Auto-optimization suggestions

[AveryClapp]

Feature Description

AI-powered suggestions for code changes to improve cache behavior.

Example

Input Code:

for (int i = 0; i < n; i++) {
    for (int j = 0; j < m; j++) {
        result[i][j] = a[i][j] + b[i][j];
    }
}

Analysis:

• L1 hit rate: 65.3%
• Issue: Row-major access conflicts with column-major storage

Suggestion:

// Loop interchange for better spatial locality
for (int j = 0; j < m; j++) {
    for (int i = 0; i < n; i++) {
        result[i][j] = a[i][j] + b[i][j];
    }
}

Optimization Categories

1. Loop transformations
   • Interchange, tiling, unrolling
2. Data layout
   • AoS → SoA conversion
   • Struct padding for false sharing
3. Prefetch hints
   • __builtin_prefetch() insertion
4. Compiler hints
   • __restrict, __assume_aligned()

Implementation

• Pattern recognition for common anti-patterns
• AST-based code transformation
• Verify suggestions with re-analysis
• Show before/after metrics

Files

• backend/optimizer/ (new component)
• backend/optimizer/PatternMatcher.cpp
• backend/optimizer/CodeGenerator.cpp

Challenges

• Requires understanding of code semantics
• Need to preserve correctness
• May need LLM integration for complex cases