(971016) g++ and Blitz -- YES

(971016) g++ and Blitz -- YES

[Mumit Khan]

I've been waiting for this for a while now -- g++'s ability to blaze
thru Blitz++'s (*) templates! Good news is that all example code build 
(qcd requires a minor fix to blitz/meta/matmat.h), and the results I've 
seen so far look ok so far; bad news is that you better have a machine 
with lots of memory to be able to compile and the resulting code size 
can be huge if you want debugging info.

Congratulations to the egcs team, and specially to the folks who have
contributed to the new template handling (thanks Mark and Jason!).

I'm within microns of getting my own code to build under egcs -- the new
rules (guiding decls) have me a bit confused, but now it's just a matter
of time.

Regards,
Mumit -- khan@xraylith.wisc.edu
http://www.xraylith.wisc.edu/~khan/

* Blitz++ is a class library for scientific computing designed by Todd 
Veldhuizen. More info at < http://monet.uwaterloo.ca/blitz/ >.

Re: (971016) g++ and Blitz -- YES

[Bill Gribble]

Mumit Khan <khan@xraylith.wisc.edu> writes:
> I'm within microns of getting my own code to build under egcs -- the new
> rules (guiding decls) have me a bit confused, but now it's just a matter
> of time.

WRT memory consumption: I have found that on our Linux machines using
heavily STL-ized code, there's one of my own source files that
reliably runs out of virtual memory even with 64M physical RAM and
over 500M swap.  After reading Jason's suggestion that -Wreturn-type
(implied by -Wall) defeated his memory-hogging fix, I started
stripping off options.

Compiling with no -W switches, no optimization, and no debugging info
shrinks memory consumption by more than an order of magnitude (at
least -- a file that failed to compile with 564M available compiles
without touching disk now).

And SGI STL brought some really, really terrible coding errors to
light that HP STL should have barfed all over but compiled without a
warning.  

Go, egcs! 

Bill Gribble

Re: (971016) g++ and Blitz -- YES

[Corey Kosak]

Bill Gribble writes:
> After reading Jason's suggestion that -Wreturn-type
> (implied by -Wall) defeated his memory-hogging fix, I started
> stripping off options.

Does this mean that the memory bloat problem is believed fixed for the cases
when -Wxxx is not used?  I have an example which uses a lot of memory using
just -O.  [ignore, for the moment, my unusual use of @ in filenames]

> limit datasize 150M
> g++ -O @.xall.cc
xservice.@: In method `communicationService_t::communicationService_t(const class mystring &, class generic_remote_t &)':
In file included from xall.@:15,
                 from @.xall.cc:1:
xservice.@:128: virtual memory exhausted


I can supply the file or try to reduce it to a small test case if desired.  If
this is still an open issue, I apologize for bringing it up again.  Perhaps a
"to do" / "open issues" list should be put on the egcs web page somewhere?