Statistiques
| Révision :

root / setup / Make.Debian_CUBLAS_OpenMPI

Historique | Voir | Annoter | Télécharger (8,98 ko)

1 9 equemene
#
2 9 equemene
#  -- High Performance Computing Linpack Benchmark (HPL)
3 9 equemene
#     HPL - 2.0 - September 10, 2008
4 9 equemene
#     Antoine P. Petitet
5 9 equemene
#     University of Tennessee, Knoxville
6 9 equemene
#     Innovative Computing Laboratory
7 9 equemene
#     (C) Copyright 2000-2008 All Rights Reserved
8 9 equemene
#
9 9 equemene
#  -- Copyright notice and Licensing terms:
10 9 equemene
#
11 9 equemene
#  Redistribution  and  use in  source and binary forms, with or without
12 9 equemene
#  modification, are  permitted provided  that the following  conditions
13 9 equemene
#  are met:
14 9 equemene
#
15 9 equemene
#  1. Redistributions  of  source  code  must retain the above copyright
16 9 equemene
#  notice, this list of conditions and the following disclaimer.
17 9 equemene
#
18 9 equemene
#  2. Redistributions in binary form must reproduce  the above copyright
19 9 equemene
#  notice, this list of conditions,  and the following disclaimer in the
20 9 equemene
#  documentation and/or other materials provided with the distribution.
21 9 equemene
#
22 9 equemene
#  3. All  advertising  materials  mentioning  features  or  use of this
23 9 equemene
#  software must display the following acknowledgement:
24 9 equemene
#  This  product  includes  software  developed  at  the  University  of
25 9 equemene
#  Tennessee, Knoxville, Innovative Computing Laboratory.
26 9 equemene
#
27 9 equemene
#  4. The name of the  University,  the name of the  Laboratory,  or the
28 9 equemene
#  names  of  its  contributors  may  not  be used to endorse or promote
29 9 equemene
#  products  derived   from   this  software  without  specific  written
30 9 equemene
#  permission.
31 9 equemene
#
32 9 equemene
#  -- Disclaimer:
33 9 equemene
#
34 9 equemene
#  THIS  SOFTWARE  IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 9 equemene
#  ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES,  INCLUDING,  BUT NOT
36 9 equemene
#  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 9 equemene
#  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY
38 9 equemene
#  OR  CONTRIBUTORS  BE  LIABLE FOR ANY  DIRECT,  INDIRECT,  INCIDENTAL,
39 9 equemene
#  SPECIAL,  EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES  (INCLUDING,  BUT NOT
40 9 equemene
#  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 9 equemene
#  DATA OR PROFITS; OR BUSINESS INTERRUPTION)  HOWEVER CAUSED AND ON ANY
42 9 equemene
#  THEORY OF LIABILITY, WHETHER IN CONTRACT,  STRICT LIABILITY,  OR TORT
43 9 equemene
#  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 9 equemene
#  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 9 equemene
# ######################################################################
46 9 equemene
#
47 9 equemene
# ----------------------------------------------------------------------
48 9 equemene
# - shell --------------------------------------------------------------
49 9 equemene
# ----------------------------------------------------------------------
50 9 equemene
#
51 9 equemene
SHELL        = /bin/sh
52 9 equemene
#
53 9 equemene
CD           = cd
54 9 equemene
CP           = cp
55 9 equemene
LN_S         = ln -s
56 9 equemene
MKDIR        = mkdir
57 9 equemene
RM           = /bin/rm -f
58 9 equemene
TOUCH        = touch
59 9 equemene
#
60 9 equemene
# ----------------------------------------------------------------------
61 9 equemene
# - Platform identifier ------------------------------------------------
62 9 equemene
# ----------------------------------------------------------------------
63 9 equemene
#
64 9 equemene
ARCH         = $(arch)
65 9 equemene
#
66 9 equemene
# ----------------------------------------------------------------------
67 9 equemene
# - HPL Directory Structure / HPL library ------------------------------
68 9 equemene
# ----------------------------------------------------------------------
69 9 equemene
#
70 9 equemene
TOPdir       = $(topdir)
71 9 equemene
INCdir       = $(TOPdir)/include
72 9 equemene
BINdir       = $(TOPdir)/bin/$(ARCH)
73 9 equemene
LIBdir       = $(TOPdir)/lib/$(ARCH)
74 9 equemene
#
75 9 equemene
HPLlib       = $(LIBdir)/libhpl.a
76 9 equemene
#
77 9 equemene
# ----------------------------------------------------------------------
78 9 equemene
# - Message Passing library (MPI) --------------------------------------
79 9 equemene
# ----------------------------------------------------------------------
80 9 equemene
# MPinc tells the  C  compiler where to find the Message Passing library
81 9 equemene
# header files,  MPlib  is defined  to be the name of  the library to be
82 9 equemene
# used. The variable MPdir is only used for defining MPinc and MPlib.
83 9 equemene
#
84 9 equemene
MPdir        = /usr/lib/openmpi
85 9 equemene
MPinc        = -I/include/include/openmpi
86 9 equemene
MPlib        = -L$(MPdir)/lib/libmpi.so
87 9 equemene
#
88 9 equemene
# ----------------------------------------------------------------------
89 9 equemene
# - Linear Algebra library (BLAS or VSIPL) -----------------------------
90 9 equemene
# ----------------------------------------------------------------------
91 9 equemene
# LAinc tells the  C  compiler where to find the Linear Algebra  library
92 9 equemene
# header files,  LAlib  is defined  to be the name of  the library to be
93 9 equemene
# used. The variable LAdir is only used for defining LAinc and LAlib.
94 9 equemene
#
95 9 equemene
LAdir        = /opt/cuda/lib64
96 9 equemene
LAinc        = -I/opt/cuda/include
97 9 equemene
LAlib        = $(LAdir)/libcublas.so
98 9 equemene
#
99 9 equemene
# ----------------------------------------------------------------------
100 9 equemene
# - F77 / C interface --------------------------------------------------
101 9 equemene
# ----------------------------------------------------------------------
102 9 equemene
# You can skip this section  if and only if  you are not planning to use
103 9 equemene
# a  BLAS  library featuring a Fortran 77 interface.  Otherwise,  it  is
104 9 equemene
# necessary  to  fill out the  F2CDEFS  variable  with  the  appropriate
105 9 equemene
# options.  **One and only one**  option should be chosen in **each** of
106 9 equemene
# the 3 following categories:
107 9 equemene
#
108 9 equemene
# 1) name space (How C calls a Fortran 77 routine)
109 9 equemene
#
110 9 equemene
# -DAdd_              : all lower case and a suffixed underscore  (Suns,
111 9 equemene
#                       Intel, ...),                           [default]
112 9 equemene
# -DNoChange          : all lower case (IBM RS6000),
113 9 equemene
# -DUpCase            : all upper case (Cray),
114 9 equemene
# -DAdd__             : the FORTRAN compiler in use is f2c.
115 9 equemene
#
116 9 equemene
# 2) C and Fortran 77 integer mapping
117 9 equemene
#
118 9 equemene
# -DF77_INTEGER=int   : Fortran 77 INTEGER is a C int,         [default]
119 9 equemene
# -DF77_INTEGER=long  : Fortran 77 INTEGER is a C long,
120 9 equemene
# -DF77_INTEGER=short : Fortran 77 INTEGER is a C short.
121 9 equemene
#
122 9 equemene
# 3) Fortran 77 string handling
123 9 equemene
#
124 9 equemene
# -DStringSunStyle    : The string address is passed at the string loca-
125 9 equemene
#                       tion on the stack, and the string length is then
126 9 equemene
#                       passed as  an  F77_INTEGER  after  all  explicit
127 9 equemene
#                       stack arguments,                       [default]
128 9 equemene
# -DStringStructPtr   : The address  of  a  structure  is  passed  by  a
129 9 equemene
#                       Fortran 77  string,  and the structure is of the
130 9 equemene
#                       form: struct {char *cp; F77_INTEGER len;},
131 9 equemene
# -DStringStructVal   : A structure is passed by value for each  Fortran
132 9 equemene
#                       77 string,  and  the  structure is  of the form:
133 9 equemene
#                       struct {char *cp; F77_INTEGER len;},
134 9 equemene
# -DStringCrayStyle   : Special option for  Cray  machines,  which  uses
135 9 equemene
#                       Cray  fcd  (fortran  character  descriptor)  for
136 9 equemene
#                       interoperation.
137 9 equemene
#
138 9 equemene
F2CDEFS      =
139 9 equemene
#
140 9 equemene
# ----------------------------------------------------------------------
141 9 equemene
# - HPL includes / libraries / specifics -------------------------------
142 9 equemene
# ----------------------------------------------------------------------
143 9 equemene
#
144 9 equemene
HPL_INCLUDES = -I$(INCdir) -I$(INCdir)/$(ARCH) $(LAinc) $(MPinc)
145 9 equemene
HPL_LIBS     = $(HPLlib) $(LAlib) $(MPlib)
146 9 equemene
#
147 9 equemene
# - Compile time options -----------------------------------------------
148 9 equemene
#
149 9 equemene
# -DHPL_COPY_L           force the copy of the panel L before bcast;
150 9 equemene
# -DHPL_CALL_CBLAS       call the cblas interface;
151 9 equemene
# -DHPL_CALL_VSIPL       call the vsip  library;
152 9 equemene
# -DHPL_CALL_CUBLAS       call the CuBLAS library from Nvidia Cuda;
153 9 equemene
154 9 equemene
# -DHPL_DETAILED_TIMING  enable detailed timers;
155 9 equemene
#
156 9 equemene
# By default HPL will:
157 9 equemene
#    *) not copy L before broadcast,
158 9 equemene
#    *) call the BLAS Fortran 77 interface,
159 9 equemene
#    *) not display detailed timing information.
160 9 equemene
#
161 9 equemene
HPL_OPTS     = -DHPL_CALL_CUBLAS
162 9 equemene
#
163 9 equemene
# ----------------------------------------------------------------------
164 9 equemene
#
165 9 equemene
HPL_DEFS     = $(F2CDEFS) $(HPL_OPTS) $(HPL_INCLUDES)
166 9 equemene
#
167 9 equemene
# ----------------------------------------------------------------------
168 9 equemene
# - Compilers / linkers - Optimization flags ---------------------------
169 9 equemene
# ----------------------------------------------------------------------
170 9 equemene
#
171 9 equemene
CC           = /usr/bin/mpicc.openmpi
172 9 equemene
CCNOOPT      = $(HPL_DEFS)
173 9 equemene
CCFLAGS      = $(HPL_DEFS) -fomit-frame-pointer -O3 -funroll-loops -W -Wall
174 9 equemene
#
175 9 equemene
# On some platforms,  it is necessary  to use the Fortran linker to find
176 9 equemene
# the Fortran internals used in the BLAS library.
177 9 equemene
#
178 9 equemene
LINKER       = /usr/bin/mpicc.openmpi
179 9 equemene
# The object fortran_thunking.o has to be compiled in root HPL folder
180 9 equemene
LINKFLAGS    = $(CCFLAGS) $(TOPdir)/fortran_thunking.o
181 9 equemene
#
182 9 equemene
ARCHIVER     = ar
183 9 equemene
ARFLAGS      = r
184 9 equemene
RANLIB       = echo
185 9 equemene
#
186 9 equemene
# ----------------------------------------------------------------------