RFR: 8310190: C2 SuperWord: AlignVector is broken, generates misaligned packs

[Emanuel Peter]

On Mon, 13 Nov 2023 08:31:27 GMT, Fei Gao <fgao at openjdk.org> wrote:

>>> @fg1417 @reinrich @zifeihan
>>> 
>>> I am getting more confident with the updated version. I reran testing, and the first performance testing results look promising (i.e. no regression).
>>> 
>>> Would you mind re-running testing on your platforms? (I ran on x64 and aarch64 without SVE)
>>> 
>>> @fg1417 I don't know aarch64 very well. Would you want to create a matching-rule for aarch64 for `VerifyVectorAlignment`, similar to what I did in `x86.ad`? It is not necessary, but could be nice.
>> 
>> Hi, I have performed tier1-3 test on linux-riscv64 using QEMU with -XX:+UseRVV. Still good.
>
>> Would you mind re-running testing on your platforms? (I ran on x64 and aarch64 without SVE)
> 
> Hi @eme64 , I've submitted testing on aarch64 (sve) and arm32 separately. Let's wait for results.
> 
>> @fg1417 I don't know aarch64 very well. Would you want to create a matching-rule for aarch64 for `VerifyVectorAlignment`, similar to what I did in `x86.ad`? It is not necessary, but could be nice.
> 
> Of course, I will create the matching rule for aarch64. BTW, what kind of testing have you done for `VerifyVectorAlignment` to see if it works as expected? Then I can verify it as well on aarch64.

@fg1417 thanks for the help!
I ran tier1-6 with `-XX:+AlignVector -XX:+IgnoreUnrecognizedVMOptions -XX:+VerifyAlignVector`.
Though `VerifyAlignVector` only has an effect if there is a matching rule in the `ad` files.
And I ran quite a few repetitions of `compiler/loopopts/superword/TestAlignVectorFuzzer.java`.

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

PR Comment: https://git.openjdk.org/jdk/pull/14785#issuecomment-1807678116