lamina

Lamina serves as a high-performance compiler backend for the Lamina Intermediate Representation (IR), a statically-typed, SSA-based language designed to facilitate seamless communication between high-level languages and low-level backends such as LLVM and Cranelift.

Overview

Lamina IR exemplifies a mid-level intermediate representation characterized by:

• Human-readable structure promoting accessibility.
• Block-based control flow accompanied by explicit type allocations.
• Static Single Assignment (SSA) form applied to all variables, ensuring clarity in code.
• A strong type system supporting both primitive and composite types.
• An explicit memory model differentiating stack and heap allocations.
• Architecture designed for straightforward lowering to machine code.

Features

Core Features

• Type System: Supports both primitive and composite types.
• Memory Model: Efficient handling of stack and heap allocations.
• Control Flow: Includes basic blocks and branching mechanisms.
• SSA Representation: Maintains single assignment form for all variables.
• Function System: Ensures typed parameters and return values.

Advanced Capabilities

• Multi-Architecture Support: Directly generates code for x86_64 and AArch64 (ARM64).
• Native Code Generation: Produces machine code without relying on external backends.
• Comprehensive Testing: Features over 20 automated test cases for robustness.
• IRBuilder API: Provides a programmatic approach for constructing IR.
• Performance Benchmarks: Demonstrates competitive performance levels comparable to established systems languages.

Direct Code Generation

Lamina distinguishes itself by generating machine code without dependence on external backends like LLVM or Cranelift, allowing for:

• Easy DSL Creation: Since it can support custom codegen, it can be used to create a DSL Language
• Complete Control: An end-to-end approach to code generation and optimization.
• Reduced Dependencies: Simplified architecture with no external backend requirements.
• Customized Optimizations: Tailored optimizations specific to the IR.
• Faster Compilation: Direct assembly generation tailored for performance.
• Cross-Platform Compatibility: Supports macOS, Linux, and Windows environments.

Key Features Included

• Basic arithmetic operations.
• Function calls without recursion and dynamic control flows.
• Effective stack and heap memory management.
• Robust integration of print statements for debugging.

Building Compilers with Lamina

The IRBuilder API simplifies the compiler-building process, encompassing:

1. Parsing the source language into an Abstract Syntax Tree (AST).
2. Converting the AST into Lamina IR using the IRBuilder.
3. Leveraging Lamina's capabilities for optimization and machine code generation.

Example of creating a simple arithmetic function:

use lamina::ir::*;

fn create_add_function() -> Result<Module, Error> {
    let mut builder = IRBuilder::new();
    let module = builder.create_module("math_example");
    let params = vec![
        FunctionParameter { ty: Type::I32, name: "a".to_string() },
        FunctionParameter { ty: Type::I32, name: "b".to_string() },
    ];
    let function = builder.create_function("add", params, Type::I32);
    let entry_block = builder.create_block("entry");
    builder.set_current_block(entry_block);
    let param_a = builder.get_parameter(0);
    let param_b = builder.get_parameter(1);
    let result = builder.add_instruction(
        Instruction::BinaryOp {
            op: BinaryOperator::Add,
            ty: Type::I32,
            left: param_a,
            right: param_b,
        }
    );
    builder.add_return(result);
    Ok(module)
}

This function generates corresponding Lamina IR effortlessly.

Performance Benchmarks

Lamina achieves remarkable performance across computational tasks, demonstrating competitiveness in benchmarks like 256×256 2D matrix multiplication. The average performance metrics against other languages reveal:

Insights from Benchmarks

• Lamina performs competitively amongst systems languages.
• The memory footprint remains low, akin to languages like C and Rust.
• Consistency in performance metrics through rigorous statistical analyses.

Future Directions

Lamina is continually evolving with plans for enhanced optimizations, broader architecture support, and advanced debugging tools to solidify its place as a leading advanced compiler backend.