[PING] [3.4] Fix PR11707

[Roger Sayle]

On Thu, 10 Feb 2005, Richard Guenther wrote:

> Please consider
> http://gcc.gnu.org/ml/gcc-patches/2005-01/msg00656.html
> or tell me to close the bug as WONTFIX.


I think that the biggest issue with this patch is that the decision
to reorder RTL passes is never to be taken lightly, it can expose
latent RTL optimizer problems on obscure platforms and have unexpected
performance impacts on different targets.

Consider the effort that went into introducing jump bypassing in the
first place: http://gcc.gnu.org/ml/gcc/2002-09/msg01129.html and
the decisions as to whether to place it before or after LOOP.  That
patch took nine months to get approved, had been bootstrapped and
regression tested on no less than eight platforms and thanks to
Andreas Jaeger over ten variants had been fully SPECcpu2000 tested.


By comparison, you're attempting to make a similar change on a
highly stable release branch with a fraction of the testing.


I don't think PR11707 needs to be closed as WONTFIX (for example the
problem could be solved in some other less intrusive way) but I think
Mark would need significantly more convincing of your proposed patch
than has currently been presented.

My 2c.

Roger
--