Request for reviews (M): 7201026: add vector for shift count

[Vladimir Kozlov]

Thanks, Christian

I think the problem (at least for vectors) with usage of multiple-match-rules is 
it can't be used if mach nodes have different predicates.

Thanks,
Vladimir

Christian Thalinger wrote:
> On Oct 1, 2012, at 4:28 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
> 
>> I updated webrev with suggested changes:
>>
>> http://cr.openjdk.java.net/~kvn/7201026/webrev.01
> 
> Looks better.  Great we re-discovered the multiple-match-rules.  It's funny that you added it but we never used it.
> 
> -- Chris
> 
>> Thanks,
>> Vladimir
>>
>> Christian Thalinger wrote:
>>> On Sep 25, 2012, at 6:52 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
>>>> http://cr.openjdk.java.net/~kvn/7201026/webrev
>>>>
>>>> Add generation of vectors for scalar shift count (same for all shift vector's elements).
>>> src/cpu/sparc/vm/sparc.ad:
>>> + const int Matcher::vector_shift_count_ideal_reg(int size) {
>>> +   fatal("vector shift is not supported");
>>> +   return Op_RegD;
>>> + }
>>> It's just a nit but I think we should return -1 here.  Someone in the future might be confused.
>>> src/share/vm/opto/vectornode.cpp:
>>> +   ShouldNotReachHere();
>>> Replace with default case and fatal?
>>> Otherwise this looks good.
>>> -- Chris
>>>> Fixed Test7200264.sh test for RShiftVI case.
>>>>
>>>> Thanks,
>>>> Vladimir
>