Expand description
GPU kernel execution implementations for sparse matrix operations
This module provides comprehensive GPU kernel execution logic with optimized memory management and multi-backend support.
Structs§
- GpuKernel
Config - High-performance GPU kernel configuration
- GpuMemory
Manager - Advanced GPU memory management and optimization utilities with smart caching
- GpuMemory
Stats - GPU memory usage statistics for optimization decisions
- GpuPerformance
Profiler - Performance profiling and optimization utilities
- Operation
Metrics - Detailed metrics for a specific GPU operation
Enums§
- Access
Pattern - Memory access patterns for optimization
- Memory
Layout - Memory layout optimization strategies for GPU access patterns
- Memory
Strategy - Memory access strategies for optimal GPU performance
- Transfer
Priority - Priority levels for GPU memory transfers
Functions§
- calculate_
adaptive_ workgroup_ size - Adaptive GPU workgroup sizing based on matrix characteristics
- execute_
spmv_ kernel - Fallback implementations when GPU feature is not enabled
- execute_
symmetric_ spmv_ kernel - execute_
triangular_ solve_ kernel - optimize_
memory_ bandwidth - GPU memory bandwidth optimization utility
- prefetch_
matrix_ data - Advanced GPU memory prefetching for sparse matrix operations