From 3694b9e82116051e48b484d9d803c209197370d3 Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Sun, 22 May 2022 17:49:22 +0200
Subject: [PATCH 1/7] corrected split optimisation
---
pystencils/astnodes.py | 2 +-
pystencils/backends/cbackend.py | 3 +-
pystencils/transformations.py | 9 ++--
pystencils_tests/test_transformations.py | 63 ++++++++++++++++++++++--
4 files changed, 68 insertions(+), 9 deletions(-)
diff --git a/pystencils/astnodes.py b/pystencils/astnodes.py
index ef0bcc6d7..f3ed2711c 100644
--- a/pystencils/astnodes.py
+++ b/pystencils/astnodes.py
@@ -448,7 +448,7 @@ class LoopOverCoordinate(Node):
def new_loop_with_different_body(self, new_body):
result = LoopOverCoordinate(new_body, self.coordinate_to_loop_over, self.start, self.stop,
self.step, self.is_block_loop)
- result.prefix_lines = [l for l in self.prefix_lines]
+ result.prefix_lines = [prefix_line for prefix_line in self.prefix_lines]
return result
def subs(self, subs_dict):
diff --git a/pystencils/backends/cbackend.py b/pystencils/backends/cbackend.py
index 486bd1260..499a80498 100644
--- a/pystencils/backends/cbackend.py
+++ b/pystencils/backends/cbackend.py
@@ -444,7 +444,8 @@ class CustomSympyPrinter(CCodePrinter):
def _print_Pow(self, expr):
"""Don't use std::pow function, for small integer exponents, write as multiplication"""
if isinstance(expr.exp, sp.Integer) and (-8 < expr.exp < 8):
- raise NotImplementedError("This pow should be simplified already?")
+ raise ValueError(f"This expression: {expr} contains a pow function that should be simplified already with "
+ f"a sequence of multiplications")
return super(CustomSympyPrinter, self)._print_Pow(expr)
# TODO don't print ones in sp.Mul
diff --git a/pystencils/transformations.py b/pystencils/transformations.py
index 43beefd25..06005cbf7 100644
--- a/pystencils/transformations.py
+++ b/pystencils/transformations.py
@@ -13,6 +13,7 @@ from pystencils.typing import (CastFunc, PointerType, StructType, TypedSymbol, g
ReinterpretCastFunc, get_next_parent_of_type, parents_of_type)
from pystencils.field import Field, FieldType
from pystencils.typing import FieldPointerSymbol
+from pystencils.sympyextensions import fast_subs
from pystencils.simp.assignment_collection import AssignmentCollection
from pystencils.slicing import normalize_slice
from pystencils.integer_functions import int_div
@@ -650,11 +651,11 @@ def split_inner_loop(ast_node: ast.Node, symbol_groups):
and which no symbol in a symbol group depends on, are not updated!
"""
all_loops = ast_node.atoms(ast.LoopOverCoordinate)
- inner_loop = [l for l in all_loops if l.is_innermost_loop]
+ inner_loop = [loop for loop in all_loops if loop.is_innermost_loop]
assert len(inner_loop) == 1, "Error in AST: multiple innermost loops. Was split transformation already called?"
inner_loop = inner_loop[0]
assert type(inner_loop.body) is ast.Block
- outer_loop = [l for l in all_loops if l.is_outermost_loop]
+ outer_loop = [loop for loop in all_loops if loop.is_outermost_loop]
assert len(outer_loop) == 1, "Error in AST, multiple outermost loops."
outer_loop = outer_loop[0]
@@ -688,8 +689,8 @@ def split_inner_loop(ast_node: ast.Node, symbol_groups):
assignment_group = []
for assignment in inner_loop.body.args:
if assignment.lhs in symbols_resolved:
- new_rhs = assignment.rhs.subs(
- symbols_with_temporary_array.items())
+ # use fast_subs here because it checks if multiplications should be evaluated or not
+ new_rhs = fast_subs(assignment.rhs, symbols_with_temporary_array)
if not isinstance(assignment.lhs, Field.Access) and assignment.lhs in symbol_group:
assert type(assignment.lhs) is TypedSymbol
new_ts = TypedSymbol(assignment.lhs.name, PointerType(assignment.lhs.dtype))
diff --git a/pystencils_tests/test_transformations.py b/pystencils_tests/test_transformations.py
index 3ede70a85..d6e6888b5 100644
--- a/pystencils_tests/test_transformations.py
+++ b/pystencils_tests/test_transformations.py
@@ -1,5 +1,7 @@
+import sympy as sp
+
import pystencils as ps
-from pystencils import TypedSymbol
+from pystencils import fields, TypedSymbol
from pystencils.astnodes import LoopOverCoordinate, SympyAssignment
from pystencils.typing import create_type
from pystencils.transformations import filtered_tree_iteration, get_loop_hierarchy, get_loop_counter_symbol_hierarchy
@@ -10,8 +12,8 @@ def test_loop_information():
update_rule = ps.Assignment(g[0, 0], f[0, 0])
ast = ps.create_kernel(update_rule)
- inner_loops = [l for l in filtered_tree_iteration(ast, LoopOverCoordinate, stop_type=SympyAssignment)
- if l.is_innermost_loop]
+ inner_loops = [loop for loop in filtered_tree_iteration(ast, LoopOverCoordinate, stop_type=SympyAssignment)
+ if loop.is_innermost_loop]
loop_order = []
for i in get_loop_hierarchy(inner_loops[0].args[0]):
@@ -23,3 +25,58 @@ def test_loop_information():
assert loop_symbols == [TypedSymbol("ctr_1", create_type("int"), nonnegative=True),
TypedSymbol("ctr_0", create_type("int"), nonnegative=True)]
+
+
+def test_split_optimisation():
+ src, dst = fields(f"src(9), dst(9): [2D]", layout='fzyx')
+
+ stencil = ((0, 0), (0, 1), (0, -1), (-1, 0), (1, 0), (-1, 1), (1, 1), (-1, -1), (1, -1))
+ w = [sp.Rational(4, 9)]
+ w += [sp.Rational(1, 9)] * 4
+ w += [sp.Rational(1, 36)] * 4
+ cs = sp.Rational(1, 3)
+
+ subexpressions = []
+ main_assignements = []
+
+ rho = sp.symbols("rho")
+ velo = sp.symbols("u_:2")
+
+ density = 0
+ velocity_x = 0
+ velocity_y = 0
+ for d in stencil:
+ density += src[d]
+ velocity_x += d[0] * src[d]
+ velocity_y += d[1] * src[d]
+
+ subexpressions.append(ps.Assignment(rho, density))
+ subexpressions.append(ps.Assignment(velo[0], velocity_x))
+ subexpressions.append(ps.Assignment(velo[1], velocity_y))
+
+ for i, d in enumerate(stencil):
+ u_d = velo[0] * d[0] + velo[1] * d[1]
+ u_2 = velo[0] * velo[0] + velo[1] * velo[1]
+
+ expr = w[i] * rho * (1 + u_d / cs + u_d ** 2 / (2 * cs ** 2) - u_2 / (2 * cs))
+
+ main_assignements.append(ps.Assignment(dst.center_vector[i], expr))
+
+ ac = ps.AssignmentCollection(main_assignments=main_assignements, subexpressions=subexpressions)
+
+ simplification_hint = {'density': rho,
+ 'velocity': (velo[0], velo[1]),
+ 'split_groups': [[rho, velo[0], velo[1], dst.center_vector[0]],
+ [dst.center_vector[1], dst.center_vector[2]],
+ [dst.center_vector[3], dst.center_vector[4]],
+ [dst.center_vector[5], dst.center_vector[6]],
+ [dst.center_vector[7], dst.center_vector[8]]]}
+
+ ac.simplification_hints = simplification_hint
+ ast = ps.create_kernel(ac)
+
+ code = ps.get_code_str(ast)
+ # after the split optimisation the two for loops are split into 6
+ assert code.count("for") == 6
+
+ print(code)
--
GitLab
From 49150419560cc096574414bef3af9dc465654f53 Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Mon, 23 May 2022 11:41:12 +0200
Subject: [PATCH 2/7] Added non evaluation
---
pystencils/cpu/vectorization.py | 26 ++++++++++++++------------
1 file changed, 14 insertions(+), 12 deletions(-)
diff --git a/pystencils/cpu/vectorization.py b/pystencils/cpu/vectorization.py
index b3236a3c5..78b6aac0f 100644
--- a/pystencils/cpu/vectorization.py
+++ b/pystencils/cpu/vectorization.py
@@ -3,13 +3,13 @@ from typing import Container, Union
import numpy as np
import sympy as sp
+from sympy import evaluate
from sympy.logic.boolalg import BooleanFunction, BooleanAtom
import pystencils.astnodes as ast
from pystencils.backends.simd_instruction_sets import get_supported_instruction_sets, get_vector_instruction_set
from pystencils.typing import (BasicType, PointerType, TypedSymbol, VectorType, CastFunc, collate_types,
get_type_of_expression, VectorMemoryAccess)
-from pystencils.fast_approximation import fast_division, fast_inv_sqrt, fast_sqrt
from pystencils.functions import DivFunc
from pystencils.field import Field
from pystencils.integer_functions import modulo_ceil, modulo_floor
@@ -133,7 +133,6 @@ def vectorize(kernel_ast: ast.KernelFunction, instruction_set: str = 'best',
vectorize_inner_loops_and_adapt_load_stores(kernel_ast, assume_aligned, nontemporal,
strided, keep_loop_stop, assume_sufficient_line_padding,
default_float_type)
- # is in vectorize_inner_loops_and_adapt_load_stores.. insert_vector_casts(kernel_ast, default_float_type)
def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontemporal_fields,
@@ -143,8 +142,8 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
vector_width = ast_node.instruction_set['width']
all_loops = filtered_tree_iteration(ast_node, ast.LoopOverCoordinate, stop_type=ast.SympyAssignment)
- inner_loops = [n for n in all_loops if n.is_innermost_loop]
- zero_loop_counters = {l.loop_counter_symbol: 0 for l in all_loops}
+ inner_loops = [loop for loop in all_loops if loop.is_innermost_loop]
+ zero_loop_counters = {loop.loop_counter_symbol: 0 for loop in all_loops}
for loop_node in inner_loops:
loop_range = loop_node.stop - loop_node.start
@@ -158,7 +157,9 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
loop_node.stop = new_stop
else:
cutting_point = modulo_floor(loop_range, vector_width) + loop_node.start
- loop_nodes = [l for l in cut_loop(loop_node, [cutting_point]).args if isinstance(l, ast.LoopOverCoordinate)]
+ with evaluate(False):
+ loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
+ if isinstance(loop, ast.LoopOverCoordinate)]
assert len(loop_nodes) in (0, 1, 2) # 2 for main and tail loop, 1 if loop range divisible by vector width
if len(loop_nodes) == 0:
continue
@@ -179,8 +180,7 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
successful = False
break
typed_symbol = base.label
- assert type(typed_symbol.dtype) is PointerType, \
- f"Type of access is {typed_symbol.dtype}, {indexed}"
+ assert type(typed_symbol.dtype) is PointerType, f"Type of access is {typed_symbol.dtype}, {indexed}"
vec_type = VectorType(typed_symbol.dtype.base_type, vector_width)
use_aligned_access = aligned_access and assume_aligned
@@ -202,7 +202,7 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
continue
loop_node.step = vector_width
- loop_node.subs(substitutions)
+ loop_node.fast_subs(substitutions)
vector_int_width = ast_node.instruction_set['intwidth']
arg_1 = CastFunc(loop_counter_symbol, VectorType(loop_counter_symbol.dtype, vector_int_width))
arg_2 = CastFunc(tuple(range(vector_int_width if type(vector_int_width) is int else 2)),
@@ -254,8 +254,7 @@ def mask_conditionals(loop_body):
def insert_vector_casts(ast_node, instruction_set, default_float_type='double'):
"""Inserts necessary casts from scalar values to vector values."""
- handled_functions = (sp.Add, sp.Mul, fast_division, fast_sqrt, fast_inv_sqrt, vec_any, vec_all, DivFunc,
- sp.UnevaluatedExpr, sp.Abs)
+ handled_functions = (sp.Add, sp.Mul, vec_any, vec_all, DivFunc, sp.UnevaluatedExpr, sp.Abs)
def visit_expr(expr, default_type='double'): # TODO Vectorization Revamp: get rid of default_type
if isinstance(expr, VectorMemoryAccess):
@@ -286,7 +285,7 @@ def insert_vector_casts(ast_node, instruction_set, default_float_type='double'):
if dtype is not int:
if dtype is np.float32:
default_type = 'float'
- expr = sp.Mul(dtype(expr.args[0]), *expr.args[1:])
+ expr = sp.Mul(dtype(expr.args[0]), *expr.args[1:], evaluate=False)
new_args = [visit_expr(a, default_type) for a in expr.args]
arg_types = [get_type_of_expression(a, default_float_type=default_type) for a in new_args]
if not any(type(t) is VectorType for t in arg_types):
@@ -296,7 +295,10 @@ def insert_vector_casts(ast_node, instruction_set, default_float_type='double'):
casted_args = [
CastFunc(a, target_type) if t != target_type and not isinstance(a, VectorMemoryAccess) else a
for a, t in zip(new_args, arg_types)]
- return expr.func(*casted_args)
+ if isinstance(expr.func, (sp.Mul, sp.Add)):
+ return expr.func(*casted_args, evaluate=False)
+ else:
+ return expr.func(*casted_args)
elif expr.func is sp.Pow:
new_arg = visit_expr(expr.args[0], default_type)
return expr.func(new_arg, expr.args[1])
--
GitLab
From 1fcf409f479c6476ee22c71b97fc38c73132f70e Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Mon, 23 May 2022 11:46:53 +0200
Subject: [PATCH 3/7] Removed fast approximations
---
pystencils_tests/test_vectorization.py | 31 --------------------------
1 file changed, 31 deletions(-)
diff --git a/pystencils_tests/test_vectorization.py b/pystencils_tests/test_vectorization.py
index 19f266b12..f526341ec 100644
--- a/pystencils_tests/test_vectorization.py
+++ b/pystencils_tests/test_vectorization.py
@@ -8,7 +8,6 @@ import sympy as sp
import pystencils as ps
from pystencils.backends.simd_instruction_sets import get_supported_instruction_sets, get_vector_instruction_set
from pystencils.cpu.vectorization import vectorize
-from pystencils.fast_approximation import insert_fast_sqrts, insert_fast_divisions
from pystencils.enums import Target
from pystencils.transformations import replace_inner_stride_with_one
@@ -19,7 +18,6 @@ else:
instruction_set = None
-
# TODO: Skip tests if no instruction set is available and check all codes if they are really vectorised !
def test_vector_type_propagation(instruction_set=instruction_set):
a, b, c, d, e = sp.symbols("a b c d e")
@@ -276,35 +274,6 @@ def test_vectorised_pow(instruction_set=instruction_set):
ast.compile()
-def test_vectorised_fast_approximations(instruction_set=instruction_set):
- # fast_approximations are a gpu thing
- arr = np.zeros((24, 24))
- f, g = ps.fields(f=arr, g=arr)
-
- expr = sp.sqrt(f[0, 0] + f[1, 0])
- assignment = ps.Assignment(g[0, 0], insert_fast_sqrts(expr))
- ast = ps.create_kernel(assignment)
- vectorize(ast, instruction_set=instruction_set)
-
- with pytest.raises(Exception):
- ast.compile()
-
- expr = f[0, 0] / f[1, 0]
- assignment = ps.Assignment(g[0, 0], insert_fast_divisions(expr))
- ast = ps.create_kernel(assignment)
- vectorize(ast, instruction_set=instruction_set)
-
- with pytest.raises(Exception):
- ast.compile()
-
- assignment = ps.Assignment(sp.Symbol("tmp"), 3 / sp.sqrt(f[0, 0] + f[1, 0]))
- ast = ps.create_kernel(insert_fast_sqrts(assignment))
- vectorize(ast, instruction_set=instruction_set)
-
- with pytest.raises(Exception):
- ast.compile()
-
-
def test_issue40(*_):
"""https://i10git.cs.fau.de/pycodegen/pystencils/-/issues/40"""
opt = {'instruction_set': "avx512", 'assume_aligned': False,
--
GitLab
From 4da1ce82b245b328f921aec43f04b4729a74b99a Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Mon, 23 May 2022 11:55:51 +0200
Subject: [PATCH 4/7] Try
---
pystencils/cpu/vectorization.py | 5 ++---
pystencils_tests/test_vectorization_specific.py | 3 ++-
2 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/pystencils/cpu/vectorization.py b/pystencils/cpu/vectorization.py
index 78b6aac0f..63552bbc7 100644
--- a/pystencils/cpu/vectorization.py
+++ b/pystencils/cpu/vectorization.py
@@ -157,9 +157,8 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
loop_node.stop = new_stop
else:
cutting_point = modulo_floor(loop_range, vector_width) + loop_node.start
- with evaluate(False):
- loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
- if isinstance(loop, ast.LoopOverCoordinate)]
+ loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
+ if isinstance(loop, ast.LoopOverCoordinate)]
assert len(loop_nodes) in (0, 1, 2) # 2 for main and tail loop, 1 if loop range divisible by vector width
if len(loop_nodes) == 0:
continue
diff --git a/pystencils_tests/test_vectorization_specific.py b/pystencils_tests/test_vectorization_specific.py
index 367250dda..49152a420 100644
--- a/pystencils_tests/test_vectorization_specific.py
+++ b/pystencils_tests/test_vectorization_specific.py
@@ -120,7 +120,8 @@ def test_alignment_and_correct_ghost_layers(gl_field, gl_kernel, instruction_set
update_rule = ps.Assignment(dst[0, 0], src[0, 0])
opt = {'instruction_set': instruction_set, 'assume_aligned': True,
'nontemporal': True, 'assume_inner_stride_one': True}
- config = pystencils.config.CreateKernelConfig(target=dh.default_target, cpu_vectorize_info=opt, ghost_layers=gl_kernel)
+ config = pystencils.config.CreateKernelConfig(target=dh.default_target,
+ cpu_vectorize_info=opt, ghost_layers=gl_kernel)
ast = ps.create_kernel(update_rule, config=config)
kernel = ast.compile()
if gl_kernel != gl_field:
--
GitLab
From 10b2337304df76b020b666748af6e8bd149a5eff Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Mon, 23 May 2022 12:05:33 +0200
Subject: [PATCH 5/7] Testing
---
pystencils/cpu/vectorization.py | 14 ++++++--------
1 file changed, 6 insertions(+), 8 deletions(-)
diff --git a/pystencils/cpu/vectorization.py b/pystencils/cpu/vectorization.py
index 63552bbc7..1daf1beee 100644
--- a/pystencils/cpu/vectorization.py
+++ b/pystencils/cpu/vectorization.py
@@ -157,8 +157,9 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
loop_node.stop = new_stop
else:
cutting_point = modulo_floor(loop_range, vector_width) + loop_node.start
- loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
- if isinstance(loop, ast.LoopOverCoordinate)]
+ with evaluate(False):
+ loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
+ if isinstance(loop, ast.LoopOverCoordinate)]
assert len(loop_nodes) in (0, 1, 2) # 2 for main and tail loop, 1 if loop range divisible by vector width
if len(loop_nodes) == 0:
continue
@@ -201,7 +202,7 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
continue
loop_node.step = vector_width
- loop_node.fast_subs(substitutions)
+ loop_node.subs(substitutions)
vector_int_width = ast_node.instruction_set['intwidth']
arg_1 = CastFunc(loop_counter_symbol, VectorType(loop_counter_symbol.dtype, vector_int_width))
arg_2 = CastFunc(tuple(range(vector_int_width if type(vector_int_width) is int else 2)),
@@ -284,7 +285,7 @@ def insert_vector_casts(ast_node, instruction_set, default_float_type='double'):
if dtype is not int:
if dtype is np.float32:
default_type = 'float'
- expr = sp.Mul(dtype(expr.args[0]), *expr.args[1:], evaluate=False)
+ expr = sp.Mul(dtype(expr.args[0]), *expr.args[1:])
new_args = [visit_expr(a, default_type) for a in expr.args]
arg_types = [get_type_of_expression(a, default_float_type=default_type) for a in new_args]
if not any(type(t) is VectorType for t in arg_types):
@@ -294,10 +295,7 @@ def insert_vector_casts(ast_node, instruction_set, default_float_type='double'):
casted_args = [
CastFunc(a, target_type) if t != target_type and not isinstance(a, VectorMemoryAccess) else a
for a, t in zip(new_args, arg_types)]
- if isinstance(expr.func, (sp.Mul, sp.Add)):
- return expr.func(*casted_args, evaluate=False)
- else:
- return expr.func(*casted_args)
+ return expr.func(*casted_args)
elif expr.func is sp.Pow:
new_arg = visit_expr(expr.args[0], default_type)
return expr.func(new_arg, expr.args[1])
--
GitLab
From 3e5566949e510289c50d4bbf939fbd76584cbff7 Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Mon, 23 May 2022 14:03:10 +0200
Subject: [PATCH 6/7] Testing
---
pystencils/backends/cbackend.py | 10 ++++++++--
pystencils/cpu/vectorization.py | 5 ++---
pystencils/transformations.py | 2 +-
3 files changed, 11 insertions(+), 6 deletions(-)
diff --git a/pystencils/backends/cbackend.py b/pystencils/backends/cbackend.py
index 499a80498..de9cb0d31 100644
--- a/pystencils/backends/cbackend.py
+++ b/pystencils/backends/cbackend.py
@@ -768,7 +768,10 @@ class VectorizedCustomSympyPrinter(CustomSympyPrinter):
return processed
def _print_Pow(self, expr):
- result = self._scalarFallback('_print_Pow', expr)
+ try:
+ result = self._scalarFallback('_print_Pow', expr)
+ except ValueError:
+ result = None
if result:
return result
@@ -800,7 +803,10 @@ class VectorizedCustomSympyPrinter(CustomSympyPrinter):
# noinspection PyProtectedMember
from sympy.core.mul import _keep_coeff
- result = self._scalarFallback('_print_Mul', expr)
+ if not inside_add:
+ result = self._scalarFallback('_print_Mul', expr)
+ else:
+ result = None
if result:
return result
diff --git a/pystencils/cpu/vectorization.py b/pystencils/cpu/vectorization.py
index 1daf1beee..d733822ca 100644
--- a/pystencils/cpu/vectorization.py
+++ b/pystencils/cpu/vectorization.py
@@ -157,9 +157,8 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
loop_node.stop = new_stop
else:
cutting_point = modulo_floor(loop_range, vector_width) + loop_node.start
- with evaluate(False):
- loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
- if isinstance(loop, ast.LoopOverCoordinate)]
+ loop_nodes = [loop for loop in cut_loop(loop_node, [cutting_point]).args
+ if isinstance(loop, ast.LoopOverCoordinate)]
assert len(loop_nodes) in (0, 1, 2) # 2 for main and tail loop, 1 if loop range divisible by vector width
if len(loop_nodes) == 0:
continue
diff --git a/pystencils/transformations.py b/pystencils/transformations.py
index 06005cbf7..cce06c2c8 100644
--- a/pystencils/transformations.py
+++ b/pystencils/transformations.py
@@ -2,7 +2,7 @@ import hashlib
import pickle
import warnings
from collections import OrderedDict
-from copy import deepcopy
+from copy import deepcopy, copy
from types import MappingProxyType
import sympy as sp
--
GitLab
From 5e0075341d83c66547f9d79206109c6a0fd33673 Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Mon, 23 May 2022 14:07:11 +0200
Subject: [PATCH 7/7] Fix flake8
---
pystencils/cpu/vectorization.py | 1 -
pystencils/transformations.py | 2 +-
2 files changed, 1 insertion(+), 2 deletions(-)
diff --git a/pystencils/cpu/vectorization.py b/pystencils/cpu/vectorization.py
index d733822ca..3141eb400 100644
--- a/pystencils/cpu/vectorization.py
+++ b/pystencils/cpu/vectorization.py
@@ -3,7 +3,6 @@ from typing import Container, Union
import numpy as np
import sympy as sp
-from sympy import evaluate
from sympy.logic.boolalg import BooleanFunction, BooleanAtom
import pystencils.astnodes as ast
diff --git a/pystencils/transformations.py b/pystencils/transformations.py
index cce06c2c8..06005cbf7 100644
--- a/pystencils/transformations.py
+++ b/pystencils/transformations.py
@@ -2,7 +2,7 @@ import hashlib
import pickle
import warnings
from collections import OrderedDict
-from copy import deepcopy, copy
+from copy import deepcopy
from types import MappingProxyType
import sympy as sp
--
GitLab