Re: [PATCH] Improve string benchtest timing

[Wilco Dijkstra <Wilco dot Dijkstra at arm dot com>]

Hi Siddhesh,

>> Reading the initial benchmark comment, it seems the original intention is
>> indeed to measure some sizes different than the multiple of START_SIZE
>> (my guess is to measure some tail handling on memcpy).  So either adjust
>> the comment accordingly to reflect what the benchmark is now measuring
>> or maybe add some sizes non multiple of START_SIZE.
>> 
>> Siddhesh, do you recall why did you added the 'misaligning' sizes?
>
> The misalignment was a simulation of a proprietary workload that was
> used to tune memcpy when I wrote it.  This proposed change skews the
> numbers in favour of aligned inputs, which is not really a bad thing.
> I'm OK with this bit.

Well the test doesn't actually test misaligned copies - both source and
destination are always mutually aligned, so any memcpy implementation which
aligns either source or destination will only do aligned copies.

In any case I'm not sure what the test is supposed to measure - the scores are
identical across all memcpy implementations. The time taken for double the
copy size is exactly twice as much.

Wilco