Enum OptimizationLevel

Namespace

Orleans.GpuBridge.Abstractions.Enums.Compilation

Assembly

Orleans.GpuBridge.Abstractions.dll

Defines the optimization levels available for GPU kernel compilation.

public enum OptimizationLevel

Fields

O0 = 0

No optimization applied during compilation.

Disables most compiler optimizations to preserve the original code structure as much as possible. This level provides:

• Fastest compilation time
• Best debugging experience with accurate line-by-line correspondence
• Predictable code generation
• Largest binary size and slowest runtime performance

Recommended for development, debugging, and educational purposes.

O1 = 1

Basic optimization level with fundamental optimizations applied.

Applies essential optimizations that provide significant performance improvements with minimal impact on compilation time. Typical optimizations include:

• Dead code elimination
• Constant folding and propagation
• Basic instruction scheduling
• Simple loop optimizations

This level maintains good debuggability while providing reasonable performance. Suitable for development builds where some optimization is desired.

O2 = 2

Standard optimization level suitable for most production scenarios.

Applies comprehensive optimizations that balance compilation time, binary size, and runtime performance. This is the recommended level for most production deployments. Optimizations typically include:

• All O1 optimizations
• Advanced loop optimizations (unrolling, vectorization)
• Function inlining
• Register allocation optimization
• Memory access pattern optimization
• Control flow optimizations

Provides excellent performance while maintaining reasonable compilation times and binary sizes.

O3 = 3

Aggressive optimization level for maximum runtime performance.

Applies the most aggressive optimizations available, prioritizing runtime performance over compilation time and binary size. May include:

• All O2 optimizations
• Aggressive function inlining
• Cross-procedural optimizations
• Advanced loop transformations
• Speculative optimizations
• Architecture-specific micro-optimizations

This level can significantly increase compilation time and binary size but provides the best runtime performance. Recommended for performance-critical kernels where maximum speed is essential.

Note: Some aggressive optimizations may make debugging difficult and could potentially affect numerical stability in edge cases.

Remarks

Optimization levels control the trade-off between compilation time and runtime performance. Higher optimization levels generally produce faster code but take longer to compile and may make debugging more difficult.

The specific optimizations applied at each level are backend-dependent, but generally follow these principles: - O0: Minimal optimization, best for debugging - O1: Basic optimizations with reasonable compilation time - O2: Standard optimizations suitable for most production use - O3: Aggressive optimizations, best runtime performance

Choose the optimization level based on your use case: - Development and debugging: Use O0 or O1 - Production with balanced performance: Use O2 - Performance-critical applications: Use O3