RFR: 8333890: Fatal error in auto-vectorizer with float16 kernel [v2]

[Jasmine Karthikeyan]

On Mon, 8 Jul 2024 10:47:29 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

> This seems to somewhat clash with an earlier comment here:

I added `ConvertNode::create_convert` and that comment last year, and I remember running into this same issue when trying to represent half float. I had to hack around it using this:
https://github.com/openjdk/jdk/blob/c34a1b7013b27a8a214f63387bd528a90342a416/src/hotspot/share/opto/cfgnode.cpp#L1886-L1895
I also think it'd be nice to have a unified representation.

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

PR Comment: https://git.openjdk.org/jdk/pull/20062#issuecomment-2213774325