MatrixVector.tex

%%
%% MatrixVector.tex
%% Login : <hoang-trong@hoang-trong-laptop>
%% Started on  Thu Oct  1 13:03:35 2009 Hoang-Trong Minh Tuan
%% $Id$
%% 
%% Copyright (C) 2009 Hoang-Trong Minh Tuan
%%

\chapter{Matrix and Vector operations}
\label{chap:matr-vect-oper}

\section{Matrix-Matrix}
\label{sec:matrix-matrix}

\subsection{M-M multiplication}
\label{sec:m-m-multiplication}

\subsubsection{DO loop}
\label{sec:do-loop}

You use 3 DO loops.

\begin{lstlisting}
  REAL, DIMENSION(*,*), ALLOCATABLE:: A, C, B
  INTEGER, INTENT(IN) :: rA, cA, rB, cB
  INTEGER, INTENT(INOUT) :: rC, cC

  rC = rA
  cC = cB
  
  DO i = 1, rC
     DO j = 1, cC
        C(i, j) = 0.
        DO k = 1, cA
           C(i, j) = C(i, j) + A(i, k)*B(k, j)
        ENDDO
     ENDDO
  ENDDO
\end{lstlisting}

\subsubsection{matmul }
\label{sec:matmul-}

MATMUL is an intrinsic subroutine whose detail implementation is
compiler-dependent. So, different compiler may give different
performance when with a program that use MATMUL.

\begin{lstlisting}
   REAL(KIND=kr8), DIMENSION(:,:), ALLOCATABLE:: A, B, C
   CALL CPU_TIME(start)
   C = MATMUL(A, B)
   CALL CPU_TIME(finish)
\end{lstlisting}
\subsubsection{BLAS level 1, 2}
\label{sec:blas-level-1}

BLAS use DGEMM or SGEMM for matrix-matrix multiplication. 
\begin{lstlisting}
   alpha = 1.0d0
   beta = 0.0d0
   rC = rA
   cC = cB
   IF (ALLOCATED(C)) THEN
      DEALLOCATE(C)
   END IF
   ALLOCATE(C(rC, cC))
   CALL CPU_TIME(start)
   CALL dgemm('N', 'N', rA, cB, cA, alpha, A, rA, B, &
        rB, beta, C, rA)
   CALL CPU_TIME(finish)
\end{lstlisting}


%%% Local Variables: 
%%% mode: latex
%%% TeX-master: "gpucomputing"
%%% End: