Re: ALLOC_IN_POOL, Re: lynx-dev lynx2.8.5dev.11

[Leonid Pauzner]

2-Dec-2002 00:24 Bela Lubkin wrote:
> I wrote:

>> Thomas Dickey wrote:
>>
>> > 2002-11-11 (2.8.5dev.10)
>>
>> > * modify GridText.c to store lines, anchors, and forms in the same HText 
>> > memory
>> >   pool as styles.  This will optimize memory allocation/deallocation by 8Kb
>> >   units.  The down side:  lines in TRST mode will be stored twice.  Some
>> >   structs are made a bit more compact -LP

> BTW, regarding the effectiveness of this (or Leonid's other efficiency
> changes)...

The above changes gives no visible speedup except HText_free()
which became *significantly* faster, by factor >100.

The quadratic vs linear complexity changes shows roughly 2 times improvement
on my test file I wrote about.  Numbers of persents were from gprof flat
profile, just a guidline.

> I tested on the pages:

>   http://www.armory.com/cache/userinfo.html
>   http://www.armory.com/cache/userinfo-t.html

> which are two different versions of 450 lines of data with multiple
> links per line.  I tested rendering speed on the basic data files, then
> on versions where I had duplicated the 450 data lines an additional 7 or
> 15 times (8 or 16 total copies, 3600 or 7200 total lines).  I compared
> performance of an old Lynx 2.8.4dev.20 binary vs. today's 2.8.5dev.11.
> I also compared output, verifying that they were in fact identical.

> The results show that Leonid grossly underestimated the benefits of his
> changes:

>    File                   2.8.4dev.20     2.8.5dev.11
>                                 -dump            -dump
>                          -dump  -nolist   -dump  -nolist
>   userinfo450.html        0.10   0.08      0.08   0.06
>   userinfo450-t.html      0.10   0.11      0.09   0.10
>   userinfo3600.html       4.95   2.50      0.48   0.42
>   userinfo3600-t.html     5.30   4.56      0.54   2.50
>   userinfo7200.html      28.24  10.54      0.93   0.83
>   userinfo7200-t.html    27.93  20.25      1.08  10.29

> (CPU-seconds on 1GHz Pentium III)

> Leonid's changes make performance (on this data set) linear in the
> number of lines, vs. some sort of quadratic behavior.  30 times as fast,
> in the worst case here.  Bravo.

Thanks to gprof...

> Oddly, `lynx -dump -nolist` is now slower than `lynx -dump` in some
> cases...

>>Bela<

> ; To UNSUBSCRIBE: Send "unsubscribe lynx-dev" to address@hidden



; To UNSUBSCRIBE: Send "unsubscribe lynx-dev" to address@hidden