Inheritance diagram for psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans:

[legend]

Collaboration diagram for psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans:

[legend]

Public Member Functions
def	__init__ (self, omp_directive="do", omp_schedule="auto")

def	__str__ (self)

def	omp_directive (self)

def	omp_directive (self, value)

def	omp_schedule (self)

def	omp_schedule (self, value)

def	apply (self, node, options=None)

Public Member Functions inherited from psyclone.psyir.transformations.parallel_loop_trans.ParallelLoopTrans
def	validate (self, node, options=None)

Public Member Functions inherited from psyclone.psyir.transformations.loop_trans.LoopTrans
def	name (self)

Public Attributes
	omp_schedule

	omp_directive

Additional Inherited Members
Static Public Attributes inherited from psyclone.psyir.transformations.parallel_loop_trans.ParallelLoopTrans
tuple	excluded_node_types = (nodes.Return, psyGen.HaloExchange, nodes.CodeBlock)

Static Public Attributes inherited from psyclone.psyir.transformations.loop_trans.LoopTrans
tuple	excluded_node_types = ()

Detailed Description

Adds an OpenMP directive to parallelise this loop. It can insert different
directives such as "omp do/for", "omp parallel do/for", "omp teams
distribute parallel do/for" or "omp loop" depending on the provided
parameters.
The OpenMP schedule to use can also be specified, but this will be ignored
in case of the "omp loop" (as the 'schedule' clause is not valid for this
specific directive). The configuration-defined 'reprod' parameter
also specifies whether a manual reproducible reproduction is to be used.
Note, reproducible in this case means obtaining the same results with the
same number of OpenMP threads, not for different numbers of OpenMP threads.

:param str omp_schedule: the OpenMP schedule to use. Defaults to 'auto'.
:param str omp_directive: choose which OpenMP loop directive to use. \
    Defaults to "omp do"

For example:

>>> from psyclone.psyir.frontend.fortran import FortranReader
>>> from psyclone.psyir.backend.fortran import FortranWriter
>>> from psyclone.psyir.nodes import Loop
>>> from psyclone.transformations import OMPLoopTrans, OMPParallelTrans
>>>
>>> psyir = FortranReader().psyir_from_source("""
...     subroutine my_subroutine()
...         integer, dimension(10, 10) :: A
...         integer :: i
...         integer :: j
...         do i = 1, 10
...             do j = 1, 10
...                 A(i, j) = 0
...             end do
...         end do
...     end subroutine
...     """)
>>> loop = psyir.walk(Loop)[0]
>>> omplooptrans1 = OMPLoopTrans(omp_schedule="dynamic",
...                              omp_directive="paralleldo")
>>> omplooptrans1.apply(loop)
>>> print(FortranWriter()(psyir))
subroutine my_subroutine()
  integer, dimension(10,10) :: a
  integer :: i
  integer :: j
<BLANKLINE>
  !$omp parallel do default(shared), private(i,j), schedule(dynamic)
  do i = 1, 10, 1
    do j = 1, 10, 1
      a(i,j) = 0
    enddo
  enddo
  !$omp end parallel do
<BLANKLINE>
end subroutine my_subroutine
<BLANKLINE>

Definition at line 55 of file omp_loop_trans.py.

Member Function Documentation

◆ apply()

def psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans.apply	(	self,
		node,
		options = `None`
	)

Apply the OMPLoopTrans transformation to the specified PSyIR Loop.

:param node: the supplied node to which we will apply the \
             OMPLoopTrans transformation
:type node: :py:class:`psyclone.psyir.nodes.Node`
:param options: a dictionary with options for transformations\
                and validation.
:type options: Optional[Dict[str, Any]]
:param bool options["reprod"]:
        indicating whether reproducible reductions should be used. \
        By default the value from the config file will be used.

Reimplemented from psyclone.psyir.transformations.parallel_loop_trans.ParallelLoopTrans.

Reimplemented in psyclone.transformations.Dynamo0p3OMPLoopTrans, psyclone.transformations.GOceanOMPParallelLoopTrans, and psyclone.transformations.OMPParallelLoopTrans.

Definition at line 227 of file omp_loop_trans.py.

     def apply(self, node, options=None):
         '''Apply the OMPLoopTrans transformation to the specified PSyIR Loop.
  
         :param node: the supplied node to which we will apply the \
                      OMPLoopTrans transformation
         :type node: :py:class:`psyclone.psyir.nodes.Node`
         :param options: a dictionary with options for transformations\
                         and validation.
         :type options: Optional[Dict[str, Any]]
         :param bool options["reprod"]:
                 indicating whether reproducible reductions should be used. \
                 By default the value from the config file will be used.
  
         '''
         if not options:
             options = {}
         self._reprod = options.get("reprod",
                                    Config.get().reproducible_reductions)
  
         if self._reprod:
             # When reprod is True, the variables th_idx and nthreads are
             # expected to be declared in the scope.
             root = node.ancestor(Routine)
  
             symtab = root.symbol_table
             try:
                 symtab.lookup_with_tag("omp_thread_index")
             except KeyError:
                 symtab.new_symbol(
                     "th_idx", tag="omp_thread_index",
                     symbol_type=DataSymbol, datatype=INTEGER_TYPE)
             try:
                 symtab.lookup_with_tag("omp_num_threads")
             except KeyError:
                 symtab.new_symbol(
                     "nthreads", tag="omp_num_threads",
                     symbol_type=DataSymbol, datatype=INTEGER_TYPE)
  
         super().apply(node, options)

References psyclone.psyir.nodes.omp_directives.OMPDoDirective._reprod, and psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._reprod.

◆ omp_directive() [1/2]

def psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans.omp_directive ( self )

:returns: the type of OMP directive that this transformation will \
    insert.
:rtype: str

Definition at line 131 of file omp_loop_trans.py.

     def omp_directive(self):
         '''
         :returns: the type of OMP directive that this transformation will \
             insert.
         :rtype: str
         '''
         return self._omp_directive
  

References psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._omp_directive.

◆ omp_directive() [2/2]

def psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans.omp_directive	(	self,
		value
	)

:param str value: the type of OMP directive to add.

:raises TypeError: if the provided value is not a valid str.

Definition at line 140 of file omp_loop_trans.py.

     def omp_directive(self, value):
         '''
         :param str value: the type of OMP directive to add.
  
         :raises TypeError: if the provided value is not a valid str.
         '''
         if not isinstance(value, str) or value not in VALID_OMP_DIRECTIVES:
             raise TypeError(
                 f"The {type(self).__name__}.omp_directive property must be "
                 f"a str with the value of {VALID_OMP_DIRECTIVES}"
                 f" but found a '{type(value).__name__}' with value '{value}'.")
         self._omp_directive = value
  

References psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._omp_directive.

◆ omp_schedule() [1/2]

def psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans.omp_schedule ( self )

:returns: the OpenMP schedule that will be specified by \
    this transformation.
:rtype: str

Definition at line 154 of file omp_loop_trans.py.

     def omp_schedule(self):
         '''
         :returns: the OpenMP schedule that will be specified by \
             this transformation.
         :rtype: str
  
         '''
         return self._omp_schedule
  

References psyclone.psyir.nodes.omp_directives.OMPDoDirective._omp_schedule, and psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._omp_schedule.

Here is the caller graph for this function:

◆ omp_schedule() [2/2]

def psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans.omp_schedule	(	self,
		value
	)

:param str value: Sets the OpenMP schedule value that will be \
    specified by this transformation, unless adding an OMP Loop \
    directive (in which case it is not applicable).

:raises TypeError: if the provided value is not a string.
:raises ValueError: if the provided string is not a valid OpenMP \
    schedule format.

Definition at line 164 of file omp_loop_trans.py.

     def omp_schedule(self, value):
         '''
         :param str value: Sets the OpenMP schedule value that will be \
             specified by this transformation, unless adding an OMP Loop \
             directive (in which case it is not applicable).
  
         :raises TypeError: if the provided value is not a string.
         :raises ValueError: if the provided string is not a valid OpenMP \
             schedule format.
         '''
  
         if not isinstance(value, str):
             raise TypeError(
                 f"The OMPLoopTrans.omp_schedule property must be a 'str'"
                 f" but found a '{type(value).__name__}'.")
  
         # Some schedules have an optional chunk size following a ','
         value_parts = value.split(',')
         if value_parts[0].lower() not in OMPScheduleClause.VALID_OMP_SCHEDULES:
             raise ValueError(
                 f"Valid OpenMP schedules are "
                 f"{OMPScheduleClause.VALID_OMP_SCHEDULES} but got "
                 f"'{value_parts[0]}'.")
  
         if len(value_parts) > 1:
             if value_parts[0] == "auto":
                 raise ValueError("Cannot specify a chunk size when using an "
                                  "OpenMP schedule of 'auto'.")
             try:
                 int(value_parts[1].strip())
             except ValueError as err:
                 raise ValueError(f"Supplied OpenMP schedule '{value}' has an "
                                  f"invalid chunk-size.") from err
  
         self._omp_schedule = value
  

References psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._omp_directive, psyclone.psyir.nodes.omp_directives.OMPDoDirective._omp_schedule, psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._omp_schedule, psyclone.psyir.nodes.omp_directives.OMPDoDirective._reprod, and psyclone.psyir.transformations.omp_loop_trans.OMPLoopTrans._reprod.

Here is the caller graph for this function:

The documentation for this class was generated from the following file:

/home/docs/checkouts/readthedocs.org/user_builds/psyclone-ref/checkouts/latest/src/psyclone/psyir/transformations/omp_loop_trans.py

Public Member Functions

Public Attributes