RFR: 8351902: RISC-V: Several tests fail after JDK-8351145

[Hamlin Li]

On Tue, 18 Mar 2025 12:22:00 GMT, Hamlin Li <mli at openjdk.org> wrote:

>> Hi,
>> Can you help to review this simple patch?
>> These client tests seems not that useful to me, so the simple solution could be just disable them on riscv.
>> 
>> Thanks!
>
> I think these tests are to verify the crypto intrinsics to be enable based on only CPU features.
> 
> For these failed tests, they depends on no cpu feature, but `AvoidUnalignedAccesses`, so we should skip them.
> For the tests such as `TestUseSHA256IntrinsicsOptionOnSupportedCPU` and `TestUseSHA512IntrinsicsOptionOnSupportedCPU`, they depends on CPU feature `zvkn`, so they run successfully.
> 
> We could modify the test framework to support passing `AvoidUnalignedAccesses` as `unsupportedCPUFeatures`, but seems to me it's not worth to do so, something like below:
> 
> --- a/test/hotspot/jtreg/compiler/testlibrary/sha/predicate/IntrinsicPredicates.java
> +++ b/test/hotspot/jtreg/compiler/testlibrary/sha/predicate/IntrinsicPredicates.java
> @@ -61,7 +61,7 @@ public class IntrinsicPredicates {
>  
>      public static final BooleanSupplier MD5_INSTRUCTION_AVAILABLE
>              = new OrPredicate(new CPUSpecificPredicate("aarch64.*", null, null),
> -              new OrPredicate(new CPUSpecificPredicate("riscv64.*", null, null),
> +              new OrPredicate(new CPUSpecificPredicate("riscv64.*", null, new String[] { "AvoidUnalignedAccesses" }),
>                // x86 variants
>                new OrPredicate(new CPUSpecificPredicate("amd64.*",   null, null),
>                new OrPredicate(new CPUSpecificPredicate("i386.*",    null, null),
> @@ -70,7 +70,7 @@ public class IntrinsicPredicates {
>      public static final BooleanSupplier SHA1_INSTRUCTION_AVAILABLE
>              = new OrPredicate(new CPUSpecificPredicate("aarch64.*", new String[] { "sha1" }, null),
>                // SHA-1 intrinsic is implemented with scalar instructions on riscv64
> -              new OrPredicate(new CPUSpecificPredicate("riscv64.*", null, null),
> +              new OrPredicate(new CPUSpecificPredicate("riscv64.*", null, new String[] { "AvoidUnalignedAccesses" }),
>                new OrPredicate(new CPUSpecificPredicate("s390.*",    new String[] { "sha1" }, null),
>                // x86 variants
>                new OrPredicate(new CPUSpecificPredicate("amd64.*",   new String[] { "sha" },  null),
> 
> 
> So, I'll add some comment about why we disable these tests on riscv.

> Yes, I got the test wrong. Thank @Hamlin-Li explaining it to me. I'm fine with just disabling it, as we basically need to predicate on AvoidUnalignedAccesses. Which in its turn is a predicate for UseMD5Instrinsic. So did we even test anything ? That is not clear to me and if there was something tested it seems to be of very little value.

Thank you!

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

PR Comment: https://git.openjdk.org/jdk/pull/24027#issuecomment-2733535899