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

[Roland Westrelin]

Here is a new webrev that addresses your comments:

http://cr.openjdk.java.net/~roland/8185265/webrev.01/

I changed my mind wrt UseTLAB: I changed the code in macro.cpp so it
works with UseTLAB off.

> templateInterpreterGenerator_x86.cpp
> - As I understand, a pack/unpack handler is always set if the value type can be returned in registers. Why do we need 
> the pack_handler != 0 check? Isn't the __ testptr(rax, 1) sufficient?

Yes, you're right about that.

Roland.