`psyclone.psyir.transformations.intrinsics.product2loop_trans`#

Module providing a transformation from a PSyIR PRODUCT intrinsic to an equivalent PSyIR loop structure. This could be useful if the PRODUCT operator is not supported by the back-end, the required parallelisation approach, or if the performance in the inline code is better than the intrinsic.

Classes #

Product2LoopTrans: Provides a transformation from a PSyIR PRODUCT IntrinsicCall node to

class psyclone.psyir.transformations.intrinsics.product2loop_trans.Product2LoopTrans[source]#

Provides a transformation from a PSyIR PRODUCT IntrinsicCall node to an equivalent PSyIR loop structure that is suitable for running in parallel on CPUs and GPUs. Validity checks are also performed.

If PRODUCT contains a single positional argument which is an array, the maximum value of all of the elements in the array is returned in the the scalar R.

R = PRODUCT(ARRAY)

For example, if the array is two dimensional, the equivalent code for real data is:

R = 1.0
DO J=LBOUND(ARRAY,2),UBOUND(ARRAY,2)
  DO I=LBOUND(ARRAY,1),UBOUND(ARRAY,1)
    R = R * ARRAY(I,J)

If the mask argument is provided then the mask is used to determine whether the product is applied:

R = PRODUCT(ARRAY, mask=MOD(ARRAY, 2.0)==1)

If the array is two dimensional, the equivalent code for real data is:

R = 1.0
DO J=LBOUND(ARRAY,2),UBOUND(ARRAY,2)
  DO I=LBOUND(ARRAY,1),UBOUND(ARRAY,1)
    IF (MOD(ARRAY(I,J), 2.0)==1) THEN
      R = R * ARRAY(I,J)

The dimension argument is currently not supported and will result in a TransformationError exception being raised.

R = PRODUCT(ARRAY, dimension=2)

The array passed to PRODUCT may use any combination of array syntax, array notation, array sections and scalar bounds:

R = PRODUCT(ARRAY) ! array syntax
R = PRODUCT(ARRAY(:,:)) ! array notation
R = PRODUCT(ARRAY(1:10,lo:hi)) ! array sections
R = PRODUCT(ARRAY(1:10,:)) ! mix of array section and array notation
R = PRODUCT(ARRAY(1:10,2)) ! mix of array section and scalar bound

An example use of this transformation is given below:

>>> from psyclone.psyir.backend.fortran import FortranWriter
>>> from psyclone.psyir.frontend.fortran import FortranReader
>>> from psyclone.psyir.transformations import Product2LoopTrans
>>> code = ("subroutine product_test(array)\n"
...         "  real :: array(10,10)\n"
...         "  real :: result\n"
...         "  result = product(array)\n"
...         "end subroutine\n")
>>> psyir = FortranReader().psyir_from_source(code)
>>> product_node = psyir.children[0].children[0].children[1]
>>> Product2LoopTrans().apply(product_node)
>>> print(FortranWriter()(psyir))
subroutine product_test(array)
  real, dimension(10,10) :: array
  real :: result
  integer :: idx
  integer :: idx_1

  result = 1.0
  do idx = 1, 10, 1
    do idx_1 = 1, 10, 1
      result = result * array(idx_1,idx)
    enddo
  enddo

end subroutine product_test

Inheritance

apply(node, options=None, **kwargs)[source]#

Apply the Product2LoopTrans to the input node.

Parameters:

node (psyclone.psyir.nodes.IntrinsicCall) – the PRODUCT intrinsic to transform.
options (Optional[Dict[str, Any]]) – options for the transformation.

psyclone.psyir.transformations.intrinsics.product2loop_trans#

Classes#

`psyclone.psyir.transformations.intrinsics.product2loop_trans`#

Classes #