dendra-solvers 0.3.1__tar.gz

dendra_solvers-0.3.1/CHANGELOG.md

@@ -0,0 +1,14 @@
+# Changelog
+
+Release notes are generated from conventional commits by Commitizen.
+
+## v0.3.1 (2026-09-24)
+
+### Fix
+
+- **release**: verify automatic version increment
+
+## v0.3.0
+
+Baseline version before automated release notes. Earlier changes are recorded
+in the Git history.

dendra_solvers-0.3.1/CONTRIBUTING.md

@@ -0,0 +1,76 @@
+# Contributing
+
+Internal development currently takes place on GitLab. Keep changes focused and
+include tests for changes to solver behavior. See the [README](README.md#testing)
+for building the extension and running the solver tests.
+
+## Development tools
+
+From a checkout, install the development dependency group in a Python 3.11+
+environment (release-tooling CI uses Python 3.12):
+
+```sh
+python -m pip install --upgrade pip
+python -m pip install --group dev
+```
+
+This installs the pinned Commitizen version and pytest without building the
+extension. Installing and testing the solver itself still requires the source
+build steps in the README.
+
+## Commit messages and versions
+
+Use [Conventional Commits](https://www.conventionalcommits.org/en/v1.0.0/) so
+Commitizen can produce useful changelog entries and choose version increments.
+The subject should describe the change, for example:
+
+```text
+fix(solvers): preserve gradients for broadcast inputs
+feat(solvers): support an additional solver variant
+docs: clarify source installation requirements
+```
+
+You can create a commit interactively with `cz commit`, or write the message
+yourself. If a merge request is squash-merged, its final squash commit message
+must follow the same convention. Check the most recent commit with:
+
+```sh
+cz check --rev-range HEAD~1..HEAD
+```
+
+The project stays on `0.x.y` versions during initial development:
+
+| Change | Version effect |
+| --- | --- |
+| `fix:`, `perf:`, or `refactor:` | Patch increment, such as `0.3.0` → `0.3.1` |
+| `feat:` | Minor increment, such as `0.3.0` → `0.4.0` |
+| A breaking change marked with `!` or a `BREAKING CHANGE:` footer | Minor increment while the project remains below `1.0` |
+| `docs:`, `test:`, `ci:`, `build:`, or `chore:` without a breaking change | No automatic version increment |
+
+Here, “pre-1.0” means versions such as `0.3.1`; it does not add Python package
+prerelease suffixes such as `a1`, `b1`, or `rc1`.
+
+GitLab CI updates `pyproject.toml`, `CHANGELOG.md`, and the corresponding version
+tag after the main-branch checks pass and a version increment is warranted.
+Contributors normally leave version changes and release tags to CI.
+
+Maintainers must enable same-project job-token pushes in GitLab and allow the
+triggering user to update the default branch and version tags. CI also requires
+an existing `v<current-version>` tag at a tested commit containing that package
+version. For this integration, establish `v0.3.0` once after the selected
+baseline commit passes its build and test jobs; CI will not invent that tag.
+
+## Testing release tooling
+
+The release tooling tests use temporary Git repositories and the pinned
+Commitizen installation. They do not require the native solver extension:
+
+```sh
+python -m pytest tools/tests -q -ra
+```
+
+Run the solver suite separately after building the extension:
+
+```sh
+python -m pytest -q -ra
+```

dendra_solvers-0.3.1/MANIFEST.in

@@ -0,0 +1,6 @@
+include README.md
+include CHANGELOG.md CONTRIBUTING.md
+recursive-include src/dendra_solvers *.cpp *.h
+recursive-include tests *.py
+recursive-exclude src *.egg-info/*
+global-exclude __pycache__ *.py[cod] .DS_Store

dendra_solvers-0.3.1/PKG-INFO

@@ -0,0 +1,75 @@
+Metadata-Version: 2.4
+Name: dendra-solvers
+Version: 0.3.1
+Summary: Fast implicit solvers (CPU) for cable models, written in C++.
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: C++
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Operating System :: MacOS
+Classifier: Operating System :: Microsoft :: Windows
+Classifier: Topic :: Scientific/Engineering
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: torch>=2.12.0
+
+# dendra-solvers
+
+C++ PyTorch extensions for fast, differentiable implicit solvers in Dendra.
+
+## Installation
+
+```bash
+pip install dendra-solvers
+```
+
+Python 3.10+ and PyTorch 2.12+ are required. When a compatible wheel is
+available, installation does not require a compiler. Stable-ABI wheels work
+across supported Python and PyTorch versions on the same operating system and
+CPU architecture.
+
+For an HPC environment or custom PyTorch build, install PyTorch first and build
+the extension against that installation:
+
+```bash
+python -m pip install "torch>=2.12.0" "setuptools>=70.1,<82" ninja wheel
+MAX_JOBS=1 python -m pip install --no-build-isolation --no-cache-dir \
+  --no-deps --no-binary=dendra-solvers dendra-solvers
+```
+
+The build uses the C++ standard selected by your installed PyTorch. Use a
+C++20-capable compiler for current PyTorch releases. Stable-ABI builds do not
+need rebuilding just because PyTorch is upgraded; see
+[installation and clean rebuilds](docs/installation.md) for wheel availability,
+source builds, and upgrading older installations.
+
+Useful build options:
+
+- `DENDRA_NATIVE=1` enables host-specific compiler optimizations. Do not use it
+  for redistributable builds.
+- `DENDRA_OMP=1` enables OpenMP when its runtime is available.
+- `MAX_JOBS=1` limits compilation concurrency on memory-constrained systems.
+
+## Testing
+
+For a development checkout:
+
+```bash
+python -m pip install "torch>=2.12.0" "setuptools>=70.1,<82" ninja wheel numpy pytest
+MAX_JOBS=1 python -m pip install -v --no-build-isolation --no-deps -e .
+python -c "import dendra_solvers._ext as ext; print(ext.__file__)"
+python -m pytest -q -ra
+```
+
+Solver contracts and PyTorch transform usage are documented in [`docs/`](docs/).
+
+See [Contributing](CONTRIBUTING.md) for development tools, commit conventions,
+and versioning.

dendra_solvers-0.3.1/README.md

@@ -0,0 +1,52 @@
+# dendra-solvers
+
+C++ PyTorch extensions for fast, differentiable implicit solvers in Dendra.
+
+## Installation
+
+```bash
+pip install dendra-solvers
+```
+
+Python 3.10+ and PyTorch 2.12+ are required. When a compatible wheel is
+available, installation does not require a compiler. Stable-ABI wheels work
+across supported Python and PyTorch versions on the same operating system and
+CPU architecture.
+
+For an HPC environment or custom PyTorch build, install PyTorch first and build
+the extension against that installation:
+
+```bash
+python -m pip install "torch>=2.12.0" "setuptools>=70.1,<82" ninja wheel
+MAX_JOBS=1 python -m pip install --no-build-isolation --no-cache-dir \
+  --no-deps --no-binary=dendra-solvers dendra-solvers
+```
+
+The build uses the C++ standard selected by your installed PyTorch. Use a
+C++20-capable compiler for current PyTorch releases. Stable-ABI builds do not
+need rebuilding just because PyTorch is upgraded; see
+[installation and clean rebuilds](docs/installation.md) for wheel availability,
+source builds, and upgrading older installations.
+
+Useful build options:
+
+- `DENDRA_NATIVE=1` enables host-specific compiler optimizations. Do not use it
+  for redistributable builds.
+- `DENDRA_OMP=1` enables OpenMP when its runtime is available.
+- `MAX_JOBS=1` limits compilation concurrency on memory-constrained systems.
+
+## Testing
+
+For a development checkout:
+
+```bash
+python -m pip install "torch>=2.12.0" "setuptools>=70.1,<82" ninja wheel numpy pytest
+MAX_JOBS=1 python -m pip install -v --no-build-isolation --no-deps -e .
+python -c "import dendra_solvers._ext as ext; print(ext.__file__)"
+python -m pytest -q -ra
+```
+
+Solver contracts and PyTorch transform usage are documented in [`docs/`](docs/).
+
+See [Contributing](CONTRIBUTING.md) for development tools, commit conventions,
+and versioning.

dendra_solvers-0.3.1/pyproject.toml

@@ -0,0 +1,52 @@
+[build-system]
+requires = ["setuptools>=70.1,<82", "ninja", "torch>=2.12.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "dendra-solvers"
+version = "0.3.1"
+description = "Fast implicit solvers (CPU) for cable models, written in C++."
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = ["torch>=2.12.0"]
+classifiers = [
+  "Development Status :: 3 - Alpha",
+  "Intended Audience :: Science/Research",
+  "Intended Audience :: Developers",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3 :: Only",
+  "Programming Language :: Python :: 3.10",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Programming Language :: Python :: 3.13",
+  "Programming Language :: Python :: 3.14",
+  "Programming Language :: C++",
+  "Operating System :: POSIX :: Linux",
+  "Operating System :: MacOS",
+  "Operating System :: Microsoft :: Windows",
+  "Topic :: Scientific/Engineering",
+]
+
+[dependency-groups]
+# Install development tools without building the native extension.
+dev = ["commitizen==4.16.3", "pytest>=8"]
+
+[tool.commitizen]
+name = "cz_conventional_commits"
+version_provider = "pep621"
+version_scheme = "pep440"
+tag_format = "v$version"
+annotated_tag = true
+update_changelog_on_bump = true
+changelog_file = "CHANGELOG.md"
+bump_message = "build(release): bump version to $new_version"
+major_version_zero = true
+
+[tool.setuptools]
+package-dir = {"" = "src"}
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]

dendra_solvers-0.3.1/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

dendra_solvers-0.3.1/setup.py

@@ -0,0 +1,178 @@
+# setup.py
+import os
+import platform
+import sys
+import subprocess
+
+from setuptools import setup
+
+try:
+    from torch.utils.cpp_extension import BuildExtension, CppExtension
+except Exception as e:
+    raise RuntimeError(
+        "dendra-solvers needs PyTorch to build C++ extensions. "
+        "Install torch first (matching the version you want to use) and "
+        "re-run `pip install dendra-solvers --no-build-isolation`."
+    ) from e
+
+# -----------------------------------------------------------------------------
+# Sources
+# -----------------------------------------------------------------------------
+sources = [
+    "src/dendra_solvers/_thomas_cpu_bt.cpp",
+    "src/dendra_solvers/_thomas_cpu_bt_spd_fwd.cpp",              
+    "src/dendra_solvers/_thomas_cpu_bt_spd_solve_with_chol.cpp", 
+    "src/dendra_solvers/_thomas_cpu_tridiag_spd.cpp",
+    "src/dendra_solvers/_thomas_cpu_t.cpp",
+    "src/dendra_solvers/_pcr_cpu.cpp",
+    "src/dendra_solvers/_dhs_cpu.cpp",
+    "src/dendra_solvers/_dhs_cpu_bt.cpp",
+    "src/dendra_solvers/_dhs_cpu_multi.cpp",
+    "src/dendra_solvers/bindings.cpp",
+]
+
+# Base compile/link args
+#
+# This project uses a PyTorch C++ extension. On Windows the compiler is MSVC
+# (cl.exe), which does *not* understand GCC/Clang-only flags (e.g. -Wno-...,
+# -ffast-math). Passing those flags causes hard build failures.
+#
+# We therefore select MSVC-compatible flags on Windows and keep the existing
+# GCC/Clang flags on other platforms.
+# Leave the C++ language standard to PyTorch's BuildExtension, which supplies
+# the minimum required by the installed torch headers. An explicit standard
+# here overrides that default and can break builds against newer PyTorch.
+
+is_windows = (os.name == "nt")
+
+if is_windows:
+    # MSVC (cl.exe)
+    # NOTE: distutils/BuildExtension typically add /O2 for Release builds; we
+    # include it here for clarity. /fp:fast approximates -ffast-math.
+    base_cxx_args = ["/O2", "/fp:fast"]
+    # If a Visual Studio Developer Command Prompt (or vcvars*.bat) has already
+    # activated the VC environment, PyTorch expects DISTUTILS_USE_SDK=1 so distutils
+    # doesn't try to "activate" MSVC again. PyTorch raises this as a UserWarning
+    # exception and aborts the build if missing.
+    vc_env_active = any(k in os.environ for k in ("VSCMD_VER", "VisualStudioVersion", "VSINSTALLDIR"))
+    if vc_env_active:
+        os.environ.setdefault("DISTUTILS_USE_SDK", "1")
+        # Some distutils toolchains also consult MSSdk; harmless if unused.
+        os.environ.setdefault("MSSdk", "1")
+else:
+    # GCC / Clang
+    base_cxx_args = [
+        "-O3",
+        "-ffast-math",
+        "-fno-math-errno",
+        "-fno-trapping-math",
+        "-fstrict-aliasing",
+        "-Wno-nan-infinity-disabled",
+    ]
+
+
+is_darwin = (sys.platform == "darwin")
+is_arm64  = (platform.machine() == "arm64")
+use_native = os.environ.get("DENDRA_NATIVE", "0") == "1"
+
+if is_darwin and is_arm64:
+    # Clang on arm64 ignores -march; use -mcpu (enables M1/M2/M3 specific pipes, FMA, etc.)
+    base_cxx_args += ["-ffp-contract=fast"]
+    if use_native:
+        base_cxx_args += ["-mcpu=native"]
+    # Keep the runtime target consistent to avoid conservative codegen / linker warnings.
+    os.environ.setdefault("MACOSX_DEPLOYMENT_TARGET", "14.0")
+elif use_native and os.name != "nt":
+    # Enable host-tuned code on opt-in builds; default wheels stay portable.
+    base_cxx_args += ["-march=native"]
+
+extra_compile_args = {"cxx": list(base_cxx_args)}
+extra_link_args = []
+include_dirs = []
+library_dirs = []
+define_macros = [
+    # Only expose the stable APIs available in our minimum supported releases.
+    ("TORCH_TARGET_VERSION", "0x020c000000000000"),
+    ("Py_LIMITED_API", "0x030A0000"),
+]
+
+# ----------------------------------------------------------------------------
+# MSVC compatibility
+# ----------------------------------------------------------------------------
+# The solver kernels use GCC/Clang's `__restrict__` qualifier. MSVC uses
+# `__restrict` instead, and will otherwise parse it as an identifier, leading
+# to confusing syntax errors (e.g., 'missing ")" before identifier D').
+if is_windows:
+    define_macros.append(("__restrict__", "__restrict"))
+
+# -----------------------------------------------------------------------------
+# Optional OpenMP (enable with: DENDRA_OMP=1)
+# -----------------------------------------------------------------------------
+use_openmp = os.environ.get("DENDRA_OMP", "0") == "1"
+
+def _brew_prefix(pkg, default):
+    try:
+        return subprocess.check_output(["brew", "--prefix", pkg], text=True).strip()
+    except Exception:
+        return default
+
+if use_openmp:
+    if sys.platform == "darwin":
+        libomp_prefix = _brew_prefix("libomp", "/opt/homebrew/opt/libomp")
+
+        # Only libomp, not brew llvm (avoids libc++ header/runtime mismatches)
+        include_dirs += [f"{libomp_prefix}/include"]
+        library_dirs += [f"{libomp_prefix}/lib"]
+
+        # If using Apple clang, use -Xpreprocessor -fopenmp
+        extra_compile_args["cxx"] += ["-Xpreprocessor", "-fopenmp"]
+        extra_link_args += ["-lomp"]
+
+        # Ensure the runtime can find libomp
+        extra_link_args += [f"-Wl,-rpath,{libomp_prefix}/lib"]
+
+        define_macros.append(("DENDRA_USE_OMP", "1"))
+
+    elif os.name == "nt":
+        # MSVC (Visual Studio). If you use LLVM toolchain on Windows you can switch this.
+        # Newer MSVC supports /openmp:llvm; classic is /openmp (no tasking features).
+        extra_compile_args["cxx"] += ["/openmp"]
+        define_macros.append(("DENDRA_USE_OMP", "1"))
+    else:
+        # Linux / other Unix with GCC/Clang that support OpenMP out of the box
+        extra_compile_args["cxx"] += ["-fopenmp"]
+        extra_link_args += ["-fopenmp"]
+        define_macros.append(("DENDRA_USE_OMP", "1"))
+
+# -----------------------------------------------------------------------------
+# Extension
+# -----------------------------------------------------------------------------
+ext_modules = [
+    CppExtension(
+        name="dendra_solvers._ext",
+        sources=sources,
+        depends=[
+            "src/dendra_solvers/stable_utils.h",
+            "src/dendra_solvers/parallel_utils.h",
+            "src/dendra_solvers/dhs_validation.h",
+        ],
+        extra_compile_args=extra_compile_args,
+        extra_link_args=extra_link_args,
+        include_dirs=include_dirs,
+        library_dirs=library_dirs,
+        define_macros=define_macros,
+        language="c++",
+        py_limited_api=True,
+    )
+]
+
+# -----------------------------------------------------------------------------
+# Setup
+# -----------------------------------------------------------------------------
+# Package metadata lives in pyproject.toml. Keeping setup.py focused on the
+# PyTorch extension avoids duplicate/overwritten metadata during PEP 517 builds.
+setup(
+    ext_modules=ext_modules,
+    cmdclass={"build_ext": BuildExtension},
+    options={"bdist_wheel": {"py_limited_api": "cp310"}},
+)