Skip to content

plan.md

Latest commit

 

History

History
27 lines (21 loc) · 865 Bytes

plan.md

File metadata and controls

27 lines (21 loc) · 865 Bytes

Experiment 11: Coalesced vs Strided Access

1. Focus

  • Measure the direct cost of breaking contiguous lane access into wider strides.

2. Question

  • How quickly does throughput fall as address stride increases and coalescing quality drops?

3. Variants

  • stride_1
  • stride_2
  • stride_4
  • stride_8
  • stride_16

4. Method

  • Keep arithmetic, output count, and useful bytes fixed while changing only the source index stride.
  • Run the sweep at sizes that are large enough to stay in the bandwidth-bound region.

5. Outputs

  • Median GPU time by stride.
  • Throughput and effective GB/s by stride.
  • Slowdown curve relative to stride_1.

6. Interpretation

  • Coalescing is a first-order performance rule on bandwidth-bound kernels.
  • The resulting curve is a concrete demonstration of why strided access wastes transactions and cache lines.