Skip to content

Lock AZI_CHUNK default to 32 in rev13; add rev11↔rev13 validator and rev10↔rev13 benchmark#10

Open
nicklasorte wants to merge 1 commit intomainfrom
codex/create-revision-13-for-performance-benchmarking
Open

Lock AZI_CHUNK default to 32 in rev13; add rev11↔rev13 validator and rev10↔rev13 benchmark#10
nicklasorte wants to merge 1 commit intomainfrom
codex/create-revision-13-for-performance-benchmarking

Conversation

@nicklasorte
Copy link
Copy Markdown
Owner

Motivation

  • Lock in the measured real-input chunk-size win with a minimal, auditable revision derived from rev11 so runtime improvements are preserved without broader refactors.
  • Provide statistical validation and a compact real-input benchmark to ensure behavior is preserved and the speedup is measurable and repeatable.

Description

  • Added subchunk_agg_check_maxazi_rev13.m which is a copy of rev11 with only the default AZI_CHUNK_DEFAULT changed to 32 and an inline comment noting the choice came from a real-input chunk sweep benchmark, while preserving RNG strategy, helper calls, off-axis logic, and overall structure.
  • Added validate_subchunk_agg_check_maxazi_rev11_rev13_statistical.m which runs rev11 and rev13 on identical inputs, measures runtime, computes summary and tail statistics (mean, std, min, max, median, p90, p95, p99, and optional p99.9), prints a PASS/FAIL summary, returns a results struct, and fails-closed if configured thresholds (0.50 dB absolute or 5% relative) are exceeded.
  • Added benchmark_subchunk_agg_check_maxazi_rev10_rev13_real.m which runs rev10 and rev13 on identical inputs, measures runtime_rev10 and runtime_rev13, reports speedup = runtime_rev10 / runtime_rev13, prints a concise summary, and returns a results struct.
  • Intentionally did not modify rev10, rev11, or rev12, nor change RNG reseeding, helper-function behavior, or off-axis calculations.

Testing

  • Performed repository-level checks to confirm the three new files exist and inspected their contents with automated file listing and text-display commands, and confirmed they implement the required interfaces and logic.
  • Attempted to detect MATLAB/Octave runtime with octave --version, but the runtime was not available in the environment so no executions of the new MATLAB scripts were performed.
  • No numeric runtime or statistical tests were executed in this environment; the validation and benchmark scripts are in place to be run in a proper MATLAB/Octave environment and will print/return the requested metrics when executed.

Codex Task

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant