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

[Fei Gao]

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`.

Hi @eme64, we have [ `tst`](https://developer.arm.com/documentation/ddi0602/2023-09/Base-Instructions/TST--immediate---Test-bits--immediate---an-alias-of-ANDS--immediate--?lang=en) on aarch64.

instruct verify_vector_alignment(iRegP r, immL_positive_bitmaskI mask, rFlagsReg cr) %{
  match(Set r (VerifyVectorAlignment r mask));
  effect(KILL cr);
  format %{ "verify_vector_alignment $r $mask \t! verify alignment" %}
  ins_encode %{
    Label Lskip;
    // check if masked bits of r are zero
    __ tst($r$$Register, $mask$$constant);
    __ br(Assembler::EQ, Lskip);
    __ stop("verify_vector_alignment found a misaligned vector memory access");
    __ bind(Lskip);
  %}
  ins_pipe( pipe_slow );
%}


I tested tier1-tier3 with `-XX:+AlignVector -XX:+IgnoreUnrecognizedVMOptions -XX:+VerifyAlignVector` on aarch64(neon and 128-bit sve) platforms, and several repetitions of `compiler/loopopts/superword/TestAlignVectorFuzzer.java`. No new failures found. Hope it works for you. Thanks.

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

PR Comment: https://git.openjdk.org/jdk/pull/14785#issuecomment-1809626686