RFR(L): 8185265 [MVT] improve performance of return of value types with new calling convention

[Tobias Hartmann]

Hi Roland,

On 09.08.2017 13:48, Roland Westrelin wrote:
>> http://cr.openjdk.java.net/~roland/8185265/webrev.00/
> 
> The following patch on top of the webrev above is needed to not break
> the 32 bit build.

This looks reasonable.

Thanks,
Tobias

> --- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp
> +++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp
> @@ -3147,6 +3147,6 @@
>   }
>   
>   BufferedValueTypeBlob* SharedRuntime::generate_buffered_value_type_adapter(const ValueKlass* vk) {
> -  ShouldNotReachHere();
> +  Unimplemented();
>     return NULL;
>   }
> diff --git a/src/share/vm/oops/valueKlass.cpp b/src/share/vm/oops/valueKlass.cpp
> --- a/src/share/vm/oops/valueKlass.cpp
> +++ b/src/share/vm/oops/valueKlass.cpp
> @@ -344,6 +344,7 @@
>   }
>   
>   void ValueKlass::initialize_calling_convention() {
> +  if (ValueTypeReturnedAsFields || ValueTypePassFieldsAsArgs) {
>     Thread* THREAD = Thread::current();
>     assert(!HAS_PENDING_EXCEPTION, "should have no exception");
>     ResourceMark rm;
> @@ -355,6 +356,7 @@
>       extended_sig->at_put(i, sig_vk.at(i));
>     }
>   
> +    if (ValueTypeReturnedAsFields) {
>     BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, nb_fields);
>     sig_bt[0] = T_METADATA;
>     SigEntry::fill_sig_bt(sig_vk, sig_bt+1, nb_fields-1, true);
> @@ -373,6 +375,8 @@
>       *((address*)adr_unpack_handler()) = buffered_blob->unpack_fields();
>       assert(CodeCache::find_blob(pack_handler()) == buffered_blob, "lost track of blob");
>     }
> +    }
> +  }
>     assert(vtable_length() == 0, "a value klass with a vtable?");
>   }
>   
>