[llvm] [GlobalIsel] Combine G_PTR_ADD. (PR #95647)

[Amara Emerson via llvm-commits]

Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>,
Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/95647 at github.com>


aemerson wrote:

> > Until we can show benefit this needs work (also AMDGPU regressions in the tests).
> 
> I have a hard time believing we don't need G_PTR_ADD + offsetting computation combines. Any indexing patterns resulting from introducing stack temporaries, and intrinsics that lower to pointer + offset would benefit

If realistic IR tests can be added that shows the benefits then feel free to reopen this. I couldn't tell from the AMDGPU tests if these were improvements or not.

https://github.com/llvm/llvm-project/pull/95647