Re: GNAT & Windows NT

Re: GNAT & Windows NT

[Michael Feldman]

In article <3h3j4e$9kn@news.cais.com>,
Lance Kibblewhite <lance@cais.com> wrote:
>I see that the currently available version of GNAT is intended for Pentium
>systems only.

Huh? This is definitely not so. Where did you get that idea?
>
>Is anybody working on a version for 486's?  There can't be too many things
>within GNAT (or GCC) that would necessitate a Pentium rather then a 486.

You don't need a Pentium. I used GNAT for a year on a 386/33, then 
upgraded to a 486/66.
>
>Actually, I'd prefer an Alpha AXP version, but I would guess that is
>really a problem with getting a GCC port first.

A garden-variety DOS clone with at least a 386 and at least 4mb RAM will
be fine. The more RAM the better; djgpp (the gcc for DOS) uses virtual
memory and will use what you give it, but of course the more the better.
I have 8 and no trouble.

Mike Feldman
------------------------------------------------------------------------
Michael B. Feldman -  chair, SIGAda Education Working Group
Professor, Dept. of Electrical Engineering and Computer Science
The George Washington University -  Washington, DC 20052 USA
202-994-5919 (voice) - 202-994-0227 (fax) - mfeldman@seas.gwu.edu (Internet)
------------------------------------------------------------------------
One, two, three ways an underdog: Ada fan, Mac fan, Old Liberal Democrat
------------------------------------------------------------------------
         Ada on the World-Wide Web: http://lglwww.epfl.ch/Ada/
------------------------------------------------------------------------

Re: GNAT & Windows NT

[Robert Dewar]

Lance says

"I see that the currently available version of GNAT is intended for Pentium
systems only.

Is anybody working on a version for 486's?  There can't be too many things
within GNAT (or GCC) that would necessitate a Pentium rather then a 486.

Actually, I'd prefer an Alpha AXP version, but I would guess that is
really a problem with getting a GCC port first.


Where did you get the idea that the current version of GNAT is only for
the Pentium. This is nonsense, it is a 386/486/Pentium port (remember
that these architectures are equivalent at the user level, modulo Pentium's
well publicized incapability of guarateeing correct division results).

Indeed there has been some discussion of a version that would compensate
for the Pentium deficiencies, but we have not done anything with this yet
(though we know how).

There already is a version for Alpha AXP.

Re: GNAT & Windows NT

[Timothy Halloran]

In <3h6o5j$ge8@gnat.cs.nyu.edu> dewar@cs.nyu.edu (Robert Dewar) writes: 
>Robert Dewer says:
>Where did you get the idea that the current version of GNAT is only for
>the Pentium. This is nonsense, it is a 386/486/Pentium port (remember
>that these architectures are equivalent at the user level, modulo Pentium's
>well publicized incapability of guarateeing correct division results).
>
>Indeed there has been some discussion of a version that would compensate
>for the Pentium deficiencies, but we have not done anything with this yet
>(though we know how).

I can see how this mistake was made (since I made it also): the binary
file on your server (cs.nyu.edu/pub/gnat) lists the file with "586" in
the file name.

I thought that it wouldn't run on my 486 and have continued to use
the DOS version (with no tasking).

Maybe someone (GNAT developers anyone?, anyone?) could comment on the
differences between GNAT and gcc between DOS and NT.  I know that
Windows 3.1 uses the "new executable file" format for programs rather
than what is used in DOS (I don't know what Win32 programs use).  I
also suspect that GNAT under dos is generating 32-bit
executables (some sort of 32-bit extender maybe?).

Anywhere where a curious programmer can find answers to the above
questions?

So may Microsoft Operating systems,
So little time,

Tim Halloran

Re: GNAT & Windows NT

[Vladimir Vukicevic]

In article <3h3j4e$9kn@news.cais.com> lance@cais.com (Lance
Kibblewhite) writes:

> I see that the currently available version of GNAT is intended for
> Pentium systems only.
>
> Is anybody working on a version for 486's?  There can't be too many
> things within GNAT (or GCC) that would necessitate a Pentium rather
> then a 486.

pragma Not_Helpful (On);

First of all, a few points.
	1) Dump Windows
	2) Dump Windows
	3) Dump Windows
	4) Dump Microsoft (along with Windows. and Bob needs to go, too.).

Then, get a decent OS, like Linux. or {Net,Free}BSD. or OS/2. (OS/2 is
supported by gnat straight out of nyu!) or anything without "Microsoft"
in front of it.

pragma Not_Helpul (Off);

If none of these are possibilites, then I'd say to try using the
pentium-only code on your 486. It might work, unless it has a
dependancy on using FDIV as a random number generator... :-)

If it doesn't find a pentium machine and recompile gcc for 486 instead
of pentium.

	- Vladimir

Re: GNAT & Windows NT

[Richard Kenner]

In article <3h3j4e$9kn@news.cais.com> lance@cais.com (Lance Kibblewhite) writes:
>I see that the currently available version of GNAT is intended for Pentium
>systems only.
>
>Is anybody working on a version for 486's?  There can't be too many things
>within GNAT (or GCC) that would necessitate a Pentium rather then a 486.

In fact there are none.  I think (but am not 100% sure) that that's also
true for the 486.  So the "Pentium" version should also run on a 486 and
probably 386.

The "pentium" in the name refers just to where it was built and optimized
for (though there are few Pentium-specific optimizations in GCC right
now).

>Actually, I'd prefer an Alpha AXP version, but I would guess that is
>really a problem with getting a GCC port first.

Right. There are code-generation issues on Alpha/NT.  I hope to have
time to deal with them in the next month or so.  GCC and GNAT should
be easily ported to the other NT platforms (MIPS and PowerPC) by just
splicing config files together, but nobody has done this yet as far as
I know.

Re: GNAT & Windows NT

[Lance Kibblewhite]

Noting the subject line, you will see that I was referring to GNAT on 
Windows NT.  I am well aware that the DOS GNAT will run on a at least a 
386, but I am looking for an NT specific version.

The only NT version of GNAT I could find was labeled as an i586 version. 
I am told there is also an alpha NT version somewhere, so any pointers to
where it is would be appreciated.  I'll check again on cs.nyu, but I
didn't see it last time I looked. 

Note also that the installation instructions for for the NT version also
referred to a pentium.  If the code is not Pentium (i586) specific, then I
would suggest they rename the distribution to correctly indicate the
platform. 

-- Lance.

Re: GNAT & Windows NT

[Richard Kenner]

In article <3h7vo0$19m@news.cais.com> lance@cais.cais.com (Lance Kibblewhite) writes:
>I am told there is also an alpha NT version somewhere,

Not yet.  try again in a month or two.

>Note also that the installation instructions for for the NT version also
>referred to a pentium.  If the code is not Pentium (i586) specific, then I
>would suggest they rename the distribution to correctly indicate the
>platform. 

Renaming the distribution to, say i486, would just prompt people to ask
if it would run on a Pentium!  I agree about the installation instructions.

Re: GNAT & Windows NT

[Douglas Rupp]

In article <3h72or$p1g@ixnews2.ix.netcom.com> halloran@ix.netcom.com (Timothy Halloran) writes:
>
>Maybe someone (GNAT developers anyone?, anyone?) could comment on the
>differences between GNAT and gcc between DOS and NT.  I know that
>Windows 3.1 uses the "new executable file" format for programs rather
>than what is used in DOS (I don't know what Win32 programs use).  I
>also suspect that GNAT under dos is generating 32-bit
>executables (some sort of 32-bit extender maybe?).
>

GNAT and Gcc on both DOS and NT use COFF executables.  On DOS, the go32
memory extender is prepended onto the COFF file to give you an .exe file
recognizable to the OS.  Since the Microsoft linker is used on the NT
versions, it does whatever to make it recognizable to NT.

Re: GNAT & Windows NT

[Robert Dewar]

"They rename the distribution to correctly identify the platform"

As you know, there are a zillion different PC platforms. All are user
level compatible (assuming we are talking 386 or above). Unless you
do something VERY weird (we have not), code running on the Pentium will
run on the 386 and 486 and vice versa (actually there are more differences
between the 386 and 486 at this level than between the 486 and Pentium).

Generally you want to distribute code that is optimized for the Pentium, 
since this can produce substantial advantages in execution efficiency. 
As Richard Kenner has noted, GCC does not yet do much in the way of
Pentium specific optimizations yet, but this may well change in the
future. Pentium optimized code runs fine on a 486, i.e. there is no
noticable penalty for such optimizations, which is why you always prefer
to ship Pentium optimized code.

The correct identification of the platform for our PC related releases
is indeed 586, since these releases were built on a Pentium. Renaming
them would be incorrect and misleading. Supplying additional ports that
are built on the 486 would be extra work and pointless, since the code
would not run faster on a 486.

Understanding this will be important to the entire PC world as we begin
to see Pentium-optimized code appear for general PC applications, and I
am sure the same confusion will arise. We will add notes to our readme
files to make sure that people (at least those who read the file!) do
not get confused on this point.

There are two ports of GNAT to NT, one built and distributed by NYU, which
is available from the NYU FTP area, and one from Labtek. The Labtek folks
specialize in NT support for GNAT, and are building some interesting NT
specific tools, so if you are interested in using GNAT on NT, you should
definitely contact them. 

Re: GNAT & Windows NT

[David O'Brien]

Robert Dewar (dewar@cs.nyu.edu) wrote:
: "They rename the distribution to correctly identify the platform"

: As you know, there are a zillion different PC platforms. All are user
: level compatible (assuming we are talking 386 or above). Unless you
: do something VERY weird (we have not), code running on the Pentium will
: run on the 386 and 486 and vice versa (actually there are more differences
: between the 386 and 486 at this level than between the 486 and Pentium).

The name of the dist file is "...-i586-ibm-winnt-...".  May I suggest
changing the name to "intel-i386" and removing the "ibm"?  IBM has
nothing to do with the OS nor the machine (anymore).  Granted you state
this is because of the optimizations, but [most] everyone knows that
something that claims to run the i386 will run on the i486 and Pentium.
Anyway, Intel doesn't use or like [I believe] "i586" for the Pentium CPU.
Another reason to go with "i386" for this file is that is the convention
Microsoft uses with NT distributions.  [Can't believe that I'm
*actually* advocating Microsoft!].

Anyway just a opinion from a Microsoft NT user and administrator [yuck!].

-- david o'brien	(dobrien@seas.gwu.edu)

Re: GNAT & Windows NT

[Tom Griest]

In article <3hbds1$pav@cronkite.seas.gwu.edu> dobrien@seas.gwu.edu (David O'Brien) writes:
>Robert Dewar (dewar@cs.nyu.edu) wrote:
>: "They rename the distribution to correctly identify the platform"
>
>: As you know, there are a zillion different PC platforms. All are user
>: level compatible (assuming we are talking 386 or above). Unless you
>: do something VERY weird (we have not), code running on the Pentium will
>: run on the 386 and 486 and vice versa (actually there are more differences
>: between the 386 and 486 at this level than between the 486 and Pentium).
>
>The name of the dist file is "...-i586-ibm-winnt-...".  May I suggest
>changing the name to "intel-i386" and removing the "ibm"?  

[snip]

It is pretty clear what the problem is here.  NYU is naming the
distribution based on the platform it was BUILT on.  Whereas
everyone who distributes software labels software based on what
platforms it will run on.   Although it would be pretty unusual
to find 586 programs that won't run on a 486, it certainly is
possible and Intel states that there are three "application instructions"
that are new from the 486 to Pentium.  For example:
   CMPXCHG8B (compare and exchange 8-byte)

Personally, I don't care what it is called.  Maybe you should put
in filename "links" and ln 386 and 486 names to the same file...
then see which one people download :^).

Maybe just label it X86 and let them guess?

-Tom

Re: GNAT & Windows NT

[Lance Kibblewhite]

Robert Dewar (dewar@cs.nyu.edu) wrote:
: Lance says

: "I see that the currently available version of GNAT is intended for Pentium
: systems only.

: Is anybody working on a version for 486's?  There can't be too many things
: within GNAT (or GCC) that would necessitate a Pentium rather then a 486.

: Actually, I'd prefer an Alpha AXP version, but I would guess that is
: really a problem with getting a GCC port first.


: Where did you get the idea that the current version of GNAT is only for
: the Pentium. This is nonsense, it is a 386/486/Pentium port (remember
: that these architectures are equivalent at the user level, modulo Pentium's
: well publicized incapability of guarateeing correct division results).

The idea that it was for a Pentium was a combination of two factors.
Firstly, it is labelled as an i586 version, and secondly, I had the
assumption that while members of the x86 family were upward compatible, 
they were necessarily downward comptible.  This assumption is obviously
incorrect (for 386, 486, and Pentium).

Now that I have been enlightened, I will proceed to install GNAT on my
486 NT system.

: There already is a version for Alpha AXP.

I've also been told this one is a month or so away.  I need this one for
work, but I can wait.

-- lance.

Re: GNAT & Windows NT

[Jules]

In article <3h6o5j$ge8@gnat.cs.nyu.edu>,
	dewar@cs.nyu.edu (Robert Dewar) writes:

>             This is nonsense, it is a 386/486/Pentium port (remember
>that these architectures are equivalent at the user level, modulo Pentium's
>well publicized incapability of guarateeing correct division results).

Note that a 386 is not necesarily compatible with 486/Pentium software, as
a 486DX or higher is guaranteed to have a floating-point coprocessor unit,
whereas 386s and 486SXs do not always have such a unit, and consequently
software emulation must be supplied. If software emulation is not supplied,
they will not be compatible.


-- 
/* Julian R Hall				csusb@csv.warwick.ac.uk
   
   Flames should be redirected to /dev/null - I don't know what
   I'm saying myself so don't expect it to make sense all the time!         */