Skip to content

CHANGELOG.md

Latest commit

 

History

History
236 lines (198 loc) · 10.5 KB

CHANGELOG.md

File metadata and controls

236 lines (198 loc) · 10.5 KB

[v0.3.1] - 2026-04-05

CLA3P Module

New Features

  • LapackSyev Eigensolver: Symmetric/Hermitian eigenvalue decomposition on CPU
    • Added syev, heev, and geev LAPACK proxies
    • Implemented LapackSyev class for symmetric and Hermitian eigenproblems
    • Added examples and tutorial/module documentation for eigsol
  • VirtualStrided Support: Dense matrix diagonal can now be referenced via VirtualStrided class

Bug Fixes

  • Fixed multiplication dimension checks in mult_dim_check() for matrix-vector operations
  • Fixed update operators for both dense and sparse cases

Improvements

  • Header Include Restructuring: All internal include paths reorganized into dedicated subdirectories (algebra, bulk, checks, dense, error, generic, linsol, lra, perms, proxies, qr, sparse, support, svd, types, virtuals); all includes now use angle-bracket form
  • Renamed dns?D file family to dns?d for consistent naming conventions
  • Replaced cla3p::svdPolicy_t::Limited with Economy throughout the codebase
  • Replaced tabs with spaces in example source files

cuLite Module

New Features

  • Dense Object Fill: Added fill() member function and operator=(Scalar) for dense vectors and matrices
  • Extended cuSOLVER Support: Additional cuSOLVER operations exposed through the handler interface
  • SVD and GEEV Improvements:
    • Re-implemented GESVD fix for S-size bug on CUDA
    • Stabilized DefaultSVD and LapackGeev constructors
    • Added additional members to LapackGeev
    • Fixed complex SVD implementation
    • Added examples and tutorial/module documentation for geev and gesvd
  • New Utility Class: GridXD for GPU kernel launch configuration

Bug Fixes

  • Fixed complex SVD implementation for correctness
  • Fixed complex_t class arithmetic operators (+, -)
  • Fixed update operators for both dense and sparse cases

Improvements

  • Implementation Split: Moved CuBlasHandler, CuSolverHandler, CuSparseHandler, and cuSPARSE proxy template instantiations from header-only to .cpp files with explicit specializations, reducing compile times
  • Header Include Restructuring: All internal include paths reorganized into dedicated subdirectories; all includes now use angle-bracket form
  • Replaced tabs with spaces in example source files
  • Updated README with SVD support documentation

[v0.3.0] - 2026-03-26

CLA3P Module

New Features

  • Enhanced CSR Matrix Support: Pardiso now accepts CSR matrices in Full and Upper storage formats
  • Template-Only Checks System: Outer checks now use meta classes and template-only implementations for improved compile-time optimization

Improvements

  • Code Maintenance: General code cleanup and refactoring
    • Use type aliases throughout codebase
    • Removed obsolete comments
    • Enhanced virtual expression system maintenance
    • Added extra validation in outer product operations
    • Updated namespace usage in syntax
  • Examples: Tidied up example programs for better clarity
  • Documentation:
    • Fixed documentation regarding the "bind" parameter
    • Updated syntax documentation

Bug Fixes

  • Improved dimension checking using meta classes in mult_dim_check()
  • Enhanced property validation code

cuLite Module

New Features

  • Eigenvalue Decomposition (GEEV): Complete eigenvalue/eigenvector computation for general matrices
    • Added geev() support in CuSolverHandler
    • New eigenvalue solver directory and solver implementations
    • LapackGeev proxy functions for CPU-based eigenvalue decomposition
    • Dedicated cuSolver proxy file for eigenvalue operations
    • Support for both eigenvalues-only and full eigenpair computation
  • Virtual Expression System: Lazy evaluation for GPU matrices and vectors
    • Virtual row vector class for transposed views
    • Virtual conjugate operations
    • Virtual transpose and conjugate transpose
    • Virtual arithmetic operations (+, -, scaling)
    • Outer product virtual expressions
    • Constructor support for creating objects from virtual expressions
    • Transpose and conjugate-transpose members for sparse matrices
  • Functional Algebra Operations: High-level functional API for common linear algebra operations
    • Inner product (dot, dotc)
    • Vector addition and scaling (axpy)
    • Matrix-vector multiplication for symmetric/Hermitian matrices
    • Matrix-matrix multiplication for symmetric/Hermitian matrices
    • Outer product operations (ger, gerc, syr, her)
    • Rank-1 updates for symmetric/Hermitian matrices
    • Functional update operations for dense and sparse matrices
    • Diagonal matrix-vector multiplication for CSX matrices
  • Sparse Matrix Operations:
    • CSR to CSC conversion via cuSPARSE proxy
    • Custom CSX addition kernels for sparse matrix arithmetic
    • Host-device copy operations for CSR matrices
    • Inner and outer algebra operators for sparse matrices
  • cuBLAS Handler Enhancements: Added multiple new operations
    • copy() member for array copying
    • dotc() for conjugate dot product
    • axpy() for vector scaling and addition
    • ger(), gerc() for rank-1 updates
    • syr(), her() for symmetric/Hermitian rank-1 updates

Improvements

  • Documentation:
    • Added comprehensive Doxygen documentation for modules
    • Enhanced documentation for update operations
    • Added documentation for LapackGeev functions
    • Fixed various documentation references
  • Code Organization:
    • Extracted legacy cuSPARSE proxies to separate file for better maintainability
    • Improved visibility control for virtual expressions
    • Enhanced dns1D code structure
  • Utilities:
    • Added memSetZero() utility function for GPU memory initialization
    • Added complex number arithmetic operators (+, -)

Bug Fixes

  • Fixed compilation error in operator>>() for stream operations
  • Fixed CuSolver GEEV implementation issues
  • Added const casts to fix compilation issues in cuSPARSE proxies
  • Resolved virtual expression conflicts by reimplementing Virtual Expression system
  • Fixed argument handling in update operations

API Changes

  • Modified add() function argument lists for consistency
  • Enhanced operator support for update operations (+=, -=, *=)
  • Added multiplication operators for matrix-vector and matrix-matrix operations

[v0.2.0] - 2026-02-07

CLA3P Module

New Features

  • CSR (Compressed Sparse Row) Matrix Support (BETA): Added full CSR matrix class with comprehensive operations matching CSC functionality
    • CSR-specific bulk operations and math functions
    • Sparse proxy support for both MKL and Arm Performance Libraries
    • Introduced CSX base class to unify common CSR/CSC functionality
    • Enhanced sparse matrix conversion capabilities (COO to CSR/CSC)
  • Virtual Expression System: Enhanced lazy evaluation capabilities
    • New VirtualStrided class for strided memory access patterns
    • Improved virtual transpose, scale, and plus/minus operations
    • Better support for complex algebraic expressions with deferred evaluation
  • Copy and Move Semantics: Refined constructors and assignment operators for better memory management
  • Template-based Checks: Converted basic validation checks to pure templates for improved compile-time optimization and cuLite integration

Improvements

  • Documentation: Comprehensive updates across all modules and tutorials
  • Memory Allocators: Renamed allocator template functions (i_?alloc<>() → i_?alloc_t<>()) for consistency
  • Vector Notation: Updated examples and documentation to use lowercase for vector variables (standard mathematical convention)
  • Code Quality: Various refactoring and cleanup improvements
    • Extracted base classes from containers (XxContainerBase)
    • Improved const-correctness throughout the codebase
    • Enhanced hermitian coefficient handling

Bug Fixes

  • Fixed COO matrix to CSR conversion
  • Fixed compilation issues on macOS
  • Corrected various compiler warnings
  • Fixed RankModerator documentation
  • Corrected examples compilation issues

API Changes

  • Renamed internal allocator functions for better naming conventions
  • Enhanced sparse matrix algebra operators to support CSR matrices
  • Extended permutation matrix operations to work with new sparse formats

cuLite Module

New Features (Initial Release, Linux/Windows)

  • CUDA-Accelerated Dense Linear Algebra (BETA): Complete GPU-based dense matrix and vector operations
    • Dense vector and matrix classes with GPU memory management
    • Support for real and complex arithmetic (float, double, complex types)
    • Block operations and shallow copy capabilities
  • Dense Linear System Solvers: GPU-accelerated linear equation solving
    • LU decomposition solver with cuSOLVER backend
    • Automatic memory management between host and device
  • Sparse Matrix Support: GPU sparse matrix operations
    • CSR (Compressed Sparse Row) matrix class
    • CSC (Compressed Sparse Column) matrix class
    • Sparse matrix-vector and matrix-matrix multiplication via cuSPARSE
  • CUDA Library Integration: Comprehensive wrappers for NVIDIA libraries
    • cuBLAS handler and proxies for dense linear algebra operations (GEMV, GEMM, TRSV, etc.)
    • cuSOLVER handler and proxies for decompositions and linear solvers
    • cuSPARSE handler and proxies for sparse operations
  • Utility Classes and Infrastructure:
    • CUDA stream management for asynchronous operations
    • Scoped Pointer Mode (SPM) utility for cuBLAS and cuSPARSE pointer mode handling
    • Device memory allocators with automatic cleanup
    • Custom CUDA kernels for specialized operations
  • Matrix Operations: Comprehensive set of GPU operations
    • Matrix norms (one-norm, infinity-norm, Frobenius, max-norm)
    • Matrix conjugate and real/imaginary extractors
    • Scaling, addition, and multiplication operations
    • 2D CUDA kernel launcher for custom operations

Documentation & Examples

  • Complete tutorial and API documentation
  • Example programs demonstrating:
    • Dense vector and matrix creation
    • Data transfer between host and device
    • Linear system solving
  • CMake integration and build scripts
  • Comprehensive README with installation instructions

Development Infrastructure

  • Full CMake build system with CUDA support (nvcc compiler)
  • Example builder and executor scripts
  • Integration with CLA3P design patterns and allocators

Architecture

  • Consistent API design mirroring CLA3P module structure
  • Bulk data operations (dns1D, dns2D) for low-level GPU memory operations
  • Type traits and enums for type-safe GPU operations
  • Exception handling with CUDA-specific error reporting

[v0.1.0] - 2026-01-02

SimuliCore official launch.