llvm-project/polly/lib/External/isl/isl_test_python.py

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

450 lines
11 KiB
Python
Raw Normal View History

# Copyright 2016-2017 Tobias Grosser
#
# Use of this software is governed by the MIT license
#
# Written by Tobias Grosser, Weststrasse 47, CH-8003, Zurich
import sys
import isl
# Test that isl objects can be constructed.
#
# This tests:
# - construction from a string
# - construction from an integer
# - static constructor without a parameter
# - conversion construction
[Polly] Update ISL to isl-0.22.1-87-gfee05a13. The primary motivation is to fix an assertion failure in isl_basic_map_alloc_equality: isl_assert(ctx, room_for_con(bmap, 1), return -1); Although the assertion does not occur anymore, I could not identify which of ISL's commits fixed it. Compared to the previous ISL version, Polly requires some changes for this update * Since ISL commit 20d3574 "perform parameter alignment by modifying both arguments to function" isl_*_gist_* and similar functions do not always align the paramter list anymore. This caused the parameter lists in JScop files to become out-of-sync. Since many regression tests use JScop files with a fixed parameter list and order, we explicitly call align_params to ensure a predictable parameter list. * ISL changed some return types to isl_size, a typedef of (signed) int. This caused some issues where the return type was unsigned int before: - No overload for std::max(unsigned,isl_size) - It cause additional 'mixed signed/unsigned comparison' warnings. Since they do not break compilation, and sizes larger than 2^31 were never supported, I am going to fix it separately. * With the change to isl_size, commit 57d547 "isl_*_list_size: return isl_size" also changed the return value in case of an error from 0 to -1. This caused undefined looping over isl_iterator since the 'end iterator' got index -1, never reached from the 'begin iterator' with index 0. * Some internal changes in ISL caused the number of operations to increase when determining access ranges to determine aliasing overlaps. In one test, this caused exceeding the default limit of 800000. The operations-limit was disabled for this test.
2020-02-11 04:51:33 +08:00
# - construction of empty union set
#
# The tests to construct from integers and strings cover functionality that
# is also tested in the parameter type tests, but here the presence of
# multiple overloaded constructors and overload resolution is tested.
#
def test_constructors():
zero1 = isl.val("0")
assert(zero1.is_zero())
zero2 = isl.val(0)
assert(zero2.is_zero())
zero3 = isl.val.zero()
assert(zero3.is_zero())
bs = isl.basic_set("{ [1] }")
result = isl.set("{ [1] }")
s = isl.set(bs)
assert(s.is_equal(result))
[Polly] Update ISL to isl-0.22.1-87-gfee05a13. The primary motivation is to fix an assertion failure in isl_basic_map_alloc_equality: isl_assert(ctx, room_for_con(bmap, 1), return -1); Although the assertion does not occur anymore, I could not identify which of ISL's commits fixed it. Compared to the previous ISL version, Polly requires some changes for this update * Since ISL commit 20d3574 "perform parameter alignment by modifying both arguments to function" isl_*_gist_* and similar functions do not always align the paramter list anymore. This caused the parameter lists in JScop files to become out-of-sync. Since many regression tests use JScop files with a fixed parameter list and order, we explicitly call align_params to ensure a predictable parameter list. * ISL changed some return types to isl_size, a typedef of (signed) int. This caused some issues where the return type was unsigned int before: - No overload for std::max(unsigned,isl_size) - It cause additional 'mixed signed/unsigned comparison' warnings. Since they do not break compilation, and sizes larger than 2^31 were never supported, I am going to fix it separately. * With the change to isl_size, commit 57d547 "isl_*_list_size: return isl_size" also changed the return value in case of an error from 0 to -1. This caused undefined looping over isl_iterator since the 'end iterator' got index -1, never reached from the 'begin iterator' with index 0. * Some internal changes in ISL caused the number of operations to increase when determining access ranges to determine aliasing overlaps. In one test, this caused exceeding the default limit of 800000. The operations-limit was disabled for this test.
2020-02-11 04:51:33 +08:00
us = isl.union_set("{ A[1]; B[2, 3] }")
empty = isl.union_set.empty()
assert(us.is_equal(us.union(empty)))
# Test integer function parameters for a particular integer value.
#
def test_int(i):
val_int = isl.val(i)
val_str = isl.val(str(i))
assert(val_int.eq(val_str))
# Test integer function parameters.
#
# Verify that extreme values and zero work.
#
def test_parameters_int():
test_int(sys.maxsize)
test_int(-sys.maxsize - 1)
test_int(0)
# Test isl objects parameters.
#
# Verify that isl objects can be passed as lvalue and rvalue parameters.
# Also verify that isl object parameters are automatically type converted if
# there is an inheritance relation. Finally, test function calls without
# any additional parameters, apart from the isl object on which
# the method is called.
#
def test_parameters_obj():
a = isl.set("{ [0] }")
b = isl.set("{ [1] }")
c = isl.set("{ [2] }")
expected = isl.set("{ [i] : 0 <= i <= 2 }")
tmp = a.union(b)
res_lvalue_param = tmp.union(c)
assert(res_lvalue_param.is_equal(expected))
res_rvalue_param = a.union(b).union(c)
assert(res_rvalue_param.is_equal(expected))
a2 = isl.basic_set("{ [0] }")
assert(a.is_equal(a2))
two = isl.val(2)
half = isl.val("1/2")
res_only_this_param = two.inv()
assert(res_only_this_param.eq(half))
# Test different kinds of parameters to be passed to functions.
#
# This includes integer and isl object parameters.
#
def test_parameters():
test_parameters_int()
test_parameters_obj()
# Test that isl objects are returned correctly.
#
# This only tests that after combining two objects, the result is successfully
# returned.
#
def test_return_obj():
one = isl.val("1")
two = isl.val("2")
three = isl.val("3")
res = one.add(two)
assert(res.eq(three))
# Test that integer values are returned correctly.
#
def test_return_int():
one = isl.val("1")
neg_one = isl.val("-1")
zero = isl.val("0")
assert(one.sgn() > 0)
assert(neg_one.sgn() < 0)
assert(zero.sgn() == 0)
# Test that isl_bool values are returned correctly.
#
# In particular, check the conversion to bool in case of true and false.
#
def test_return_bool():
empty = isl.set("{ : false }")
univ = isl.set("{ : }")
b_true = empty.is_empty()
b_false = univ.is_empty()
assert(b_true)
assert(not b_false)
# Test that strings are returned correctly.
# Do so by calling overloaded isl.ast_build.from_expr methods.
#
def test_return_string():
context = isl.set("[n] -> { : }")
build = isl.ast_build.from_context(context)
pw_aff = isl.pw_aff("[n] -> { [n] }")
set = isl.set("[n] -> { : n >= 0 }")
expr = build.expr_from(pw_aff)
expected_string = "n"
assert(expected_string == expr.to_C_str())
expr = build.expr_from(set)
expected_string = "n >= 0"
assert(expected_string == expr.to_C_str())
# Test that return values are handled correctly.
#
# Test that isl objects, integers, boolean values, and strings are
# returned correctly.
#
def test_return():
test_return_obj()
test_return_int()
test_return_bool()
test_return_string()
# Test that foreach functions are modeled correctly.
#
# Verify that closures are correctly called as callback of a 'foreach'
# function and that variables captured by the closure work correctly. Also
# check that the foreach function handles exceptions thrown from
# the closure and that it propagates the exception.
#
def test_foreach():
s = isl.set("{ [0]; [1]; [2] }")
list = []
def add(bs):
list.append(bs)
s.foreach_basic_set(add)
assert(len(list) == 3)
assert(list[0].is_subset(s))
assert(list[1].is_subset(s))
assert(list[2].is_subset(s))
assert(not list[0].is_equal(list[1]))
assert(not list[0].is_equal(list[2]))
assert(not list[1].is_equal(list[2]))
def fail(bs):
raise "fail"
caught = False
try:
s.foreach_basic_set(fail)
except:
caught = True
assert(caught)
[Polly] Update ISL to isl-0.22.1-87-gfee05a13. The primary motivation is to fix an assertion failure in isl_basic_map_alloc_equality: isl_assert(ctx, room_for_con(bmap, 1), return -1); Although the assertion does not occur anymore, I could not identify which of ISL's commits fixed it. Compared to the previous ISL version, Polly requires some changes for this update * Since ISL commit 20d3574 "perform parameter alignment by modifying both arguments to function" isl_*_gist_* and similar functions do not always align the paramter list anymore. This caused the parameter lists in JScop files to become out-of-sync. Since many regression tests use JScop files with a fixed parameter list and order, we explicitly call align_params to ensure a predictable parameter list. * ISL changed some return types to isl_size, a typedef of (signed) int. This caused some issues where the return type was unsigned int before: - No overload for std::max(unsigned,isl_size) - It cause additional 'mixed signed/unsigned comparison' warnings. Since they do not break compilation, and sizes larger than 2^31 were never supported, I am going to fix it separately. * With the change to isl_size, commit 57d547 "isl_*_list_size: return isl_size" also changed the return value in case of an error from 0 to -1. This caused undefined looping over isl_iterator since the 'end iterator' got index -1, never reached from the 'begin iterator' with index 0. * Some internal changes in ISL caused the number of operations to increase when determining access ranges to determine aliasing overlaps. In one test, this caused exceeding the default limit of 800000. The operations-limit was disabled for this test.
2020-02-11 04:51:33 +08:00
# Test the functionality of "every" functions.
#
# In particular, test the generic functionality and
# test that exceptions are properly propagated.
#
def test_every():
us = isl.union_set("{ A[i]; B[j] }")
def is_empty(s):
return s.is_empty()
assert(not us.every_set(is_empty))
def is_non_empty(s):
return not s.is_empty()
assert(us.every_set(is_non_empty))
def in_A(s):
return s.is_subset(isl.set("{ A[x] }"))
assert(not us.every_set(in_A))
def not_in_A(s):
return not s.is_subset(isl.set("{ A[x] }"))
assert(not us.every_set(not_in_A))
def fail(s):
raise "fail"
caught = False
try:
us.ever_set(fail)
except:
caught = True
assert(caught)
# Check basic construction of spaces.
#
def test_space():
unit = isl.space.unit()
set_space = unit.add_named_tuple("A", 3)
map_space = set_space.add_named_tuple("B", 2)
set = isl.set.universe(set_space)
map = isl.map.universe(map_space)
assert(set.is_equal(isl.set("{ A[*,*,*] }")))
assert(map.is_equal(isl.map("{ A[*,*,*] -> B[*,*] }")))
# Construct a simple schedule tree with an outer sequence node and
# a single-dimensional band node in each branch, with one of them
# marked coincident.
#
def construct_schedule_tree():
A = isl.union_set("{ A[i] : 0 <= i < 10 }")
B = isl.union_set("{ B[i] : 0 <= i < 20 }")
node = isl.schedule_node.from_domain(A.union(B))
node = node.child(0)
filters = isl.union_set_list(A).add(B)
node = node.insert_sequence(filters)
f_A = isl.multi_union_pw_aff("[ { A[i] -> [i] } ]")
node = node.child(0)
node = node.child(0)
node = node.insert_partial_schedule(f_A)
node = node.member_set_coincident(0, True)
node = node.ancestor(2)
f_B = isl.multi_union_pw_aff("[ { B[i] -> [i] } ]")
node = node.child(1)
node = node.child(0)
node = node.insert_partial_schedule(f_B)
node = node.ancestor(2)
return node.schedule()
# Test basic schedule tree functionality.
#
# In particular, create a simple schedule tree and
# - check that the root node is a domain node
# - test map_descendant_bottom_up
# - test foreach_descendant_top_down
# - test every_descendant
#
def test_schedule_tree():
schedule = construct_schedule_tree()
root = schedule.root()
assert(type(root) == isl.schedule_node_domain)
count = [0]
def inc_count(node):
count[0] += 1
return node
root = root.map_descendant_bottom_up(inc_count)
assert(count[0] == 8)
def fail_map(node):
raise "fail"
return node
caught = False
try:
root.map_descendant_bottom_up(fail_map)
except:
caught = True
assert(caught)
count = [0]
def inc_count(node):
count[0] += 1
return True
root.foreach_descendant_top_down(inc_count)
assert(count[0] == 8)
count = [0]
def inc_count(node):
count[0] += 1
return False
root.foreach_descendant_top_down(inc_count)
assert(count[0] == 1)
def is_not_domain(node):
return type(node) != isl.schedule_node_domain
assert(root.child(0).every_descendant(is_not_domain))
assert(not root.every_descendant(is_not_domain))
def fail(node):
raise "fail"
caught = False
try:
root.every_descendant(fail)
except:
caught = True
assert(caught)
domain = root.domain()
filters = [isl.union_set("{}")]
def collect_filters(node):
if type(node) == isl.schedule_node_filter:
filters[0] = filters[0].union(node.filter())
return True
root.every_descendant(collect_filters)
assert(domain.is_equal(filters[0]))
# Test marking band members for unrolling.
# "schedule" is the schedule created by construct_schedule_tree.
# It schedules two statements, with 10 and 20 instances, respectively.
# Unrolling all band members therefore results in 30 at-domain calls
# by the AST generator.
#
def test_ast_build_unroll(schedule):
root = schedule.root()
def mark_unroll(node):
if type(node) == isl.schedule_node_band:
node = node.member_set_ast_loop_unroll(0)
return node
root = root.map_descendant_bottom_up(mark_unroll)
schedule = root.schedule()
count_ast = [0]
def inc_count_ast(node, build):
count_ast[0] += 1
return node
build = isl.ast_build()
build = build.set_at_each_domain(inc_count_ast)
ast = build.node_from(schedule)
assert(count_ast[0] == 30)
# Test basic AST generation from a schedule tree.
#
# In particular, create a simple schedule tree and
# - generate an AST from the schedule tree
# - test at_each_domain
# - test unrolling
#
def test_ast_build():
schedule = construct_schedule_tree()
count_ast = [0]
def inc_count_ast(node, build):
count_ast[0] += 1
return node
build = isl.ast_build()
build_copy = build.set_at_each_domain(inc_count_ast)
ast = build.node_from(schedule)
assert(count_ast[0] == 0)
count_ast[0] = 0
ast = build_copy.node_from(schedule)
assert(count_ast[0] == 2)
build = build_copy
count_ast[0] = 0
ast = build.node_from(schedule)
assert(count_ast[0] == 2)
do_fail = True
count_ast_fail = [0]
def fail_inc_count_ast(node, build):
count_ast_fail[0] += 1
if do_fail:
raise "fail"
return node
build = isl.ast_build()
build = build.set_at_each_domain(fail_inc_count_ast)
caught = False
try:
ast = build.node_from(schedule)
except:
caught = True
assert(caught)
assert(count_ast_fail[0] > 0)
build_copy = build
build_copy = build_copy.set_at_each_domain(inc_count_ast)
count_ast[0] = 0
ast = build_copy.node_from(schedule)
assert(count_ast[0] == 2)
count_ast_fail[0] = 0
do_fail = False
ast = build.node_from(schedule)
assert(count_ast_fail[0] == 2)
test_ast_build_unroll(schedule)
# Test basic AST expression generation from an affine expression.
#
def test_ast_build_expr():
pa = isl.pw_aff("[n] -> { [n + 1] }")
build = isl.ast_build.from_context(pa.domain())
op = build.expr_from(pa)
assert(type(op) == isl.ast_expr_op_add)
assert(op.n_arg() == 2)
# Test the isl Python interface
#
# This includes:
# - Object construction
# - Different parameter types
# - Different return types
# - Foreach functions
[Polly] Update ISL to isl-0.22.1-87-gfee05a13. The primary motivation is to fix an assertion failure in isl_basic_map_alloc_equality: isl_assert(ctx, room_for_con(bmap, 1), return -1); Although the assertion does not occur anymore, I could not identify which of ISL's commits fixed it. Compared to the previous ISL version, Polly requires some changes for this update * Since ISL commit 20d3574 "perform parameter alignment by modifying both arguments to function" isl_*_gist_* and similar functions do not always align the paramter list anymore. This caused the parameter lists in JScop files to become out-of-sync. Since many regression tests use JScop files with a fixed parameter list and order, we explicitly call align_params to ensure a predictable parameter list. * ISL changed some return types to isl_size, a typedef of (signed) int. This caused some issues where the return type was unsigned int before: - No overload for std::max(unsigned,isl_size) - It cause additional 'mixed signed/unsigned comparison' warnings. Since they do not break compilation, and sizes larger than 2^31 were never supported, I am going to fix it separately. * With the change to isl_size, commit 57d547 "isl_*_list_size: return isl_size" also changed the return value in case of an error from 0 to -1. This caused undefined looping over isl_iterator since the 'end iterator' got index -1, never reached from the 'begin iterator' with index 0. * Some internal changes in ISL caused the number of operations to increase when determining access ranges to determine aliasing overlaps. In one test, this caused exceeding the default limit of 800000. The operations-limit was disabled for this test.
2020-02-11 04:51:33 +08:00
# - Every functions
# - Spaces
# - Schedule trees
# - AST generation
# - AST expression generation
#
test_constructors()
test_parameters()
test_return()
test_foreach()
[Polly] Update ISL to isl-0.22.1-87-gfee05a13. The primary motivation is to fix an assertion failure in isl_basic_map_alloc_equality: isl_assert(ctx, room_for_con(bmap, 1), return -1); Although the assertion does not occur anymore, I could not identify which of ISL's commits fixed it. Compared to the previous ISL version, Polly requires some changes for this update * Since ISL commit 20d3574 "perform parameter alignment by modifying both arguments to function" isl_*_gist_* and similar functions do not always align the paramter list anymore. This caused the parameter lists in JScop files to become out-of-sync. Since many regression tests use JScop files with a fixed parameter list and order, we explicitly call align_params to ensure a predictable parameter list. * ISL changed some return types to isl_size, a typedef of (signed) int. This caused some issues where the return type was unsigned int before: - No overload for std::max(unsigned,isl_size) - It cause additional 'mixed signed/unsigned comparison' warnings. Since they do not break compilation, and sizes larger than 2^31 were never supported, I am going to fix it separately. * With the change to isl_size, commit 57d547 "isl_*_list_size: return isl_size" also changed the return value in case of an error from 0 to -1. This caused undefined looping over isl_iterator since the 'end iterator' got index -1, never reached from the 'begin iterator' with index 0. * Some internal changes in ISL caused the number of operations to increase when determining access ranges to determine aliasing overlaps. In one test, this caused exceeding the default limit of 800000. The operations-limit was disabled for this test.
2020-02-11 04:51:33 +08:00
test_every()
test_space()
test_schedule_tree()
test_ast_build()
test_ast_build_expr()