noxfile.py

+from __future__ import annotations
+from typing import Sequence
+from argparse import ArgumentParser
+
+import os
+import nox
+import subprocess
+import re
+
+nox.options.sessions = ["lint", "typecheck"]
+
+
+def get_cuda_version(session: nox.Session) -> None | tuple[int, ...]:
+    query_args = ["nvcc", "--version"]
+
+    try:
+        query_result = subprocess.run(query_args, capture_output=True)
+    except FileNotFoundError:
+        return None
+
+    matches = re.findall(r"release \d+\.\d+", str(query_result.stdout))
+    if matches:
+        match = matches[0]
+        version_string = match.split()[-1]
+        try:
+            return tuple(int(v) for v in version_string.split("."))
+        except ValueError:
+            pass
+
+    session.warn("nvcc was found, but I am unable to determine the CUDA version.")
+    return None
+
+
+def install_cupy(
+    session: nox.Session, cupy_version: str, skip_if_no_cuda: bool = False
+):
+    if cupy_version is not None:
+        cuda_version = get_cuda_version(session)
+        if cuda_version is None or cuda_version[0] not in (11, 12):
+            if skip_if_no_cuda:
+                session.skip(
+                    "No compatible installation of CUDA found - Need either CUDA 11 or 12"
+                )
+            else:
+                session.warn(
+                    "Running without cupy: no compatbile installation of CUDA found. Need either CUDA 11 or 12."
+                )
+                return
+
+        cuda_major = cuda_version[0]
+        cupy_package = f"cupy-cuda{cuda_major}x=={cupy_version}"
+        session.install(cupy_package)
+
+
+def check_external_doc_dependencies(session: nox.Session):
+    dot_args = ["dot", "--version"]
+    try:
+        _ = subprocess.run(dot_args, capture_output=True)
+    except FileNotFoundError:
+        session.error(
+            "Unable to build documentation: "
+            "Command `dot` from the `graphviz` package (https://www.graphviz.org/) is not available"
+        )
+
+
+def editable_install(session: nox.Session, opts: Sequence[str] = ()):
+    if opts:
+        opts_str = "[" + ",".join(opts) + "]"
+    else:
+        opts_str = ""
+    session.install("-e", f".{opts_str}")
+
+
+def install_pystencils_master(session: nox.Session):
+    session.install("git+https://i10git.cs.fau.de/pycodegen/pystencils.git@master")
+
+
+def install_sympy_master(session: nox.Session):
+    session.install("--upgrade", "git+https://github.com/sympy/sympy.git@master")
+
+
+@nox.session(python="3.10", tags=["qa", "code-quality"])
+def lint(session: nox.Session):
+    """Lint code using flake8"""
+
+    session.install("flake8")
+    session.run("flake8", "src/lbmpy")
+
+
+@nox.session(python="3.10", tags=["qa", "code-quality"])
+def typecheck(session: nox.Session):
+    """Run MyPy for static type checking"""
+    editable_install(session)
+    session.install("mypy")
+    session.run("mypy", "src/lbmpy")
+
+
+def run_testsuite(session: nox.Session, coverage: bool = True):
+    num_cores = os.cpu_count()
+
+    args = [
+        "pytest",
+        "-v",
+        "-n",
+        str(num_cores),
+        "-m",
+        "not longrun",
+        "--html",
+        "test-report/index.html",
+        "--junitxml=report.xml",
+    ]
+
+    if coverage:
+        args += [
+            "--cov-report=term",
+            "--cov=.",
+        ]
+
+    session.run(*args)
+
+    if coverage:
+        session.run("coverage", "html")
+        session.run("coverage", "xml")
+
+
+@nox.session(python=["3.10", "3.11", "3.12", "3.13"])
+def testsuite_cpu(session: nox.Session):
+    install_pystencils_master(session)
+    editable_install(session, ["alltrafos", "use_cython", "interactive", "tests"])
+    run_testsuite(session, coverage=False)
+
+
+@nox.session(python=["3.10", "3.11", "3.12", "3.13"])
+@nox.parametrize("cupy_version", ["12", "13"], ids=["cupy12", "cupy13"])
+def testsuite_gpu(session: nox.Session, cupy_version: str | None):
+    install_cupy(session, cupy_version, skip_if_no_cuda=True)
+    install_pystencils_master(session)
+    editable_install(session, ["alltrafos", "use_cython", "interactive", "tests"])
+    run_testsuite(session)
+
+
+@nox.parametrize("cupy_version", [None, "12", "13"], ids=["cpu", "cupy12", "cupy13"])
+@nox.session(python="3.10", tags=["test"])
+def testsuite_pystencils2(session: nox.Session, cupy_version: str | None):
+    if cupy_version is not None:
+        install_cupy(session, cupy_version, skip_if_no_cuda=True)
+
+    session.install(
+        "git+https://i10git.cs.fau.de/pycodegen/pystencils.git@v2.0-dev"
+    )
+    editable_install(session, ["alltrafos", "use_cython", "interactive", "tests"])
+
+    run_testsuite(session)
+
+
+@nox.session
+def quicktest(session: nox.Session):
+    parser = ArgumentParser()
+    parser.add_argument(
+        "--sympy-master", action="store_true", help="Use latest SymPy master revision"
+    )
+    args = parser.parse_args(session.posargs)
+
+    install_pystencils_master(session)
+    editable_install(session)
+
+    if args.sympy_master:
+        install_sympy_master(session)
+
+    session.run("python", "quicktest.py")

pyproject.toml

+[project]
+name = "lbmpy"
+description = "Code Generation for Lattice Boltzmann Methods"
+dynamic = ["version"]
+readme = "README.md"
+authors = [
+    { name = "Martin Bauer" },
+    { name = "Markus Holzer" },
+    { name = "Frederik Hennig" },
+    { email = "cs10-codegen@fau.de" },
+]
+license = { file = "COPYING.txt" }
+requires-python = ">=3.10"
+dependencies = ["pystencils>=1.3", "sympy>=1.12", "numpy>=1.8.0", "appdirs", "joblib", "packaging"]
+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/lbmpy/-/issues"
+"Documentation" = "https://pycodegen.pages.i10git.cs.fau.de/lbmpy/"
+"Source Code" = "https://i10git.cs.fau.de/pycodegen/lbmpy"
+
+[project.optional-dependencies]
+gpu = ['cupy']
+alltrafos = ['islpy', 'py-cpuinfo']
+bench_db = ['blitzdb', 'pymongo', 'pandas']
+interactive = [
+    'matplotlib',
+    'ipy_table',
+    'imageio',
+    'jupyter',
+    'pyevtk',
+    'rich',
+    'graphviz',
+    'scipy',
+    'scikit-image'
+]
+use_cython = [
+    'Cython'
+]
+doc = [
+    'sphinx',
+    'sphinx_rtd_theme',
+    'nbsphinx',
+    'sphinxcontrib-bibtex',
+    'sphinx_autodoc_typehints',
+    'pandoc',
+]
+tests = [
+    'pytest',
+    'pytest-cov',
+    'pytest-html',
+    'ansi2html',
+    'pytest-xdist',
+    'flake8',
+    'nbformat',
+    'nbconvert',
+    'ipython',
+    'randomgen>=1.18',
+]
+
+[build-system]
+requires = [
+    "setuptools>=69",
+    "versioneer[toml]>=0.29",
+]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.package-data]
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["lbmpy", "lbmpy.*"]
+namespaces = false
+
+[tool.versioneer]
+# 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.
+VCS = "git"
+style = "pep440"
+versionfile_source = "src/lbmpy/_version.py"
+versionfile_build = "lbmpy/_version.py"
+tag_prefix = "release/"
+parentdir_prefix = "lbmpy-"

pytest.ini

 [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
+addopts = 
+       --doctest-modules --durations=20
+       --cov-config pytest.ini
+       --ignore=src/lbmpy/custom_code_nodes.py
+       --ignore=src/lbmpy/lookup_tables.py
+       --ignore=src/lbmpy/phasefield_allen_cahn/contact_angle.py
 markers =
        longrun: tests only run at night since they have large execution time
-       notebook: jupyter notebooks
+       notebook: mark for notebooks
 # these warnings all come from third party libraries.
 filterwarnings =
        ignore:the imp module is deprecated in favour of importlib:DeprecationWarning
@@ -14,15 +21,19 @@ filterwarnings =
 
 [run]
 branch = True
-source = lbmpy
-         lbmpy_tests
+source = src/lbmpy
+         tests
 
 omit = doc/*
-       lbmpy_tests/*
+       tests/*
        setup.py
        conftest.py
        versioneer.py
-       lbmpy/_version.py
+       quicktest.py
+       noxfile.py
+       src/lbmpy/_version.py
+       src/lbmpy/_compat.py
+       venv/
 
 [report]
 exclude_lines =
@@ -35,6 +46,7 @@ exclude_lines =
        # Don't complain if tests don't hit defensive assertion code:
        raise AssertionError
        raise NotImplementedError
+       NotImplementedError()
        #raise ValueError
 
        # Don't complain if non-runnable code isn't run:
@@ -43,7 +55,7 @@ exclude_lines =
        if __name__ == .__main__.:
 
 skip_covered = True
-fail_under = 88
+fail_under = 87
 
 [html]
 directory = coverage_report

quicktest.py

+#!/usr/bin/env python3
+
+from contextlib import redirect_stdout
+import io
+from tests.test_quicktests import (
+    test_poiseuille_channel_quicktest,
+    test_entropic_methods,
+    test_cumulant_ldc
+)
+
+quick_tests = [
+    test_poiseuille_channel_quicktest,
+    test_entropic_methods,
+    test_cumulant_ldc,
+]
+
+if __name__ == "__main__":
+    print("Running lbmpy quicktests")
+    for qt in quick_tests:
+        print(f"   -> {qt.__name__}")
+        with redirect_stdout(io.StringIO()):
+            qt()

setup.cfg

-# 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 = lbmpy/_version.py
-versionfile_build = lbmpy/_version.py
-tag_prefix = release/
-parentdir_prefix = lbmpy-
\ No newline at end of file

setup.py

-import os
-import io
-from setuptools import setup, find_packages
-import distutils
-from contextlib import redirect_stdout
-from importlib import import_module
+from setuptools import setup, __version__ as setuptools_version
 
-import versioneer
-
-try:
-    import cython  # noqa
-
-    USE_CYTHON = True
-except ImportError:
-    USE_CYTHON = False
-
-quick_tests = [
-    'test_quicktests.test_poiseuille_channel_quicktest',
-    'test_quicktests.test_entropic_methods',
-    'test_quicktests.test_cumulant_ldc',
-]
-
-
-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 = 'lbmpy_tests.' + test
-        mod, function_name = test.rsplit('.', 1)
-        if isinstance(mod, str):
-            mod = import_module(mod)
+if int(setuptools_version.split('.')[0]) < 61:
+    raise Exception(
+        "[ERROR] lbmpy 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 lbmpy into a virtual environment instead."
+    )
 
-        func = getattr(mod, function_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
+import versioneer
 
 
 def get_cmdclass():
-    cmdclass = {"quicktest": SimpleTestRunner}
-    cmdclass.update(versioneer.get_cmdclass())
-    return cmdclass
+    return versioneer.get_cmdclass()
 
 
-major_version = versioneer.get_version().split("+")[0]
-setup(name='lbmpy',
-      version=versioneer.get_version(),
-      description='Code Generation for Lattice Boltzmann Methods',
-      long_description=readme(),
-      long_description_content_type="text/markdown",
-      author='Martin Bauer, Markus Holzer, Frederik Hennig',
-      license='AGPLv3',
-      author_email='cs10-codegen@fau.de',
-      url='https://i10git.cs.fau.de/pycodegen/lbmpy/',
-      packages=['lbmpy'] + ['lbmpy.' + s for s in find_packages('lbmpy')],
-      install_requires=[f'pystencils>=0.4.0,<={major_version}', 'sympy>=1.5.1,<=1.11.1', 'numpy>=1.11.0'],
-      package_data={'lbmpy': ['phasefield/simplex_projection.pyx', 'phasefield/simplex_projection.c']},
-      ext_modules=cython_extensions("lbmpy.phasefield.simplex_projection"),
-      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+)',
-      ],
-      python_requires=">=3.8",
-      extras_require={
-          'gpu': ['pycuda'],
-          'opencl': ['pyopencl'],
-          'alltrafos': ['islpy', 'py-cpuinfo'],
-          'interactive': ['scipy', 'scikit-image', 'cython', 'matplotlib',
-                          'ipy_table', 'imageio', 'jupyter', 'pyevtk'],
-          'doc': ['sphinx', 'sphinx_rtd_theme', 'nbsphinx',
-                  'sphinxcontrib-bibtex', 'sphinx_autodoc_typehints', 'pandoc'],
-          'phasefield': ['Cython']
-      },
-      cmdclass=get_cmdclass()
-      )
+setup(
+    version=versioneer.get_version(),
+    cmdclass=get_cmdclass(),
+)

src/lbmpy/__init__.py

+from .creationfunctions import (
+    create_lb_ast,
+    create_lb_collision_rule,
+    create_lb_function,
+    create_lb_method,
+    create_lb_update_rule,
+    LBMConfig,
+    LBMOptimisation,
+)
+from .enums import Stencil, Method, ForceModel, CollisionSpace, SubgridScaleModel
+from .lbstep import LatticeBoltzmannStep
+from .macroscopic_value_kernels import (
+    pdf_initialization_assignments,
+    macroscopic_values_getter,
+    strain_rate_tensor_getter,
+    compile_macroscopic_values_getter,
+    compile_macroscopic_values_setter,
+    create_advanced_velocity_setter_collision_rule,
+)
+from .maxwellian_equilibrium import get_weights
+from .relaxationrates import (
+    relaxation_rate_from_lattice_viscosity,
+    lattice_viscosity_from_relaxation_rate,
+    relaxation_rate_from_magic_number,
+)
+from .scenarios import create_lid_driven_cavity, create_fully_periodic_flow
+from .stencils import LBStencil
+
+
+__all__ = [
+    "create_lb_ast",
+    "create_lb_collision_rule",
+    "create_lb_function",
+    "create_lb_method",
+    "create_lb_update_rule",
+    "LBMConfig",
+    "LBMOptimisation",
+    "Stencil",
+    "Method",
+    "ForceModel",
+    "CollisionSpace",
+    "SubgridScaleModel",
+    "LatticeBoltzmannStep",
+    "pdf_initialization_assignments",
+    "macroscopic_values_getter",
+    "strain_rate_tensor_getter",
+    "compile_macroscopic_values_getter",
+    "compile_macroscopic_values_setter",
+    "create_advanced_velocity_setter_collision_rule",
+    "get_weights",
+    "relaxation_rate_from_lattice_viscosity",
+    "lattice_viscosity_from_relaxation_rate",
+    "relaxation_rate_from_magic_number",
+    "create_lid_driven_cavity",
+    "create_fully_periodic_flow",
+    "LBStencil",
+]
+
+
+from . import _version
+__version__ = _version.get_versions()['version']

src/lbmpy/_compat.py

+from pystencils import __version__ as ps_version
+
+#   Determine if we're running pystencils 1.x or 2.x
+version_tokes = ps_version.split(".")
+
+PYSTENCILS_VERSION_MAJOR = int(version_tokes[0])
+IS_PYSTENCILS_2 = PYSTENCILS_VERSION_MAJOR == 2
+
+if IS_PYSTENCILS_2:
+    from pystencils.defaults import DEFAULTS
+
+    def get_loop_counter_symbol(coord: int):
+        return DEFAULTS.spatial_counters[coord]
+
+    def get_supported_instruction_sets():
+        from pystencils import Target
+        vector_targets = Target.available_vector_cpu_targets()
+        isas = []
+        for target in vector_targets:
+            tokens = target.name.split("_")
+            isas.append(tokens[-1].lower())
+        return isas
+
+else:
+    from pystencils.backends.simd_instruction_sets import (
+        get_supported_instruction_sets as get_supported_instruction_sets_,
+    )
+
+    get_supported_instruction_sets = get_supported_instruction_sets_
+
+    def get_loop_counter_symbol(coord: int):
+        from pystencils.astnodes import LoopOverCoordinate
+
+        return LoopOverCoordinate.get_loop_counter_symbol(coord)
+
+
+def import_guard_pystencils1(feature):
+    if IS_PYSTENCILS_2:
+        raise ImportError(
+            f"The following feature is not yet available when running pystencils 2.x: {feature}"
+        )
+    return True

lbmpy/advanced_streaming/__init__.py → src/lbmpy/advanced_streaming/__init__.py

-from .indexing import BetweenTimestepsIndexing, NeighbourOffsetArrays
+from .indexing import BetweenTimestepsIndexing
 from .communication import get_communication_slices, LBMPeriodicityHandling
 from .utility import Timestep, get_accessor, is_inplace, get_timesteps, \
     numeric_index, numeric_offsets, inverse_dir_index, AccessPdfValues
 
-__all__ = ['BetweenTimestepsIndexing', 'NeighbourOffsetArrays',
+__all__ = ['BetweenTimestepsIndexing',
            'get_communication_slices', 'LBMPeriodicityHandling',
            'Timestep', 'get_accessor', 'is_inplace', 'get_timesteps',
            'numeric_index', 'numeric_offsets', 'inverse_dir_index', 'AccessPdfValues']

lbmpy/advanced_streaming/communication.py → src/lbmpy/advanced_streaming/communication.py

 import itertools
-from pystencils import Field, Assignment
-from pystencils.slicing import shift_slice, get_slice_before_ghost_layer, normalize_slice
-from lbmpy.advanced_streaming.utility import is_inplace, get_accessor, numeric_index, \
-    numeric_offsets, Timestep, get_timesteps
+from pystencils import CreateKernelConfig, Field, Assignment, AssignmentCollection, Target
+from pystencils.slicing import (
+    shift_slice,
+    get_slice_before_ghost_layer,
+    normalize_slice,
+)
+from lbmpy.advanced_streaming.utility import (
+    is_inplace,
+    get_accessor,
+    numeric_index,
+    Timestep,
+    get_timesteps,
+    numeric_offsets,
+)
 from pystencils.datahandling import SerialDataHandling
-from pystencils.enums import Target
 from itertools import chain
 
 
-def _trim_slice_in_direction(slices, direction):
-    assert len(slices) == len(direction)
+class LBMPeriodicityHandling:
 
-    result = []
-    for s, d in zip(slices, direction):
-        if isinstance(s, int):
-            result.append(s)
-            continue
-        start = s.start + 1 if d == -1 else s.start
-        stop = s.stop - 1 if d == 1 else s.stop
-        result.append(slice(start, stop, s.step))
+    def __init__(
+        self,
+        stencil,
+        data_handling,
+        pdf_field_name,
+        streaming_pattern="pull",
+        ghost_layers=1,
+        cupy_direct_copy=True,
+    ):
+        """
+        Periodicity Handling for Lattice Boltzmann Streaming.
+
+        **On the usage with cuda:**
+        - cupy allows the copying of sliced arrays within device memory using the numpy syntax,
+        e.g. `dst[:,0] = src[:,-1]`. In this implementation, this is the default for periodicity
+        handling. Alternatively, if you set `cupy_direct_copy=False`, GPU kernels are generated and
+        compiled. The compiled kernels are almost twice as fast in execution as cupy array copying,
+        but especially for large stencils like D3Q27, their compilation can take up to 20 seconds.
+        Choose your weapon depending on your use case.
+        """
+        if not isinstance(data_handling, SerialDataHandling):
+            raise ValueError("Only serial data handling is supported!")
 
-    return tuple(result)
+        self.stencil = stencil
+        self.dim = stencil.D
+        self.dh = data_handling
 
+        assert data_handling.default_target in [Target.CPU, Target.GPU]
+        self.target = data_handling.default_target
 
-def _extend_dir(direction):
-    if len(direction) == 0:
-        yield tuple()
-    elif direction[0] == 0:
-        for d in [-1, 0, 1]:
-            for rest in _extend_dir(direction[1:]):
-                yield (d, ) + rest
-    else:
-        for rest in _extend_dir(direction[1:]):
-            yield (direction[0], ) + rest
+        self.pdf_field_name = pdf_field_name
+        self.ghost_layers = ghost_layers
+        self.periodicity = data_handling.periodicity
+        self.inplace_pattern = is_inplace(streaming_pattern)
 
+        self.cpu = self.target == Target.CPU
+        self.cupy_direct_copy = self.target == Target.GPU and cupy_direct_copy
 
-def _get_neighbour_transform(direction, ghost_layers):
-    return tuple(d * (ghost_layers + 1) for d in direction)
+        def is_copy_direction(direction):
+            s = 0
+            for d, p in zip(direction, self.periodicity):
+                s += abs(d)
+                if d != 0 and not p:
+                    return False
 
+            return s != 0
 
-def _fix_length_one_slices(slices):
-    """Slices of length one are replaced by their start value for correct periodic shifting"""
-    if isinstance(slices, int):
-        return slices
-    elif isinstance(slices, slice):
-        if slices.stop is not None and abs(slices.start - slices.stop) == 1:
-            return slices.start
-        elif slices.stop is None and slices.start == -1:
-            return -1  # [-1:] also has length one
+        full_stencil = itertools.product(*([-1, 0, 1] for _ in range(self.dim)))
+        copy_directions = tuple(filter(is_copy_direction, full_stencil))
+        self.comm_slices = []
+        timesteps = get_timesteps(streaming_pattern)
+        for timestep in timesteps:
+            slices_per_comm_dir = get_communication_slices(
+                stencil=stencil,
+                comm_stencil=copy_directions,
+                streaming_pattern=streaming_pattern,
+                prev_timestep=timestep,
+                ghost_layers=ghost_layers,
+            )
+            self.comm_slices.append(
+                list(chain.from_iterable(v for k, v in slices_per_comm_dir.items()))
+            )
+
+        if self.target == Target.GPU and not cupy_direct_copy:
+            self.device_copy_kernels = list()
+            for timestep in timesteps:
+                self.device_copy_kernels.append(self._compile_copy_kernels(timestep))
+
+    def __call__(self, prev_timestep=Timestep.BOTH):
+        if self.cpu:
+            self._periodicity_handling_cpu(prev_timestep)
         else:
-            return slices
-    else:
-        return tuple(_fix_length_one_slices(s) for s in slices)
+            self._periodicity_handling_gpu(prev_timestep)
+
+    def _periodicity_handling_cpu(self, prev_timestep):
+        arr = self.dh.cpu_arrays[self.pdf_field_name]
+        comm_slices = self.comm_slices[prev_timestep.idx]
+        for src, dst in comm_slices:
+            arr[dst] = arr[src]
+
+    def _compile_copy_kernels(self, timestep):
+        assert self.target == Target.GPU
+        pdf_field = self.dh.fields[self.pdf_field_name]
+        kernels = []
+        for src, dst in self.comm_slices[timestep.idx]:
+            kernels.append(periodic_pdf_gpu_copy_kernel(pdf_field, src, dst))
+        return kernels
+
+    def _periodicity_handling_gpu(self, prev_timestep):
+        arr = self.dh.gpu_arrays[self.pdf_field_name]
+        if self.cupy_direct_copy:
+            for src, dst in self.comm_slices[prev_timestep.idx]:
+                arr[dst] = arr[src]
+        else:
+            kernel_args = {self.pdf_field_name: arr}
+            for kernel in self.device_copy_kernels[prev_timestep.idx]:
+                kernel(**kernel_args)
 
 
 def get_communication_slices(
-        stencil, comm_stencil=None, streaming_pattern='pull', prev_timestep=Timestep.BOTH, ghost_layers=1):
+    stencil,
+    comm_stencil=None,
+    streaming_pattern="pull",
+    prev_timestep=Timestep.BOTH,
+    ghost_layers=1,
+):
     """
     Return the source and destination slices for periodicity handling or communication between blocks.
 
     :param stencil: The stencil used by the LB method.
-    :param comm_stencil: The stencil defining the communication directions. If None, it will be set to the 
+    :param comm_stencil: The stencil defining the communication directions. If None, it will be set to the
                          full stencil (D2Q9 in 2D, D3Q27 in 3D, etc.).
     :param streaming_pattern: The streaming pattern.
     :param prev_timestep: Timestep after which communication is run.
@@ -71,7 +140,9 @@ def get_communication_slices(
     if comm_stencil is None:
         comm_stencil = itertools.product(*([-1, 0, 1] for _ in range(stencil.D)))
 
-    pdfs = Field.create_generic('pdfs', spatial_dimensions=len(stencil[0]), index_shape=(stencil.Q,))
+    pdfs = Field.create_generic(
+        "pdfs", spatial_dimensions=len(stencil[0]), index_shape=(stencil.Q,)
+    )
     write_accesses = get_accessor(streaming_pattern, prev_timestep).write(pdfs, stencil)
     slices_per_comm_direction = dict()
 
@@ -83,19 +154,27 @@ def get_communication_slices(
 
         for streaming_dir in set(_extend_dir(comm_dir)) & set(stencil):
             d = stencil.index(streaming_dir)
-            write_offsets = numeric_offsets(write_accesses[d])
             write_index = numeric_index(write_accesses[d])[0]
 
+            origin_slice = get_slice_before_ghost_layer(
+                comm_dir, ghost_layers=ghost_layers, thickness=1
+            )
+            src_slice = _fix_length_one_slices(origin_slice)
+
+            write_offsets = numeric_offsets(write_accesses[d])
             tangential_dir = tuple(s - c for s, c in zip(streaming_dir, comm_dir))
-            origin_slice = get_slice_before_ghost_layer(comm_dir, ghost_layers=ghost_layers, thickness=1)
-            origin_slice = _fix_length_one_slices(origin_slice)
-            src_slice = shift_slice(_trim_slice_in_direction(origin_slice, tangential_dir), write_offsets)
+
+            # TODO: this is just a hotfix. _trim_slice_in_direction breaks FreeSlip BC with adjacent periodic side
+            if streaming_pattern != "pull":
+                src_slice = shift_slice(
+                    _trim_slice_in_direction(src_slice, tangential_dir), write_offsets
+                )
 
             neighbour_transform = _get_neighbour_transform(comm_dir, ghost_layers)
             dst_slice = shift_slice(src_slice, neighbour_transform)
 
-            src_slice = src_slice + (write_index, )
-            dst_slice = dst_slice + (write_index, )
+            src_slice = src_slice + (write_index,)
+            dst_slice = dst_slice + (write_index,)
 
             slices_for_dir.append((src_slice, dst_slice))
 
@@ -103,10 +182,10 @@ def get_communication_slices(
     return slices_per_comm_direction
 
 
-def periodic_pdf_copy_kernel(pdf_field, src_slice, dst_slice,
-                             domain_size=None, target=Target.GPU):
-    """Copies a rectangular array slice onto another non-overlapping array slice"""
-    from pystencils.gpucuda.kernelcreation import create_cuda_kernel
+def periodic_pdf_gpu_copy_kernel(pdf_field, src_slice, dst_slice, domain_size=None):
+    """Generate a GPU kernel which copies all values from one slice of a field
+    to another non-overlapping slice."""
+    from pystencils import create_kernel
 
     pdf_idx = src_slice[-1]
     assert isinstance(pdf_idx, int), "PDF index needs to be an integer constant"
@@ -114,6 +193,7 @@ def periodic_pdf_copy_kernel(pdf_field, src_slice, dst_slice,
     src_slice = src_slice[:-1]
     dst_slice = dst_slice[:-1]
 
+    # TODO this is the domain_size with GL
     if domain_size is None:
         domain_size = pdf_field.spatial_shape
 
@@ -126,105 +206,71 @@ def periodic_pdf_copy_kernel(pdf_field, src_slice, dst_slice,
     def _stop(s):
         return s.stop if isinstance(s, slice) else s
 
-    offset = [_start(s1) - _start(s2) for s1, s2 in zip(normalized_from_slice, normalized_to_slice)]
-    assert offset == [_stop(s1) - _stop(s2) for s1, s2 in zip(normalized_from_slice, normalized_to_slice)], \
-        "Slices have to have same size"
+    offset = [
+        _start(s1) - _start(s2)
+        for s1, s2 in zip(normalized_from_slice, normalized_to_slice)
+    ]
+    assert offset == [
+        _stop(s1) - _stop(s2)
+        for s1, s2 in zip(normalized_from_slice, normalized_to_slice)
+    ], "Slices have to have same size"
+
+    copy_eq = AssignmentCollection(
+        main_assignments=[
+            Assignment(pdf_field(pdf_idx), pdf_field[tuple(offset)](pdf_idx))
+        ]
+    )
+    config = CreateKernelConfig(
+        iteration_slice=dst_slice,
+        skip_independence_check=True,
+        target=Target.GPU,
+    )
+
+    ast = create_kernel(copy_eq, config=config)
+    return ast.compile()
 
-    copy_eq = Assignment(pdf_field(pdf_idx), pdf_field[tuple(offset)](pdf_idx))
-    ast = create_cuda_kernel([copy_eq], iteration_slice=dst_slice, skip_independence_check=True)
-    if target == Target.GPU:
-        from pystencils.gpucuda import make_python_function
-        return make_python_function(ast)
-    else:
-        raise ValueError('Invalid target:', target)
-
-
-class LBMPeriodicityHandling:
-
-    def __init__(self, stencil, data_handling, pdf_field_name,
-                 streaming_pattern='pull', ghost_layers=1,
-                 pycuda_direct_copy=True):
-        """
-            Periodicity Handling for Lattice Boltzmann Streaming.
-
-            **On the usage with cuda:**
-            - pycuda allows the copying of sliced arrays within device memory using the numpy syntax,
-            e.g. `dst[:,0] = src[:,-1]`. In this implementation, this is the default for periodicity
-            handling. Alternatively, if you set `pycuda_direct_copy=False`, GPU kernels are generated and
-            compiled. The compiled kernels are almost twice as fast in execution as pycuda array copying,
-            but especially for large stencils like D3Q27, their compilation can take up to 20 seconds. 
-            Choose your weapon depending on your use case.
-        """
-        if not isinstance(data_handling, SerialDataHandling):
-            raise ValueError('Only serial data handling is supported!')
-
-        self.stencil = stencil
-        self.dim = stencil.D
-        self.dh = data_handling
-
-        target = data_handling.default_target
-        assert target in [Target.CPU, Target.GPU]
 
-        self.pdf_field_name = pdf_field_name
-        self.ghost_layers = ghost_layers
-        periodicity = data_handling.periodicity
-        self.inplace_pattern = is_inplace(streaming_pattern)
-        self.target = target
-        self.cpu = target == Target.CPU
-        self.pycuda_direct_copy = target == Target.GPU and pycuda_direct_copy
+def _extend_dir(direction):
+    if len(direction) == 0:
+        yield tuple()
+    elif direction[0] == 0:
+        for d in [-1, 0, 1]:
+            for rest in _extend_dir(direction[1:]):
+                yield (d,) + rest
+    else:
+        for rest in _extend_dir(direction[1:]):
+            yield (direction[0],) + rest
 
-        def is_copy_direction(direction):
-            s = 0
-            for d, p in zip(direction, periodicity):
-                s += abs(d)
-                if d != 0 and not p:
-                    return False
 
-            return s != 0
+def _get_neighbour_transform(direction, ghost_layers):
+    return tuple(d * (ghost_layers + 1) for d in direction)
 
-        full_stencil = itertools.product(*([-1, 0, 1] for _ in range(self.dim)))
-        copy_directions = tuple(filter(is_copy_direction, full_stencil))
-        self.comm_slices = []
-        timesteps = get_timesteps(streaming_pattern)
-        for timestep in timesteps:
-            slices_per_comm_dir = get_communication_slices(stencil=stencil,
-                                                           comm_stencil=copy_directions,
-                                                           streaming_pattern=streaming_pattern,
-                                                           prev_timestep=timestep,
-                                                           ghost_layers=ghost_layers)
-            self.comm_slices.append(list(chain.from_iterable(v for k, v in slices_per_comm_dir.items())))
-
-        if target == Target.GPU and not pycuda_direct_copy:
-            self.device_copy_kernels = []
-            for timestep in timesteps:
-                self.device_copy_kernels.append(self._compile_copy_kernels(timestep))
 
-    def __call__(self, prev_timestep=Timestep.BOTH):
-        if self.cpu:
-            self._periodicity_handling_cpu(prev_timestep)
+def _fix_length_one_slices(slices):
+    """Slices of length one are replaced by their start value for correct periodic shifting"""
+    if isinstance(slices, int):
+        return slices
+    elif isinstance(slices, slice):
+        if slices.stop is not None and abs(slices.start - slices.stop) == 1:
+            return slices.start
+        elif slices.stop is None and slices.start == -1:
+            return -1  # [-1:] also has length one
         else:
-            self._periodicity_handling_gpu(prev_timestep)
+            return slices
+    else:
+        return tuple(_fix_length_one_slices(s) for s in slices)
 
-    def _periodicity_handling_cpu(self, prev_timestep):
-        arr = self.dh.cpu_arrays[self.pdf_field_name]
-        comm_slices = self.comm_slices[prev_timestep.idx]
-        for src, dst in comm_slices:
-            arr[dst] = arr[src]
 
-    def _compile_copy_kernels(self, timestep):
-        pdf_field = self.dh.fields[self.pdf_field_name]
-        kernels = []
-        for src, dst in self.comm_slices[timestep.idx]:
-            kernels.append(
-                periodic_pdf_copy_kernel(pdf_field, src, dst, target=self.target))
-        return kernels
+def _trim_slice_in_direction(slices, direction):
+    assert len(slices) == len(direction)
 
-    def _periodicity_handling_gpu(self, prev_timestep):
-        arr = self.dh.gpu_arrays[self.pdf_field_name]
-        if self.pycuda_direct_copy:
-            for src, dst in self.comm_slices[prev_timestep.idx]:
-                arr[dst] = arr[src]
-        else:
-            kernel_args = {self.pdf_field_name: arr}
-            for kernel in self.device_copy_kernels[prev_timestep.idx]:
-                kernel(**kernel_args)
+    result = []
+    for s, d in zip(slices, direction):
+        if isinstance(s, int):
+            result.append(s)
+            continue
+        start = s.start + 1 if d == -1 else s.start
+        stop = s.stop - 1 if d == 1 else s.stop
+        result.append(slice(start, stop, s.step))
+
+    return tuple(result)

lbmpy/advanced_streaming/utility.py → src/lbmpy/advanced_streaming/utility.py

@@ -58,24 +58,27 @@ odd_accessors = {
 }
 
 
+def is_inplace(streaming_pattern):
+    if streaming_pattern not in streaming_patterns:
+        raise ValueError('Invalid streaming pattern', streaming_pattern)
+
+    return streaming_pattern in ['aa', 'esotwist', 'esopull', 'esopush']
+
+
 def get_accessor(streaming_pattern: str, timestep: Timestep) -> PdfFieldAccessor:
     if streaming_pattern not in streaming_patterns:
         raise ValueError(
             "Invalid value of parameter 'streaming_pattern'.", streaming_pattern)
 
+    if is_inplace(streaming_pattern) and (timestep == Timestep.BOTH):
+        raise ValueError(f"Invalid timestep for streaming pattern {streaming_pattern}: {str(timestep)}")
+
     if timestep == Timestep.EVEN:
         return even_accessors[streaming_pattern]
     else:
         return odd_accessors[streaming_pattern]
 
 
-def is_inplace(streaming_pattern):
-    if streaming_pattern not in streaming_patterns:
-        raise ValueError('Invalid streaming pattern', streaming_pattern)
-
-    return streaming_pattern in ['aa', 'esotwist', 'esopull', 'esopush']
-
-
 def get_timesteps(streaming_pattern):
     return (Timestep.EVEN, Timestep.ODD) if is_inplace(streaming_pattern) else (Timestep.BOTH, )
 

src/lbmpy/boundaries/__init__.py

+from lbmpy.boundaries.boundaryconditions import (
+    UBB, FixedDensity, DiffusionDirichlet, SimpleExtrapolationOutflow, WallFunctionBounce,
+    ExtrapolationOutflow, NeumannByCopy, NoSlip, NoSlipLinearBouzidi, QuadraticBounceBack, StreamInConstant, FreeSlip)
+from lbmpy.boundaries.boundaryhandling import LatticeBoltzmannBoundaryHandling
+from lbmpy.boundaries.wall_function_models import MoninObukhovSimilarityTheory, LogLaw, MuskerLaw, SpaldingsLaw
+
+__all__ = ['NoSlip', 'NoSlipLinearBouzidi', 'QuadraticBounceBack', 'FreeSlip', 'WallFunctionBounce',
+           'UBB', 'FixedDensity',
+           'SimpleExtrapolationOutflow', 'ExtrapolationOutflow',
+           'DiffusionDirichlet', 'NeumannByCopy', 'StreamInConstant',
+           'LatticeBoltzmannBoundaryHandling',
+           'MoninObukhovSimilarityTheory', 'LogLaw', 'MuskerLaw', 'SpaldingsLaw']