Object and array inlining

 The tradeoff between the benefits of inlining or not is demonstrated by the length of the time interval between the overwrite of the fields and the garbage collection which restores the optimized field order. If the interval is too long, the benefits fade away when a revert to non-inlined field access performs better as before. If the interval is short, then the restoration of optimized field order for subsequent access saves at least one instruction to load each time. 

The garbage collector is based on aging and compaction of heap allocations. The younger generation is collected frequently since they are not used on a prolonged basis.  The stop-and-copy collector copies young objects between two alternating spaces and it increments a field for age in each activity. When the age exceeds a threshold, the object is promoted to the older generation. If the objects don’t linger long, they don’t make it to the next generation and this works in the favor of the inlining. This is true for all new arrays and objects but in order for it to be true for parent arrays, the parent must also be in the younger generation. These collocated objects can be restored collectively in the young generation if there were a way to keep them all there and for this purpose the factors that determine the aging and the threshold can be tweaked in favor of inlining.

The check for array bounds can be eliminated in favor of optimized access if it can be guaranteed that all accesses based on the index of the array are safe. For example, if the check is in a loop and the loop is invariant, the check can be completely omitted. Similarly, if the array field accesses are known which explains all changes to the array, then the bounds check can be eliminated say outside the loop but it is just the opposite. The bounds check is used by the program to make modifications to the array in which case an optimized access is not practical.

Since the savings in the optimized code has been referred to as the number of instructions avoided, the performance of array inlining is directly proportional to the number of field loads eliminated. Against a comparison of baseline, array inlining, object inlining and dynamic array inlining, a count of the load of references against the same distribution in each case clearly explains the differences in performance between them.