Skip to content

OpenPFC release v0.1.4 (2025-12-18)

Latest
Latest

Choose a tag to compare

View all tags
@ahojukka5 ahojukka5 released this 18 Dec 10:28
· 87 commits to master since this release
v0.1.4
8891a6a

This major release brings GPU acceleration, modern visualization output, improved configuration formats, and a complete CI/CD pipeline — transforming OpenPFC into a production-ready, GPU-enabled phase field simulation framework with comprehensive testing and developer tooling.

Added: Complete CUDA/GPU support with DataBuffer memory abstraction, VTK parallel output for visualization, TOML configuration format, runtime FFT backend selection, parameter validation system, modular CMake architecture, comprehensive CI/CD workflows, SparseVector with MPI exchange, and type-safe World API. 257 commits including 10 new API examples, integration tests, benchmarks, and research tools.

Changed: Modular code organization (CMake, Tungsten, UI, World split into focused components), single-invocation test model, optional benchmark builds, enhanced compiler warnings, and updated dependencies (nixpkgs 24.05, tomlplusplus integration).

Fixed: Build system improvements (CMake warnings, test discovery, CI reliability), memory safety (initialization, template instantiation), MPI synchronization, code quality (shadowing, const correctness, clang-format), and documentation (SPDX headers, Doxygen configuration).

Deprecated: Old world::create(Int3, Real3, Real3) in favor of type-safe strong types API.

Breaking changes: None — all deprecated APIs remain functional.

Added

  • GPU/CUDA Support: Complete CUDA implementation enabling GPU-accelerated PFC simulations.
    Added DataBuffer for backend-agnostic memory management with CPU/GPU memory traits,
    CUDA FFT integration via HeFFTe, GPU kernels for element-wise operations, and GPUVector
    RAII container. Implemented full Tungsten model on GPU with optimized kernel launches and
    CPU-GPU synchronization for FieldModifiers and VTK output. Runtime backend selection API
    allows choosing between CPU and CUDA FFT backends via configuration. Comprehensive test
    coverage includes GPU device detection, memory allocation, FFT operations, and CPU vs CUDA
    result comparison. Build system supports optional OpenPFC_ENABLE_CUDA flag.
  • VTK Output: New VTK ImageData writer in include/openpfc/results/vtk_writer.hpp and
    src/results/vtk_writer.cpp for parallel visualization output. Generates .vti files
    for each rank and .pvti parallel metadata files for ParaView/VisIt. Includes comprehensive
    test suite with MPI-aware tests and single-invocation test model to prevent cleanup races.
  • TOML Configuration: Added TOML config file support alongside JSON. New
    feat(utils): Add TOML to JSON conversion utility enables .toml input files with
    automatic conversion. Integrated tomlplusplus library via CMake find module. All example
    configurations converted to TOML format. Unit tests validate conversion accuracy.
  • Modular CMake Architecture: Refactored monolithic CMakeLists.txt into 12 focused modules
    in cmake/ directory: ProjectSetup, CompilerSettings, CudaSupport, Dependencies,
    LibraryConfiguration, BuildOptions, CodeCoverage, Installation, PackageConfig, BuildSummary.
    Improves maintainability and reusability. Documented in cmake/README.md.
  • CI/CD Pipelines: Comprehensive GitHub Actions workflows for build matrix (GCC/Clang,
    multiple OS), documentation deployment, code coverage analysis with Codecov integration,
    and REUSE license compliance. Status badges added to README. Documentation includes
    workflow descriptions and troubleshooting guides.
  • Parameter Validation System: New UI subsystem for configuration validation with
    ParameterMetadata, ParameterValidator, and ValidationResult classes. Supports nested
    path validation, finite checks, type validation, and helpful error messages. Integrated
    into Tungsten app with comprehensive test coverage (300+ assertions).
  • FFT Backend Selection: Runtime FFT backend selection API allowing users to choose
    between available HeFFTe backends (FFTW, MKL, cuFFT) via configuration. New
    examples/fft_backend_benchmark.cpp demonstrates performance comparison. Backend field
    added to config schema with parsing and validation.
  • SparseVector & MPI Exchange: New SparseVector container with halo exchange patterns
    for domain decomposition. Includes gather/scatter operations, neighbor exchange with MPI,
    and halo pattern creation utilities. Comprehensive test suite validates exchange correctness.
  • Testing Infrastructure: First integration test suite for diffusion model validating
    complete simulation pipeline against analytical solutions (4 test cases, 331 assertions).
    Added benchmark subdirectory with microbenchmarks for World coordinate operations.
    Comprehensive unit tests for UI validation (300+ assertions), VTK writer (MPI-aware),
    DataBuffer, GPUVector, and SparseVector. Switched to single-invocation test model to
    prevent MPI initialization issues. Test coverage improvements across all modules.
  • World API: Type-safe World construction using strong types from strong_types.hpp.
    Added new create(GridSize, PhysicalOrigin, GridSpacing) overload preventing parameter
    confusion at compile time. Old create(Int3, Real3, Real3) API deprecated. Zero overhead -
    strong types compile away completely. Updated all examples and helper functions. Test suite
    with 71 assertions covering type safety, zero overhead, and backward compatibility.
  • Documentation: Added 10 comprehensive API examples (World, FFT, Simulator, Time,
    Decomposition, ResultsWriter, FieldModifier, DiscreteField, Model, custom field initializer).
    Added CITATION.cff for standardized citations. Improved Doxygen configuration. README
    sections on configuration validation, FFT backend selection, and extending OpenPFC.
  • Research Tools: Added power consumption benchmarks for FFT operations (CPU and GPU),
    multi-GPU HeFFTe examples, and scalability testing applications for Tungsten model.

Changed

  • CMake Structure: Root project() moved to top-level CMakeLists.txt. Build options
    reorganized into logical modules. Test discovery switched to single-invocation model.
    Benchmark compilation now optional via OpenPFC_BUILD_BENCHMARKS.
  • Tungsten Structure: Split monolithic tungsten code into modular headers and separate
    JSON inputs into inputs_json/ subdirectory. Restructured JSON schema to nested format.
    Renamed 'origo' field to 'origin' for consistency.
  • UI Module: Split monolithic ui.hpp into modular components. Made plan_options
    optional in app config. Added error formatting utilities for better user messages.
  • World Module: Split world.hpp into modular headers. Added query helper examples.
    Updated coordinate benchmark documentation.
  • Test Organization: Split monolithic parameter validation tests. Serialize VTK writer
    tests to prevent cleanup races. Make MPI_Worker static to persist MPI per process.
    Normalize test commands under single-invocation model.
  • Build Warnings: Enabled additional compiler warnings for code quality in Debug builds.
    Added -Werror=format-security. Made GCC-specific warnings conditional. Format check
    warns instead of fails in Nix builds.
  • Dependencies: Updated nixpkgs from 23.11 to 24.05. Added git and tomlplusplus to
    Nix build dependencies. Integrated Catch2 test discovery.

Fixed

  • Build System: Fixed CMake warnings by moving project() to root. Fixed Catch2 test
    discovery and optional MPI suites. Made documentation comment posting optional in CI.
    Cleaned up clang-format artifacts before REUSE checks. Improved error reporting in Nix tests.
  • Test Fixes: Fixed narrowing conversions in sparse vector tests. Fixed GridSpacing
    initializers in FFT tests. Fixed syntax errors in world benchmark and CUDA tests. Added
    missing pfc namespace qualifiers. Suppressed unused variable/parameter warnings with
    [[maybe_unused]]. Fixed incorrectly converted world::create calls.
  • Application Fixes: Fixed missing set_fft() call in diffusion example causing runtime
    errors. Removed unused fields (verbose in Diffusion, m_first in Aluminum). Fixed array
    initialization in SeedFCC. Added MPI-aware main to tungsten CPU vs CUDA test.
  • MPI Fixes: Fixed MPI_Worker to be safe for test frameworks. Query current MPI
    rank/size when generating PVTI instead of using stale values. Synchronize ranks before
    cleanup in VTK writer test to prevent races.
  • Memory Safety: Initialize all params struct members in aluminum to prevent undefined
    behavior. Add explicit template instantiation for World constructor. Fix CPU FFT
    std::vector interface to call HeFFTe directly.
  • Code Quality: Removed redundant const qualifiers. Added missing override keywords.
    Fixed variable shadowing in multiple files. Removed variable shadowing in timing collection.
    Fixed clang-format violations across codebase.
  • CI/CD: Removed ubuntu-20.04 from test matrix. Removed Cachix binary cache step.
    Initialized git submodules in all workflows. Made clang-format check warning instead of
    error. Used forked clang-format-action with fail-on-error option.
  • Documentation: Removed internal tracking references from code. Added SPDX headers for
    REUSE compliance to all test READMEs. Fixed Doxygen file headers for better doc generation.

Deprecated

  • World API: Old world::create(Int3, Real3, Real3) deprecated in favor of type-safe
    create(GridSize, PhysicalOrigin, GridSpacing). Migration guide in documentation.

Breaking Changes

None - all deprecated APIs remain functional with warnings.

Assets 2
Loading