Skip to content
Snippets Groups Projects

Compare revisions

Changes are shown as if the source revision was being merged into the target revision. Learn more about comparing revisions.

Source

Select target project
No results found

Target

Select target project
No results found
Show changes
Hide whitespace changes
Inline Side-by-side
Showing
with 449 additions and 1024 deletions
pystencils_tests/test_print_unsupported_node.py deleted 100644 → 0
View file @ 1e7a1e6f
# -*- coding: utf-8 -*-
#
# Copyright © 2019 Stephan Seitz <stephan.seitz@fau.de>
#
# Distributed under terms of the GPLv3 license.
"""
"""
import pytest
import pystencils
from pystencils.backends.cbackend import CBackend
class UnsupportedNode(pystencils.astnodes.Node):
def __init__(self):
super().__init__()
def test_print_unsupported_node():
with pytest.raises(NotImplementedError, match='CBackend does not support node of type UnsupportedNode'):
CBackend()(UnsupportedNode())
pystencils_tests/test_size_and_layout_checks_llvm.py deleted 100644 → 0
View file @ 1e7a1e6f
import numpy as np
import pytest
from pystencils import Assignment, Field
try:
from pystencils.llvm import create_kernel, make_python_function
except ModuleNotFoundError:
pytest.importorskip("llvmlite")
def test_size_check():
"""Kernel with two fixed-sized fields creating with same size but calling with wrong size"""
src = np.zeros((20, 21, 9))
dst = np.zeros_like(src)
sym_src = Field.create_from_numpy_array("src", src, index_dimensions=1)
sym_dst = Field.create_from_numpy_array("dst", dst, index_dimensions=1)
update_rule = Assignment(sym_dst(0),
sym_src[-1, 1](1) + sym_src[1, -1](2))
ast = create_kernel([update_rule])
func = make_python_function(ast)
# change size of src field
new_shape = [a - 7 for a in src.shape]
src = np.zeros(new_shape)
dst = np.zeros(new_shape)
try:
func(src=src, dst=dst)
assert False, "Expected ValueError because fields with different sized where passed"
except ValueError:
pass
def test_fixed_size_mismatch_check():
"""Create kernel with two differently sized but constant fields """
src = np.zeros((20, 21, 9))
dst = np.zeros((21, 21, 9))
sym_src = Field.create_from_numpy_array("src", src, index_dimensions=1)
sym_dst = Field.create_from_numpy_array("dst", dst, index_dimensions=1)
update_rule = Assignment(sym_dst(0),
sym_src[-1, 1](1) + sym_src[1, -1](2))
try:
create_kernel([update_rule])
assert False, "Expected ValueError because fields with different sized where passed"
except ValueError:
pass
def test_fixed_and_variable_field_check():
"""Create kernel with two variable sized fields - calling them with different sizes"""
src = np.zeros((20, 21, 9))
sym_src = Field.create_from_numpy_array("src", src, index_dimensions=1)
sym_dst = Field.create_generic("dst", spatial_dimensions=2, index_dimensions=1)
update_rule = Assignment(sym_dst(0),
sym_src[-1, 1](1) + sym_src[1, -1](2))
try:
create_kernel([update_rule])
assert False, "Expected ValueError because fields with different sized where passed"
except ValueError:
pass
def test_two_variable_shaped_fields():
src = np.zeros((20, 21, 9))
dst = np.zeros((22, 21, 9))
sym_src = Field.create_generic("src", spatial_dimensions=2, index_dimensions=1)
sym_dst = Field.create_generic("dst", spatial_dimensions=2, index_dimensions=1)
update_rule = Assignment(sym_dst(0),
sym_src[-1, 1](1) + sym_src[1, -1](2))
ast = create_kernel([update_rule])
func = make_python_function(ast)
try:
func(src=src, dst=dst)
assert False, "Expected ValueError because fields with different sized where passed"
except ValueError:
pass
pystencils_tests/test_sliced_iteration.py deleted 100644 → 0
View file @ 1e7a1e6f
import numpy as np
import sympy as sp
from pystencils import Assignment, Field, TypedSymbol, create_kernel, make_slice
from pystencils.simp import sympy_cse_on_assignment_list
def test_sliced_iteration():
size = (4, 4)
src_arr = np.ones(size)
dst_arr = np.zeros_like(src_arr)
src_field = Field.create_from_numpy_array('src', src_arr)
dst_field = Field.create_from_numpy_array('dst', dst_arr)
a, b = sp.symbols("a b")
update_rule = Assignment(dst_field[0, 0],
(a * src_field[0, 1] + a * src_field[0, -1] +
b * src_field[1, 0] + b * src_field[-1, 0]) / 4)
x_end = TypedSymbol("x_end", "int")
s = make_slice[1:x_end, 1]
x_end_value = size[1] - 1
kernel = create_kernel(sympy_cse_on_assignment_list([update_rule]), iteration_slice=s).compile()
kernel(src=src_arr, dst=dst_arr, a=1.0, b=1.0, x_end=x_end_value)
expected_result = np.zeros(size)
expected_result[1:x_end_value, 1] = 1
np.testing.assert_almost_equal(expected_result, dst_arr)
def test_sliced_iteration_llvm():
import pytest
pytest.importorskip("llvmlite")
size = (4, 4)
src_arr = np.ones(size)
dst_arr = np.zeros_like(src_arr)
src_field = Field.create_from_numpy_array('src', src_arr)
dst_field = Field.create_from_numpy_array('dst', dst_arr)
a, b = sp.symbols("a b")
update_rule = Assignment(dst_field[0, 0],
(a * src_field[0, 1] + a * src_field[0, -1] +
b * src_field[1, 0] + b * src_field[-1, 0]) / 4)
x_end = TypedSymbol("x_end", "int")
s = make_slice[1:x_end, 1]
x_end_value = size[1] - 1
import pystencils.llvm as llvm_generator
ast = llvm_generator.create_kernel(sympy_cse_on_assignment_list([update_rule]), iteration_slice=s)
kernel = llvm_generator.make_python_function(ast)
kernel(src=src_arr, dst=dst_arr, a=1.0, b=1.0, x_end=x_end_value)
expected_result = np.zeros(size)
expected_result[1:x_end_value, 1] = 1
np.testing.assert_almost_equal(expected_result, dst_arr)
pystencils_tests/test_small_block_benchmark.ipynb deleted 100644 → 0
View file @ 1e7a1e6f
%% Cell type:code id: tags:
``` python
import pytest
pytest.importorskip('waLBerla')
```
%% Cell type:code id: tags:
``` python
from pystencils.session import *
from time import perf_counter
from statistics import median
from functools import partial
```
%% Cell type:markdown id: tags:
## Benchmark for Python call overhead
%% Cell type:code id: tags:
``` python
inner_repeats = 100
outer_repeats = 5
sizes = [2**i for i in range(1, 8)]
sizes
```
%% Output
$\displaystyle \left[ 2, \ 4, \ 8, \ 16, \ 32, \ 64, \ 128\right]$
[2, 4, 8, 16, 32, 64, 128]
%% Cell type:code id: tags:
``` python
def benchmark_pure(domain_size, extract_first=False):
src = np.zeros(domain_size)
dst = np.zeros_like(src)
f_src, f_dst = ps.fields("src, dst", src=src, dst=dst)
kernel = ps.create_kernel(ps.Assignment(f_dst.center, f_src.center)).compile()
if extract_first:
kernel = kernel.kernel
start = perf_counter()
for i in range(inner_repeats):
kernel(src=src, dst=dst)
src, dst = dst, src
end = perf_counter()
else:
start = perf_counter()
for i in range(inner_repeats):
kernel(src=src, dst=dst)
src, dst = dst, src
end = perf_counter()
return (end - start) / inner_repeats
def benchmark_datahandling(domain_size, parallel=False):
dh = ps.create_data_handling(domain_size, parallel=parallel)
f_src = dh.add_array('src')
f_dst = dh.add_array('dst')
kernel = ps.create_kernel(ps.Assignment(f_dst.center, f_src.center)).compile()
start = perf_counter()
for i in range(inner_repeats):
dh.run_kernel(kernel)
dh.swap('src', 'dst')
end = perf_counter()
return (end - start) / inner_repeats
name_to_func = {
'pure_extract': partial(benchmark_pure, extract_first=True),
'pure_no_extract': partial(benchmark_pure, extract_first=False),
'dh_serial': partial(benchmark_datahandling, parallel=False),
'dh_parallel': partial(benchmark_datahandling, parallel=True),
}
```
%% Cell type:code id: tags:
``` python
result = {'block_size': [],
'name': [],
'time': []}
for bs in sizes:
print("Computing size ", bs)
for name, func in name_to_func.items():
for i in range(outer_repeats):
time = func((bs, bs))
result['block_size'].append(bs)
result['name'].append(name)
result['time'].append(time)
```
%% Output
Computing size 2
Computing size 4
Computing size 8
Computing size 16
Computing size 32
Computing size 64
Computing size 128
%% Cell type:code id: tags:
``` python
if 'is_test_run' not in globals():
import pandas as pd
import seaborn as sns
data = pd.DataFrame.from_dict(result)
plt.subplot(1,2,1)
sns.barplot(x='block_size', y='time', hue='name', data=data, alpha=0.6)
plt.yscale('log')
plt.subplot(1,2,2)
data = pd.DataFrame.from_dict(result)
sns.barplot(x='block_size', y='time', hue='name', data=data, alpha=0.6)
```
%% Output
pystencils_tests/test_sum_prod.py deleted 100644 → 0
View file @ 1e7a1e6f
# -*- coding: utf-8 -*-
#
# Copyright © 2019 Stephan Seitz <stephan.seitz@fau.de>
#
# Distributed under terms of the GPLv3 license.
"""
"""
import numpy as np
import sympy
from sympy.abc import k
import pystencils
from pystencils.data_types import create_type
def test_sum():
sum = sympy.Sum(k, (k, 1, 100))
expanded_sum = sum.doit()
print(sum)
print(expanded_sum)
x = pystencils.fields('x: float32[1d]')
assignments = pystencils.AssignmentCollection({
x.center(): sum
})
ast = pystencils.create_kernel(assignments)
code = str(pystencils.get_code_obj(ast))
kernel = ast.compile()
print(code)
assert 'double sum' in code
array = np.zeros((10,), np.float32)
kernel(x=array)
assert np.allclose(array, int(expanded_sum) * np.ones_like(array))
def test_sum_use_float():
sum = sympy.Sum(k, (k, 1, 100))
expanded_sum = sum.doit()
print(sum)
print(expanded_sum)
x = pystencils.fields('x: float32[1d]')
assignments = pystencils.AssignmentCollection({
x.center(): sum
})
ast = pystencils.create_kernel(assignments, data_type=create_type('float32'))
code = str(pystencils.get_code_obj(ast))
kernel = ast.compile()
print(code)
print(pystencils.get_code_obj(ast))
assert 'float sum' in code
array = np.zeros((10,), np.float32)
kernel(x=array)
assert np.allclose(array, int(expanded_sum) * np.ones_like(array))
def test_product():
k = pystencils.TypedSymbol('k', create_type('int64'))
sum = sympy.Product(k, (k, 1, 10))
expanded_sum = sum.doit()
print(sum)
print(expanded_sum)
x = pystencils.fields('x: int64[1d]')
assignments = pystencils.AssignmentCollection({
x.center(): sum
})
ast = pystencils.create_kernel(assignments)
code = pystencils.get_code_str(ast)
kernel = ast.compile()
print(code)
assert 'int64_t product' in code
array = np.zeros((10,), np.int64)
kernel(x=array)
assert np.allclose(array, int(expanded_sum) * np.ones_like(array))
def test_prod_var_limit():
k = pystencils.TypedSymbol('k', create_type('int64'))
limit = pystencils.TypedSymbol('limit', create_type('int64'))
sum = sympy.Sum(k, (k, 1, limit))
expanded_sum = sum.replace(limit, 100).doit()
print(sum)
print(expanded_sum)
x = pystencils.fields('x: int64[1d]')
assignments = pystencils.AssignmentCollection({
x.center(): sum
})
ast = pystencils.create_kernel(assignments)
pystencils.show_code(ast)
kernel = ast.compile()
array = np.zeros((10,), np.int64)
kernel(x=array, limit=100)
assert np.allclose(array, int(expanded_sum) * np.ones_like(array))
pystencils_tests/test_sympy_optimizations.py deleted 100644 → 0
View file @ 1e7a1e6f
import pytest
import sympy as sp
import pystencils
from pystencils.math_optimizations import HAS_REWRITING, optimize_assignments, optims_pystencils_cpu, optimize_ast
@pytest.mark.skipif(not HAS_REWRITING, reason="need sympy.codegen.rewriting")
def test_sympy_optimizations():
for target in ('cpu', 'gpu'):
for op_ast in (True, False):
x, y, z = pystencils.fields('x, y, z: float32[2d]')
# Triggers Sympy's expm1 optimization
# Sympy's expm1 optimization is tedious to use and the behaviour is highly depended on the sympy version. In
# some cases the exp expression has to be encapsulated in brackets or multiplied with 1 or 1.0
# for sympy to work properly ...
assignments = pystencils.AssignmentCollection({
x[0, 0]: 1.0 * (sp.exp(y[0, 0]) - 1)
})
if not op_ast:
assignments = optimize_assignments(assignments, optims_pystencils_cpu)
print(assignments)
ast = pystencils.create_kernel(assignments, target=target)
if op_ast:
optimize_ast(ast, optims_pystencils_cpu)
code = pystencils.get_code_str(ast)
assert 'expm1(' in code
@pytest.mark.skipif(not HAS_REWRITING, reason="need sympy.codegen.rewriting")
def test_evaluate_constant_terms():
for target in ('cpu', 'gpu'):
x, y, z = pystencils.fields('x, y, z: float32[2d]')
# Triggers Sympy's cos optimization
assignments = pystencils.AssignmentCollection({
x[0, 0]: -sp.cos(1) + y[0, 0]
})
assignments = optimize_assignments(assignments, optims_pystencils_cpu)
ast = pystencils.create_kernel(assignments, target=target)
code = pystencils.get_code_str(ast)
assert 'cos(' not in code
print(code)
@pytest.mark.skipif(not HAS_REWRITING, reason="need sympy.codegen.rewriting")
def test_do_not_evaluate_constant_terms():
optimizations = pystencils.math_optimizations.optims_pystencils_cpu
optimizations.remove(pystencils.math_optimizations.evaluate_constant_terms)
for target in ('cpu', 'gpu'):
x, y, z = pystencils.fields('x, y, z: float32[2d]')
assignments = pystencils.AssignmentCollection({
x[0, 0]: -sp.cos(1) + y[0, 0]
})
ast = pystencils.create_kernel(assignments, target=target)
code = pystencils.get_code_str(ast)
assert 'cos(' in code
print(code)
pystencils_tests/test_transformations.py deleted 100644 → 0
View file @ 1e7a1e6f
import pystencils as ps
from pystencils import TypedSymbol
from pystencils.astnodes import LoopOverCoordinate, SympyAssignment
from pystencils.data_types import create_type
from pystencils.transformations import filtered_tree_iteration, get_loop_hierarchy, get_loop_counter_symbol_hierarchy
def test_loop_information():
f, g = ps.fields("f, g: double[2D]")
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]
loop_order = []
for i in get_loop_hierarchy(inner_loops[0].args[0]):
loop_order.append(i)
assert loop_order == [0, 1]
loop_symbols = get_loop_counter_symbol_hierarchy(inner_loops[0].args[0])
assert loop_symbols == [TypedSymbol("ctr_1", create_type("int"), nonnegative=True),
TypedSymbol("ctr_0", create_type("int"), nonnegative=True)]
pystencils_tests/test_type_interference.py deleted 100644 → 0
View file @ 1e7a1e6f
from sympy.abc import a, b, c, d, e, f
import pystencils
from pystencils.data_types import cast_func, create_type
def test_type_interference():
x = pystencils.fields('x: float32[3d]')
assignments = pystencils.AssignmentCollection({
a: cast_func(10, create_type('float64')),
b: cast_func(10, create_type('uint16')),
e: 11,
c: b,
f: c + b,
d: c + b + x.center + e,
x.center: c + b + x.center
})
ast = pystencils.create_kernel(assignments)
code = str(pystencils.get_code_str(ast))
assert 'double a' in code
assert 'uint16_t b' in code
assert 'uint16_t f' in code
assert 'int64_t e' in code
pystencils_tests/test_vectorization_specific.py deleted 100644 → 0
View file @ 1e7a1e6f
import pytest
import numpy as np
import sympy as sp
import pystencils as ps
from pystencils.backends.simd_instruction_sets import (get_cacheline_size, get_supported_instruction_sets,
get_vector_instruction_set)
from pystencils.data_types import cast_func, VectorType
supported_instruction_sets = get_supported_instruction_sets() if get_supported_instruction_sets() else []
@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
def test_vectorisation_varying_arch(instruction_set):
shape = (9, 9, 3)
arr = np.ones(shape, order='f')
@ps.kernel
def update_rule(s):
f = ps.fields("f(3) : [2D]", f=arr)
s.tmp0 @= f(0)
s.tmp1 @= f(1)
s.tmp2 @= f(2)
f0, f1, f2 = f(0), f(1), f(2)
f0 @= 2 * s.tmp0
f1 @= 2 * s.tmp0
f2 @= 2 * s.tmp0
ast = ps.create_kernel(update_rule, cpu_vectorize_info={'instruction_set': instruction_set})
kernel = ast.compile()
kernel(f=arr)
np.testing.assert_equal(arr, 2)
@pytest.mark.parametrize('dtype', ('float', 'double'))
@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
def test_vectorized_abs(instruction_set, dtype):
"""Some instructions sets have abs, some don't.
Furthermore, the special treatment of unary minus makes this data type-sensitive too.
"""
arr = np.ones((2 ** 2 + 2, 2 ** 3 + 2), dtype=np.float64 if dtype == 'double' else np.float32)
arr[-3:, :] = -1
f, g = ps.fields(f=arr, g=arr)
update_rule = [ps.Assignment(g.center(), sp.Abs(f.center()))]
ast = ps.create_kernel(update_rule, cpu_vectorize_info={'instruction_set': instruction_set})
func = ast.compile()
dst = np.zeros_like(arr)
func(g=dst, f=arr)
np.testing.assert_equal(np.sum(dst[1:-1, 1:-1]), 2 ** 2 * 2 ** 3)
@pytest.mark.parametrize('dtype', ('float', 'double'))
@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
def test_strided(instruction_set, dtype):
f, g = ps.fields(f"f, g : float{64 if dtype == 'double' else 32}[2D]")
update_rule = [ps.Assignment(g[0, 0], f[0, 0] + f[-1, 0] + f[1, 0] + f[0, 1] + f[0, -1] + 42.0)]
if 'storeS' not in get_vector_instruction_set(dtype, instruction_set) and not instruction_set in ['avx512', 'rvv'] and not instruction_set.startswith('sve'):
with pytest.warns(UserWarning) as warn:
ast = ps.create_kernel(update_rule, cpu_vectorize_info={'instruction_set': instruction_set})
assert 'Could not vectorize loop' in warn[0].message.args[0]
else:
with pytest.warns(None) as warn:
ast = ps.create_kernel(update_rule, cpu_vectorize_info={'instruction_set': instruction_set})
assert len(warn) == 0
func = ast.compile()
ref_func = ps.create_kernel(update_rule).compile()
arr = np.random.random((23 + 2, 17 + 2)).astype(np.float64 if dtype == 'double' else np.float32)
dst = np.zeros_like(arr, dtype=np.float64 if dtype == 'double' else np.float32)
ref = np.zeros_like(arr, dtype=np.float64 if dtype == 'double' else np.float32)
func(g=dst, f=arr)
ref_func(g=ref, f=arr)
np.testing.assert_almost_equal(dst, ref, 13 if dtype == 'double' else 5)
@pytest.mark.parametrize('dtype', ('float', 'double'))
@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
@pytest.mark.parametrize('gl_field, gl_kernel', [(1, 0), (0, 1), (1, 1)])
def test_alignment_and_correct_ghost_layers(gl_field, gl_kernel, instruction_set, dtype):
dtype = np.float64 if dtype == 'double' else np.float32
domain_size = (128, 128)
dh = ps.create_data_handling(domain_size, periodicity=(True, True), default_target='cpu')
src = dh.add_array("src", values_per_cell=1, dtype=dtype, ghost_layers=gl_field, alignment=True)
dh.fill(src.name, 1.0, ghost_layers=True)
dst = dh.add_array("dst", values_per_cell=1, dtype=dtype, ghost_layers=gl_field, alignment=True)
dh.fill(dst.name, 1.0, ghost_layers=True)
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}
ast = ps.create_kernel(update_rule, target=dh.default_target, cpu_vectorize_info=opt, ghost_layers=gl_kernel)
kernel = ast.compile()
if gl_kernel != gl_field:
with pytest.raises(ValueError):
dh.run_kernel(kernel)
else:
dh.run_kernel(kernel)
@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
def test_cacheline_size(instruction_set):
cacheline_size = get_cacheline_size(instruction_set)
if cacheline_size is None and instruction_set in ['sse', 'avx', 'avx512', 'rvv']:
pytest.skip()
instruction_set = get_vector_instruction_set('double', instruction_set)
vector_size = instruction_set['bytes']
assert cacheline_size > 8 and cacheline_size < 0x100000, "Cache line size is implausible"
if type(vector_size) is int:
assert cacheline_size % vector_size == 0, "Cache line size should be multiple of vector size"
assert cacheline_size & (cacheline_size - 1) == 0, "Cache line size is not a power of 2"
# test_vectorization is not parametrized because it is supposed to run without pytest, so we parametrize it here
from pystencils_tests import test_vectorization
@pytest.mark.parametrize('instruction_set', sorted(set(supported_instruction_sets) - set([test_vectorization.instruction_set])))
@pytest.mark.parametrize('function', [f for f in test_vectorization.__dict__ if f.startswith('test_') and f != 'test_hardware_query'])
def test_vectorization_other(instruction_set, function):
test_vectorization.__dict__[function](instruction_set)
pytest.ini
View file @ 5600b6b6
[pytest]
testpaths = src tests doc/notebooks
pythonpath = src
python_files = test_*.py *_test.py scenario_*.py
norecursedirs = *.egg-info .git .cache .ipynb_checkpoints htmlcov
addopts = --doctest-modules --durations=20 --cov-config pytest.ini
markers =
kerncraft: tests depending on kerncraft
longrun: tests only run at night since they have large execution time
notebook: mark for notebooks
# these warnings all come from third party libraries.
filterwarnings =
ignore:an integer is required:DeprecationWarning
ignore:\s*load will be removed, use:PendingDeprecationWarning
ignore:the imp module is deprecated in favour of importlib:DeprecationWarning
ignore:.*is a deprecated alias for the builtin `bool`:DeprecationWarning
ignore:'contextfilter' is renamed to 'pass_context':DeprecationWarning
ignore:Using or importing the ABCs from 'collections' instead of from 'collections.abc':DeprecationWarning
ignore:Animation was deleted without rendering anything:UserWarning
[run]
branch = True
source = pystencils
pystencils_tests
source = src/pystencils
tests
omit = doc/*
pystencils_tests/*
tests/*
setup.py
quicktest.py
conftest.py
versioneer.py
pystencils/jupytersetup.py
pystencils/cpu/msvc_detection.py
pystencils/sympy_gmpy_bug_workaround.py
pystencils/cache.py
pystencils/pacxx/benchmark.py
pystencils/_version.py
src/pystencils/jupytersetup.py
src/pystencils/cpu/msvc_detection.py
src/pystencils/sympy_gmpy_bug_workaround.py
src/pystencils/cache.py
src/pystencils/pacxx/benchmark.py
src/pystencils/_version.py
venv/
[report]
exclude_lines =
......@@ -43,7 +56,7 @@ exclude_lines =
if __name__ == .__main__.:
skip_covered = True
fail_under = 87
fail_under = 85
[html]
directory = coverage_report
quicktest.py 0 → 100644
View file @ 5600b6b6
#!/usr/bin/env python3
from contextlib import redirect_stdout
import io
from tests.test_quicktests import (
test_basic_kernel,
test_basic_blocking_staggered,
test_basic_vectorization,
)
quick_tests = [
test_basic_kernel,
test_basic_blocking_staggered,
test_basic_vectorization,
]
if __name__ == "__main__":
print("Running pystencils quicktests")
for qt in quick_tests:
print(f" -> {qt.__name__}")
with redirect_stdout(io.StringIO()):
qt()
setup.cfg deleted 100644 → 0
View file @ 1e7a1e6f
# See the docstring in versioneer.py for instructions. Note that you must
# re-run 'versioneer.py setup' after changing this section, and commit the
# resulting files.
[versioneer]
VCS = git
style = pep440
versionfile_source = pystencils/_version.py
versionfile_build = pystencils/_version.py
tag_prefix = release/
parentdir_prefix = pystencils-
setup.py
View file @ 5600b6b6
import distutils
import io
import os
import sys
from contextlib import redirect_stdout
from importlib import import_module
from setuptools import setup, __version__ as setuptools_version
import setuptools
if int(setuptools_version.split('.')[0]) < 61:
raise Exception(
"[ERROR] pystencils requires at least setuptools version 61 to install.\n"
"If this error occurs during an installation via pip, it is likely that there is a conflict between "
"versions of setuptools installed by pip and the system package manager. "
"In this case, it is recommended to install pystencils into a virtual environment instead."
)
import versioneer
try:
import cython # noqa
USE_CYTHON = True
except ImportError:
USE_CYTHON = False
quick_tests = [
'test_datahandling.test_kernel',
'test_blocking_staggered.test_blocking_staggered',
'test_blocking_staggered.test_blocking_staggered',
'test_vectorization.test_vectorization_variable_size',
]
class SimpleTestRunner(distutils.cmd.Command):
"""A custom command to run selected tests"""
description = 'run some quick tests'
user_options = []
@staticmethod
def _run_tests_in_module(test):
"""Short test runner function - to work also if py.test is not installed."""
test = 'pystencils_tests.' + test
mod, function_name = test.rsplit('.', 1)
if isinstance(mod, str):
mod = import_module(mod)
func = getattr(mod, function_name)
print(f" -> {function_name} in {mod.__name__}" )
with redirect_stdout(io.StringIO()):
func()
def initialize_options(self):
pass
def finalize_options(self):
pass
def run(self):
"""Run command."""
for test in quick_tests:
self._run_tests_in_module(test)
def readme():
with open('README.md') as f:
return f.read()
def cython_extensions(*extensions):
from distutils.extension import Extension
if USE_CYTHON:
ext = '.pyx'
result = [Extension(e, [os.path.join(*e.split(".")) + ext]) for e in extensions]
from Cython.Build import cythonize
result = cythonize(result, language_level=3)
return result
elif all([os.path.exists(os.path.join(*e.split(".")) + '.c') for e in extensions]):
ext = '.c'
result = [Extension(e, [os.path.join(*e.split(".")) + ext]) for e in extensions]
return result
else:
return None
def get_cmdclass():
cmdclass={"quicktest": SimpleTestRunner}
cmdclass.update(versioneer.get_cmdclass())
return cmdclass
return versioneer.get_cmdclass()
setuptools.setup(name='pystencils',
description='Speeding up stencil computations on CPUs and GPUs',
version=versioneer.get_version(),
long_description=readme(),
long_description_content_type="text/markdown",
author='Martin Bauer',
license='AGPLv3',
author_email='martin.bauer@fau.de',
url='https://i10git.cs.fau.de/pycodegen/pystencils/',
packages=['pystencils'] + ['pystencils.' + s for s in setuptools.find_packages('pystencils')],
install_requires=['sympy>=1.2', 'numpy>=1.8.0', 'appdirs', 'joblib'],
package_data={'pystencils': ['include/*.h',
'kerncraft_coupling/templates/*',
'backends/cuda_known_functions.txt',
'backends/opencl1.1_known_functions.txt',
'boundaries/createindexlistcython.c',
'boundaries/createindexlistcython.pyx']},
ext_modules=cython_extensions("pystencils.boundaries.createindexlistcython"),
classifiers=[
'Development Status :: 4 - Beta',
'Framework :: Jupyter',
'Topic :: Software Development :: Code Generators',
'Topic :: Scientific/Engineering :: Physics',
'Intended Audience :: Developers',
'Intended Audience :: Science/Research',
'License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)',
],
project_urls={
"Bug Tracker": "https://i10git.cs.fau.de/pycodegen/pystencils/issues",
"Documentation": "http://pycodegen.pages.walberla.net/pystencils/",
"Source Code": "https://i10git.cs.fau.de/pycodegen/pystencils",
},
extras_require={
'gpu': ['pycuda'],
'opencl': ['pyopencl'],
'alltrafos': ['islpy', 'py-cpuinfo'],
'bench_db': ['blitzdb', 'pymongo', 'pandas'],
'interactive': ['matplotlib', 'ipy_table', 'imageio', 'jupyter', 'pyevtk', 'rich', 'graphviz'],
'autodiff': ['pystencils-autodiff'],
'doc': ['sphinx', 'sphinx_rtd_theme', 'nbsphinx',
'sphinxcontrib-bibtex', 'sphinx_autodoc_typehints', 'pandoc'],
'use_cython': ['Cython']
},
tests_require=['pytest',
'pytest-cov',
'pytest-html',
'ansi2html',
'pytest-xdist',
'flake8',
'nbformat',
'nbconvert',
'ipython',
'randomgen>=1.18'],
python_requires=">=3.6",
cmdclass=get_cmdclass()
)
setup(
version=versioneer.get_version(),
cmdclass=get_cmdclass(),
)
pystencils/__init__.py → src/pystencils/__init__.py
View file @ 5600b6b6
"""Module to generate stencil kernels in C or CUDA using sympy expressions and call them as Python functions"""
from .enums import Backend, Target
from . import fd
from . import stencil as stencil
from .assignment import Assignment, assignment_from_stencil
from .data_types import TypedSymbol
from .datahandling import create_data_handling
from .display_utils import show_code, get_code_obj, get_code_str, to_dot
from .assignment import Assignment, AddAugmentedAssignment, assignment_from_stencil
from .typing.typed_sympy import TypedSymbol
from .display_utils import get_code_obj, get_code_str, show_code, to_dot
from .field import Field, FieldType, fields
from .kernel_decorator import kernel
from .kernelcreation import create_indexed_kernel, create_kernel, create_staggered_kernel
from .config import CreateKernelConfig
from .cache import clear_cache
from .kernel_decorator import kernel, kernel_config
from .kernelcreation import create_kernel, create_staggered_kernel
from .simp import AssignmentCollection
from .slicing import make_slice
from .spatial_coordinates import x_, x_staggered, x_staggered_vector, x_vector, y_, y_staggered, z_, z_staggered
from .sympyextensions import SymbolCreator
from .spatial_coordinates import (x_, x_staggered, x_staggered_vector, x_vector,
y_, y_staggered, z_, z_staggered)
try:
import pystencils_autodiff
autodiff = pystencils_autodiff
except ImportError:
pass
from .datahandling import create_data_handling
__all__ = ['Field', 'FieldType', 'fields',
'TypedSymbol',
'make_slice',
'create_kernel', 'create_indexed_kernel', 'create_staggered_kernel',
'CreateKernelConfig',
'create_kernel', 'create_staggered_kernel',
'Target', 'Backend',
'show_code', 'to_dot', 'get_code_obj', 'get_code_str',
'AssignmentCollection',
'Assignment',
'Assignment', 'AddAugmentedAssignment',
'assignment_from_stencil',
'SymbolCreator',
'create_data_handling',
'kernel',
'clear_cache',
'kernel', 'kernel_config',
'x_', 'y_', 'z_',
'x_staggered', 'y_staggered', 'z_staggered',
'x_vector', 'x_staggered_vector',
'fd',
'stencil']
from ._version import get_versions
__version__ = get_versions()['version']
del get_versions
from . import _version
__version__ = _version.get_versions()['version']
pystencils/_version.py → src/pystencils/_version.py
View file @ 5600b6b6
......@@ -5,8 +5,9 @@
# directories (produced by setup.py build) will contain a much shorter file
# that just contains the computed version number.
# This file is released into the public domain. Generated by
# versioneer-0.19 (https://github.com/python-versioneer/python-versioneer)
# This file is released into the public domain.
# Generated by versioneer-0.29
# https://github.com/python-versioneer/python-versioneer
"""Git implementation of _version.py."""
......@@ -15,9 +16,11 @@ import os
import re
import subprocess
import sys
from typing import Any, Callable, Dict, List, Optional, Tuple
import functools
def get_keywords():
def get_keywords() -> Dict[str, str]:
"""Get the keywords needed to look up the version information."""
# these strings will be replaced by git during git-archive.
# setup.py/versioneer.py will grep for the variable names, so they must
......@@ -33,8 +36,15 @@ def get_keywords():
class VersioneerConfig:
"""Container for Versioneer configuration parameters."""
VCS: str
style: str
tag_prefix: str
parentdir_prefix: str
versionfile_source: str
verbose: bool
def get_config():
def get_config() -> VersioneerConfig:
"""Create, populate and return the VersioneerConfig() object."""
# these strings are filled in when 'setup.py versioneer' creates
# _version.py
......@@ -43,7 +53,7 @@ def get_config():
cfg.style = "pep440"
cfg.tag_prefix = "release/"
cfg.parentdir_prefix = "pystencils-"
cfg.versionfile_source = "pystencils/_version.py"
cfg.versionfile_source = "src/pystencils/_version.py"
cfg.verbose = False
return cfg
......@@ -52,13 +62,13 @@ class NotThisMethod(Exception):
"""Exception raised if a method is not valid for the current scenario."""
LONG_VERSION_PY = {}
HANDLERS = {}
LONG_VERSION_PY: Dict[str, str] = {}
HANDLERS: Dict[str, Dict[str, Callable]] = {}
def register_vcs_handler(vcs, method): # decorator
def register_vcs_handler(vcs: str, method: str) -> Callable: # decorator
"""Create decorator to mark a method as the handler of a VCS."""
def decorate(f):
def decorate(f: Callable) -> Callable:
"""Store f in HANDLERS[vcs][method]."""
if vcs not in HANDLERS:
HANDLERS[vcs] = {}
......@@ -67,22 +77,35 @@ def register_vcs_handler(vcs, method): # decorator
return decorate
def run_command(commands, args, cwd=None, verbose=False, hide_stderr=False,
env=None):
def run_command(
commands: List[str],
args: List[str],
cwd: Optional[str] = None,
verbose: bool = False,
hide_stderr: bool = False,
env: Optional[Dict[str, str]] = None,
) -> Tuple[Optional[str], Optional[int]]:
"""Call the given command(s)."""
assert isinstance(commands, list)
p = None
for c in commands:
process = None
popen_kwargs: Dict[str, Any] = {}
if sys.platform == "win32":
# This hides the console window if pythonw.exe is used
startupinfo = subprocess.STARTUPINFO()
startupinfo.dwFlags |= subprocess.STARTF_USESHOWWINDOW
popen_kwargs["startupinfo"] = startupinfo
for command in commands:
try:
dispcmd = str([c] + args)
dispcmd = str([command] + args)
# remember shell=False, so use git.cmd on windows, not just git
p = subprocess.Popen([c] + args, cwd=cwd, env=env,
stdout=subprocess.PIPE,
stderr=(subprocess.PIPE if hide_stderr
else None))
process = subprocess.Popen([command] + args, cwd=cwd, env=env,
stdout=subprocess.PIPE,
stderr=(subprocess.PIPE if hide_stderr
else None), **popen_kwargs)
break
except EnvironmentError:
e = sys.exc_info()[1]
except OSError as e:
if e.errno == errno.ENOENT:
continue
if verbose:
......@@ -93,16 +116,20 @@ def run_command(commands, args, cwd=None, verbose=False, hide_stderr=False,
if verbose:
print("unable to find command, tried %s" % (commands,))
return None, None
stdout = p.communicate()[0].strip().decode()
if p.returncode != 0:
stdout = process.communicate()[0].strip().decode()
if process.returncode != 0:
if verbose:
print("unable to run %s (error)" % dispcmd)
print("stdout was %s" % stdout)
return None, p.returncode
return stdout, p.returncode
return None, process.returncode
return stdout, process.returncode
def versions_from_parentdir(parentdir_prefix, root, verbose):
def versions_from_parentdir(
parentdir_prefix: str,
root: str,
verbose: bool,
) -> Dict[str, Any]:
"""Try to determine the version from the parent directory name.
Source tarballs conventionally unpack into a directory that includes both
......@@ -111,15 +138,14 @@ def versions_from_parentdir(parentdir_prefix, root, verbose):
"""
rootdirs = []
for i in range(3):
for _ in range(3):
dirname = os.path.basename(root)
if dirname.startswith(parentdir_prefix):
return {"version": dirname[len(parentdir_prefix):],
"full-revisionid": None,
"dirty": False, "error": None, "date": None}
else:
rootdirs.append(root)
root = os.path.dirname(root) # up a level
rootdirs.append(root)
root = os.path.dirname(root) # up a level
if verbose:
print("Tried directories %s but none started with prefix %s" %
......@@ -128,39 +154,42 @@ def versions_from_parentdir(parentdir_prefix, root, verbose):
@register_vcs_handler("git", "get_keywords")
def git_get_keywords(versionfile_abs):
def git_get_keywords(versionfile_abs: str) -> Dict[str, str]:
"""Extract version information from the given file."""
# the code embedded in _version.py can just fetch the value of these
# keywords. When used from setup.py, we don't want to import _version.py,
# so we do it with a regexp instead. This function is not used from
# _version.py.
keywords = {}
keywords: Dict[str, str] = {}
try:
f = open(versionfile_abs, "r")
for line in f.readlines():
if line.strip().startswith("git_refnames ="):
mo = re.search(r'=\s*"(.*)"', line)
if mo:
keywords["refnames"] = mo.group(1)
if line.strip().startswith("git_full ="):
mo = re.search(r'=\s*"(.*)"', line)
if mo:
keywords["full"] = mo.group(1)
if line.strip().startswith("git_date ="):
mo = re.search(r'=\s*"(.*)"', line)
if mo:
keywords["date"] = mo.group(1)
f.close()
except EnvironmentError:
with open(versionfile_abs, "r") as fobj:
for line in fobj:
if line.strip().startswith("git_refnames ="):
mo = re.search(r'=\s*"(.*)"', line)
if mo:
keywords["refnames"] = mo.group(1)
if line.strip().startswith("git_full ="):
mo = re.search(r'=\s*"(.*)"', line)
if mo:
keywords["full"] = mo.group(1)
if line.strip().startswith("git_date ="):
mo = re.search(r'=\s*"(.*)"', line)
if mo:
keywords["date"] = mo.group(1)
except OSError:
pass
return keywords
@register_vcs_handler("git", "keywords")
def git_versions_from_keywords(keywords, tag_prefix, verbose):
def git_versions_from_keywords(
keywords: Dict[str, str],
tag_prefix: str,
verbose: bool,
) -> Dict[str, Any]:
"""Get version information from git keywords."""
if not keywords:
raise NotThisMethod("no keywords at all, weird")
if "refnames" not in keywords:
raise NotThisMethod("Short version file found")
date = keywords.get("date")
if date is not None:
# Use only the last line. Previous lines may contain GPG signature
......@@ -179,11 +208,11 @@ def git_versions_from_keywords(keywords, tag_prefix, verbose):
if verbose:
print("keywords are unexpanded, not using")
raise NotThisMethod("unexpanded keywords, not a git-archive tarball")
refs = set([r.strip() for r in refnames.strip("()").split(",")])
refs = {r.strip() for r in refnames.strip("()").split(",")}
# starting in git-1.8.3, tags are listed as "tag: foo-1.0" instead of
# just "foo-1.0". If we see a "tag: " prefix, prefer those.
TAG = "tag: "
tags = set([r[len(TAG):] for r in refs if r.startswith(TAG)])
tags = {r[len(TAG):] for r in refs if r.startswith(TAG)}
if not tags:
# Either we're using git < 1.8.3, or there really are no tags. We use
# a heuristic: assume all version tags have a digit. The old git %d
......@@ -192,7 +221,7 @@ def git_versions_from_keywords(keywords, tag_prefix, verbose):
# between branches and tags. By ignoring refnames without digits, we
# filter out many common branch names like "release" and
# "stabilization", as well as "HEAD" and "master".
tags = set([r for r in refs if re.search(r'\d', r)])
tags = {r for r in refs if re.search(r'\d', r)}
if verbose:
print("discarding '%s', no digits" % ",".join(refs - tags))
if verbose:
......@@ -201,6 +230,11 @@ def git_versions_from_keywords(keywords, tag_prefix, verbose):
# sorting will prefer e.g. "2.0" over "2.0rc1"
if ref.startswith(tag_prefix):
r = ref[len(tag_prefix):]
# Filter out refs that exactly match prefix or that don't start
# with a number once the prefix is stripped (mostly a concern
# when prefix is '')
if not re.match(r'\d', r):
continue
if verbose:
print("picking %s" % r)
return {"version": r,
......@@ -216,7 +250,12 @@ def git_versions_from_keywords(keywords, tag_prefix, verbose):
@register_vcs_handler("git", "pieces_from_vcs")
def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
def git_pieces_from_vcs(
tag_prefix: str,
root: str,
verbose: bool,
runner: Callable = run_command
) -> Dict[str, Any]:
"""Get version from 'git describe' in the root of the source tree.
This only gets called if the git-archive 'subst' keywords were *not*
......@@ -227,8 +266,15 @@ def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
if sys.platform == "win32":
GITS = ["git.cmd", "git.exe"]
out, rc = run_command(GITS, ["rev-parse", "--git-dir"], cwd=root,
hide_stderr=True)
# GIT_DIR can interfere with correct operation of Versioneer.
# It may be intended to be passed to the Versioneer-versioned project,
# but that should not change where we get our version from.
env = os.environ.copy()
env.pop("GIT_DIR", None)
runner = functools.partial(runner, env=env)
_, rc = runner(GITS, ["rev-parse", "--git-dir"], cwd=root,
hide_stderr=not verbose)
if rc != 0:
if verbose:
print("Directory %s not under git control" % root)
......@@ -236,24 +282,57 @@ def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
# if there is a tag matching tag_prefix, this yields TAG-NUM-gHEX[-dirty]
# if there isn't one, this yields HEX[-dirty] (no NUM)
describe_out, rc = run_command(GITS, ["describe", "--tags", "--dirty",
"--always", "--long",
"--match", "%s*" % tag_prefix],
cwd=root)
describe_out, rc = runner(GITS, [
"describe", "--tags", "--dirty", "--always", "--long",
"--match", f"{tag_prefix}[[:digit:]]*"
], cwd=root)
# --long was added in git-1.5.5
if describe_out is None:
raise NotThisMethod("'git describe' failed")
describe_out = describe_out.strip()
full_out, rc = run_command(GITS, ["rev-parse", "HEAD"], cwd=root)
full_out, rc = runner(GITS, ["rev-parse", "HEAD"], cwd=root)
if full_out is None:
raise NotThisMethod("'git rev-parse' failed")
full_out = full_out.strip()
pieces = {}
pieces: Dict[str, Any] = {}
pieces["long"] = full_out
pieces["short"] = full_out[:7] # maybe improved later
pieces["error"] = None
branch_name, rc = runner(GITS, ["rev-parse", "--abbrev-ref", "HEAD"],
cwd=root)
# --abbrev-ref was added in git-1.6.3
if rc != 0 or branch_name is None:
raise NotThisMethod("'git rev-parse --abbrev-ref' returned error")
branch_name = branch_name.strip()
if branch_name == "HEAD":
# If we aren't exactly on a branch, pick a branch which represents
# the current commit. If all else fails, we are on a branchless
# commit.
branches, rc = runner(GITS, ["branch", "--contains"], cwd=root)
# --contains was added in git-1.5.4
if rc != 0 or branches is None:
raise NotThisMethod("'git branch --contains' returned error")
branches = branches.split("\n")
# Remove the first line if we're running detached
if "(" in branches[0]:
branches.pop(0)
# Strip off the leading "* " from the list of branches.
branches = [branch[2:] for branch in branches]
if "master" in branches:
branch_name = "master"
elif not branches:
branch_name = None
else:
# Pick the first branch that is returned. Good or bad.
branch_name = branches[0]
pieces["branch"] = branch_name
# parse describe_out. It will be like TAG-NUM-gHEX[-dirty] or HEX[-dirty]
# TAG might have hyphens.
git_describe = describe_out
......@@ -270,7 +349,7 @@ def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
# TAG-NUM-gHEX
mo = re.search(r'^(.+)-(\d+)-g([0-9a-f]+)$', git_describe)
if not mo:
# unparseable. Maybe git-describe is misbehaving?
# unparsable. Maybe git-describe is misbehaving?
pieces["error"] = ("unable to parse git-describe output: '%s'"
% describe_out)
return pieces
......@@ -295,13 +374,11 @@ def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
else:
# HEX: no tags
pieces["closest-tag"] = None
count_out, rc = run_command(GITS, ["rev-list", "HEAD", "--count"],
cwd=root)
pieces["distance"] = int(count_out) # total number of commits
out, rc = runner(GITS, ["rev-list", "HEAD", "--left-right"], cwd=root)
pieces["distance"] = len(out.split()) # total number of commits
# commit date: see ISO-8601 comment in git_versions_from_keywords()
date = run_command(GITS, ["show", "-s", "--format=%ci", "HEAD"],
cwd=root)[0].strip()
date = runner(GITS, ["show", "-s", "--format=%ci", "HEAD"], cwd=root)[0].strip()
# Use only the last line. Previous lines may contain GPG signature
# information.
date = date.splitlines()[-1]
......@@ -310,14 +387,14 @@ def git_pieces_from_vcs(tag_prefix, root, verbose, run_command=run_command):
return pieces
def plus_or_dot(pieces):
def plus_or_dot(pieces: Dict[str, Any]) -> str:
"""Return a + if we don't already have one, else return a ."""
if "+" in pieces.get("closest-tag", ""):
return "."
return "+"
def render_pep440(pieces):
def render_pep440(pieces: Dict[str, Any]) -> str:
"""Build up version string, with post-release "local version identifier".
Our goal: TAG[+DISTANCE.gHEX[.dirty]] . Note that if you
......@@ -342,23 +419,71 @@ def render_pep440(pieces):
return rendered
def render_pep440_pre(pieces):
"""TAG[.post0.devDISTANCE] -- No -dirty.
def render_pep440_branch(pieces: Dict[str, Any]) -> str:
"""TAG[[.dev0]+DISTANCE.gHEX[.dirty]] .
The ".dev0" means not master branch. Note that .dev0 sorts backwards
(a feature branch will appear "older" than the master branch).
Exceptions:
1: no tags. 0.post0.devDISTANCE
1: no tags. 0[.dev0]+untagged.DISTANCE.gHEX[.dirty]
"""
if pieces["closest-tag"]:
rendered = pieces["closest-tag"]
if pieces["distance"] or pieces["dirty"]:
if pieces["branch"] != "master":
rendered += ".dev0"
rendered += plus_or_dot(pieces)
rendered += "%d.g%s" % (pieces["distance"], pieces["short"])
if pieces["dirty"]:
rendered += ".dirty"
else:
# exception #1
rendered = "0"
if pieces["branch"] != "master":
rendered += ".dev0"
rendered += "+untagged.%d.g%s" % (pieces["distance"],
pieces["short"])
if pieces["dirty"]:
rendered += ".dirty"
return rendered
def pep440_split_post(ver: str) -> Tuple[str, Optional[int]]:
"""Split pep440 version string at the post-release segment.
Returns the release segments before the post-release and the
post-release version number (or -1 if no post-release segment is present).
"""
vc = str.split(ver, ".post")
return vc[0], int(vc[1] or 0) if len(vc) == 2 else None
def render_pep440_pre(pieces: Dict[str, Any]) -> str:
"""TAG[.postN.devDISTANCE] -- No -dirty.
Exceptions:
1: no tags. 0.post0.devDISTANCE
"""
if pieces["closest-tag"]:
if pieces["distance"]:
rendered += ".post0.dev%d" % pieces["distance"]
# update the post release segment
tag_version, post_version = pep440_split_post(pieces["closest-tag"])
rendered = tag_version
if post_version is not None:
rendered += ".post%d.dev%d" % (post_version + 1, pieces["distance"])
else:
rendered += ".post0.dev%d" % (pieces["distance"])
else:
# no commits, use the tag as the version
rendered = pieces["closest-tag"]
else:
# exception #1
rendered = "0.post0.dev%d" % pieces["distance"]
return rendered
def render_pep440_post(pieces):
def render_pep440_post(pieces: Dict[str, Any]) -> str:
"""TAG[.postDISTANCE[.dev0]+gHEX] .
The ".dev0" means dirty. Note that .dev0 sorts backwards
......@@ -385,7 +510,36 @@ def render_pep440_post(pieces):
return rendered
def render_pep440_old(pieces):
def render_pep440_post_branch(pieces: Dict[str, Any]) -> str:
"""TAG[.postDISTANCE[.dev0]+gHEX[.dirty]] .
The ".dev0" means not master branch.
Exceptions:
1: no tags. 0.postDISTANCE[.dev0]+gHEX[.dirty]
"""
if pieces["closest-tag"]:
rendered = pieces["closest-tag"]
if pieces["distance"] or pieces["dirty"]:
rendered += ".post%d" % pieces["distance"]
if pieces["branch"] != "master":
rendered += ".dev0"
rendered += plus_or_dot(pieces)
rendered += "g%s" % pieces["short"]
if pieces["dirty"]:
rendered += ".dirty"
else:
# exception #1
rendered = "0.post%d" % pieces["distance"]
if pieces["branch"] != "master":
rendered += ".dev0"
rendered += "+g%s" % pieces["short"]
if pieces["dirty"]:
rendered += ".dirty"
return rendered
def render_pep440_old(pieces: Dict[str, Any]) -> str:
"""TAG[.postDISTANCE[.dev0]] .
The ".dev0" means dirty.
......@@ -407,7 +561,7 @@ def render_pep440_old(pieces):
return rendered
def render_git_describe(pieces):
def render_git_describe(pieces: Dict[str, Any]) -> str:
"""TAG[-DISTANCE-gHEX][-dirty].
Like 'git describe --tags --dirty --always'.
......@@ -427,7 +581,7 @@ def render_git_describe(pieces):
return rendered
def render_git_describe_long(pieces):
def render_git_describe_long(pieces: Dict[str, Any]) -> str:
"""TAG-DISTANCE-gHEX[-dirty].
Like 'git describe --tags --dirty --always -long'.
......@@ -447,7 +601,7 @@ def render_git_describe_long(pieces):
return rendered
def render(pieces, style):
def render(pieces: Dict[str, Any], style: str) -> Dict[str, Any]:
"""Render the given version pieces into the requested style."""
if pieces["error"]:
return {"version": "unknown",
......@@ -461,10 +615,14 @@ def render(pieces, style):
if style == "pep440":
rendered = render_pep440(pieces)
elif style == "pep440-branch":
rendered = render_pep440_branch(pieces)
elif style == "pep440-pre":
rendered = render_pep440_pre(pieces)
elif style == "pep440-post":
rendered = render_pep440_post(pieces)
elif style == "pep440-post-branch":
rendered = render_pep440_post_branch(pieces)
elif style == "pep440-old":
rendered = render_pep440_old(pieces)
elif style == "git-describe":
......@@ -479,7 +637,7 @@ def render(pieces, style):
"date": pieces.get("date")}
def get_versions():
def get_versions() -> Dict[str, Any]:
"""Get version information or return default if unable to do so."""
# I am in _version.py, which lives at ROOT/VERSIONFILE_SOURCE. If we have
# __file__, we can work backwards from there to the root. Some
......@@ -500,7 +658,7 @@ def get_versions():
# versionfile_source is the relative path from the top of the source
# tree (where the .git directory might live) to this file. Invert
# this to find the root from __file__.
for i in cfg.versionfile_source.split('/'):
for _ in cfg.versionfile_source.split('/'):
root = os.path.dirname(root)
except NameError:
return {"version": "0+unknown", "full-revisionid": None,
......
pystencils/alignedarray.py → src/pystencils/alignedarray.py
View file @ 5600b6b6
import numpy as np
from pystencils.data_types import BasicType
def aligned_empty(shape, byte_alignment=True, dtype=np.float64, byte_offset=0, order='C', align_inner_coordinate=True):
......@@ -21,26 +20,26 @@ def aligned_empty(shape, byte_alignment=True, dtype=np.float64, byte_offset=0, o
from pystencils.backends.simd_instruction_sets import (get_supported_instruction_sets, get_cacheline_size,
get_vector_instruction_set)
type_name = BasicType.numpy_name_to_c(np.dtype(dtype).name)
instruction_sets = get_supported_instruction_sets()
if instruction_sets is None:
byte_alignment = 64
elif byte_alignment == 'cacheline':
cacheline_sizes = [get_cacheline_size(is_name) for is_name in instruction_sets]
if all([s is None for s in cacheline_sizes]):
widths = [get_vector_instruction_set(type_name, is_name)['width'] * np.dtype(dtype).itemsize
if all([s is None for s in cacheline_sizes]) or \
max([s for s in cacheline_sizes if s is not None]) > 0x100000:
widths = [get_vector_instruction_set(dtype, is_name)['width'] * np.dtype(dtype).itemsize
for is_name in instruction_sets
if type(get_vector_instruction_set(type_name, is_name)['width']) is int]
if type(get_vector_instruction_set(dtype, is_name)['width']) is int]
byte_alignment = 64 if all([s is None for s in widths]) else max(widths)
else:
byte_alignment = max([s for s in cacheline_sizes if s is not None])
elif not any([type(get_vector_instruction_set(type_name, is_name)['width']) is int
elif not any([type(get_vector_instruction_set(dtype, is_name)['width']) is int
for is_name in instruction_sets]):
byte_alignment = 64
else:
byte_alignment = max([get_vector_instruction_set(type_name, is_name)['width'] * np.dtype(dtype).itemsize
byte_alignment = max([get_vector_instruction_set(dtype, is_name)['width'] * np.dtype(dtype).itemsize
for is_name in instruction_sets
if type(get_vector_instruction_set(type_name, is_name)['width']) is int])
if type(get_vector_instruction_set(dtype, is_name)['width']) is int])
if (not align_inner_coordinate) or (not hasattr(shape, '__len__')):
size = np.prod(shape)
d = np.dtype(dtype)
......@@ -78,7 +77,7 @@ def aligned_empty(shape, byte_alignment=True, dtype=np.float64, byte_offset=0, o
return tmp
def aligned_zeros(shape, byte_alignment=True, dtype=float, byte_offset=0, order='C', align_inner_coordinate=True):
def aligned_zeros(shape, byte_alignment=True, dtype=np.float64, byte_offset=0, order='C', align_inner_coordinate=True):
arr = aligned_empty(shape, dtype=dtype, byte_offset=byte_offset,
order=order, byte_alignment=byte_alignment, align_inner_coordinate=align_inner_coordinate)
x = np.zeros((), arr.dtype)
......@@ -86,7 +85,7 @@ def aligned_zeros(shape, byte_alignment=True, dtype=float, byte_offset=0, order=
return arr
def aligned_ones(shape, byte_alignment=True, dtype=float, byte_offset=0, order='C', align_inner_coordinate=True):
def aligned_ones(shape, byte_alignment=True, dtype=np.float64, byte_offset=0, order='C', align_inner_coordinate=True):
arr = aligned_empty(shape, dtype=dtype, byte_offset=byte_offset,
order=order, byte_alignment=byte_alignment, align_inner_coordinate=align_inner_coordinate)
x = np.ones((), arr.dtype)
......
pystencils/assignment.py → src/pystencils/assignment.py
View file @ 5600b6b6
import numpy as np
import sympy as sp
from sympy.codegen.ast import Assignment
from sympy.codegen.ast import Assignment, AugmentedAssignment, AddAugmentedAssignment
from sympy.printing.latex import LatexPrinter
__all__ = ['Assignment', 'assignment_from_stencil']
__all__ = ['Assignment', 'AugmentedAssignment', 'AddAugmentedAssignment', 'assignment_from_stencil']
def print_assignment_latex(printer, expr):
binop = f"{expr.binop}=" if isinstance(expr, AugmentedAssignment) else ''
"""sympy cannot print Assignments as Latex. Thus, this function is added to the sympy Latex printer"""
printed_lhs = printer.doprint(expr.lhs)
printed_rhs = printer.doprint(expr.rhs)
return r"{printed_lhs} \leftarrow {printed_rhs}".format(printed_lhs=printed_lhs, printed_rhs=printed_rhs)
return fr"{printed_lhs} \leftarrow_{{{binop}}} {printed_rhs}"
def assignment_str(assignment):
return r"{lhs} ← {rhs}".format(lhs=assignment.lhs, rhs=assignment.rhs)
op = f"{assignment.binop}=" if isinstance(assignment, AugmentedAssignment) else '←'
return fr"{assignment.lhs} {op} {assignment.rhs}"
_old_new = sp.codegen.ast.Assignment.__new__
# TODO Typing Part2 add default type, defult_float_type, default_int_type and use sane defaults
def _Assignment__new__(cls, lhs, rhs, *args, **kwargs):
if isinstance(lhs, (list, tuple, sp.Matrix)) and isinstance(rhs, (list, tuple, sp.Matrix)):
assert len(lhs) == len(rhs), f'{lhs} and {rhs} must have same length when performing vector assignment!'
......@@ -31,20 +34,10 @@ Assignment.__str__ = assignment_str
Assignment.__new__ = _Assignment__new__
LatexPrinter._print_Assignment = print_assignment_latex
sp.MutableDenseMatrix.__hash__ = lambda self: hash(tuple(self))
# Apparently, in SymPy 1.4 Assignment.__hash__ is not implemented. This has been fixed in current master
try:
sympy_version = sp.__version__.split('.')
AugmentedAssignment.__str__ = assignment_str
LatexPrinter._print_AugmentedAssignment = print_assignment_latex
if int(sympy_version[0]) <= 1 and int(sympy_version[1]) <= 4:
def hash_fun(self):
return hash((self.lhs, self.rhs))
Assignment.__hash__ = hash_fun
except Exception:
pass
sp.MutableDenseMatrix.__hash__ = lambda self: hash(tuple(self))
def assignment_from_stencil(stencil_array, input_field, output_field,
......
pystencils/astnodes.py → src/pystencils/astnodes.py
View file @ 5600b6b6
......@@ -5,11 +5,12 @@ from typing import Any, List, Optional, Sequence, Set, Union
import sympy as sp
import pystencils
from pystencils.data_types import TypedImaginaryUnit, TypedSymbol, cast_func, create_type
from pystencils.assignment import Assignment
from pystencils.enums import Target, Backend
from pystencils.field import Field
from pystencils.kernelparameters import FieldPointerSymbol, FieldShapeSymbol, FieldStrideSymbol
from pystencils.sympyextensions import fast_subs
from pystencils.typing import (create_type, get_next_parent_of_type,
FieldPointerSymbol, FieldShapeSymbol, FieldStrideSymbol, TypedSymbol, CFunction)
NodeOrExpr = Union['Node', sp.Expr]
......@@ -136,7 +137,6 @@ class Conditional(Node):
class KernelFunction(Node):
class Parameter:
"""Function parameter.
......@@ -176,7 +176,9 @@ class KernelFunction(Node):
def field_name(self):
return self.fields[0].name
def __init__(self, body, target, backend, compile_function, ghost_layers, function_name="kernel", assignments=None):
def __init__(self, body, target: Target, backend: Backend, compile_function, ghost_layers,
function_name: str = "kernel",
assignments=None):
super(KernelFunction, self).__init__()
self._body = body
body.parent = self
......@@ -191,15 +193,19 @@ class KernelFunction(Node):
# function that compiles the node to a Python callable, is set by the backends
self._compile_function = compile_function
self.assignments = assignments
# If nontemporal stores are activated together with the Neon instruction set it results in cacheline zeroing
# For cacheline zeroing the information of the field size for each field is needed. Thus, in this case
# all field sizes are kernel parameters and not just the common field size used for the loops
self.use_all_written_field_sizes = False
@property
def target(self):
"""Currently either 'cpu' or 'gpu' """
"""See pystencils.Target"""
return self._target
@property
def backend(self):
"""Backend for generating the code e.g. 'llvm', 'c', 'cuda' """
"""Backend for generating the code: `Backend`"""
return self._backend
@property
......@@ -226,13 +232,13 @@ class KernelFunction(Node):
@property
def fields_accessed(self) -> Set[Field]:
"""Set of Field instances: fields which are accessed inside this kernel function"""
from pystencils.interpolation_astnodes import InterpolatorAccess
return set(o.field for o in itertools.chain(self.atoms(ResolvedFieldAccess), self.atoms(InterpolatorAccess)))
return set(o.field for o in itertools.chain(self.atoms(ResolvedFieldAccess)))
@property
def fields_written(self) -> Set[Field]:
assignments = self.atoms(SympyAssignment)
return {a.lhs.field for a in assignments if isinstance(a.lhs, ResolvedFieldAccess)}
return set().union(itertools.chain.from_iterable([f.field for f in a.lhs.free_symbols if hasattr(f, 'field')]
for a in assignments))
@property
def fields_read(self) -> Set[Field]:
......@@ -246,6 +252,11 @@ class KernelFunction(Node):
This function is expensive, cache the result where possible!
"""
field_map = {f.name: f for f in self.fields_accessed}
sizes = set()
if self.use_all_written_field_sizes:
sizes = set().union(*(a.shape[:a.spatial_dimensions] for a in self.fields_written))
sizes = filter(lambda s: isinstance(s, FieldShapeSymbol), sizes)
def get_fields(symbol):
if hasattr(symbol, 'field_name'):
......@@ -255,9 +266,13 @@ class KernelFunction(Node):
return ()
argument_symbols = self._body.undefined_symbols - self.global_variables
argument_symbols.update(sizes)
parameters = [self.Parameter(symbol, get_fields(symbol)) for symbol in argument_symbols]
if hasattr(self, 'indexing'):
parameters += [self.Parameter(s, []) for s in self.indexing.symbolic_parameters()]
# Exclude paramters of type CFunction. These parameters will result in a C function call that will be handled
# by including a respective header file in the compute kernel. Hence, it is not a free parameter.
parameters = [p for p in parameters if not isinstance(p.symbol, CFunction)]
parameters.sort(key=lambda p: p.symbol.name)
return parameters
......@@ -291,8 +306,10 @@ class SkipIteration(Node):
class Block(Node):
def __init__(self, nodes: List[Node]):
def __init__(self, nodes: Union[Node, List[Node]]):
super(Block, self).__init__()
if not isinstance(nodes, list):
nodes = [nodes]
self._nodes = nodes
self.parent = None
for n in self._nodes:
......@@ -331,14 +348,6 @@ class Block(Node):
assert self._nodes.count(insert_before) == 1
idx = self._nodes.index(insert_before)
# move all assignment (definitions to the top)
if isinstance(new_node, SympyAssignment) and new_node.is_declaration:
while idx > 0:
pn = self._nodes[idx - 1]
if isinstance(pn, LoopOverCoordinate) or isinstance(pn, Conditional):
idx -= 1
else:
break
if not if_not_exists or self._nodes[idx] != new_node:
self._nodes.insert(idx, new_node)
......@@ -347,14 +356,6 @@ class Block(Node):
assert self._nodes.count(insert_after) == 1
idx = self._nodes.index(insert_after) + 1
# move all assignment (definitions to the top)
if isinstance(new_node, SympyAssignment) and new_node.is_declaration:
while idx > 0:
pn = self._nodes[idx - 1]
if isinstance(pn, LoopOverCoordinate) or isinstance(pn, Conditional):
idx -= 1
else:
break
if not if_not_exists or not (self._nodes[idx - 1] == new_node
or (idx < len(self._nodes) and self._nodes[idx] == new_node)):
self._nodes.insert(idx, new_node)
......@@ -389,7 +390,7 @@ class Block(Node):
def symbols_defined(self):
result = set()
for a in self.args:
if isinstance(a, pystencils.Assignment):
if isinstance(a, Assignment):
result.update(a.free_symbols)
else:
result.update(a.symbols_defined)
......@@ -400,7 +401,7 @@ class Block(Node):
result = set()
defined_symbols = set()
for a in self.args:
if isinstance(a, pystencils.Assignment):
if isinstance(a, Assignment):
result.update(a.free_symbols)
defined_symbols.update({a.lhs})
else:
......@@ -430,7 +431,7 @@ class LoopOverCoordinate(Node):
LOOP_COUNTER_NAME_PREFIX = "ctr"
BLOCK_LOOP_COUNTER_NAME_PREFIX = "_blockctr"
def __init__(self, body, coordinate_to_loop_over, start, stop, step=1, is_block_loop=False):
def __init__(self, body, coordinate_to_loop_over, start, stop, step=1, is_block_loop=False, custom_loop_ctr=None):
super(LoopOverCoordinate, self).__init__(parent=None)
self.body = body
body.parent = self
......@@ -441,11 +442,12 @@ class LoopOverCoordinate(Node):
self.body.parent = self
self.prefix_lines = []
self.is_block_loop = is_block_loop
self.custom_loop_ctr = custom_loop_ctr
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]
self.step, self.is_block_loop, self.custom_loop_ctr)
result.prefix_lines = [prefix_line for prefix_line in self.prefix_lines]
return result
def subs(self, subs_dict):
......@@ -507,10 +509,13 @@ class LoopOverCoordinate(Node):
@property
def loop_counter_name(self):
if self.is_block_loop:
return LoopOverCoordinate.get_block_loop_counter_name(self.coordinate_to_loop_over)
if self.custom_loop_ctr:
return self.custom_loop_ctr.name
else:
return LoopOverCoordinate.get_loop_counter_name(self.coordinate_to_loop_over)
if self.is_block_loop:
return LoopOverCoordinate.get_block_loop_counter_name(self.coordinate_to_loop_over)
else:
return LoopOverCoordinate.get_loop_counter_name(self.coordinate_to_loop_over)
@staticmethod
def is_loop_counter_symbol(symbol):
......@@ -534,14 +539,16 @@ class LoopOverCoordinate(Node):
@property
def loop_counter_symbol(self):
if self.is_block_loop:
return self.get_block_loop_counter_symbol(self.coordinate_to_loop_over)
if self.custom_loop_ctr:
return self.custom_loop_ctr
else:
return self.get_loop_counter_symbol(self.coordinate_to_loop_over)
if self.is_block_loop:
return self.get_block_loop_counter_symbol(self.coordinate_to_loop_over)
else:
return self.get_loop_counter_symbol(self.coordinate_to_loop_over)
@property
def is_outermost_loop(self):
from pystencils.transformations import get_next_parent_of_type
return get_next_parent_of_type(self, LoopOverCoordinate) is None
@property
......@@ -564,13 +571,14 @@ class SympyAssignment(Node):
def __init__(self, lhs_symbol, rhs_expr, is_const=True, use_auto=False):
super(SympyAssignment, self).__init__(parent=None)
self._lhs_symbol = sp.sympify(lhs_symbol)
self.rhs = sp.sympify(rhs_expr)
self._rhs = sp.sympify(rhs_expr)
self._is_const = is_const
self._is_declaration = self.__is_declaration()
self.use_auto = use_auto
self._use_auto = use_auto
def __is_declaration(self):
if isinstance(self._lhs_symbol, cast_func):
from pystencils.typing import CastFunc
if isinstance(self._lhs_symbol, CastFunc):
return False
if any(isinstance(self._lhs_symbol, c) for c in (Field.Access, sp.Indexed, TemporaryMemoryAllocation)):
return False
......@@ -580,15 +588,28 @@ class SympyAssignment(Node):
def lhs(self):
return self._lhs_symbol
@property
def rhs(self):
return self._rhs
@lhs.setter
def lhs(self, new_value):
self._lhs_symbol = new_value
self._is_declaration = self.__is_declaration()
@rhs.setter
def rhs(self, new_rhs_expr):
self._rhs = new_rhs_expr
def subs(self, subs_dict):
self.lhs = fast_subs(self.lhs, subs_dict)
self.rhs = fast_subs(self.rhs, subs_dict)
def fast_subs(self, subs_dict, skip=None):
self.lhs = fast_subs(self.lhs, subs_dict, skip)
self.rhs = fast_subs(self.rhs, subs_dict, skip)
return self
def optimize(self, optimizations):
try:
from sympy.codegen.rewriting import optimize
......@@ -598,7 +619,7 @@ class SympyAssignment(Node):
@property
def args(self):
return [self._lhs_symbol, self.rhs, sp.sympify(self._is_const)]
return [self._lhs_symbol, self.rhs]
@property
def symbols_defined(self):
......@@ -615,9 +636,10 @@ class SympyAssignment(Node):
if isinstance(symbol, Field.Access):
for i in range(len(symbol.offsets)):
loop_counters.add(LoopOverCoordinate.get_loop_counter_symbol(i))
result = {r for r in result if not isinstance(r, TypedImaginaryUnit)}
result.update(loop_counters)
result.update(self._lhs_symbol.atoms(sp.Symbol))
return result
@property
......@@ -628,6 +650,10 @@ class SympyAssignment(Node):
def is_const(self):
return self._is_const
@property
def use_auto(self):
return self._use_auto
def replace(self, child, replacement):
if child == self.lhs:
replacement.parent = self
......@@ -650,7 +676,7 @@ class SympyAssignment(Node):
return hash((self.lhs, self.rhs))
def __eq__(self, other):
return type(self) == type(other) and (self.lhs, self.rhs) == (other.lhs, other.rhs)
return type(self) is type(other) and (self.lhs, self.rhs) == (other.lhs, other.rhs)
class ResolvedFieldAccess(sp.Indexed):
......
pystencils/backends/__init__.py → src/pystencils/backends/__init__.py
View file @ 5600b6b6
......@@ -6,9 +6,3 @@ try:
__all__.append('print_dot')
except ImportError:
pass
try:
from .llvm import generate_llvm # NOQA
__all__.append('generate_llvm')
except ImportError:
pass
pystencils/backends/arm_instruction_sets.py → src/pystencils/backends/arm_instruction_sets.py
View file @ 5600b6b6
from pystencils.typing import CFunction
def get_argument_string(function_shortcut, first=''):
args = function_shortcut[function_shortcut.index('[') + 1: -1]
arg_string = "("
......@@ -16,10 +19,13 @@ def get_argument_string(function_shortcut, first=''):
def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
if instruction_set != 'neon' and not instruction_set.startswith('sve'):
if instruction_set not in ['neon', 'sme'] and not instruction_set.startswith('sve'):
raise NotImplementedError(instruction_set)
if instruction_set == 'sve':
if instruction_set in ['sve', 'sve2', 'sme']:
cmp = 'cmp'
elif instruction_set.startswith('sve2') and instruction_set not in ('sve256', 'sve2048'):
cmp = 'cmp'
bitwidth = int(instruction_set[4:])
elif instruction_set.startswith('sve'):
cmp = 'cmp'
bitwidth = int(instruction_set[3:])
......@@ -35,9 +41,7 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
'sqrt': 'sqrt[0]',
'loadU': 'ld1[0]',
'loadA': 'ld1[0]',
'storeU': 'st1[0, 1]',
'storeA': 'st1[0, 1]',
'abs': 'abs[0]',
'==': f'{cmp}eq[0, 1]',
......@@ -54,7 +58,7 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
result = dict()
if instruction_set == 'sve':
if instruction_set in ['sve', 'sve2', 'sme']:
width = 'svcntd()' if data_type == 'double' else 'svcntw()'
intwidth = 'svcntw()'
result['bytes'] = 'svcntb()'
......@@ -62,14 +66,15 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
width = bitwidth // bits[data_type]
intwidth = bitwidth // bits['int']
result['bytes'] = bitwidth // 8
if instruction_set.startswith('sve'):
if instruction_set.startswith('sve') or instruction_set == 'sme':
base_names['stream'] = 'stnt1[0, 1]'
prefix = 'sv'
suffix = f'_f{bits[data_type]}'
suffix = f'_f{bits[data_type]}'
elif instruction_set == 'neon':
prefix = 'v'
suffix = f'q_f{bits[data_type]}'
suffix = f'q_f{bits[data_type]}'
if instruction_set == 'sve':
if instruction_set in ['sve', 'sve2', 'sme']:
predicate = f'{prefix}whilelt_b{bits[data_type]}_u64({{loop_counter}}, {{loop_stop}})'
int_predicate = f'{prefix}whilelt_b{bits["int"]}_u64({{loop_counter}}, {{loop_stop}})'
else:
......@@ -88,33 +93,36 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
result[intrinsic_id] = prefix + name + suffix + undef + arg_string
if instruction_set == 'sve':
from pystencils.backends.cbackend import CFunction
if instruction_set in ['sve', 'sve2', 'sme']:
result['width'] = CFunction(width, "int")
result['intwidth'] = CFunction(intwidth, "int")
else:
result['width'] = width
result['intwidth'] = intwidth
if instruction_set.startswith('sve'):
if instruction_set.startswith('sve') or instruction_set == 'sme':
result['makeVecConst'] = f'svdup_f{bits[data_type]}' + '({0})'
result['makeVecConstInt'] = f'svdup_s{bits["int"]}' + '({0})'
result['makeVecIndex'] = f'svindex_s{bits["int"]}' + '({0}, {1})'
vindex = f'svindex_u{bits[data_type]}(0, {{0}})'
result['storeS'] = f'svst1_scatter_u{bits[data_type]}index_f{bits[data_type]}({predicate}, {{0}}, ' + \
vindex.format("{2}") + ', {1})'
result['loadS'] = f'svld1_gather_u{bits[data_type]}index_f{bits[data_type]}({predicate}, {{0}}, ' + \
vindex.format("{1}") + ')'
if instruction_set != 'sme':
vindex = f'svindex_u{bits[data_type]}(0, {{0}})'
result['storeS'] = f'svst1_scatter_u{bits[data_type]}index_f{bits[data_type]}({predicate}, {{0}}, ' + \
vindex.format("{2}") + ', {1})'
result['loadS'] = f'svld1_gather_u{bits[data_type]}index_f{bits[data_type]}({predicate}, {{0}}, ' + \
vindex.format("{1}") + ')'
if instruction_set.startswith('sve2') and instruction_set not in ('sve256', 'sve2048'):
result['streamS'] = f'svstnt1_scatter_u{bits[data_type]}offset_f{bits[data_type]}({predicate}, {{0}}, ' + \
vindex.format(f"{{2}}*{bits[data_type]//8}") + ', {1})'
result['+int'] = f"svadd_s{bits['int']}_x({int_predicate}, " + "{0}, {1})"
result['float'] = f'svfloat{bits["float"]}_{"s" if instruction_set != "sve" else ""}t'
result['double'] = f'svfloat{bits["double"]}_{"s" if instruction_set != "sve" else ""}t'
result['int'] = f'svint{bits["int"]}_{"s" if instruction_set != "sve" else ""}t'
result['bool'] = f'svbool_{"s" if instruction_set != "sve" else ""}t'
result['float'] = f'svfloat{bits["float"]}_{"s" if instruction_set not in ["sve", "sve2", "sme"] else ""}t'
result['double'] = f'svfloat{bits["double"]}_{"s" if instruction_set not in ["sve", "sve2", "sme"] else ""}t'
result['int'] = f'svint{bits["int"]}_{"s" if instruction_set not in ["sve", "sve2", "sme"] else ""}t'
result['bool'] = f'svbool_{"s" if instruction_set not in ["sve", "sve2", "sme"] else ""}t'
result['headers'] = ['<arm_sve.h>', '"arm_neon_helpers.h"']
result['headers'] = ['<arm_sve.h>', '<arm_acle.h>', '"arm_neon_helpers.h"']
result['&'] = f'svand_b_z({predicate},' + ' {0}, {1})'
result['|'] = f'svorr_b_z({predicate},' + ' {0}, {1})'
......@@ -123,10 +131,17 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
result['all'] = f'svcntp_b{bits[data_type]}({predicate}, {{0}}) == {width}'
result['maskStoreU'] = result['storeU'].replace(predicate, '{2}')
result['maskStoreA'] = result['storeA'].replace(predicate, '{2}')
result['maskStoreS'] = result['storeS'].replace(predicate, '{3}')
result['maskStream'] = result['stream'].replace(predicate, '{2}')
if instruction_set != 'sme':
result['maskStoreS'] = result['storeS'].replace(predicate, '{3}')
if instruction_set.startswith('sve2') and instruction_set not in ('sve256', 'sve2048'):
result['maskStreamS'] = result['streamS'].replace(predicate, '{3}')
if instruction_set != 'sve':
result['streamFence'] = '__dmb(15)'
if instruction_set == 'sme':
result['function_prefix'] = '__attribute__((arm_locally_streaming))'
elif instruction_set not in ['sve', 'sve2', 'sme']:
result['compile_flags'] = [f'-msve-vector-bits={bitwidth}']
else:
result['makeVecConst'] = f'vdupq_n_f{bits[data_type]}' + '({0})'
......@@ -151,9 +166,9 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
result['any'] = f'vaddlvq_u8(vreinterpretq_u8_u{bits[data_type]}({{0}})) > 0'
result['all'] = f'vaddlvq_u8(vreinterpretq_u8_u{bits[data_type]}({{0}})) == 16*0xff'
if instruction_set == 'sve' or bitwidth & (bitwidth - 1) == 0:
# only power-of-2 vector sizes will evenly divide a cacheline
result['cachelineSize'] = 'cachelineSize()'
# SVE has real nontemporal stores, so we only need to zero cachlines on Neon
result['cachelineZero'] = 'cachelineZero((void*) {0})'
result['cachelineSize'] = 'cachelineSize()'
return result