6.0-Current and gcc 4.x

[O. Hartmann]

Xin LI wrote:
> Hi,
> 
> 在 2005-06-18六的 13:46 +0200，O. Hartmann写道：
> 
>>Hi.
>>As I see in the sources, FreeBSD 6.0-Current is still based on gcc 
>>3.4.4. I think due to code freezing this version will be the platform 
>>compiler when 6.0-RELEASE comes out. Are there any plans using a more 
>>recent version of gcc, like 4.0 or 4.1? On some lists I read something 
>>about better support of the features of AMD64/EM64T architectures, 
>>especially SSE3.
> 
> 
> Currently there is no plan to upgrade to GCC 4.x.  We are in a code
> slush for preparation of RELENG_6 right now, since GCC 4.x import would
> involve a large scale code modification

All right, I think this is one of the most important aspects.

> (some of our code would not
> compile on gcc 4.x since it is more picky than 3.x series, and some of
> gcc 3.x features has been removed since they are not ISO C/C++ and
> certain part of our code has utilized these features), and causes
> performance degradation on i386 architecture, I don't think we will
> import gcc 4.x in FreeBSD 6.x.

I wasn't aware of those restrictions. I also did not recognize the 
'Nocona' switch which obviously offers the SSE3 option for Intel/EM64T 
add ons. I remember of some discussions herein about mapping this flag 
into Athlon64/Opteron world, sorry for that.

Maybe this is a bit of off topic, but I'm curious about why gcc 4.0 let 
FreeBSD run into performance penalties. May you give me a hint for 
further digging the net? I'm interested in this.

> 
> Help on making FreeBSD compile on newer compiler(s) and clean up the
> code base is always appreciated, of course :-)  Meantime you would be
> able to get latest GCC 4.x from our ports collection.

And another point i did not thought of (ports collection).

> 
> Cheers,
Also off topic:

Well, the major intention is to use a faster C and/or free Fortran 
compiler doing my numerical research on FreeBSD boxes. In most cases I 
gained success on compiling Fortran90 code with PGI and/or Intel V8 
compilers, running this code on FBSD 5.3/4 boxes within the Linuxulator. 
I found out that some C code performs not that fast as  expected using 
PGI or Intel C compiler, but Fortran code (F77) performs really faster 
(fast in the manner of: it is useable!).
On a pure AMD64-FreeBSD 5.4-STABLE box I feel like a dead man in the 
water because I do not have access to a 64Bit Linuxulator using the 
64Bit versions of Intel or PGI (which gave me a real boost on our SuSE 
Linux Cluster, but I do not wish to develop under Linux, sorry).

All right, that's a bit apart of the main question.

Thanks for your response.

Oliver