[PATCH] D26224: NewGVN

[Daniel Berlin via llvm-commits]

On Mon, Dec 12, 2016 at 7:09 PM, Daniel Berlin <dberlin at dberlin.org> wrote:

> There are about 5 starting since 2007.  You have to be careful, most focus
> on finding "equivalence among variables in the program".
>
> IE they are trying to find equivalence among *variables* and not among the
> set of *subexpressions in the program*.
>
> That is, they are useful for redundancy elimination. The will tell you
> "this instruction and that instruction are equal".
>
> But you can imagine sub-expressions that, if they were
> reassociated/forward propagated, would be equivalent.
>
> The one i've seen implemented for real was "An Efficient SSA-Based
> Algorithm for Complete Global Value Numbering" by Nie and Cheng.
>
> There is also, for example,
> http://link.springer.com/chapter/10.1007/978-3-319-26529-2_4
>
>
and
http://dl.acm.org/citation.cfm?id=2555120
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161212/ac283e92/attachment.html>