Re: [boost] Re: Cutting over to MPL - some progress.

From: Phil Richards (news_at_[hidden])
Date: 2004-01-27 16:53:02

On Fri, 23 Jan 2004 16:54:14 -0500, David Abrahams wrote:
[I wrote:]
>> I'll have a look at zip_view, but on first looks it doesn't exactly look
>> simple :-)
> Replacing lambda expressions with metafunction classes would be my first
> strategy.

Well, I replaced zip_view with transform and got the following
(non-scientific) results:

% time g++ -Wall -W -I. -I/home/pgr/src/boost -DWITH_ZIP_VIEW pq.cpp -c
real    0m59.201s
user    0m56.444s
sys     0m0.475s

[without zip_view, using transform]
% time g++ -Wall -W -I. -I/home/pgr/src/boost pq.cpp -c
real    0m14.074s
user    0m13.418s
sys     0m0.372s

Well, that's much better.  (Though there is probably still some room for
improvement yet - it's still 3 or 4 fold slower than without the MPL.)

How difficult would it be to produce a full family of mpl transform
functions?  Currently, there are the 1 and 2 sequence forms, but I could
use a 3 sequence version :-)  (The multi-sequence version is what I would
have called "zip_with" from my old F.P. days...)

Given the progress I've made, I will put together yet another release of
my library.

This version allows user choice of the base units in their dimensional
analysis (so if you want bananas/second you can have it).  It still uses
fixed-size vectors of powers, though - so if you have a banana-system and
you want to multiply a banana-system-based quantity by a "physical
system"-based quantity, it won't let you.  Personally, that's the way I
like it...

phil
-- 
change name before "@" to "phil" for email