src/hotspot/cpu/x86/c1_MacroAssembler_x86.cpp

@@ -161,28 +161,32 @@
    }
  }
  
  
  void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
-   assert_different_registers(obj, klass, len);
-   movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
+   assert_different_registers(obj, klass, len, t1, t2);
+   if (UseCompactObjectHeaders) {
+     movptr(t1, Address(klass, Klass::prototype_header_offset()));
+     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
+   } else {
+     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
  #ifdef _LP64
-   if (UseCompressedClassPointers) { // Take care not to kill klass
-     movptr(t1, klass);
-     encode_klass_not_null(t1, rscratch1);
-     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
-   } else
+     if (UseCompressedClassPointers) { // Take care not to kill klass
+       movptr(t1, klass);
+       encode_klass_not_null(t1, rscratch1);
+       movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
+     } else
  #endif
-   {
-     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
+     {
+       movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
+     }
    }
- 
    if (len->is_valid()) {
      movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
    }
  #ifdef _LP64
-   else if (UseCompressedClassPointers) {
+   else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
      xorptr(t1, t1);
      store_klass_gap(obj, t1);
    }
  #endif
  }

@@ -220,34 +224,35 @@
    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)
+     int hdr_size_aligned = align_up(hdr_size_in_bytes, BytesPerWord); // klass gap is already cleared by init_header().
      if (var_size_in_bytes != noreg) {
        mov(index, var_size_in_bytes);
-       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
+       initialize_body(obj, index, hdr_size_aligned, 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)
+       for (int i = hdr_size_aligned; i < con_size_in_bytes; i += BytesPerWord)
          movptr(Address(obj, i), t1_zero);
-     } else if (con_size_in_bytes > hdr_size_in_bytes) {
+     } else if (con_size_in_bytes > hdr_size_aligned) {
        // 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);
+       movptr(index, (con_size_in_bytes - hdr_size_aligned) >> 3);
        // initialize last object field if constant size is odd
-       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
+       if (((con_size_in_bytes - hdr_size_aligned) & 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)),
+         movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (1*BytesPerWord)),
                 t1_zero);
-         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
+         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (2*BytesPerWord)),
                 t1_zero);)
          decrement(index);
          jcc(Assembler::notZero, loop);
        }
      }

@@ -259,11 +264,11 @@
    }
  
    verify_oop(obj);
  }
  
- void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
+ void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, Address::ScaleFactor f, Register klass, Label& slow_case) {
    assert(obj == rax, "obj must be in rax, for cmpxchg");
    assert_different_registers(obj, len, t1, t2, klass);
  
    // determine alignment mask
    assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");

@@ -272,21 +277,21 @@
    cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
    jcc(Assembler::above, slow_case);
  
    const Register arr_size = t2; // okay to be the same
    // align object end
-   movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
+   movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
    lea(arr_size, Address(arr_size, len, f));
    andptr(arr_size, ~MinObjAlignmentInBytesMask);
  
    try_allocate(obj, arr_size, 0, t1, t2, slow_case);
  
    initialize_header(obj, klass, len, t1, t2);
  
    // clear rest of allocated space
    const Register len_zero = len;
-   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
+   initialize_body(obj, arr_size, base_offset_in_bytes, len_zero);
  
    if (CURRENT_ENV->dtrace_alloc_probes()) {
      assert(obj == rax, "must be");
      call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
    }