User Guide

Comprehensive guides for using HPCSeries Core effectively.

• Architecture
  • HPCSeries Core v0.7 — Logical Architecture
  • Architectural Intent
  • High-Level Layer Stack
  • Fortran Kernel Subsystems
  • Parallel & Vectorized Execution
  • C++ Fast-Path Kernels
  • Stable C ABI Layer
  • Python Binding Layer (v0.7)
  • What HPCSeries Core IS (and Is Not)
  • Architectural Stability
  • Design Decisions
  • Zero-Copy Design
  • Thread Safety
  • See Also
• Performance Guide
  • Performance Characteristics
  • Calibration
  • OpenMP Configuration
  • Array Layout Optimization
  • Benchmarking
  • Performance Tips
  • Common Performance Issues
  • Performance Monitoring
  • Expected Performance
  • Advanced Optimization
  • Troubleshooting Performance
  • See Also
• Migration Guide
  • From NumPy
  • From Pandas
  • API Compatibility Matrix
  • Common Migration Patterns
  • Migration Checklist
  • Key Differences Summary
  • Performance Expectations
  • Migration Strategy
  • See Also

Architecture Overview

HPCSeries Core is a hybrid Fortran/C/C++ library with Python bindings:

Performance Stack:

┌─────────────────────────────────────┐
│  Python API (Zero-Copy)             │  ← User Interface
├─────────────────────────────────────┤
│  Cython Bindings                    │  ← NumPy Integration
├─────────────────────────────────────┤
│  C/C++ Orchestration Layer          │  ← SIMD Dispatch
├─────────────────────────────────────┤
│  Fortran HPC Kernels + C SIMD       │  ← Core Algorithms
├─────────────────────────────────────┤
│  SIMD Instructions (AVX2/AVX-512)   │  ← Hardware Acceleration
├─────────────────────────────────────┤
│  OpenMP Parallelization             │  ← Multi-threading
└─────────────────────────────────────┘

Key Design Principles

1. Zero-Copy: Direct NumPy array access without data copying

2. SIMD Vectorization: Automatic use of AVX2/AVX-512 when available

3. OpenMP Parallelization: Multi-threaded for large datasets

4. Adaptive Tuning: Auto-calibration for optimal performance

5. Robust Statistics: MAD-based methods resistant to outliers

Contents

• Architecture - Detailed architecture and design decisions

• Performance Guide - Performance tuning and benchmarking

• Migration Guide - Migrating from NumPy/Pandas