cuda-writing

Optimize GPU kernel launch configurations by applying tier decision trees and best practices for device memory management. This Skill provides systematic kernel lifecycle management to improve CUDA performance and correctness for high-performance spatial workloads.

Manage device memory and synchronization patterns in CUDA by following structured guidance on memory tiers and stream management. This Skill outlines synchronization strategies and primitive selection to ensure correct and efficient GPU kernel execution.

Classify GPU operations by using tier decision trees to determine kernel lifecycle and stream overlap. This Skill guides decisions on GPU dispatch logic, helping you systematically categorize and execute tasks for optimal performance.

Yes, you can use this guidance for debugging and reviewing existing CUDA code. The Skill provides comprehensive instructions and best practices for modifying and reviewing GPU kernels to improve both performance and correctness.

You should reconsider specific CUDA synchronization strategies when they conflict with the tier classification of your GPU operation. This Skill helps you avoid incorrect primitive selection by mapping synchronization patterns to appropriate kernel lifecycles and stream management scenarios.