RFR: 8259610: VectorReshapeTests are not effective due to failing to intrinsify "VectorSupport.convert" [v4]

[Mai Đặng Quân Anh]

On Thu, 9 Dec 2021 17:19:39 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:

>> Mai Đặng Quân Anh has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   fix cpu pattern on Neon
>
> test/hotspot/jtreg/compiler/vectorapi/reshape/tests/TestVectorCast.java line 39:
> 
>> 37:  * In each cast, the VectorCastNode is expected to appear exactly once.
>> 38:  */
>> 39: public class TestVectorCast {
> 
> At some point i would like to explore autogenerating such files., from another Java program and a template mechanism (unfortunately we cannot use a library like Java Poet). Not now though! 
> 
> This also relates to redesigning the existing vector unit tests, moving away from a bash script to a more flexible Java program, from which we can generate unit tests and/or IR tests.
> 
> The challenge with the IR tests is knowing associated IR nodes and the set supported on various platforms. I like how you enumerated the list for the conversions. Ideally we could go to the source of truth in C2 and determine that, but it does not seem easy to determine.

I believe if we have a way to query for compiler support from Java the same way as we do in C2 then the task would be trivial. I don't know if we have a way to call into C2 from Java similar to JNI, though.

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

PR: https://git.openjdk.java.net/jdk/pull/6724