On 25 Jul 2023, at 04:56, Khem Raj via lists.openembedded.org 
<[email protected]> wrote:
> 
> Signed-off-by: Khem Raj <[email protected]>
> ---
> meta/recipes-core/glibc/glibc_2.37.bb | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/meta/recipes-core/glibc/glibc_2.37.bb 
> b/meta/recipes-core/glibc/glibc_2.37.bb
> index 8ff9801ed4b..4290f3e6e4f 100644
> --- a/meta/recipes-core/glibc/glibc_2.37.bb
> +++ b/meta/recipes-core/glibc/glibc_2.37.bb
> @@ -85,7 +85,8 @@ EXTRA_OECONF += "${@get_libc_fpu_setting(bb, d)}"
> EXTRA_OECONF:append:x86 = " ${@bb.utils.contains_any('TUNE_FEATURES', 'i586 
> c3', '--disable-cet', '--enable-cet', d)}"
> EXTRA_OECONF:append:x86-64 = " --enable-cet"
> 
> -PACKAGECONFIG ??= "nscd memory-tagging"
> +PACKAGECONFIG ??= "nscd memory-tagging ${@bb.utils.filter('TUNE_FEATURES', 
> 'sve', d)}"
> +
> PACKAGECONFIG[nscd] = "--enable-nscd,--disable-nscd"
> PACKAGECONFIG[memory-tagging] = 
> "--enable-memory-tagging,--disable-memory-tagging"
> PACKAGECONFIG[sve] = "--enable-mathvec,--disable-mathvec"

This is the wrong thing to do, imho.

mathvec has existed since glibc 2.22 for x86-64 and only with 2.38 did it gain 
aarch64 support, where it has both NEON and SVE (optional in 8.2+) 
implementations.

Fedora 38 and Debian 11 both have a libmvec.so, so it looks like we should be 
turning it on by default for the relevant architectures (x86-64 for <2.38, 
x86-64 and aarch64 for 2.38+).

Ross
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#185522): 
https://lists.openembedded.org/g/openembedded-core/message/185522
Mute This Topic: https://lists.openembedded.org/mt/100344107/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to