mirror of
https://github.com/sfilippone/psblas3.git
synced 2026-10-16 10:52:08 +00:00
Updated description of DOT and friends to account for optional GLOBAL
var.
This commit is contained in:
Salvatore Filippone
@@ -595,6 +595,66 @@ Type, kind, rank and size must agree on all processes.
|
||||
\item The \verb|dat| argument may also be a long integer scalar.
|
||||
\end{enumerate}
|
||||
|
||||
\clearpage\subsection*{psb\_nrm2 --- Global 2-norm reduction}
|
||||
\addcontentsline{toc}{subsection}{psb\_nrm2}
|
||||
|
||||
\begin{verbatim}
|
||||
call psb_nrm2(icontxt, dat, root)
|
||||
\end{verbatim}
|
||||
|
||||
This subroutine implements a 2-norm value reduction
|
||||
operation based on the underlying communication library.
|
||||
\begin{description}
|
||||
\item[Type:] Synchronous.
|
||||
\item[\bf On Entry ]
|
||||
\item[icontxt] the communication context identifying the virtual
|
||||
parallel machine.\\
|
||||
Scope: {\bf global}.\\
|
||||
Type: {\bf required}.\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: an integer variable.
|
||||
\item[dat] The local contribution to the global minimum.\\
|
||||
Scope: {\bf local}.\\
|
||||
Type: {\bf required}.\\
|
||||
Intent: {\bf inout}.\\
|
||||
Specified as: a real variable, which may be a
|
||||
scalar, or a rank 1 array. \
|
||||
Kind, rank and size must agree on all processes.
|
||||
\item[root] Process to hold the final value, or $-1$ to make it available
|
||||
on all processes.\\
|
||||
Scope: {\bf global}.\\
|
||||
Type: {\bf optional}.\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: an integer value $-1<= root <= np-1$, default -1. \\
|
||||
\end{description}
|
||||
|
||||
|
||||
\begin{description}
|
||||
\item[\bf On Return]
|
||||
\item[dat] On destination process(es), the result of the 2-norm reduction.\\
|
||||
Scope: {\bf global}.\\
|
||||
Type: {\bf required}.\\
|
||||
Intent: {\bf inout}.\\
|
||||
Specified as: a real variable, which may be a
|
||||
scalar, or a rank 1 array. \\
|
||||
Kind, rank and size must agree on all processes.
|
||||
\end{description}
|
||||
|
||||
|
||||
{\par\noindent\large\bfseries Notes}
|
||||
\begin{enumerate}
|
||||
\item This reduction is appropriate to compute the results of multiple
|
||||
(local) NRM2 operations at the same time.
|
||||
\item Denoting by $dat_i$ the value of the variable $dat$ on process
|
||||
$i$, the output $res$ is equivalent to the computation of
|
||||
\[ res = \sqrt{\sum_i dat_i^2},\]
|
||||
with care taken to avoid unnecessary overflow.
|
||||
\item The \verb|dat| argument is both input and output, and its
|
||||
value may be changed even on processes different from the final
|
||||
result destination.
|
||||
\end{enumerate}
|
||||
|
||||
|
||||
|
||||
\clearpage\subsection*{psb\_snd --- Send data}
|
||||
\addcontentsline{toc}{subsection}{psb\_snd}
|
||||
|
||||
+115
-16
@@ -105,7 +105,6 @@ Intent: {\bf out}.\\
|
||||
An integer value; 0 means no error has been detected.
|
||||
\end{description}
|
||||
|
||||
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
%
|
||||
% F90DOT PRODUCT
|
||||
@@ -129,7 +128,7 @@ Else if $x$ and $y$ are complex vectors then it computes dot-product as:
|
||||
%% \end{description}
|
||||
|
||||
\begin{verbatim}
|
||||
psb_gedot(x, y, desc_a, info)
|
||||
psb_gedot(x, y, desc_a, info [,global])
|
||||
\end{verbatim}
|
||||
%% \syntax*{psb\_gedot}{x, y, desc\_a, info, jx, jy}
|
||||
\begin{table}[h]
|
||||
@@ -176,6 +175,13 @@ Scope: {\bf local} \\
|
||||
Type: {\bf required}\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: an object of type \descdata.
|
||||
\item[global] Specifies whether the computation should include the
|
||||
global reduction across all processes.\\
|
||||
Scope: {\bf global} \\
|
||||
Type: {\bf optional}.\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: a logical scalar.
|
||||
Default: \verb|global=.true.|\\
|
||||
%% \item[jx] the column index of global dense matrix $x$,
|
||||
%% identifying the column of subvector $x$.\\
|
||||
%% Scope: {\bf global} \\
|
||||
@@ -190,7 +196,8 @@ Specified as: an object of type \descdata.
|
||||
%% Specified as: an integer variable $jy\ge 1$.
|
||||
\item[\bf On Return]
|
||||
\item[Function value] is the dot product of subvectors $x$ and $y$.\\
|
||||
Scope: {\bf global} \\
|
||||
Scope: {\bf global} unless the optional variable
|
||||
\verb|global=.false.| as been specified\\
|
||||
Specified as: a number of the data type indicated in Table~\ref{tab:f90dot}.
|
||||
\item[info] Error code.\\
|
||||
Scope: {\bf local} \\
|
||||
@@ -198,7 +205,25 @@ Type: {\bf required} \\
|
||||
Intent: {\bf out}.\\
|
||||
An integer value; 0 means no error has been detected.
|
||||
\end{description}
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
|
||||
{\par\noindent\large\bfseries Notes}
|
||||
\begin{enumerate}
|
||||
\item The computation of a global result requires a global
|
||||
communication, which entails a significant overhead. It may be
|
||||
necessary and/or advisable to compute multiple dot products at the same
|
||||
time; in this case, it is possible to improve the runtime efficiency
|
||||
by using the following scheme:
|
||||
\begin{lstlisting}
|
||||
vres(1) = psb_gedot(x1,y1,desc_a,info,global=.false.)
|
||||
vres(2) = psb_gedot(x2,y2,desc_a,info,global=.false.)
|
||||
vres(3) = psb_gedot(x3,y3,desc_a,info,global=.false.)
|
||||
call psb_sum(ictxt,vres(1:3))
|
||||
\end{lstlisting}
|
||||
In this way the global communication, which for small sizes is a
|
||||
latency-bound operation, is invoked only once.
|
||||
\end{enumerate}
|
||||
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
%
|
||||
% F90DOT PRODUCT
|
||||
%
|
||||
@@ -294,8 +319,8 @@ else if $x$ is a complex vector then it computes the infinity-norm as:
|
||||
%% \end{description}
|
||||
|
||||
\begin{verbatim}
|
||||
psb_geamax(x, desc_a, info)
|
||||
psb_normi(x, desc_a, info)
|
||||
psb_geamax(x, desc_a, info [,global])
|
||||
psb_normi(x, desc_a, info [,global])
|
||||
\end{verbatim}
|
||||
%% \syntax*{psb\_geamax}{x, desc\_a, info, jx}
|
||||
|
||||
@@ -334,7 +359,13 @@ Scope: {\bf local} \\
|
||||
Type: {\bf required}\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: an object of type \descdata.
|
||||
%% \item[jx] the column index of global dense matrix $x$,
|
||||
\item[global] Specifies whether the computation should include the
|
||||
global reduction across all processes.\\
|
||||
Scope: {\bf global} \\
|
||||
Type: {\bf optional}.\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: a logical scalar.
|
||||
Default: \verb|global=.true.|\\%% \item[jx] the column index of global dense matrix $x$,
|
||||
%% identifying the column of subvector $x$.\\
|
||||
%% Scope: {\bf global} \\
|
||||
%% Type: {\bf optional}; can only be present if $x$ is of rank 2.\\
|
||||
@@ -343,7 +374,8 @@ Specified as: an object of type \descdata.
|
||||
|
||||
\item[\bf On Return]
|
||||
\item[Function value] is the infinity norm of subvector $x$.\\
|
||||
Scope: {\bf global} \\
|
||||
Scope: {\bf global} unless the optional variable
|
||||
\verb|global=.false.| as been specified\\
|
||||
Specified as: a long precision real number.
|
||||
\item[info] Error code.\\
|
||||
Scope: {\bf local} \\
|
||||
@@ -351,7 +383,25 @@ Type: {\bf required} \\
|
||||
Intent: {\bf out}.\\
|
||||
An integer value; 0 means no error has been detected.
|
||||
\end{description}
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
|
||||
{\par\noindent\large\bfseries Notes}
|
||||
\begin{enumerate}
|
||||
\item The computation of a global result requires a global
|
||||
communication, which entails a significant overhead. It may be
|
||||
necessary and/or advisable to compute multiple norms at the same
|
||||
time; in this case, it is possible to improve the runtime efficiency
|
||||
by using the following scheme:
|
||||
\begin{lstlisting}
|
||||
vres(1) = psb_geamax(x1,desc_a,info,global=.false.)
|
||||
vres(2) = psb_geamax(x2,desc_a,info,global=.false.)
|
||||
vres(3) = psb_geamax(x3,desc_a,info,global=.false.)
|
||||
call psb_amx(ictxt,vres(1:3))
|
||||
\end{lstlisting}
|
||||
In this way the global communication, which for small sizes is a
|
||||
latency-bound operation, is invoked only once.
|
||||
\end{enumerate}
|
||||
|
||||
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
||||
%
|
||||
% Infinity norm
|
||||
%
|
||||
@@ -431,8 +481,8 @@ else if $x$ is a complex vector then it computes 1-norm as:
|
||||
|
||||
|
||||
\begin{verbatim}
|
||||
psb_geasum(x, desc_a, info)
|
||||
psb_norm1(x, desc_a, info)
|
||||
psb_geasum(x, desc_a, info [,global])
|
||||
psb_norm1(x, desc_a, info [,global])
|
||||
\end{verbatim}
|
||||
|
||||
\begin{table}[h]
|
||||
@@ -469,10 +519,18 @@ Scope: {\bf local} \\
|
||||
Type: {\bf required}\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: an object of type \descdata.
|
||||
\item[global] Specifies whether the computation should include the
|
||||
global reduction across all processes.\\
|
||||
Scope: {\bf global} \\
|
||||
Type: {\bf optional}.\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: a logical scalar.
|
||||
Default: \verb|global=.true.|\\
|
||||
|
||||
\item[\bf On Return]
|
||||
\item[Function value] is the 1-norm of vector $x$.\\
|
||||
Scope: {\bf global} \\
|
||||
Scope: {\bf global} unless the optional variable
|
||||
\verb|global=.false.| as been specified\\
|
||||
Specified as: a long precision real number.
|
||||
\item[info] Error code.\\
|
||||
Scope: {\bf local} \\
|
||||
@@ -481,6 +539,23 @@ Intent: {\bf out}.\\
|
||||
An integer value; 0 means no error has been detected.
|
||||
\end{description}
|
||||
|
||||
{\par\noindent\large\bfseries Notes}
|
||||
\begin{enumerate}
|
||||
\item The computation of a global result requires a global
|
||||
communication, which entails a significant overhead. It may be
|
||||
necessary and/or advisable to compute multiple norms at the same
|
||||
time; in this case, it is possible to improve the runtime efficiency
|
||||
by using the following scheme:
|
||||
\begin{lstlisting}
|
||||
vres(1) = psb_geasum(x1,desc_a,info,global=.false.)
|
||||
vres(2) = psb_geasum(x2,desc_a,info,global=.false.)
|
||||
vres(3) = psb_geasum(x3,desc_a,info,global=.false.)
|
||||
call psb_sum(ictxt,vres(1:3))
|
||||
\end{lstlisting}
|
||||
In this way the global communication, which for small sizes is a
|
||||
latency-bound operation, is invoked only once.
|
||||
\end{enumerate}
|
||||
|
||||
|
||||
\clearpage\subsection*{psb\_geasums --- Generalized 1-Norm of Vector}
|
||||
\addcontentsline{toc}{subsection}{psb\_geasums}
|
||||
@@ -587,8 +662,8 @@ Long Precision Real&Long Precision Complex & psb\_genrm2 \\
|
||||
\end{table}
|
||||
|
||||
\begin{verbatim}
|
||||
psb_genrm2(x, desc_a, info)
|
||||
psb_norm2(x, desc_a, info)
|
||||
psb_genrm2(x, desc_a, info [,global])
|
||||
psb_norm2(x, desc_a, info [,global])
|
||||
\end{verbatim}
|
||||
%% \syntax*{psb\_genrm2}{x, desc\_a, info, jx}
|
||||
|
||||
@@ -610,7 +685,13 @@ Scope: {\bf local} \\
|
||||
Type: {\bf required}\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: an object of type \descdata.
|
||||
%% \item[jx] the column index of global dense matrix $x$,
|
||||
\item[global] Specifies whether the computation should include the
|
||||
global reduction across all processes.\\
|
||||
Scope: {\bf global} \\
|
||||
Type: {\bf optional}.\\
|
||||
Intent: {\bf in}.\\
|
||||
Specified as: a logical scalar.
|
||||
Default: \verb|global=.true.|\\%% \item[jx] the column index of global dense matrix $x$,
|
||||
%% identifying the column of subvector $x$.\\
|
||||
%% Scope: {\bf global} \\
|
||||
%% Type: {\bf optional}; can only be present if $x$ is of rank 2.\\
|
||||
@@ -619,7 +700,8 @@ Specified as: an object of type \descdata.
|
||||
|
||||
\item[\bf On Return]
|
||||
\item[Function Value] is the 2-norm of subvector $x$.\\
|
||||
Scope: {\bf global} \\
|
||||
Scope: {\bf global} unless the optional variable
|
||||
\verb|global=.false.| as been specified\\
|
||||
Type: {\bf required} \\
|
||||
Specified as: a long precision real number.
|
||||
\item[info] Error code.\\
|
||||
@@ -629,6 +711,23 @@ Intent: {\bf out}.\\
|
||||
An integer value; 0 means no error has been detected.
|
||||
\end{description}
|
||||
|
||||
{\par\noindent\large\bfseries Notes}
|
||||
\begin{enumerate}
|
||||
\item The computation of a global result requires a global
|
||||
communication, which entails a significant overhead. It may be
|
||||
necessary and/or advisable to compute multiple norms at the same
|
||||
time; in this case, it is possible to improve the runtime efficiency
|
||||
by using the following scheme:
|
||||
\begin{lstlisting}
|
||||
vres(1) = psb_genrm2(x1,desc_a,info,global=.false.)
|
||||
vres(2) = psb_genrm2(x2,desc_a,info,global=.false.)
|
||||
vres(3) = psb_genrm2(x3,desc_a,info,global=.false.)
|
||||
call psb_nrm2(ictxt,vres(1:3))
|
||||
\end{lstlisting}
|
||||
In this way the global communication, which for small sizes is a
|
||||
latency-bound operation, is invoked only once.
|
||||
\end{enumerate}
|
||||
|
||||
|
||||
|
||||
\clearpage\subsection*{psb\_genrm2s --- Generalized 2-Norm of Vector}
|
||||
|
||||
Reference in New Issue
Block a user