add graph500 benchmark - #1754

Description

This PR adds the Graph500 benchmark application to Ramble (var/ramble/repos/builtin/applications/graph500/application.py).

Graph500 is the de-facto HPC benchmark for data-intensive, memory-bound computing and irregular graph traversal. It evaluates large-scale Breadth-First Search (BFS) and Single-Source Shortest Path (SSSP) on Kronecker power-law graphs, stressing memory bandwidth, latency, and all-to-all communication patterns. It also serves as an important proxy workload for Electronic Design Automation (EDA) and graph analytics.

Features & Capabilities

• Workloads Supported:
  • bfs: Standard Breadth-First Search reference kernel with customizable scale and edge_factor.
  • bfs_sssp: Combined BFS and Single-Source Shortest Path kernels.
  • Pre-configured scale workloads: scale_16 through scale_30 ($2^{16}$ to $2^{30}$ vertices).
• Package Manager Integration:
  • Full integration with Spack (graph500@3.0.0 +procs_not_power_of_two).
  • Enables arbitrary non-power-of-two scaling across non-standard core/rank topologies (e.g., 24, 48, 88, 96, 176 ranks).
• Comprehensive Figure of Merit (FOM) Extractors:
  • Traversal Throughput: Harmonic Mean TEPS (primary), Median TEPS, Min/Max TEPS, Quartiles, and Standard Deviation.
  • Timing: Graph generation time, CSR construction time, median BFS validation time, and SSSP traversal times.
  • Graph Validation: Number of BFS roots, traversed edges, and edge counts.

Verification & Testing

1. Execution Validation:
   • Tested on single-node and multi-node clusters (c3-standard-176 instances) using Intel OneAPI MPI (intel-oneapi-mpi@2021.17.2) and GCC 14/15 across Scales 21, 24, 26, and 27.
   • Verified accurate FOM extraction across all metrics with ramble analyze.