[poppler] [PATCH 4/4] SplashXPathScanner: Reduce complexity of sorting spans

[Adam Reichold]

Hello again,

Am 26.05.2018 um 16:22 schrieb Stefan Brüns:
> Unfortunately, it gives only a 100% speed increase (i.e. 30 minutes with 
> small_vector, 60 minutes with std::vector), so probably I should stop here.

I am unsure about the level of irony involved, but I also do not fully
understand the numbers: In your previous mail you wrote

> Before:
> - User time (seconds): 2979.80
> - Maximum resident set size (kbytes): 51208
> 
> After:
> - User time (seconds): 1773.53
> - Maximum resident set size (kbytes): 45896

i.e. ca. 60 minutes without your changes (algorithm + allocations) and
ca. 30 minutes with them. Now you write that changing only algorithm but
not changing the allocation brings you back to 60 minutes? Does this
mean that basically all of the speed-up is due to the use of small_vec
instead vector?

If I misunderstood that and you meant the 100% w.r.t. changing algorithm
+ allocations, then producing a measurement where just vector instead of
small_vec is used would probably be interesting. And it would be another
reason to break out that part of the patch.

Regards,
Adam

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/poppler/attachments/20180526/0544c534/attachment.sig>