JSOP_NEWOBJECT should use the MacroAssembler's getNewObject() path where possible. In terms of benchmarks, this mostly affects v8-splay; 3d-cube; and a few kraken tests, especially crypto-ccm.
Ahem, that should have read JSOP_NEWARRAY.
Created attachment 612418 [details] [diff] [review] Inline allocation for JSOP_NEWARRAY. Brings splay 125ms -> 112ms. Combined with JSOP_NEWOBJECT patch, splay goes from 125ms -> 82ms. 3d-cube goes from ~35ms to ~32ms, not a significant change.