OPEN CATALOG

Best practices for performance

RMK003: potential temporary variable for the loop which might be privatizable, thus enabling the loop parallelization

Issue #

Loop is a potential parallelization opportunity if it can be confirmed that variable X is a temporary

Relevance #

Determining the data scoping of all variables in a loop is critical to decide whether it can be parallelized. Just like scalars, arrays can be loop temporary variables, meaning that their value is computed and used on each iteration regardless of other iterations. Since they are only used in a specific iteration, temporaries can be safely privatized upon loop parallelization. In order to determine whether an array is a temporary, all its usages must be completely analyzed. This is not trivial and many tools struggle to do it, even when they can successfully do so for scalar temporaries. Therefore, the developer should analyze array usages to detect temporaries and ensure that their data scoping is private to create the most efficient parallel version of the loop.

Actions #

Analyze the data scoping of the variable X to determine whether it is a temporary and privatize it when parallelizing if it is.