[PATCH] Fix unfortunate interaction between foldSelectICmpAnd and urem

[Benjamin Kramer]

On 18.05.2013, at 03:31, David Majnemer <david.majnemer at gmail.com> wrote:

> We fail to convert a urem to an and if we have a select between two powers of two.
> The reason why is because it gets turned into (add (and (lshr V, C1), C2), C2) and we can not tell that it is a power of two any more.
> 
> The most natural thing seems to be to teach isKnownToBeAPowerOfTwo that (X & Y) + Y is a power of two or zero if Y is a power of two.

Making isKnownToBeAPowerOfTwo smarter is definitely the right way to fix this problem. Patch LGTM.

- Ben
> 
> A patch is attached with a test case.
> 
> -- 
> David Majnemer
> <rem-select-powers-of-2.diff>