Practice: Parallel Sum Reduction (WebGPU)medium

Practice: Parallel Sum Reduction (WebGPU)

medium ML Systems & MLOps 15–20 min

NumPy allowed · plain Python earns 10 bonus points

New to this topic? MLOps Overview

Implement parallel binary tree sum reduction, the core reduction pattern executed by GPU workgroups and CUDA kernels.

Task

Implement parallel_sum_reduction(arr). Simulate hierarchical pairwise reduction: at stride s = 1, 2, 4..., add elements pair-by-pair until a single scalar sum remains. Return 0.0 if empty.

Function Signature

parallel_sum_reduction(arr: list[float]) -> float

Examples

Example 1: Power of two length array
Input: arr = [1, 2, 3, 4, 5, 6, 7, 8]
Output: 36

Example 2: Odd length array requiring boundary checks
Input: arr = [2.5, 3.5, 4]
Output: 10

Constraints

arr is a list of numbers
Simulates tree reduction stages
Return scalar float rounded to 6 decimal places

Python3Saved ✓

def parallel_sum_reduction(arr: list[float]) -> float:

"""Simulate parallel tree sum reduction in O(log n) stages.

Return total sum.

"""

# Your code here

pass

Case 1: Power of two length array

Input: arr = [1, 2, 3, 4, 5, 6, 7, 8]

Expected: 36

Case 2: Odd length array requiring boundary checks

Input: arr = [2.5, 3.5, 4]

Expected: 10