Puzzle 10: Dot Product

Overview

Implement a kernel that computes the dot-product of vector a and vector b and stores it in output (single number).

Note: You have 1 thread per position. You only need 2 global reads per thread and 1 global write per thread block.

Implementation approaches

🔰 Raw memory approach

Learn how to implement the reduction with manual memory management and synchronization.

📐 LayoutTensor Version

Use LayoutTensor’s features for efficient reduction and shared memory management.

💡 Note: See how LayoutTensor simplifies efficient memory access patterns.