6829192: JSR 292 needs to support 64-bit x86

Summary: changes for method handles and invokedynamic
Reviewed-by: kvn

src/cpu/x86/vm/assembler_x86.cpp

@@ -7666,7 +7666,7 @@ RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_ad
 
 #ifdef ASSERT
   Label L;
-  testl(tmp, tmp);
+  testptr(tmp, tmp);
   jccb(Assembler::notZero, L);
   hlt();
   bind(L);

src/cpu/x86/vm/interp_masm_x86_32.cpp

@@ -196,6 +196,9 @@ void InterpreterMacroAssembler::get_cache_index_at_bcp(Register reg, int bcp_off
   } else {
     assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic");
     movl(reg, Address(rsi, bcp_offset));
+    // Check if the secondary index definition is still ~x, otherwise
+    // we have to change the following assembler code to calculate the
+    // plain index.
     assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line");
     notl(reg);  // convert to plain index
   }

src/cpu/x86/vm/interp_masm_x86_64.cpp

@@ -185,12 +185,30 @@ void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
 }
 
 
+void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
+                                                       int bcp_offset,
+                                                       bool giant_index) {
+  assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
+  if (!giant_index) {
+    load_unsigned_short(index, Address(r13, bcp_offset));
+  } else {
+    assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic");
+    movl(index, Address(r13, bcp_offset));
+    // Check if the secondary index definition is still ~x, otherwise
+    // we have to change the following assembler code to calculate the
+    // plain index.
+    assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line");
+    notl(index);  // convert to plain index
+  }
+}
+
+
 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
                                                            Register index,
-                                                           int bcp_offset) {
-  assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
+                                                           int bcp_offset,
+                                                           bool giant_index) {
   assert(cache != index, "must use different registers");
-  load_unsigned_short(index, Address(r13, bcp_offset));
+  get_cache_index_at_bcp(index, bcp_offset, giant_index);
   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   // convert from field index to ConstantPoolCacheEntry index
@@ -200,10 +218,10 @@ void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 
 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
                                                                Register tmp,
-                                                               int bcp_offset) {
-  assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
+                                                               int bcp_offset,
+                                                               bool giant_index) {
   assert(cache != tmp, "must use different register");
-  load_unsigned_short(tmp, Address(r13, bcp_offset));
+  get_cache_index_at_bcp(tmp, bcp_offset, giant_index);
   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   // convert from field index to ConstantPoolCacheEntry index
   // and from word offset to byte offset
@@ -1236,7 +1254,8 @@ void InterpreterMacroAssembler::profile_final_call(Register mdp) {
 
 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
                                                      Register mdp,
-                                                     Register reg2) {
+                                                     Register reg2,
+                                                     bool receiver_can_be_null) {
   if (ProfileInterpreter) {
     Label profile_continue;
 
@@ -1246,8 +1265,15 @@ void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
     // We are making a call.  Increment the count.
     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
 
+    Label skip_receiver_profile;
+    if (receiver_can_be_null) {
+      testptr(receiver, receiver);
+      jcc(Assembler::zero, skip_receiver_profile);
+    }
+
     // Record the receiver type.
     record_klass_in_profile(receiver, mdp, reg2);
+    bind(skip_receiver_profile);
 
     // The method data pointer needs to be updated to reflect the new target.
     update_mdp_by_constant(mdp,

src/cpu/x86/vm/interp_masm_x86_64.hpp

@@ -95,9 +95,10 @@ class InterpreterMacroAssembler: public MacroAssembler {
 
   void get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset);
   void get_cache_and_index_at_bcp(Register cache, Register index,
-                                  int bcp_offset);
+                                  int bcp_offset, bool giant_index = false);
   void get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
-                                      int bcp_offset);
+                                      int bcp_offset, bool giant_index = false);
+  void get_cache_index_at_bcp(Register index, int bcp_offset, bool giant_index = false);
 
 
   void pop_ptr(Register r = rax);
@@ -236,7 +237,8 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void profile_call(Register mdp);
   void profile_final_call(Register mdp);
   void profile_virtual_call(Register receiver, Register mdp,
-                            Register scratch2);
+                            Register scratch2,
+                            bool receiver_can_be_null = false);
   void profile_ret(Register return_bci, Register mdp);
   void profile_null_seen(Register mdp);
   void profile_typecheck(Register mdp, Register klass, Register scratch);

src/cpu/x86/vm/interpreter_x86_64.cpp

@@ -277,12 +277,11 @@ address InterpreterGenerator::generate_abstract_entry(void) {
   address entry_point = __ pc();
 
   // abstract method entry
-  // remove return address. Not really needed, since exception
-  // handling throws away expression stack
-  __ pop(rbx);
 
-  // adjust stack to what a normal return would do
-  __ mov(rsp, r13);
+  //  pop return address, reset last_sp to NULL
+  __ empty_expression_stack();
+  __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
+  __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
 
   // throw exception
   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
@@ -300,7 +299,10 @@ address InterpreterGenerator::generate_method_handle_entry(void) {
   if (!EnableMethodHandles) {
     return generate_abstract_entry();
   }
-  return generate_abstract_entry(); //6815692//
+
+  address entry_point = MethodHandles::generate_method_handle_interpreter_entry(_masm);
+
+  return entry_point;
 }
 
 

src/cpu/x86/vm/methodHandles_x86.cpp

@@ -448,7 +448,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
                                 rbx_index, Address::times_ptr,
                                 base + vtableEntry::method_offset_in_bytes());
       Register rbx_method = rbx_temp;
-      __ movl(rbx_method, vtable_entry_addr);
+      __ movptr(rbx_method, vtable_entry_addr);
 
       __ verify_oop(rbx_method);
       __ jmp(rbx_method_fie);

src/cpu/x86/vm/stubGenerator_x86_64.cpp

@@ -2935,6 +2935,16 @@ class StubGenerator: public StubCodeGenerator {
 
     // arraycopy stubs used by compilers
     generate_arraycopy_stubs();
+
+    // generic method handle stubs
+    if (EnableMethodHandles && SystemDictionary::MethodHandle_klass() != NULL) {
+      for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST;
+           ek < MethodHandles::_EK_LIMIT;
+           ek = MethodHandles::EntryKind(1 + (int)ek)) {
+        StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
+        MethodHandles::generate_method_handle_stub(_masm, ek);
+      }
+    }
   }
 
  public:

src/cpu/x86/vm/templateInterpreter_x86_64.cpp

@@ -100,21 +100,26 @@ address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
   return entry;
 }
 
-// Arguments are: required type in rarg1, failing object (or NULL) in rarg2
+// Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4.
 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
   address entry = __ pc();
 
   __ pop(c_rarg2);              // failing object is at TOS
   __ pop(c_rarg1);              // required type is at TOS+8
 
-  // expression stack must be empty before entering the VM if an
-  // exception happened
+  __ verify_oop(c_rarg1);
+  __ verify_oop(c_rarg2);
+
+  // Various method handle types use interpreter registers as temps.
+  __ restore_bcp();
+  __ restore_locals();
+
+  // Expression stack must be empty before entering the VM for an exception.
   __ empty_expression_stack();
 
   __ call_VM(noreg,
              CAST_FROM_FN_PTR(address,
-                              InterpreterRuntime::
-                              throw_WrongMethodTypeException),
+                              InterpreterRuntime::throw_WrongMethodTypeException),
              // pass required type, failing object (or NULL)
              c_rarg1, c_rarg2);
   return entry;
@@ -182,15 +187,29 @@ address TemplateInterpreterGenerator::generate_return_entry_for(TosState state,
   __ restore_bcp();
   __ restore_locals();
 
-  __ get_cache_and_index_at_bcp(rbx, rcx, 1);
+  Label L_got_cache, L_giant_index;
+  if (EnableInvokeDynamic) {
+    __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
+    __ jcc(Assembler::equal, L_giant_index);
+  }
+  __ get_cache_and_index_at_bcp(rbx, rcx, 1, false);
+  __ bind(L_got_cache);
   __ movl(rbx, Address(rbx, rcx,
-                       Address::times_8,
+                       Address::times_ptr,
                        in_bytes(constantPoolCacheOopDesc::base_offset()) +
                        3 * wordSize));
   __ andl(rbx, 0xFF);
   if (TaggedStackInterpreter) __ shll(rbx, 1); // 2 slots per parameter.
   __ lea(rsp, Address(rsp, rbx, Address::times_8));
   __ dispatch_next(state, step);
+
+  // out of the main line of code...
+  if (EnableInvokeDynamic) {
+    __ bind(L_giant_index);
+    __ get_cache_and_index_at_bcp(rbx, rcx, 1, true);
+    __ jmp(L_got_cache);
+  }
+
   return entry;
 }
 

src/cpu/x86/vm/templateTable_x86_32.cpp

@@ -3146,7 +3146,6 @@ void TemplateTable::invokedynamic(int byte_no) {
     __ profile_call(rsi);
   }
 
-  Label handle_unlinked_site;
   __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx)));
   __ null_check(rcx);
   __ prepare_to_jump_from_interpreted();

src/cpu/x86/vm/templateTable_x86_64.cpp

@@ -203,18 +203,15 @@ void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
     __ jcc(Assembler::notEqual, fast_patch);
     __ get_method(scratch);
     // Let breakpoint table handling rewrite to quicker bytecode
-    __ call_VM(noreg,
-               CAST_FROM_FN_PTR(address,
-                                InterpreterRuntime::set_original_bytecode_at),
-               scratch, r13, bc);
+    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, r13, bc);
 #ifndef ASSERT
     __ jmpb(patch_done);
-    __ bind(fast_patch);
-  }
 #else
     __ jmp(patch_done);
+#endif
     __ bind(fast_patch);
   }
+#ifdef ASSERT
   Label okay;
   __ load_unsigned_byte(scratch, at_bcp(0));
   __ cmpl(scratch, (int) Bytecodes::java_code(bytecode));
@@ -2054,26 +2051,28 @@ void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
   }
 }
 
-void TemplateTable::resolve_cache_and_index(int byte_no,
-                                            Register Rcache,
-                                            Register index) {
+void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) {
   assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
+  bool is_invokedynamic = (bytecode() == Bytecodes::_invokedynamic);
 
   const Register temp = rbx;
   assert_different_registers(Rcache, index, temp);
 
   const int shift_count = (1 + byte_no) * BitsPerByte;
   Label resolved;
-  __ get_cache_and_index_at_bcp(Rcache, index, 1);
-  __ movl(temp, Address(Rcache,
-                        index, Address::times_8,
-                        constantPoolCacheOopDesc::base_offset() +
-                        ConstantPoolCacheEntry::indices_offset()));
-  __ shrl(temp, shift_count);
-  // have we resolved this bytecode?
-  __ andl(temp, 0xFF);
-  __ cmpl(temp, (int) bytecode());
-  __ jcc(Assembler::equal, resolved);
+  __ get_cache_and_index_at_bcp(Rcache, index, 1, is_invokedynamic);
+  if (is_invokedynamic) {
+    // we are resolved if the f1 field contains a non-null CallSite object
+    __ cmpptr(Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()), (int32_t) NULL_WORD);
+    __ jcc(Assembler::notEqual, resolved);
+  } else {
+    __ movl(temp, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
+    __ shrl(temp, shift_count);
+    // have we resolved this bytecode?
+    __ andl(temp, 0xFF);
+    __ cmpl(temp, (int) bytecode());
+    __ jcc(Assembler::equal, resolved);
+  }
 
   // resolve first time through
   address entry;
@@ -2090,6 +2089,9 @@ void TemplateTable::resolve_cache_and_index(int byte_no,
   case Bytecodes::_invokeinterface:
     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
     break;
+  case Bytecodes::_invokedynamic:
+    entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
+    break;
   default:
     ShouldNotReachHere();
     break;
@@ -2098,7 +2100,7 @@ void TemplateTable::resolve_cache_and_index(int byte_no,
   __ call_VM(noreg, entry, temp);
 
   // Update registers with resolved info
-  __ get_cache_and_index_at_bcp(Rcache, index, 1);
+  __ get_cache_and_index_at_bcp(Rcache, index, 1, is_invokedynamic);
   __ bind(resolved);
 }
 
@@ -2832,15 +2834,14 @@ void TemplateTable::count_calls(Register method, Register temp) {
   ShouldNotReachHere();
 }
 
-void TemplateTable::prepare_invoke(Register method,
-                                   Register index,
-                                   int byte_no,
-                                   Bytecodes::Code code) {
+void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
   // determine flags
+  Bytecodes::Code code = bytecode();
   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
+  const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
-  const bool load_receiver       = code != Bytecodes::_invokestatic;
+  const bool load_receiver      = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
   const bool receiver_null_check = is_invokespecial;
   const bool save_flags = is_invokeinterface || is_invokevirtual;
   // setup registers & access constant pool cache
@@ -2858,9 +2859,13 @@ void TemplateTable::prepare_invoke(Register method,
     __ movl(recv, flags);
     __ andl(recv, 0xFF);
     if (TaggedStackInterpreter) __ shll(recv, 1);  // index*2
-    __ movptr(recv, Address(rsp, recv, Address::times_8,
-                                 -Interpreter::expr_offset_in_bytes(1)));
-    __ verify_oop(recv);
+    Address recv_addr(rsp, recv, Address::times_8, -Interpreter::expr_offset_in_bytes(1));
+    if (is_invokedynamic) {
+      __ lea(recv, recv_addr);
+    } else {
+      __ movptr(recv, recv_addr);
+      __ verify_oop(recv);
+    }
   }
 
   // do null check if needed
@@ -2878,10 +2883,14 @@ void TemplateTable::prepare_invoke(Register method,
   ConstantPoolCacheEntry::verify_tosBits();
   // load return address
   {
-    ExternalAddress return_5((address)Interpreter::return_5_addrs_by_index_table());
-    ExternalAddress return_3((address)Interpreter::return_3_addrs_by_index_table());
-    __ lea(rscratch1, (is_invokeinterface ? return_5 : return_3));
-    __ movptr(flags, Address(rscratch1, flags, Address::times_8));
+    address table_addr;
+    if (is_invokeinterface || is_invokedynamic)
+      table_addr = (address)Interpreter::return_5_addrs_by_index_table();
+    else
+      table_addr = (address)Interpreter::return_3_addrs_by_index_table();
+    ExternalAddress table(table_addr);
+    __ lea(rscratch1, table);
+    __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
   }
 
   // push return address
@@ -2947,7 +2956,7 @@ void TemplateTable::invokevirtual_helper(Register index,
 
 void TemplateTable::invokevirtual(int byte_no) {
   transition(vtos, vtos);
-  prepare_invoke(rbx, noreg, byte_no, bytecode());
+  prepare_invoke(rbx, noreg, byte_no);
 
   // rbx: index
   // rcx: receiver
@@ -2959,7 +2968,7 @@ void TemplateTable::invokevirtual(int byte_no) {
 
 void TemplateTable::invokespecial(int byte_no) {
   transition(vtos, vtos);
-  prepare_invoke(rbx, noreg, byte_no, bytecode());
+  prepare_invoke(rbx, noreg, byte_no);
   // do the call
   __ verify_oop(rbx);
   __ profile_call(rax);
@@ -2969,7 +2978,7 @@ void TemplateTable::invokespecial(int byte_no) {
 
 void TemplateTable::invokestatic(int byte_no) {
   transition(vtos, vtos);
-  prepare_invoke(rbx, noreg, byte_no, bytecode());
+  prepare_invoke(rbx, noreg, byte_no);
   // do the call
   __ verify_oop(rbx);
   __ profile_call(rax);
@@ -2983,7 +2992,7 @@ void TemplateTable::fast_invokevfinal(int byte_no) {
 
 void TemplateTable::invokeinterface(int byte_no) {
   transition(vtos, vtos);
-  prepare_invoke(rax, rbx, byte_no, bytecode());
+  prepare_invoke(rax, rbx, byte_no);
 
   // rax: Interface
   // rbx: index
@@ -3072,7 +3081,24 @@ void TemplateTable::invokedynamic(int byte_no) {
     return;
   }
 
-  __ stop("invokedynamic NYI");//6815692//
+  prepare_invoke(rax, rbx, byte_no);
+
+  // rax: CallSite object (f1)
+  // rbx: unused (f2)
+  // rcx: receiver address
+  // rdx: flags (unused)
+
+  if (ProfileInterpreter) {
+    Label L;
+    // %%% should make a type profile for any invokedynamic that takes a ref argument
+    // profile this call
+    __ profile_call(r13);
+  }
+
+  __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx)));
+  __ null_check(rcx);
+  __ prepare_to_jump_from_interpreted();
+  __ jump_to_method_handle_entry(rcx, rdx);
 }
 
 

src/cpu/x86/vm/templateTable_x86_64.hpp

@@ -22,8 +22,7 @@
  *
  */
 
-  static void prepare_invoke(Register method, Register index, int byte_no,
-                             Bytecodes::Code code);
+  static void prepare_invoke(Register method, Register index, int byte_no);
   static void invokevirtual_helper(Register index, Register recv,
                                    Register flags);
   static void volatile_barrier(Assembler::Membar_mask_bits order_constraint);

src/share/vm/classfile/classFileParser.cpp

@@ -2511,23 +2511,12 @@ void ClassFileParser::java_dyn_MethodHandle_fix_pre(constantPoolHandle cp,
       fac_ptr->nonstatic_byte_count -= 1;
       (*fields_ptr)->ushort_at_put(i + instanceKlass::signature_index_offset,
                                    word_sig_index);
-      if (wordSize == jintSize) {
-        fac_ptr->nonstatic_word_count += 1;
-      } else {
-        fac_ptr->nonstatic_double_count += 1;
-      }
+      fac_ptr->nonstatic_word_count += 1;
 
-      FieldAllocationType atype = (FieldAllocationType) (*fields_ptr)->ushort_at(i+4);
+      FieldAllocationType atype = (FieldAllocationType) (*fields_ptr)->ushort_at(i + instanceKlass::low_offset);
       assert(atype == NONSTATIC_BYTE, "");
       FieldAllocationType new_atype = NONSTATIC_WORD;
-      if (wordSize > jintSize) {
-        if (Universe::field_type_should_be_aligned(T_LONG)) {
-          atype = NONSTATIC_ALIGNED_DOUBLE;
-        } else {
-          atype = NONSTATIC_DOUBLE;
-        }
-      }
-      (*fields_ptr)->ushort_at_put(i+4, new_atype);
+      (*fields_ptr)->ushort_at_put(i + instanceKlass::low_offset, new_atype);
 
       found_vmentry = true;
       break;
@@ -3085,7 +3074,7 @@ instanceKlassHandle ClassFileParser::parseClassFile(symbolHandle name,
     int len = fields->length();
     for (int i = 0; i < len; i += instanceKlass::next_offset) {
       int real_offset;
-      FieldAllocationType atype = (FieldAllocationType) fields->ushort_at(i+4);
+      FieldAllocationType atype = (FieldAllocationType) fields->ushort_at(i + instanceKlass::low_offset);
       switch (atype) {
         case STATIC_OOP:
           real_offset = next_static_oop_offset;
@@ -3173,8 +3162,8 @@ instanceKlassHandle ClassFileParser::parseClassFile(symbolHandle name,
         default:
           ShouldNotReachHere();
       }
-      fields->short_at_put(i+4, extract_low_short_from_int(real_offset) );
-      fields->short_at_put(i+5, extract_high_short_from_int(real_offset) );
+      fields->short_at_put(i + instanceKlass::low_offset,  extract_low_short_from_int(real_offset));
+      fields->short_at_put(i + instanceKlass::high_offset, extract_high_short_from_int(real_offset));
     }
 
     // Size of instances