On Thu, Jun 20, 2024 at 5:01 PM Daniel P. Berrangé <berra...@redhat.com> wrote:
> > config_host_data.set('CONFIG_AVX2_OPT', get_option('avx2') \
> > + .enable_auto_if(get_option('x86_version') >= '3') \
> > .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot
> > enable AVX2') \
> > .require(cc.links('''
> > #include <cpuid.h>
> > @@ -2875,6 +2876,7 @@ config_host_data.set('CONFIG_AVX2_OPT',
> > get_option('avx2') \
> > '''), error_message: 'AVX2 not available').allowed())
> >
> > config_host_data.set('CONFIG_AVX512BW_OPT', get_option('avx512bw') \
> > + .enable_auto_if(get_option('x86_version') >= '4') \
> > .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot
> > enable AVX512BW') \
> > .require(cc.links('''
> > #include <cpuid.h>
>
> I'm not sure this makes sense. The CONFIG_AVX* options are used only
> to validate whether the toolchain has support for this. The QEMU
> code then has a runtime, so it automagically uses AVX2/AVX512
> if-and-only-if running on a suitably new CPU. IOW, we want this
> enabled always when the toolchain supports it, regardless of what
> x86_version is set.
The difference is that if the toolchain does not support AVX2/AVX512
intrinsics for some reason, and you require -Dx86_version={3,4}, meson
would report an error with this patch.
Paolo
