Re: [pooma-dev] Prelinker Bug in KCC-4.0e

[Dave Nystrom]

Scott Haney writes:
 > 
 > On Wednesday, June 6, 2001, at 11:52 AM, Dave Nystrom wrote:
 > 
 > > Hopefully Scott is making some progress
 > > on this.
 > 
 > Ahem, I have to admit that I haven't spent much time on this due to an 
 > internal Proximation deadline.
 > 
 > Moreover, I'm going to Disneyland  (really!) at 4pm today and will 
 > return on June 18th.

Wow.  I did not know it was possible to spend 12 days at Disneyland.  Hope
you have a nice vacation.

 > I agree that the compile times you describe are too long. When I get 
 > back, I'll look into this from the POOMA perspective. However, I've got 
 > to  think that there are physical design issues on the Blanca-side or 
 > maybe compiler issues that are playing into this.

Blanca certainly has some physical design problems but my
Pooma2Instantiations library does not depend on or know about any Blanca code
and so all the compile time problems associated with it seem to me to be
localized to either physical design problems associated with Pooma 2 or
problems with how KCC deals with template instantiation.  When I see how much
intermediate C code is being generated from my Pooma2Instantiations library,
I have to believe that most of the problem I am seeing with this particular
library is with Pooma 2's design.  It is taking about 4 hours to compile just
this library.

I am wondering if you are really the only person who can deal with this
issue.  One thing that we have just learned is that the new Blanca project
leader, Tom Gorman, has a meeting with Stephen Lee and Don McCoy of the
program office on June 18 to discuss the Blanca Project.  Tom has requested
that the four of us provide him with some input for this meeting.  As I
understand it, he would basically like us to have something he could show to
the program office by then.

I think there are two ways that the Pooma 2 team could help us regarding
compile time issues that are associated with Pooma 2.  One I would classify
as a short term item and the other as a longer term item.  The short term
item is to make it possible for us to instantiate any template that the
compiler can instantiate in some reasonable fashion.  The second and longer
term item is to evaluate the Pooma 2 logical design and see if there are
changes that can be made to factor things differently or some other design
techniques that would result in Pooma 2's template code generating less
code.  In reality, being unfamiliar with the Pooma 2 design, I don't even
know if this second item would be very successful.

I am trying everything I know to try and get a handle on this issue and get
manageable compile times in our Unix environments.  I have KAI building me a
version of KCC-4.0d for RedHat 7.1 that I can use on my laptop until they get
the prelinker bug for 4.0e fixed.  I've sent email to Arch Robison to see if
it would be possible to use gcc as the backend C compiler on bluemountain
since the SGI C compiler seems to have some problem with compiling large
files.  So far, I have resisted turning my 20 instantiation files for the
Pooma 2 assign function into 100+ files so the SGI C compiler can handle them
better but I may have to do that.  I am looking at the possiblity of using
gcc 3.0 as a development compiler in Unix environments.  If you have any
ideas of more things I could do, please let me know.  Right now, I think my
best option is to try and press on with explicitly instantiating everything
and if someone else could help out with getting Pooma 2 to support this, it
would really be great.

-- 
Dave Nystrom                    email: address@hidden
LANL X-3                        phone: 505-667-7913     fax: 505-665-3046