From 577db396135869838cc4b699d72609b3a2156471 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Sun, 8 May 2016 21:00:41 -0500
Subject: [PATCH] Add Instruction.no_sync_with
---
doc/ref_kernel.rst | 15 +++++++++---
loopy/kernel/creation.py | 49 ++++++++++++++++++++++++++++------------
loopy/kernel/data.py | 36 +++++++++++++++++++++++++----
loopy/preprocess.py | 4 +++-
loopy/schedule.py | 3 +++
loopy/version.py | 2 +-
6 files changed, 86 insertions(+), 23 deletions(-)
diff --git a/doc/ref_kernel.rst b/doc/ref_kernel.rst
index d151a2128..a323fff52 100644
--- a/doc/ref_kernel.rst
+++ b/doc/ref_kernel.rst
@@ -217,9 +217,9 @@ These are usually key-value pairs. The following attributes are recognized:
dependency is that the code generated for this instruction is required to
appear textually after all of these dependees' generated code.
- Identifiers here are allowed to be wildcards as defined by
- the Python module :mod:`fnmatchcase`. This is helpful in conjunction
- with ``id_prefix``.
+ Identifiers here are allowed to be wildcards as defined by the Python
+ function :func:`fnmatch.fnmatchcase`. This is helpful in conjunction with
+ ``id_prefix``.
.. note::
@@ -242,6 +242,15 @@ These are usually key-value pairs. The following attributes are recognized:
heuristic and indicate that the specified list of dependencies is
exhaustive.
+* ``nosync=id1:id2`` prescribes that no barrier synchronization is necessary
+ the instructions with identifiers ``id1`` and ``id2`` to the, even if
+ a dependency chain exists and variables are accessed in an apparently
+ racy way.
+
+ Identifiers here are allowed to be wildcards as defined by the Python
+ function :func:`fnmatch.fnmatchcase`. This is helpful in conjunction with
+ ``id_prefix``.
+
* ``priority=integer`` sets the instructions priority to the value
``integer``. Instructions with higher priority will be scheduled sooner,
if possible. Note that the scheduler may still schedule a lower-priority
diff --git a/loopy/kernel/creation.py b/loopy/kernel/creation.py
index aedc1edc6..034c9dd82 100644
--- a/loopy/kernel/creation.py
+++ b/loopy/kernel/creation.py
@@ -207,6 +207,7 @@ def parse_insn(insn):
if insn_match is not None:
depends_on = None
depends_on_is_final = False
+ no_sync_with = None
insn_groups = None
conflicts_with_groups = None
insn_id = None
@@ -256,6 +257,11 @@ def parse_insn(insn):
intern(dep.strip()) for dep in opt_value.split(":")
if dep.strip())
+ elif opt_key == "nosync" and opt_value is not None:
+ no_sync_with = frozenset(
+ intern(dep.strip()) for dep in opt_value.split(":")
+ if dep.strip())
+
elif opt_key == "groups" and opt_value is not None:
insn_groups = frozenset(
intern(grp.strip()) for grp in opt_value.split(":")
@@ -319,6 +325,7 @@ def parse_insn(insn):
else insn_id),
depends_on=depends_on,
depends_on_is_final=depends_on_is_final,
+ no_sync_with=no_sync_with,
groups=insn_groups,
conflicts_with_groups=conflicts_with_groups,
forced_iname_deps_is_final=forced_iname_deps_is_final,
@@ -1021,25 +1028,39 @@ def apply_default_order_to_args(kernel, default_order):
# {{{ resolve wildcard insn dependencies
+def find_matching_insn_ids(knl, dep):
+ from fnmatch import fnmatchcase
+
+ return [
+ other_insn.id
+ for other_insn in knl.instructions
+ if fnmatchcase(other_insn.id, dep)]
+
+
+def resove_wildcard_insn_ids(knl, deps):
+ new_deps = []
+ for dep in deps:
+ matches = find_matching_insn_ids(knl, dep)
+
+ if matches:
+ new_deps.extend(matches)
+ else:
+ # Uh, best we can do
+ new_deps.append(dep)
+
+ return frozenset(new_deps)
+
+
def resolve_wildcard_deps(knl):
new_insns = []
- from fnmatch import fnmatchcase
for insn in knl.instructions:
if insn.depends_on is not None:
- new_deps = set()
- for dep in insn.depends_on:
- match_count = 0
- for other_insn in knl.instructions:
- if fnmatchcase(other_insn.id, dep):
- new_deps.add(other_insn.id)
- match_count += 1
-
- if match_count == 0:
- # Uh, best we can do
- new_deps.add(dep)
-
- insn = insn.copy(depends_on=frozenset(new_deps))
+ insn = insn.copy(
+ depends_on=resove_wildcard_insn_ids(knl, insn.depends_on),
+ no_sync_with=resove_wildcard_insn_ids(
+ knl, insn.no_sync_with),
+ )
new_insns.append(insn)
diff --git a/loopy/kernel/data.py b/loopy/kernel/data.py
index 0e0638491..c4433d5e3 100644
--- a/loopy/kernel/data.py
+++ b/loopy/kernel/data.py
@@ -430,6 +430,8 @@ class InstructionBase(Record):
An (otherwise meaningless) identifier that is unique within
a :class:`loopy.kernel.LoopKernel`.
+ .. rubric:: Instruction ordering
+
.. attribute:: depends_on
a :class:`frozenset` of :attr:`id` values of :class:`Instruction` instances
@@ -460,6 +462,21 @@ class InstructionBase(Record):
(see :class:`InstructionBase.groups`) may not be active when this
instruction is scheduled.
+ .. attribute:: priority
+
+ Scheduling priority, an integer. Higher means 'execute sooner'.
+ Default 0.
+
+ .. rubric :: Synchronization
+
+ .. attribute:: no_sync_with
+
+ a :class:`frozenset` of :attr:`id` values of :class:`Instruction` instances
+ with which no barrier synchronization is necessary, even given the existence
+ of a dependency chain and apparently conflicting writes
+
+ .. rubric:: Conditionals
+
.. attribute:: predicates
a :class:`frozenset` of variable names the conjunction (logical and) of
@@ -467,6 +484,8 @@ class InstructionBase(Record):
should be run. Each variable name may, optionally, be preceded by
an exclamation point, indicating negation.
+ .. rubric:: Iname dependencies
+
.. attribute:: forced_iname_deps_is_final
A :class:`bool` determining whether :attr:`forced_iname_deps` constitutes
@@ -478,10 +497,7 @@ class InstructionBase(Record):
dependencies *or* constitute the entire list of iname dependencies,
depending on the value of :attr:`forced_iname_deps_is_final`.
- .. attribute:: priority
-
- Scheduling priority, an integer. Higher means 'execute sooner'.
- Default 0.
+ .. rubric:: Iname dependencies
.. attribute:: boostable
@@ -495,6 +511,8 @@ class InstructionBase(Record):
may need to be boosted, as a heuristic help for the scheduler.
Also allowed to be *None*.
+ .. rubric:: Tagging
+
.. attribute:: tags
A tuple of string identifiers that can be used to identify groups
@@ -512,12 +530,14 @@ class InstructionBase(Record):
fields = set("id depends_on depends_on_is_final "
"groups conflicts_with_groups "
+ "no_sync_with "
"predicates "
"forced_iname_deps_is_final forced_iname_deps "
"priority boostable boostable_into".split())
def __init__(self, id, depends_on, depends_on_is_final,
groups, conflicts_with_groups,
+ no_sync_with,
forced_iname_deps_is_final, forced_iname_deps, priority,
boostable, boostable_into, predicates, tags,
insn_deps=None, insn_deps_is_final=None):
@@ -541,6 +561,9 @@ class InstructionBase(Record):
if conflicts_with_groups is None:
conflicts_with_groups = frozenset()
+ if no_sync_with is None:
+ no_sync_with = frozenset()
+
if forced_iname_deps_is_final is None:
forced_iname_deps_is_final = False
@@ -574,6 +597,7 @@ class InstructionBase(Record):
id=id,
depends_on=depends_on,
depends_on_is_final=depends_on_is_final,
+ no_sync_with=no_sync_with,
groups=groups, conflicts_with_groups=conflicts_with_groups,
forced_iname_deps_is_final=forced_iname_deps_is_final,
forced_iname_deps=forced_iname_deps,
@@ -967,6 +991,7 @@ class Assignment(InstructionBase):
depends_on_is_final=None,
groups=None,
conflicts_with_groups=None,
+ no_sync_with=None,
forced_iname_deps_is_final=None,
forced_iname_deps=frozenset(),
boostable=None, boostable_into=None, tags=None,
@@ -980,6 +1005,7 @@ class Assignment(InstructionBase):
depends_on_is_final=depends_on_is_final,
groups=groups,
conflicts_with_groups=conflicts_with_groups,
+ no_sync_with=no_sync_with,
forced_iname_deps_is_final=forced_iname_deps_is_final,
forced_iname_deps=forced_iname_deps,
boostable=boostable,
@@ -1134,6 +1160,7 @@ class CInstruction(InstructionBase):
read_variables=frozenset(), assignees=frozenset(),
id=None, depends_on=None, depends_on_is_final=None,
groups=None, conflicts_with_groups=None,
+ no_sync_with=None,
forced_iname_deps_is_final=None, forced_iname_deps=frozenset(),
priority=0, boostable=None, boostable_into=None,
predicates=frozenset(), tags=None,
@@ -1153,6 +1180,7 @@ class CInstruction(InstructionBase):
depends_on=depends_on,
depends_on_is_final=depends_on_is_final,
groups=groups, conflicts_with_groups=conflicts_with_groups,
+ no_sync_with=no_sync_with,
forced_iname_deps_is_final=forced_iname_deps_is_final,
forced_iname_deps=forced_iname_deps,
boostable=boostable,
diff --git a/loopy/preprocess.py b/loopy/preprocess.py
index e30d3bcb3..93e898068 100644
--- a/loopy/preprocess.py
+++ b/loopy/preprocess.py
@@ -400,7 +400,9 @@ def add_default_dependencies(kernel):
% var)
if len(var_writers) == 1:
- auto_deps.update(var_writers - set([insn.id]))
+ auto_deps.update(
+ var_writers
+ - set([insn.id]))
# }}}
diff --git a/loopy/schedule.py b/loopy/schedule.py
index 4bacc43b9..b606ba360 100644
--- a/loopy/schedule.py
+++ b/loopy/schedule.py
@@ -1043,6 +1043,9 @@ def get_barrier_needing_dependency(kernel, target, source, reverse, var_kind):
if reverse:
source, target = target, source
+ if source.id in target.no_sync_with:
+ return None
+
# {{{ check that a dependency exists
dep_descr = None
diff --git a/loopy/version.py b/loopy/version.py
index 9fa881d02..cd9f45ac3 100644
--- a/loopy/version.py
+++ b/loopy/version.py
@@ -32,4 +32,4 @@ except ImportError:
else:
_islpy_version = islpy.version.VERSION_TEXT
-DATA_MODEL_VERSION = "v24-islpy%s" % _islpy_version
+DATA_MODEL_VERSION = "v25-islpy%s" % _islpy_version
--
GitLab