diff --git a/loopy/codegen/bounds.py b/loopy/codegen/bounds.py
index 51c0dca36b4159670e0b04ab7cd02f7e5275dcd5..17641ec019e0eccaec74951070ddb401cf3d11f1 100644
--- a/loopy/codegen/bounds.py
+++ b/loopy/codegen/bounds.py
@@ -23,14 +23,10 @@ THE SOFTWARE.
"""
-
-
import islpy as isl
from islpy import dim_type
-
-
def constraint_to_code(ccm, cns):
if cns.is_equality():
comp_op = "=="
@@ -40,6 +36,7 @@ def constraint_to_code(ccm, cns):
from loopy.symbolic import constraint_to_expr
return "%s %s 0" % (ccm(constraint_to_expr(cns), 'i'), comp_op)
+
# {{{ bounds check generator
def get_bounds_checks(domain, check_inames, implemented_domain,
@@ -71,6 +68,7 @@ def get_bounds_checks(domain, check_inames, implemented_domain,
# }}}
+
# {{{ on which inames may a conditional depend?
def get_usable_inames_for_conditional(kernel, sched_index):
@@ -102,7 +100,4 @@ def get_usable_inames_for_conditional(kernel, sched_index):
# }}}
-
-
-
# vim: foldmethod=marker
diff --git a/loopy/codegen/control.py b/loopy/codegen/control.py
index fa42edf88ab6c42681a3dba494b6d15ed1cfe3e6..57257fe1ba50596724267f4825da6e1efea9a860 100644
--- a/loopy/codegen/control.py
+++ b/loopy/codegen/control.py
@@ -25,16 +25,10 @@ THE SOFTWARE.
"""
-
-
-
-
from loopy.codegen import CodeGenerationState, gen_code_block
import islpy as isl
-
-
def get_admissible_conditional_inames_for(kernel, sched_index):
"""This function disallows conditionals on local-idx tagged
inames if there is a barrier nested somewhere within.
@@ -55,8 +49,6 @@ def get_admissible_conditional_inames_for(kernel, sched_index):
return result
-
-
def generate_code_for_sched_index(kernel, sched_index, codegen_state):
from loopy.schedule import (EnterLoop, RunInstruction, Barrier)
@@ -76,8 +68,8 @@ def generate_code_for_sched_index(kernel, sched_index, codegen_state):
elif tag is None or isinstance(tag, (LoopedIlpTag, ForceSequentialTag)):
func = generate_sequential_loop_dim_code
else:
- raise RuntimeError("encountered (invalid) EnterLoop for '%s', tagged '%s'"
- % (sched_item.iname, tag))
+ raise RuntimeError("encountered (invalid) EnterLoop "
+ "for '%s', tagged '%s'" % (sched_item.iname, tag))
return func(kernel, sched_index, codegen_state)
@@ -105,8 +97,6 @@ def generate_code_for_sched_index(kernel, sched_index, codegen_state):
% type(sched_item))
-
-
def remove_inames_for_shared_hw_axes(kernel, cond_inames):
"""
See if cond_inames contains references to two (or more) inames that
@@ -138,8 +128,6 @@ def remove_inames_for_shared_hw_axes(kernel, cond_inames):
return frozenset(cond_inames - multi_use_inames)
-
-
def build_loop_nest(kernel, sched_index, codegen_state):
# Most of the complexity of this function goes towards finding groups of
# instructions that can be nested inside a shared conditional.
@@ -180,8 +168,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
# {{{ pass 2: find admissible conditional inames for each sibling schedule item
admissible_cond_inames = [
- get_admissible_conditional_inames_for(kernel, sched_index)
- for sched_index in my_sched_indices]
+ get_admissible_conditional_inames_for(kernel, i)
+ for i in my_sched_indices]
# }}}
@@ -210,7 +198,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
# so we can safely overapproximate here.
overapproximate=True)
- def build_insn_group(sched_indices_and_cond_inames, codegen_state, done_group_lengths=set()):
+ def build_insn_group(sched_indices_and_cond_inames, codegen_state,
+ done_group_lengths=set()):
# done_group_lengths serves to prevent infinite recursion by imposing a
# bigger and bigger minimum size on the group of shared inames found.
@@ -224,7 +213,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
# Keep growing schedule item group as long as group fulfills minimum
# size requirement.
- bounds_check_cache = BoundsCheckCache(kernel, codegen_state.implemented_domain)
+ bounds_check_cache = BoundsCheckCache(
+ kernel, codegen_state.implemented_domain)
current_iname_set = cond_inames
@@ -236,7 +226,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
candidate_group_length += 1
continue
- other_sched_index, other_cond_inames = sched_indices_and_cond_inames[candidate_group_length-1]
+ other_sched_index, other_cond_inames = \
+ sched_indices_and_cond_inames[candidate_group_length-1]
current_iname_set = current_iname_set & other_cond_inames
# {{{ see which inames are actually used in group
@@ -244,7 +235,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
# And only generate conditionals for those.
from loopy.schedule import find_used_inames_within
used_inames = set()
- for subsched_index, _ in sched_indices_and_cond_inames[0:candidate_group_length]:
+ for subsched_index, _ in \
+ sched_indices_and_cond_inames[0:candidate_group_length]:
used_inames |= find_used_inames_within(kernel, subsched_index)
# }}}
@@ -255,7 +247,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
bounds_checks = bounds_check_cache(only_unshared_inames)
if bounds_checks or candidate_group_length == 1:
- # length-1 must always be an option to reach the recursion base case below
+ # length-1 must always be an option to reach the recursion base
+ # case below
found_hoists.append((candidate_group_length, bounds_checks))
candidate_group_length += 1
@@ -283,7 +276,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
if group_length == 1:
# group only contains starting schedule item
- result = [generate_code_for_sched_index(kernel, sched_index, new_codegen_state)]
+ result = [generate_code_for_sched_index(
+ kernel, sched_index, new_codegen_state)]
else:
# recurse with a bigger done_group_lengths
result = build_insn_group(
@@ -295,7 +289,8 @@ def build_loop_nest(kernel, sched_index, codegen_state):
from loopy.codegen import wrap_in_if
from loopy.codegen.bounds import constraint_to_code
result = [wrap_in_if(
- [constraint_to_code(codegen_state.c_code_mapper, cns) for cns in bounds_checks],
+ [constraint_to_code(codegen_state.c_code_mapper, cns)
+ for cns in bounds_checks],
gen_code_block(result))]
return result + build_insn_group(
diff --git a/loopy/codegen/instruction.py b/loopy/codegen/instruction.py
index 0858a67d933c63e05582a02efea9741e5c792f1e..ab603aa6f6fd334a851ff3a58bb29af52b2ccafb 100644
--- a/loopy/codegen/instruction.py
+++ b/loopy/codegen/instruction.py
@@ -24,20 +24,17 @@ THE SOFTWARE.
"""
-
-
import islpy as isl
-
-
def wrap_in_bounds_checks(ccm, domain, check_inames, implemented_domain, stmt):
from loopy.codegen.bounds import get_bounds_checks, constraint_to_code
bounds_checks = get_bounds_checks(
domain, check_inames,
implemented_domain, overapproximate=False)
- bounds_check_set = isl.Set.universe(domain.get_space()).add_constraints(bounds_checks)
+ bounds_check_set = isl.Set.universe(domain.get_space()) \
+ .add_constraints(bounds_checks)
bounds_check_set, new_implemented_domain = isl.align_two(
bounds_check_set, implemented_domain)
new_implemented_domain = new_implemented_domain & bounds_check_set
@@ -51,8 +48,6 @@ def wrap_in_bounds_checks(ccm, domain, check_inames, implemented_domain, stmt):
return stmt, new_implemented_domain
-
-
def generate_instruction_code(kernel, insn, codegen_state):
from loopy.codegen import GeneratedInstruction
@@ -91,7 +86,9 @@ def generate_instruction_code(kernel, insn, codegen_state):
ast=S(r'printf("write %s[%s]\n", %s);'
% (insn.get_assignee_var_name(),
",".join(len(idx) * ["%d"]),
- ",".join(ccm(i, prec=None, type_context="i") for i in idx))),
+ ",".join(
+ ccm(i, prec=None, type_context="i")
+ for i in idx))),
implemented_domain=None),
result
])
@@ -99,7 +96,4 @@ def generate_instruction_code(kernel, insn, codegen_state):
return result
-
-
-
# vim: foldmethod=marker
diff --git a/loopy/codegen/loop.py b/loopy/codegen/loop.py
index 41118cbf173c2ede733da51a1960f4676bd31e16..a20ec818bd18e83719c5ea7b7f26fd6ff41f4956 100644
--- a/loopy/codegen/loop.py
+++ b/loopy/codegen/loop.py
@@ -23,17 +23,12 @@ THE SOFTWARE.
"""
-
-
from loopy.codegen import gen_code_block
import islpy as isl
from islpy import dim_type
from loopy.codegen.control import build_loop_nest
-
-
-
# {{{ conditional-minimizing slab decomposition
def get_slab_decomposition(kernel, iname, sched_index, codegen_state):
@@ -115,6 +110,7 @@ def get_slab_decomposition(kernel, iname, sched_index, codegen_state):
# }}}
+
# {{{ unrolled loops
def generate_unroll_loop(kernel, sched_index, codegen_state):
@@ -144,6 +140,7 @@ def generate_unroll_loop(kernel, sched_index, codegen_state):
# }}}
+
def intersect_kernel_with_slab(kernel, slab, iname):
hdi = kernel.get_home_domain_index(iname)
home_domain = kernel.domains[hdi]
@@ -156,8 +153,10 @@ def intersect_kernel_with_slab(kernel, slab, iname):
# {{{ hw-parallel loop
-def set_up_hw_parallel_loops(kernel, sched_index, codegen_state, hw_inames_left=None):
- from loopy.kernel.data import UniqueTag, HardwareParallelTag, LocalIndexTag, GroupIndexTag
+def set_up_hw_parallel_loops(kernel, sched_index, codegen_state,
+ hw_inames_left=None):
+ from loopy.kernel.data import (
+ UniqueTag, HardwareParallelTag, LocalIndexTag, GroupIndexTag)
if hw_inames_left is None:
hw_inames_left = [iname
@@ -235,6 +234,7 @@ def set_up_hw_parallel_loops(kernel, sched_index, codegen_state, hw_inames_left=
# }}}
+
# {{{ sequential loop
def generate_sequential_loop_dim_code(kernel, sched_index, codegen_state):