8132051: Better byte behavior

Co-authored-by: Roland Westerlin <roland.westerlin@oracle.com>
Co-authored-by: Vladimir Kozlov <vladimir.kozlov@oracle.com>
Co-authored-by: John Rose <john.r.rose@oracle.com>
Reviewed-by: bdelsart, roland, kvn, jrose, ahgross

hotspot/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp

@@ -344,7 +344,7 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) {
     length.set_instruction(x->length());
     length.load_item();
   }
-  if (needs_store_check) {
+  if (needs_store_check || x->check_boolean()) {
     value.load_item();
   } else {
     value.load_for_store(x->elt_type());
@@ -389,7 +389,8 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) {
     pre_barrier(LIR_OprFact::address(array_addr), LIR_OprFact::illegalOpr /* pre_val */,
                 true /* do_load */, false /* patch */, NULL);
   }
-  __ move(value.result(), array_addr, null_check_info);
+  LIR_Opr result = maybe_mask_boolean(x, array.result(), value.result(), null_check_info);
+  __ move(result, array_addr, null_check_info);
   if (obj_store) {
     // Precise card mark
     post_barrier(LIR_OprFact::address(array_addr), value.result());

hotspot/src/cpu/sparc/vm/interp_masm_sparc.cpp

@@ -208,6 +208,7 @@ void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
   case atos: ld_ptr(oop_addr, Otos_l);
              st_ptr(G0, oop_addr);                        break;
   case btos:                                           // fall through
+  case ztos:                                           // fall through
   case ctos:                                           // fall through
   case stos:                                           // fall through
   case itos: ld(val_addr, Otos_l1);                       break;
@@ -452,9 +453,10 @@ void InterpreterMacroAssembler::push(TosState state) {
   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
   switch (state) {
     case atos: push_ptr();            break;
-    case btos: push_i();              break;
+    case btos:                        // fall through
-    case ctos:
+    case ztos:                        // fall through
-    case stos: push_i();              break;
+    case ctos:                        // fall through
+    case stos:                        // fall through
     case itos: push_i();              break;
     case ltos: push_l();              break;
     case ftos: push_f();              break;
@@ -468,9 +470,10 @@ void InterpreterMacroAssembler::push(TosState state) {
 void InterpreterMacroAssembler::pop(TosState state) {
   switch (state) {
     case atos: pop_ptr();            break;
-    case btos: pop_i();              break;
+    case btos:                       // fall through
-    case ctos:
+    case ztos:                       // fall through
-    case stos: pop_i();              break;
+    case ctos:                       // fall through
+    case stos:                       // fall through
     case itos: pop_i();              break;
     case ltos: pop_l();              break;
     case ftos: pop_f();              break;
@@ -1103,6 +1106,49 @@ void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 }
 
+void InterpreterMacroAssembler::narrow(Register result) {
+
+  ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch);
+  ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch);
+
+  Label notBool, notByte, notChar, done;
+
+  // common case first
+  cmp(G3_scratch, T_INT);
+  br(Assembler::equal, true, pn, done);
+  delayed()->nop();
+
+  cmp(G3_scratch, T_BOOLEAN);
+  br(Assembler::notEqual, true, pn, notBool);
+  delayed()->cmp(G3_scratch, T_BYTE);
+  and3(result, 1, result);
+  ba(done);
+  delayed()->nop();
+
+  bind(notBool);
+  // cmp(G3_scratch, T_BYTE);
+  br(Assembler::notEqual, true, pn, notByte);
+  delayed()->cmp(G3_scratch, T_CHAR);
+  sll(result, 24, result);
+  sra(result, 24, result);
+  ba(done);
+  delayed()->nop();
+
+  bind(notByte);
+  // cmp(G3_scratch, T_CHAR);
+  sll(result, 16, result);
+  br(Assembler::notEqual, true, pn, done);
+  delayed()->sra(result, 16, result);
+  // sll(result, 16, result);
+  srl(result, 16, result);
+
+  // bind(notChar);
+  // must be short, instructions already executed in delay slot
+  // sll(result, 16, result);
+  // sra(result, 16, result);
+
+  bind(done);
+}
 
 // remove activation
 //
@@ -1151,6 +1197,7 @@ void InterpreterMacroAssembler::remove_activation(TosState state,
   case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through  // O1 -> I1
 #endif
   case btos:                                      // fall through
+  case ztos:                                      // fall through
   case ctos:
   case stos:                                      // fall through
   case atos:                                      // fall through

hotspot/src/cpu/sparc/vm/interp_masm_sparc.hpp

@@ -103,6 +103,8 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void dispatch_via (TosState state, address* table);
 
 
+  void narrow(Register result);
+
   // Removes the current activation (incl. unlocking of monitors).
   // Additionally this code is used for earlyReturn in which case we
   // want to skip throwing an exception and installing an exception.

hotspot/src/cpu/sparc/vm/templateTable_sparc.cpp

@@ -165,6 +165,7 @@ void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
   switch (bc) {
   case Bytecodes::_fast_aputfield:
   case Bytecodes::_fast_bputfield:
+  case Bytecodes::_fast_zputfield:
   case Bytecodes::_fast_cputfield:
   case Bytecodes::_fast_dputfield:
   case Bytecodes::_fast_fputfield:
@@ -922,8 +923,20 @@ void TemplateTable::bastore() {
   transition(itos, vtos);
   __ pop_i(O2); // index
   // Otos_i: val
+  // O2: index
   // O3: array
   __ index_check(O3, O2, 0, G3_scratch, O2);
+  // Need to check whether array is boolean or byte
+  // since both types share the bastore bytecode.
+  __ load_klass(O3, G4_scratch);
+  __ ld(G4_scratch, in_bytes(Klass::layout_helper_offset()), G4_scratch);
+  __ set(Klass::layout_helper_boolean_diffbit(), G3_scratch);
+  __ andcc(G3_scratch, G4_scratch, G0);
+  Label L_skip;
+  __ br(Assembler::zero, false, Assembler::pn, L_skip);
+  __ delayed()->nop();
+  __ and3(Otos_i, 1, Otos_i);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
+  __ bind(L_skip);
   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
 }
 
@@ -2008,6 +2021,12 @@ void TemplateTable::_return(TosState state) {
     __ bind(skip_register_finalizer);
   }
 
+  // Narrow result if state is itos but result type is smaller.
+  // Need to narrow in the return bytecode rather than in generate_return_entry
+  // since compiled code callers expect the result to already be narrowed.
+  if (state == itos) {
+    __ narrow(Otos_i);
+  }
   __ remove_activation(state, /* throw_monitor_exception */ true);
 
   // The caller's SP was adjusted upon method entry to accomodate
@@ -2218,7 +2237,7 @@ void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteContr
   Label checkVolatile;
 
   // compute field type
-  Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
+  Label notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
   // Make sure we don't need to mask Rflags after the above shift
   ConstantPoolCacheEntry::verify_tos_state_shift();
@@ -2273,7 +2292,7 @@ void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteContr
 
   // cmp(Rflags, btos);
   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
-  __ delayed() ->cmp(Rflags, ctos);
+  __ delayed() ->cmp(Rflags, ztos);
 
   // btos
   __ ldsb(Rclass, Roffset, Otos_i);
@@ -2286,6 +2305,22 @@ void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteContr
 
   __ bind(notByte);
 
+  // cmp(Rflags, ztos);
+  __ br(Assembler::notEqual, false, Assembler::pt, notBool);
+  __ delayed() ->cmp(Rflags, ctos);
+
+  // ztos
+  __ ldsb(Rclass, Roffset, Otos_i);
+  __ push(itos);
+  if (!is_static && rc == may_rewrite) {
+    // use btos rewriting, no truncating to t/f bit is needed for getfield.
+    patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
+  }
+  __ ba(checkVolatile);
+  __ delayed()->tst(Lscratch);
+
+  __ bind(notBool);
+
   // cmp(Rflags, ctos);
   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
   __ delayed() ->cmp(Rflags, stos);
@@ -2449,6 +2484,7 @@ void TemplateTable::jvmti_post_fast_field_mod() {
     switch (bytecode()) {  // save tos values before call_VM() clobbers them
     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
     case Bytecodes::_fast_bputfield: // fall through
+    case Bytecodes::_fast_zputfield: // fall through
     case Bytecodes::_fast_sputfield: // fall through
     case Bytecodes::_fast_cputfield: // fall through
     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
@@ -2466,6 +2502,7 @@ void TemplateTable::jvmti_post_fast_field_mod() {
     switch (bytecode()) {             // restore tos values
     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
     case Bytecodes::_fast_bputfield: // fall through
+    case Bytecodes::_fast_zputfield: // fall through
     case Bytecodes::_fast_sputfield: // fall through
     case Bytecodes::_fast_cputfield: // fall through
     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
@@ -2581,7 +2618,7 @@ void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteContr
   ConstantPoolCacheEntry::verify_tos_state_shift();
 
   // compute field type
-  Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
+  Label notInt, notShort, notChar, notObj, notByte, notBool, notLong, notFloat;
 
   if (is_static) {
     // putstatic with object type most likely, check that first
@@ -2649,7 +2686,7 @@ void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteContr
 
   // cmp(Rflags, btos);
   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
-  __ delayed()->cmp(Rflags, ltos);
+  __ delayed()->cmp(Rflags, ztos);
 
   // btos
   {
@@ -2664,6 +2701,25 @@ void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteContr
   }
 
   __ bind(notByte);
+
+  // cmp(Rflags, btos);
+  __ br(Assembler::notEqual, false, Assembler::pt, notBool);
+  __ delayed()->cmp(Rflags, ltos);
+
+  // ztos
+  {
+    __ pop_i();
+    if (!is_static) pop_and_check_object(Rclass);
+    __ and3(Otos_i, 1, Otos_i);
+    __ stb(Otos_i, Rclass, Roffset);
+    if (!is_static && rc == may_rewrite) {
+      patch_bytecode(Bytecodes::_fast_zputfield, G3_scratch, G4_scratch, true, byte_no);
+    }
+    __ ba(checkVolatile);
+    __ delayed()->tst(Lscratch);
+  }
+
+  __ bind(notBool);
   // cmp(Rflags, ltos);
   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
   __ delayed()->cmp(Rflags, ctos);
@@ -2787,6 +2843,7 @@ void TemplateTable::fast_storefield(TosState state) {
   pop_and_check_object(Rclass);
 
   switch (bytecode()) {
+    case Bytecodes::_fast_zputfield: __ and3(Otos_i, 1, Otos_i);  // fall through to bputfield
     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
     case Bytecodes::_fast_cputfield: /* fall through */
     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;

hotspot/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp

@@ -284,7 +284,7 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) {
     length.load_item();
 
   }
-  if (needs_store_check) {
+  if (needs_store_check || x->check_boolean()) {
     value.load_item();
   } else {
     value.load_for_store(x->elt_type());
@@ -332,7 +332,8 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) {
     // Seems to be a precise
     post_barrier(LIR_OprFact::address(array_addr), value.result());
   } else {
-    __ move(value.result(), array_addr, null_check_info);
+    LIR_Opr result = maybe_mask_boolean(x, array.result(), value.result(), null_check_info);
+    __ move(result, array_addr, null_check_info);
   }
 }
 

hotspot/src/cpu/x86/vm/interp_masm_x86.cpp

@@ -349,6 +349,7 @@ void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
                verify_oop(rax, state);              break;
     case ltos: movptr(rax, val_addr);                 break;
     case btos:                                   // fall through
+    case ztos:                                   // fall through
     case ctos:                                   // fall through
     case stos:                                   // fall through
     case itos: movl(rax, val_addr);                 break;
@@ -370,6 +371,7 @@ void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
     case ltos:
                movl(rdx, val_addr1);               // fall through
     case btos:                                     // fall through
+    case ztos:                                     // fall through
     case ctos:                                     // fall through
     case stos:                                     // fall through
     case itos: movl(rax, val_addr);                   break;
@@ -616,6 +618,7 @@ void InterpreterMacroAssembler::pop(TosState state) {
   switch (state) {
   case atos: pop_ptr();                 break;
   case btos:
+  case ztos:
   case ctos:
   case stos:
   case itos: pop_i();                   break;
@@ -633,6 +636,7 @@ void InterpreterMacroAssembler::push(TosState state) {
   switch (state) {
   case atos: push_ptr();                break;
   case btos:
+  case ztos:
   case ctos:
   case stos:
   case itos: push_i();                  break;
@@ -668,6 +672,7 @@ void InterpreterMacroAssembler::pop(TosState state) {
   switch (state) {
     case atos: pop_ptr(rax);                                 break;
     case btos:                                               // fall through
+    case ztos:                                               // fall through
     case ctos:                                               // fall through
     case stos:                                               // fall through
     case itos: pop_i(rax);                                   break;
@@ -716,6 +721,7 @@ void InterpreterMacroAssembler::push(TosState state) {
   switch (state) {
     case atos: push_ptr(rax); break;
     case btos:                                               // fall through
+    case ztos:                                               // fall through
     case ctos:                                               // fall through
     case stos:                                               // fall through
     case itos: push_i(rax);                                    break;
@@ -849,6 +855,51 @@ void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
   dispatch_base(state, table);
 }
 
+void InterpreterMacroAssembler::narrow(Register result) {
+
+  // Get method->_constMethod->_result_type
+  movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
+  movptr(rcx, Address(rcx, Method::const_offset()));
+  load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
+
+  Label done, notBool, notByte, notChar;
+
+  // common case first
+  cmpl(rcx, T_INT);
+  jcc(Assembler::equal, done);
+
+  // mask integer result to narrower return type.
+  cmpl(rcx, T_BOOLEAN);
+  jcc(Assembler::notEqual, notBool);
+  andl(result, 0x1);
+  jmp(done);
+
+  bind(notBool);
+  cmpl(rcx, T_BYTE);
+  jcc(Assembler::notEqual, notByte);
+  LP64_ONLY(movsbl(result, result);)
+  NOT_LP64(shll(result, 24);)      // truncate upper 24 bits
+  NOT_LP64(sarl(result, 24);)      // and sign-extend byte
+  jmp(done);
+
+  bind(notByte);
+  cmpl(rcx, T_CHAR);
+  jcc(Assembler::notEqual, notChar);
+  LP64_ONLY(movzwl(result, result);)
+  NOT_LP64(andl(result, 0xFFFF);)  // truncate upper 16 bits
+  jmp(done);
+
+  bind(notChar);
+  // cmpl(rcx, T_SHORT);  // all that's left
+  // jcc(Assembler::notEqual, done);
+  LP64_ONLY(movswl(result, result);)
+  NOT_LP64(shll(result, 16);)      // truncate upper 16 bits
+  NOT_LP64(sarl(result, 16);)      // and sign-extend short
+
+  // Nothing to do for T_INT
+  bind(done);
+}
+
 // remove activation
 //
 // Unlock the receiver if this is a synchronized method.

hotspot/src/cpu/x86/vm/interp_masm_x86.hpp

@@ -192,6 +192,9 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void prepare_to_jump_from_interpreted();
   void jump_from_interpreted(Register method, Register temp);
 
+  // narrow int return value
+  void narrow(Register result);
+
   // Returning from interpreted functions
   //
   // Removes the current activation (incl. unlocking of monitors)

hotspot/src/cpu/x86/vm/templateTable_x86.cpp

@@ -243,6 +243,7 @@ void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
   switch (bc) {
   case Bytecodes::_fast_aputfield:
   case Bytecodes::_fast_bputfield:
+  case Bytecodes::_fast_zputfield:
   case Bytecodes::_fast_cputfield:
   case Bytecodes::_fast_dputfield:
   case Bytecodes::_fast_fputfield:
@@ -1082,6 +1083,16 @@ void TemplateTable::bastore() {
   // rbx: index
   // rdx: array
   index_check(rdx, rbx); // prefer index in rbx
+  // Need to check whether array is boolean or byte
+  // since both types share the bastore bytecode.
+  __ load_klass(rcx, rdx);
+  __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
+  int diffbit = Klass::layout_helper_boolean_diffbit();
+  __ testl(rcx, diffbit);
+  Label L_skip;
+  __ jccb(Assembler::zero, L_skip);
+  __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
+  __ bind(L_skip);
   __ movb(Address(rdx, rbx,
                   Address::times_1,
                   arrayOopDesc::base_offset_in_bytes(T_BYTE)),
@@ -2540,13 +2551,12 @@ void TemplateTable::fast_binaryswitch() {
 void TemplateTable::_return(TosState state) {
   transition(state, state);
 
-  Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
-
   assert(_desc->calls_vm(),
          "inconsistent calls_vm information"); // call in remove_activation
 
   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
     assert(state == vtos, "only valid state");
+    Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
     __ movptr(robj, aaddress(0));
     __ load_klass(rdi, robj);
     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
@@ -2559,7 +2569,14 @@ void TemplateTable::_return(TosState state) {
     __ bind(skip_register_finalizer);
   }
 
+  // Narrow result if state is itos but result type is smaller.
+  // Need to narrow in the return bytecode rather than in generate_return_entry
+  // since compiled code callers expect the result to already be narrowed.
+  if (state == itos) {
+    __ narrow(rax);
+  }
   __ remove_activation(state, rbcp);
+
   __ jmp(rbcp);
 }
 
@@ -2754,7 +2771,7 @@ void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteContr
   const Address field(obj, off, Address::times_1, 0*wordSize);
   NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
 
-  Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
+  Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
 
   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
   // Make sure we don't need to mask edx after the above shift
@@ -2773,6 +2790,20 @@ void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteContr
   __ jmp(Done);
 
   __ bind(notByte);
+  __ cmpl(flags, ztos);
+  __ jcc(Assembler::notEqual, notBool);
+
+  // ztos (same code as btos)
+  __ load_signed_byte(rax, field);
+  __ push(ztos);
+  // Rewrite bytecode to be faster
+  if (!is_static && rc == may_rewrite) {
+    // use btos rewriting, no truncating to t/f bit is needed for getfield.
+    patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
+  }
+  __ jmp(Done);
+
+  __ bind(notBool);
   __ cmpl(flags, atos);
   __ jcc(Assembler::notEqual, notObj);
   // atos
@@ -3006,7 +3037,7 @@ void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteContr
   const Address field(obj, off, Address::times_1, 0*wordSize);
   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
 
-  Label notByte, notInt, notShort, notChar,
+  Label notByte, notBool, notInt, notShort, notChar,
         notLong, notFloat, notObj, notDouble;
 
   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
@@ -3027,6 +3058,22 @@ void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteContr
   }
 
   __ bind(notByte);
+  __ cmpl(flags, ztos);
+  __ jcc(Assembler::notEqual, notBool);
+
+  // ztos
+  {
+    __ pop(ztos);
+    if (!is_static) pop_and_check_object(obj);
+    __ andl(rax, 0x1);
+    __ movb(field, rax);
+    if (!is_static && rc == may_rewrite) {
+      patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
+    }
+    __ jmp(Done);
+  }
+
+  __ bind(notBool);
   __ cmpl(flags, atos);
   __ jcc(Assembler::notEqual, notObj);
 
@@ -3214,6 +3261,7 @@ void TemplateTable::jvmti_post_fast_field_mod() {
     switch (bytecode()) {          // load values into the jvalue object
     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
     case Bytecodes::_fast_bputfield: // fall through
+    case Bytecodes::_fast_zputfield: // fall through
     case Bytecodes::_fast_sputfield: // fall through
     case Bytecodes::_fast_cputfield: // fall through
     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
@@ -3238,6 +3286,7 @@ void TemplateTable::jvmti_post_fast_field_mod() {
     switch (bytecode()) {             // restore tos values
     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
     case Bytecodes::_fast_bputfield: // fall through
+    case Bytecodes::_fast_zputfield: // fall through
     case Bytecodes::_fast_sputfield: // fall through
     case Bytecodes::_fast_cputfield: // fall through
     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
@@ -3297,6 +3346,9 @@ void TemplateTable::fast_storefield(TosState state) {
   case Bytecodes::_fast_iputfield:
     __ movl(field, rax);
     break;
+  case Bytecodes::_fast_zputfield:
+    __ andl(rax, 0x1);  // boolean is true if LSB is 1
+    // fall through to bputfield
   case Bytecodes::_fast_bputfield:
     __ movb(field, rax);
     break;

hotspot/src/cpu/zero/vm/cppInterpreter_zero.cpp

@@ -82,6 +82,29 @@ int CppInterpreter::normal_entry(Method* method, intptr_t UNUSED, TRAPS) {
   return 0;
 }
 
+intptr_t narrow(BasicType type, intptr_t result) {
+  // mask integer result to narrower return type.
+  switch (type) {
+    case T_BOOLEAN:
+      return result&1;
+    case T_BYTE:
+      return (intptr_t)(jbyte)result;
+    case T_CHAR:
+      return (intptr_t)(uintptr_t)(jchar)result;
+    case T_SHORT:
+      return (intptr_t)(jshort)result;
+    case T_OBJECT:  // nothing to do fall through
+    case T_LONG:
+    case T_INT:
+    case T_FLOAT:
+    case T_DOUBLE:
+    case T_VOID:
+      return result;
+    default  : ShouldNotReachHere();
+  }
+}
+
+
 void CppInterpreter::main_loop(int recurse, TRAPS) {
   JavaThread *thread = (JavaThread *) THREAD;
   ZeroStack *stack = thread->zero_stack();
@@ -195,8 +218,14 @@ void CppInterpreter::main_loop(int recurse, TRAPS) {
   stack->set_sp(stack->sp() + method->max_locals());
 
   // Push our result
-  for (int i = 0; i < result_slots; i++)
+  for (int i = 0; i < result_slots; i++) {
-    stack->push(result[-i]);
+    // Adjust result to smaller
+    intptr_t res = result[-i];
+    if (result_slots == 1) {
+      res = narrow(result_type_of(method), res);
+    }
+    stack->push(res);
+  }
 }
 
 int CppInterpreter::native_entry(Method* method, intptr_t UNUSED, TRAPS) {
@@ -532,6 +561,7 @@ int CppInterpreter::accessor_entry(Method* method, intptr_t UNUSED, TRAPS) {
       break;
 
     case btos:
+    case ztos:
       SET_LOCALS_INT(object->byte_field_acquire(entry->f2_as_index()), 0);
       break;
 
@@ -570,6 +600,7 @@ int CppInterpreter::accessor_entry(Method* method, intptr_t UNUSED, TRAPS) {
       break;
 
     case btos:
+    case ztos:
       SET_LOCALS_INT(object->byte_field(entry->f2_as_index()), 0);
       break;
 

hotspot/src/share/vm/c1/c1_Canonicalizer.cpp

@@ -305,7 +305,8 @@ void Canonicalizer::do_StoreIndexed   (StoreIndexed*    x) {
     // limit this optimization to current block
     if (value != NULL && in_current_block(conv)) {
       set_canonical(new StoreIndexed(x->array(), x->index(), x->length(),
-                                     x->elt_type(), value, x->state_before()));
+                                     x->elt_type(), value, x->state_before(),
+                                     x->check_boolean()));
       return;
     }
   }

hotspot/src/share/vm/c1/c1_GraphBuilder.cpp

@@ -976,7 +976,19 @@ void GraphBuilder::store_indexed(BasicType type) {
       (array->as_NewArray() && array->as_NewArray()->length() && array->as_NewArray()->length()->type()->is_constant())) {
     length = append(new ArrayLength(array, state_before));
   }
-  StoreIndexed* result = new StoreIndexed(array, index, length, type, value, state_before);
+  ciType* array_type = array->declared_type();
+  bool check_boolean = false;
+  if (array_type != NULL) {
+    if (array_type->is_loaded() &&
+      array_type->as_array_klass()->element_type()->basic_type() == T_BOOLEAN) {
+      assert(type == T_BYTE, "boolean store uses bastore");
+      Value mask = append(new Constant(new IntConstant(1)));
+      value = append(new LogicOp(Bytecodes::_iand, value, mask));
+    }
+  } else if (type == T_BYTE) {
+    check_boolean = true;
+  }
+  StoreIndexed* result = new StoreIndexed(array, index, length, type, value, state_before, check_boolean);
   append(result);
   _memory->store_value(value);
 
@@ -1443,6 +1455,36 @@ void GraphBuilder::method_return(Value x) {
     need_mem_bar = true;
   }
 
+  BasicType bt = method()->return_type()->basic_type();
+  switch (bt) {
+    case T_BYTE:
+    {
+      Value shift = append(new Constant(new IntConstant(24)));
+      x = append(new ShiftOp(Bytecodes::_ishl, x, shift));
+      x = append(new ShiftOp(Bytecodes::_ishr, x, shift));
+      break;
+    }
+    case T_SHORT:
+    {
+      Value shift = append(new Constant(new IntConstant(16)));
+      x = append(new ShiftOp(Bytecodes::_ishl, x, shift));
+      x = append(new ShiftOp(Bytecodes::_ishr, x, shift));
+      break;
+    }
+    case T_CHAR:
+    {
+      Value mask = append(new Constant(new IntConstant(0xFFFF)));
+      x = append(new LogicOp(Bytecodes::_iand, x, mask));
+      break;
+    }
+    case T_BOOLEAN:
+    {
+      Value mask = append(new Constant(new IntConstant(1)));
+      x = append(new LogicOp(Bytecodes::_iand, x, mask));
+      break;
+    }
+  }
+
   // Check to see whether we are inlining. If so, Return
   // instructions become Gotos to the continuation point.
   if (continuation() != NULL) {
@@ -1611,6 +1653,10 @@ void GraphBuilder::access_field(Bytecodes::Code code) {
       if (state_before == NULL) {
         state_before = copy_state_for_exception();
       }
+      if (field->type()->basic_type() == T_BOOLEAN) {
+        Value mask = append(new Constant(new IntConstant(1)));
+        val = append(new LogicOp(Bytecodes::_iand, val, mask));
+      }
       append(new StoreField(append(obj), offset, field, val, true, state_before, needs_patching));
       break;
     }
@@ -1672,6 +1718,10 @@ void GraphBuilder::access_field(Bytecodes::Code code) {
       if (state_before == NULL) {
         state_before = copy_state_for_exception();
       }
+      if (field->type()->basic_type() == T_BOOLEAN) {
+        Value mask = append(new Constant(new IntConstant(1)));
+        val = append(new LogicOp(Bytecodes::_iand, val, mask));
+      }
       StoreField* store = new StoreField(obj, offset, field, val, false, state_before, needs_patching);
       if (!needs_patching) store = _memory->store(store);
       if (store != NULL) {
@@ -4134,7 +4184,12 @@ void GraphBuilder::append_unsafe_put_obj(ciMethod* callee, BasicType t, bool is_
 #ifndef _LP64
   offset = append(new Convert(Bytecodes::_l2i, offset, as_ValueType(T_INT)));
 #endif
-  Instruction* op = append(new UnsafePutObject(t, args->at(1), offset, args->at(3), is_volatile));
+  Value val = args->at(3);
+  if (t == T_BOOLEAN) {
+    Value mask = append(new Constant(new IntConstant(1)));
+    val = append(new LogicOp(Bytecodes::_iand, val, mask));
+  }
+  Instruction* op = append(new UnsafePutObject(t, args->at(1), offset, val, is_volatile));
   compilation()->set_has_unsafe_access(true);
   kill_all();
 }
@@ -4208,7 +4263,7 @@ void GraphBuilder::append_char_access(ciMethod* callee, bool is_store) {
   Value index = args->at(1);
   if (is_store) {
     Value value = args->at(2);
-    Instruction* store = append(new StoreIndexed(array, index, NULL, T_CHAR, value, state_before));
+    Instruction* store = append(new StoreIndexed(array, index, NULL, T_CHAR, value, state_before, false));
     store->set_flag(Instruction::NeedsRangeCheckFlag, false);
     _memory->store_value(value);
   } else {

hotspot/src/share/vm/c1/c1_Instruction.hpp

@@ -974,11 +974,13 @@ LEAF(StoreIndexed, AccessIndexed)
 
   ciMethod* _profiled_method;
   int       _profiled_bci;
+  bool      _check_boolean;
+
  public:
   // creation
-  StoreIndexed(Value array, Value index, Value length, BasicType elt_type, Value value, ValueStack* state_before)
+  StoreIndexed(Value array, Value index, Value length, BasicType elt_type, Value value, ValueStack* state_before, bool check_boolean)
   : AccessIndexed(array, index, length, elt_type, state_before)
-  , _value(value), _profiled_method(NULL), _profiled_bci(0)
+  , _value(value), _profiled_method(NULL), _profiled_bci(0), _check_boolean(check_boolean)
   {
     set_flag(NeedsWriteBarrierFlag, (as_ValueType(elt_type)->is_object()));
     set_flag(NeedsStoreCheckFlag, (as_ValueType(elt_type)->is_object()));
@@ -990,6 +992,7 @@ LEAF(StoreIndexed, AccessIndexed)
   Value value() const                            { return _value; }
   bool needs_write_barrier() const               { return check_flag(NeedsWriteBarrierFlag); }
   bool needs_store_check() const                 { return check_flag(NeedsStoreCheckFlag); }
+  bool check_boolean() const                     { return _check_boolean; }
   // Helpers for MethodData* profiling
   void set_should_profile(bool value)                { set_flag(ProfileMDOFlag, value); }
   void set_profiled_method(ciMethod* method)         { _profiled_method = method;   }

hotspot/src/share/vm/c1/c1_LIRGenerator.cpp

@@ -3680,3 +3680,26 @@ void LIRGenerator::do_MemBar(MemBar* x) {
     }
   }
 }
+
+LIR_Opr LIRGenerator::maybe_mask_boolean(StoreIndexed* x, LIR_Opr array, LIR_Opr value, CodeEmitInfo*& null_check_info) {
+  if (x->check_boolean()) {
+    LIR_Opr value_fixed = rlock_byte(T_BYTE);
+    if (TwoOperandLIRForm) {
+      __ move(value, value_fixed);
+      __ logical_and(value_fixed, LIR_OprFact::intConst(1), value_fixed);
+    } else {
+      __ logical_and(value, LIR_OprFact::intConst(1), value_fixed);
+    }
+    LIR_Opr klass = new_register(T_METADATA);
+    __ move(new LIR_Address(array, oopDesc::klass_offset_in_bytes(), T_ADDRESS), klass, null_check_info);
+    null_check_info = NULL;
+    LIR_Opr layout = new_register(T_INT);
+    __ move(new LIR_Address(klass, in_bytes(Klass::layout_helper_offset()), T_INT), layout);
+    int diffbit = Klass::layout_helper_boolean_diffbit();
+    __ logical_and(layout, LIR_OprFact::intConst(diffbit), layout);
+    __ cmp(lir_cond_notEqual, layout, LIR_OprFact::intConst(0));
+    __ cmove(lir_cond_notEqual, value_fixed, value, value_fixed, T_BYTE);
+    value = value_fixed;
+  }
+  return value;
+}

hotspot/src/share/vm/c1/c1_LIRGenerator.hpp

@@ -448,6 +448,7 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
   void profile_arguments(ProfileCall* x);
   void profile_parameters(Base* x);
   void profile_parameters_at_call(ProfileCall* x);
+  LIR_Opr maybe_mask_boolean(StoreIndexed* x, LIR_Opr array, LIR_Opr value, CodeEmitInfo*& null_check_info);
 
  public:
   Compilation*  compilation() const              { return _compilation; }

hotspot/src/share/vm/classfile/classFileParser.cpp

@@ -2713,11 +2713,9 @@ Method* ClassFileParser::parse_method(const ClassFileStream* const cfs,
   m->set_constants(_cp);
   m->set_name_index(name_index);
   m->set_signature_index(signature_index);
-#ifdef CC_INTERP
+
-  // hmm is there a gc issue here??
   ResultTypeFinder rtf(cp->symbol_at(signature_index));
-  m->set_result_index(rtf.type());
+  m->constMethod()->set_result_type(rtf.type());
-#endif
 
   if (args_size >= 0) {
     m->set_size_of_parameters(args_size);

hotspot/src/share/vm/classfile/defaultMethods.cpp

@@ -860,10 +860,8 @@ static Method* new_method(
   m->set_constants(NULL); // This will get filled in later
   m->set_name_index(cp->utf8(name));
   m->set_signature_index(cp->utf8(sig));
-#ifdef CC_INTERP
   ResultTypeFinder rtf(sig);
-  m->set_result_index(rtf.type());
+  m->constMethod()->set_result_type(rtf.type());
-#endif
   m->set_size_of_parameters(params);
   m->set_max_stack(max_stack);
   m->set_max_locals(params);

hotspot/src/share/vm/interpreter/bytecodeInterpreter.cpp

@@ -1767,8 +1767,19 @@ run:
           ((objArrayOop) arrObj)->obj_at_put(index, rhsObject);
           UPDATE_PC_AND_TOS_AND_CONTINUE(1, -3);
       }
-      CASE(_bastore):
+      CASE(_bastore): {
-          ARRAY_STOREFROM32(T_BYTE, jbyte,  "%d",   STACK_INT, 0);
+          ARRAY_INTRO(-3);
+          int item = STACK_INT(-1);
+          // if it is a T_BOOLEAN array, mask the stored value to 0/1
+          if (arrObj->klass() == Universe::boolArrayKlassObj()) {
+            item &= 1;
+          } else {
+            assert(arrObj->klass() == Universe::byteArrayKlassObj(),
+                   "should be byte array otherwise");
+          }
+          ((typeArrayOop)arrObj)->byte_at_put(index, item);
+          UPDATE_PC_AND_TOS_AND_CONTINUE(1, -3);
+      }
       CASE(_castore):
           ARRAY_STOREFROM32(T_CHAR, jchar,  "%d",   STACK_INT, 0);
       CASE(_sastore):
@@ -1999,7 +2010,7 @@ run:
             } else if (tos_type == ltos) {
               SET_STACK_LONG(obj->long_field_acquire(field_offset), 0);
               MORE_STACK(1);
-            } else if (tos_type == btos) {
+            } else if (tos_type == btos || tos_type == ztos) {
               SET_STACK_INT(obj->byte_field_acquire(field_offset), -1);
             } else if (tos_type == ctos) {
               SET_STACK_INT(obj->char_field_acquire(field_offset), -1);
@@ -2020,7 +2031,7 @@ run:
             } else if (tos_type == ltos) {
               SET_STACK_LONG(obj->long_field(field_offset), 0);
               MORE_STACK(1);
-            } else if (tos_type == btos) {
+            } else if (tos_type == btos || tos_type == ztos) {
               SET_STACK_INT(obj->byte_field(field_offset), -1);
             } else if (tos_type == ctos) {
               SET_STACK_INT(obj->char_field(field_offset), -1);
@@ -2109,6 +2120,9 @@ run:
               obj->release_obj_field_put(field_offset, STACK_OBJECT(-1));
             } else if (tos_type == btos) {
               obj->release_byte_field_put(field_offset, STACK_INT(-1));
+            } else if (tos_type == ztos) {
+              int bool_field = STACK_INT(-1);  // only store LSB
+              obj->release_byte_field_put(field_offset, (bool_field & 1));
             } else if (tos_type == ltos) {
               obj->release_long_field_put(field_offset, STACK_LONG(-1));
             } else if (tos_type == ctos) {
@@ -2129,6 +2143,9 @@ run:
               obj->obj_field_put(field_offset, STACK_OBJECT(-1));
             } else if (tos_type == btos) {
               obj->byte_field_put(field_offset, STACK_INT(-1));
+            } else if (tos_type == ztos) {
+              int bool_field = STACK_INT(-1);  // only store LSB
+              obj->byte_field_put(field_offset, (bool_field & 1));
             } else if (tos_type == ltos) {
               obj->long_field_put(field_offset, STACK_LONG(-1));
             } else if (tos_type == ctos) {

hotspot/src/share/vm/interpreter/bytecodes.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -496,6 +496,7 @@ void Bytecodes::initialize() {
 
   def(_fast_aputfield      , "fast_aputfield"      , "bJJ"  , NULL    , T_OBJECT ,  0, true , _putfield       );
   def(_fast_bputfield      , "fast_bputfield"      , "bJJ"  , NULL    , T_INT    ,  0, true , _putfield       );
+  def(_fast_zputfield      , "fast_zputfield"      , "bJJ"  , NULL    , T_INT    ,  0, true , _putfield       );
   def(_fast_cputfield      , "fast_cputfield"      , "bJJ"  , NULL    , T_CHAR   ,  0, true , _putfield       );
   def(_fast_dputfield      , "fast_dputfield"      , "bJJ"  , NULL    , T_DOUBLE ,  0, true , _putfield       );
   def(_fast_fputfield      , "fast_fputfield"      , "bJJ"  , NULL    , T_FLOAT  ,  0, true , _putfield       );

hotspot/src/share/vm/interpreter/bytecodes.hpp

@@ -257,6 +257,7 @@ class Bytecodes: AllStatic {
 
     _fast_aputfield       ,
     _fast_bputfield       ,
+    _fast_zputfield       ,
     _fast_cputfield       ,
     _fast_dputfield       ,
     _fast_fputfield       ,

hotspot/src/share/vm/interpreter/interpreterRuntime.cpp

@@ -1077,7 +1077,8 @@ IRT_ENTRY(void, InterpreterRuntime::post_field_modification(JavaThread *thread,
   char sig_type = '\0';
 
   switch(cp_entry->flag_state()) {
-    case btos: sig_type = 'Z'; break;
+    case btos: sig_type = 'B'; break;
+    case ztos: sig_type = 'Z'; break;
     case ctos: sig_type = 'C'; break;
     case stos: sig_type = 'S'; break;
     case itos: sig_type = 'I'; break;

hotspot/src/share/vm/interpreter/templateInterpreter.cpp

@@ -89,8 +89,9 @@ void TemplateInterpreter::initialize() {
 // Implementation of EntryPoint
 
 EntryPoint::EntryPoint() {
-  assert(number_of_states == 9, "check the code below");
+  assert(number_of_states == 10, "check the code below");
   _entry[btos] = NULL;
+  _entry[ztos] = NULL;
   _entry[ctos] = NULL;
   _entry[stos] = NULL;
   _entry[atos] = NULL;
@@ -102,9 +103,10 @@ EntryPoint::EntryPoint() {
 }
 
 
-EntryPoint::EntryPoint(address bentry, address centry, address sentry, address aentry, address ientry, address lentry, address fentry, address dentry, address ventry) {
+EntryPoint::EntryPoint(address bentry, address zentry, address centry, address sentry, address aentry, address ientry, address lentry, address fentry, address dentry, address ventry) {
-  assert(number_of_states == 9, "check the code below");
+  assert(number_of_states == 10, "check the code below");
   _entry[btos] = bentry;
+  _entry[ztos] = zentry;
   _entry[ctos] = centry;
   _entry[stos] = sentry;
   _entry[atos] = aentry;
@@ -155,6 +157,7 @@ EntryPoint DispatchTable::entry(int i) const {
   return
     EntryPoint(
       _table[btos][i],
+      _table[ztos][i],
       _table[ctos][i],
       _table[stos][i],
       _table[atos][i],
@@ -169,8 +172,9 @@ EntryPoint DispatchTable::entry(int i) const {
 
 void DispatchTable::set_entry(int i, EntryPoint& entry) {
   assert(0 <= i && i < length, "index out of bounds");
-  assert(number_of_states == 9, "check the code below");
+  assert(number_of_states == 10, "check the code below");
   _table[btos][i] = entry.entry(btos);
+  _table[ztos][i] = entry.entry(ztos);
   _table[ctos][i] = entry.entry(ctos);
   _table[stos][i] = entry.entry(stos);
   _table[atos][i] = entry.entry(atos);

hotspot/src/share/vm/interpreter/templateInterpreter.hpp

@@ -47,7 +47,7 @@ class EntryPoint VALUE_OBJ_CLASS_SPEC {
  public:
   // Construction
   EntryPoint();
-  EntryPoint(address bentry, address centry, address sentry, address aentry, address ientry, address lentry, address fentry, address dentry, address ventry);
+  EntryPoint(address bentry, address zentry, address centry, address sentry, address aentry, address ientry, address lentry, address fentry, address dentry, address ventry);
 
   // Attributes
   address entry(TosState state) const;                // return target address for a given tosca state

hotspot/src/share/vm/interpreter/templateInterpreterGenerator.cpp

@@ -74,6 +74,7 @@ void TemplateInterpreterGenerator::generate_all() {
         Interpreter::_trace_code =
           EntryPoint(
                      generate_trace_code(btos),
+                     generate_trace_code(ztos),
                      generate_trace_code(ctos),
                      generate_trace_code(stos),
                      generate_trace_code(atos),
@@ -94,6 +95,7 @@ void TemplateInterpreterGenerator::generate_all() {
                        generate_return_entry_for(itos, i, index_size),
                        generate_return_entry_for(itos, i, index_size),
                        generate_return_entry_for(itos, i, index_size),
+                       generate_return_entry_for(itos, i, index_size),
                        generate_return_entry_for(atos, i, index_size),
                        generate_return_entry_for(itos, i, index_size),
                        generate_return_entry_for(ltos, i, index_size),
@@ -105,13 +107,16 @@ void TemplateInterpreterGenerator::generate_all() {
       }
 
       { CodeletMark cm(_masm, "invoke return entry points");
-        const TosState states[] = {itos, itos, itos, itos, ltos, ftos, dtos, atos, vtos};
+        // These states are in order specified in TosState, except btos/ztos/ctos/stos are
+        // really the same as itos since there is no top of stack optimization for these types
+        const TosState states[] = {itos, itos, itos, itos, itos, ltos, ftos, dtos, atos, vtos, ilgl};
         const int invoke_length = Bytecodes::length_for(Bytecodes::_invokestatic);
         const int invokeinterface_length = Bytecodes::length_for(Bytecodes::_invokeinterface);
         const int invokedynamic_length = Bytecodes::length_for(Bytecodes::_invokedynamic);
 
         for (int i = 0; i < Interpreter::number_of_return_addrs; i++) {
           TosState state = states[i];
+          assert(state != ilgl, "states array is wrong above");
           Interpreter::_invoke_return_entry[i] = generate_return_entry_for(state, invoke_length, sizeof(u2));
           Interpreter::_invokeinterface_return_entry[i] = generate_return_entry_for(state, invokeinterface_length, sizeof(u2));
           Interpreter::_invokedynamic_return_entry[i] = generate_return_entry_for(state, invokedynamic_length, sizeof(u4));
@@ -122,6 +127,7 @@ void TemplateInterpreterGenerator::generate_all() {
         Interpreter::_earlyret_entry =
           EntryPoint(
                      generate_earlyret_entry_for(btos),
+                     generate_earlyret_entry_for(ztos),
                      generate_earlyret_entry_for(ctos),
                      generate_earlyret_entry_for(stos),
                      generate_earlyret_entry_for(atos),
@@ -140,6 +146,7 @@ void TemplateInterpreterGenerator::generate_all() {
                        generate_deopt_entry_for(itos, i),
                        generate_deopt_entry_for(itos, i),
                        generate_deopt_entry_for(itos, i),
+                       generate_deopt_entry_for(itos, i),
                        generate_deopt_entry_for(atos, i),
                        generate_deopt_entry_for(itos, i),
                        generate_deopt_entry_for(ltos, i),
@@ -167,6 +174,7 @@ void TemplateInterpreterGenerator::generate_all() {
         Interpreter::_continuation_entry =
           EntryPoint(
                      generate_continuation_for(btos),
+                     generate_continuation_for(ztos),
                      generate_continuation_for(ctos),
                      generate_continuation_for(stos),
                      generate_continuation_for(atos),
@@ -182,6 +190,7 @@ void TemplateInterpreterGenerator::generate_all() {
         Interpreter::_safept_entry =
           EntryPoint(
                      generate_safept_entry_for(btos, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint)),
+                     generate_safept_entry_for(ztos, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint)),
                      generate_safept_entry_for(ctos, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint)),
                      generate_safept_entry_for(stos, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint)),
                      generate_safept_entry_for(atos, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint)),
@@ -301,7 +310,7 @@ void TemplateInterpreterGenerator::set_safepoints_for_all_bytes() {
 
 void TemplateInterpreterGenerator::set_unimplemented(int i) {
   address e = _unimplemented_bytecode;
-  EntryPoint entry(e, e, e, e, e, e, e, e, e);
+  EntryPoint entry(e, e, e, e, e, e, e, e, e, e);
   Interpreter::_normal_table.set_entry(i, entry);
   Interpreter::_wentry_point[i] = _unimplemented_bytecode;
 }
@@ -316,6 +325,7 @@ void TemplateInterpreterGenerator::set_entry_points(Bytecodes::Code code) {
   assert(_unimplemented_bytecode    != NULL, "should have been generated before");
   assert(_illegal_bytecode_sequence != NULL, "should have been generated before");
   address bep = _illegal_bytecode_sequence;
+  address zep = _illegal_bytecode_sequence;
   address cep = _illegal_bytecode_sequence;
   address sep = _illegal_bytecode_sequence;
   address aep = _illegal_bytecode_sequence;
@@ -337,7 +347,7 @@ void TemplateInterpreterGenerator::set_entry_points(Bytecodes::Code code) {
     set_wide_entry_point(t, wep);
   }
   // set entry points
-  EntryPoint entry(bep, cep, sep, aep, iep, lep, fep, dep, vep);
+  EntryPoint entry(bep, zep, cep, sep, aep, iep, lep, fep, dep, vep);
   Interpreter::_normal_table.set_entry(code, entry);
   Interpreter::_wentry_point[code] = wep;
   CodeCacheExtensions::completed_template_interpreter_entries(_masm, code);
@@ -355,6 +365,7 @@ void TemplateInterpreterGenerator::set_short_entry_points(Template* t, address&
   assert(t->is_valid(), "template must exist");
   switch (t->tos_in()) {
     case btos:
+    case ztos:
     case ctos:
     case stos:
       ShouldNotReachHere();  // btos/ctos/stos should use itos.

hotspot/src/share/vm/interpreter/templateTable.cpp

@@ -488,6 +488,7 @@ void TemplateTable::initialize() {
 
   def(Bytecodes::_fast_aputfield      , ubcp|____|____|____, atos, vtos, fast_storefield ,   atos        );
   def(Bytecodes::_fast_bputfield      , ubcp|____|____|____, itos, vtos, fast_storefield ,   itos        );
+  def(Bytecodes::_fast_zputfield      , ubcp|____|____|____, itos, vtos, fast_storefield ,   itos        );
   def(Bytecodes::_fast_cputfield      , ubcp|____|____|____, itos, vtos, fast_storefield  ,  itos        );
   def(Bytecodes::_fast_dputfield      , ubcp|____|____|____, dtos, vtos, fast_storefield  ,  dtos        );
   def(Bytecodes::_fast_fputfield      , ubcp|____|____|____, ftos, vtos, fast_storefield  ,  ftos        );

hotspot/src/share/vm/oops/constMethod.cpp

@@ -66,6 +66,7 @@ ConstMethod::ConstMethod(int byte_code_size,
   set_max_locals(0);
   set_method_idnum(0);
   set_size_of_parameters(0);
+  set_result_type(T_VOID);
 }
 
 // Accessor that copies to metadata.

hotspot/src/share/vm/oops/constMethod.hpp

@@ -210,6 +210,7 @@ private:
 
   int               _constMethod_size;
   u2                _flags;
+  u1                _result_type;                 // BasicType of result
 
   // Size of Java bytecodes allocated immediately after Method*.
   u2                _code_size;
@@ -494,6 +495,8 @@ public:
   static ByteSize size_of_parameters_offset()
                             { return byte_offset_of(ConstMethod, _size_of_parameters); }
 
+  static ByteSize result_type_offset()
+                            { return byte_offset_of(ConstMethod, _result_type); }
 
   // Unique id for the method
   static const u2 MAX_IDNUM;
@@ -516,6 +519,8 @@ public:
   int  size_of_parameters() const                { return _size_of_parameters; }
   void set_size_of_parameters(int size)          { _size_of_parameters = size; }
 
+  void set_result_type(BasicType rt)             { assert(rt < 16, "result type too large");
+                                                   _result_type = (u1)rt; }
   // Deallocation for RedefineClasses
   void deallocate_contents(ClassLoaderData* loader_data);
   bool is_klass() const { return false; }

hotspot/src/share/vm/oops/cpCache.hpp

@@ -77,18 +77,19 @@ class PSPromotionManager;
 // f2 flag true if f2 contains an oop (e.g., virtual final method)
 // fv flag true if invokeinterface used for method in class Object
 //
-// The flags 31, 30, 29, 28 together build a 4 bit number 0 to 8 with the
+// The flags 31, 30, 29, 28 together build a 4 bit number 0 to 16 with the
 // following mapping to the TosState states:
 //
 // btos: 0
-// ctos: 1
+// ztos: 1
-// stos: 2
+// ctos: 2
-// itos: 3
+// stos: 3
-// ltos: 4
+// itos: 4
-// ftos: 5
+// ltos: 5
-// dtos: 6
+// ftos: 6
-// atos: 7
+// dtos: 7
-// vtos: 8
+// atos: 8
+// vtos: 9
 //
 // Entry specific: field entries:
 // _indices = get (b1 section) and put (b2 section) bytecodes, original constant pool index
@@ -352,14 +353,8 @@ class ConstantPoolCacheEntry VALUE_OBJ_CLASS_SPEC {
   bool has_method_type() const                   { return (!is_f1_null()) && (_flags & (1 << has_method_type_shift))   != 0; }
   bool is_method_entry() const                   { return (_flags & (1 << is_field_entry_shift))    == 0; }
   bool is_field_entry() const                    { return (_flags & (1 << is_field_entry_shift))    != 0; }
-  bool is_byte() const                           { return flag_state() == btos; }
-  bool is_char() const                           { return flag_state() == ctos; }
-  bool is_short() const                          { return flag_state() == stos; }
-  bool is_int() const                            { return flag_state() == itos; }
   bool is_long() const                           { return flag_state() == ltos; }
-  bool is_float() const                          { return flag_state() == ftos; }
   bool is_double() const                         { return flag_state() == dtos; }
-  bool is_object() const                         { return flag_state() == atos; }
   TosState flag_state() const                    { assert((uint)number_of_states <= (uint)tos_state_mask+1, "");
                                                    return (TosState)((_flags >> tos_state_shift) & tos_state_mask); }
 

hotspot/src/share/vm/oops/klass.hpp

@@ -342,6 +342,21 @@ protected:
     assert(btvalue >= T_BOOLEAN && btvalue <= T_OBJECT, "sanity");
     return (BasicType) btvalue;
   }
+
+  // Want a pattern to quickly diff against layout header in register
+  // find something less clever!
+  static int layout_helper_boolean_diffbit() {
+    jint zlh = array_layout_helper(T_BOOLEAN);
+    jint blh = array_layout_helper(T_BYTE);
+    assert(zlh != blh, "array layout helpers must differ");
+    int diffbit = 1;
+    while ((diffbit & (zlh ^ blh)) == 0 && (diffbit & zlh) == 0) {
+      diffbit <<= 1;
+      assert(diffbit != 0, "make sure T_BOOLEAN has a different bit than T_BYTE");
+    }
+    return diffbit;
+  }
+
   static int layout_helper_log2_element_size(jint lh) {
     assert(lh < (jint)_lh_neutral_value, "must be array");
     int l2esz = (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;

hotspot/src/share/vm/oops/method.cpp

@@ -1220,10 +1220,8 @@ methodHandle Method::make_method_handle_intrinsic(vmIntrinsics::ID iid,
   m->set_signature_index(_imcp_invoke_signature);
   assert(MethodHandles::is_signature_polymorphic_name(m->name()), "");
   assert(m->signature() == signature, "");
-#ifdef CC_INTERP
   ResultTypeFinder rtf(signature);
-  m->set_result_index(rtf.type());
+  m->constMethod()->set_result_type(rtf.type());
-#endif
   m->compute_size_of_parameters(THREAD);
   m->init_intrinsic_id();
   assert(m->is_method_handle_intrinsic(), "");

hotspot/src/share/vm/oops/oop.inline.hpp

@@ -443,7 +443,7 @@ jchar oopDesc::char_field(int offset) const                   { return (jchar) *
 void oopDesc::char_field_put(int offset, jchar contents)      { *char_field_addr(offset) = (jint) contents; }
 
 jboolean oopDesc::bool_field(int offset) const                { return (jboolean) *bool_field_addr(offset); }
-void oopDesc::bool_field_put(int offset, jboolean contents)   { *bool_field_addr(offset) = (jint) contents; }
+void oopDesc::bool_field_put(int offset, jboolean contents)   { *bool_field_addr(offset) = (((jint) contents) & 1); }
 
 jint oopDesc::int_field(int offset) const                     { return *int_field_addr(offset);        }
 void oopDesc::int_field_put(int offset, jint contents)        { *int_field_addr(offset) = contents;    }
@@ -483,7 +483,7 @@ jchar oopDesc::char_field_acquire(int offset) const                   { return O
 void oopDesc::release_char_field_put(int offset, jchar contents)      { OrderAccess::release_store(char_field_addr(offset), contents); }
 
 jboolean oopDesc::bool_field_acquire(int offset) const                { return OrderAccess::load_acquire(bool_field_addr(offset));     }
-void oopDesc::release_bool_field_put(int offset, jboolean contents)   { OrderAccess::release_store(bool_field_addr(offset), contents); }
+void oopDesc::release_bool_field_put(int offset, jboolean contents)   { OrderAccess::release_store(bool_field_addr(offset), (contents & 1)); }
 
 jint oopDesc::int_field_acquire(int offset) const                     { return OrderAccess::load_acquire(int_field_addr(offset));      }
 void oopDesc::release_int_field_put(int offset, jint contents)        { OrderAccess::release_store(int_field_addr(offset), contents);  }

hotspot/src/share/vm/opto/memnode.cpp

@@ -2389,7 +2389,7 @@ StoreNode* StoreNode::make(PhaseGVN& gvn, Node* ctl, Node* mem, Node* adr, const
          ctl != NULL, "raw memory operations should have control edge");
 
   switch (bt) {
-  case T_BOOLEAN:
+  case T_BOOLEAN: val = gvn.transform(new AndINode(val, gvn.intcon(0x1))); // Fall through to T_BYTE case
   case T_BYTE:    return new StoreBNode(ctl, mem, adr, adr_type, val, mo);
   case T_INT:     return new StoreINode(ctl, mem, adr, adr_type, val, mo);
   case T_CHAR:

hotspot/src/share/vm/opto/parse1.cpp

@@ -730,6 +730,26 @@ void Parse::do_all_blocks() {
 #endif
 }
 
+static Node* mask_int_value(Node* v, BasicType bt, PhaseGVN* gvn) {
+  switch (bt) {
+  case T_BYTE:
+    v = gvn->transform(new LShiftINode(v, gvn->intcon(24)));
+    v = gvn->transform(new RShiftINode(v, gvn->intcon(24)));
+    break;
+  case T_SHORT:
+    v = gvn->transform(new LShiftINode(v, gvn->intcon(16)));
+    v = gvn->transform(new RShiftINode(v, gvn->intcon(16)));
+    break;
+  case T_CHAR:
+    v = gvn->transform(new AndINode(v, gvn->intcon(0xFFFF)));
+    break;
+  case T_BOOLEAN:
+    v = gvn->transform(new AndINode(v, gvn->intcon(0x1)));
+    break;
+  }
+  return v;
+}
+
 //-------------------------------build_exits----------------------------------
 // Build normal and exceptional exit merge points.
 void Parse::build_exits() {
@@ -754,6 +774,16 @@ void Parse::build_exits() {
   // Add a return value to the exit state.  (Do not push it yet.)
   if (tf()->range()->cnt() > TypeFunc::Parms) {
     const Type* ret_type = tf()->range()->field_at(TypeFunc::Parms);
+    if (ret_type->isa_int()) {
+      BasicType ret_bt = method()->return_type()->basic_type();
+      if (ret_bt == T_BOOLEAN ||
+          ret_bt == T_CHAR ||
+          ret_bt == T_BYTE ||
+          ret_bt == T_SHORT) {
+        ret_type = TypeInt::INT;
+      }
+    }
+
     // Don't "bind" an unloaded return klass to the ret_phi. If the klass
     // becomes loaded during the subsequent parsing, the loaded and unloaded
     // types will not join when we transform and push in do_exits().
@@ -1014,6 +1044,10 @@ void Parse::do_exits() {
       }
       return;
     }
+    if (ret_type->isa_int()) {
+      BasicType ret_bt = method()->return_type()->basic_type();
+      ret_phi = mask_int_value(ret_phi, ret_bt, &_gvn);
+    }
     _exits.push_node(ret_type->basic_type(), ret_phi);
   }
 

hotspot/src/share/vm/opto/parse2.cpp

@@ -64,11 +64,15 @@ void Parse::array_load(BasicType elem_type) {
 
 //--------------------------------array_store----------------------------------
 void Parse::array_store(BasicType elem_type) {
-  Node* adr = array_addressing(elem_type, 1);
+  const Type* elem = Type::TOP;
+  Node* adr = array_addressing(elem_type, 1, &elem);
   if (stopped())  return;     // guaranteed null or range check
   Node* val = pop();
   dec_sp(2);                  // Pop array and index
   const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(elem_type);
+  if (elem == TypeInt::BOOL) {
+    elem_type = T_BOOLEAN;
+  }
   store_to_memory(control(), adr, val, elem_type, adr_type, StoreNode::release_if_reference(elem_type));
 }
 

hotspot/src/share/vm/opto/type.cpp

@@ -1820,14 +1820,16 @@ const TypeTuple *TypeTuple::make_domain(ciInstanceKlass* recv, ciSignature* sig)
       break;
     case T_OBJECT:
     case T_ARRAY:
-    case T_BOOLEAN:
-    case T_CHAR:
     case T_FLOAT:
-    case T_BYTE:
-    case T_SHORT:
     case T_INT:
       field_array[pos++] = get_const_type(type);
       break;
+    case T_BOOLEAN:
+    case T_CHAR:
+    case T_BYTE:
+    case T_SHORT:
+      field_array[pos++] = TypeInt::INT;
+      break;
     default:
       ShouldNotReachHere();
     }

hotspot/src/share/vm/prims/jni.cpp

@@ -2188,6 +2188,7 @@ JNI_QUICK_ENTRY(void, jni_Set##Result##Field(JNIEnv *env, jobject obj, jfieldID
     field_value.unionType = value; \
     o = JvmtiExport::jni_SetField_probe_nh(thread, obj, o, k, fieldID, false, SigType, (jvalue *)&field_value); \
   } \
+  if (SigType == 'Z') { value = ((jboolean)value) & 1; } \
   o->Fieldname##_field_put(offset, value); \
   ReturnProbe; \
 JNI_END
@@ -2387,6 +2388,7 @@ JNI_ENTRY(void, jni_SetStatic##Result##Field(JNIEnv *env, jclass clazz, jfieldID
     field_value.unionType = value; \
     JvmtiExport::jni_SetField_probe(thread, NULL, NULL, id->holder(), fieldID, true, SigType, (jvalue *)&field_value); \
   } \
+  if (SigType == 'Z') { value = ((jboolean)value) & 1; } \
   id->holder()->java_mirror()-> Fieldname##_field_put (id->offset(), value); \
   ReturnProbe;\
 JNI_END

hotspot/src/share/vm/prims/jvmtiEnvBase.cpp

@@ -1354,7 +1354,7 @@ JvmtiEnvBase::check_top_frame(JavaThread* current_thread, JavaThread* java_threa
   ResultTypeFinder rtf(signature);
   TosState fr_tos = as_TosState(rtf.type());
   if (fr_tos != tos) {
-    if (tos != itos || (fr_tos != btos && fr_tos != ctos && fr_tos != stos)) {
+    if (tos != itos || (fr_tos != btos && fr_tos != ztos && fr_tos != ctos && fr_tos != stos)) {
       return JVMTI_ERROR_TYPE_MISMATCH;
     }
   }

hotspot/src/share/vm/prims/jvmtiExport.cpp

@@ -1702,7 +1702,7 @@ void JvmtiExport::post_raw_field_modification(JavaThread *thread, Method* method
   address location, KlassHandle field_klass, Handle object, jfieldID field,
   char sig_type, jvalue *value) {
 
-  if (sig_type == 'I' || sig_type == 'Z' || sig_type == 'C' || sig_type == 'S') {
+  if (sig_type == 'I' || sig_type == 'Z' || sig_type == 'B' || sig_type == 'C' || sig_type == 'S') {
     // 'I' instructions are used for byte, char, short and int.
     // determine which it really is, and convert
     fieldDescriptor fd;

hotspot/src/share/vm/prims/unsafe.cpp

@@ -133,13 +133,22 @@ jlong Unsafe_field_offset_from_byte_offset(jlong byte_offset) {
 
 ///// Data in the Java heap.
 
+#define truncate_jboolean(x) ((x) & 1)
+#define truncate_jbyte(x) (x)
+#define truncate_jshort(x) (x)
+#define truncate_jchar(x) (x)
+#define truncate_jint(x) (x)
+#define truncate_jlong(x) (x)
+#define truncate_jfloat(x) (x)
+#define truncate_jdouble(x) (x)
+
 #define GET_FIELD(obj, offset, type_name, v) \
   oop p = JNIHandles::resolve(obj); \
   type_name v = *(type_name*)index_oop_from_field_offset_long(p, offset)
 
 #define SET_FIELD(obj, offset, type_name, x) \
   oop p = JNIHandles::resolve(obj); \
-  *(type_name*)index_oop_from_field_offset_long(p, offset) = x
+  *(type_name*)index_oop_from_field_offset_long(p, offset) = truncate_##type_name(x)
 
 #define GET_FIELD_VOLATILE(obj, offset, type_name, v) \
   oop p = JNIHandles::resolve(obj); \
@@ -150,7 +159,7 @@ jlong Unsafe_field_offset_from_byte_offset(jlong byte_offset) {
 
 #define SET_FIELD_VOLATILE(obj, offset, type_name, x) \
   oop p = JNIHandles::resolve(obj); \
-  OrderAccess::release_store_fence((volatile type_name*)index_oop_from_field_offset_long(p, offset), x);
+  OrderAccess::release_store_fence((volatile type_name*)index_oop_from_field_offset_long(p, offset), truncate_##type_name(x));
 
 
 // Get/SetObject must be special-cased, since it works with handles.

hotspot/src/share/vm/runtime/reflection.cpp

@@ -1025,7 +1025,7 @@ static BasicType basic_type_mirror_to_basic_type(oop basic_type_mirror, TRAPS) {
 static void narrow(jvalue* value, BasicType narrow_type, TRAPS) {
   switch (narrow_type) {
   case T_BOOLEAN:
-    value->z = (jboolean)value->i;
+    value->z = (jboolean) (value->i & 1);
     return;
   case T_BYTE:
     value->b = (jbyte)value->i;

hotspot/src/share/vm/utilities/globalDefinitions.hpp

@@ -846,14 +846,15 @@ class JavaValue {
 
 enum TosState {         // describes the tos cache contents
   btos = 0,             // byte, bool tos cached
-  ctos = 1,             // char tos cached
+  ztos = 1,             // byte, bool tos cached
-  stos = 2,             // short tos cached
+  ctos = 2,             // char tos cached
-  itos = 3,             // int tos cached
+  stos = 3,             // short tos cached
-  ltos = 4,             // long tos cached
+  itos = 4,             // int tos cached
-  ftos = 5,             // float tos cached
+  ltos = 5,             // long tos cached
-  dtos = 6,             // double tos cached
+  ftos = 6,             // float tos cached
-  atos = 7,             // object cached
+  dtos = 7,             // double tos cached
-  vtos = 8,             // tos not cached
+  atos = 8,             // object cached
+  vtos = 9,             // tos not cached
   number_of_states,
   ilgl                  // illegal state: should not occur
 };
@@ -862,7 +863,7 @@ enum TosState {         // describes the tos cache contents
 inline TosState as_TosState(BasicType type) {
   switch (type) {
     case T_BYTE   : return btos;
-    case T_BOOLEAN: return btos; // FIXME: Add ztos
+    case T_BOOLEAN: return ztos;
     case T_CHAR   : return ctos;
     case T_SHORT  : return stos;
     case T_INT    : return itos;
@@ -878,8 +879,8 @@ inline TosState as_TosState(BasicType type) {
 
 inline BasicType as_BasicType(TosState state) {
   switch (state) {
-    //case ztos: return T_BOOLEAN;//FIXME
     case btos : return T_BYTE;
+    case ztos : return T_BOOLEAN;
     case ctos : return T_CHAR;
     case stos : return T_SHORT;
     case itos : return T_INT;