Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

(Closes #2499) scalarization transformation implementation #2563

Open
wants to merge 9 commits into
base: master
Choose a base branch
from
Open
3 changes: 3 additions & 0 deletions src/psyclone/psyir/transformations/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -103,6 +103,8 @@
ReplaceInductionVariablesTrans
from psyclone.psyir.transformations.reference2arrayrange_trans import \
Reference2ArrayRangeTrans
from psyclone.psyir.transformations.scalarization_trans import \
ScalarizationTrans


# For AutoAPI documentation generation
Expand Down Expand Up @@ -143,4 +145,5 @@
'Reference2ArrayRangeTrans',
'RegionTrans',
'ReplaceInductionVariablesTrans',
'ScalarizationTrans',
'TransformationError']
219 changes: 219 additions & 0 deletions src/psyclone/psyir/transformations/scalarization_trans.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,219 @@
# -----------------------------------------------------------------------------
# BSD 3-Clause License
#
# Copyright (c) 2017-2024, Science and Technology Facilities Council.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
#
# * Redistributions of source code must retain the above copyright notice, this
# list of conditions and the following disclaimer.
#
# * Redistributions in binary form must reproduce the above copyright notice,
# this list of conditions and the following disclaimer in the documentation
# and/or other materials provided with the distribution.
#
# * Neither the name of the copyright holder nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
# FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
# COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
# INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
# BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
# LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
# ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
# POSSIBILITY OF SUCH DAMAGE.
# -----------------------------------------------------------------------------
# Author: A. B. G. Chalk, STFC Daresbury Lab

'''This module provides the sclarization transformation class.'''

import itertools

from psyclone.core import VariablesAccessInfo
from psyclone.psyGen import Kern
from psyclone.psyir.nodes import Assignment, Call, CodeBlock, IfBlock, \
Reference, Routine
from psyclone.psyir.symbols import DataSymbol
from psyclone.psyir.transformations.loop_trans import LoopTrans


class ScalarizationTrans(LoopTrans):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

When you add the docstring, can you bring what you wrote in apply here and add a simple example, e.g. like src/psyclone/psyir/transformations/hoist_trans.py . So then we just need to list the autoclass under the available transformations section of doc/user_guide/transformations.rst and can be tested with python -m doctest src/psyclone/psyir/transformations/scalarization_trans.py. The apply docstring can be a oneliner then.


@staticmethod
def _is_local_array(signature, var_accesses):
if not var_accesses[signature].is_array():
return False
base_symbol = var_accesses[signature].all_accesses[0].node.symbol
if not base_symbol.is_automatic:
return False

return True

@staticmethod
def _have_same_unmodified_index(signature, var_accesses):
array_indices = None
scalarizable = True
for access in var_accesses[signature].all_accesses:
if array_indices is None:
array_indices = access.component_indices
# For some reason using == on the component_lists doesn't work
elif array_indices[:] != access.component_indices[:]:
scalarizable = False
break
# For each index, we need to check they're not written to in
# the loop.
flattened_indices = list(itertools.chain.from_iterable(
array_indices))
for index in flattened_indices:
# Index may not be a Reference, so we need to loop over the
# References
for ref in index.walk(Reference):
sig, _ = ref.get_signature_and_indices()
if var_accesses[sig].is_written():
scalarizable = False
break

return scalarizable

@staticmethod
def _check_first_access_is_write(sig, var_accesses):
if var_accesses[sig].is_written_first():
return True
return False

@staticmethod
def _value_unused_after_loop(sig, node, var_accesses):
# Find the last access of the signature
last_access = var_accesses[sig].all_accesses[-1].node
# Find the next access to this symbol
next_access = last_access.next_access()
# If we don't use this again then this can be scalarized
if next_access is None:
return True

# If the next_access has an ancestor IfBlock and
# that isn't an ancestor of the loop then its not valid since
# we aren't tracking down what the condition-dependent next
# use really is.
if_ancestor = next_access.ancestor(IfBlock)
# If abs_position of if_ancestor is > node.abs_position
# its not an ancestor of us.
# Handles:
# if (some_condition) then
# x = next_access[i] + 1
if (if_ancestor is not None and
if_ancestor.abs_position > node.abs_position):
# Not a valid next_access pattern.
return False

# If next access is the RHS of an assignment then we need to
# skip it
# Handles:
# a = next_access[i] + 1
ancestor_assign = next_access.ancestor(Assignment)
if (ancestor_assign is not None and
ancestor_assign.lhs is not next_access):
return False

# If it has an ancestor that is a CodeBlock or Call or Kern
# then we can't guarantee anything, so we remove it.
# Handles: call my_func(next_access)
if (next_access.ancestor((CodeBlock, Call, Kern))
is not None):
return False

return True

def apply(self, node, options=None):
'''Apply the scalarization transformation to a loop.
All of the array accesses that are identified as being able to be
scalarized will be transformed by this transformation.

An array access will be scalarized if:
1. All accesses to the array use the same indexing statement.
2. All References contained in the indexing statement are not modified
inside of the loop (loop variables are ok).
3. The array symbol is either not accessed again or is written to
as its next access. If the next access is inside a conditional
that is not an ancestor of the input loop, then PSyclone will
assume that we cannot scalarize that value instead of attempting to
understand the control flow.
4. TODO - The array symbol is a local variable.

:param node: the supplied loop to apply scalarization to.
:type node: :py:class:`psyclone.psyir.nodes.Loop`
:param options: a dictionary with options for transformations.
:type options: Optional[Dict[str, Any]]

'''
# For each array reference in the Loop:
# Find every access to the same symbol in the loop
# They all have to be accessed with the same index statement, and
# that index needs to not be written to inside the loop body.
# For each symbol that meets this criteria, we then need to check the
# first access is a write
# Then, for each symbol still meeting this criteria, we need to find
# the next access outside of this loop. If its inside an ifblock that
# is not an ancestor of this loop then we refuse to scalarize for
# simplicity. Otherwise if its a read we can't scalarize safely.
# If its a write then this symbol can be scalarized.

var_accesses = VariablesAccessInfo(nodes=node.loop_body)

# Find all the ararys that are only accessed by a single index, and
# that index is only read inside the loop.
potential_targets = filter(
lambda sig:
ScalarizationTrans._is_local_array(sig, var_accesses),
var_accesses)
potential_targets = filter(
lambda sig:
ScalarizationTrans._have_same_unmodified_index(sig,
var_accesses),
potential_targets)
# potential_targets = self._find_potential_scalarizable_array_symbols(
# node, var_accesses)

# Now we need to check the first access is a write and remove those
# that aren't.
potential_targets = filter(
lambda sig:
ScalarizationTrans._check_first_access_is_write(sig,
var_accesses),
potential_targets)
# potential_targets = self._check_first_access_is_write(
# node, var_accesses, potential_targets)

# Check the values written to these arrays are not used after this loop
finalised_targets = filter(
lambda sig:
ScalarizationTrans._value_unused_after_loop(sig, node,
var_accesses),
potential_targets)
# finalised_targets = self._check_valid_following_access(
# node, var_accesses, potential_targets)

routine_table = node.ancestor(Routine).symbol_table
# For each finalised target we can replace them with a scalarized
# symbol
for target in finalised_targets:
target_accesses = var_accesses[target].all_accesses
first_access = target_accesses[0].node
symbol_type = first_access.symbol.datatype.datatype
symbol_name = first_access.symbol.name
scalar_symbol = routine_table.new_symbol(
root_name=f"{symbol_name}_scalar",
symbol_type=DataSymbol,
datatype=symbol_type)
ref_to_copy = Reference(scalar_symbol)
for access in target_accesses:
node = access.node
node.replace_with(ref_to_copy.copy())
Loading
Loading