[llvm] r212444 - [x86] Teach the new vector shuffle lowering code to handle what is

[Chandler Carruth]

On Mon, Jul 7, 2014 at 2:23 PM, Hal Finkel <hfinkel at anl.gov> wrote:

> > [x86] Teach the new vector shuffle lowering code to handle what is
> > essentially a DAG combine that never gets a chance to run.
>
> Can you please explain what this means? We run DAGCombine in
> AfterLegalizeDAG mode just before instruction selection.


But by then, we've finished selecting x86 instructions for the
VECTOR_SHUFFLE node. The problem is that we see new VECTOR_SHUFFLE nodes
created during instruction selection, and yet instruction selection has to
remove all such nodes before it finishes and the combiner runs again.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140707/f41f5679/attachment.html>