2014-07-03 14:56 GMT+04:00 Jakub Jelinek <ja...@redhat.com>:
> On Thu, Jul 03, 2014 at 02:49:10PM +0400, Ilya Enkovich wrote:
>> 2014-07-02 20:21 GMT+04:00 Andi Kleen <a...@firstfloor.org>:
>> > Ilya Enkovich <enkovich....@gmail.com> writes:
>> >
>> >> Silvermont processors have penalty for instructions having 4+ bytes of
>> >> prefixes (including escape bytes in opcode). This situation happens
>> >> when REX prefix is used in SSE4 instructions. This patch tries to
>> >> avoid such situation by preferring xmm0-xmm7 usage over xmm8-xmm15 in
>> >> those instructions. I achieved it by adding new tuning flag and new
>> >> alternatives affected by tuning.
>> >
>> > Why make it a tuning flag? Shouldn't this help unconditionally for code
>> > size everywhere? Or is there some drawback?
>>
>> There is already a higher priority for registers not requiring REX.
>> My patch affects cases when compiler has to use xmm8-15 and it just
>> tries to say LRA to assign them for non SSE4 instructions. I doubt it
>> would have some use for other targets than Silvermont.
>
> When it is just a hint, shouldn't there be something like Ya,???x
> or Ya,!x or similar in the SSE4 constraints? I mean, xmm{8-15} can be used,
> just is costly.
>
I made it Ya,?x. I do not know how many '?' would be reasonable here.
One was enough to get gain on tests where I expected it.
Ilya
> Jakub
