Patch for stricter implicit conversions between vectors

[Andrew Pinski]

On Tue, 2006-11-14 at 17:23 +0000, Mark Shinwell wrote:
> If the consensus is not to have the flag, so be it -- but after the previous
> discussion (http://gcc.gnu.org/ml/gcc/2006-11/msg00051.html for example) it
> seemed that people were more in favour of the flag rather than just the more
> strict behaviour.  It isn't entirely clear to me that just because this issue
> is a regression then we should reject any such broken code in question with
> no option: that sort of behaviour might only serve to annoy.  I suppose the
> deciding factor is really whether there are significant quantities of
> broken code out there that people will be unwilling to fix -- others here
> are much better placed than I to make a judgement on that one -- or possibly
> how willing we are to risk alienating users :-)

I should mention the PS3's GCC has already been fixed for this issue.
So the remaining vector users I know of, usually also try to have it
compile with 3.3 also so the alienating users is not going to happen.  

We (SCE) have already got complaints that GCC was not strict with
respect of implicit conversions which is why the PS3's GCC was fixed.
Now we should have pushed the bug upstream but we did not and I only
found about the problem when you (Mark) mentioned it on the list.  This
is one of the reasons why I have been advocating we just fix the
regression without adding an option to allow the implicit conversions.

The main reason why I don't like the option is that it allows people to
write bad code and then just work around that issue but using the
option.  The other reason is the code is no longer portable between
compilers (or even compiler versions for that matter since we rejected
it in 3.4.0 and before) and yes other compilers have vectors and some
even support attributes and vector_size.

Thanks,
Andrew Pinski