[cfe-dev] C++ devirtualization

[Joerg Sonnenberger]

On Wed, Apr 23, 2014 at 06:02:34PM +0200, Matthieu Monrocq wrote:
> TL;DR: where I wonder about the viability of a seemingly simple change to
> the code generation in Clang that seem to trick LLVM in doing a lot more
> devirtualization work than it's doing today... without any change to LLVM
> itself.

Have you tried using llvm.invariant.start / llvm.invariant.end markers
instead of the stores?

Joerg