RFR: 8323220: Reassociate loop invariants involved in Cmps and Add/Subs [v6]
Joshua Cao
duke at openjdk.org
Thu Feb 8 07:48:54 UTC 2024
On Wed, 7 Feb 2024 07:43:01 GMT, Emanuel Peter <epeter at openjdk.org> wrote:
>> @eme64 can you take a look at the updates?
>
> @caojoshua I don't think you have added what I requested. Do you have any test that does the reassociation, and where you verify that the computed values are correct?
OK thanks I think I misread. I now have some correctness tests for tests with and without reassociation.
We can't add correctness tests for tests with random inputs because we do not know how many iterations there will be. There are two tests with reassociation `equalsVariantSubInvarint(Int|Long)` that have a drastically different graph with the correctness change that messes up the IR matching, so I opted out of changes those as well.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/17375#discussion_r1482540647
More information about the hotspot-compiler-dev
mailing list