[llvm-dev] RFC: Adding argument allocas

[Hans Wennborg via llvm-dev]

On Thu, Dec 8, 2016 at 5:05 PM, Reid Kleckner via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> If the semantics are the same, it begs the question, why don’t we pattern
> match the alloca and store to elide dead stores and reuse existing argument
> stack slots?

One could reverse that argument: if we're going to emit the
alloca+store, as we currently do, only to pattern match it away, the
question is why emit the alloca+store in the first place? :-)

Also, pattern-matching would need to happen in the back-end since we
can't represent this in IR, right? Which means all the backends:
SelectionDAG, FastISel (and GlobalISel going forward).