Request for reviews (XXL): 6893268: additional dynamic language related optimizations in C2

[Tom Rodriguez]

Ok.

tom

On Dec 3, 2009, at 3:16 AM, Christian Thalinger wrote:

> On Wed, 2009-12-02 at 15:19 -0800, Tom Rodriguez wrote:
>> doCall.cpp:
>> 
>> +       if (n->is_Con()) {
>> +         assert(n->Opcode() == Op_ConP, "sanity");
>> +         ConPNode* con = (ConPNode*) n;  // XXX Is there another way?
>> +         const Type* t = con->type();
>> +         const TypeOopPtr* oop_ptr = t->isa_oopptr();
>> 
>> You could also write this as:
>> 
>> if (n->Opcode() == Op_ConP) {
>>  const TypeOopPtr* oop_ptr = n->bottom_type()->is_oopptr();
>> 
>> is_Con() would return true for top if that somehow managed to sneak in there.
> 
> Ahh, OK.  Thanks for that.
> 
>> what this about?
>> 
>> +         // XXX The returned direct call to target_method is not a MH
>> +         // invoke and does not work.
>> +         //if (hit_cg != NULL)
>> +         if (hit_cg != NULL && hit_cg->is_inline())
>> +           return hit_cg;
>> 
>> could the commented piece simply be deleted or is there some lingering question here?
> 
> In the beginning I hoped that I can use the returned CallGenerator, even
> if it's not an inlined one, because it would be a direct call to the
> generated adapters.  But I think this would not work anyway.  I will
> remove that comment.
> 
>> otherwise this looks good.
> 
> Here is the updated webrev:
> 
> http://cr.openjdk.java.net/~twisti/6893268/webrev.03/
> 
> -- Christian
>