RFR (S): 8014286: failed java/lang/Math/DivModTests.java after 6934604 changes

[Vladimir Kozlov]

Thanks, Christian

How about this?:

static bool is_init_with_ea(ciMethod* callee_method,
                             ciMethod* caller_method, Compile* C) {
   if (!C->do_escape_analysis() || !EliminateAllocations) {
     return false; // EA is off
   }
   if (callee_method->is_initializer()) {
     return true;
   }
   if (caller_method->is_initializer() &&
       caller_method != C->method() &&
       caller_method->holder()->is_subclass_of(callee_method->holder())) {
     return true; // super constructor is called from inlined constructor
   }
   if (C->eliminate_boxing() && callee_method->is_boxing_method()) {
     return true;
   }
   return false;
}

Vladimir

On 5/13/13 12:36 PM, Christian Thalinger wrote:
> Looks good.  Just one comment:  could we change is_init_with_ea to have some if statements returning true and false?  It has almost reached its maximum complexity.
>
> -- Chris
>
> On May 10, 2013, at 5:36 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
>
>> http://cr.openjdk.java.net/~kvn/8014286/webrev/
>>
>> Corrected escape state for the result of boxing method. Returned object escapes if it is loaded from boxing cache.
>> Also added force inlining executed boxing methods.
>>
>> Tested with jprt, compiler jtreg, jdk/testjava/lang/Math (-Xcomp) tests.
>>
>> Thanks,
>> Vladimir
>>
>>
>>
>>
>>
>>
>>
>