Install the platform » Historique » Révision 36
Révision 35 (Damien Gratadour, 10/11/2015 19:07) → Révision 36/62 (Damien Gratadour, 10/11/2015 19:08)
{{toc}}
h1. Install MAGMA
h2. Why MAGMA ?
The MAGMA project aims to develop a dense linear algebra library similar to LAPACK but for heterogeneous/hybrid architectures, starting with current "Multicore+GPU" systems.
Unlike CULA, MAGMA propose a dense linear algebra library handling double for free.
But MAGMA needs a LAPACK and a BLAS implementation. Actually, we try two options : openBLAS (free, easy to install) and MKL (free, need a registration but better optimized on Intel processors) more powerful)
h2. Dependencies : gfortran
Use your package manager to install dependencies:
* on scientific linux : yum install gcc-gfortran libgfortran
* on debian : apt-get install gfortran gfortran-multilib
h2. Configure MAGMA with openBLAS
h3. Dependencies : openblas (http://www.openblas.net)
First, clone the GIT repository:
<pre>
git clone https://github.com/xianyi/OpenBLAS.git
</pre>
compile it:
<pre>
cd OpenBLAS/
make
</pre>
install it:
<pre>
sudo make install PREFIX=/usr/local/openblas-haswellp-r0.2.14.a
</pre>
add to you .bashrc:
<pre>
export OPENBLAS_ROOT=/usr/local/openblas-haswellp-r0.2.14.a
</pre>
h3. extraction
MAGMA is available here : http://icl.cs.utk.edu/magma/software/index.html
extract the tgz file and go into the new directory
> ~$ tar xf magma-1.7.0-b.tar.gz
> ~$ cd magma-1.7.0
h3. configuration
You have to create your own make.inc based on make.inc.openblas:
example : *please verify GPU_TARGET, LAPACKDIR, ATLASDIR, CUDADIR*
<pre><code class="Makefile">
#//////////////////////////////////////////////////////////////////////////////
# -- MAGMA (version 1.7.0) --
# Univ. of Tennessee, Knoxville
# Univ. of California, Berkeley
# Univ. of Colorado, Denver
# @date September 2015
#//////////////////////////////////////////////////////////////////////////////
# GPU_TARGET contains one or more of Tesla, Fermi, or Kepler,
# to specify for which GPUs you want to compile MAGMA:
# Tesla - NVIDIA compute capability 1.x cards (no longer supported in CUDA 6.5)
# Fermi - NVIDIA compute capability 2.x cards
# Kepler - NVIDIA compute capability 3.x cards
# The default is "Fermi Kepler".
# See http://developer.nvidia.com/cuda-gpus
#
GPU_TARGET ?= Kepler
# --------------------
# programs
CC = gcc
CXX = g++
NVCC = nvcc
FORT = gfortran
ARCH = ar
ARCHFLAGS = cr
RANLIB = ranlib
# --------------------
# flags
# Use -fPIC to make shared (.so) and static (.a) library;
# can be commented out if making only static library.
FPIC = -fPIC
CFLAGS = -O3 $(FPIC) -DADD_ -Wall -fopenmp
FFLAGS = -O3 $(FPIC) -DADD_ -Wall -Wno-unused-dummy-argument
F90FLAGS = -O3 $(FPIC) -DADD_ -Wall -Wno-unused-dummy-argument -x f95-cpp-input
NVCCFLAGS = -O3 -DADD_ -Xcompiler "$(FPIC)"
LDFLAGS = $(FPIC) -fopenmp
# --------------------
# libraries
# gcc with OpenBLAS (includes LAPACK)
LIB = -lopenblas
LIB += -lcublas -lcudart
# --------------------
# directories
# define library directories preferably in your environment, or here.
OPENBLASDIR = /usr/local/openblas-haswellp-r0.2.14.a
CUDADIR = /usr/local/cuda
-include make.check-openblas
-include make.check-cuda
LIBDIR = -L$(CUDADIR)/lib64 \
-L$(OPENBLASDIR)/lib
INC = -I$(CUDADIR)/include
</code></pre>
h2. Configure MAGMA with MKL
h3. extraction
To download MKL, you have to create a account here : https://registrationcenter.intel.com/RegCenter/NComForm.aspx?ProductID=1517
extract l_ccompxe_2013_sp1.1.106.tgz and go into l_ccompxe_2013_sp1.1.106
install it with ./install_GUI.sh and add IPP stuff to default choices
h3. configuration
You have to create your own make.inc based on make.inc.mkl-gcc-ilp64:
example: *please verify GPU_TARGET, MKLROOT, CUDADIR*
<pre><code class="Makefile">
#//////////////////////////////////////////////////////////////////////////////
# -- MAGMA (version 1.7.0) --
# Univ. of Tennessee, Knoxville
# Univ. of California, Berkeley
# Univ. of Colorado, Denver
# @date September 2015
#//////////////////////////////////////////////////////////////////////////////
# GPU_TARGET contains one or more of Tesla, Fermi, or Kepler,
# to specify for which GPUs you want to compile MAGMA:
# Tesla - NVIDIA compute capability 1.x cards (no longer supported in CUDA 6.5)
# Fermi - NVIDIA compute capability 2.x cards
# Kepler - NVIDIA compute capability 3.x cards
# The default is "Fermi Kepler".
# See http://developer.nvidia.com/cuda-gpus
#
#GPU_TARGET ?= Fermi Kepler
# --------------------
# programs
CC = gcc
CXX = g++
NVCC = nvcc
FORT = gfortran
ARCH = ar
ARCHFLAGS = cr
RANLIB = ranlib
# --------------------
# flags
# Use -fPIC to make shared (.so) and static (.a) library;
# can be commented out if making only static library.
FPIC = -fPIC
CFLAGS = -O3 $(FPIC) -DADD_ -Wall -Wshadow -fopenmp -DMAGMA_WITH_MKL
FFLAGS = -O3 $(FPIC) -DADD_ -Wall -Wno-unused-dummy-argument
F90FLAGS = -O3 $(FPIC) -DADD_ -Wall -Wno-unused-dummy-argument -x f95-cpp-input
NVCCFLAGS = -O3 -DADD_ -Xcompiler "$(FPIC) -Wall -Wno-unused-function"
LDFLAGS = $(FPIC) -fopenmp
# Defining MAGMA_ILP64 or MKL_ILP64 changes magma_int_t to int64_t in include/magma_types.h
CFLAGS += -DMKL_ILP64
FFLAGS += -fdefault-integer-8
F90FLAGS += -fdefault-integer-8
NVCCFLAGS += -DMKL_ILP64
# Options to do extra checks for non-standard things like variable length arrays;
# it is safe to disable all these
CFLAGS += -pedantic -Wno-long-long
#CFLAGS += -Werror # uncomment to ensure all warnings are dealt with
CXXFLAGS := $(CFLAGS) -std=c++98
CFLAGS += -std=c99
# --------------------
# libraries
# IMPORTANT: this link line is for 64-bit int !!!!
# For regular 64-bit builds using 64-bit pointers and 32-bit int,
# use the lp64 library, not the ilp64 library. See make.inc.mkl-gcc or make.inc.mkl-icc.
# see MKL Link Advisor at http://software.intel.com/sites/products/mkl/
# gcc with MKL 10.3, Intel threads, 64-bit int
# note -DMAGMA_ILP64 or -DMKL_ILP64, and -fdefault-integer-8 in FFLAGS above
LIB = -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -lpthread -lstdc++ -lm -liomp5 -lgfortran
LIB += -lcublas -lcudart
# --------------------
# directories
# define library directories preferably in your environment, or here.
# for MKL run, e.g.: source /opt/intel/composerxe/mkl/bin/mklvars.sh intel64
#MKLROOT ?= /opt/intel/composerxe/mkl
#CUDADIR ?= /usr/local/cuda
-include make.check-mkl
-include make.check-cuda
LIBDIR = -L$(CUDADIR)/lib64 \
-L$(MKLROOT)/lib/intel64
INC = -I$(CUDADIR)/include \
-I$(MKLROOT)/include
</code></pre>
In this example, I use gcc but with MKL, you can use icc instead of gcc. In this case, you have to compile yorick with icc. For this, you have to change the CC flag in Make.cfg
h2. compilation and installation
h3. compilation
just compile the shared target (and test if you want)
> ~$ make -j 8 shared sparse
h3. installation
To install libraries and include files in a given prefix, run:
> ~$ make install prefix=/usr/local/magma
The default prefix is /usr/local/magma. You can also set prefix in make.inc.
h3. tune (not tested)
For multi-GPU functions, set $MAGMA_NUM_GPUS to set the number of GPUs to use.
For multi-core BLAS libraries, set $OMP_NUM_THREADS or $MKL_NUM_THREADS or $VECLIB_MAXIMUM_THREADS to set the number of CPU threads, depending on your BLAS library.
h1. Install the platform
The COMPASS platform is distributed as a single bundle of CArMA and SuTrA libraries and NAGA & SHESHA and its extensions for Python.
h2. Hardware requirements
The system must contain at least an x86 CPU and a CUDA capable GPU. list of compatible GPUs can be found here http://www.nvidia.com/object/cuda_gpus.html. Specific requirements apply to clusters (to be updated).
h2. Environment requirements
The system must be running a 64 bit distribution of Linux with the latest NVIDIA drivers and "CUDA toolkit":https://developer.nvidia.com/cuda-downloads. The following installation instructions are valid if the default installation paths have been selected for these components.
Additionally, to benefit from the user-oriented features of the platform, Anaconda should be installed.
In the last versions of compass (r608+), Yorick is no more supported.
For the widget, you also need pyQTgraph. You can install it like this :
<pre>
pip install pyqtgraph
</pre>
h2. Installation process
First check out the latest version from the svn repository :
<pre>
svn co https://version-lesia.obspm.fr/repos/compass compass
</pre>
then go in the newly created directory and then trunk:
<pre>
cd compass/trunk
</pre>
once there, you need to modify system variables in our .bashrc :
<pre>
# CUDA default definitions
export CUDA_ROOT=$CUDA_ROOT #/usr/local/cuda
export CUDA_INC_PATH=$CUDA_ROOT/include
export CUDA_LIB_PATH=$CUDA_ROOT/lib
export CUDA_LIB_PATH_64=$CUDA_ROOT/lib64
export CPLUS_INCLUDE_PATH=$CUDA_INC_PATH
export PATH=$CUDA_ROOT/bin:$PATH
</pre>
in this file, you also have to indicate the proper architecture of your GPU so as the compiler will generate the appropriate code.
<pre>
export GENCODE="arch=compute_52,code=sm_52"
</pre>
and change both 52 to your architecture : for instance a Tesla Fermi will have 2.0 computing capabilities so change 52 to 20, a Kepler GPU will have 3.0 or 3.5 (K20) computing capabilities, change 52 to 30 (or 35), a Maxwell GPU have 5.2 (M6000).
If you are using CULA, you have to specify it:
<pre>
# CULA default definitions
export CULA_ROOT= /usr/local/cula
export CULA_INC_PATH= $CULA_ROOT/include
export CULA_LIB_PATH= $CULA_ROOT/lib
export CULA_LIB_PATH_64= $CULA_ROOT/lib64
</pre>
If you are using MAGMA, you have to specify it:
<pre>
# MAGMA definitions (uncomment this line if MAGMA is installed)
export MAGMA_ROOT=/usr/local/magma
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$MAGMA_ROOT/lib
export PKG_CONFIG_PATH=$MAGMA_ROOT/lib/pkgconfig
</pre>
Last variables to define:
<pre>
export COMPASS_ROOT=/path/to/compass/trunk
export NAGA_ROOT=$COMPASS_ROOT/naga
export SHESHA_ROOT=$COMPASS_ROOT/shesha
export LD_LIBRARY_PATH=$COMPASS_ROOT/libcarma:$COMPASS_ROOT/libsutra:$CUDA_LIB_PATH_64:$CUDA_LIB_PATH:$CULA_LIB_PATH_64:$CULA_LIB_PATH:$LD_LIBRARY_PATH
</pre>
Once this is done, you're ready to compile the whole library:
<pre>
make clean install
</pre>
If you did not get any error, CArMA, SuTrA, NAGA and SHESHA are now installed on your machine. You can check that everything is working by launching a GUI to test a simulation:
<pre>
cd $SHESHA_ROOT/widgets && ipython -i widget_ao.py
</pre>