hpc-strategies

name	hpc-strategies
description	Evolutionary flat-directory exploration strategy for HPC optimization. Use when designing directory hierarchies or planning multi-generation parallelization strategies.

HPC Optimization Strategies

Evolutionary Flat-Directory Design

Generations

1. Seed: Individual techniques (/OpenMP/, /MPI/, /CUDA/)
2. Crossover: Fuse promising results (/OpenMP_MPI/, /OpenMP_AVX2/)
3. Breeding: Combine best hybrids (/OpenMP_MPI_AVX2/)

Naming

• Parallelization strategies: _ separator in implementation order (OpenMP_MPI_AVX2)
• Metadata: - separator (CUDA-sharedMem)
• One worker per flat directory

Directory Structure

<hardware>/<compiler>/
├── OpenMP/       # Gen 1
├── MPI/          # Gen 1
├── CUDA/         # Gen 1
├── OpenMP_MPI/   # Gen 2
└── OpenMP_MPI_AVX2/  # Gen 3

Why Flat

Nested hierarchies cause duplicate implementations when combining A+B+C. Flat structure with visibility control (via PG_visible_dir.md) prevents this.

Details: references/evolutionary_flat_dir.md, references/typical_hpc_structure.md