JDK-8319773 updated the x86/x86_64 inlined hash code intrinsic for C1 (generate_native_wrapper) to be aware that the fast path (reading the hash code directly from the markWord) is possible on fast_locked objects when LM_LIGHTWEIGHT is used.

arm also has this optimisation in their generate_native_wrapper but it only takes the fast path when the object is unlocked regardless of locking mode.