[llvm] [X86] Fold some (truncate (srl (add X, C1), C2)) patterns to (add (truncate (srl X, C2)), C1') (PR #126448)

[Phoebe Wang via llvm-commits]

=?utf-8?q?João?= Gouveia <jtalonegouveia at gmail.com>,
=?utf-8?q?João?= Gouveia <jtalonegouveia at gmail.com>,
=?utf-8?q?João?= Gouveia <jtalonegouveia at gmail.com>,
=?utf-8?q?João?= Gouveia <jtalonegouveia at gmail.com>,
=?utf-8?q?João?= Gouveia <jtalonegouveia at gmail.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/126448 at github.com>


phoebewang wrote:

> > > [eb2f176d67cdf1955a90e71e25d6d39910d723d4e0b8a9bf8dfa229d3a6b2c1eR53685-R53686](https://github.com/llvm/llvm-project/commit/b773400981d8ca97de016f6acbe02e71b30b5779#diff-eb2f176d67cdf1955a90e71e25d6d39910d723d4e0b8a9bf8dfa229d3a6b2c1eR53685-R53686)), should I remove it entirely?
> > 
> > 
> > This reminds me can we expand it to AND/XOR/OR/SUB, maybe in a follow up patch? Then I think we can remove it. For i64, it seems only the constant fold is benefit.
> 
> This transformation is correct for those cases as well. Expanding it to include those ops would require adding a couple of tests and expanding the `sd_match` to cover them, so it should be pretty simple. I can do it in this patch or in a follow up, depending on how wide you want the scope of this patch to be.

Thanks! The general preference is to use multiple patches and keep each simple.

> It looks like `AND` is already being transformed (https://godbolt.org/z/a5d5vhdsb), so for that specific operation maybe just adding a test would suffice?

Seems it's due to the number is happen to be power of 2, https://godbolt.org/z/1za939PKc

https://github.com/llvm/llvm-project/pull/126448