plan.md

Experiment 23: Histogram and Atomic Contention

1. Focus

• Measure how atomic-heavy histogram construction responds to input skew and privatization.

2. Question

• How much does shared privatization reduce contention relative to direct global atomics?

3. Variants

• global_atomics
• privatized_shared
• input distributions uniform, mixed_hotset, and hot_bin_90

4. Method

• Build the same histogram from deterministic inputs under several contention regimes.
• Compare direct global updates against a workgroup-private accumulation path with controlled flush behavior.

5. Outputs

• Median GPU time by variant and distribution.
• Speedup of privatized_shared vs global_atomics.
• Contention sensitivity across the input distributions.

6. Interpretation

• Histogram performance is dominated by contention shape, not just by the number of input elements.
• Privatization should be judged by both its best-case speedup and its cost on low-contention inputs.