loki.transformations.argument_shape

Inter-procedural analysis passes to derive and augment argument array shapes.

A pair of utility Transformation classes that allows thea shape of array arguments with deferred dimensions to be derived from the calling context via inter-procedural analysis.

To infer the declared dimensions of array arguments ArgumentArrayShapeAnalysis needs to be applied first to set the shape property on respective Array symbols, before ExplicitArgumentArrayShapeTransformation can be applied in a reverse traversal order to apply the necessary changes to argument declarations and call signatures.

Classes

ArgumentArrayShapeAnalysis()	Infer shape of array arguments with deferred shape.
ExplicitArgumentArrayShapeTransformation()	Add dimensions to array arguments and adjust call signatures.

class ArgumentArrayShapeAnalysis

Bases: Transformation

Infer shape of array arguments with deferred shape.

An inter-procedural analysis pass that passively infers the shape symbols for argument arrays from calling contexts and sets the shape attribute on Array symbols accordingly.

The shape information is propagated from a caller to the called subroutine in a forward traversal of the call-tree. If the call-side shape of an array argument is either set, or has already been derived (possibly with conflicting information), this transformation will have no effect.

Note: This transformation does not affect the generated source code, as it only sets the shape property, which is ignored during the code generation step (fgen). To actively change the argument array declarations and the resulting source code, the ExplicitArgumentArrayShapeTransformation needs to be applied after this transformation.

transform_subroutine(routine, **kwargs)

Defines the transformation to apply to Subroutine items.

For transformations that modify Subroutine objects, this method should be implemented. It gets called via the dispatch method apply().

Parameters:

• routine (Subroutine) – The subroutine to be transformed.

• **kwargs (optional) – Keyword arguments for the transformation.

class ExplicitArgumentArrayShapeTransformation

Bases: Transformation

Add dimensions to array arguments and adjust call signatures.

Adjusts array argument declarations by inserting explicit shape variables according to the shape property of the Array symbol. This property can be derived from the calling context via ArgumentArrayShapeAnalysis.

If the Scalar symbol defining an array dimension is not yet known in the local Subroutine, it gets added to the call signature. In the caller routine, the respective Scalar argument is added to the CallStatement via keyword-argument notation.

Note: Since the CallStatement needs updating after the called Subroutine signature, this transformation has to be applied in reverse order via Scheduler.process(..., reverse=True).

reverse_traversal = True

transform_subroutine(routine, **kwargs)

Defines the transformation to apply to Subroutine items.

For transformations that modify Subroutine objects, this method should be implemented. It gets called via the dispatch method apply().

Parameters:

• routine (Subroutine) – The subroutine to be transformed.

• **kwargs (optional) – Keyword arguments for the transformation.