[gambit-list] enable-single-host and gcc 4.2.3

Bradley Lucier lucier at math.purdue.edu
Thu Jul 10 13:42:15 EDT 2008


On Jul 10, 2008, at 1:01 PM, Jeremie Lasalle Ratelle wrote:

> Hi, I just installed ubuntu 8.04 and it seems that gcc 4.2.3 isn't  
> able
> to compile large files using --enable-single-host.
>
> We tried compiling a 6000 lines file and a 16000 lines file.  
> (Generated from Jazz modules)
>
> Using enable-single-host, both of them are still compiling after 10  
> minutes.
> Without enable-single-host it takes 1 minute for the 6000 one and  
> the 16000 one is still going on after 10 minutes.
>
> On Apple gcc 4.0.1 it takes 1.5 minute to compile the 16000 lines  
> file with enable-single-host
> In windows on gcc-mingw-3.4.5 it takes about 20 seconds with enable- 
> single-host.
>
> Could this be some serious regression?...

Yes.

I've been reporting inefficiencies in gcc compiling Gambit-generated  
C code for years.  See the open bugzilla report

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26854

opened in 2006; the most recent improvements were checked in in  
January (to the branch that became 4.3.0) and in May (to the branch  
that is developing the new register allocator, likely to be  
integrated into 4.4.0).  The main problem now is that gobs of memory  
are needed to compile truly large Scheme->C files with recent  
versions of gcc; the example in the bugzilla report takes 7.3GB (but  
< 3 minutes on a 2.2 GHz Opteron, so it should take about 90 sec on a  
recent Core 2 Duo).

I've just added a note with recent timing results to the bugzilla  
report, noting that the compilers that

I'd suggest looking at top to see how much memory cc1 is taking when  
compiling; if you run out of real memory it probably won't finish in  
any reasonable amount of time.

If you like I can try sometime in the next few weeks to see if some  
options on the gcc side or on the gambit side could reduce the  
compilation time to something reasonable.

Brad



More information about the Gambit-list mailing list