RFR: 8346664: C2: Optimize mask check with constant offset [v12]

[Emanuel Peter]

On Fri, 31 Jan 2025 01:32:52 GMT, Quan Anh Mai <qamai at openjdk.org> wrote:

> I mean disabling the @IR annotation in case my statement is ambiguous. The method seems to not vectorize on master, either. I suspect the reason is the same as that of [JDK-8348096](https://bugs.openjdk.org/browse/JDK-8348096) when we push the OrI through Phi:

Ah, that's probably true.

> I think you should disable dst[i] = src[i] * (i | 7) and we can investigate the situation there later.

I was hoping we could find at least some op that works. But it's not that important. Well, it just shows that we need to look a little harder into the issue of [JDK-8348096](https://bugs.openjdk.org/browse/JDK-8348096), i.e. pushing things through phi.

-> I'm fine with removing the IR rule there.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/22856#issuecomment-2626385923