| 1 |
** dissection solver Ver 1.0.0 : written in C++ with qd libiray or ** |
| 2 |
** GNU __float128 ** |
| 3 |
** Atsushi Suzuki, 25 Jun. 2015 |
| 4 |
** 30 Nov. 2016 |
| 5 |
|
| 6 |
How to compile: |
| 7 |
|
| 8 |
- directory structure needs to be the follownig. |
| 9 |
include/ lib/ src/ are needed to be the same place called as $(PROJ_DIR) |
| 10 |
|
| 11 |
- copy src/Defs.inc.Intel to src/Defs.inc |
| 12 |
- copy src/Makfile.Intel to src/Makefile |
| 13 |
|
| 14 |
- src/Defs.inc contains definition of the place of header files |
| 15 |
METIS_INC = $(PROJ_DIR)/metis-5.1.0/include |
| 16 |
SCOTCH_INC = $(PROJ_DIR)/scotch_6.0.4/include |
| 17 |
QD_INC = $(PROJ_DIR)/include |
| 18 |
and libraries |
| 19 |
METIS_DIR = $(PROJ_DIR)/metis-5.0.2/build/Darwin-x86_64/libmetis |
| 20 |
LIB_DIR_SCOTCH = $(PROJ_DIR)/scotch_6.0.4/lib |
| 21 |
LIB_DIR_QD = $(PROJ_DIR)/qd-2.3.17/lib |
| 22 |
|
| 23 |
- to use intel MKL specify BLAS in Def.inc |
| 24 |
BLAS = MKL |
| 25 |
On the MacOSX (Mavericks or Yosemite), |
| 26 |
BLAS = VECLIB |
| 27 |
attains 90 % of the speed of MKL. For preparation to use vecLib |
| 28 |
framework, command line Xcode needs to be installed by |
| 29 |
% xcode-select --install |
| 30 |
|
| 31 |
|
| 32 |
- qd library needs to be compiled with patches to handle complex class of |
| 33 |
LLVM clang++. |
| 34 |
qd-2.3.17.tar.gz will be obtained from |
| 35 |
http://crd-legacy.lbl.gov/~dhbailey/mpdist/qd-2.3.17.tar.gz |
| 36 |
to extract tar ball and apply patches |
| 37 |
% tar xzf qd-2.3.17.tar.gz |
| 38 |
% cd qd-2.3.17 |
| 39 |
% patch -p1 < qd-2.3.17-for-LLVM.patch |
| 40 |
configure needs to be done with specifying install directory $(PROJ_DIR) |
| 41 |
% ./configure --prefix=$(PROJ_DIR) CXX=clang++ CC=clang |
| 42 |
% make install |
| 43 |
|
| 44 |
- to clean directories under $(PROJ_DIR)/src, |
| 45 |
% make distclean |
| 46 |
% make distclean |
| 47 |
|
| 48 |
- to build $(PROJ_DIR)/libDissection.dyld, in $(PROJ_DIR)/src |
| 49 |
% make |
| 50 |
|
| 51 |
- a test program is located in $(PROJ_DIR)/src/C-test |
| 52 |
% make |
| 53 |
% ./MM-Dissection ../MM-matrix/MM.testA.data 0 1 1.e-2 8 2 |
| 54 |
the first argument specifies 0 for SCOTCH and 1 for METIS |
| 55 |
the second specifies number of threads, which should be less or equal to |
| 56 |
number of the physical CPU cores |
| 57 |
the third specifies pivot threshold |
| 58 |
the fourth specifies number of level of dissection |
| 59 |
the fifth specifies scaling strategy : 0 for no-scaling, 1 for diagonal |
| 60 |
2 for scaling for KKT-type matrix |
| 61 |
|