🚀 go-pugleaf

"Atle Smelv�r" wrote:

> Refering to my post: "Help for optimization" :)

The thing it that you haven't really given us a proper problem yet.
Your code is fairly decent given the algorithm.  What we really would
like to do is check the algorithm and we need a compiling program with,
if not actual sample data, at least an idea of what sample data would be
like.


--
Bob Lee
High Performance Delphi - http://www.optimalcode.com/

Ok, here are the consts and types involved:

const
  AntallDagkoder = 365

type

TTurNode = record // 16 bytes
    NodeID : integer;
    idTid  : integer;
    maxTid : integer;
    Tider  : TTidNoder;
    Turer  : TKeyValArr;
  end;

TTurNoder = array of TTurNode;

TTidNoder = array of TTidNode;

TTidNode = record // 8 bytes
    StartTid  : word;      // Start on trip in format, minutes from
midnight, max 1439
    Tid       : word;        // Time used on the trip
    ShortHeap : integer; // Index pointer to a record in THeapNoder
  end;

THeapNode = record // size 64 bytes, left out the ones not used here
    Finished       : boolean;
    FraDeltadag    : shortint;
    DkoBufferPtr   : pDkoBuffer; // Pointer to DayCodeBuffer
  end;
THeapNoder = array of THeapNode;

  TDkoBuffer = array[0..AntallDagkoder-1] of boolean;

  pDkoBuffer = ^TDkoBuffer;

//NodeTider points to a record in TTurNoder, id points to a record in
TTurNode's TTidNoder,
// day holds active day, return value is the first valid time backwards that
runs at active day
// in most lucky cases this would be the same as id, else it travels back at
max 7 days (visibledager)
function TReiseMotor.ValidBackTime(NodeTider: TTurNode; id: integer; var
dag: integer): integer;
var
  i,
  nydag,
  maxdag : integer;
begin
  result := -1;
  maxdag := dag - VisibleDager;
  nydag := dag;
  if maxdag < 0 then maxdag := 0;
  i := id;
  while (nydag >= maxdag) do with fHeapPtr^[NodeTider.Tider[i].ShortHeap] do
  begin
    if not Finished and DkoBufferPtr^[nydag+FraDeltadag] then
    begin
      result := i;
      break;
    end;
    dec(i);
    if i < 0 then begin i := NodeTider.idTid; dec(nydag); end;
  end;
  if result >= 0 then dag := nydag;
end;

// This little proc is executed several million times from withing other
procs in the main algorithm.
// Assembly version is in my prev. post..

To test it,. copy these types etc. and push some values inside,. then
execute the proc. within a for loop,.
for i := 0 to 2000000 do ValidBackTime(....);

Hope this is enough for a testbed. Thanx for any suggestions and tips.

-Atle


"Robert Lee" <rhlee@optimalcode.com> wrote in message
news:3D6572A1.8192E172@optimalcode.com...
> "Atle Smelv�r" wrote:
>
> > Refering to my post: "Help for optimization" :)
>
> The thing it that you haven't really given us a proper problem yet.
> Your code is fairly decent given the algorithm.  What we really would
> like to do is check the algorithm and we need a compiling program with,
> if not actual sample data, at least an idea of what sample data would be
> like.
>
>
> --
> Bob Lee
> High Performance Delphi - http://www.optimalcode.com/
>
>

"Atle Smelv�r" wrote:
> Ok, here are the consts and types involved:

Your code still doesn't compile.

> To test it,. copy these types etc. and push some values inside

I'm afraid you have to write that code too.

> Hope this is enough for a testbed.

It is not, IMO. You expect people to do too much
work just trying to replicate your test-bench.

I did not read the problem, but I noticed the scandinavian language like
in "FraDeltaDag".
Being a dane, I can read the variables, but I guess you would get more
listeners if you used english variablenames like "FromDeltaDay".

Browsing the code, I would say that the best initial optimization has to
be done in pascal only.

Maybe fix the function from
  ValidBackTime(NodeTider: TTurNode; id: integer; var dag: integer):
integer;
to
  ValidBackTime(VAR NodeTider: TTurNode; id: integer; var dag: integer):
integer;
or
  ValidBackTime(CONST NodeTider: TTurNode; id: integer; var dag:
integer): integer;

as I guess you are using much time just copying data around.

--
Christen Fihl
http://HSPascal.Fihl.net/

Thanx for the tips. By fixing the parameter I got some more speed.

-Atle

In article <3d661237$1@newsgroups.borland.com>, Christen Fihl says...

> I did not read the problem, but I noticed the scandinavian language like
> in "FraDeltaDag".
> Being a dane, I can read the variables, but I guess you would get more
> listeners if you used english variablenames like "FromDeltaDay".

Oh, yes, indeed. I have never gotten used to localised variable names, 
not even in Dutch or German. I always think and write in English, when 
programming, although it is not my first or second language.
-- 
Rudy Velthuis (TeamB)

"If we knew what it was we were doing, it would not be called research,
 would it?" -- Albert Einstein

"Rudy Velthuis (TeamB)" <rvelthuis@gmx.de> wrote in message
news:MPG.17d0b2f77326567898df99@207.105.83.65...
> Oh, yes, indeed. I have never gotten used to localised variable names,
> not even in Dutch or German. I always think and write in English, when
> programming, although it is not my first or second language.

Same here.

Guido

Hi

    I volunteer but it's gonna cost you. It's cheap, I only charge 9 cents
per microsecond.

Tomaz

"Atle Smelv�r" <atle.smelvaer@broadpark.no> wrote in message
news:3d65671c$1@newsgroups.borland.com...
> Refering to my post: "Help for optimization" :)
>
> -Atle
>
>

"Tomaz Koritnik" <tomaz.koritnik1@guest.arnes.si> wrote in message
news:3d6e8ee9@newsgroups.borland.com...
>     I volunteer but it's gonna cost you. It's cheap, I only charge 9 cents
> per microsecond.

LOL!

Guido


--
Guido Gybels
http://www.optimalcode.com/Guido
(Disclaimer: all remarks made apply to Intel processors unless specified
otherwise)

"Atle Smelv�r" <atle.smelvaer@broadpark.no> wrote in message
news:3d70c7ab@newsgroups.borland.com...
> and it sure doesn't look like anyone here has the abillity to help me..

You are wrong there.

Guido


--
Guido Gybels
http://www.optimalcode.com/Guido
(Disclaimer: all remarks made apply to Intel processors unless specified
otherwise)

"Atle Smelv�r" wrote:
> it sure doesn't look like anyone here has the abillity to help me..

You haven't bothered to post a full compiling reference
implementation with test data yet, have you?