参考链接:linux下安装blas和lapack
BLAS 和 LAPACK 这两个数学库是很多 Linux 科学计算软件需要调用的,所以经常会用到。
LAPACK,其名为Linear Algebra PACKage的缩写,是一以Fortran编程语言编写,用于数值计算的函式集。LAPACK提供了丰富的工具函式,可用于诸如解多元线性方程式、线性系统方程组的最小平方解、计算特征向量、用于计算矩阵QR分解的Householder转换、以及奇异值分解等问题。
LAPACK的源码可以从http://www.netlib.org/lapack/处下载,BLAS也包含在其中。
BLAS,全称Basic Linear AlgebraSubprograms,即基础线性代数子程序库,里面拥有大量已经编写好的关于线性代数运算的程序。
BLAS的源码可以从 http://www.netlib.org/blas/ 下载,但实际上LAPACK中已经包含了BLAS。
0. 写在前面的:
之前采用gfortran来编译生成了lapack的库文件,但是在后续采用pgf90命令(pgf90 -llapack)来编译其它文件时,产生了以下的类似错误:
[She@she-centos7 TEC_She]$ make pgf90 -g -fast -c m_bern.f90 pgf90 -g -fast -c d_inpkey.f90 pgf90 -g -fast -c p_menaux.f90 ... pgf90 -g -fast -o main_igsTec -L/usr/local/lib m_bern.o d_inpkey.o ... -llapack -lblas /usr/local/lib/liblapack.a(dormlq.o):在函数‘dormlq_’中: dormlq.f:(.text+0x32b):对‘_gfortran_concat_string’未定义的引用 dormlq.f:(.text+0x887):对‘_gfortran_concat_string’未定义的引用 /usr/local/lib/liblapack.a(dormqr.o):在函数‘dormqr_’中: dormqr.f:(.text+0x2f8):对‘_gfortran_concat_string’未定义的引用 dormqr.f:(.text+0x81c):对‘_gfortran_concat_string’未定义的引用 /usr/local/lib/liblapack.a(ilaenv.o):在函数‘ilaenv_’中: ilaenv.f:(.text+0x58):对‘_gfortran_compare_string’未定义的引用 ilaenv.f:(.text+0x287):对‘_gfortran_compare_string’未定义的引用 ilaenv.f:(.text+0x2b4):对‘_gfortran_compare_string’未定义的引用 ilaenv.f:(.text+0x2d5):对‘_gfortran_compare_string’未定义的引用 ilaenv.f:(.text+0x2f4):对‘_gfortran_compare_string’未定义的引用 /usr/local/lib/liblapack.a(ilaenv.o):ilaenv.f:(.text+0x313): more undefined references to `_gfortran_compare_string' follow /usr/local/lib/liblapack.a(xerbla.o):在函数‘xerbla_’中: xerbla.f:(.text+0x49):对‘_gfortran_st_write’未定义的引用 xerbla.f:(.text+0x54):对‘_gfortran_string_len_trim’未定义的引用 xerbla.f:(.text+0x66):对‘_gfortran_transfer_character_write’未定义的引用 xerbla.f:(.text+0x76):对‘_gfortran_transfer_integer_write’未定义的引用 xerbla.f:(.text+0x7e):对‘_gfortran_st_write_done’未定义的引用 xerbla.f:(.text+0x87):对‘_gfortran_stop_string’未定义的引用 /usr/local/lib/liblapack.a(iparmq.o):在函数‘iparmq_’中: iparmq.f:(.text+0x150):对‘_gfortran_compare_string’未定义的引用 iparmq.f:(.text+0x16f):对‘_gfortran_compare_string’未定义的引用 iparmq.f:(.text+0x18f):对‘_gfortran_compare_string’未定义的引用 iparmq.f:(.text+0x273):对‘_gfortran_compare_string’未定义的引用 iparmq.f:(.text+0x28e):对‘_gfortran_compare_string’未定义的引用 /usr/local/lib/liblapack.a(iparam2stage.o):iparam2stage.F:(.text+0x263): more undefined references to `_gfortran_compare_string' follow make: *** [all] 错误 2
这是由于gfortran和pgf90编译命令不同导致的,因而本文以PGI编译器来执行lapack源代码的编译,即,Fortran程序采用pgf90命令,C程序采用pgcc命令来编译,相应的具体过程及参数记录如下。
1. 确保机器上安装了PGI gfortran编译器。如果没有安装的话,手动安装:
sudo yum install gfortran
PGI编译器需要去官网下载,具体安装过程参见我的另一篇博客《CentOS 7上安装PGI 2017编译器》。
2. 下载blas, cblas, lapack 源代码, 这些源码都可以在http://www.netlib.org 上找到,下载并解压。
我下载的版本是lapack-3.7.1.tgz,解压之后会有一个文件夹,lapack-3.7.1,它含有BLAS,CBLAS,LAPACKE等文件夹,其中BLAS是BLAS的源码,CBLAS是BLAS的C语言接口。
3. 这里就是具体的编译步骤
(0) 复制lapack目录下的make.in.example文件,并修改其中的内容
首先进入lapack-3.7.1文件夹,然后根据平台的特点,将该目录下对应的 make.inc.example 文件另存为 make.inc。
cd .. cp make.inc.example make.inc vi make.inc #################################################################### # LAPACK make include file. # # LAPACK, Version 3.7.1 # # June 2017 # #################################################################### SHELL = /bin/sh # CC is the C compiler, normally invoked with options CFLAGS. # CC = pgcc # gcc CFLAGS = -O3 # Modify the FORTRAN and OPTS definitions to refer to the compiler # and desired compiler options for your machine. NOOPT refers to # the compiler options desired when NO OPTIMIZATION is selected. # # Note: During a regular execution, LAPACK might create NaN and Inf # and handle these quantities appropriately. As a consequence, one # should not compile LAPACK with flags such as -ffpe-trap=overflow. # FORTRAN = pgf90 # gfortran OPTS = -O2 -Mrecursive # -frecursive DRVOPTS = $(OPTS) NOOPT = -O0 -Mrecursive # -frecursive # Define LOADER and LOADOPTS to refer to the loader and desired # load options for your machine. # LOADER = pgf90 # gfortran LOADOPTS = # The archiver and the flag(s) to use when building an archive # (library). If your system has no ranlib, set RANLIB = echo. # ARCH = ar ARCHFLAGS = cr RANLIB = ranlib # Timer for the SECOND and DSECND routines # # Default: SECOND and DSECND will use a call to the # EXTERNAL FUNCTION ETIME #TIMER = EXT_ETIME # For RS6K: SECOND and DSECND will use a call to the # EXTERNAL FUNCTION ETIME_ #TIMER = EXT_ETIME_ # For gfortran compiler: SECOND and DSECND will use a call to the # INTERNAL FUNCTION ETIME #TIMER = INT_ETIME # If your Fortran compiler does not provide etime (like Nag Fortran # Compiler, etc...) SECOND and DSECND will use a call to the # INTERNAL FUNCTION CPU_TIME TIMER = INT_CPU_TIME # If none of these work, you can use the NONE value. # In that case, SECOND and DSECND will always return 0. #TIMER = NONE # Uncomment the following line to include deprecated routines in # the LAPACK library. # #BUILD_DEPRECATED = Yes # LAPACKE has the interface to some routines from tmglib. # If LAPACKE_WITH_TMG is defined, add those routines to LAPACKE. # #LAPACKE_WITH_TMG = Yes # Location of the extended-precision BLAS (XBLAS) Fortran library # used for building and testing extended-precision routines. The # relevant routines will be compiled and XBLAS will be linked only # if USEXBLAS is defined. # #USEXBLAS = Yes #XBLASLIB = -lxblas # The location of the libraries to which you will link. (The # machine-specific, optimized BLAS library should be used whenever # possible.) # BLASLIB = ../../librefblas.a CBLASLIB = ../../libcblas.a LAPACKLIB = liblapack.a TMGLIB = libtmglib.a LAPACKELIB = liblapacke.a
(1) 编译blas
进入 BLAS/SRC 文件夹,执行以下几条命令
cd BLAS/SRC
# gfortran -c -O3 *.f # 编译所有的 .f 文件,生成 .o文件,这里采用PGI编译器的pgf90命令来编译 pgf90 -c -O3 *.f # 编译所有的 .f 文件,生成 .o文件,这个pgf90编译命令与~/lapack*/make.inc保持一致 ar rv libblas.a *.o # 链接所有的 .o文件,生成.a 文件 sudo cp libblas.a /usr/local/lib #将库文件复制到系统库目录
sudo cp libblas.a /usr/lib
(2) 编译cblas
进入CBLAS 文件夹,首先根据你自己的计算机平台,将目录下某个 Makefile.XXX复制为 Makefile.in , XXX表示计算机的平台,如果是linux,那么就将Makefile.LINUX 复制为Makefile.in,然后执行以下命令
cd .. && cd ../CBLAS
cp ../BLAS/SRC/libblas.a ./testing/ # 将上一步编译成功的 libblas.a复制到 CBLAS目录下的testing子目录 make # 编译所有的目录 sudo cp ../libcblas.a /usr/local/lib #将库文件复制到系统库目录下
sudo cp ../libcblas.a /usr/lib
(3) 编译 lapack 以及 lapacke
这一步比较麻烦,首先进入lapack-3.7.1文件夹,根据平台的特点,编辑 Makefile,编译 lapack 和 lapacke 文件,并将 lapacke 目录下的头文件、lapack 目录下生成的 *.a 文件拷贝到系统目录(/usr/local/lib, /usr/lib)下。
cd ..
vi Makefile # 修改 lapack-3.7.1/Makefile 文件,因为 lapack 依赖于 blas 库 # 旧版本
lib: lapacklib tmglib #lib: blaslib variants lapacklib tmglib # 新版本
#lib: lapacklib tmglib lib: blaslib variants lapacklib tmglib
make # 编译所有的lapack文件 cd LAPACKE # 进入LAPACKE 文件夹,这个文件夹包含lapack的C语言接口文件
make # 编译lapacke
sudo cp include/*.h /usr/local/include #将lapacke的头文件复制到系统头文件目录,
# 包括: lapacke.h, lapacke_config.h, lapacke_mangling.h,lapacke_mangling_with_flags.h lapacke_utils.h
cd .. # 返回到 lapack-3.7.1 目录
sudo cp *.a /usr/local/lib # 将生成的所有库文件复制到系统库目录,
# 包括:liblapack.a, liblapacke.a, librefblas.a,libtmglib.a。
sudo cp *.a /usr/lib
Ques: 事实上,编译 lapack 时生成的 librefblas.a 文件与编译 BLAS 时生成的 libblas.a 文件大小基本一样,这里生成了两次,是否可以省去第(1)-(2)步?
至此blas,cblas 和 lapack 就成功安装到你的电脑上了。
4. lapack子程序测试
测试程序 Console.f
! 测试程序来自:http://blog.sina.com.cn/s/blog_5f350c9601014ejc.html program Console1 external dgesv integer n, lda, nrhs, ldb parameter (n=2,lda=2,nrhs=1,ldb=2) double precision A(lda,n) double precision b(ldb,nrhs) character byebye integer ipiv(n), info, i, j A(1,1)=1 A(1,2)=2 A(2,1)=3 A(2,2)=4 B(1,1)=5 B(2,1)=6 write(*,*) 'Hello World' call dgesv(n,nrhs,A,lda,ipiv,b,ldb,info) write(*,*) 'INFO =', info write(*,*) ((A(i,j),i=1,lda),j=1,n) write(*,*) ((B(i,j),i=1,ldb),j=1,nrhs) write(*,*) "END OF PROGRAM..." end program Console1
使用pgf90来编译该程序,可以得到预期的结果,而使用 gfortran 来编译则会报错:
[She@she-centos7 LSQtest]$ pgf90 Console1.f -lblas -llapack # 编译及运行正常 [She@she-centos7 LSQtest]$ ./a.out Hello World INFO = 0 3.000000000000000 0.3333333333333333 4.000000000000000 0.6666666666666667 -3.999999999999999 4.499999999999999 END OF PROGRAM... [She@she-centos7 LSQtest]$ [She@she-centos7 LSQtest]$ [She@she-centos7 LSQtest]$ [She@she-centos7 LSQtest]$ gfortran Console1.f -lblas -llapack # 编译报错 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgesv.o):在函数‘.C1_322’中: dgesv.f:(.data+0x18):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgetrf.o):在函数‘dgetrf_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrf.f:206:对‘dtrsm_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrf.f:213:对‘dgemm_’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgetrf.o):在函数‘.C1_331’中: dgetrf.f:(.data+0x38):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgetrs.o):在函数‘dgetrs_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrs.f:191:对‘dtrsm_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrs.f:191:对‘dtrsm_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrs.f:202:对‘dtrsm_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrs.f:202:对‘dtrsm_’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgetrs.o):在函数‘.C1_292’中: dgetrs.f:(.data+0x48):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dlaswp.o):(.data+0x0):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(ilaenv.o):在函数‘ilaenv_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:703:对‘pgf90_str_copy’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:261:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:274:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:687:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:353:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:479:对‘pgf90_strcmp’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(ilaenv.o):/home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./ilaenv.f:485: more undefined references to `pgf90_strcmp' follow /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(ilaenv.o):在函数‘.STATICS1’中: ilaenv.f:(.data+0x128):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(ieeeck.o):在函数‘.C1_352’中: ieeeck.f:(.data+0x10):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(xerbla.o):在函数‘xerbla_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90io_src_info03’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90io_fmtw_init’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90_lentrim’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90io_fmt_write’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90io_sc_i_fmt_write’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90io_fmtw_end’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./xerbla.f:90:对‘pgf90_stop08’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(xerbla.o):在函数‘.STATICS1’中: xerbla.f:(.data+0xa0):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(iparmq.o):在函数‘iparmq_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:265:对‘__gss_log’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:265:对‘__gss_log’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:265:对‘__mth_i_nint’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:321:对‘pgf90_str_copy’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:336:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:369:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:387:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:379:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparmq.f:379:对‘pgf90_strcmp’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(iparmq.o):在函数‘.C1_289’中: iparmq.f:(.data+0x20):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(iparam2stage.o):在函数‘iparam2stage_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:207:对‘pgf90_str_copy’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:341:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:350:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:359:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:360:对‘pgf90_strcmp’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:378:对‘pgf90_strcmp’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(iparam2stage.o):/home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./iparam2stage.F:378: more undefined references to `pgf90_strcmp' follow /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(iparam2stage.o):在函数‘.C1_327’中: iparam2stage.F:(.data+0x48):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(lsame.o):(.data+0x0):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgetrf2.o):在函数‘dgetrf2_’中: /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrf2.f:242:对‘dtrsm_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrf2.f:247:对‘dgemm_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrf2.f:192:对‘idamax_’未定义的引用 /home/She/Software/Fortran/Lapack/lapack-3.7.1/SRC/./dgetrf2.f:207:对‘dscal_’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dgetrf2.o):在函数‘.C1_292’中: dgetrf2.f:(.data+0x40):对‘f90_compiled’未定义的引用 /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../liblapack.a(dlamch.o):在函数‘.C1_291’中: dlamch.f:(.data+0x68):对‘f90_compiled’未定义的引用 collect2: 错误:ld 返回 1
测试完毕。
5. 结论和心得
在同一台电脑上,最好对Fortran程序和C程序使用一致的编译命令,库文件和源代码都遵循这样的做法,可以避免不必要的奇怪bug。
如果一组程序中,某些文件采用了gfortran 来编译,而一些文件采用了 pgf90 命令来编译,则链接时容易产生一些难以检查的错误,浪费生命!