Floating point emulation

Floating point emulation

[Frank LoPinto]

I am using a DDC-I Ada 83 compiler that is hosted on a SunOS platform.
The compiler cross-compiles to an 80X86 target.

My application is an embedded system, and the target has no 80x87 math
co-processor. The DDC-I compiler only supports floating point math in
conjunction with a math co-processor.

Where can I get a 80x87 math co-processor emulation package that I can
link into my application?

Thanks in advance for any help you may be able to provide.

--Frank

--------------------------------------------------------
Frank LoPinto
BFGoodrich Aerospace
Aircraft Integrated Systems
Vergennes, VT
flopinto@aisvt.bfg.com

Re: Floating point emulation

[Michael F Brenner]

Is this an arbitrary restriction by an ivory tower? Or have you 
costed out the cost of upgrading the chip to have a floating point
unit versus the cost of emulating floating point in software?

Re: Floating point emulation

[Robert Dewar]

Michael asks

<<Is this an arbitrary restriction by an ivory tower? Or have you
costed out the cost of upgrading the chip to have a floating point
unit versus the cost of emulating floating point in software?>>

It is just something we (a) don't support and (b) none of our customers
have needed. Last I knew, ivory tower was not the usual phrase that people
use for the principle that you do what your customers are willing to pay for!

As I said in an earlier post, it is probably not that difficult a problem
to solve (get DOS and GNAT working with floating-point emulation), we have
invested zero effort in trying to solve this problem, because there is
zero demand. Well that's not quite fair, there are a couple of users with
educational licenses who would like it (notably Mike Feldman), but that's
not enough for us to justify working on it.

One of the changes that happened in the GNAT project when it moved from
being directly supported by the DoD to being commercially supported is that
the choices on such matters have to get far harder nosed. We have little
choice but to work on things that people will pay for. 

It would be nice if there were some general free support for educational
users, or if we could afford to provide this kind of free support. There
isn't, and we can't (at least not yet).

The one thing that we DO contribute on our own nickel is the continued
work to make public versions of GNAT in binary form for various targets.
Often people think that the GPL somehow mysteriously requires this kind
of effort -- it definitely does not! But as part of our contribution to
helping to spread Ada 95 use, particularly in universities, we think this
is a worthwhile effort, and will continue to spend our time doing it.

A lot of folks, particularly in universities, would like free support, and
that is understandable (sometimes people rant and rave, complete with
obscenities, *demanding* that we provide free support, and that is not
understandable :-) As i say, it would be nice if we could afford to provide
such support free, or if someone could fund it, but right now that's not
the way things are, and we are super busy.

Especially this week, preparing for four more 100% validatoins of GNAT for
four new targets that are just about to begin (the validators arrive from
Germany on Monday :-)

Robert Dewar
Ada Core Technologies

Re: Floating point emulation

[Andrew Dunstan]

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

: One of the changes that happened in the GNAT project when it moved from
: being directly supported by the DoD to being commercially supported is that
: the choices on such matters have to get far harder nosed. We have little
: choice but to work on things that people will pay for. 

: It would be nice if there were some general free support for educational
: users, or if we could afford to provide this kind of free support. There
: isn't, and we can't (at least not yet).

Good and reasonable point.

Most universities will not be able to purchase a meaningful support
contract, I guess (in Australia they are being severely squeezed by a
neanderthal government, and _every_ dollar has to be watched). I was
surprised, however, to discover that my old university doesn't appear
to be using GNAT, but a commercial compiler of moderate quality which
I suspect (will try to verify) is still an Ada83 version, on Solaris
platforms. 

I have no idea what (if anything) they pay for this, but one idea did
occur to me. Perhaps a GNAT Educational Fund could be established, to
fund development that would be of educational interest. Of course,
since I am no longer in the University world, this idea is put forward
with the "recklessness of the non-combatant" :-). Not only could
universities contribute the funds they might be paying for other,
possibly old, compilers, but corporate donations could also be sought.  

Just a thought.

cheers

andrew
--

-------------------------------------------------------------------------
  There's nothing either good or bad, but thinking makes it so - Hamlet
  http://www.gr-lakes.com/~andrew (including PGP key)
  PGP Key fingerprint = 5C 44 7D E4 76 A3 31 DE  3D 11 FA 15 4D 87 1F 5E
-------------------------------------------------------------------------

Re: Floating point emulation

[Joel VanLaven]

Andrew Dunstan (amd001@its.maynick.com.au) wrote:
[snip]
: Most universities will not be able to purchase a meaningful support
: contract, I guess (in Australia they are being severely squeezed by a
: neanderthal government, and _every_ dollar has to be watched). I was
: surprised, however, to discover that my old university doesn't appear
: to be using GNAT, but a commercial compiler of moderate quality which
: I suspect (will try to verify) is still an Ada83 version, on Solaris
: platforms. 

: I have no idea what (if anything) they pay for this, but one idea did
: occur to me. Perhaps a GNAT Educational Fund could be established, to
: fund development that would be of educational interest. Of course,
: since I am no longer in the University world, this idea is put forward
: with the "recklessness of the non-combatant" :-). Not only could
: universities contribute the funds they might be paying for other,
: possibly old, compilers, but corporate donations could also be sought.

  Interesting idea.  Do not jump to conclusions about poor little
commercial compiler vendors (like us :) though.  Many compiler vendors
offer either severely discounted or (like us) free compilers and tools
for educational purposes.  Some are even Ada95.  Just as in the
commercial arena, educators should have multiple options when it comes
to compilers and tools.  An introductory cs course might need an easy
to use gui and debugger but not need all of the annexes.  Another course
might find the ability to produce java applets a big plus and one might
need cobol interfacing and functionality capabilities.  Many of us have
a favorite compiler and development environment.  Each has advantages
and drawbacks and sometimes which we use is chosen for us.

  I would rather see more compiler-independant packages and so on put
forward than see a lot of special enhancements to one compiler.  I
think that these standard libraries are what we are missing, not the
compilers.

: Just a thought.

  Likewise.  (It does seem like it might be a good idea)

[snip]

-- 
-- Joel VanLaven

Re: Floating point emulation

[Robert Dewar]

<<Andrew Dunstan (amd001@its.maynick.com.au) wrote:
[snip]
: Most universities will not be able to purchase a meaningful support
: contract, I guess (in Australia they are being severely squeezed by a
: neanderthal government, and _every_ dollar has to be watched). I was
: surprised, however, to discover that my old university doesn't appear
: to be using GNAT, but a commercial compiler of moderate quality which
: I suspect (will try to verify) is still an Ada83 version, on Solaris
: platforms.>>

I am not quite sure of the context of this (since the original has not yet
shown up at our site here), but a relevant piece of information is that
Ada Core Technologies does provide inexpensive site licenses for universities
covering unlimited users on multiple platforms, and a number of universities
hold such contracts. For more information, contact sales@gnat.com.

Note that other Ada companies also provide free or inexpensive Ada
products for use in universities. The days when the statement above
was true are past!

Robert Dewar
Ada Core Technologies

Ada Support for Universities (was: Floating point emulation)

[Larry Kilgallen]

In article <dewar.858686090@merv>, dewar@merv.cs.nyu.edu (Robert Dewar) writes:
> <<Andrew Dunstan (amd001@its.maynick.com.au) wrote:
> [snip]
> : Most universities will not be able to purchase a meaningful support
> : contract, I guess (in Australia they are being severely squeezed by a
> : neanderthal government, and _every_ dollar has to be watched). I was
> : surprised, however, to discover that my old university doesn't appear
> : to be using GNAT, but a commercial compiler of moderate quality which
> : I suspect (will try to verify) is still an Ada83 version, on Solaris
> : platforms.>>
> 
> I am not quite sure of the context of this (since the original has not yet
> shown up at our site here), but a relevant piece of information is that
> Ada Core Technologies does provide inexpensive site licenses for universities
> covering unlimited users on multiple platforms, and a number of universities
> hold such contracts. For more information, contact sales@gnat.com.
> 
> Note that other Ada companies also provide free or inexpensive Ada
> products for use in universities. The days when the statement above
> was true are past!

And I hope this battle for mindshare will persist until the total
students reached by various Ada compilers reaches 100%.

Computer hardware companies over the past 30 years have used this
technique to gain fans entering industry.  One recent change
however is the prevalence of home computers, possibly allowing
children to be led down the path of darkness before they get
to University.

Larry Kilgallen

Re: Floating point emulation

[Frank LoPinto]

Michael F Brenner wrote:

Frank LoPinto wrote:
>>I am using a DDC-I Ada 83 compiler that is hosted on a SunOS platform.
>>The compiler cross-compiles to an 80X86 target.

>>Where can I get a 80x87 math co-processor emulation package that I can
>>link into my application?
> 
> Is this an arbitrary restriction by an ivory tower? Or have you
> costed out the cost of upgrading the chip to have a floating point
> unit versus the cost of emulating floating point in software?

No this is not an Ivory Tower restriction, in fact it is very much a
"real world" restriction. 

I am using and embedded version of the 80186, formally the 80186EB. The
system in which this processor resides will be placed on a helicopter.
We do not have an FPU beacuse of weight and power consumption
restrictions placed on the total system. There is no room. What is more,
the hardware is done. I have no choices with regards to harware.

Formerly, I was using the old DOS based Janus Ada compiler with the
embedded toolkit, for the pre-production and development versions of the
flight software. It had a floating point emulator such that if I told
the linker that I did not have an FPU, it would simply link in the FPU
emulator. This compiler, however, can no longer be used because I now
have a table that is over 4k in size and this compiler can not handle
it.

Now I could go and buy a new compiler that has everything I need, but
our company already has a license for the DDC-I that I can use. But as I
said, this compiler does not have FPU emulation.

Now there are a number of things I COULD do, but all options cost time
or money. I would rather spend neither, or at least reduce them to a
minimum. Hence the question.

--Frank

Frank LoPinto
BFGoodrich Aerospace
Aircraft Integrated Systems
Vergennes, VT
flopinto@aisvt.bfg.com

Re: Floating point emulation

[Michael F Brenner]

Frank LoPinto wrote:

Frank wrote:
>>Where can I get a 80x87 math co-processor emulation package that I can
>>link into my application?

Mike wrote:
> Is this an arbitrary restriction by an ivory tower? Or have you
> costed out the cost of upgrading the chip to have a floating point
> unit versus the cost of emulating floating point in software?

Frank Replied:
    > No this is not an Ivory Tower restriction, in fact it is very much a
    > "real world" restriction.
    > I am using and embedded version of the 80186, formally the 80186EB. The
    > system in which this processor resides will be placed on a helicopter.
    > We do not have an FPU beacuse of weight and power consumption
    > restrictions placed on the total system. There is no room. What is more,
    > the hardware is done. I have no choices with regards to harware.

Mike Explains:
Sorry I was not clear enough what I meant by Ivory Tower. In my 
opinion, one of the symptoms of Ivory Tower restrictions is crytallizing 
the hardware before the software is done, and this is not a criticism
of any one particular program, since so many of ourprograms do this. It is
just an observation that we would be better without such restrictions. When
engineers believe they are not permitted to cost alternatives, then what
they are doing is not engineering, but a form of politics, that sometimes
results in system failures.

Re: Floating point emulation

[Nick Roberts]

It's always been the case that engineering is politics.