initial commit

Signed-off-by: Kamal Tufekcic <kamal@lo.sh>

benches/compare-flac.sh

@@ -0,0 +1,69 @@
+#!/usr/bin/env bash
+#
+# Wall-clock + compressed-size comparison between FLAC and LAC on the
+# local corpus directory. Diagnostic only — not part of CI.
+#
+# Usage:
+#   benches/compare-flac.sh [corpus_dir]
+#
+# Output columns (tab-separated, stable for piping into column -t):
+#
+#   file   flac_ms   flac_bytes   [matching LAC line from `cargo test`]
+#
+# For the LAC side, run `cargo test --test corpus --release -- --nocapture`
+# and correlate `lac_enc_ms` / `lac=<bytes>` values against the filenames
+# printed here. Two separate invocations because automating the join is
+# more fragile than eyeballing it for the six files the corpus contains.
+
+set -euo pipefail
+
+CORPUS_DIR="${1:-corpus}"
+
+if ! command -v flac > /dev/null 2>&1; then
+    echo "flac CLI not found in PATH; install the flac package" >&2
+    exit 1
+fi
+if [[ ! -d "$CORPUS_DIR" ]]; then
+    echo "corpus directory not found: $CORPUS_DIR" >&2
+    exit 1
+fi
+
+# Header. Columns cover both FLAC modes — default (`-5`, what most
+# production pipelines actually use) and `--best` (`-8`, the ceiling
+# `tests/corpus.rs` asserts its ratios against). `column -t` on the
+# output aligns to the printf format below.
+printf "%-50s\t%12s\t%14s\t%12s\t%14s\n" \
+    "file" "flac_d_ms" "flac_d_bytes" "flac_b_ms" "flac_b_bytes"
+
+# Shell globs are unordered across filesystems; sort for stable output.
+shopt -s nullglob
+files=("$CORPUS_DIR"/*.wav)
+IFS=$'\n' files=($(sort <<< "${files[*]}"))
+unset IFS
+
+# Warm-up invocation against the first file: the very first `flac` exec
+# in a shell session pays dynamic-linker + page-fault costs that aren't
+# representative of steady-state. Subsequent runs don't repay them.
+if [[ ${#files[@]} -gt 0 ]]; then
+    flac --stdout --best --silent "${files[0]}" > /dev/null 2>&1 || true
+fi
+
+for f in "${files[@]}"; do
+    # `date +%s%N` gives nanoseconds since epoch on GNU coreutils. Not
+    # portable to BSD `date`, but this script is Linux-only by design
+    # (matches the CI runner environment).
+    #
+    # Two invocations per file: default (`-5`) first, then `--best`.
+    # Ordering is deliberate: the default pass also warms the OS file
+    # cache, so `--best` sees warm-cache I/O and its time reflects the
+    # compute cost, not disk read.
+    start_ns=$(date +%s%N)
+    flac_d_bytes=$(flac --stdout --silent "$f" 2> /dev/null | wc -c)
+    mid_ns=$(date +%s%N)
+    flac_b_bytes=$(flac --stdout --best --silent "$f" 2> /dev/null | wc -c)
+    end_ns=$(date +%s%N)
+    flac_d_ms=$(( (mid_ns - start_ns) / 1000000 ))
+    flac_b_ms=$(( (end_ns - mid_ns) / 1000000 ))
+    printf "%-50s\t%12d\t%14d\t%12d\t%14d\n" \
+        "$(basename "$f")" "$flac_d_ms" "$flac_d_bytes" "$flac_b_ms" "$flac_b_bytes"
+done