RFR: 8317121: vector_masked_load instruction is moved too early after JDK-8286941

[Tobias Hartmann]

On Tue, 3 Oct 2023 12:55:23 GMT, David Schlosnagle <duke at openjdk.org> wrote:

>> We have internal setup which run all vectorization tests with different AVX and SSE settings.
>> I don't know about production Aarch64 machine which supports SVE.
>
> [AWS Graviton 3 instance types support 4x Neon 128bit / 2x SVE 256bit vectors and Graviton 2 instance types support 2x Neon 128bit vectors](https://github.com/aws/aws-graviton-getting-started#building-for-graviton2-graviton3-and-graviton3e)

I think such flags/settings should only be part of an individual test if they are required to trigger a specific bug. Otherwise, we would need to add a lot of different flag combinations to a lot of tests. Instead, tests/tiers should be executed with such flag values regularly, for example, as part of a continuous integration system like we have here at Oracle.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16015#discussion_r1344152519