[llvm] [SLP][Revec] Consider revectorization of the following shuffle pattern (PR #206903)

[Alexey Bataev via llvm-commits]

alexey-bataev wrote:

> > > Some high-level question: can the same stuff be done in VectorCombine pass?
> > 
> > 
> > Yes, VectorCombine does this optimisation.
> 
> This was my 1st reply. Apologies if this misconveyed the information. So to reiterate, vector combine is doing this transformation.
> 
> Caveat is doing it in SLP leads to better codegen than doing this in vector combine.
> 
> So, doing this in SLP may sound redundant but proves beneficial, does this now make sense?

Why the codegen is better? It should not be better, it should be the same, not matter where we're doing this transformation

https://github.com/llvm/llvm-project/pull/206903