Re: tdf#109097 - mdds accelerating random lookups

> On August 28, 2017 at 7:26 AM Michael Meeks <michael.meeks@collabora.com> wrote:

>       Not entirely convinced here - -but- I expect that dropping one
> indirection eg.
>
>       typedef std::vector<block> blocks_type;
>
>       rather than that extra indirection should significantly improve cache
> locality for iteration.

Funny you should mention that, because I was thinking of exactly the same thing after I posted my 
reply earlier.

>       Kohei - does having that extra indirection help with the implementation
> of iterators ?

No, not really.  I believe I initially did that to make the block movement operation as cheap as 
just moving the pointer value.  I was especially thinking of moving blocks from one container to 
another.  But since that struct is so small, making it a non-pointer and storing their states 
directly in std::vector will (as you say) improve the locality of whatever is stored therein 
without sacrificing the block swapping operations, and it should in theory give us better runtime 
performance in all the other operations.

> I assume that transferring the mp_data pointer as well as
> the size is much the same as transferring a block* ?

Yup.  You and I are in agreement here.  But if I can propose one thing, let's do that change 
separately (or maybe do that first before the other change) just to make things easier to track.

Kohei

--
Kohei Yoshida, LibreOffice Calc hacker