Using LAPACK in C on Linux

Instruction for using LAPACK in c on Linux.

Installation

Install BLAS
- Download latest version from http://www.netlib.org/blas/.
- Untar
- $cd BLAS-*/
- $make
- get lib blas*.a
Download latest version from http://www.netlib.org/lapack/.
Untar
$cd lapack-*/
$cp make.inc.example make.inc and change line 81 to point to blas*.a
$make lib
$make lapackelib
$make cblaslib
Move files in LAPACKE/include/ and CBLAS/include/ to some include/ you are familiar with.
Move ‘lapacke’, ‘lapack’, ‘blas’, ‘cblas’ librarys to some lib/ you are familiar with, and rename them as liblapacke.a, liblapack.a, libblas.a, libcblas.a.

Test

Get test program from http://www.netlib.org/lapack/lapacke.html.
Compile with $gcc test_lapack.c -L lib/ -llapacke -llapack -lcblas -lblas -lgfortran -lm -I include/ -o test_lapack

Documentation

http://www.netlib.org/lapack/explore-html/index.html
This documentation have specific defination for all lapack functions. But, if you want to use some function f() in lapack, you should use LAPACKE_f() or cblas_f(), you can see parameters of LAPACKE_f() or cblas_f() in this documentation as well, but there are no details.

Functions

#include <lapacke.h>for lapacke functions, #include <cblas.h>for cblas functions

cblas_dtrsm

void cblas_dtrsm (const CBLAS_LAYOUT layout, const CBLAS_SIDE Side, const CBLAS_UPLO Uplo, const CBLAS_TRANSPOSE TransA, const CBLAS_DIAG Diag, const int M, const int N, const double alpha, const double \*A, const int lda, double \*B, const int ldb)

layout: CblasRowMajor or CblasRowMajor
Side: CblasLeft or CblasRight, same meaning as SIDE in dtrsm()
Uplo: CblasLower or CblasUpper, same meaning as UPLO in dtrsm()
TransA: CblasNoTrans or CblasTrans, same meaning as TRANSA in dtrsm()
Diag: CblasNonUnit or CblasUnit, same meaning as DIAG in dtrsm()
alpha: same meaning as ALPHA in dtrsm()
A: same meaning as A in dtrsm()
B: same meaning as B in dtrsm()

If you are using CblasRowMajor, and A is a square matrix:

M: length of each edge of A
N: number of columns in B
LDA: the same as M
LDB: the same as N

FAQ

What is LDA?

(http://icl.cs.utk.edu/lapack-forum/viewtopic.php?f=2&t=1897)

On LDA, the doc says:

LDA (input) INTEGER

The leading dimension of the array A. LDA >= max(1,N).

Now N is the number of columns in the input matrix. But the input matrix is symmetric, so we’d expect it to be N x N, right? In that case LDA = N. The “>=” part in the comment just allows for the possibility that the input matrix A is in fact part of a larger array, with a greater number of rows than are actually used in this function. But unless you’re doing something tricksy, “LDA” is just the rows of A, and equals N.

Understunding of mine: While storing a matrix A in memory, the jth element of ith row A[i][j] can also be A[i*n+j], which means the (i*n+j)th element in A, where n is the number of elements in each row. However, while you are using part of a matrix to compute, for example part of A, A', A'[i][j] is the (i*n+j)th element of A', but n is no longer the number of elements in each row of A'. As a result, you need to tell the computer A' is part of a larger matrix A using LDA=n.

In conclution, LDA is the number of elements in each row of the matrix you stored in memory.