Enum CudaGraphOptimizationLevel

Namespace

DotCompute.Backends.CUDA.Types

Assembly

DotCompute.Backends.CUDA.dll

Defines the optimization levels available for CUDA graph operations. Higher levels provide more aggressive optimizations but may increase compilation time.

public enum CudaGraphOptimizationLevel

Fields

Aggressive = 3

Aggressive optimization with all available techniques applied. May significantly increase compilation time but provides maximum performance. Best for performance-critical paths where compilation cost can be amortized.

Balanced = 2

Balanced optimization providing good performance improvements without excessive compilation overhead. Includes kernel fusion, memory coalescing, and shared memory optimization. Recommended for most production scenarios.

Basic = 1

Basic optimizations including dead code elimination and simple kernel merging. Minimal impact on compilation time.

None = 0

No optimization applied. Graphs are executed as-is without modification. Best for debugging and development.