[Rd] O2 optimization produces wrong code (PR#5315)
dmurdoch at pair.com
Tue Nov 25 22:45:20 MET 2003
On Tue, 25 Nov 2003 21:22:08 +0000 (GMT), Prof Brian Ripley
<ripley at stats.ox.ac.uk> wrote :
>On 25 Nov 2003, Peter Dalgaard wrote:
>> Duncan Murdoch <dmurdoch at pair.com> writes:
>> > On Tue, 25 Nov 2003 15:47:01 +0000 (GMT), Prof Brian Ripley
>> > <ripley at stats.ox.ac.uk> wrote :
>> > >This is a long-known problem: one example is the MASS example in the
>> > >script ch15.R, and that has gone wrong on platforms other than Windows.
>> > >It has been reported to the maintainer in the past.
>> > Should we back off to something less aggressive than -O2?
>> Traditionally, we've prioritized performance and used cases like this
>> to file bugs for the compiler writers. But is this really a compiler
>> problem or a problem with badly written code in akima??
>I suspect the latter, in particular I suspect uninitialized data areas.
>As far as I recall, this has gone wrong even without optimization.
I think I misread Brian as saying that this is a long-known problem
with gcc, rather than a long-known problem with akima. Back to my
original suggestion: give the akima maintainer a reproducible
More information about the R-devel