Neoverse V2: remove sm3, sm4 and svesm4

[paolotricerri]

fixes #102

[paolotricerri]

@dslarm

[giordano]

Can you please edit the title of the PR to be self descriptive? Just referencing an issue number isn't useful.

[alalazo]

Can you please edit the title of the PR to be self descriptive? Just referencing an issue number isn't useful.

Asked the same thing in #105 😅 A brief description of the PR would be welcome too.

[dslarm]

I can confirm that this patch fixes my initial issue: the sm2/3 and svesm4 are not present on the V2 system that I am using. with this patch,

poetry run python
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import archspec.cpu
>>> print (archspec.cpu.host())
neoverse_v2
>>> 

and neoverse_n1 without it.

[alalazo]

Can you add a sample of the misidentified cpu to the tests? This would allow us to unit test this failed detection in archspec

[paolotricerri]

@alalazo , just as a test, so will have to confirm with @dslarm properly, I have taken the archspec repo, removed the SM crypto flags from neoverse_v2 as well as from the linux-rhel9-neoverse_v2 test file and the function cpu.detect.host() returns a microarchitecture for neoverse_v2 because of the sorting function (and so the length of ancestors and features) and not neoverse_n1.

I'll confirm @dslarm 's test and try to add a similar test to the archspec repo.

[dslarm]

Can you add a sample of the misidentified cpu to the tests? This would allow us to unit test this failed detection in archspec

@alalazo - this is the flags excerpt from cpuinfo:

Flags:              fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asim
                    dhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 asimddp sha51
                    2 sve asimdfhm dit uscat ilrcpc flagm ssbs sb paca pacg dcpo
                    dp sve2 sveaes svepmull svebitperm svesha3 flagm2 frint svei
                    8mm svebf16 i8mm bf16 dgh rng bti

[paolotricerri]

Hi @alalazo,
It should be safe to remove these flags from Neoverse V2 now because of the test on cpuid I propose in another PR.

What do you think?

Thanks
P

[alalazo]

@paolotricerri I think using cpuid is the way to disambiguate. On #106 (comment), can you add a relevant sample of /proc/cpuinfo as done, for instance in:

• https://github.com/archspec/archspec-json/blob/master/tests/targets/linux-amazon-neoverse_v1

?

[paolotricerri]

@alalazo , in the test folder I already see a file with data for the neoverse_V2 case. That example has got the feature that this patch is trying to remove. Do you mean adding another one taking the features from the comment above #106 (comment) ?

[alalazo]

Do you mean adding another one taking the features from the comment above #106 (comment) ?

Yes, they would be both two real cases of neoverse_v2 that we need to detect.

[alalazo]

@paolotricerri Can you add a sample of /proc/cpuinfo for the Neoverse V2 without the crypto features?

[alalazo]

Another consideration: do we need to remove sm3 and sm4 also from noeverse-v1? I would assume so, but seeking confirmation @dslarm

[paolotricerri]

I have moved the question I had over to #115 as that PR closes this one. Thanks