--- old/src/cpu/x86/vm/c1_MacroAssembler_x86.cpp	2016-01-08 09:42:42.813566923 +0100
+++ new/src/cpu/x86/vm/c1_MacroAssembler_x86.cpp	2016-01-08 09:42:42.541566935 +0100
@@ -182,54 +182,13 @@
 
 // preserves obj, destroys len_in_bytes
 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
+  assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
   Label done;
-  assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different");
-  assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord");
-  Register index = len_in_bytes;
-  // index is positive and ptr sized
-  subptr(index, hdr_size_in_bytes);
-  jcc(Assembler::zero, done);
-  // initialize topmost word, divide index by 2, check if odd and test if zero
-  // note: for the remaining code to work, index must be a multiple of BytesPerWord
-#ifdef ASSERT
-  { Label L;
-    testptr(index, BytesPerWord - 1);
-    jcc(Assembler::zero, L);
-    stop("index is not a multiple of BytesPerWord");
-    bind(L);
-  }
-#endif
-  xorptr(t1, t1);    // use _zero reg to clear memory (shorter code)
-  if (UseIncDec) {
-    shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
-  } else {
-    shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
-    shrptr(index, 1);
-  }
-#ifndef _LP64
-  // index could have been not a multiple of 8 (i.e., bit 2 was set)
-  { Label even;
-    // note: if index was a multiple of 8, than it cannot
-    //       be 0 now otherwise it must have been 0 before
-    //       => if it is even, we don't need to check for 0 again
-    jcc(Assembler::carryClear, even);
-    // clear topmost word (no jump needed if conditional assignment would work here)
-    movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1);
-    // index could be 0 now, need to check again
-    jcc(Assembler::zero, done);
-    bind(even);
-  }
-#endif // !_LP64
-  // initialize remaining object fields: rdx is a multiple of 2 now
-  { Label loop;
-    bind(loop);
-    movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 1*BytesPerWord), t1);
-    NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 2*BytesPerWord), t1);)
-    decrement(index);
-    jcc(Assembler::notZero, loop);
-  }
 
-  // done
+  // len_in_bytes is positive and ptr sized
+  subptr(len_in_bytes, hdr_size_in_bytes);
+  jcc(Assembler::zero, done);
+  zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
   bind(done);
 }
 
@@ -241,47 +200,49 @@
 
   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
 
-  initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2);
+  initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
 }
 
-void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) {
+void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
          "con_size_in_bytes is not multiple of alignment");
   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
 
   initialize_header(obj, klass, noreg, t1, t2);
 
-  // clear rest of allocated space
-  const Register t1_zero = t1;
-  const Register index = t2;
-  const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
-  if (var_size_in_bytes != noreg) {
-    mov(index, var_size_in_bytes);
-    initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
-  } else if (con_size_in_bytes <= threshold) {
-    // use explicit null stores
-    // code size = 2 + 3*n bytes (n = number of fields to clear)
-    xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
-    for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
-      movptr(Address(obj, i), t1_zero);
-  } else if (con_size_in_bytes > hdr_size_in_bytes) {
-    // use loop to null out the fields
-    // code size = 16 bytes for even n (n = number of fields to clear)
-    // initialize last object field first if odd number of fields
-    xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
-    movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
-    // initialize last object field if constant size is odd
-    if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
-      movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
-    // initialize remaining object fields: rdx is a multiple of 2
-    { Label loop;
-      bind(loop);
-      movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
-             t1_zero);
-      NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
-             t1_zero);)
-      decrement(index);
-      jcc(Assembler::notZero, loop);
+  if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
+    // clear rest of allocated space
+    const Register t1_zero = t1;
+    const Register index = t2;
+    const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
+    if (var_size_in_bytes != noreg) {
+      mov(index, var_size_in_bytes);
+      initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
+    } else if (con_size_in_bytes <= threshold) {
+      // use explicit null stores
+      // code size = 2 + 3*n bytes (n = number of fields to clear)
+      xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
+      for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
+        movptr(Address(obj, i), t1_zero);
+    } else if (con_size_in_bytes > hdr_size_in_bytes) {
+      // use loop to null out the fields
+      // code size = 16 bytes for even n (n = number of fields to clear)
+      // initialize last object field first if odd number of fields
+      xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
+      movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
+      // initialize last object field if constant size is odd
+      if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
+        movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
+      // initialize remaining object fields: rdx is a multiple of 2
+      { Label loop;
+        bind(loop);
+        movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
+               t1_zero);
+        NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
+               t1_zero);)
+        decrement(index);
+        jcc(Assembler::notZero, loop);
+      }
     }
   }