--- old/src/hotspot/cpu/arm/gc/shared/barrierSetAssembler_arm.cpp	2018-09-14 14:46:13.683823408 +0300
+++ new/src/hotspot/cpu/arm/gc/shared/barrierSetAssembler_arm.cpp	2018-09-14 14:46:13.487810979 +0300
@@ -24,6 +24,8 @@
 
 #include "precompiled.hpp"
 #include "gc/shared/barrierSetAssembler.hpp"
+#include "gc/shared/collectedHeap.hpp"
+#include "runtime/thread.hpp"
 
 #define __ masm->
 
@@ -166,3 +168,118 @@
                                      Register obj1, Register obj2) {
   __ cmp(obj1, obj2);
 }
+
+// Puts address of allocated object into register `obj` and end of allocated object into register `obj_end`.
+void BarrierSetAssembler::eden_allocate(MacroAssembler* masm, Register obj, Register obj_end, Register tmp1, Register tmp2,
+                                 RegisterOrConstant size_expression, Label& slow_case) {
+  if (!Universe::heap()->supports_inline_contig_alloc()) {
+    __ b(slow_case);
+    return;
+  }
+
+  CollectedHeap* ch = Universe::heap();
+
+  const Register top_addr = tmp1;
+  const Register heap_end = tmp2;
+
+  if (size_expression.is_register()) {
+    assert_different_registers(obj, obj_end, top_addr, heap_end, size_expression.as_register());
+  } else {
+    assert_different_registers(obj, obj_end, top_addr, heap_end);
+  }
+
+  bool load_const = AARCH64_ONLY(false) NOT_AARCH64(VM_Version::supports_movw() ); // TODO-AARCH64 check performance
+  if (load_const) {
+    __ mov_address(top_addr, (address)Universe::heap()->top_addr(), symbolic_Relocation::eden_top_reference);
+  } else {
+    __ ldr(top_addr, Address(Rthread, JavaThread::heap_top_addr_offset()));
+  }
+  // Calculate new heap_top by adding the size of the object
+  Label retry;
+  __ bind(retry);
+
+#ifdef AARCH64
+  __ ldxr(obj, top_addr);
+#else
+  __ ldr(obj, Address(top_addr));
+#endif // AARCH64
+
+  __ ldr(heap_end, Address(top_addr, (intptr_t)ch->end_addr() - (intptr_t)ch->top_addr()));
+  __ add_rc(obj_end, obj, size_expression);
+  // Check if obj_end wrapped around, i.e., obj_end < obj. If yes, jump to the slow case.
+  __ cmp(obj_end, obj);
+  __ b(slow_case, lo);
+  // Update heap_top if allocation succeeded
+  __ cmp(obj_end, heap_end);
+  __ b(slow_case, hi);
+
+#ifdef AARCH64
+  __ stxr(heap_end/*scratched*/, obj_end, top_addr);
+  __ cbnz_w(heap_end, retry);
+#else
+  __ atomic_cas_bool(obj, obj_end, top_addr, 0, heap_end/*scratched*/);
+  __ b(retry, ne);
+#endif // AARCH64
+
+  incr_allocated_bytes(masm, size_expression, tmp1);
+}
+
+// Puts address of allocated object into register `obj` and end of allocated object into register `obj_end`.
+void BarrierSetAssembler::tlab_allocate(MacroAssembler* masm, Register obj, Register obj_end, Register tmp1,
+                                 RegisterOrConstant size_expression, Label& slow_case) {
+  const Register tlab_end = tmp1;
+  assert_different_registers(obj, obj_end, tlab_end);
+
+  __ ldr(obj, Address(Rthread, JavaThread::tlab_top_offset()));
+  __ ldr(tlab_end, Address(Rthread, JavaThread::tlab_end_offset()));
+  __ add_rc(obj_end, obj, size_expression);
+  __ cmp(obj_end, tlab_end);
+  __ b(slow_case, hi);
+  __ str(obj_end, Address(Rthread, JavaThread::tlab_top_offset()));
+}
+
+void BarrierSetAssembler::incr_allocated_bytes(MacroAssembler* masm, RegisterOrConstant size_in_bytes, Register tmp) {
+#ifdef AARCH64
+  __ ldr(tmp, Address(Rthread, in_bytes(JavaThread::allocated_bytes_offset())));
+  __ add_rc(tmp, tmp, size_in_bytes);
+  __ str(tmp, Address(Rthread, in_bytes(JavaThread::allocated_bytes_offset())));
+#else
+  // Bump total bytes allocated by this thread
+  Label done;
+
+  // Borrow the Rthread for alloc counter
+  Register Ralloc = Rthread;
+  __ add(Ralloc, Ralloc, in_bytes(JavaThread::allocated_bytes_offset()));
+  __ ldr(tmp, Address(Ralloc));
+  __ adds(tmp, tmp, size_in_bytes);
+  __ str(tmp, Address(Ralloc), cc);
+  __ b(done, cc);
+
+  // Increment the high word and store single-copy atomically (that is an unlikely scenario on typical embedded systems as it means >4GB has been allocated)
+  // To do so ldrd/strd instructions used which require an even-odd pair of registers. Such a request could be difficult to satisfy by
+  // allocating those registers on a higher level, therefore the routine is ready to allocate a pair itself.
+  Register low, high;
+  // Select ether R0/R1 or R2/R3
+
+  if (size_in_bytes.is_register() && (size_in_bytes.as_register() == R0 || size_in_bytes.as_register() == R1)) {
+    low = R2;
+    high  = R3;
+  } else {
+    low = R0;
+    high  = R1;
+  }
+  __ push(RegisterSet(low, high));
+
+  __ ldrd(low, Address(Ralloc));
+  __ adds(low, low, size_in_bytes);
+  __ adc(high, high, 0);
+  __ strd(low, Address(Ralloc));
+
+  __ pop(RegisterSet(low, high));
+
+  __ bind(done);
+
+  // Unborrow the Rthread
+  __ sub(Rthread, Ralloc, in_bytes(JavaThread::allocated_bytes_offset()));
+#endif // AARCH64
+}