Projet

Général

Profil

Install the platform » Historique » Version 18

Arnaud Sevin, 08/01/2014 11:14

1 14 Arnaud Sevin
{{toc}}
2 1 Damien Gratadour
3 14 Arnaud Sevin
h1. Install the platform without MAGMA
4
5 1 Damien Gratadour
The COMPASS platform is distributed as a single bundle of CArMA and SuTrA libraries and YoGA and its AO extension for Yorick. 
6
7
h2. Hardware requirements
8 2 Damien Gratadour
9 10 Damien Gratadour
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).
10 1 Damien Gratadour
11
h2. Environment requirements
12 2 Damien Gratadour
13 11 Damien Gratadour
The system must be running a 64 bit distribution of Linux or Mac OS with the latest NVIDIA drivers and "CUDA toolkit":https://developer.nvidia.com/cuda-downloads. The installation of the corresponding version of the "CULA tools":http://www.culatools.com/downloads/dense/ is also required. The following installation instructions are valid if the default installation paths have been selected for these components.
14 1 Damien Gratadour
15 5 Damien Gratadour
Additionally, to benefit from the user-oriented features of the platform, Yorick should be installed as well as the latest version of Python and the associated pygtk module. 
16 1 Damien Gratadour
17 3 Damien Gratadour
To install Yorick, download the latest version from the github repository:
18
<pre>
19
git clone https://github.com/dhmunro/yorick.git yorick.git
20
</pre>
21
then cd onto the created directory and install:
22
<pre>
23
./configure && make && make install
24
</pre>
25
once Yorick is locally installed, you will have to add this directory : yorick.git/relocate/bin to your PATH to have an easy access to the yorick executable. You may want to add support for command history by using rlwrap and alias the yorick executable as :
26
<pre>
27 4 Damien Gratadour
alias yorick='rlwrap path_to_yorick_executable/yorick'
28 3 Damien Gratadour
</pre>
29 1 Damien Gratadour
30 3 Damien Gratadour
31 1 Damien Gratadour
h2. Installation process
32
33
First check out the latest version from the svn repository :
34
<pre>
35 3 Damien Gratadour
svn co https://version-lesia.obspm.fr/repos/compass compass
36 1 Damien Gratadour
</pre>
37
then go in the newly created directory and then trunk:
38
<pre>
39
cd compass/trunk
40
</pre>
41
once there, you need to modify system variables in the define_var.sh executable :
42
<pre>
43
emacs define_var.sh
44
</pre>
45
in this file define properly CUDA_ROOT, CULA_ROOT and YoGA path. Note that for the latter, as YoGA is distributed with SUTrA you should just point to the newly created trunk directory. On a Linux system you should normally have:
46
<pre>
47
export CUDA_ROOT=/usr/local/cuda
48
export CULA_ROOT=/usr/local/cula
49
export YOGA_DIR=/home/MyUserName/path2compass/trunk
50
</pre>
51 12 Damien Gratadour
in this file, you also have to indicate the proper architecture of your GPU so as the compiler will generate the appropriate code. Modify the following line:
52 13 Damien Gratadour
<pre>
53 12 Damien Gratadour
export GENCODE="arch=compute_12,code=sm_12"
54
</pre>
55
and change both 12 to your architecture : for instance a Tesla Fermi will have 3.0 computing capabilities so change 12 to 30, a Kepler GPU will have 3.5 computing capabilities, change 12 to 35
56 1 Damien Gratadour
57
Once this is done, you're ready to compile the whole library. First run define_var.sh to define the system variables that will be used during the compilation process:
58
<pre>
59
./define_var.sh
60
</pre>
61
62
then identify the absolute path to your Yorick executable using: 
63
<pre>
64
which yorick
65
</pre>
66
and run the compilation script:
67
<pre>
68 6 Damien Gratadour
./reinstall absolute_path_to_yorick
69 1 Damien Gratadour
</pre>
70
71 7 Damien Gratadour
If you did not get any error, CArMA, SuTrA and YoGA are now installed on your machine. You can check that everything is working by launching a GUI to test a simulation:
72 1 Damien Gratadour
<pre>
73
yorick -i yoga_ao/ywidgets/widget_ao.i
74
</pre>
75 14 Arnaud Sevin
76
h1. Install the platform with MAGMA
77
78
h2. Why MAGMA ?
79
80
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.
81
82
Unlike CULA, MAGMA propose a dense linear algebra library handling double for free.
83
84
But MAGMA needs a LAPACK and a BLAS implementation. Actually, we try two options : ATLAS BLAS (free, easy to install) and MKL (free, need a registration but more powerful)
85
86 18 Arnaud Sevin
h2. Dependencies : gfortran
87
88
Use your package manager to install dependencies:
89
* on scientific linux : yum install gcc-gfortran libgfortran
90
* on debian : apt-get install gfortran gfortran-multilib
91
92 14 Arnaud Sevin
h2. Configure MAGMA with ATLAS
93
94
h3. Dependencies : blas, lapack, atlas
95
96
Use your package manager to install dependencies:
97
* on scientific linux : yum install blas-devel lapack-devel atlas-devel
98 16 Arnaud Sevin
* on debian : apt-get install libblas-dev liblapack-dev libatlas-base-dev libatlas-dev
99 14 Arnaud Sevin
100
h3. extraction
101
102
MAGMA is available here : http://icl.cs.utk.edu/magma/software/index.html
103
104
extract the tgz file and go into the new directory
105
> ~$ tar xf magma-1.4.1-beta.tar.gz
106
> ~$ cd magma-1.4.1
107
108
h3. configuration
109
110
You have to create your own make.inc :
111
112
* example on a scientific linux : *please verify GPU_TARGET, LAPACKDIR, ATLASDIR, CUDADIR*
113
114
<pre><code class="Makefile">
115
#//////////////////////////////////////////////////////////////////////////////
116
#   -- MAGMA (version 1.4.1) --
117
#      Univ. of Tennessee, Knoxville
118
#      Univ. of California, Berkeley
119
#      Univ. of Colorado, Denver
120
#      November 2013
121
#//////////////////////////////////////////////////////////////////////////////
122
123
# GPU_TARGET specifies for which GPU you want to compile MAGMA:
124
#     "Tesla"  (NVIDIA compute capability 1.x cards)
125
#     "Fermi"  (NVIDIA compute capability 2.x cards)
126
#     "Kepler" (NVIDIA compute capability 3.x cards)
127
# See http://developer.nvidia.com/cuda-gpus
128
129
GPU_TARGET ?= Fermi
130
131
CC        = gcc
132
NVCC      = nvcc
133
FORT      = gfortran
134
135
ARCH      = ar
136
ARCHFLAGS = cr
137
RANLIB    = ranlib
138
139
OPTS      = -fPIC -O3 -DADD_ -fopenmp -DMAGMA_SETAFFINITY
140
F77OPTS   = -fPIC -O3 -DADD_
141
FOPTS     = -fPIC -O3 -DADD_ -x f95-cpp-input
142
NVOPTS    =       -O3 -DADD_ -Xcompiler "-fno-strict-aliasing -fPIC"
143
LDOPTS    = -fPIC -fopenmp
144
145
# Depending on how ATLAS and LAPACK were compiled, you may need one or more of:
146
LIB       = -llapack -lf77blas -latlas -lcblas -lcublas -lcudart -lstdc++ -lm -lgfortran
147
148
# define library directories here or in your environment
149
LAPACKDIR = /usr/lib64
150
ATLASDIR  = /usr/lib64/atlas
151
CUDADIR   = /usr/local/cuda
152
153
LIBDIR    = -L$(LAPACKDIR) \
154
            -L$(ATLASDIR) \
155
            -L$(CUDADIR)/lib64
156
157
INC       = -I$(CUDADIR)/include
158
</code></pre>
159
160 17 Arnaud Sevin
* on debian : *please verify GPU_TARGET, LAPACKDIR, ATLASDIR, CUDADIR*
161
<pre><code class="Makefile">
162
#//////////////////////////////////////////////////////////////////////////////
163
#   -- MAGMA (version 1.4.1) --
164
#      Univ. of Tennessee, Knoxville
165
#      Univ. of California, Berkeley
166
#      Univ. of Colorado, Denver
167
#      November 2013
168
#//////////////////////////////////////////////////////////////////////////////
169
170
# GPU_TARGET specifies for which GPU you want to compile MAGMA:
171
#     "Tesla"  (NVIDIA compute capability 1.x cards)
172
#     "Fermi"  (NVIDIA compute capability 2.x cards)
173
#     "Kepler" (NVIDIA compute capability 3.x cards)
174
# See http://developer.nvidia.com/cuda-gpus
175
176
GPU_TARGET ?= Fermi
177
178
CC        = gcc
179
NVCC      = nvcc
180
FORT      = gfortran
181
182
ARCH      = ar
183
ARCHFLAGS = cr
184
RANLIB    = ranlib
185
186
OPTS      = -fPIC -O3 -DADD_ -fopenmp -DMAGMA_SETAFFINITY
187
F77OPTS   = -fPIC -O3 -DADD_
188
FOPTS     = -fPIC -O3 -DADD_ -x f95-cpp-input
189
NVOPTS    =       -O3 -DADD_ -Xcompiler "-fno-strict-aliasing -fPIC" 
190
LDOPTS    = -fPIC -fopenmp
191
192
# Depending on how ATLAS and LAPACK were compiled, you may need one or more of:
193
LIB       = -llapack -lf77blas -latlas -lcblas -lcublas -lcudart -lstdc++ -lm -lgfortran
194
195
# define library directories here or in your environment
196
LAPACKDIR = /usr/lib
197
ATLASDIR  = /usr/lib
198
CUDADIR   = /usr/local/cuda
199
200
LIBDIR    = -L$(LAPACKDIR) \
201
            -L$(ATLASDIR) \
202
            -L$(CUDADIR)/lib64
203
204
INC       = -I$(CUDADIR)/include
205
</code></pre>
206 14 Arnaud Sevin
207
h2. Configure MAGMA with MKL
208
209
h3. extraction
210
211
To download MKL, you have to create a account here : https://registrationcenter.intel.com/RegCenter/NComForm.aspx?ProductID=1517
212
213
extract l_ccompxe_2013_sp1.1.106.tgz and go into l_ccompxe_2013_sp1.1.106
214
215
install it with ./install_GUI.sh and add IPP stuff to default choices
216
217
h3. configuration
218
219
Just copy make.inc.mkl-gcc to make.inc *please verify GPU_TARGET, MKLDIR, CUDADIR*
220
221
h2. compilation and installation
222
223
h3. compilation
224
225
just compile the shared target (and test if you want)
226
> ~$ make -j 8 shared
227
228
h3. installation
229
230
To install libraries and include files in a given prefix, run:
231
> ~$ make install prefix=/usr/local/magma
232
  
233
The default prefix is /usr/local/magma. You can also set prefix in make.inc.
234
235
h3. tune (not tested)
236
237 15 Arnaud Sevin
For multi-GPU functions, set $MAGMA_NUM_GPUS to set the number of GPUs to use.
238
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.
239
240
h2. Platform installation
241
242
Just just define $MAGMA_PATH and use the standard procedure