🚀 go-pugleaf

Tony Werten wrote:

> Hi,
>
> does anybody know if there is a difference in code optimization by Delphi 5,
> 6 and 7 ?

There is no difference between 5 and 6 and it looks like there is no difference
between 7 and the previous/

> ie Do applications run faster on newer Delphi's ( maybe combined
> with newer operating systems )

OS is a separate issue, generally performance improves but there can be specific
areas where it can get worse.


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

> OS is a separate issue, generally performance improves but there can be specific
> areas where it can get worse.

That is an answer from "Yes, Minister" book :) - "yes, and no" :))

As far as my experience tells me - if performance went better - you bought more
powerful machine. If you did not upgrade your machine - perfomance goes (much, sometimes-
other times very much) worse. Very picturesque example - Adobe Photoshop: 6th version is much slower
than 5th (yeah, I know there is 7th version, but its requirements are above my machine, so I just
do not care)

Back to original question: there was not any smallest thing about compiler re-designment in the
D7 feature matrix (and that made me crazy), so I assumed that is the same Borland C++ 5.1 code generator. It was cool
in its time - back 6-7 years ago, but now? (I think you all know the answer ;) - Actually - :((((

--
Andrew Rybenkov.

create labels, logos, image lists
for your programs and websites with
  "Imagizer: Click and Get It",
  http://www.geocities.com/arybenkov/imagizer/

And unless we put massive pressure on Borlane, D8 won't have a better
compiler either...

Frank


"Robert Lee" <rhlee@optimalcode.com> wrote in message
news:3D6BAC8A.4589833D@optimalcode.com...
>
> There is no difference between 5 and 6 and it looks like there is no
difference
> between 7 and the previous/

> [...] It was cool in its time [...]

Yes, even not much improvement would be good, just to tweak 
the code output so that instead of compiling

	...
	FSTP [Somewhere]
	FWAIT
	FLD  [The same Somewhere]
	...

it would only compile to

	...
	FST [Somewhere]
	FWAIT
	...

And the same can happen for integers:

	MOV [Somewhere], EAX
	MOV EAX, [The same Somewhere]

How many times have I seen this... 
a memory stall for nothing.

Eric Grange
http://glscene.org

In article <3d6be8d6@newsgroups.borland.com>, Frank de Groot says...

> And unless we put massive pressure on Borlane, D8 won't have a better
> compiler either...

You don't know. That is pure speculation on your part.
-- 
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) scripsit:
>> And unless we put massive pressure on Borlane, D8 won't have a better
>> compiler either...
> You don't know. That is pure speculation on your part.

Taking into account that Borland is focusing on business applications, his
speculations seem realistic to me. In this respect, developing a new
compiler while having a stable one which still meets most requirements is
a waste of money. They always saw it that way, why should that change?

And as time goes by, this gets even more improbable, because optimizing
for Intel processors is not very farsighted. I think they will concentrate
on dot net, as they have already shown a preview dot net compiler.

Mab�

By the way, that does not mean that I support this policy. I wished coding 
seriously speed-related programs would be easier with Delphi. Using floating 
point calculations reduces performance so much sometimes that - in order to 
avoid the need for assembler - one has to do fixed point arithmetics if 
possible, which is slower compared to good FPU code (which e. g. uses FPU 
registers during a loop). Better optimization could even persuade some C++ 
hard-liners to use Object Pascal.

-- 
 "Zwei, eins, Risiko!" - Beruf Programmierer

In article <Xns927C1B31871C8membuede@news.mbue.de>, =?iso-8859-1?Q?
Matthias_B=FCchse?= says...

> > You don't know. That is pure speculation on your part.
> 
> Taking into account that Borland is focusing on business applications, his 
> speculations seem realistic to me.

Still it is pure speculation. Pure speculation should never be presented 
as fact. A simple "I think" or "IMO" is often enough to make that clear.
-- 
Rudy Velthuis (TeamB)

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

In article <Xns927C1B31871C8membuede@news.mbue.de>, =?iso-8859-1?Q?
Matthias_B=FCchse?= says...

> Taking into account that Borland is focusing on business applications, his 
> speculations seem realistic to me. Is this respect, developing a new compiler 
> while having a stable one

Why a *new* one? Why not improving the existent one?

I don't think it is realistic.
-- 
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) scripsit:
>> Taking into account that Borland is focusing on business applications,
>> his speculations seem realistic to me. Is this respect, developing a
>> new compiler while having a stable one
> Why a *new* one? Why not improving the existent one?

Andrew said that the compiler was between 6 and 7 years old.
If improving was it that easy, why haven't they done it yet?
So many features and new keywords have been added since then, so it is 
possible that a redesign is necessary.

Anyways, we (at least I) don't know.

Mab�

-- 
 "Zwei, eins, Risiko!" - Beruf Programmierer

In article <Xns927CECED023A9membuede@news.mbue.de>, =?iso-8859-1?Q?
Matthias_B=FCchse?= says...

> Andrew said that the compiler was between 6 and 7 years old.

When did D2 come out? He is probably right.

> If improving was it that easy, why haven't they done it yet?

Huh? They have constantly improved it, even if this was not easily 
visible at the surface.

> So many features and new keywords have been added since then, so it is 
> possible that a redesign is necessary.

Did you mean better optimisation, or general improvement?
-- 
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)" wrote:

> In article <Xns927C1B31871C8membuede@news.mbue.de>, =?iso-8859-1?Q?
> Matthias_B=FCchse?= says...
>
> > Taking into account that Borland is focusing on business applications, his
> > speculations seem realistic to me. Is this respect, developing a new compiler
> > while having a stable one
>
> Why a *new* one? Why not improving the existent one?

It would seem that Borland has been in "new" rather than "improved" mode for some
time (i.e. start over rather than "fix" the old).  This isn't necessarily bad, and
may even be resource efficient, but the lack of consideration for performance in
the recent past (D2 thru D7) makes the idea of a "new" compiler disconcerting.

In keeping with that, my prediction is that compiler optimization in the x86 32bit
compiler won't ever be changed.  That means that the next real chance for some
compiler improvements is the transition to 64 bit....assuming that Dot net type
interpreters don't take over the world.

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

In article <3D72920D.C184C3A6@optimalcode.com>, Robert Lee says...

> assuming that Dot net type interpreters don't take over the world.

I've been told that

  - .NET IL code is normally JITted, i.e. compiled to native 
    on the spot and then run, on a routine by routine base;
  - optimisation of the JITter is quite good.

Calling them interpreters is a bit beside the point.
-- 
Rudy Velthuis (TeamB)

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

: Robert Lee <rhlee@optimalcode.com> wrote:

>In keeping with that, my prediction is that compiler optimization in the x86 32bit
>compiler won't ever be changed.  That means that the next real chance for some
>compiler improvements is the transition to 64 bit....assuming that Dot net type
>interpreters don't take over the world.

You should read up on dynamic compilation a bit. HP (or whatever it is
called nowadays) has interesting papers on that and a research
implementation, as do various universities.

Given enough type information and enough strictness, it is possible to
really create fun things.

Rudy Velthuis (TeamB) (rvelthuis@gmx.de) scripsit:
>> If improving was it that easy, why haven't they done it yet?
> Huh? They have constantly improved it, even if this was not easily 
> visible at the surface.

This time, I meant optimization.

>> So many features and new keywords have been added since then, so it is 
>> possible that a redesign is necessary.
> Did you mean better optimisation, or general improvement?

The latter including the first.
Maybe some really new things could be added then.

By the way, is it correct that the Delphi 7 compiler does inline small 
functions?

Mab�

-- 
 "Zwei, eins, Risiko!" - Beruf Programmierer

> may even be resource efficient, but the lack of consideration for performance in
> the recent past (D2 thru D7) makes the idea of a "new" compiler disconcerting.

Robert, Imho, you may need to remove that "D7 hype" from your great site :)

--
Andrew Rybenkov.

create labels, logos, image lists
for your programs and websites with
  "Imagizer: Click and Get It",
  http://www.geocities.com/arybenkov/imagizer/

Stefan Hoffmeister wrote:

> Given enough type information and enough strictness, it is possible to
> really create fun things.

I don't disagree and I know that net isn't an interpreter in the traditional sense.
And there are "fun" things that can come out of it.  But that is "can" and not "will"
In short, the jury is still out.

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

Andrew Rybenkov wrote:

> > may even be resource efficient, but the lack of consideration for performance in
> > the recent past (D2 thru D7) makes the idea of a "new" compiler disconcerting.
>
> Robert, Imho, you may need to remove that "D7 hype" from your great site :)

Nope,  hype is a double-edged sword.

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

The reason they haven't touched the compiler for 7 years is the same as why
they haven't touched the editor for 7 years. "The guy who wrote it is gone
and nobody dares to touch the code".

Frank


"Matthias B�chse" <me@mbue.de> wrote in message
news:Xns927CECED023A9membuede@news.mbue.de...
>
> Andrew said that the compiler was between 6 and 7 years old.
> If improving was it that easy, why haven't they done it yet?

: "Frank de Groot" <nospam@nospam.com> wrote:

>The reason they haven't touched the compiler for 7 years is the same as why
>they haven't touched the editor for 7 years. "The guy who wrote it is gone
>and nobody dares to touch the code".

Frank, you don't know what you are talking about.

"Per Larsen" wrote:
> Here's an excerpt of some fairly typical code generated by the JIT'er:
> mov esi, eax (no, I'm
> mov ebx, esi   not
> mov esi, ebx   making
> mov eax, esi   this up)

That's nice! Probably a security measure to make sure
you don't loose the value - safer to keep it in four
registers :-P!

> future updates to the JIT'er, but such improvements
> will inevitably slow down it down thus making initial load times even
> longer.

Yes. I still don't get it why a standard installation of
a .Net application dont use ngen (or similar) to pre-
compile it at install time. They could add powerful
optimizations to that without incurring runtime costs.

I merely repeat what has been said by people who said they *did* know what
they were talking about
on these groups. I would not make this up, because I would love to stick
with Delphi, and things are getting worse, IMO there will never be 64-bit
CPU support in the Delphi compiler for example.
Borland thinks that 'enterprise customers' who make 'enterprise software' do
not need speed.
Arrogant.

Frank

"Stefan Hoffmeister" <no@email.here> wrote in message
news:73c6nusj194486lh5i20smhedcja3tdl9k@4ax.com...
>
> Frank, you don't know what you are talking about.

"Hallvard Vassbotn" <hallvard.vassbotn@nospamthanks-c2i.net> wrote in
message news:3d733740@newsgroups.borland.com...
> That's nice! Probably a security measure to make sure
> you don't loose the value - safer to keep it in four
> registers :-P!

LOL!

Guido

: "Frank de Groot" <nospam@nospam.com> wrote:

>I merely repeat what has been said by people who said they *did* know what
>they were talking about on these groups. 

In that case you might want to qualify your statements with "this and
that person said that" and simply repeat what "this and that person"
said, without interpreting it.

This way, "this and that person" gets all the blame for the
misinformation and it is much easier to identify who spreads rumors just
to hurt Delphi and Borland.

You might also want to figure out how exactly the Kylix compiler was
created. Out of thin air? Or by a Borland Delphi compiler R&D team?

What about the inline assembler rewrite in D6? Another creation by
OutOfTheBlue Productions? Or the result of intense R&D work?

> The reason they haven't touched the compiler for 7 years is the same as why
> they haven't touched the editor for 7 years. "The guy who wrote it is gone
> and nobody dares to touch the code".

They maybe have someone, _BUT_ (as I see it) they do not care about Delphi programs execution times.
Let's again look at this picture
http://homepages.borland.com/aohlsson/BorCon2002/a107-0800_IMGs.jpg
(note URL, btw)

 "new compiler architecture
 priority on optimization
 multiple hardware platforms"

So those are priorities for CPP (CBuilder). Because why? - because C++ folks are less that
VB/Delphi (original idea by Clarion) care about DB stuff (shit ;). So they need to get something they really care about.
But Delphi is positioning, exactly, for DB "OnClick and get it" folks (and I have arguments for this!),
and, thus, does not need "best of all" optimizer, nor appropriate compiler pragmas.
Moreover, as you know, they eliminated TP "inline" directive, nothing giving back instead of it.


--
Andrew Rybenkov.

create labels, logos, image lists
for your programs and websites with
  "Imagizer: Click and Get It",
  http://www.geocities.com/arybenkov/imagizer/

In article <3d73390a$1@newsgroups.borland.com>, Frank de Groot says...

> I merely repeat what has been said by people who said they *did* know what
> they were talking about on these groups.

I am sure Stefan knows much better than anyone else on this group.
-- 
Rudy Velthuis (TeamB)

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

> You might also want to figure out how exactly the Kylix compiler was
> created. Out of thin air? Or by a Borland Delphi compiler R&D team?
>
> What about the inline assembler rewrite in D6? Another creation by
> OutOfTheBlue Productions? Or the result of intense R&D work?

Let's, please, not be so picky about words (especially from our foreign mouthes :).
He (as many of us) meant "OPTIMIZER".

Kylix was created?? AFAIK Kylix is Delphi for Linux Intel '86. So what makes it different from
Windows' one? Library - yes, but compiler?

Inline asm - ehem, very simple task for Lex/Yacc or CocoR, no?

--
Andrew Rybenkov.

create labels, logos, image lists
for your programs and websites with
  "Imagizer: Click and Get It",
  http://www.geocities.com/arybenkov/imagizer/

> misinformation and it is much easier to identify who spreads rumors just
> to hurt Delphi and Borland.

Ignoring CPUs after 80386 Borland hurts to itself itself.

--
Andrew Rybenkov.

create labels, logos, image lists
for your programs and websites with
  "Imagizer: Click and Get It",
  http://www.geocities.com/arybenkov/imagizer/

In article <3d734bf5@newsgroups.borland.com>, Andrew Rybenkov says...

> Kylix was created?? AFAIK Kylix is Delphi for Linux Intel '86. So what makes it different from
> Windows' one? Library - yes, but compiler?

One example: the original compiler only generated PE executables, and 
was not capable of generating PIC.

> Inline asm - ehem, very simple task for Lex/Yacc or CocoR, no?

That depends on what you define as "very simple".
-- 
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.17dd71decb37abd898e26a@207.105.83.65...
> That depends on what you define as "very simple".

Indeed, and even though I also would love to see a new compiler that uses
all of the features on Pentium II and above to the fullest, building such a
compiler is by no means trivial.

Guido


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

"Guido Gybels" <guido.gybels@rnid.org.uk> schreef in bericht
news:3d7354a0$1@newsgroups.borland.com...
> "Rudy Velthuis (TeamB)" <rvelthuis@gmx.de> wrote in message
> news:MPG.17dd71decb37abd898e26a@207.105.83.65...
> > That depends on what you define as "very simple".
>
> Indeed, and even though I also would love to see a new compiler that uses
> all of the features on Pentium II and above to the fullest, building such
a
> compiler is by no means trivial.

In addition. Target processor based optimization and profiler based
optimization would be very nice too.

Tony

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

> Here's an excerpt of some fairly typical code generated by the JIT'er:

ROFL!
I've seen interpreters that can best that hands down :D

Eric Grange
http://glscene.org

"Andrew Rybenkov" <arybenkov@hotmail.com> wrote in message
news:3d734bfc@newsgroups.borland.com...
> > misinformation and it is much easier to identify who spreads rumors just
> > to hurt Delphi and Borland.
>
> Ignoring CPUs after 80386 Borland hurts to itself itself.
>

I doubt it.

If you look at the development of Delphi it's been focused on IDE and
integration features. The compiler has not so much neglected as been
prioritised appropriately for the majority of customers.

Unfortunately there are a few of us who have different priorities.

In mitigation -- barring register stalls when using 16 bits -- the P-II
and later run Delphi integer code quite well.

> Calling them interpreters is a bit beside the point.

OK. So it's an "interpreter that parses the code so that it can compile it
before running it". Sounds even slower than a traditional interpreter.

Olivier

"Rudy Velthuis (TeamB)" wrote:

> In article <3d7724da@newsgroups.borland.com>, Olivier Beltrami says...
>
> > > Calling them interpreters is a bit beside the point.
> >
> > OK. So it's an "interpreter that parses the code so that it can compile it
> > before running it". Sounds even slower than a traditional interpreter.
>
> It is not slower than the Delphi compiler. And compilation is done on a
> routine by routine basis, i.e. when it is needed, not the entire program
> at once.

Unless it is subsequently saved, (and it may be, I don't know)  then you are
still going to have a huge amount to process on startup.   Not the end of the
world, but an irritation nonetheless.


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

>the P-II and later run Delphi integer code quite well.

Yep, PII and later CPUs "saves" Delphi code performance with out-of-order execution.

--
Andrew Rybenkov.

create labels, logos, image lists
for your programs and websites with
  "Imagizer: Click and Get It",
  http://www.geocities.com/arybenkov/imagizer/

"Andrew Rybenkov" <arybenkov@hotmail.com> wrote in message
news:3d794e24@newsgroups.borland.com...
> > Out of order execution was introduced by Intel in the Pentium Pro (P6
core).
>
> But who saw PPro alive (except military folks)? IMHO it went to shadow as
80186
> did in its time.
>

There are a few about -- still. Mainly old servers IME.

Lots of 80186s went into embedded systems. I can only think of one
PC clone(sort of) that used them.