[PATCH] x86: Clean up and improve FPU inline assembly

[Uros Bizjak]

On Fri, Sep 19, 2025 at 8:35 PM Florian Weimer <fweimer@redhat.com> wrote:
>
> * Uros Bizjak:
>
> > Clean up and improve x86 FPU inline assembly:
> >
> >   * Remove obsolete "*&" GCC asm memory operand workaround
> >
> >   * Use %v prefix to emit VEX prefixed insns for AVX targets
> >
> >   * Use \t and \n\t separators consistently
> >
> > Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
>
> Reviewed-by: Florian Weimer <fweimer@redhat.com>
>
> I'm not particularly fond of the \t, but I probably won't change any of
> this code, so it doesn't matter to me.

Insn stream looks somewhat nicer in the assembly dumps.

> The %v changes make the VEX prefix more consistent.  I suppose that's
> okay.  But it's puzzling to me my the VEX variant is advantageous, given
> that its encoding is one byte longer.  I wouldn't consider it a SIMD
> instruction, so the VEX vs legacy mixing shouldn't apply?

Actually, it avoids mode switching penalties, especially since the
decoder doesn’t have to switch between legacy and VEX instruction
decode paths.

Uros.