components that are used in other Parasol projects to detect a STAPL application is being compiled rather than a sequential application
-I $(STAPL_ROOT)/$(stl): required so that versions of the STL and Boost
-I $(STAPL_ROOT)/tools: headers with define_type methods (used in serialization) are available
-I $(STAPL_ROOT)/: required so the stapl headers in stapl/ are available. Include statements are of the form
-I $(BOOST_ROOT)/include: STAPL requires Boost 1.56`
-DBOOST_RESULT_OF_USE_TR1_WITH_DECLTYPE_FALLBACK: Required by mechanism that detects work function return types
-std=c++11: STAPL requires c++11 support
-L$(STAPL_ROOT)/lib -lstapl -lrt: path to libstapl.a and the libraries required
-L$(BOOST_ROOT)/lib -lboost_serialization -lboost_thread -lboost_system: Boost libraries required by STAPL runtime. These options must come after
-lstaplin the compile line.
There are also several optional compile flags:
-DSTAPL_NDEBUG: disable runtime assertions, which can perform expensive computations as part of their check. This flag should be used when compiling executables for timing experiments.
-DSTAPL_PRODUCTION: enable single core optimizations
-DSTAPL_PER_VIEW_LOCALIZATION: enable independent localization of views passed to algorithms. This flag should be used when compiling executables for timing experiments.
-DSTAPL_RUNTIME_USE_OMP -fopenmp: explicitly specify runtime used for mixed-mode
To see a full command line cd to examples/statistics and run
gmake mean_standard_deviation. Note that in addition to the Boost libraries mentioned above, this test also requires the Boost program options library to link (i.e.,
Additional compiler options can be specified as part of the make command through the USER_CXXFLAGS and USER_LIB variables if the STAPL Makefiles are used.
Running STAPL programs
STAPL programs execute in mixed-mode at all times. To run MPI-only with a single thread per process you set
STAPL_NUM_THREADS to 1. If
STAPL_NUM_THREADS is not set execution will terminate.
Where to start
Please see the following resources:
February 2015 CERT Deep Dive meeting slides.
Example programs using STAPL. Each directory contains different types of simple examples. All Project Euler examples include a report by the student that developed the example including pseudocode, design with STAPL components, and a scaling study using a Cray system in the TAMU CSE department.
Tutorial guide to walk the reader through progressively more advanced uses of STAPL components. The
docs/tutorial_guide/directory includes the source code for those examples.
Implementations of benchmarks using STAPL. The benchmarks have been collected from various sources. For example, the runtime OSU benchmarks are STAPL ARMI implementations of the OSU microbenchmarks for message passing.
Each code directory has a GNUmakefile in it that is setup to allow compilation given that the directory hasn't been moved outside the working copy.
The STAPL development community can be found online at https://gitlab.com/parasol-lab/stapl . If you need more direct support please send email to firstname.lastname@example.org.