Add rev20 optimization with vectorized Pr helper and validation suite

[nicklasorte]

Summary

This PR introduces subchunk_agg_check_maxazi_rev20, a performance-optimized branch that replaces the Pr helper function with a vectorized implementation (monte_carlo_Pr_dBm_rev3_app), along with comprehensive profiling and statistical validation tools to ensure correctness against the golden rev11 baseline.

Key Changes

• New optimized function: subchunk_agg_check_maxazi_rev20.m

  • Semantic equivalent to rev19 but with improved Pr helper (rev2 → rev3)
  • Configurable azimuth chunking (default 128) for memory/throughput tuning
  • Precomputes off-axis gain matrix once instead of per-iteration
  • Pre-generates all MC random numbers upfront for better cache locality

• Vectorized Pr helper: monte_carlo_Pr_dBm_rev3_app.m

  • Replaces nearestpoint_app bracket search with discretize for speed
  • Fully vectorized per-TX interpolation (proven pattern from clutter_rev5)
  • RNG-free, maintains exact rev2 output contract and units
  • Handles edge cases (NaN, Inf, zero-span interpolation) with rev2 semantics

• Profiling tool: profile_subchunk_agg_check_maxazi_rev20_real.m

  • Direct timing comparison of rev20 vs rev11 golden baseline on identical inputs
  • Detailed function-level profiling with top-20 contributors by total time
  • Tracks material Pr helper performance drop and identifies new bottlenecks
  • Returns structured results with full profile tables for analysis

• Statistical validation: validate_subchunk_agg_check_maxazi_rev11_rev20_statistical.m

  • Fail-closed validation: rev20 must stay within thresholds of rev11
  • Separate core (mean, std, min, max, median) and tail (p95, p99) metrics
  • Configurable absolute and relative drift tolerances
  • Multiple timing runs to assess runtime stability

Implementation Details

• Azimuth chunking in rev20 processes simulation azimuths in configurable blocks to balance memory usage and vectorization efficiency
• Discretize-based bracket search in rev3 Pr helper eliminates two nearestpoint_app calls per TX per MC iteration
• Pre-generation of random numbers decouples RNG from inner loops, enabling better parallelization in future revisions
• Validation thresholds use max(absolute_dB, relative_fraction) to handle both small and large signal magnitudes

https://claude.ai/code/session_018i3rp6VALpD9dvwAoYQmMZ

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4a69d247d0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use rev11 EIRP helper when summarizing baseline profile

This baseline summary uses eirp_pattern selected from file existence, so when monte_carlo_super_bs_eirp_dist_rev8.m is present it matches rev8 in the rev11 profiler table even though subchunk_agg_check_maxazi_rev11 calls monte_carlo_super_bs_eirp_dist_rev5 (see subchunk_agg_check_maxazi_rev11.m line 80). In that common setup, base.monte_carlo_super_bs_eirp_dist_valid is recorded as not visible/zero, which silently skews baseline-vs-rev20 profiling outputs and can mislead performance conclusions.

Useful? React with 👍 / 👎.