Module gpu 

GPU Storage Layer — Optional acceleration using CUDA.

Provides utilities to transfer Arrow RecordBatches to GPU memory and execute custom kernels.

Advanced Optimizations (007)

• CUDA Streams: Asynchronous data transfer and kernel execution overlap
• Multi-GPU: Distributed processing with NVLink support
• Histogram Optimization: Fast aggregation for small cardinality keys
• Unified Memory: Automatic memory management with prefetching
• Persistent Kernels: Reduced kernel launch overhead

Structs

GpuManager

Manager for GPU-accelerated operations.

Enums

GpuGroupByStrategy

Adaptive algorithm selection strategy

GpuHashStrategy

GPU Hash Strategy for GROUP BY operations

GpuReductionStrategy

Reduction strategy for SUM/COUNT operations