Re: [Xen-devel] [PATCH v4 1/4] x86: use PDEP/PEXT for maddr/direct-map-offset conversion when available

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 31.10.18 at 18:03, <wei.liu2@xxxxxxxxxx> wrote:
> On Tue, Sep 18, 2018 at 06:36:21AM -0600, Jan Beulich wrote:
>> This allows to fold 6 instructions into a single one, reducing code size
>> quite a bit, especially when not considering the fallback functions
>> (which won't ever need to be brought into iCache or their mappings into
>> iTLB on systems supporting BMI2).
>> 
>> Make use of gcc's new V operand modifier, even if that results in a
>> slightly odd dependency in the sources (but I also didn't want to
>> introduce yet another manifest constant). This assumes that backports of
>> support for this relatively new modifier have only been done to tool
>> chains with not too old a binutils (gas) version. If this turns out to
>> be a false assumption, we'll have to add HAVE_AS_BMI2 as a qualifier.
>> 
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> 
> I'm afraid this patch series smells like  "micro optimisation gone too
> far" to me. :-/

Well, as explained in replies to Andrew as well, the primary motivation
here is two-fold:
1) The binary code involved in these translations is rather hard to
follow in some cases. Analysis of crashes would be a bit easier with
the simpler resulting insns.
2) The ultimate use of a single MOV is, considering the large number
of instances, quite certainly an improvement on the vast majority
of systems.

That said, I think I've decided to not pursue this series following
Andrew's comments, which in part have matched your remark.

Jan



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel