[jdk21u-dev] RFR: 8310308: IR Framework: check for type and size of vector nodes [v3]

[Goetz Lindenmaier]

On Mon, 18 Dec 2023 10:58:18 GMT, Goetz Lindenmaier <goetz at openjdk.org> wrote:

>> I backport this to improve testing of C2 in 22u. It will simplify backporting tests of fixes to C2, as, e.g. " 8313720: C2 SuperWord: wrong result with -XX:+UseVectorCmov -XX:+UseCMoveUnconditionally"
>> 
>> I had to trivially resolve three files:
>> 
>> test/hotspot/jtreg/compiler/loopopts/superword/TestDependencyOffsets.java
>> Resolved the bugid in comment.
>> 
>> patching file test/hotspot/jtreg/compiler/vectorization/runner/BasicDoubleOpTest.java
>> patching file test/hotspot/jtreg/compiler/vectorization/runner/BasicFloatOpTest.java
>> These had to be resolved because "8308340: C2: Idealize Fma nodes" is not in 21. Trivial.
>> That change is huge thus not a candidate for backport.
>> 
>> Further I include "8314513: [IR Framework] Some internal IR Framework tests are failing after JDK-8310308 on PPC and Cascade Lake".  This is a follow-up fix of the tests.
>> 
>> All tests passed locally. 
>> 
>> It's a test-only change.
>
> Goetz Lindenmaier has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Fix failing test - revert to status of 21

TestVectorizationMismatchedAccess.java was failing.
This is because in head, subtest testByteByte2 was enabled by 8310308. But in 21, it is disabled. 
If I enable it in 21 without this change, it fails with the same message as with the change. So I disabled it again.

As I understand, 21 does not generate vector nodes for the test method, and in 22 there was a functional improvement.

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

PR Comment: https://git.openjdk.org/jdk21u-dev/pull/29#issuecomment-1860371133