x86_64: TargetReg update for x86

Also includes changes in common code. Elimination of use of TargetReg
with one parameter and direct access to special target registers.

Change-Id: Ied2c1f87d4d1e4345248afe74bca40487a46a371
Signed-off-by: Serguei Katkov <serguei.i.katkov@intel.com>
Signed-off-by: Chao-ying Fu <chao-ying.fu@intel.com>
diff --git a/compiler/dex/quick/x86/call_x86.cc b/compiler/dex/quick/x86/call_x86.cc
index bfbfa0e..6ca220c 100644
--- a/compiler/dex/quick/x86/call_x86.cc
+++ b/compiler/dex/quick/x86/call_x86.cc
@@ -158,29 +158,33 @@
 
   // Making a call - use explicit registers
   FlushAllRegs();   /* Everything to home location */
-  LoadValueDirectFixed(rl_src, rs_rX86_ARG0);
+  RegStorage array_ptr = TargetRefReg(kArg0);
+  RegStorage payload = TargetPtrReg(kArg1);
+  RegStorage method_start = TargetPtrReg(kArg2);
+
+  LoadValueDirectFixed(rl_src, array_ptr);
   // Materialize a pointer to the fill data image
   if (base_of_code_ != nullptr) {
     // We can use the saved value.
     RegLocation rl_method = mir_graph_->GetRegLocation(base_of_code_->s_reg_low);
     if (rl_method.wide) {
-      LoadValueDirectWide(rl_method, rs_rX86_ARG2);
+      LoadValueDirectWide(rl_method, method_start);
     } else {
-      LoadValueDirect(rl_method, rs_rX86_ARG2);
+      LoadValueDirect(rl_method, method_start);
     }
     store_method_addr_used_ = true;
   } else {
     // TODO(64) force to be 64-bit
-    NewLIR1(kX86StartOfMethod, rs_rX86_ARG2.GetReg());
+    NewLIR1(kX86StartOfMethod, method_start.GetReg());
   }
-  NewLIR2(kX86PcRelAdr, rs_rX86_ARG1.GetReg(), WrapPointer(tab_rec));
-  NewLIR2(cu_->target64 ? kX86Add64RR : kX86Add32RR, rs_rX86_ARG1.GetReg(), rs_rX86_ARG2.GetReg());
+  NewLIR2(kX86PcRelAdr, payload.GetReg(), WrapPointer(tab_rec));
+  OpRegReg(kOpAdd, payload, method_start);
   if (cu_->target64) {
-    CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(8, pHandleFillArrayData), rs_rX86_ARG0,
-                            rs_rX86_ARG1, true);
+    CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(8, pHandleFillArrayData), array_ptr,
+                            payload, true);
   } else {
-    CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(4, pHandleFillArrayData), rs_rX86_ARG0,
-                            rs_rX86_ARG1, true);
+    CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(4, pHandleFillArrayData), array_ptr,
+                            payload, true);
   }
 }
 
@@ -291,11 +295,12 @@
   FlushIns(ArgLocs, rl_method);
 
   if (base_of_code_ != nullptr) {
+    RegStorage method_start = TargetPtrReg(kArg0);
     // We have been asked to save the address of the method start for later use.
-    setup_method_address_[0] = NewLIR1(kX86StartOfMethod, rs_rX86_ARG0.GetReg());
+    setup_method_address_[0] = NewLIR1(kX86StartOfMethod, method_start.GetReg());
     int displacement = SRegOffset(base_of_code_->s_reg_low);
     // Native pointer - must be natural word size.
-    setup_method_address_[1] = StoreBaseDisp(rs_rX86_SP, displacement, rs_rX86_ARG0,
+    setup_method_address_[1] = StoreBaseDisp(rs_rX86_SP, displacement, method_start,
                                              cu_->target64 ? k64 : k32, kNotVolatile);
   }
 
diff --git a/compiler/dex/quick/x86/codegen_x86.h b/compiler/dex/quick/x86/codegen_x86.h
index afb6184..646da7f 100644
--- a/compiler/dex/quick/x86/codegen_x86.h
+++ b/compiler/dex/quick/x86/codegen_x86.h
@@ -34,9 +34,11 @@
 
   class InToRegStorageX86_64Mapper : public InToRegStorageMapper {
    public:
-    InToRegStorageX86_64Mapper() : cur_core_reg_(0), cur_fp_reg_(0) {}
+    explicit InToRegStorageX86_64Mapper(Mir2Lir* ml) : ml_(ml), cur_core_reg_(0), cur_fp_reg_(0) {}
     virtual ~InToRegStorageX86_64Mapper() {}
     virtual RegStorage GetNextReg(bool is_double_or_float, bool is_wide);
+   protected:
+    Mir2Lir* ml_;
    private:
     int cur_core_reg_;
     int cur_fp_reg_;
@@ -85,7 +87,22 @@
   void MarkGCCard(RegStorage val_reg, RegStorage tgt_addr_reg);
 
   // Required for target - register utilities.
-  RegStorage TargetReg(SpecialTargetRegister reg);
+  RegStorage TargetReg(SpecialTargetRegister reg) OVERRIDE;
+  RegStorage TargetReg32(SpecialTargetRegister reg);
+  RegStorage TargetReg(SpecialTargetRegister symbolic_reg, bool is_wide) OVERRIDE {
+    RegStorage reg = TargetReg32(symbolic_reg);
+    if (is_wide) {
+      return (reg.Is64Bit()) ? reg : As64BitReg(reg);
+    } else {
+      return (reg.Is32Bit()) ? reg : As32BitReg(reg);
+    }
+  }
+  RegStorage TargetRefReg(SpecialTargetRegister symbolic_reg) OVERRIDE {
+    return TargetReg(symbolic_reg, cu_->target64);
+  }
+  RegStorage TargetPtrReg(SpecialTargetRegister symbolic_reg) OVERRIDE {
+    return TargetReg(symbolic_reg, cu_->target64);
+  }
   RegStorage GetArgMappingToPhysicalReg(int arg_num);
   RegStorage GetCoreArgMappingToPhysicalReg(int core_arg_num);
   RegLocation GetReturnAlt();
@@ -388,6 +405,43 @@
   std::vector<uint8_t>* ReturnCallFrameInformation();
 
  protected:
+  // Casting of RegStorage
+  RegStorage As32BitReg(RegStorage reg) {
+    DCHECK(!reg.IsPair());
+    if ((kFailOnSizeError || kReportSizeError) && !reg.Is64Bit()) {
+      if (kFailOnSizeError) {
+        LOG(FATAL) << "Expected 64b register " << reg.GetReg();
+      } else {
+        LOG(WARNING) << "Expected 64b register " << reg.GetReg();
+        return reg;
+      }
+    }
+    RegStorage ret_val = RegStorage(RegStorage::k32BitSolo,
+                                    reg.GetRawBits() & RegStorage::kRegTypeMask);
+    DCHECK_EQ(GetRegInfo(reg)->FindMatchingView(RegisterInfo::k32SoloStorageMask)
+                             ->GetReg().GetReg(),
+              ret_val.GetReg());
+    return ret_val;
+  }
+
+  RegStorage As64BitReg(RegStorage reg) {
+    DCHECK(!reg.IsPair());
+    if ((kFailOnSizeError || kReportSizeError) && !reg.Is32Bit()) {
+      if (kFailOnSizeError) {
+        LOG(FATAL) << "Expected 32b register " << reg.GetReg();
+      } else {
+        LOG(WARNING) << "Expected 32b register " << reg.GetReg();
+        return reg;
+      }
+    }
+    RegStorage ret_val = RegStorage(RegStorage::k64BitSolo,
+                                    reg.GetRawBits() & RegStorage::kRegTypeMask);
+    DCHECK_EQ(GetRegInfo(reg)->FindMatchingView(RegisterInfo::k64SoloStorageMask)
+                             ->GetReg().GetReg(),
+              ret_val.GetReg());
+    return ret_val;
+  }
+
   size_t ComputeSize(const X86EncodingMap* entry, int32_t raw_reg, int32_t raw_index,
                      int32_t raw_base, int32_t displacement);
   void CheckValidByteRegister(const X86EncodingMap* entry, int32_t raw_reg);
diff --git a/compiler/dex/quick/x86/fp_x86.cc b/compiler/dex/quick/x86/fp_x86.cc
index 7454475..4414d7c 100644
--- a/compiler/dex/quick/x86/fp_x86.cc
+++ b/compiler/dex/quick/x86/fp_x86.cc
@@ -145,12 +145,12 @@
     } else {
       // It must have been register promoted if it is not a temp but is still in physical
       // register. Since we need it to be in memory to convert, we place it there now.
-      StoreBaseDisp(TargetReg(kSp), src_v_reg_offset, rl_src.reg, k64, kNotVolatile);
+      StoreBaseDisp(rs_rX86_SP, src_v_reg_offset, rl_src.reg, k64, kNotVolatile);
     }
   }
 
   // Push the source virtual register onto the x87 stack.
-  LIR *fild64 = NewLIR2NoDest(kX86Fild64M, TargetReg(kSp).GetReg(),
+  LIR *fild64 = NewLIR2NoDest(kX86Fild64M, rs_rX86_SP.GetReg(),
                               src_v_reg_offset + LOWORD_OFFSET);
   AnnotateDalvikRegAccess(fild64, (src_v_reg_offset + LOWORD_OFFSET) >> 2,
                           true /* is_load */, true /* is64bit */);
@@ -158,7 +158,7 @@
   // Now pop off x87 stack and store it in the destination VR's stack location.
   int opcode = is_double ? kX86Fstp64M : kX86Fstp32M;
   int displacement = is_double ? dest_v_reg_offset + LOWORD_OFFSET : dest_v_reg_offset;
-  LIR *fstp = NewLIR2NoDest(opcode, TargetReg(kSp).GetReg(), displacement);
+  LIR *fstp = NewLIR2NoDest(opcode, rs_rX86_SP.GetReg(), displacement);
   AnnotateDalvikRegAccess(fstp, displacement >> 2, false /* is_load */, is_double);
 
   /*
@@ -179,11 +179,11 @@
      */
     rl_result = EvalLoc(rl_dest, kFPReg, true);
     if (is_double) {
-      LoadBaseDisp(TargetReg(kSp), dest_v_reg_offset, rl_result.reg, k64, kNotVolatile);
+      LoadBaseDisp(rs_rX86_SP, dest_v_reg_offset, rl_result.reg, k64, kNotVolatile);
 
       StoreFinalValueWide(rl_dest, rl_result);
     } else {
-      Load32Disp(TargetReg(kSp), dest_v_reg_offset, rl_result.reg);
+      Load32Disp(rs_rX86_SP, dest_v_reg_offset, rl_result.reg);
 
       StoreFinalValue(rl_dest, rl_result);
     }
@@ -364,7 +364,7 @@
     } else {
       // It must have been register promoted if it is not a temp but is still in physical
       // register. Since we need it to be in memory to convert, we place it there now.
-      StoreBaseDisp(TargetReg(kSp), src1_v_reg_offset, rl_src1.reg, is_double ? k64 : k32,
+      StoreBaseDisp(rs_rX86_SP, src1_v_reg_offset, rl_src1.reg, is_double ? k64 : k32,
                     kNotVolatile);
     }
   }
@@ -375,7 +375,7 @@
       FlushSpecificReg(reg_info);
       ResetDef(rl_src2.reg);
     } else {
-      StoreBaseDisp(TargetReg(kSp), src2_v_reg_offset, rl_src2.reg, is_double ? k64 : k32,
+      StoreBaseDisp(rs_rX86_SP, src2_v_reg_offset, rl_src2.reg, is_double ? k64 : k32,
                     kNotVolatile);
     }
   }
@@ -383,12 +383,12 @@
   int fld_opcode = is_double ? kX86Fld64M : kX86Fld32M;
 
   // Push the source virtual registers onto the x87 stack.
-  LIR *fld_2 = NewLIR2NoDest(fld_opcode, TargetReg(kSp).GetReg(),
+  LIR *fld_2 = NewLIR2NoDest(fld_opcode, rs_rX86_SP.GetReg(),
                              src2_v_reg_offset + LOWORD_OFFSET);
   AnnotateDalvikRegAccess(fld_2, (src2_v_reg_offset + LOWORD_OFFSET) >> 2,
                           true /* is_load */, is_double /* is64bit */);
 
-  LIR *fld_1 = NewLIR2NoDest(fld_opcode, TargetReg(kSp).GetReg(),
+  LIR *fld_1 = NewLIR2NoDest(fld_opcode, rs_rX86_SP.GetReg(),
                              src1_v_reg_offset + LOWORD_OFFSET);
   AnnotateDalvikRegAccess(fld_1, (src1_v_reg_offset + LOWORD_OFFSET) >> 2,
                           true /* is_load */, is_double /* is64bit */);
@@ -417,7 +417,7 @@
   // Now store result in the destination VR's stack location.
   int displacement = dest_v_reg_offset + LOWORD_OFFSET;
   int opcode = is_double ? kX86Fst64M : kX86Fst32M;
-  LIR *fst = NewLIR2NoDest(opcode, TargetReg(kSp).GetReg(), displacement);
+  LIR *fst = NewLIR2NoDest(opcode, rs_rX86_SP.GetReg(), displacement);
   AnnotateDalvikRegAccess(fst, displacement >> 2, false /* is_load */, is_double /* is64bit */);
 
   // Pop ST(1) and ST(0).
@@ -436,10 +436,10 @@
   if (rl_result.location == kLocPhysReg) {
     rl_result = EvalLoc(rl_dest, kFPReg, true);
     if (is_double) {
-      LoadBaseDisp(TargetReg(kSp), dest_v_reg_offset, rl_result.reg, k64, kNotVolatile);
+      LoadBaseDisp(rs_rX86_SP, dest_v_reg_offset, rl_result.reg, k64, kNotVolatile);
       StoreFinalValueWide(rl_dest, rl_result);
     } else {
-      Load32Disp(TargetReg(kSp), dest_v_reg_offset, rl_result.reg);
+      Load32Disp(rs_rX86_SP, dest_v_reg_offset, rl_result.reg);
       StoreFinalValue(rl_dest, rl_result);
     }
   }
@@ -627,7 +627,7 @@
     // Operate directly into memory.
     int displacement = SRegOffset(rl_dest.s_reg_low);
     ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
-    LIR *lir = NewLIR3(kX86And32MI, TargetReg(kSp).GetReg(), displacement, 0x7fffffff);
+    LIR *lir = NewLIR3(kX86And32MI, rs_rX86_SP.GetReg(), displacement, 0x7fffffff);
     AnnotateDalvikRegAccess(lir, displacement >> 2, false /*is_load */, false /* is_64bit */);
     AnnotateDalvikRegAccess(lir, displacement >> 2, true /* is_load */, false /* is_64bit*/);
     return true;
@@ -682,7 +682,7 @@
     // Operate directly into memory.
     int displacement = SRegOffset(rl_dest.s_reg_low);
     ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
-    LIR *lir = NewLIR3(kX86And32MI, TargetReg(kSp).GetReg(), displacement  + HIWORD_OFFSET, 0x7fffffff);
+    LIR *lir = NewLIR3(kX86And32MI, rs_rX86_SP.GetReg(), displacement  + HIWORD_OFFSET, 0x7fffffff);
     AnnotateDalvikRegAccess(lir, (displacement + HIWORD_OFFSET) >> 2, true /* is_load */, true /* is_64bit*/);
     AnnotateDalvikRegAccess(lir, (displacement + HIWORD_OFFSET) >> 2, false /*is_load */, true /* is_64bit */);
     return true;
diff --git a/compiler/dex/quick/x86/int_x86.cc b/compiler/dex/quick/x86/int_x86.cc
index 14a18e5..7a4ea26 100644
--- a/compiler/dex/quick/x86/int_x86.cc
+++ b/compiler/dex/quick/x86/int_x86.cc
@@ -93,7 +93,7 @@
 }
 
 LIR* X86Mir2Lir::OpCmpBranch(ConditionCode cond, RegStorage src1, RegStorage src2, LIR* target) {
-  NewLIR2(kX86Cmp32RR, src1.GetReg(), src2.GetReg());
+  NewLIR2(src1.Is64Bit() ? kX86Cmp64RR : kX86Cmp32RR, src1.GetReg(), src2.GetReg());
   X86ConditionCode cc = X86ConditionEncoding(cond);
   LIR* branch = NewLIR2(kX86Jcc8, 0 /* lir operand for Jcc offset */ ,
                         cc);
@@ -105,9 +105,13 @@
                                 int check_value, LIR* target) {
   if ((check_value == 0) && (cond == kCondEq || cond == kCondNe)) {
     // TODO: when check_value == 0 and reg is rCX, use the jcxz/nz opcode
-    NewLIR2(kX86Test32RR, reg.GetReg(), reg.GetReg());
+    NewLIR2(reg.Is64Bit() ? kX86Test64RR: kX86Test32RR, reg.GetReg(), reg.GetReg());
   } else {
-    NewLIR2(IS_SIMM8(check_value) ? kX86Cmp32RI8 : kX86Cmp32RI, reg.GetReg(), check_value);
+    if (reg.Is64Bit()) {
+      NewLIR2(IS_SIMM8(check_value) ? kX86Cmp64RI8 : kX86Cmp64RI, reg.GetReg(), check_value);
+    } else {
+      NewLIR2(IS_SIMM8(check_value) ? kX86Cmp32RI8 : kX86Cmp32RI, reg.GetReg(), check_value);
+    }
   }
   X86ConditionCode cc = X86ConditionEncoding(cond);
   LIR* branch = NewLIR2(kX86Jcc8, 0 /* lir operand for Jcc offset */ , cc);
@@ -241,7 +245,7 @@
     // FIXME: depending on how you use registers you could get a false != mismatch when dealing
     // with different views of the same underlying physical resource (i.e. solo32 vs. solo64).
     const bool result_reg_same_as_src =
-        (rl_src.location == kLocPhysReg && rl_src.reg.GetReg() == rl_result.reg.GetReg());
+        (rl_src.location == kLocPhysReg && rl_src.reg.GetRegNum() == rl_result.reg.GetRegNum());
     const bool true_zero_case = (true_val == 0 && false_val != 0 && !result_reg_same_as_src);
     const bool false_zero_case = (false_val == 0 && true_val != 0 && !result_reg_same_as_src);
     const bool catch_all_case = !(true_zero_case || false_zero_case);
@@ -846,14 +850,14 @@
     ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
     const size_t push_offset = (push_si ? 4u : 0u) + (push_di ? 4u : 0u);
     if (!obj_in_si && !obj_in_di) {
-      LoadWordDisp(TargetReg(kSp), SRegOffset(rl_src_obj.s_reg_low) + push_offset, rs_obj);
+      LoadWordDisp(rs_rX86_SP, SRegOffset(rl_src_obj.s_reg_low) + push_offset, rs_obj);
       // Dalvik register annotation in LoadBaseIndexedDisp() used wrong offset. Fix it.
       DCHECK(!DECODE_ALIAS_INFO_WIDE(last_lir_insn_->flags.alias_info));
       int reg_id = DECODE_ALIAS_INFO_REG(last_lir_insn_->flags.alias_info) - push_offset / 4u;
       AnnotateDalvikRegAccess(last_lir_insn_, reg_id, true, false);
     }
     if (!off_in_si && !off_in_di) {
-      LoadWordDisp(TargetReg(kSp), SRegOffset(rl_src_offset.s_reg_low) + push_offset, rs_off);
+      LoadWordDisp(rs_rX86_SP, SRegOffset(rl_src_offset.s_reg_low) + push_offset, rs_off);
       // Dalvik register annotation in LoadBaseIndexedDisp() used wrong offset. Fix it.
       DCHECK(!DECODE_ALIAS_INFO_WIDE(last_lir_insn_->flags.alias_info));
       int reg_id = DECODE_ALIAS_INFO_REG(last_lir_insn_->flags.alias_info) - push_offset / 4u;
@@ -1008,23 +1012,23 @@
       RegStorage new_index = index_;
       // Move index out of kArg1, either directly to kArg0, or to kArg2.
       // TODO: clean-up to check not a number but with type
-      if (index_.GetRegNum() == m2l_->TargetReg(kArg1).GetRegNum()) {
-        if (array_base_.GetRegNum() == m2l_->TargetReg(kArg0).GetRegNum()) {
-          m2l_->OpRegCopy(m2l_->TargetReg(kArg2), index_);
-          new_index = m2l_->TargetReg(kArg2);
+      if (index_ == m2l_->TargetReg(kArg1, false)) {
+        if (array_base_ == m2l_->TargetRefReg(kArg0)) {
+          m2l_->OpRegCopy(m2l_->TargetReg(kArg2, false), index_);
+          new_index = m2l_->TargetReg(kArg2, false);
         } else {
-          m2l_->OpRegCopy(m2l_->TargetReg(kArg0), index_);
-          new_index = m2l_->TargetReg(kArg0);
+          m2l_->OpRegCopy(m2l_->TargetReg(kArg0, false), index_);
+          new_index = m2l_->TargetReg(kArg0, false);
         }
       }
       // Load array length to kArg1.
-      m2l_->OpRegMem(kOpMov, m2l_->TargetReg(kArg1), array_base_, len_offset_);
+      m2l_->OpRegMem(kOpMov, m2l_->TargetReg(kArg1, false), array_base_, len_offset_);
       if (cu_->target64) {
         m2l_->CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(8, pThrowArrayBounds),
-                                      new_index, m2l_->TargetReg(kArg1), true);
+                                      new_index, m2l_->TargetReg(kArg1, false), true);
       } else {
         m2l_->CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(4, pThrowArrayBounds),
-                                      new_index, m2l_->TargetReg(kArg1), true);
+                                      new_index, m2l_->TargetReg(kArg1, false), true);
       }
     }
 
@@ -1057,14 +1061,14 @@
       GenerateTargetLabel(kPseudoThrowTarget);
 
       // Load array length to kArg1.
-      m2l_->OpRegMem(kOpMov, m2l_->TargetReg(kArg1), array_base_, len_offset_);
-      m2l_->LoadConstant(m2l_->TargetReg(kArg0), index_);
+      m2l_->OpRegMem(kOpMov, m2l_->TargetReg(kArg1, false), array_base_, len_offset_);
+      m2l_->LoadConstant(m2l_->TargetReg(kArg0, false), index_);
       if (cu_->target64) {
         m2l_->CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(8, pThrowArrayBounds),
-                                      m2l_->TargetReg(kArg0), m2l_->TargetReg(kArg1), true);
+                                      m2l_->TargetReg(kArg0, false), m2l_->TargetReg(kArg1, false), true);
       } else {
         m2l_->CallRuntimeHelperRegReg(QUICK_ENTRYPOINT_OFFSET(4, pThrowArrayBounds),
-                                      m2l_->TargetReg(kArg0), m2l_->TargetReg(kArg1), true);
+                                      m2l_->TargetReg(kArg0, false), m2l_->TargetReg(kArg1, false), true);
       }
     }
 
@@ -1406,7 +1410,7 @@
   // RHS is in memory.
   DCHECK((rl_src.location == kLocDalvikFrame) ||
          (rl_src.location == kLocCompilerTemp));
-  int r_base = TargetReg(kSp).GetReg();
+  int r_base = rs_rX86_SP.GetReg();
   int displacement = SRegOffset(rl_src.s_reg_low);
 
   ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
@@ -1440,7 +1444,7 @@
 
   // Operate directly into memory.
   X86OpCode x86op = GetOpcode(op, rl_dest, rl_src, false);
-  int r_base = TargetReg(kSp).GetReg();
+  int r_base = rs_rX86_SP.GetReg();
   int displacement = SRegOffset(rl_dest.s_reg_low);
 
   ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
@@ -2122,7 +2126,7 @@
 
     if ((rl_dest.location == kLocDalvikFrame) ||
         (rl_dest.location == kLocCompilerTemp)) {
-      int r_base = TargetReg(kSp).GetReg();
+      int r_base = rs_rX86_SP.GetReg();
       int displacement = SRegOffset(rl_dest.s_reg_low);
 
       ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
@@ -2153,7 +2157,7 @@
   // Can we just do this into memory?
   if ((rl_dest.location == kLocDalvikFrame) ||
       (rl_dest.location == kLocCompilerTemp)) {
-    int r_base = TargetReg(kSp).GetReg();
+    int r_base = rs_rX86_SP.GetReg();
     int displacement = SRegOffset(rl_dest.s_reg_low);
 
     ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
@@ -2271,7 +2275,8 @@
   RegStorage result_reg = rl_result.reg;
 
   // For 32-bit, SETcc only works with EAX..EDX.
-  if (result_reg == object.reg || !IsByteRegister(result_reg)) {
+  RegStorage object_32reg = object.reg.Is64Bit() ? As32BitReg(object.reg) : object.reg;
+  if (result_reg == object_32reg || !IsByteRegister(result_reg)) {
     result_reg = AllocateByteRegister();
   }
 
@@ -2337,8 +2342,10 @@
   FlushAllRegs();
   // May generate a call - use explicit registers.
   LockCallTemps();
-  LoadCurrMethodDirect(TargetReg(kArg1));  // kArg1 gets current Method*.
-  RegStorage class_reg = TargetReg(kArg2);  // kArg2 will hold the Class*.
+  RegStorage method_reg = TargetRefReg(kArg1);  // kArg1 gets current Method*.
+  LoadCurrMethodDirect(method_reg);
+  RegStorage class_reg = TargetRefReg(kArg2);  // kArg2 will hold the Class*.
+  RegStorage ref_reg = TargetRefReg(kArg0);  // kArg2 will hold the ref.
   // Reference must end up in kArg0.
   if (needs_access_check) {
     // Check we have access to type_idx and if not throw IllegalAccessError,
@@ -2350,16 +2357,16 @@
       CallRuntimeHelperImm(QUICK_ENTRYPOINT_OFFSET(4, pInitializeTypeAndVerifyAccess),
                            type_idx, true);
     }
-    OpRegCopy(class_reg, TargetReg(kRet0));
-    LoadValueDirectFixed(rl_src, TargetReg(kArg0));
+    OpRegCopy(class_reg, TargetRefReg(kRet0));
+    LoadValueDirectFixed(rl_src, ref_reg);
   } else if (use_declaring_class) {
-    LoadValueDirectFixed(rl_src, TargetReg(kArg0));
-    LoadRefDisp(TargetReg(kArg1), mirror::ArtMethod::DeclaringClassOffset().Int32Value(),
+    LoadValueDirectFixed(rl_src, ref_reg);
+    LoadRefDisp(method_reg, mirror::ArtMethod::DeclaringClassOffset().Int32Value(),
                 class_reg, kNotVolatile);
   } else {
     // Load dex cache entry into class_reg (kArg2).
-    LoadValueDirectFixed(rl_src, TargetReg(kArg0));
-    LoadRefDisp(TargetReg(kArg1), mirror::ArtMethod::DexCacheResolvedTypesOffset().Int32Value(),
+    LoadValueDirectFixed(rl_src, ref_reg);
+    LoadRefDisp(method_reg, mirror::ArtMethod::DexCacheResolvedTypesOffset().Int32Value(),
                 class_reg, kNotVolatile);
     int32_t offset_of_type =
         mirror::Array::DataOffset(sizeof(mirror::HeapReference<mirror::Class*>)).Int32Value() +
@@ -2374,8 +2381,8 @@
       } else {
         CallRuntimeHelperImm(QUICK_ENTRYPOINT_OFFSET(4, pInitializeType), type_idx, true);
       }
-      OpRegCopy(TargetReg(kArg2), TargetReg(kRet0));  // Align usage with fast path.
-      LoadValueDirectFixed(rl_src, TargetReg(kArg0));  /* Reload Ref. */
+      OpRegCopy(class_reg, TargetRefReg(kRet0));  // Align usage with fast path.
+      LoadValueDirectFixed(rl_src, ref_reg);  /* Reload Ref. */
       // Rejoin code paths
       LIR* hop_target = NewLIR0(kPseudoTargetLabel);
       hop_branch->target = hop_target;
@@ -2386,33 +2393,34 @@
 
   // On x86-64 kArg0 is not EAX, so we have to copy ref from kArg0 to EAX.
   if (cu_->target64) {
-    OpRegCopy(rl_result.reg, TargetReg(kArg0));
+    OpRegCopy(rl_result.reg, ref_reg);
   }
 
   // For 32-bit, SETcc only works with EAX..EDX.
   DCHECK_LT(rl_result.reg.GetRegNum(), 4);
 
   // Is the class NULL?
-  LIR* branch1 = OpCmpImmBranch(kCondEq, TargetReg(kArg0), 0, NULL);
+  LIR* branch1 = OpCmpImmBranch(kCondEq, ref_reg, 0, NULL);
 
+  RegStorage ref_class_reg = TargetRefReg(kArg1);  // kArg2 will hold the Class*.
   /* Load object->klass_. */
   DCHECK_EQ(mirror::Object::ClassOffset().Int32Value(), 0);
-  LoadRefDisp(TargetReg(kArg0),  mirror::Object::ClassOffset().Int32Value(), TargetReg(kArg1),
+  LoadRefDisp(ref_reg,  mirror::Object::ClassOffset().Int32Value(), ref_class_reg,
               kNotVolatile);
   /* kArg0 is ref, kArg1 is ref->klass_, kArg2 is class. */
   LIR* branchover = nullptr;
   if (type_known_final) {
     // Ensure top 3 bytes of result are 0.
     LoadConstant(rl_result.reg, 0);
-    OpRegReg(kOpCmp, TargetReg(kArg1), TargetReg(kArg2));
+    OpRegReg(kOpCmp, ref_class_reg, class_reg);
     // Set the low byte of the result to 0 or 1 from the compare condition code.
     NewLIR2(kX86Set8R, rl_result.reg.GetReg(), kX86CondEq);
   } else {
     if (!type_known_abstract) {
       LoadConstant(rl_result.reg, 1);     // Assume result succeeds.
-      branchover = OpCmpBranch(kCondEq, TargetReg(kArg1), TargetReg(kArg2), NULL);
+      branchover = OpCmpBranch(kCondEq, ref_class_reg, class_reg, NULL);
     }
-    OpRegCopy(TargetReg(kArg0), TargetReg(kArg2));
+    OpRegCopy(TargetRefReg(kArg0), class_reg);
     if (cu_->target64) {
       OpThreadMem(kOpBlx, QUICK_ENTRYPOINT_OFFSET(8, pInstanceofNonTrivial));
     } else {
@@ -2552,7 +2560,7 @@
   } else {
     if (shift_op) {
       // X86 doesn't require masking and must use ECX.
-      RegStorage t_reg = TargetReg(kCount);  // rCX
+      RegStorage t_reg = TargetReg(kCount, false);  // rCX
       LoadValueDirectFixed(rl_rhs, t_reg);
       if (is_two_addr) {
         // Can we do this directly into memory?
@@ -2740,7 +2748,7 @@
   }
 
   // X86 doesn't require masking and must use ECX.
-  RegStorage t_reg = TargetReg(kCount);  // rCX
+  RegStorage t_reg = TargetReg(kCount, false);  // rCX
   LoadValueDirectFixed(rl_shift, t_reg);
   if (is_two_addr) {
     // Can we do this directly into memory?
diff --git a/compiler/dex/quick/x86/target_x86.cc b/compiler/dex/quick/x86/target_x86.cc
index 730a271..f80e200 100644
--- a/compiler/dex/quick/x86/target_x86.cc
+++ b/compiler/dex/quick/x86/target_x86.cc
@@ -154,8 +154,7 @@
 }
 
 RegLocation X86Mir2Lir::LocCReturnRef() {
-  // FIXME: return x86_loc_c_return_wide for x86_64 when wide refs supported.
-  return x86_loc_c_return;
+  return cu_->target64 ? x86_64_loc_c_return_ref : x86_loc_c_return_ref;
 }
 
 RegLocation X86Mir2Lir::LocCReturnWide() {
@@ -170,8 +169,8 @@
   return x86_loc_c_return_double;
 }
 
-// Return a target-dependent special register.
-RegStorage X86Mir2Lir::TargetReg(SpecialTargetRegister reg) {
+// Return a target-dependent special register for 32-bit.
+RegStorage X86Mir2Lir::TargetReg32(SpecialTargetRegister reg) {
   RegStorage res_reg = RegStorage::InvalidReg();
   switch (reg) {
     case kSelf: res_reg = RegStorage::InvalidReg(); break;
@@ -204,6 +203,11 @@
   return res_reg;
 }
 
+RegStorage X86Mir2Lir::TargetReg(SpecialTargetRegister reg) {
+  LOG(FATAL) << "Do not use this function!!!";
+  return RegStorage::InvalidReg();
+}
+
 /*
  * Decode the register id.
  */
@@ -832,7 +836,7 @@
       (rl_dest.location == kLocCompilerTemp)) {
     int32_t val_lo = Low32Bits(value);
     int32_t val_hi = High32Bits(value);
-    int r_base = TargetReg(kSp).GetReg();
+    int r_base = rs_rX86_SP.GetReg();
     int displacement = SRegOffset(rl_dest.s_reg_low);
 
     ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
@@ -887,7 +891,7 @@
   uintptr_t target_method_id_ptr = reinterpret_cast<uintptr_t>(&target_method_id);
 
   // Generate the move instruction with the unique pointer and save index, dex_file, and type.
-  LIR *move = RawLIR(current_dalvik_offset_, kX86Mov32RI, TargetReg(symbolic_reg).GetReg(),
+  LIR *move = RawLIR(current_dalvik_offset_, kX86Mov32RI, TargetReg(symbolic_reg, false).GetReg(),
                      static_cast<int>(target_method_id_ptr), target_method_idx,
                      WrapPointer(const_cast<DexFile*>(target_dex_file)), type);
   AppendLIR(move);
@@ -904,7 +908,7 @@
   uintptr_t ptr = reinterpret_cast<uintptr_t>(&id);
 
   // Generate the move instruction with the unique pointer and save index and type.
-  LIR *move = RawLIR(current_dalvik_offset_, kX86Mov32RI, TargetReg(symbolic_reg).GetReg(),
+  LIR *move = RawLIR(current_dalvik_offset_, kX86Mov32RI, TargetReg(symbolic_reg, false).GetReg(),
                      static_cast<int>(ptr), type_idx);
   AppendLIR(move);
   class_type_address_insns_.Insert(move);
@@ -1746,29 +1750,22 @@
 
 // ------------ ABI support: mapping of args to physical registers -------------
 RegStorage X86Mir2Lir::InToRegStorageX86_64Mapper::GetNextReg(bool is_double_or_float, bool is_wide) {
-  const RegStorage coreArgMappingToPhysicalReg[] = {rs_rX86_ARG1, rs_rX86_ARG2, rs_rX86_ARG3, rs_rX86_ARG4, rs_rX86_ARG5};
-  const int coreArgMappingToPhysicalRegSize = sizeof(coreArgMappingToPhysicalReg) / sizeof(RegStorage);
-  const RegStorage fpArgMappingToPhysicalReg[] = {rs_rX86_FARG0, rs_rX86_FARG1, rs_rX86_FARG2, rs_rX86_FARG3,
-                                                  rs_rX86_FARG4, rs_rX86_FARG5, rs_rX86_FARG6, rs_rX86_FARG7};
-  const int fpArgMappingToPhysicalRegSize = sizeof(fpArgMappingToPhysicalReg) / sizeof(RegStorage);
+  const SpecialTargetRegister coreArgMappingToPhysicalReg[] = {kArg1, kArg2, kArg3, kArg4, kArg5};
+  const int coreArgMappingToPhysicalRegSize = sizeof(coreArgMappingToPhysicalReg) / sizeof(SpecialTargetRegister);
+  const SpecialTargetRegister fpArgMappingToPhysicalReg[] = {kFArg0, kFArg1, kFArg2, kFArg3,
+                                                  kFArg4, kFArg5, kFArg6, kFArg7};
+  const int fpArgMappingToPhysicalRegSize = sizeof(fpArgMappingToPhysicalReg) / sizeof(SpecialTargetRegister);
 
-  RegStorage result = RegStorage::InvalidReg();
   if (is_double_or_float) {
     if (cur_fp_reg_ < fpArgMappingToPhysicalRegSize) {
-      result = fpArgMappingToPhysicalReg[cur_fp_reg_++];
-      if (result.Valid()) {
-        result = is_wide ? RegStorage::FloatSolo64(result.GetReg()) : RegStorage::FloatSolo32(result.GetReg());
-      }
+      return ml_->TargetReg(fpArgMappingToPhysicalReg[cur_fp_reg_++], is_wide);
     }
   } else {
     if (cur_core_reg_ < coreArgMappingToPhysicalRegSize) {
-      result = coreArgMappingToPhysicalReg[cur_core_reg_++];
-      if (result.Valid()) {
-        result = is_wide ? RegStorage::Solo64(result.GetReg()) : RegStorage::Solo32(result.GetReg());
-      }
+      return ml_->TargetReg(coreArgMappingToPhysicalReg[cur_core_reg_++], is_wide);
     }
   }
-  return result;
+  return RegStorage::InvalidReg();
 }
 
 RegStorage X86Mir2Lir::InToRegStorageMapping::Get(int in_position) {
@@ -1806,7 +1803,7 @@
     int start_vreg = cu_->num_dalvik_registers - cu_->num_ins;
     RegLocation* arg_locs = &mir_graph_->reg_location_[start_vreg];
 
-    InToRegStorageX86_64Mapper mapper;
+    InToRegStorageX86_64Mapper mapper(this);
     in_to_reg_storage_mapping_.Initialize(arg_locs, cu_->num_ins, &mapper);
   }
   return in_to_reg_storage_mapping_.Get(arg_num);
@@ -1847,13 +1844,13 @@
 
   RegLocation rl_src = rl_method;
   rl_src.location = kLocPhysReg;
-  rl_src.reg = TargetReg(kArg0);
+  rl_src.reg = TargetRefReg(kArg0);
   rl_src.home = false;
   MarkLive(rl_src);
   StoreValue(rl_method, rl_src);
   // If Method* has been promoted, explicitly flush
   if (rl_method.location == kLocPhysReg) {
-    StoreRefDisp(TargetReg(kSp), 0, TargetReg(kArg0), kNotVolatile);
+    StoreRefDisp(rs_rX86_SP, 0, As32BitReg(TargetRefReg(kArg0)), kNotVolatile);
   }
 
   if (cu_->num_ins == 0) {
@@ -1890,9 +1887,9 @@
       } else {
         // Needs flush.
         if (t_loc->ref) {
-          StoreRefDisp(TargetReg(kSp), SRegOffset(start_vreg + i), reg, kNotVolatile);
+          StoreRefDisp(rs_rX86_SP, SRegOffset(start_vreg + i), reg, kNotVolatile);
         } else {
-          StoreBaseDisp(TargetReg(kSp), SRegOffset(start_vreg + i), reg, t_loc->wide ? k64 : k32,
+          StoreBaseDisp(rs_rX86_SP, SRegOffset(start_vreg + i), reg, t_loc->wide ? k64 : k32,
                         kNotVolatile);
         }
       }
@@ -1900,9 +1897,9 @@
       // If arriving in frame & promoted.
       if (t_loc->location == kLocPhysReg) {
         if (t_loc->ref) {
-          LoadRefDisp(TargetReg(kSp), SRegOffset(start_vreg + i), t_loc->reg, kNotVolatile);
+          LoadRefDisp(rs_rX86_SP, SRegOffset(start_vreg + i), t_loc->reg, kNotVolatile);
         } else {
-          LoadBaseDisp(TargetReg(kSp), SRegOffset(start_vreg + i), t_loc->reg,
+          LoadBaseDisp(rs_rX86_SP, SRegOffset(start_vreg + i), t_loc->reg,
                        t_loc->wide ? k64 : k32, kNotVolatile);
         }
       }
@@ -1974,7 +1971,7 @@
 
   const int start_index = skip_this ? 1 : 0;
 
-  InToRegStorageX86_64Mapper mapper;
+  InToRegStorageX86_64Mapper mapper(this);
   InToRegStorageMapping in_to_reg_storage_mapping;
   in_to_reg_storage_mapping.Initialize(info->args, info->num_arg_words, &mapper);
   const int last_mapped_in = in_to_reg_storage_mapping.GetMaxMappedIn();
@@ -1993,14 +1990,14 @@
         loc = UpdateLocWide(loc);
         if (loc.location == kLocPhysReg) {
           ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
-          StoreBaseDisp(TargetReg(kSp), SRegOffset(loc.s_reg_low), loc.reg, k64, kNotVolatile);
+          StoreBaseDisp(rs_rX86_SP, SRegOffset(loc.s_reg_low), loc.reg, k64, kNotVolatile);
         }
         next_arg += 2;
       } else {
         loc = UpdateLoc(loc);
         if (loc.location == kLocPhysReg) {
           ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
-          StoreBaseDisp(TargetReg(kSp), SRegOffset(loc.s_reg_low), loc.reg, k32, kNotVolatile);
+          StoreBaseDisp(rs_rX86_SP, SRegOffset(loc.s_reg_low), loc.reg, k32, kNotVolatile);
         }
         next_arg++;
       }
@@ -2057,23 +2054,23 @@
 
         ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
         if (src_is_16b_aligned) {
-          ld1 = OpMovRegMem(temp, TargetReg(kSp), current_src_offset, kMovA128FP);
+          ld1 = OpMovRegMem(temp, rs_rX86_SP, current_src_offset, kMovA128FP);
         } else if (src_is_8b_aligned) {
-          ld1 = OpMovRegMem(temp, TargetReg(kSp), current_src_offset, kMovLo128FP);
-          ld2 = OpMovRegMem(temp, TargetReg(kSp), current_src_offset + (bytes_to_move >> 1),
+          ld1 = OpMovRegMem(temp, rs_rX86_SP, current_src_offset, kMovLo128FP);
+          ld2 = OpMovRegMem(temp, rs_rX86_SP, current_src_offset + (bytes_to_move >> 1),
                             kMovHi128FP);
         } else {
-          ld1 = OpMovRegMem(temp, TargetReg(kSp), current_src_offset, kMovU128FP);
+          ld1 = OpMovRegMem(temp, rs_rX86_SP, current_src_offset, kMovU128FP);
         }
 
         if (dest_is_16b_aligned) {
-          st1 = OpMovMemReg(TargetReg(kSp), current_dest_offset, temp, kMovA128FP);
+          st1 = OpMovMemReg(rs_rX86_SP, current_dest_offset, temp, kMovA128FP);
         } else if (dest_is_8b_aligned) {
-          st1 = OpMovMemReg(TargetReg(kSp), current_dest_offset, temp, kMovLo128FP);
-          st2 = OpMovMemReg(TargetReg(kSp), current_dest_offset + (bytes_to_move >> 1),
+          st1 = OpMovMemReg(rs_rX86_SP, current_dest_offset, temp, kMovLo128FP);
+          st2 = OpMovMemReg(rs_rX86_SP, current_dest_offset + (bytes_to_move >> 1),
                             temp, kMovHi128FP);
         } else {
-          st1 = OpMovMemReg(TargetReg(kSp), current_dest_offset, temp, kMovU128FP);
+          st1 = OpMovMemReg(rs_rX86_SP, current_dest_offset, temp, kMovU128FP);
         }
 
         // TODO If we could keep track of aliasing information for memory accesses that are wider
@@ -2107,11 +2104,11 @@
 
         // Instead of allocating a new temp, simply reuse one of the registers being used
         // for argument passing.
-        RegStorage temp = TargetReg(kArg3);
+        RegStorage temp = TargetReg(kArg3, false);
 
         // Now load the argument VR and store to the outs.
-        Load32Disp(TargetReg(kSp), current_src_offset, temp);
-        Store32Disp(TargetReg(kSp), current_dest_offset, temp);
+        Load32Disp(rs_rX86_SP, current_src_offset, temp);
+        Store32Disp(rs_rX86_SP, current_dest_offset, temp);
       }
 
       current_src_offset += bytes_to_move;
@@ -2123,8 +2120,8 @@
 
   // Now handle rest not registers if they are
   if (in_to_reg_storage_mapping.IsThereStackMapped()) {
-    RegStorage regSingle = TargetReg(kArg2);
-    RegStorage regWide = RegStorage::Solo64(TargetReg(kArg3).GetReg());
+    RegStorage regSingle = TargetReg(kArg2, false);
+    RegStorage regWide = TargetReg(kArg3, true);
     for (int i = start_index;
          i < last_mapped_in + size_of_the_last_mapped + regs_left_to_pass_via_stack; i++) {
       RegLocation rl_arg = info->args[i];
@@ -2137,17 +2134,17 @@
           ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
           if (rl_arg.wide) {
             if (rl_arg.location == kLocPhysReg) {
-              StoreBaseDisp(TargetReg(kSp), out_offset, rl_arg.reg, k64, kNotVolatile);
+              StoreBaseDisp(rs_rX86_SP, out_offset, rl_arg.reg, k64, kNotVolatile);
             } else {
               LoadValueDirectWideFixed(rl_arg, regWide);
-              StoreBaseDisp(TargetReg(kSp), out_offset, regWide, k64, kNotVolatile);
+              StoreBaseDisp(rs_rX86_SP, out_offset, regWide, k64, kNotVolatile);
             }
           } else {
             if (rl_arg.location == kLocPhysReg) {
-              StoreBaseDisp(TargetReg(kSp), out_offset, rl_arg.reg, k32, kNotVolatile);
+              StoreBaseDisp(rs_rX86_SP, out_offset, rl_arg.reg, k32, kNotVolatile);
             } else {
               LoadValueDirectFixed(rl_arg, regSingle);
-              StoreBaseDisp(TargetReg(kSp), out_offset, regSingle, k32, kNotVolatile);
+              StoreBaseDisp(rs_rX86_SP, out_offset, regSingle, k32, kNotVolatile);
             }
           }
         }
@@ -2183,13 +2180,13 @@
                            direct_code, direct_method, type);
   if (pcrLabel) {
     if (cu_->compiler_driver->GetCompilerOptions().GetExplicitNullChecks()) {
-      *pcrLabel = GenExplicitNullCheck(TargetReg(kArg1), info->opt_flags);
+      *pcrLabel = GenExplicitNullCheck(TargetRefReg(kArg1), info->opt_flags);
     } else {
       *pcrLabel = nullptr;
       // In lieu of generating a check for kArg1 being null, we need to
       // perform a load when doing implicit checks.
       RegStorage tmp = AllocTemp();
-      Load32Disp(TargetReg(kArg1), 0, tmp);
+      Load32Disp(TargetRefReg(kArg1), 0, tmp);
       MarkPossibleNullPointerException(info->opt_flags);
       FreeTemp(tmp);
     }
diff --git a/compiler/dex/quick/x86/x86_lir.h b/compiler/dex/quick/x86/x86_lir.h
index 7ff4f72..a52e842 100644
--- a/compiler/dex/quick/x86/x86_lir.h
+++ b/compiler/dex/quick/x86/x86_lir.h
@@ -353,6 +353,12 @@
 const RegLocation x86_loc_c_return_wide
     {kLocPhysReg, 1, 0, 0, 0, 0, 0, 0, 1,
      RegStorage(RegStorage::k64BitPair, rAX, rDX), INVALID_SREG, INVALID_SREG};
+const RegLocation x86_loc_c_return_ref
+    {kLocPhysReg, 0, 0, 0, 0, 0, 1, 0, 1,
+     RegStorage(RegStorage::k32BitSolo, rAX), INVALID_SREG, INVALID_SREG};
+const RegLocation x86_64_loc_c_return_ref
+    {kLocPhysReg, 0, 0, 0, 0, 0, 1, 0, 1,
+     RegStorage(RegStorage::k64BitSolo, rAX), INVALID_SREG, INVALID_SREG};
 const RegLocation x86_64_loc_c_return_wide
     {kLocPhysReg, 1, 0, 0, 0, 0, 0, 0, 1,
      RegStorage(RegStorage::k64BitSolo, rAX), INVALID_SREG, INVALID_SREG};