Request for reviews(M): 7193318: C2: remove number of inputs requirement from Node's new operator

[Vladimir Kozlov]

 > Did you find any slowdown in compilation speed (with PrintCompilation2)?

PrintCompilation2 is difficult to compare. I would build "optimized" version and 
run with TimeCompiler. Or use -XX:+LogCompilation and then run our 
src/share/tools/LogCompilation tool with -S flag.

Vladimir

Christian Thalinger wrote:
> On Sep 24, 2012, at 11:09 AM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
> 
>> I regenerated webrev using Tom's webrev script so that udiffs are more clear:
>>
>> http://cr.openjdk.java.net/~kvn/7193318/webrev/
>>
>> Typo "necassary":
>>
>> +   // Allocate memory for the necassary number of edges.
>>
>> We do such alignment for sparc only. See comment in allocation.hpp. Please, change question to statement:
>>
>> +     // Do we really need to allocate space for _in array to have
>> +     // double alignment? May be we do since we are casting it to be
>> +     // Node**.
>>
>> Otherwise looks good.
>>
>> thanks,
>> Vladimir
>>
>> Bharadwaj Yadavalli wrote:
>>> De-couple memory allocation for edges from memory allocation for Node object.
>>> This allows use of placement new operator of Node, viz., new(size_t, Compile *)
>>> instead of node(size_t, Compile *, int) thereby eliminating the need to specify
>>> the number of edges in the new operator.
>>> Deleted placement new operator of Node - node(size_t, Compile *, int).
>>> Testing done: jtreg, and JPRT
>>> Benchmarking: refworkload.
> 
> Did you find any slowdown in compilation speed (with PrintCompilation2)?
> 
> -- Chris
> 
>>> Thanks,
>>> Bharadwaj
>