[C++] Use existing library for CPU detection

Describe the enhancement requested

We currently use our own home-grown module for detecting CPU features, but it mainly supports x86, which implies that for now we cannot do dynamic dispatch on non-x86 CPUs (for example ARM where we might want to detect ARM or even SVE support at runtime).

We could probably reuse https://github.com/google/cpu_features instead. It doesn't seem to have any dependencies, and could be bundled as a private static library.

This was already proposed years ago as #16715

Component(s)

C++

@cyb70289

Looking at https://github.com/google/cpu_features/blob/main/include/cpuinfo_aarch64.h , it's not obvious to me how we could get the supported SVE vector widths. Do you have any idea @cyb70289 ?

Edit: apparently there's an intrinsic to get the vector width: https://lemire.me/blog/2022/11/29/how-big-are-your-sve-registers-aws-graviton/

+1

The library itself looks good. My main concern is the slowdown of its release cadence: https://github.com/google/cpu_features/releases

Hmm, that's a good point. Maintenance also seems to have slowed down a bit: https://github.com/google/cpu_features/graphs/contributors?from=20%2F01%2F2024

There's also https://github.com/anrieff/libcpuid , which only supports x86 and ARM but exposes a lot of information.

And by the way, xsimd has a more limited functionality for CPU feature detection: https://xsimd.readthedocs.io/en/latest/api/arch.html . It only supports x86 and ARM, does not offer detailed CPU information, does not expose things like BMI2 etc.

For completeness, you might be interested in https://github.com/archspec/archspec - it supports x64, aarch, ppc & risc, plus a lot of meta-information about the (partial) inclusion ordering of the various CPUs in terms of feature sets.

I think it doesn't have a C/C++ API though, so this is probably a long shot (slow to call python from C++), but OTOH, querying CPU features is not a common occurrence (once per session...?) and could be cached.

Numpy does it the other way around, detecting individual CPU features, rather than processor models, with a result à la:

Supported SIMD extensions in this NumPy install:
    baseline = SSE,SSE2,SSE3
    found = SSSE3,SSE41,POPCNT,SSE42,AVX,F16C,FMA3,AVX2
    not found = AVX512F,AVX512CD,AVX512_KNL,AVX512_KNM,AVX512_SKX,AVX512_CLX,AVX512_CNL,AVX512_ICL

It's implemented in C, but I don't think it has a public API.

@h-vetinari

@h-vetinari Well, we don't want the core Arrow C++ library to depend on the existence of a Python interpreter and Python libraries. As for the NumPy C code, since it's internal it's probably not trivial to vendor, and I suppose its APIs have no stability guarantee either.

Yeah, I understand. That's why I said "for completeness". :)

The library itself looks good. My main concern is the slowdown of its release cadence: https://github.com/google/cpu_features/releases

For the record, a new release was just made (v0.10.0).

@AntoinePrv

Also cc @AntoinePrv

My 2c as not an arrow maintainer: if we want to use more of xsimd in arrow, it could be relevant to extend support for what we lack in xsimd. It would make static/dynamic dispatch of xsimd code in arrow more convenient as opposed to requiring a conversion layer and another dependency.

Perhaps, but some information such as cache sizes might be out of scope for xsimd anyway.

@AntoinePrv

So, IMHO, we have two reasonable options:

Use xsimd. Pros: header-only, already a dependency. Cons: might not have all desired functionality, so we'd either need to add them to xsimd (if that's ok to the xsimd maintainers) or keep maintaining code for it in Arrow.
Use Google's cpu-features. Pros: it's more featureful. Cons: it's an additional dependency, and requires a compilation step as it's not header-only; therefore it will some complication to the build system.

@AntoinePrv Do you want to do a bit more research to find out which functionality is needed in xsimd and whether it can be added there? (and feel free it to contribute it of course :))

@pitrou

Update on this, the current idea discussed with @pitrou is to use xsimd (already a dependency).

We've done some work on xsimd to match Arrow CPU detection needs (excluding cache size, but that is a separate topic and apparently only used for tests at the moment). There is a good basis to easily add more feature as needed.

GH-49921: [C++] Bump xsimd to 14.2.0 (#49922)

GH-45331: [C++] Use xsimd for CPU feature detection (#49940)

Issue resolved by pull request 49940
#49940

Navigation Menu

Repository navigation

[C++] Use existing library for CPU detection #45331

Describe the enhancement requested

Component(s)

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[C++] Use existing library for CPU detection #45331

Description

Describe the enhancement requested

Component(s)

Activity

pitrou commented on Jan 22, 2025

pitrou commented on Jan 22, 2025

kou commented on Jan 22, 2025

cyb70289 commented on Jan 23, 2025

wgtmac commented on Jan 23, 2025

pitrou commented on Jan 23, 2025

pitrou commented on Jan 23, 2025

pitrou commented on Jan 23, 2025

h-vetinari commented on Jan 23, 2025

pitrou commented on Jan 23, 2025

h-vetinari commented on Jan 23, 2025

pitrou commented on May 6, 2025

pitrou commented on Jul 15, 2025

AntoinePrv commented on Jul 15, 2025

pitrou commented on Jul 15, 2025

pitrou commented on Jul 17, 2025

AntoinePrv commented on Apr 10, 2026

pitrou commented on Jun 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions