[15] RFR (S): 8238696: x86: Enumerate all detected CPU features in VM_Version feature string

[David Holmes]

On 11/02/2020 11:58 pm, Vladimir Ivanov wrote:
> 
>> Now I'm wondering about using the dot as a terminator. Doesn't the 
>> return value from jio_snprintf tell you whether you exceeded the buffer?
> 
> Yes, very good point. How about the following?
> 
>    http://cr.openjdk.java.net/~vlivanov/8238696/webrev.03

Functional part seems fine.

> Considering some tests have hard-coded feature names, I decided to 
> enhance the new test and check that all reported features are known (at 
> least, on x86).

Not sure about the test - I don't know enough about which features are 
"always" presents on x86/x64. I'd suggest using @requires x86 rather 
than the internal Platform checks.

Thanks,
David

> Best regards,
> Vladimir Ivanov
> 
>>> On 10.02.2020 17:12, Vladimir Ivanov wrote:
>>>> Thanks for taking a look, David.
>>>>
>>>> Updated webrev:
>>>>    http://cr.openjdk.java.net/~vlivanov/8238696/webrev.01/
>>>>
>>>>>> http://cr.openjdk.java.net/~vlivanov/8238696/webrev.00/
>>>>>> https://bugs.openjdk.java.net/browse/JDK-8238696
>>>>>>
>>>>>> Abstract_VM_Version::features_string() doesn't contain all the CPU 
>>>>>> features detected by the JVM on x86 (for example, for AVX-512 only 
>>>>>> AVX-512F is included as "evex").
>>>>>>
>>>>>> The fix adds missing features to the list [1].
>>>>>
>>>>> The function naming that you have changed for 
>>>>> supports_avx512_vpclmulqdq was explicitly requested by Vladimir K.:
>>>>>
>>>>> http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2019-December/036326.html 
>>>>>
>>>>>
>>>>> If anything I would suggest supports_vpopcntdq() should have the 
>>>>> avx512 part added.
>>>>
>>>> Makes sense. Also, decided to rename VNNI, VAES, and VBMI2. They are 
>>>> listed in the manual as part of AVX512.
>>>>
>>>>> The actual expansion of the feature string seems okay. It does make 
>>>>> me wonder why these items have been left out previously.
>>>>
>>>> I believe it was just overlooked. Added a comment near cpu feature 
>>>> constant declarations.
>>>>
>>>> Also, introduced a check to ensure the temporary on-stack buffer is 
>>>> large enough.
>>>>
>>>> Best regards,
>>>> Vladimir Ivanov
>>>>>
>>>>> Thanks,
>>>>> David
>>>>>
>>>>>> Testing: manual (-Xlog:os+cpu=trace output on different hardware),
>>>>>>           hs-precheckin-comp, hs-tier1, hs-tier2
>>>>>>
>>>>>> Best regards,
>>>>>> Vladimir Ivanov
>>>>>>
>>>>>> [1] $ java -Xlog:os+cpu=trace -XX:UseAVX=3 -version
>>>>>>
>>>>>> ...
>>>>>> [0.010s][info][os,cpu] UseSSE=4  UseAVX=3  UseAES=1  MaxVectorSize=64
>>>>>> ...
>>>>>> [0.011s][info][os,cpu] CPU:total 8 (initial active 8) (4 cores per 
>>>>>> cpu, 2 threads per core) family 6 model 85 stepping 4, cmov, cx8, 
>>>>>> fxsr, mmx, sse, sse2, sse3, ssse3, sse4.1, sse4.2, popcnt, 
>>>>>> vzeroupper, avx, avx2, aes, clmul, erms, rtm, 3dnowpref, lzcnt, 
>>>>>> ht, tsc, bmi1, bmi2, adx, avx512f, avx512dq, avx512cd, avx512bw, 
>>>>>> avx512vl, fma, clflush, clflushopt, clwb
>>>>>> ...
>>>>>> [0.011s][info][os,cpu] flags        : fpu vme de pse tsc msr pae 
>>>>>> mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse 
>>>>>> sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon 
>>>>>> rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq vmx 
>>>>>> ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt 
>>>>>> tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm 
>>>>>> abm 3dnowprefetch cpuid_fault invpcid_single pti ssbd ibrs ibpb 
>>>>>> stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust 
>>>>>> bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx avx512f avx512dq 
>>>>>> rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl 
>>>>>> xsaveopt xsavec xgetbv1 xsaves arat umip pku ospke md_clear
>>>>>> ...