StarPU Handbook
|
One of the StarPU developers being a Debian Developer, the packages are well integrated and very uptodate. To see which packages are available, simply type:
$ apt-cache search starpu
To install what you need, type for example:
$ sudo apt-get install libstarpu-1.3 libstarpu-dev
StarPU can be built and installed by the standard means of the GNU autotools. The following chapter is intended to briefly remind how these tools can be used to install StarPU.
The hwloc
(http://www.open-mpi.org/software/hwloc) topology discovery library is not mandatory to use StarPU but strongly recommended. It allows for topology aware scheduling, which improves performance. hwloc
is available in major free operating system distributions, and for most operating systems. Make sure to not only install a hwloc
or libhwloc
package, but also hwloc-devel
or libhwloc-dev
so as to have hwloc headers etc.
If libhwloc
is installed in a standard location, no option is required, it will be detected automatically, otherwise --with-hwloc=<directory> should be used to specify its location.
If libhwloc
is not available on your system, the option --without-hwloc should be explicitely given when calling the script configure
.
StarPU's sources can be obtained from the download page of the StarPU website (https://starpu.gitlabpages.inria.fr/files/).
All releases and the development tree of StarPU are freely available on StarPU SCM server under the LGPL license. Some releases are available under the BSD license.
The latest release can be downloaded from the StarPU download page (https://starpu.gitlabpages.inria.fr/files/).
The latest nightly snapshot can be downloaded from the StarPU website (https://starpu.gitlabpages.inria.fr/files/testing/).
And finally, current development version is also accessible via git. It should only be used if you need the very latest changes (i.e. less than a day old!).
$ git clone git@gitlab.inria.fr:starpu/starpu.git
Running autogen.sh
is not necessary when using the tarball releases of StarPU. However when using the source code from the git repository, you first need to generate the script configure
and the different Makefiles. This requires the availability of autoconf
and automake
>= 2.60.
$ ./autogen.sh
You then need to configure StarPU. Details about options that are useful to give to configure
are given in Compilation Configuration.
$ ./configure
If configure
does not detect some software or produces errors, please make sure to post the contents of the file config.log
when reporting the issue.
By default, the files produced during the compilation are placed in the source directory. As the compilation generates a lot of files, it is advised to put them all in a separate directory. It is then easier to cleanup, and this allows to compile several configurations out of the same source tree. To do so, simply enter the directory where you want the compilation to produce its files, and invoke the script configure
located in the StarPU source directory.
$ mkdir build $ cd build $ ../configure
By default, StarPU will be installed in /usr/local/bin
, /usr/local/lib
, etc. You can specify an installation prefix other than /usr/local
using the option –prefix
, for instance:
$ ../configure --prefix=$HOME/starpu
$ make
Once everything is built, you may want to test the result. An extensive set of regression tests is provided with StarPU. Running the tests is done by calling make check
. These tests are run every night and the result from the main profile is publicly available (https://starpu.gitlabpages/files/testing/master/).
$ make check
In order to install StarPU at the location which was specified during configuration:
$ make install
If you have let StarPU install in /usr/local/
, you additionally need to run
$ sudo ldconfig
so the libraries can be found by the system.
Libtool interface versioning information are included in libraries names (libstarpu-1.3.so
, libstarpumpi-1.3.so
and libstarpufft-1.3.so
).
StarPU provides a pkg-config
executable to obtain relevant compiler and linker flags. As compiling and linking an application against StarPU may require to use specific flags or libraries (for instance CUDA
or libspe2
).
If StarPU was not installed at some standard location, the path of StarPU's library must be specified in the environment variable PKG_CONFIG_PATH
to allow pkg-config
to find it. For example if StarPU was installed in $STARPU_PATH
:
$ export PKG_CONFIG_PATH=$PKG_CONFIG_PATH:$STARPU_PATH/lib/pkgconfig
The flags required to compile or link against StarPU are then accessible with the following commands:
$ pkg-config --cflags starpu-1.3 # options for the compiler $ pkg-config --libs starpu-1.3 # options for the linker
Note that it is still possible to use the API provided in the version 1.0 of StarPU by calling pkg-config
with the starpu-1.0
package. Similar packages are provided for starpumpi-1.0
and starpufft-1.0
. It is also possible to use the API provided in the version 0.9 of StarPU by calling pkg-config
with the libstarpu
package. Similar packages are provided for libstarpumpi
and libstarpufft
.
Make sure that pkg-config –libs starpu-1.3
actually produces some output before going further: PKG_CONFIG_PATH
has to point to the place where starpu-1.3.pc
was installed during make install
.
Also pass the option –static
if the application is to be linked statically.
It is also necessary to set the environment variable LD_LIBRARY_PATH
to locate dynamic libraries at runtime.
$ export LD_LIBRARY_PATH=$STARPU_PATH/lib:$LD_LIBRARY_PATH
And it is useful to get access to the StarPU tools:
$ export PATH=$PATH:$STARPU_PATH/bin
It is then useful to check that StarPU executes correctly and finds your hardware:
$ starpu_machine_display
If it does not, please check the output of lstopo
from hwloc
and report the issue to the hwloc
project, since this is what StarPU uses to detect the hardware.
A tool is provided to help setting all the environment variables needed by StarPU. Once StarPU is installed in a specific directory, calling the script bin/starpu_env
will set in your current environment the variables STARPU_PATH
, LD_LIBRARY_PATH
, PKG_CONFIG_PATH
, PATH
and MANPATH
.
$ source $STARPU_PATH/bin/starpu_env
When using a Makefile, the following lines can be added to set the options for the compiler and the linker:
CFLAGS += $$(pkg-config --cflags starpu-1.3) LDLIBS += $$(pkg-config --libs starpu-1.3)
If you have a test-starpu.c
file containing for instance:
You can build it with make test-starpu
and run it with ./test-starpu
This section shows a minimal example integrating StarPU in an existing application's CMake build system.
Let's assume we want to build an executable from the following source code using CMake:
The CMakeLists.txt
file below uses the Pkg-Config support from CMake to autodetect the StarPU installation and library dependences (such as libhwloc
) provided that the PKG_CONFIG_PATH
variable is set, and is sufficient to build a statically-linked executable. This example has been successfully tested with CMake 3.2, though it may work with earlier CMake 3.x versions.
The following CMakeLists.txt
implements an alternative, more complex strategy, still relying on Pkg-Config, but also taking into account additional flags. While more complete, this approach makes CMake's build types (Debug, Release, ...) unavailable because of the direct affectation to variable CMAKE_C_FLAGS
. If both the full flags support and the build types support are needed, the CMakeLists.txt
below may be altered to work with CMAKE_C_FLAGS_RELEASE
, CMAKE_C_FLAGS_DEBUG
, and others as needed. This example has been successfully tested with CMake 3.2, though it may work with earlier CMake 3.x versions.
Basic examples using StarPU are built in the directory examples/basic_examples/
(and installed in $STARPU_PATH/lib/starpu/examples/
). You can for example run the example vector_scal
.
$ ./examples/basic_examples/vector_scal BEFORE: First element was 1.000000 AFTER: First element is 3.140000
When StarPU is used for the first time, the directory $STARPU_HOME/.starpu/
is created, performance models will be stored in this directory (STARPU_HOME).
Please note that buses are benchmarked when StarPU is launched for the first time. This may take a few minutes, or less if libhwloc
is installed. This step is done only once per user and per machine.
Batch files are provided to run StarPU applications under Microsoft Visual C. They are installed in $STARPU_PATH/bin/msvc
.
To execute a StarPU application, you first need to set the environment variable STARPU_PATH.
c:\....> cd c:\cygwin\home\ci\starpu\ c:\....> set STARPU_PATH=c:\cygwin\home\ci\starpu\ c:\....> cd bin\msvc c:\....> starpu_open.bat starpu_simple.c
The batch script will run Microsoft Visual C with a basic project file to run the given application.
The batch script starpu_clean.bat
can be used to delete all compilation generated files.
The batch script starpu_exec.bat
can be used to compile and execute a StarPU application from the command prompt.
c:\....> cd c:\cygwin\home\ci\starpu\ c:\....> set STARPU_PATH=c:\cygwin\home\ci\starpu\ c:\....> cd bin\msvc c:\....> starpu_exec.bat ..\..\..\..\examples\basic_examples\hello_world.c
MSVC StarPU Execution ... /out:hello_world.exe ... Hello world (params = {1, 2.00000}) Callback function got argument 0000042 c:\....>
StarPU automatically binds one thread per CPU core. It does not use SMT/hyperthreading because kernels are usually already optimized for using a full core, and using hyperthreading would make kernel calibration rather random.
Since driving GPUs is a CPU-consuming task, StarPU dedicates one core per GPU.
While StarPU tasks are executing, the application is not supposed to do computations in the threads it starts itself, tasks should be used instead.
If the application needs to reserve some cores for its own computations, it can do so with the field starpu_conf::reserve_ncpus, get the core IDs with starpu_get_next_bindid(), and bind to them with starpu_bind_thread_on().
Another option is for the application to pause StarPU by calling starpu_pause(), then to perform its own computations, and then to resume StarPU by calling starpu_resume() so that StarPU can execute tasks.
When both CUDA and OpenCL drivers are enabled, StarPU will launch an OpenCL worker for NVIDIA GPUs only if CUDA is not already running on them. This design choice was necessary as OpenCL and CUDA can not run at the same time on the same NVIDIA GPU, as there is currently no interoperability between them.
To enable OpenCL, you need either to disable CUDA when configuring StarPU:
$ ./configure --disable-cuda
or when running applications:
$ STARPU_NCUDA=0 ./application
OpenCL will automatically be started on any device not yet used by CUDA. So on a machine running 4 GPUS, it is therefore possible to enable CUDA on 2 devices, and OpenCL on the 2 other devices by doing so:
$ STARPU_NCUDA=2 ./application
Some interesting benchmarks are installed among examples in $STARPU_PATH/lib/starpu/examples/
. Make sure to try various schedulers, for instance STARPU_SCHED=dmda
.
This benchmark gives a glimpse into how long a task should be (in µs) for StarPU overhead to be low enough to keep efficiency. Running tasks_size_overhead.sh
generates a plot of the speedup of tasks of various sizes, depending on the number of CPUs being used.
local_pingpong
performs a ping-pong between the first two CUDA nodes, and prints the measured latency.
sgemm
and dgemm
perform a blocked matrix-matrix multiplication using BLAS and cuBLAS. They output the obtained GFlops.
cholesky_*
perform a Cholesky factorization (single precision). They use different dependency primitives.
lu_*
perform an LU factorization. They use different dependency primitives.
It can also be convenient to try simulated benchmarks, if you want to give a try at CPU-GPU scheduling without actually having a GPU at hand. This can be done by using the SimGrid version of StarPU: first install the SimGrid simulator from http://simgrid.gforge.inria.fr/ (we tested with SimGrid from 3.11 to 3.16, and 3.18 to 3.30. SimGrid versions 3.25 and above need to be configured with -Denable_msg=ON
. Other versions may have compatibility issues, 3.17 notably does not build at all. MPI simulation does not work with version 3.22). Then configure StarPU with --enable-simgrid and rebuild and install it, and then you can simulate the performance for a few virtualized systems shipped along StarPU: attila, mirage, idgraf, and sirocco.
For instance:
$ export STARPU_PERF_MODEL_DIR=$STARPU_PATH/share/starpu/perfmodels/sampling $ export STARPU_HOSTNAME=attila $ $STARPU_PATH/lib/starpu/examples/cholesky_implicit -size $((960*20)) -nblocks 20
Will show the performance of the cholesky factorization with the attila system. It will be interesting to try with different matrix sizes and schedulers.
Performance models are available for cholesky_*
, lu_*
, *gemm
, with block sizes 320, 640, or 960 (plus 1440 for sirocco), and for stencil
with block size 128x128x128, 192x192x192, and 256x256x256.
Read the chapter SimGrid Support for more information on the SimGrid support.