`batuta optimize`

Optimize transpiled Rust code using MoE (Mixture-of-Experts) backend selection and Cargo profile tuning (Phase 3).

Synopsis

batuta optimize [OPTIONS]

The optimize command analyzes your transpiled Rust code for compute-intensive patterns and recommends optimal backends (Scalar, SIMD, or GPU) using the 5x PCIe dispatch rule (Gregg & Hazelwood, 2011). It also configures Cargo release profiles based on the selected optimization level.

This is Phase 3 of the 5-phase transpilation pipeline. It requires Phase 2 (Transpilation) to be completed first.

Options

Option	Description
`--enable-gpu`	Enable GPU acceleration for large matrix operations
`--enable-simd`	Enable SIMD vectorization via Trueno
`--profile <PROFILE>`	Optimization profile: `fast`, `balanced` (default), `aggressive`
`--gpu-threshold <N>`	GPU dispatch threshold in matrix size (default: 500)
`-v, --verbose`	Enable verbose output
`-h, --help`	Print help

Optimization Profiles

Profile	`opt-level`	LTO	`codegen-units`	Use Case
Fast	2	off	16	Quick iteration during development
Balanced	3	thin	4	Default production builds
Aggressive	3	full	1	Maximum performance (slow compile)

What It Does

Scans for compute patterns in .rs files under the transpiled output directory:
- matmul/gemm/dot_product → High complexity (GPU candidate)
- .sum()/.fold()/reduce → Medium complexity (SIMD candidate)
- .iter().map()/.zip() → Low complexity (Scalar)
Runs MoE backend analysis using BackendSelector::select_with_moe() to recommend Scalar, SIMD, or GPU for each pattern found.
Applies Cargo profile by writing [profile.release] settings to the transpiled project’s Cargo.toml.

Examples

Default Optimization

$ batuta optimize

⚡ Optimizing code...

Optimization Settings:
  • Profile: Balanced
  • SIMD vectorization: disabled
  • GPU acceleration: disabled

Scanning for compute patterns in ./rust-output...
Found 3 optimization targets:
  src/model.rs: High (matmul) → GPU recommended
  src/loss.rs: Medium (reduce) → SIMD recommended
  src/utils.rs: Low (iter/map) → Scalar

Applied balanced profile to Cargo.toml

GPU + SIMD Enabled

$ batuta optimize --enable-gpu --enable-simd --profile aggressive

Quick Development Iteration

$ batuta optimize --profile fast

The Batuta Book