We could devise a simple enough benchmark to test the impact of alignment on element-by-element operations (e.g addition).
Take a DoubleArray and generate four F64Arrays with different alignments (starting at element 0, 1, 2, 3). Then benchmark adding each of them to another, fixed F64Array. We should see a spike in throughput where alignments coincide.
(This, however, would only be valid if the garbage collector doesn't move the array around between benchmark iterations and shift the alignment.)
We could devise a simple enough benchmark to test the impact of alignment on element-by-element operations (e.g addition).
Take a
DoubleArrayand generate fourF64Arrays with different alignments (starting at element 0, 1, 2, 3). Then benchmark adding each of them to another, fixedF64Array. We should see a spike in throughput where alignments coincide.(This, however, would only be valid if the garbage collector doesn't move the array around between benchmark iterations and shift the alignment.)