[jdk17] RFR: 8269825: [TESTBUG] Missing testing for x86 KNL platforms

Vladimir Kozlov kvn at openjdk.java.net
Tue Jul 6 16:56:47 UTC 2021


On Fri, 2 Jul 2021 19:35:56 GMT, Vladimir Kozlov <kvn at openjdk.org> wrote:

> Knights family of X86 Intel CPU (KNL) does not support some of avx512 features (AVX512VL/BW) and have other restrictions. We may not have such kind of machines in our testing environment and may miss bugs as JBS history shows (look recent fixes for KNL).
> On other hand we have some Windows VM instances which seem emulate KNL configuration and limit avx512 instructions on CPU which supports full set. Recent bug JDK-8269775 shows such example.
> I suggest to add -XX:+UseKNLSetting x86 diagnostic flag to emulate KNL settings in HotSpot VM to test such configuration.

Interesting. Excluding AVX512DQ does help to pass tests from 8269828 (at least on my local machine).

Yes, my listed before CPUID was incorrect (was generated with my patch). Here is correct one from one of crashes we see (no AVX512BW, AVX512VL, AVX512DQ, CLWB, FLUSHOPT):


CPU: total 8 (initial active 8) (4 cores per cpu, 2 threads per core) family 6 model 85 stepping 4 microcode 0x1, cx8, cmov, fxsr, ht, mmx, 3dnowpref, sse, sse2, sse3, ssse3, sse4.1, sse4.2, popcnt, lzcnt, tsc, avx, avx2, aes, erms, clmul, bmi1, bmi2, rtm, adx, avx512f, avx512cd, fma, vzeroupper, clflush, hv


I am currently testing update for this PR which excludes AVX512DQ too.

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

PR: https://git.openjdk.java.net/jdk17/pull/205


More information about the hotspot-compiler-dev mailing list