[llvm] r201275 - Remove a very old instcombine where we would turn sequences of selects into

[Chandler Carruth]

On Wed, Feb 12, 2014 at 4:10 PM, Nadav Rotem <nrotem at apple.com> wrote:

> I don’t have a specific testcase that regressed, this is purely
> theoretical.


I think we need test cases to drive these types of changes. Without them,
we're operating in the dark, and cannot check our assumptions.

Notably, this is a *canonicalization* issue. It may be significantly more
appropriate to transform back to the efficient form when lowering a vector
for a target where it is more efficient. Test cases would help the analysis
of these kinds of tradeoffs.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140212/39e2c163/attachment.html>