correctness and comprehensiveness of compilers [was: Re: C compiletime]

[Vladimir N. Makarov]

"Wolf, Joe" wrote:

> Vlad,
>
> 7.1 does correctly build and run SPECint2000.  The config file we use
> for IA-32 SPEC numbers should be on the spec.org site.  It turns out
> that vpr and gcc are sensitive to floating-point precision.  By, default
> our compiler does not round, but truncates assignments and casts
> (-fp_port-).  Gcc and vpr does not work unless we round fp results,
> though.  -xW (P4 optimization/vectorization/scheduling) enables rounding
> (-fp_port) and allows gcc and vpr to run correctly.  So, I suspect this
> is what you are running into and if you do not want run at -O2 or -O3
> without P4-specific optimization, just add -fp_port.
>

  I am sorry.  I should have analyzed the problem of miscompilation of the
tests before jumping to the conclusion.  As I remember, the previous
versions of icc (6 and 7.0) did not require special options to compile gcc
and vpr correctly with -O2.

Actually some versions of x86 gcc have analogous problem for eon when
temporary fp results are not truncated to IEEE doubles (but saved in Intel
80-bits fp format).

My appologies again.

Vlad


>
> Please let me know what options you are using and we can follow up
> further if needed.
>
> Joe Wolf
> Intel Compiler Technical Consulting & Support Manager
> Software Products Division
> Intel Corporation
> joe.wolf@intel.com
>
> -----Original Message-----
> From: Vladimir Makarov [mailto:vmakarov@redhat.com]
> Sent: Wednesday, June 18, 2003 7:16 AM
> To: Karel Gardas
> Cc: Scott Robert Ladd; Dara Hazeghi; gcc mailing list
> Subject: Re: correctness and comprehensiveness of compilers [was: Re: C
> compiletime]
>
> Karel Gardas wrote:
> >
> > On Wed, 18 Jun 2003, Scott Robert Ladd wrote:
> >
> > Just one example: all last three releases of Intel C++ compiler for
> > Linux (I mean 6.0, 7.0 and 7.1) miscompile MICO (http://www.mico.org)
> > such that its IDL compiler is either crashing (6.0 + 7.0) or loops
> > infinitely (7.1).
> >
>
>   I don't know about MICO test.  But it is amazing for me how such
> respectable company as Intel could make a release (7.1) which can not
> compile correctly some SPECINT2000 tests (vpr and gcc) with -O2 and -O3
> for P4 (their mainstream processor).  I found that a week ago using the
> demo CD which we got on the gcc summit.
>
> Vlad