Re: Optimising std::find on x86 and PPC

[Paolo Carlini <pcarlini at suse dot de>]

Zdenek Dvorak wrote:

> as Andrew noted, the problem is that the loop actually looks like
>
> for (i = a; i < b; i += 4)
>  something;
>
> At rtl level we do not have sufficient information to determine the
> number of iterations of this loop precisely at runtime -- we cannot
> determine whether it is not infinite.  This however should not block
> loop unrolling, and should be relatively simple to fix; I will try.
>  
Thanks in advance, Zdenek.

> Note however that the fact that 3.3 unrolls this loop very likely means
> that 3.3 is buggy; I guess that 3.3 would also unroll
>
> for (i = a; i < b; i += 5)
>  something;
>
> Which would be obviously wrong, unless you are very careful with the way
> number of iterations is determined (which 3.3 is not).
>  
I see, thanks for the explanation. Something similar already happened in 
other
circumstances, I seem to remember.

Thanks again,
Paolo.