https://bugs.kde.org/show_bug.cgi?id=520753

--- Comment #4 from Mark Wielaard <[email protected]> ---
(In reply to mcermak from comment #3)
> Created attachment 192784 [details]
> proposed patch
> 
> This proposed patch seems to address the requirement. 

Yes. The code looks good.
Please do update the commit message a little.
"sowftare" -> "software"

I find "Instead it is run in the HW, but indirectly - via the BSR instruction. 
In fact, LZCNT was introduced with BMI in the Haswell processor." a little
confusing because it was originally a AMD extension, and we don't implement BMI
for x86. Maybe just say:
"The LZCNT instruction is only recognized if the host cpu supports LZCNT"

> I've tested in in
> QEMU/KVM by setting:
> 
>   <cpu mode="host-passthrough" check="none" migratable="on">
>     <feature policy="disable" name="abm"/>
>   </cpu>

Nice, I didn't know you could mask specific cpuid feature bits with qemu.

> This testing however shows that similar update should be done also for the
> amd64 case, where the LZCNT instruction is always present under valgrind.  
> Does it make sense to do analogous update for amd64?  If so, may I do that
> as part of this bug?

That makes sense, it can be the same bug (or a new one). I would do it in two
separate patches.

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to