RFR: 8298935: fix cyclic dependency bug in create_pack logic in SuperWord::find_adjacent_refs [v9]

Emanuel Peter epeter at openjdk.org
Thu Mar 2 07:48:19 UTC 2023


On Thu, 2 Mar 2023 05:47:26 GMT, Jatin Bhateja <jbhateja at openjdk.org> wrote:

>> Sounds good.
>
>> @jatin-bhateja I now have a first version out of the `test/hotspot/jtreg/compiler/loopopts/superword/TestDependencyOffsets.java`. It seems to work for `sse4.1 ... avx512`. I'm now testing it for `asimd`. And then I will proceed to add the features for `-XX:+AlignVector`, with the modulo check.
> 
> I am seeing lots of IR violations with UseSSE=4.

I realized I have lots of negative IR rules that check that we do NOT vectorize if I expect cyclic dependency. But these negative rules are difficult, there may always be some other factor that leads to shorter vector sizes than what I expect. And then it vectorizes, and does not encounter a cyclic dependency. So I will have to remove all these negative IR rules.

@jatin-bhateja was there any positive IR rule that failed? One that did expect vectorization, but it did not in fact vectorize?

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

PR: https://git.openjdk.org/jdk/pull/12350


More information about the hotspot-compiler-dev mailing list