simd-bench

name	simd-bench
description	Benchmark SIMD vs scalar paths in dashtable probing, vector distance, and protocol parsing. Args: dashtable|distance|parse|all.

SIMD performance verification suite for moon.

Usage

• /simd-bench dashtable — SSE2 group matching vs scalar fallback
• /simd-bench distance — SIMD distance kernels vs scalar
• /simd-bench parse — RESP protocol parsing with SIMD
• /simd-bench all — full SIMD benchmark suite

Steps

dashtable

1. Run existing Criterion benchmarks that exercise dashtable:

   RUSTFLAGS="-C target-cpu=native" cargo bench --bench get_hotpath -- "3_dashtable"
   

2. Check if AVX2 is available:

   sysctl -a 2>/dev/null | grep avx2 || lscpu 2>/dev/null | grep avx2
   

3. Compare SSE2 baseline vs scalar (force scalar via cfg):

   # Baseline (SIMD enabled)
   cargo bench --bench get_hotpath -- "3_dashtable" --save-baseline simd
   # Report improvement factor
   

distance (for vector search)

1. Benchmark L2/cosine/dot with dimensions 128/256/512/1024:

   RUSTFLAGS="-C target-cpu=native" cargo bench --bench vector_distance 2>/dev/null || echo "No vector_distance bench yet — skip"
   

parse

1. Benchmark RESP parsing:

   RUSTFLAGS="-C target-cpu=native" cargo bench --bench resp_parsing
   

Expected Results

• SIMD dashtable probing: ≥2x faster than scalar for group match operations
• SSE2 is baseline x86_64 — always available
• AVX2 (if detected): potentially 2x over SSE2 for wider group matching