27 lines (21 loc) · 857 Bytes

Experiment 24: Stream Compaction

1. Focus

Measure the baseline compaction pipeline across several keep ratios.

2. Question

How does valid-element ratio change the cost of filtering and compact-writing a stream?

3. Variants

keep_5
keep_25
keep_50
keep_75
keep_95

4. Method

Run the same compaction pipeline on deterministic inputs with different valid ratios.
Keep output semantics fixed and validate counts and compacted values after each run.

5. Outputs

Median GPU time by keep ratio.
Effective compacted-elements throughput.
Stage-sensitive interpretation of where compaction cost shifts across sparsity regimes.

6. Interpretation

Stream compaction cost is shaped by both the scan path and the amount of output that survives.
This is the baseline for later subgroup-assisted compaction work.