On Wednesday, 17 August 2016 at 19:36:17 UTC, Walter Bright wrote:
On 8/17/2016 5:20 AM, deadalnix wrote:
Controlling aliasing is really the #1 optimization barrier
these days, so I
don't think it's that good of a thing.
Almost every single one case where Rust end up being faster
than C++ is because
their type system allow for more AA information available for
the optimizer.
AA is also key to do non GC memory management at language
level.
At least for this case, as I mentioned in another post, if the
pointer and length of the global is cached in a local, it can
be cached in a register. The contents of locals don't have
aliasing problems because if their addresses are not taken,
nobody can point to them. Optimization relies heavily on that.
But doing so would be incorrect if D doesn't provide strong
aliasing guarantees. And if D does provide these guarantees, we
won't need to do this manually.
