(Closes #3032, #3031) updates to improve profiling

changelog

@@ -1,3 +1,7 @@
+   46) PR 3033 for #3031 and #3032. Prevent inserting profiling callipers to
+   NEMO functions and any elemental function. Also remove pure attribute when
+   inserting psy_data wrappers.
+
    45) PR #2948 for #2950. Update extraction transformation to perform
    extraction and driver creation at lowering
 

examples/nemo/scripts/utils.py

@@ -35,10 +35,13 @@
 
 ''' Utilities file to parallelise Nemo code. '''
 
+from typing import List, Union
+
 from psyclone.domain.common.transformations import KernelModuleInlineTrans
 from psyclone.psyir.nodes import (
-    Assignment, Loop, Directive, Reference, CodeBlock, ArrayReference,
-    Call, Return, IfBlock, Routine, IntrinsicCall, StructureReference)
+    Assignment, Loop, Directive, Node, Reference, CodeBlock, ArrayReference,
+    Call, Return, IfBlock, Routine, Schedule, IntrinsicCall,
+    StructureReference)
 from psyclone.psyir.symbols import (
     DataSymbol, INTEGER_TYPE, ScalarType, RoutineSymbol)
 from psyclone.psyir.transformations import (
@@ -501,19 +504,28 @@ def insert_explicit_loop_parallelism(
             continue
 
 
-def add_profiling(children):
+def add_profiling(children: Union[List[Node], Schedule]):
     '''
-    Walks down the PSyIR and inserts the largest possible profiling regions.
-    Code that contains directives is excluded.
+    Walks down the PSyIR and inserts the largest possible profiling regions
+    in place. Code inside functions or that contains directives is excluded.
 
-    :param children: sibling nodes in the PSyIR to which to attempt to add \
-                     profiling regions.
-    :type children: list of :py:class:`psyclone.psyir.nodes.Node`
+    :param children: a Schedule or sibling nodes in the PSyIR to which to
+        attempt to add profiling regions.
 
     '''
+    if children and isinstance(children, Schedule):
+        # If we are given a Schedule, we look at its children.
+        children = children.children
+
     if not children:
         return
 
+    # We do not want profiling calipers inside functions (such as the
+    # PSyclone-generated comparison functions).
+    parent_routine = children[0].ancestor(Routine)
+    if parent_routine and parent_routine.return_symbol:
+        return
+
     node_list = []
     for child in children[:]:
         # Do we want this node to be included in a profiling region?

src/psyclone/psyir/transformations/psy_data_trans.py

@@ -201,6 +201,8 @@ def validate(self, nodes, options=None):
         :raises TransformationError: if there will be a name clash between \
             any existing symbols and those that must be imported from the \
             appropriate PSyData library.
+        :raises TransformationError: if the target nodes are within an
+                                     ELEMENTAL routine.
 
         '''
         # pylint: disable=too-many-branches
@@ -263,6 +265,13 @@ def validate(self, nodes, options=None):
                 except KeyError:
                     pass
 
+        parent_routine = node_list[0].ancestor(Routine)
+        if parent_routine and parent_routine.symbol.is_elemental:
+            raise TransformationError(
+                f"Cannot add PSyData calls inside ELEMENTAL routine "
+                f"'{parent_routine.symbol.name}' because it would change its "
+                f"semantics.")
+
         super().validate(node_list, options)
 
     # ------------------------------------------------------------------------
@@ -272,6 +281,9 @@ def apply(self, nodes, options=None):
         schedule - i.e. enclose the specified Nodes in the
         schedule within a single PSyData region.
 
+        Note that if the nodes are within a routine that previously had the
+        `pure` attribute, this attribute is removed.
+
         :param nodes: can be a single node or a list of nodes.
         :type nodes: :py:obj:`psyclone.psyir.nodes.Node` or list of \
                      :py:obj:`psyclone.psyir.nodes.Node`
@@ -314,6 +326,12 @@ def apply(self, nodes, options=None):
             node_list, symbol_table=table, options=options)
         parent.addchild(psy_data_node, position)
 
+        # If we've added PSyData calls to a pure routine then it is
+        # no longer pure.
+        parent_routine = node_list[0].ancestor(Routine)
+        if parent_routine and parent_routine.symbol.is_pure:
+            parent_routine.symbol.is_pure = False
+
 
 # =============================================================================
 # For AutoAPI documentation generation

src/psyclone/tests/psyir/transformations/psy_data_trans_test.py

@@ -40,10 +40,9 @@
 
 from psyclone.configuration import Config
 from psyclone.errors import InternalError
-from psyclone.psyir.nodes import PSyDataNode
-from psyclone.psyir.transformations import (PSyDataTrans,
-                                            ReadOnlyVerifyTrans,
-                                            TransformationError)
+from psyclone.psyir.nodes import Assignment, Loop, PSyDataNode, Routine
+from psyclone.psyir.transformations import (
+    OMPLoopTrans, PSyDataTrans, ReadOnlyVerifyTrans, TransformationError)
 from psyclone.tests.utilities import get_invoke
 
 
@@ -85,7 +84,44 @@ def test_psy_data_trans_basic():
         children[0] is node
 
 
-# -----------------------------------------------------------------------------
+def test_psy_data_trans_validate_not_inside_loop_directive(fortran_reader):
+    '''
+    Check that the transformation refuses to add caliper nodes between
+    a loop-directive and the associated loop.
+
+    '''
+    otrans = OMPLoopTrans()
+    psytrans = PSyDataTrans()
+    psyir = fortran_reader.psyir_from_source('''\
+    subroutine a_test()
+      integer :: i, va(10)
+      do i = 1, 10
+        va(i) = 5
+      end do
+    end subroutine a_test''')
+    loop = psyir.walk(Loop)[0]
+    otrans.apply(loop)
+    with pytest.raises(TransformationError) as err:
+        psytrans.validate(loop)
+    assert ("A PSyData node cannot be inserted between an OpenMP/ACC "
+            "directive and the loop(s)" in str(err.value))
+
+
+def test_psy_data_trans_validate_no_elemental(fortran_reader):
+    '''Check that the transformation refuses to act on an elemental routine.'''
+    data_trans = PSyDataTrans()
+    psyir = fortran_reader.psyir_from_source('''\
+    elemental real function a_test(var)
+      real, intent(in) :: var
+      a_test = var*var
+    end function a_test''')
+    assign = psyir.walk(Assignment)[0]
+    with pytest.raises(TransformationError) as err:
+        data_trans.validate(assign)
+    assert ("Cannot add PSyData calls inside ELEMENTAL routine 'a_test' "
+            "because it would change its semantics" in str(err.value))
+
+
 def test_class_definitions(fortran_writer):
     '''Tests if the class-prefix can be set and behaves as expected.
     '''
@@ -202,3 +238,22 @@ def test_trans_with_shape_function(monkeypatch, fortran_reader,
     out = fortran_writer(psyir)
     assert 'PreDeclareVariable("dummy", dummy)' in out
     assert 'ProvideVariable("dummy", dummy)' in out
+
+
+def test_psy_data_trans_remove_pure(fortran_reader):
+    '''
+    Test that applying the transformation to a pure routine causes that
+    attribute to be removed.
+
+    '''
+    psyir = fortran_reader.psyir_from_source('''\
+    pure subroutine so_clean(var)
+      integer, intent(inout) :: var
+      var = var*var
+    end subroutine so_clean
+    ''')
+    routine = psyir.walk(Routine)[0]
+    assert routine.symbol.is_pure
+    psytrans = PSyDataTrans()
+    psytrans.apply(routine.children)
+    assert not routine.symbol.is_pure