RFR: 8309894: compiler/vectorapi/VectorLogicalOpIdentityTest.java fails on SVE system with UseSVE=0

[Andrew Haley]

On Sun, 25 Jun 2023 07:02:07 GMT, Pengfei Li <pli at openjdk.org> wrote:

> > I discussed it with @chhagedorn and @TobiHartmann . They also think it is better to have flags and cpu features in sync. Because the flags really should restrict what the VM uses everywhere. We want to make sure the restrictions apply in the VM, if we check for features or flags.

I agree.

> Having flags and cpu features in sync sounds a good idea. However, of all platforms supported by HotSpot, only x86 does in this way. Even on x86, only AVX & SSE related features have such sync at present. There is no sync for other feature strings and flags, like `UseSHA`, `UseAES` and etc. That's the reason we think current approach of x86 is more like a workaround for IR tests only. If you believe keeping this kind of sync is a better approach, I'd suggest doing this for all CPU features on all platforms in another RFE.

That's an interesting suggestion, but having to keep all of the back ends in lock-step is an intolerable constraint.

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

PR Comment: https://git.openjdk.org/jdk/pull/14533#issuecomment-1606016037