Trident Benchmarks

Per-function compiler overhead analysis: compiles real library modules from std/, vm/, and os/, then compares instruction counts and execution costs against hand-optimized TASM baselines.

Directory Structure

benches/
  end_to_end.rs              Criterion bench
  harnesses/                 Live execution programs (.tri + .inputs)
    std/compiler/lexer.tri
    std/compiler/lexer.inputs
    std/trinity/inference.tri
    ...
  references/                Rust ground truth
    std/crypto/poseidon2.rs
    std/nn/tensor.rs
    ...

baselines/triton/            Hand-optimized TASM (separate top-level dir)
  std/crypto/poseidon2.tasm
  vm/core/field.tasm
  os/neptune/kernel.tasm
  ...

• harnesses/ — .tri source + .inputs files for live execution benchmarks
• references/ — Rust programs that generate inputs and expected outputs (ground truth)
• baselines/triton/ — hand-optimized TASM baselines (top-level, outside benches/)

How It Works

1. trident bench scans baselines/triton/ for .tasm files
2. Each baseline maps to a source module by path: baselines/triton/std/crypto/auth.tasm -> std/crypto/auth.tri
3. The source module is compiled through the full pipeline (resolve, parse, typecheck, TIR, optimize, lower) without linking
4. Both compiled output and baseline are parsed into per-function instruction maps
5. Functions are matched by label name and instruction counts compared

Metrics

Running

trident bench                         # from project root
trident bench baselines/triton/       # explicit directory
trident bench baselines/triton/std/   # subdirectory

Works from any subdirectory -- walks up to find baselines/.

Adding a Baseline

1. Write the .tri module in std/, vm/, or os/ (real library code)
2. Create the matching baseline path: baselines/triton/std/crypto/newmod.tasm
3. Write hand-optimized TASM with __funcname: labels matching the module's public functions
4. Run trident bench to see the comparison

Baseline Format

// Hand-optimized TASM baseline: std.crypto.example

__function_name:
    instruction1
    instruction2
    return

__another_function:
    instruction1
    return

Rules:

• Labels use __funcname: format (matching compiler output)
• Comments (//) are not counted
• Labels (ending with :) are not counted
• halt is not counted
• Blank lines are not counted
• Everything else is counted