Translating array sections

The rules for translating array sections are more complicated than any other part of the basic translation scheme.

We will break it down into three cases: the case where there are no scalar subscripts--integer or distributed index; the case where integer scalar subscripts appear in sequential dimensions only; and the general case where scalar subscripts may appear in distributed dimensions. The scheme for translating the first case is illustrated in Figure A.27.

The macro PROCESS_SUBSCRIPTS will be defined here in a tail-recursive way. The intention is that it should be expanded to a compile-time loop over $s$.

Let $R$ be the rank of the subscripted array. If $s = R$, then the macro $\mbox{\it PROCESS\_SUBSCRIPTS}\left(v, r, a, s\right)$ is empty. Otherwise, if $\mbox{\it subs}_s$ is the degenerate triplet, :, then

$$\begin{minipage}[t]{\linewidth}\begin{tabbing} $\mbox{\it PR... ...SCRIPTS}\left(v, r+1, a, s+1\right)$\end{tabbing}\end{minipage}$$

Otherwise, if $\mbox{\it subs}_s$ is the triplet, $e_{\mbox{\small lo}}$: $e_{\mbox{\small hi}}$: $e_{\mbox{\small stp}}$, and the $s$th dimension of $a$ is distributed, then

$$\begin{minipage}[t]{\linewidth}\begin{tabbing} $\mbox{\it PR... ...SCRIPTS}\left(v, r+1, a, s+1\right)$\end{tabbing}\end{minipage}$$

where $a'_s = {\bf T}_{s}\left[{a}\right]$, $e_{\mbox{\small lo}}' = {\bf T}\left[{e_{\mbox{\small lo}}}\right]$, $e_{\mbox{\small hi}}' = {\bf T}\left[{e_{\mbox{\small hi}}}\right]$, and $e_{\mbox{\small stp}}' = {\bf T}\left[{e_{\mbox{\small stp}}}\right]$. Otherwise, if $\mbox{\it subs}_s$ is the triplet, $e_{\mbox{\small lo}}$: $e_{\mbox{\small hi}}$: $e_{\mbox{\small stp}}$, and the $s$th dimension of $a$ is sequential, then

$$\begin{minipage}[t]{\linewidth}\begin{tabbing} $\mbox{\it PR... ...SCRIPTS}\left(v, r+1, a, s+1\right)$\end{tabbing}\end{minipage}$$

with definitions as above. Two similar cases using the two-argument form of subrng() take care of triplets of the form $e_{\mbox{\small lo}}$: $e_{\mbox{\small hi}}$. Otherwise, if $\mbox{\it subs}_s$ is the splitting subscript, <>, then

$$\begin{minipage}[t]{\linewidth}\begin{tabbing} $\mbox{\it PR... ...SCRIPTS}\left(v, r+2, a, s+1\right)$\end{tabbing}\end{minipage}$$

where $x$, $u$, and $z$ are the names of new temporaries.

Figure A.28: Translation of array section without any scalar subscripts in distributed dimensions.

SOURCE:

$$\begin{minipage}[t]{\linewidth}\begin{tabbing} $v$\verb$ = $... ... $\mbox{\it subs}_{R-1}$\verb$]] ;$ \end{tabbing}\end{minipage}$$

TRANSLATION:

$$\begin{minipage}[t]{\linewidth}\begin{tabbing} \verb$int $$b... ...t]$\verb$.group, $$b$\verb$) ;$ \\ \end{tabbing}\end{minipage}$$

where:

$$\begin{array}{l} \mbox{The expression $v$ is the assigned a... ... {\it PROCESS\_SUBSCRIPTS} is defined in the text.} \end{array}$$