doc/sphinx/api.rst

@@ -5,10 +5,12 @@ API Reference
    :maxdepth: 3
 
    kernel_compile_and_call.rst
+   enums.rst
    simplifications.rst
    datahandling.rst
    configuration.rst
    field.rst
+   stencil.rst
    finite_differences.rst
    plot.rst
    ast.rst

doc/sphinx/enums.rst

+************
+Enumerations
+************
+
+.. automodule:: pystencils.enums
+   :members:

doc/sphinx/kernel_compile_and_call.rst

@@ -8,9 +8,14 @@ Creating kernels
 
 .. autofunction:: pystencils.create_kernel
 
-.. autofunction:: pystencils.create_indexed_kernel
+.. autoclass:: pystencils.CreateKernelConfig
+    :members:
 
-.. autofunction:: pystencils.create_staggered_kernel
+.. autofunction:: pystencils.kernelcreation.create_domain_kernel
+
+.. autofunction:: pystencils.kernelcreation.create_indexed_kernel
+
+.. autofunction:: pystencils.kernelcreation.create_staggered_kernel
 
 
 Code printing
@@ -22,11 +27,11 @@ Code printing
 GPU Indexing
 -------------
 
-.. autoclass:: pystencils.gpucuda.AbstractIndexing
+.. autoclass:: pystencils.gpu.AbstractIndexing
    :members:
 
-.. autoclass:: pystencils.gpucuda.BlockIndexing
+.. autoclass:: pystencils.gpu.BlockIndexing
    :members:
 
-.. autoclass:: pystencils.gpucuda.LineIndexing
+.. autoclass:: pystencils.gpu.LineIndexing
    :members:

doc/sphinx/plot.rst

@@ -2,7 +2,7 @@
 Plotting and Animation
 **********************
 
-.. automodule:: pystencils.plot2d
+.. automodule:: pystencils.plot
    :members:
 
 

doc/sphinx/simplifications.rst

@@ -10,13 +10,27 @@ AssignmentCollection
    :members:
 
 
+SimplificationStrategy
+======================
+
+.. autoclass:: pystencils.simp.SimplificationStrategy
+    :members:
+
 Simplifications
 ===============
 
-.. automodule:: pystencils.simp
-   :members:
+.. automodule:: pystencils.simp.simplifications
+    :members:
+
+Subexpression insertion
+=======================
 
+The subexpression insertions have the goal to insert subexpressions which will not reduce the number of FLOPs.
+For example a constant value kept as subexpression will lead to a new variable in the code which will occupy
+a register slot. On the other side a single variable could just be inserted in all assignments.
 
+.. automodule:: pystencils.simp.subexpression_insertion
+    :members:
 
 
 

doc/sphinx/stencil.rst

+*******
+Stencil
+*******
+
+.. automodule:: pystencils.stencil
+   :members:

doc/sphinx/tutorials.rst

@@ -15,6 +15,7 @@ It is a good idea to download them and run them directly to be able to play arou
    /notebooks/05_tutorial_phasefield_spinodal_decomposition.ipynb
    /notebooks/06_tutorial_phasefield_dentritic_growth.ipynb
    /notebooks/demo_assignment_collection.ipynb
+   /notebooks/demo_plotting_and_animation.ipynb
+   /notebooks/demo_derivatives.ipynb
    /notebooks/demo_benchmark.ipynb
    /notebooks/demo_wave_equation.ipynb
-

pyproject.toml

+[project]
+name = "pystencils"
+description = "Speeding up stencil computations on CPUs and GPUs"
+dynamic = ["version"]
+readme = "README.md"
+authors = [
+    { name = "Martin Bauer" },
+    { name = "Jan Hönig " },
+    { name = "Markus Holzer" },
+    { name = "Frederik Hennig" },
+    { email = "cs10-codegen@fau.de" },
+]
+license = { file = "COPYING.txt" }
+requires-python = ">=3.10"
+dependencies = ["sympy>=1.9,<=1.12.1", "numpy>=1.8.0", "appdirs", "joblib", "pyyaml", "fasteners"]
+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" = "https://pycodegen.pages.i10git.cs.fau.de/pystencils/"
+"Source Code" = "https://i10git.cs.fau.de/pycodegen/pystencils"
+
+[project.optional-dependencies]
+gpu = ['cupy']
+alltrafos = ['islpy', 'py-cpuinfo']
+bench_db = ['blitzdb', 'pymongo', 'pandas']
+interactive = [
+    'matplotlib',
+    'ipy_table',
+    'imageio',
+    'jupyter',
+    'pyevtk',
+    'rich',
+    'graphviz',
+]
+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',
+    'matplotlib',
+    'py-cpuinfo',
+    'randomgen>=1.18',
+]
+
+[build-system]
+requires = [
+    "setuptools>=61",
+    "versioneer[toml]>=0.29",
+    # 'Cython'
+]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.package-data]
+pystencils = [
+    "include/*.h",
+    "boundaries/createindexlistcython.pyx"
+]
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["pystencils", "pystencils.*"]
+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/pystencils/_version.py"
+versionfile_build = "pystencils/_version.py"
+tag_prefix = "release/"
+parentdir_prefix = "pystencils-"

pystencils/backends/simd_instruction_sets.py

-
-
-# noinspection SpellCheckingInspection
-def get_vector_instruction_set(data_type='double', instruction_set='avx'):
-    comparisons = {
-        '==': '_CMP_EQ_UQ',
-        '!=': '_CMP_NEQ_UQ',
-        '>=': '_CMP_GE_OQ',
-        '<=': '_CMP_LE_OQ',
-        '<': '_CMP_NGE_UQ',
-        '>': '_CMP_NLE_UQ',
-    }
-    base_names = {
-        '+': 'add[0, 1]',
-        '-': 'sub[0, 1]',
-        '*': 'mul[0, 1]',
-        '/': 'div[0, 1]',
-        '&': 'and[0, 1]',
-        '|': 'or[0, 1]',
-        'blendv': 'blendv[0, 1, 2]',
-
-        'sqrt': 'sqrt[0]',
-
-        'makeVec': 'set[]',
-        'makeZero': 'setzero[]',
-
-        'loadU': 'loadu[0]',
-        'loadA': 'load[0]',
-        'storeU': 'storeu[0,1]',
-        'storeA': 'store[0,1]',
-        'stream': 'stream[0,1]',
-    }
-    for comparison_op, constant in comparisons.items():
-        base_names[comparison_op] = 'cmp[0, 1, %s]' % (constant,)
-
-    headers = {
-        'avx512': ['<immintrin.h>'],
-        'avx': ['<immintrin.h>'],
-        'sse': ['<immintrin.h>', '<xmmintrin.h>', '<emmintrin.h>', '<pmmintrin.h>',
-                '<tmmintrin.h>', '<smmintrin.h>', '<nmmintrin.h>']
-    }
-
-    suffix = {
-        'double': 'pd',
-        'float': 'ps',
-    }
-    prefix = {
-        'sse': '_mm',
-        'avx': '_mm256',
-        'avx512': '_mm512',
-    }
-
-    width = {
-        ("double", "sse"): 2,
-        ("float", "sse"): 4,
-        ("double", "avx"): 4,
-        ("float", "avx"): 8,
-        ("double", "avx512"): 8,
-        ("float", "avx512"): 16,
-    }
-
-    result = {
-        'width': width[(data_type, instruction_set)],
-    }
-    pre = prefix[instruction_set]
-    suf = suffix[data_type]
-    for intrinsic_id, function_shortcut in base_names.items():
-        function_shortcut = function_shortcut.strip()
-        name = function_shortcut[:function_shortcut.index('[')]
-
-        if intrinsic_id == 'makeVec':
-            arg_string = "({})".format(",".join(["{0}"] * result['width']))
-        else:
-            args = function_shortcut[function_shortcut.index('[') + 1: -1]
-            arg_string = "("
-            for arg in args.split(","):
-                arg = arg.strip()
-                if not arg:
-                    continue
-                if arg in ('0', '1', '2', '3', '4', '5'):
-                    arg_string += "{" + arg + "},"
-                else:
-                    arg_string += arg + ","
-            arg_string = arg_string[:-1] + ")"
-        mask_suffix = '_mask' if instruction_set == 'avx512' and intrinsic_id in comparisons.keys() else ''
-        result[intrinsic_id] = pre + "_" + name + "_" + suf + mask_suffix + arg_string
-
-    result['dataTypePrefix'] = {
-        'double': "_" + pre + 'd',
-        'float': "_" + pre,
-    }
-
-    result['rsqrt'] = None
-    bit_width = result['width'] * (64 if data_type == 'double' else 32)
-    result['double'] = "__m%dd" % (bit_width,)
-    result['float'] = "__m%d" % (bit_width,)
-    result['int'] = "__m%di" % (bit_width,)
-    result['bool'] = "__m%dd" % (bit_width,)
-
-    result['headers'] = headers[instruction_set]
-    result['any'] = "%s_movemask_%s({0}) > 0" % (pre, suf)
-    result['all'] = "%s_movemask_%s({0}) == 0xF" % (pre, suf)
-
-    if instruction_set == 'avx512':
-        size = 8 if data_type == 'double' else 16
-        result['&'] = '_kand_mask%d({0}, {1})' % (size,)
-        result['|'] = '_kor_mask%d({0}, {1})' % (size,)
-        result['any'] = '!_ktestz_mask%d_u8({0}, {0})' % (size, )
-        result['all'] = '_kortestc_mask%d_u8({0}, {0})' % (size, )
-        result['blendv'] = '%s_mask_blend_%s({2}, {0}, {1})' % (pre, suf)
-        result['rsqrt'] = "_mm512_rsqrt14_%s({0})" % (suf,)
-        result['bool'] = "__mmask%d" % (size,)
-
-    if instruction_set == 'avx' and data_type == 'float':
-        result['rsqrt'] = "_mm256_rsqrt_ps({0})"
-
-    return result
-
-
-def get_supported_instruction_sets():
-    """List of supported instruction sets on current hardware, or None if query failed."""
-    try:
-        from cpuinfo import get_cpu_info
-    except ImportError:
-        return None
-
-    result = []
-    required_sse_flags = {'sse', 'sse2', 'ssse3', 'sse4_1', 'sse4_2'}
-    required_avx_flags = {'avx'}
-    required_avx512_flags = {'avx512f'}
-    flags = set(get_cpu_info()['flags'])
-    if flags.issuperset(required_sse_flags):
-        result.append("sse")
-    if flags.issuperset(required_avx_flags):
-        result.append("avx")
-    if flags.issuperset(required_avx512_flags):
-        result.append("avx512")
-    return result