Re: Artificial front-end variables?

[Tobias Schlüter <tobias dot schlueter at physik dot uni-muenchen dot de>]

Andrew Pinski wrote:
> > Another PR that comes to mind (PR12840):
> > m = SUM((/ (m+1,m=0,order) /))
> > -> t = (/ (m+1,m=0,order) /); m = sum(t)
>
>
> There was one for matmul (13152) but I closed it as a dup of bug 12840 
> because I thought
> it would be cured when PR12840 was also fixed.

Ah, I remember that. I thought they were separate, because the compiler 
takes quite different ways until it hits that failure, so I didn't think 
of them as dups. In the SUM case, the frontend expands the function when 
generating the GENERIC trees, so it is much more likely that the actual 
bug, which later causes the problem, is in the SUM expansion than that 
the bug is in the scalarizer. But I didn't think about this enough (and 
it could be that there is an enhancement to the scalarizer which fixes 
both) to want to override your assessment.

Thanks,
- Tobi