Request for review 7178145: Change constMethodOop::_exception_table to optionally inlined u2 table

[keith mcguigan]

Hi Jiangli,

In methodOop.hpp, the class ExceptionTable holds a raw pointer into the 
heap (the pointer to the embedded table).  This is used in 
methodOop::fast_exception_handler_bci_for() in a loop when searching for 
the bci.  If class loading is initiated , a Full GC could occur which 
could invalidate the pointer in ExceptionTable on the next pass through 
the loop.  Need some sort of handle or recalculation in that loop. 
Perhaps a No_Safepoint_Verifier to catch any other places like this?

Also in ClassVerifier::verify_exception_handler_targets().

I didn't look to see if this was possible in generateOopMaps code too, 
but you might want to take a look at that too -- it's definitely looping 
with that raw pointer (I didn't follow through to see if it could 
safepoint).

Other than that one issue, I think this looks really good.

--
- Keith

On 6/19/2012 5:24 PM, Jiangli Zhou wrote:
> Hi,
>
> Please review the following webrev for method exception handler table
> memory reduction:
>
> http://javaweb.sfbay.sun.com/~jianzhou/webrev.exceptiontable/
>
> Per VM spec the method exception handler table is tuples of u2's, but in
> hotspot the data are stored as ints in constMethodOop::_exception_table
> (a typeArray). The change reduces the exception handler table memory
> usage by:
>
>   * Changing the exception handler table from int array to u2 array.
>   * Making the exception handler table as conditionally allocated
>     inlined table in constMethod, and eliminate the _exception_table
>     field. The inlined table is only allocated when the exception table
>     attribute exists for the method.
>   * Removing the overhead of the extra typeArray, the data is directly
>     stored in the inlined table when exception handler exists for a method.
>
> Tested with runthese, jprt and vm.quick.testlist. No noticeable
> performance degradation from specjvm98 and specjbb2005.
>
> ==============================================================================
> logs.exceptiontable_baseline.1:
> Benchmark Samples Mean Stdev Geomean Weight
> specjbb2005 8 56372.58 763.16
> specjvm98 8 534.50 19.50
> ==============================================================================
> logs.exceptiontable.1:
> Benchmark Samples Mean Stdev %Diff P Significant
> specjbb2005 8 56734.05 880.76 0.64 0.395 *
> specjvm98 8 531.05 21.00 -0.65 0.738 *
> ==============================================================================
>
> ==============================================================================
> logs.exceptiontable_baseline.2:
> Benchmark Samples Mean Stdev Geomean Weight
> specjvm98 8 527.63 17.10
> ==============================================================================
> logs.exceptiontable.2:
> Benchmark Samples Mean Stdev %Diff P Significant
> specjvm98 8 536.79 20.54 1.74 0.349 *
> ==============================================================================
>
> Thanks,
>
> Jiangli
>