Re: wip-cse

[Ludovic Courtès]

Hi,

Andy Wingo <address@hidden> skribis:

> On Tue 24 Apr 2012 18:31, address@hidden (Ludovic Courtès) writes:
>
>> Andy Wingo <address@hidden> skribis:
>>
>>> I spent some time poking at vlists and ended up squeezing out a little
>>> bit of performance.
>>
>> Any figures?
>
> Well, last time I checked, with the patch compiling peval.scm was 2.10
> seconds, compared to 2.35 without.  Not tiny, not huge.

OK.  Would be nice to check with the micro-benchs in vlists.bm as well.

>> Did you try fiddling with ‘block-growth-factor’, as suggested earlier?
>
> No I did not.  Another thing one could fiddle with would be the block
> size for a new list head.  I feel like we need more metrics though: a
> distribution of sizes of blocks, numbers of blocks in a vlist, hash
> collisions per block, etc.  I'll merge this to master soon and we can
> check what impact the new hash function has on performance.

Nice.

Thanks,
Ludo’.