C2: Pre/Main/Post for foreach leaves never-executed post loop (AArch64)
=======================================================================

In the foreach case, the disassembly shows:

On foreach case disassembly we see:
- a hot loop
- a post loop that checks uncommon cases and can finish the loop if any iterations remain

Two observations
- The induction variable (i) is shared between the hot loop and the post loop. This keeps it live-out (forces a spill/rename) and makes the hot loop slightly less efficient.
- In this example, the post loop is never executed; it is effectively dead code.

------------------------------------------------------------
// main hot counted loop
for (; i < size; i++) {
    bh.consume(a[i]);
}
------------------------------------------------------------
  ;; B12: #	out( B13 ) <- in( B13 ) top-of-loop Freq: 986885
  0x000000010cf0e8b0:   	mov	w13, w11            ;*getfield cursor {reexecute=0 rethrow=0 return_oop=0}
 ;; B13: #	out( B12 B14 ) <- in( B10 B12 ) Loop( B13-B12 inner main of N65) Freq: 986886
  0x000000010cf0e8b4:   	add	x11, x10, w13, sxtw #2
  0x000000010cf0e8b8:   	ldr	w14, [x11, #0x10]
  0x000000010cf0e8bc:   	lsl	x11, x14, #3        ;*invokestatic consumeCompiler {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e8c0:   	add	w11, w13, #0x1      ;*iadd {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e8c4:   	cmp	w11, w12
  0x000000010cf0e8c8:   	b.lt	#-0x18              ;*ifeq {reexecute=0 rethrow=0 return_oop=0}
------------------------------------------------------------
// post-loop
if (i < size) {
    if (i >= len)
        // B25/B26 → uncommon_trap("loop_limit_check");

    if (modCount != l->modCount)
        // B22 → uncommon_trap("predicate");

    // B17–B20 (inner post counted loop)
    do {
        bh.consume(a[i]);
        i++;
    } while (i < size);
}
------------------------------------------------------------
 ;; B14: #	out( B21 B15 ) <- in( B23 B13 )  Freq: 0,999991
  0x000000010cf0e8cc:   	cmp	w11, w16
  0x000000010cf0e8d0:   	b.ge	#0x48
  0x000000010cf0e8d4:   	cmp	w11, w29
  0x000000010cf0e8d8:   	b.hs	#0xc0
  0x000000010cf0e8dc:   	cmp	w11, w29
  0x000000010cf0e8e0:   	b.hs	#0xc8               ;*getfield cursor {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e8e4:   	cmp	w11, w16
  0x000000010cf0e8e8:   	b.ge	#0x110              ;*if_icmplt {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e8ec:   	add	x12, x10, w11, sxtw #2
  0x000000010cf0e8f0:   	ldr	w14, [x12, #0x10]
  0x000000010cf0e8f4:   	cmp	w11, w29
  0x000000010cf0e8f8:   	b.ge	#0xd0               ;*if_icmplt {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e8fc:   	lsl	x12, x14, #3        ;*aaload {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e900:   	add	w14, w11, #0x1      ;*invokestatic consumeCompiler {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e904:   	cmp	w14, w16
  0x000000010cf0e908:   	b.ge	#0x10               ;*ifeq {reexecute=0 rethrow=0 return_oop=0}
  0x000000010cf0e90c:   	mov	w13, w11
  0x000000010cf0e910:   	mov	w11, w14
  0x000000010cf0e914:   	b	#-0x30              ;*return {reexecute=0 rethrow=0 return_oop=0}
------------------------------------------------------------



Experiment in HotSpot: disable pre–main–post split
(e.g., force should_rce := false in iteration_split_impl)
====================================================================

Result: the post loop disappears, but the hot loop becomes less efficient (two per-iteration checks remain, blocking a clean counted loop and related optimizations):

  29.15%    0x000000011040c340:   	mov	w16, w14
   0.53%    0x000000011040c344:   	mov	w14, w13            ;*getfield cursor {reexecute=0 rethrow=0 return_oop=0}
   0.83%    0x000000011040c348:   	cmp	w14, w10
   0.65%    0x000000011040c34c:   	b.ge	#0x3c               ;*if_icmplt {reexecute=0 rethrow=0 return_oop=0}
  30.81%    0x000000011040c350:   	cmp	w14, w15
   0.41%    0x000000011040c354:   	b.ge	#0x68               ;*if_icmplt {reexecute=0 rethrow=0 return_oop=0}
            0x000000011040c358:   	add	x12, x11, w14, sxtw #2
  27.38%    0x000000011040c35c:   	ldr	w12, [x12, #0x10]
   0.41%    0x000000011040c360:   	add	w13, w14, #0x1      ;*iadd {reexecute=0 rethrow=0 return_oop=0}
   0.59%    0x000000011040c364:   	lsl	x12, x12, #3        ;*invokestatic consumeCompiler {reexecute=0 rethrow=0 return_oop=0}
   0.41%    0x000000011040c368:   	cmp	w13, w10
            0x000000011040c36c:   	b.lt	#-0x2c              ;*return {reexecute=0 rethrow=0 return_oop=0}