Add IntelliKit Metrix profiling script for Example 7 (GEMM all-scatter)

[Copilot]

Adds a standalone IntelliKit Metrix profiling wrapper for the GEMM all-scatter benchmark to collect GPU hardware metrics (HBM bandwidth, cache hit rates, compute throughput, arithmetic intensity).

Changes

• examples/07_gemm_all_scatter/profile_intellikit.py: New script that runs benchmark.py under IntelliKit Metrix via torchrun, exposes all benchmark parameters plus Metrix-specific options, and prints grouped metric results with min/max/avg statistics

Usage

# All metrics, 2 GPUs
python examples/07_gemm_all_scatter/profile_intellikit.py --num_ranks 2

# Memory preset, filter to GEMM kernels, save to JSON
python examples/07_gemm_all_scatter/profile_intellikit.py --num_ranks 2 \
    --profile memory --kernel gemm --output profile.json

# Custom metrics
python examples/07_gemm_all_scatter/profile_intellikit.py --num_ranks 2 \
    --metrics memory.hbm_bandwidth_utilization,compute.hbm_gflops,memory.l2_hit_rate

# Timing only (no hardware counter overhead)
python examples/07_gemm_all_scatter/profile_intellikit.py --num_ranks 2 --time_only

Output is grouped by metric category (COMPUTE, MEMORY) per kernel, with optional export to JSON, CSV, or plain text.

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.