Suspicous code in kdelibs-3.5.2
steleman at nyc.rr.com
Thu Apr 6 12:59:47 BST 2006
On Thursday 06 April 2006 03:04, Oswald Buddenhagen wrote:
> > I don't know if Intel EMT64 (Xeon/Pentium 4 && friends) are NUMA.
> > Need input from an Intel guru.
> yes? and how many compilers do you know that would be even remotely
> able to optimize such cases? i'd be too interested in knowing how
> they do it. honestly, i think trying to apply NUMA as an example
> here is a dead end. using complex structures transparently might be
> used on segmented architectures where offset calculations with
> segments are impossible (e.g., i286 (16 bit) protected mode). i
> don't think this applies to any architecture [prospectively]
> supported by kde.
All the commercial compilers from UNIX vendors support NUMA. If the
hardware supports NUMA, the compiler will support it as well. That's
what the compiler teams get paid to do: write compilers which support
the underlying architecture. We are not talking about segmented
memory architectures like the i286, we are talking about code which
violates 8.3.4 of the C++ Standard. Since the Standard clearly states
that the first subscript of a multidimensional array plays no part in
subscript calculations, NUMAs are prime candidate architectures for
which the assumption that multidimensional arrays must be allocated
in contiguous memory will fail.
What exactly do you mean by "architectures [prospectively] supported
by KDE" ? Is it an official policy that KDE only runs on VAX ?
If you are interested in finding out how commercial compilers
implement NUMA, the only thing you have to do is get a job with one
of the compiler teams at either IBM, Sun, HP or Apple.
Here's what i think is a dead end: discussions predicated on the
assumption that "what i don't know doesn't exist".
Stefan Teleman 'Nobody Expects the Spanish Inquisition'
steleman at nyc.rr.com -Monty Python
More information about the kde-core-devel