x86_32.ad

//
// Copyright (c) 1997, 2013, 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
// under the terms of the GNU General Public License version 2 only, as
// published by the Free Software Foundation.
//
// This code is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
// FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
// version 2 for more details (a copy is included in the LICENSE file that
// accompanied this code).
//
// You should have received a copy of the GNU General Public License version
// 2 along with this work; if not, write to the Free Software Foundation,
// Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
//
// Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
// or visit www.oracle.com if you need additional information or have any
// questions.
//
//

// X86 Architecture Description File

//----------REGISTER DEFINITION BLOCK------------------------------------------
// This information is used by the matcher and the register allocator to
// describe individual registers and classes of registers within the target
// archtecture.

register %{
//----------Architecture Description Register Definitions----------------------
// General Registers
// "reg_def"  name ( register save type, C convention save type,
//                   ideal register type, encoding );
// Register Save Types:
//
// NS  = No-Save:       The register allocator assumes that these registers
//                      can be used without saving upon entry to the method, &
//                      that they do not need to be saved at call sites.
//
// SOC = Save-On-Call:  The register allocator assumes that these registers
//                      can be used without saving upon entry to the method,
//                      but that they must be saved at call sites.
//
// SOE = Save-On-Entry: The register allocator assumes that these registers
//                      must be saved before using them upon entry to the
//                      method, but they do not need to be saved at call
//                      sites.
//
// AS  = Always-Save:   The register allocator assumes that these registers
//                      must be saved before using them upon entry to the
//                      method, & that they must be saved at call sites.
//
// Ideal Register Type is used to determine how to save & restore a
// register.  Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
// spilled with LoadP/StoreP.  If the register supports both, use Op_RegI.
//
// The encoding number is the actual bit-pattern placed into the opcodes.

// General Registers
// Previously set EBX, ESI, and EDI as save-on-entry for java code
// Turn off SOE in java-code due to frequent use of uncommon-traps.
// Now that allocator is better, turn on ESI and EDI as SOE registers.

reg_def EBX(SOC, SOE, Op_RegI, 3, rbx->as_VMReg());
reg_def ECX(SOC, SOC, Op_RegI, 1, rcx->as_VMReg());
reg_def ESI(SOC, SOE, Op_RegI, 6, rsi->as_VMReg());
reg_def EDI(SOC, SOE, Op_RegI, 7, rdi->as_VMReg());
// now that adapter frames are gone EBP is always saved and restored by the prolog/epilog code
reg_def EBP(NS, SOE, Op_RegI, 5, rbp->as_VMReg());
reg_def EDX(SOC, SOC, Op_RegI, 2, rdx->as_VMReg());
reg_def EAX(SOC, SOC, Op_RegI, 0, rax->as_VMReg());
reg_def ESP( NS,  NS, Op_RegI, 4, rsp->as_VMReg());

// Float registers.  We treat TOS/FPR0 special.  It is invisible to the
// allocator, and only shows up in the encodings.
reg_def FPR0L( SOC, SOC, Op_RegF, 0, VMRegImpl::Bad());
reg_def FPR0H( SOC, SOC, Op_RegF, 0, VMRegImpl::Bad());
// Ok so here's the trick FPR1 is really st(0) except in the midst
// of emission of assembly for a machnode. During the emission the fpu stack
// is pushed making FPR1 == st(1) temporarily. However at any safepoint
// the stack will not have this element so FPR1 == st(0) from the
// oopMap viewpoint. This same weirdness with numbering causes
// instruction encoding to have to play games with the register
// encode to correct for this 0/1 issue. See MachSpillCopyNode::implementation
// where it does flt->flt moves to see an example
//
reg_def FPR1L( SOC, SOC, Op_RegF, 1, as_FloatRegister(0)->as_VMReg());
reg_def FPR1H( SOC, SOC, Op_RegF, 1, as_FloatRegister(0)->as_VMReg()->next());
reg_def FPR2L( SOC, SOC, Op_RegF, 2, as_FloatRegister(1)->as_VMReg());
reg_def FPR2H( SOC, SOC, Op_RegF, 2, as_FloatRegister(1)->as_VMReg()->next());
reg_def FPR3L( SOC, SOC, Op_RegF, 3, as_FloatRegister(2)->as_VMReg());
reg_def FPR3H( SOC, SOC, Op_RegF, 3, as_FloatRegister(2)->as_VMReg()->next());
reg_def FPR4L( SOC, SOC, Op_RegF, 4, as_FloatRegister(3)->as_VMReg());
reg_def FPR4H( SOC, SOC, Op_RegF, 4, as_FloatRegister(3)->as_VMReg()->next());
reg_def FPR5L( SOC, SOC, Op_RegF, 5, as_FloatRegister(4)->as_VMReg());
reg_def FPR5H( SOC, SOC, Op_RegF, 5, as_FloatRegister(4)->as_VMReg()->next());
reg_def FPR6L( SOC, SOC, Op_RegF, 6, as_FloatRegister(5)->as_VMReg());
reg_def FPR6H( SOC, SOC, Op_RegF, 6, as_FloatRegister(5)->as_VMReg()->next());
reg_def FPR7L( SOC, SOC, Op_RegF, 7, as_FloatRegister(6)->as_VMReg());
reg_def FPR7H( SOC, SOC, Op_RegF, 7, as_FloatRegister(6)->as_VMReg()->next());

// Specify priority of register selection within phases of register
// allocation.  Highest priority is first.  A useful heuristic is to
// give registers a low priority when they are required by machine
// instructions, like EAX and EDX.  Registers which are used as
// pairs must fall on an even boundary (witness the FPR#L's in this list).
// For the Intel integer registers, the equivalent Long pairs are
// EDX:EAX, EBX:ECX, and EDI:EBP.
alloc_class chunk0( ECX,   EBX,   EBP,   EDI,   EAX,   EDX,   ESI, ESP,
                    FPR0L, FPR0H, FPR1L, FPR1H, FPR2L, FPR2H,
                    FPR3L, FPR3H, FPR4L, FPR4H, FPR5L, FPR5H,
                    FPR6L, FPR6H, FPR7L, FPR7H );


//----------Architecture Description Register Classes--------------------------
// Several register classes are automatically defined based upon information in
// this architecture description.
// 1) reg_class inline_cache_reg           ( /* as def'd in frame section */ )
// 2) reg_class compiler_method_oop_reg    ( /* as def'd in frame section */ )
// 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
// 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
//
// Class for all registers
reg_class any_reg(EAX, EDX, EBP, EDI, ESI, ECX, EBX, ESP);
// Class for general registers
reg_class int_reg(EAX, EDX, EBP, EDI, ESI, ECX, EBX);
// Class for general registers which may be used for implicit null checks on win95
// Also safe for use by tailjump. We don't want to allocate in rbp,
reg_class int_reg_no_rbp(EAX, EDX, EDI, ESI, ECX, EBX);
// Class of "X" registers
reg_class int_x_reg(EBX, ECX, EDX, EAX);
// Class of registers that can appear in an address with no offset.
// EBP and ESP require an extra instruction byte for zero offset.
// Used in fast-unlock
reg_class p_reg(EDX, EDI, ESI, EBX);
// Class for general registers not including ECX
reg_class ncx_reg(EAX, EDX, EBP, EDI, ESI, EBX);
// Class for general registers not including EAX
reg_class nax_reg(EDX, EDI, ESI, ECX, EBX);
// Class for general registers not including EAX or EBX.
reg_class nabx_reg(EDX, EDI, ESI, ECX, EBP);
// Class of EAX (for multiply and divide operations)
reg_class eax_reg(EAX);
// Class of EBX (for atomic add)
reg_class ebx_reg(EBX);
// Class of ECX (for shift and JCXZ operations and cmpLTMask)
reg_class ecx_reg(ECX);
// Class of EDX (for multiply and divide operations)
reg_class edx_reg(EDX);
// Class of EDI (for synchronization)
reg_class edi_reg(EDI);
// Class of ESI (for synchronization)
reg_class esi_reg(ESI);
// Singleton class for interpreter's stack pointer
reg_class ebp_reg(EBP);
// Singleton class for stack pointer
reg_class sp_reg(ESP);
// Singleton class for instruction pointer
// reg_class ip_reg(EIP);
// Class of integer register pairs
reg_class long_reg( EAX,EDX, ECX,EBX, EBP,EDI );
// Class of integer register pairs that aligns with calling convention
reg_class eadx_reg( EAX,EDX );
reg_class ebcx_reg( ECX,EBX );
// Not AX or DX, used in divides
reg_class nadx_reg( EBX,ECX,ESI,EDI,EBP );

// Floating point registers.  Notice FPR0 is not a choice.
// FPR0 is not ever allocated; we use clever encodings to fake
// a 2-address instructions out of Intels FP stack.
reg_class fp_flt_reg( FPR1L,FPR2L,FPR3L,FPR4L,FPR5L,FPR6L,FPR7L );

reg_class fp_dbl_reg( FPR1L,FPR1H, FPR2L,FPR2H, FPR3L,FPR3H,
                      FPR4L,FPR4H, FPR5L,FPR5H, FPR6L,FPR6H,
                      FPR7L,FPR7H );

reg_class fp_flt_reg0( FPR1L );
reg_class fp_dbl_reg0( FPR1L,FPR1H );
reg_class fp_dbl_reg1( FPR2L,FPR2H );
reg_class fp_dbl_notreg0( FPR2L,FPR2H, FPR3L,FPR3H, FPR4L,FPR4H,
                          FPR5L,FPR5H, FPR6L,FPR6H, FPR7L,FPR7H );

%}


//----------SOURCE BLOCK-------------------------------------------------------
// This is a block of C++ code which provides values, functions, and
// definitions necessary in the rest of the architecture description
source_hpp %{
// Must be visible to the DFA in dfa_x86_32.cpp
extern bool is_operand_hi32_zero(Node* n);
%}

source %{
#define   RELOC_IMM32    Assembler::imm_operand
#define   RELOC_DISP32   Assembler::disp32_operand

#define __ _masm.

// How to find the high register of a Long pair, given the low register
#define   HIGH_FROM_LOW(x) ((x)+2)

// These masks are used to provide 128-bit aligned bitmasks to the XMM
// instructions, to allow sign-masking or sign-bit flipping.  They allow
// fast versions of NegF/NegD and AbsF/AbsD.

// Note: 'double' and 'long long' have 32-bits alignment on x86.
static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
  // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
  // of 128-bits operands for SSE instructions.
  jlong *operand = (jlong*)(((uintptr_t)adr)&((uintptr_t)(~0xF)));
  // Store the value to a 128-bits operand.
  operand[0] = lo;
  operand[1] = hi;
  return operand;
}

// Buffer for 128-bits masks used by SSE instructions.
static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)

// Static initialization during VM startup.
static jlong *float_signmask_pool  = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
static jlong *float_signflip_pool  = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));

// Offset hacking within calls.
static int pre_call_resets_size() {
  int size = 0;
  Compile* C = Compile::current();
  if (C->in_24_bit_fp_mode()) {
    size += 6; // fldcw
  }
  if (C->max_vector_size() > 16) {
    size += 3; // vzeroupper
  }
  return size;
}

static int preserve_SP_size() {
  return 2;  // op, rm(reg/reg)
}

// !!!!! Special hack to get all type of calls to specify the byte offset
//       from the start of the call to the point where the return address
//       will point.
int MachCallStaticJavaNode::ret_addr_offset() {
  int offset = 5 + pre_call_resets_size();  // 5 bytes from start of call to where return address points
  if (_method_handle_invoke)
    offset += preserve_SP_size();
  return offset;
}

int MachCallDynamicJavaNode::ret_addr_offset() {
  return 10 + pre_call_resets_size();  // 10 bytes from start of call to where return address points
}

static int sizeof_FFree_Float_Stack_All = -1;

int MachCallRuntimeNode::ret_addr_offset() {
  assert(sizeof_FFree_Float_Stack_All != -1, "must have been emitted already");
  return sizeof_FFree_Float_Stack_All + 5 + pre_call_resets_size();
}

// Indicate if the safepoint node needs the polling page as an input.
// Since x86 does have absolute addressing, it doesn't.
bool SafePointNode::needs_polling_address_input() {
  return false;
}

//
// Compute padding required for nodes which need alignment
//

// The address of the call instruction needs to be 4-byte aligned to
// ensure that it does not span a cache line so that it can be patched.
int CallStaticJavaDirectNode::compute_padding(int current_offset) const {
  current_offset += pre_call_resets_size();  // skip fldcw, if any
  current_offset += 1;      // skip call opcode byte
  return round_to(current_offset, alignment_required()) - current_offset;
}

// The address of the call instruction needs to be 4-byte aligned to
// ensure that it does not span a cache line so that it can be patched.
int CallStaticJavaHandleNode::compute_padding(int current_offset) const {
  current_offset += pre_call_resets_size();  // skip fldcw, if any
  current_offset += preserve_SP_size();   // skip mov rbp, rsp
  current_offset += 1;      // skip call opcode byte
  return round_to(current_offset, alignment_required()) - current_offset;
}

// The address of the call instruction needs to be 4-byte aligned to
// ensure that it does not span a cache line so that it can be patched.
int CallDynamicJavaDirectNode::compute_padding(int current_offset) const {
  current_offset += pre_call_resets_size();  // skip fldcw, if any
  current_offset += 5;      // skip MOV instruction
  current_offset += 1;      // skip call opcode byte
  return round_to(current_offset, alignment_required()) - current_offset;
}

// EMIT_RM()
void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3) {
  unsigned char c = (unsigned char)((f1 << 6) | (f2 << 3) | f3);
  cbuf.insts()->emit_int8(c);
}

// EMIT_CC()
void emit_cc(CodeBuffer &cbuf, int f1, int f2) {
  unsigned char c = (unsigned char)( f1 | f2 );
  cbuf.insts()->emit_int8(c);
}

// EMIT_OPCODE()
void emit_opcode(CodeBuffer &cbuf, int code) {
  cbuf.insts()->emit_int8((unsigned char) code);
}

// EMIT_OPCODE() w/ relocation information
void emit_opcode(CodeBuffer &cbuf, int code, relocInfo::relocType reloc, int offset = 0) {
  cbuf.relocate(cbuf.insts_mark() + offset, reloc);
  emit_opcode(cbuf, code);
}

// EMIT_D8()
void emit_d8(CodeBuffer &cbuf, int d8) {
  cbuf.insts()->emit_int8((unsigned char) d8);
}

// EMIT_D16()
void emit_d16(CodeBuffer &cbuf, int d16) {
  cbuf.insts()->emit_int16(d16);
}

// EMIT_D32()
void emit_d32(CodeBuffer &cbuf, int d32) {
  cbuf.insts()->emit_int32(d32);
}

// emit 32 bit value and construct relocation entry from relocInfo::relocType
void emit_d32_reloc(CodeBuffer &cbuf, int d32, relocInfo::relocType reloc,
        int format) {
  cbuf.relocate(cbuf.insts_mark(), reloc, format);
  cbuf.insts()->emit_int32(d32);
}

// emit 32 bit value and construct relocation entry from RelocationHolder
void emit_d32_reloc(CodeBuffer &cbuf, int d32, RelocationHolder const& rspec,
        int format) {
#ifdef ASSERT
  if (rspec.reloc()->type() == relocInfo::oop_type && d32 != 0 && d32 != (int)Universe::non_oop_word()) {
    assert(cast_to_oop(d32)->is_oop() && (ScavengeRootsInCode || !cast_to_oop(d32)->is_scavengable()), "cannot embed scavengable oops in code");
  }
#endif
  cbuf.relocate(cbuf.insts_mark(), rspec, format);
  cbuf.insts()->emit_int32(d32);
}

// Access stack slot for load or store
void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp) {
  emit_opcode( cbuf, opcode );               // (e.g., FILD   [ESP+src])
  if( -128 <= disp && disp <= 127 ) {
    emit_rm( cbuf, 0x01, rm_field, ESP_enc );  // R/M byte
    emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);    // SIB byte
    emit_d8 (cbuf, disp);     // Displacement  // R/M byte
  } else {
    emit_rm( cbuf, 0x02, rm_field, ESP_enc );  // R/M byte
    emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);    // SIB byte
    emit_d32(cbuf, disp);     // Displacement  // R/M byte
  }
}

   // rRegI ereg, memory mem) %{    // emit_reg_mem
void encode_RegMem( CodeBuffer &cbuf, int reg_encoding, int base, int index, int scale, int displace, relocInfo::relocType disp_reloc ) {
  // There is no index & no scale, use form without SIB byte
  if ((index == 0x4) &&
      (scale == 0) && (base != ESP_enc)) {
    // If no displacement, mode is 0x0; unless base is [EBP]
    if ( (displace == 0) && (base != EBP_enc) ) {
      emit_rm(cbuf, 0x0, reg_encoding, base);
    }
    else {                    // If 8-bit displacement, mode 0x1
      if ((displace >= -128) && (displace <= 127)
          && (disp_reloc == relocInfo::none) ) {
        emit_rm(cbuf, 0x1, reg_encoding, base);
        emit_d8(cbuf, displace);
      }
      else {                  // If 32-bit displacement
        if (base == -1) { // Special flag for absolute address
          emit_rm(cbuf, 0x0, reg_encoding, 0x5);
          // (manual lies; no SIB needed here)
          if ( disp_reloc != relocInfo::none ) {
            emit_d32_reloc(cbuf, displace, disp_reloc, 1);
          } else {
            emit_d32      (cbuf, displace);
          }
        }
        else {                // Normal base + offset
          emit_rm(cbuf, 0x2, reg_encoding, base);
          if ( disp_reloc != relocInfo::none ) {
            emit_d32_reloc(cbuf, displace, disp_reloc, 1);
          } else {
            emit_d32      (cbuf, displace);
          }
        }
      }
    }
  }
  else {                      // Else, encode with the SIB byte
    // If no displacement, mode is 0x0; unless base is [EBP]
    if (displace == 0 && (base != EBP_enc)) {  // If no displacement
      emit_rm(cbuf, 0x0, reg_encoding, 0x4);
      emit_rm(cbuf, scale, index, base);
    }
    else {                    // If 8-bit displacement, mode 0x1
      if ((displace >= -128) && (displace <= 127)
          && (disp_reloc == relocInfo::none) ) {
        emit_rm(cbuf, 0x1, reg_encoding, 0x4);
        emit_rm(cbuf, scale, index, base);
        emit_d8(cbuf, displace);
      }
      else {                  // If 32-bit displacement
        if (base == 0x04 ) {
          emit_rm(cbuf, 0x2, reg_encoding, 0x4);
          emit_rm(cbuf, scale, index, 0x04);
        } else {
          emit_rm(cbuf, 0x2, reg_encoding, 0x4);
          emit_rm(cbuf, scale, index, base);
        }
        if ( disp_reloc != relocInfo::none ) {
          emit_d32_reloc(cbuf, displace, disp_reloc, 1);
        } else {
          emit_d32      (cbuf, displace);
        }
      }
    }
  }
}


void encode_Copy( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
  if( dst_encoding == src_encoding ) {
    // reg-reg copy, use an empty encoding
  } else {
    emit_opcode( cbuf, 0x8B );
    emit_rm(cbuf, 0x3, dst_encoding, src_encoding );
  }
}

void emit_cmpfp_fixup(MacroAssembler& _masm) {
  Label exit;
  __ jccb(Assembler::noParity, exit);
  __ pushf();
  //
  // comiss/ucomiss instructions set ZF,PF,CF flags and
  // zero OF,AF,SF for NaN values.
  // Fixup flags by zeroing ZF,PF so that compare of NaN
  // values returns 'less than' result (CF is set).
  // Leave the rest of flags unchanged.
  //
  //    7 6 5 4 3 2 1 0
  //   |S|Z|r|A|r|P|r|C|  (r - reserved bit)
  //    0 0 1 0 1 0 1 1   (0x2B)
  //
  __ andl(Address(rsp, 0), 0xffffff2b);
  __ popf();
  __ bind(exit);
}

void emit_cmpfp3(MacroAssembler& _masm, Register dst) {
  Label done;
  __ movl(dst, -1);
  __ jcc(Assembler::parity, done);
  __ jcc(Assembler::below, done);
  __ setb(Assembler::notEqual, dst);
  __ movzbl(dst, dst);
  __ bind(done);
}


//=============================================================================
const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;

int Compile::ConstantTable::calculate_table_base_offset() const {
  return 0;  // absolute addressing, no offset
}

void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
  // Empty encoding
}

uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
  return 0;
}

#ifndef PRODUCT
void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
  st->print("# MachConstantBaseNode (empty encoding)");
}
#endif


//=============================================================================
#ifndef PRODUCT
void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
  Compile* C = ra_->C;

  int framesize = C->frame_slots() << LogBytesPerInt;
  assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
  // Remove wordSize for return addr which is already pushed.
  framesize -= wordSize;

  if (C->need_stack_bang(framesize)) {
    framesize -= wordSize;
    st->print("# stack bang");
    st->print("\n\t");
    st->print("PUSH   EBP\t# Save EBP");
    if (framesize) {
      st->print("\n\t");
      st->print("SUB    ESP, #%d\t# Create frame",framesize);
    }
  } else {
    st->print("SUB    ESP, #%d\t# Create frame",framesize);
    st->print("\n\t");
    framesize -= wordSize;
    st->print("MOV    [ESP + #%d], EBP\t# Save EBP",framesize);
  }

  if (VerifyStackAtCalls) {
    st->print("\n\t");
    framesize -= wordSize;
    st->print("MOV    [ESP + #%d], 0xBADB100D\t# Majik cookie for stack depth check",framesize);
  }

  if( C->in_24_bit_fp_mode() ) {
    st->print("\n\t");
    st->print("FLDCW  \t# load 24 bit fpu control word");
  }
  if (UseSSE >= 2 && VerifyFPU) {
    st->print("\n\t");
    st->print("# verify FPU stack (must be clean on entry)");
  }

#ifdef ASSERT
  if (VerifyStackAtCalls) {
    st->print("\n\t");
    st->print("# stack alignment check");
  }
#endif
  st->cr();
}
#endif


void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  Compile* C = ra_->C;
  MacroAssembler _masm(&cbuf);

  int framesize = C->frame_slots() << LogBytesPerInt;

  __ verified_entry(framesize, C->need_stack_bang(framesize), C->in_24_bit_fp_mode());

  C->set_frame_complete(cbuf.insts_size());

  if (C->has_mach_constant_base_node()) {
    // NOTE: We set the table base offset here because users might be
    // emitted before MachConstantBaseNode.
    Compile::ConstantTable& constant_table = C->constant_table();
    constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
  }
}

uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
  return MachNode::size(ra_); // too many variables; just compute it the hard way
}

int MachPrologNode::reloc() const {
  return 0; // a large enough number
}

//=============================================================================
#ifndef PRODUCT
void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
  Compile *C = ra_->C;
  int framesize = C->frame_slots() << LogBytesPerInt;
  assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
  // Remove two words for return addr and rbp,
  framesize -= 2*wordSize;

  if (C->max_vector_size() > 16) {
    st->print("VZEROUPPER");
    st->cr(); st->print("\t");
  }
  if (C->in_24_bit_fp_mode()) {
    st->print("FLDCW  standard control word");
    st->cr(); st->print("\t");
  }
  if (framesize) {
    st->print("ADD    ESP,%d\t# Destroy frame",framesize);
    st->cr(); st->print("\t");
  }
  st->print_cr("POPL   EBP"); st->print("\t");
  if (do_polling() && C->is_method_compilation()) {
    st->print("TEST   PollPage,EAX\t! Poll Safepoint");
    st->cr(); st->print("\t");
  }
}
#endif

void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  Compile *C = ra_->C;

  if (C->max_vector_size() > 16) {
    // Clear upper bits of YMM registers when current compiled code uses
    // wide vectors to avoid AVX <-> SSE transition penalty during call.
    MacroAssembler masm(&cbuf);
    masm.vzeroupper();
  }
  // If method set FPU control word, restore to standard control word
  if (C->in_24_bit_fp_mode()) {
    MacroAssembler masm(&cbuf);
    masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
  }

  int framesize = C->frame_slots() << LogBytesPerInt;
  assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
  // Remove two words for return addr and rbp,
  framesize -= 2*wordSize;

  // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here

  if (framesize >= 128) {
    emit_opcode(cbuf, 0x81); // add  SP, #framesize
    emit_rm(cbuf, 0x3, 0x00, ESP_enc);
    emit_d32(cbuf, framesize);
  } else if (framesize) {
    emit_opcode(cbuf, 0x83); // add  SP, #framesize
    emit_rm(cbuf, 0x3, 0x00, ESP_enc);
    emit_d8(cbuf, framesize);
  }

  emit_opcode(cbuf, 0x58 | EBP_enc);

  if (do_polling() && C->is_method_compilation()) {
    cbuf.relocate(cbuf.insts_end(), relocInfo::poll_return_type, 0);
    emit_opcode(cbuf,0x85);
    emit_rm(cbuf, 0x0, EAX_enc, 0x5); // EAX
    emit_d32(cbuf, (intptr_t)os::get_polling_page());
  }
}

uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
  Compile *C = ra_->C;
  // If method set FPU control word, restore to standard control word
  int size = C->in_24_bit_fp_mode() ? 6 : 0;
  if (C->max_vector_size() > 16) size += 3; // vzeroupper
  if (do_polling() && C->is_method_compilation()) size += 6;

  int framesize = C->frame_slots() << LogBytesPerInt;
  assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
  // Remove two words for return addr and rbp,
  framesize -= 2*wordSize;

  size++; // popl rbp,

  if (framesize >= 128) {
    size += 6;
  } else {
    size += framesize ? 3 : 0;
  }
  return size;
}

int MachEpilogNode::reloc() const {
  return 0; // a large enough number
}

const Pipeline * MachEpilogNode::pipeline() const {
  return MachNode::pipeline_class();
}

int MachEpilogNode::safepoint_offset() const { return 0; }

//=============================================================================

enum RC { rc_bad, rc_int, rc_float, rc_xmm, rc_stack };
static enum RC rc_class( OptoReg::Name reg ) {

  if( !OptoReg::is_valid(reg)  ) return rc_bad;
  if (OptoReg::is_stack(reg)) return rc_stack;

  VMReg r = OptoReg::as_VMReg(reg);
  if (r->is_Register()) return rc_int;
  if (r->is_FloatRegister()) {
    assert(UseSSE < 2, "shouldn't be used in SSE2+ mode");
    return rc_float;
  }
  assert(r->is_XMMRegister(), "must be");
  return rc_xmm;
}

static int impl_helper( CodeBuffer *cbuf, bool do_size, bool is_load, int offset, int reg,
                        int opcode, const char *op_str, int size, outputStream* st ) {
  if( cbuf ) {
    emit_opcode  (*cbuf, opcode );
    encode_RegMem(*cbuf, Matcher::_regEncode[reg], ESP_enc, 0x4, 0, offset, relocInfo::none);
#ifndef PRODUCT
  } else if( !do_size ) {
    if( size != 0 ) st->print("\n\t");
    if( opcode == 0x8B || opcode == 0x89 ) { // MOV
      if( is_load ) st->print("%s   %s,[ESP + #%d]",op_str,Matcher::regName[reg],offset);
      else          st->print("%s   [ESP + #%d],%s",op_str,offset,Matcher::regName[reg]);
    } else { // FLD, FST, PUSH, POP
      st->print("%s [ESP + #%d]",op_str,offset);
    }
#endif
  }
  int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
  return size+3+offset_size;
}

// Helper for XMM registers.  Extra opcode bits, limited syntax.
static int impl_x_helper( CodeBuffer *cbuf, bool do_size, bool is_load,
                         int offset, int reg_lo, int reg_hi, int size, outputStream* st ) {
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    if (reg_lo+1 == reg_hi) { // double move?
      if (is_load) {
        __ movdbl(as_XMMRegister(Matcher::_regEncode[reg_lo]), Address(rsp, offset));
      } else {
        __ movdbl(Address(rsp, offset), as_XMMRegister(Matcher::_regEncode[reg_lo]));
      }
    } else {
      if (is_load) {
        __ movflt(as_XMMRegister(Matcher::_regEncode[reg_lo]), Address(rsp, offset));
      } else {
        __ movflt(Address(rsp, offset), as_XMMRegister(Matcher::_regEncode[reg_lo]));
      }
    }
#ifndef PRODUCT
  } else if (!do_size) {
    if (size != 0) st->print("\n\t");
    if (reg_lo+1 == reg_hi) { // double move?
      if (is_load) st->print("%s %s,[ESP + #%d]",
                              UseXmmLoadAndClearUpper ? "MOVSD " : "MOVLPD",
                              Matcher::regName[reg_lo], offset);
      else         st->print("MOVSD  [ESP + #%d],%s",
                              offset, Matcher::regName[reg_lo]);
    } else {
      if (is_load) st->print("MOVSS  %s,[ESP + #%d]",
                              Matcher::regName[reg_lo], offset);
      else         st->print("MOVSS  [ESP + #%d],%s",
                              offset, Matcher::regName[reg_lo]);
    }
#endif
  }
  int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
  // VEX_2bytes prefix is used if UseAVX > 0, so it takes the same 2 bytes as SIMD prefix.
  return size+5+offset_size;
}


static int impl_movx_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
                            int src_hi, int dst_hi, int size, outputStream* st ) {
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    if (src_lo+1 == src_hi && dst_lo+1 == dst_hi) { // double move?
      __ movdbl(as_XMMRegister(Matcher::_regEncode[dst_lo]),
                as_XMMRegister(Matcher::_regEncode[src_lo]));
    } else {
      __ movflt(as_XMMRegister(Matcher::_regEncode[dst_lo]),
                as_XMMRegister(Matcher::_regEncode[src_lo]));
    }
#ifndef PRODUCT
  } else if (!do_size) {
    if (size != 0) st->print("\n\t");
    if (UseXmmRegToRegMoveAll) {//Use movaps,movapd to move between xmm registers
      if (src_lo+1 == src_hi && dst_lo+1 == dst_hi) { // double move?
        st->print("MOVAPD %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
      } else {
        st->print("MOVAPS %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
      }
    } else {
      if( src_lo+1 == src_hi && dst_lo+1 == dst_hi ) { // double move?
        st->print("MOVSD  %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
      } else {
        st->print("MOVSS  %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
      }
    }
#endif
  }
  // VEX_2bytes prefix is used if UseAVX > 0, and it takes the same 2 bytes as SIMD prefix.
  // Only MOVAPS SSE prefix uses 1 byte.
  int sz = 4;
  if (!(src_lo+1 == src_hi && dst_lo+1 == dst_hi) &&
      UseXmmRegToRegMoveAll && (UseAVX == 0)) sz = 3;
  return size + sz;
}

static int impl_movgpr2x_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
                            int src_hi, int dst_hi, int size, outputStream* st ) {
  // 32-bit
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    __ movdl(as_XMMRegister(Matcher::_regEncode[dst_lo]),
             as_Register(Matcher::_regEncode[src_lo]));
#ifndef PRODUCT
  } else if (!do_size) {
    st->print("movdl   %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
#endif
  }
  return 4;
}


static int impl_movx2gpr_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
                                 int src_hi, int dst_hi, int size, outputStream* st ) {
  // 32-bit
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    __ movdl(as_Register(Matcher::_regEncode[dst_lo]),
             as_XMMRegister(Matcher::_regEncode[src_lo]));
#ifndef PRODUCT
  } else if (!do_size) {
    st->print("movdl   %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
#endif
  }
  return 4;
}

static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int size, outputStream* st ) {
  if( cbuf ) {
    emit_opcode(*cbuf, 0x8B );
    emit_rm    (*cbuf, 0x3, Matcher::_regEncode[dst], Matcher::_regEncode[src] );
#ifndef PRODUCT
  } else if( !do_size ) {
    if( size != 0 ) st->print("\n\t");
    st->print("MOV    %s,%s",Matcher::regName[dst],Matcher::regName[src]);
#endif
  }
  return size+2;
}

static int impl_fp_store_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int src_hi, int dst_lo, int dst_hi,
                                 int offset, int size, outputStream* st ) {
  if( src_lo != FPR1L_num ) {      // Move value to top of FP stack, if not already there
    if( cbuf ) {
      emit_opcode( *cbuf, 0xD9 );  // FLD (i.e., push it)
      emit_d8( *cbuf, 0xC0-1+Matcher::_regEncode[src_lo] );
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print("FLD    %s",Matcher::regName[src_lo]);
#endif
    }
    size += 2;
  }

  int st_op = (src_lo != FPR1L_num) ? EBX_num /*store & pop*/ : EDX_num /*store no pop*/;
  const char *op_str;
  int op;
  if( src_lo+1 == src_hi && dst_lo+1 == dst_hi ) { // double store?
    op_str = (src_lo != FPR1L_num) ? "FSTP_D" : "FST_D ";
    op = 0xDD;
  } else {                   // 32-bit store
    op_str = (src_lo != FPR1L_num) ? "FSTP_S" : "FST_S ";
    op = 0xD9;
    assert( !OptoReg::is_valid(src_hi) && !OptoReg::is_valid(dst_hi), "no non-adjacent float-stores" );
  }

  return impl_helper(cbuf,do_size,false,offset,st_op,op,op_str,size, st);
}

// Next two methods are shared by 32- and 64-bit VM. They are defined in x86.ad.
static int vec_mov_helper(CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
                          int src_hi, int dst_hi, uint ireg, outputStream* st);

static int vec_spill_helper(CodeBuffer *cbuf, bool do_size, bool is_load,
                            int stack_offset, int reg, uint ireg, outputStream* st);

static int vec_stack_to_stack_helper(CodeBuffer *cbuf, bool do_size, int src_offset,
                                     int dst_offset, uint ireg, outputStream* st) {
  int calc_size = 0;
  int src_offset_size = (src_offset == 0) ? 0 : ((src_offset < 0x80) ? 1 : 4);
  int dst_offset_size = (dst_offset == 0) ? 0 : ((dst_offset < 0x80) ? 1 : 4);
  switch (ireg) {
  case Op_VecS:
    calc_size = 3+src_offset_size + 3+dst_offset_size;
    break;
  case Op_VecD:
    calc_size = 3+src_offset_size + 3+dst_offset_size;
    src_offset += 4;
    dst_offset += 4;
    src_offset_size = (src_offset == 0) ? 0 : ((src_offset < 0x80) ? 1 : 4);
    dst_offset_size = (dst_offset == 0) ? 0 : ((dst_offset < 0x80) ? 1 : 4);
    calc_size += 3+src_offset_size + 3+dst_offset_size;
    break;
  case Op_VecX:
    calc_size = 6 + 6 + 5+src_offset_size + 5+dst_offset_size;
    break;
  case Op_VecY:
    calc_size = 6 + 6 + 5+src_offset_size + 5+dst_offset_size;
    break;
  default:
    ShouldNotReachHere();
  }
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    int offset = __ offset();
    switch (ireg) {
    case Op_VecS:
      __ pushl(Address(rsp, src_offset));
      __ popl (Address(rsp, dst_offset));
      break;
    case Op_VecD:
      __ pushl(Address(rsp, src_offset));
      __ popl (Address(rsp, dst_offset));
      __ pushl(Address(rsp, src_offset+4));
      __ popl (Address(rsp, dst_offset+4));
      break;
    case Op_VecX:
      __ movdqu(Address(rsp, -16), xmm0);
      __ movdqu(xmm0, Address(rsp, src_offset));
      __ movdqu(Address(rsp, dst_offset), xmm0);
      __ movdqu(xmm0, Address(rsp, -16));
      break;
    case Op_VecY:
      __ vmovdqu(Address(rsp, -32), xmm0);
      __ vmovdqu(xmm0, Address(rsp, src_offset));
      __ vmovdqu(Address(rsp, dst_offset), xmm0);
      __ vmovdqu(xmm0, Address(rsp, -32));
      break;
    default:
      ShouldNotReachHere();
    }
    int size = __ offset() - offset;
    assert(size == calc_size, "incorrect size calculattion");
    return size;
#ifndef PRODUCT
  } else if (!do_size) {
    switch (ireg) {
    case Op_VecS:
      st->print("pushl   [rsp + #%d]\t# 32-bit mem-mem spill\n\t"
                "popl    [rsp + #%d]",
                src_offset, dst_offset);
      break;
    case Op_VecD:
      st->print("pushl   [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
                "popq    [rsp + #%d]\n\t"
                "pushl   [rsp + #%d]\n\t"
                "popq    [rsp + #%d]",
                src_offset, dst_offset, src_offset+4, dst_offset+4);
      break;
     case Op_VecX:
      st->print("movdqu  [rsp - #16], xmm0\t# 128-bit mem-mem spill\n\t"
                "movdqu  xmm0, [rsp + #%d]\n\t"
                "movdqu  [rsp + #%d], xmm0\n\t"
                "movdqu  xmm0, [rsp - #16]",
                src_offset, dst_offset);
      break;
    case Op_VecY:
      st->print("vmovdqu [rsp - #32], xmm0\t# 256-bit mem-mem spill\n\t"
                "vmovdqu xmm0, [rsp + #%d]\n\t"
                "vmovdqu [rsp + #%d], xmm0\n\t"
                "vmovdqu xmm0, [rsp - #32]",
                src_offset, dst_offset);
      break;
    default:
      ShouldNotReachHere();
    }
#endif
  }
  return calc_size;
}

uint MachSpillCopyNode::implementation( CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, outputStream* st ) const {
  // Get registers to move
  OptoReg::Name src_second = ra_->get_reg_second(in(1));
  OptoReg::Name src_first = ra_->get_reg_first(in(1));
  OptoReg::Name dst_second = ra_->get_reg_second(this );
  OptoReg::Name dst_first = ra_->get_reg_first(this );

  enum RC src_second_rc = rc_class(src_second);
  enum RC src_first_rc = rc_class(src_first);
  enum RC dst_second_rc = rc_class(dst_second);
  enum RC dst_first_rc = rc_class(dst_first);

  assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" );

  // Generate spill code!
  int size = 0;

  if( src_first == dst_first && src_second == dst_second )
    return size;            // Self copy, no move

  if (bottom_type()->isa_vect() != NULL) {
    uint ireg = ideal_reg();
    assert((src_first_rc != rc_int && dst_first_rc != rc_int), "sanity");
    assert((src_first_rc != rc_float && dst_first_rc != rc_float), "sanity");
    assert((ireg == Op_VecS || ireg == Op_VecD || ireg == Op_VecX || ireg == Op_VecY), "sanity");
    if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
      // mem -> mem
      int src_offset = ra_->reg2offset(src_first);
      int dst_offset = ra_->reg2offset(dst_first);
      return vec_stack_to_stack_helper(cbuf, do_size, src_offset, dst_offset, ireg, st);
    } else if (src_first_rc == rc_xmm && dst_first_rc == rc_xmm ) {
      return vec_mov_helper(cbuf, do_size, src_first, dst_first, src_second, dst_second, ireg, st);
    } else if (src_first_rc == rc_xmm && dst_first_rc == rc_stack ) {
      int stack_offset = ra_->reg2offset(dst_first);
      return vec_spill_helper(cbuf, do_size, false, stack_offset, src_first, ireg, st);
    } else if (src_first_rc == rc_stack && dst_first_rc == rc_xmm ) {
      int stack_offset = ra_->reg2offset(src_first);
      return vec_spill_helper(cbuf, do_size, true,  stack_offset, dst_first, ireg, st);
    } else {
      ShouldNotReachHere();
    }
  }

  // --------------------------------------
  // Check for mem-mem move.  push/pop to move.
  if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
    if( src_second == dst_first ) { // overlapping stack copy ranges
      assert( src_second_rc == rc_stack && dst_second_rc == rc_stack, "we only expect a stk-stk copy here" );
      size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),ESI_num,0xFF,"PUSH  ",size, st);
      size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),EAX_num,0x8F,"POP   ",size, st);
      src_second_rc = dst_second_rc = rc_bad;  // flag as already moved the second bits
    }
    // move low bits
    size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),ESI_num,0xFF,"PUSH  ",size, st);
    size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),EAX_num,0x8F,"POP   ",size, st);
    if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) { // mov second bits
      size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),ESI_num,0xFF,"PUSH  ",size, st);
      size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),EAX_num,0x8F,"POP   ",size, st);
    }
    return size;
  }

  // --------------------------------------
  // Check for integer reg-reg copy
  if( src_first_rc == rc_int && dst_first_rc == rc_int )
    size = impl_mov_helper(cbuf,do_size,src_first,dst_first,size, st);

  // Check for integer store
  if( src_first_rc == rc_int && dst_first_rc == rc_stack )
    size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first,0x89,"MOV ",size, st);

  // Check for integer load
  if( dst_first_rc == rc_int && src_first_rc == rc_stack )
    size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first,0x8B,"MOV ",size, st);

  // Check for integer reg-xmm reg copy
  if( src_first_rc == rc_int && dst_first_rc == rc_xmm ) {
    assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad),
            "no 64 bit integer-float reg moves" );
    return impl_movgpr2x_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
  }
  // --------------------------------------
  // Check for float reg-reg copy
  if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
    assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad) ||
            (src_first+1 == src_second && dst_first+1 == dst_second), "no non-adjacent float-moves" );
    if( cbuf ) {

      // Note the mucking with the register encode to compensate for the 0/1
      // indexing issue mentioned in a comment in the reg_def sections
      // for FPR registers many lines above here.

      if( src_first != FPR1L_num ) {
        emit_opcode  (*cbuf, 0xD9 );           // FLD    ST(i)
        emit_d8      (*cbuf, 0xC0+Matcher::_regEncode[src_first]-1 );
        emit_opcode  (*cbuf, 0xDD );           // FSTP   ST(i)
        emit_d8      (*cbuf, 0xD8+Matcher::_regEncode[dst_first] );
     } else {
        emit_opcode  (*cbuf, 0xDD );           // FST    ST(i)
        emit_d8      (*cbuf, 0xD0+Matcher::_regEncode[dst_first]-1 );
     }
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      if( src_first != FPR1L_num ) st->print("FLD    %s\n\tFSTP   %s",Matcher::regName[src_first],Matcher::regName[dst_first]);
      else                      st->print(             "FST    %s",                            Matcher::regName[dst_first]);
#endif
    }
    return size + ((src_first != FPR1L_num) ? 2+2 : 2);
  }

  // Check for float store
  if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
    return impl_fp_store_helper(cbuf,do_size,src_first,src_second,dst_first,dst_second,ra_->reg2offset(dst_first),size, st);
  }

  // Check for float load
  if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
    int offset = ra_->reg2offset(src_first);
    const char *op_str;
    int op;
    if( src_first+1 == src_second && dst_first+1 == dst_second ) { // double load?
      op_str = "FLD_D";
      op = 0xDD;
    } else {                   // 32-bit load
      op_str = "FLD_S";
      op = 0xD9;
      assert( src_second_rc == rc_bad && dst_second_rc == rc_bad, "no non-adjacent float-loads" );
    }
    if( cbuf ) {
      emit_opcode  (*cbuf, op );
      encode_RegMem(*cbuf, 0x0, ESP_enc, 0x4, 0, offset, relocInfo::none);
      emit_opcode  (*cbuf, 0xDD );           // FSTP   ST(i)
      emit_d8      (*cbuf, 0xD8+Matcher::_regEncode[dst_first] );
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print("%s  ST,[ESP + #%d]\n\tFSTP   %s",op_str, offset,Matcher::regName[dst_first]);
#endif
    }
    int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
    return size + 3+offset_size+2;
  }

  // Check for xmm reg-reg copy
  if( src_first_rc == rc_xmm && dst_first_rc == rc_xmm ) {
    assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad) ||
            (src_first+1 == src_second && dst_first+1 == dst_second),
            "no non-adjacent float-moves" );
    return impl_movx_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
  }

  // Check for xmm reg-integer reg copy
  if( src_first_rc == rc_xmm && dst_first_rc == rc_int ) {
    assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad),
            "no 64 bit float-integer reg moves" );
    return impl_movx2gpr_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
  }

  // Check for xmm store
  if( src_first_rc == rc_xmm && dst_first_rc == rc_stack ) {
    return impl_x_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first, src_second, size, st);
  }

  // Check for float xmm load
  if( dst_first_rc == rc_xmm && src_first_rc == rc_stack ) {
    return impl_x_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first, dst_second, size, st);
  }

  // Copy from float reg to xmm reg
  if( dst_first_rc == rc_xmm && src_first_rc == rc_float ) {
    // copy to the top of stack from floating point reg
    // and use LEA to preserve flags
    if( cbuf ) {
      emit_opcode(*cbuf,0x8D);  // LEA  ESP,[ESP-8]
      emit_rm(*cbuf, 0x1, ESP_enc, 0x04);
      emit_rm(*cbuf, 0x0, 0x04, ESP_enc);
      emit_d8(*cbuf,0xF8);
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print("LEA    ESP,[ESP-8]");
#endif
    }
    size += 4;

    size = impl_fp_store_helper(cbuf,do_size,src_first,src_second,dst_first,dst_second,0,size, st);

    // Copy from the temp memory to the xmm reg.
    size = impl_x_helper(cbuf,do_size,true ,0,dst_first, dst_second, size, st);

    if( cbuf ) {
      emit_opcode(*cbuf,0x8D);  // LEA  ESP,[ESP+8]
      emit_rm(*cbuf, 0x1, ESP_enc, 0x04);
      emit_rm(*cbuf, 0x0, 0x04, ESP_enc);
      emit_d8(*cbuf,0x08);
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print("LEA    ESP,[ESP+8]");
#endif
    }
    size += 4;
    return size;
  }

  assert( size > 0, "missed a case" );

  // --------------------------------------------------------------------
  // Check for second bits still needing moving.
  if( src_second == dst_second )
    return size;               // Self copy; no move
  assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" );

  // Check for second word int-int move
  if( src_second_rc == rc_int && dst_second_rc == rc_int )
    return impl_mov_helper(cbuf,do_size,src_second,dst_second,size, st);

  // Check for second word integer store
  if( src_second_rc == rc_int && dst_second_rc == rc_stack )
    return impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),src_second,0x89,"MOV ",size, st);

  // Check for second word integer load
  if( dst_second_rc == rc_int && src_second_rc == rc_stack )
    return impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),dst_second,0x8B,"MOV ",size, st);


  Unimplemented();
}

#ifndef PRODUCT
void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const {
  implementation( NULL, ra_, false, st );
}
#endif

void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  implementation( &cbuf, ra_, false, NULL );
}

uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
  return implementation( NULL, ra_, true, NULL );
}


//=============================================================================
#ifndef PRODUCT
void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
  int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  int reg = ra_->get_reg_first(this);
  st->print("LEA    %s,[ESP + #%d]",Matcher::regName[reg],offset);
}
#endif

void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  int reg = ra_->get_encode(this);
  if( offset >= 128 ) {
    emit_opcode(cbuf, 0x8D);      // LEA  reg,[SP+offset]
    emit_rm(cbuf, 0x2, reg, 0x04);
    emit_rm(cbuf, 0x0, 0x04, ESP_enc);
    emit_d32(cbuf, offset);
  }
  else {
    emit_opcode(cbuf, 0x8D);      // LEA  reg,[SP+offset]
    emit_rm(cbuf, 0x1, reg, 0x04);
    emit_rm(cbuf, 0x0, 0x04, ESP_enc);
    emit_d8(cbuf, offset);
  }
}

uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
  int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  if( offset >= 128 ) {
    return 7;
  }
  else {
    return 4;
  }
}

//=============================================================================
#ifndef PRODUCT
void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
  st->print_cr(  "CMP    EAX,[ECX+4]\t# Inline cache check");
  st->print_cr("\tJNE    SharedRuntime::handle_ic_miss_stub");
  st->print_cr("\tNOP");
  st->print_cr("\tNOP");
  if( !OptoBreakpoint )
    st->print_cr("\tNOP");
}
#endif

void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  MacroAssembler masm(&cbuf);
#ifdef ASSERT
  uint insts_size = cbuf.insts_size();
#endif
  masm.cmpptr(rax, Address(rcx, oopDesc::klass_offset_in_bytes()));
  masm.jump_cc(Assembler::notEqual,
               RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
  /* WARNING these NOPs are critical so that verified entry point is properly
     aligned for patching by NativeJump::patch_verified_entry() */
  int nops_cnt = 2;
  if( !OptoBreakpoint ) // Leave space for int3
     nops_cnt += 1;
  masm.nop(nops_cnt);

  assert(cbuf.insts_size() - insts_size == size(ra_), "checking code size of inline cache node");
}

uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
  return OptoBreakpoint ? 11 : 12;
}


//=============================================================================
uint size_exception_handler() {
  // NativeCall instruction size is the same as NativeJump.
  // exception handler starts out as jump and can be patched to
  // a call be deoptimization.  (4932387)
  // Note that this value is also credited (in output.cpp) to
  // the size of the code section.
  return NativeJump::instruction_size;
}

// Emit exception handler code.  Stuff framesize into a register
// and call a VM stub routine.
int emit_exception_handler(CodeBuffer& cbuf) {

  // Note that the code buffer's insts_mark is always relative to insts.
  // That's why we must use the macroassembler to generate a handler.
  MacroAssembler _masm(&cbuf);
  address base =
  __ start_a_stub(size_exception_handler());
  if (base == NULL)  return 0;  // CodeBuffer::expand failed
  int offset = __ offset();
  __ jump(RuntimeAddress(OptoRuntime::exception_blob()->entry_point()));
  assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
  __ end_a_stub();
  return offset;
}

uint size_deopt_handler() {
  // NativeCall instruction size is the same as NativeJump.
  // exception handler starts out as jump and can be patched to
  // a call be deoptimization.  (4932387)
  // Note that this value is also credited (in output.cpp) to
  // the size of the code section.
  return 5 + NativeJump::instruction_size; // pushl(); jmp;
}

// Emit deopt handler code.
int emit_deopt_handler(CodeBuffer& cbuf) {

  // Note that the code buffer's insts_mark is always relative to insts.
  // That's why we must use the macroassembler to generate a handler.
  MacroAssembler _masm(&cbuf);
  address base =
  __ start_a_stub(size_exception_handler());
  if (base == NULL)  return 0;  // CodeBuffer::expand failed
  int offset = __ offset();
  InternalAddress here(__ pc());
  __ pushptr(here.addr());

  __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
  assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
  __ end_a_stub();
  return offset;
}

int Matcher::regnum_to_fpu_offset(int regnum) {
  return regnum - 32; // The FP registers are in the second chunk
}

// This is UltraSparc specific, true just means we have fast l2f conversion
const bool Matcher::convL2FSupported(void) {
  return true;
}

// Is this branch offset short enough that a short branch can be used?
//
// NOTE: If the platform does not provide any short branch variants, then
//       this method should return false for offset 0.
bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
  // The passed offset is relative to address of the branch.
  // On 86 a branch displacement is calculated relative to address
  // of a next instruction.
  offset -= br_size;

  // the short version of jmpConUCF2 contains multiple branches,
  // making the reach slightly less
  if (rule == jmpConUCF2_rule)
    return (-126 <= offset && offset <= 125);
  return (-128 <= offset && offset <= 127);
}

const bool Matcher::isSimpleConstant64(jlong value) {
  // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
  return false;
}

// The ecx parameter to rep stos for the ClearArray node is in dwords.
const bool Matcher::init_array_count_is_in_bytes = false;

// Threshold size for cleararray.
const int Matcher::init_array_short_size = 8 * BytesPerLong;

// Needs 2 CMOV's for longs.
const int Matcher::long_cmove_cost() { return 1; }

// No CMOVF/CMOVD with SSE/SSE2
const int Matcher::float_cmove_cost() { return (UseSSE>=1) ? ConditionalMoveLimit : 0; }

// Should the Matcher clone shifts on addressing modes, expecting them to
// be subsumed into complex addressing expressions or compute them into
// registers?  True for Intel but false for most RISCs
const bool Matcher::clone_shift_expressions = true;

// Do we need to mask the count passed to shift instructions or does
// the cpu only look at the lower 5/6 bits anyway?
const bool Matcher::need_masked_shift_count = false;

bool Matcher::narrow_oop_use_complex_address() {
  ShouldNotCallThis();
  return true;
}

bool Matcher::narrow_klass_use_complex_address() {
  ShouldNotCallThis();
  return true;
}


// Is it better to copy float constants, or load them directly from memory?
// Intel can load a float constant from a direct address, requiring no
// extra registers.  Most RISCs will have to materialize an address into a
// register first, so they would do better to copy the constant from stack.
const bool Matcher::rematerialize_float_constants = true;

// If CPU can load and store mis-aligned doubles directly then no fixup is
// needed.  Else we split the double into 2 integer pieces and move it
// piece-by-piece.  Only happens when passing doubles into C code as the
// Java calling convention forces doubles to be aligned.
const bool Matcher::misaligned_doubles_ok = true;


void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
  // Get the memory operand from the node
  uint numopnds = node->num_opnds();        // Virtual call for number of operands
  uint skipped  = node->oper_input_base();  // Sum of leaves skipped so far
  assert( idx >= skipped, "idx too low in pd_implicit_null_fixup" );
  uint opcnt     = 1;                 // First operand
  uint num_edges = node->_opnds[1]->num_edges(); // leaves for first operand
  while( idx >= skipped+num_edges ) {
    skipped += num_edges;
    opcnt++;                          // Bump operand count
    assert( opcnt < numopnds, "Accessing non-existent operand" );
    num_edges = node->_opnds[opcnt]->num_edges(); // leaves for next operand
  }

  MachOper *memory = node->_opnds[opcnt];
  MachOper *new_memory = NULL;
  switch (memory->opcode()) {
  case DIRECT:
  case INDOFFSET32X:
    // No transformation necessary.
    return;
  case INDIRECT:
    new_memory = new (C) indirect_win95_safeOper( );
    break;
  case INDOFFSET8:
    new_memory = new (C) indOffset8_win95_safeOper(memory->disp(NULL, NULL, 0));
    break;
  case INDOFFSET32:
    new_memory = new (C) indOffset32_win95_safeOper(memory->disp(NULL, NULL, 0));
    break;
  case INDINDEXOFFSET:
    new_memory = new (C) indIndexOffset_win95_safeOper(memory->disp(NULL, NULL, 0));
    break;
  case INDINDEXSCALE:
    new_memory = new (C) indIndexScale_win95_safeOper(memory->scale());
    break;
  case INDINDEXSCALEOFFSET:
    new_memory = new (C) indIndexScaleOffset_win95_safeOper(memory->scale(), memory->disp(NULL, NULL, 0));
    break;
  case LOAD_LONG_INDIRECT:
  case LOAD_LONG_INDOFFSET32:
    // Does not use EBP as address register, use { EDX, EBX, EDI, ESI}
    return;
  default:
    assert(false, "unexpected memory operand in pd_implicit_null_fixup()");
    return;
  }
  node->_opnds[opcnt] = new_memory;
}

// Advertise here if the CPU requires explicit rounding operations
// to implement the UseStrictFP mode.
const bool Matcher::strict_fp_requires_explicit_rounding = true;

// Are floats conerted to double when stored to stack during deoptimization?
// On x32 it is stored with convertion only when FPU is used for floats.
bool Matcher::float_in_double() { return (UseSSE == 0); }

// Do ints take an entire long register or just half?
const bool Matcher::int_in_long = false;

// Return whether or not this register is ever used as an argument.  This
// function is used on startup to build the trampoline stubs in generateOptoStub.
// Registers not mentioned will be killed by the VM call in the trampoline, and
// arguments in those registers not be available to the callee.
bool Matcher::can_be_java_arg( int reg ) {
  if(  reg == ECX_num   || reg == EDX_num   ) return true;
  if( (reg == XMM0_num  || reg == XMM1_num ) && UseSSE>=1 ) return true;
  if( (reg == XMM0b_num || reg == XMM1b_num) && UseSSE>=2 ) return true;
  return false;
}

bool Matcher::is_spillable_arg( int reg ) {
  return can_be_java_arg(reg);
}

bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
  // Use hardware integer DIV instruction when
  // it is faster than a code which use multiply.
  // Only when constant divisor fits into 32 bit
  // (min_jint is excluded to get only correct
  // positive 32 bit values from negative).
  return VM_Version::has_fast_idiv() &&
         (divisor == (int)divisor && divisor != min_jint);
}

// Register for DIVI projection of divmodI
RegMask Matcher::divI_proj_mask() {
  return EAX_REG_mask();
}

// Register for MODI projection of divmodI
RegMask Matcher::modI_proj_mask() {
  return EDX_REG_mask();
}

// Register for DIVL projection of divmodL
RegMask Matcher::divL_proj_mask() {
  ShouldNotReachHere();
  return RegMask();
}

// Register for MODL projection of divmodL
RegMask Matcher::modL_proj_mask() {
  ShouldNotReachHere();
  return RegMask();
}

const RegMask Matcher::method_handle_invoke_SP_save_mask() {
  return EBP_REG_mask();
}

const RegMask Matcher::mathExactI_result_proj_mask() {
  return EAX_REG_mask();
}

const RegMask Matcher::mathExactL_result_proj_mask() {
  ShouldNotReachHere();
  return RegMask();
}

const RegMask Matcher::mathExactI_flags_proj_mask() {
  return INT_FLAGS_mask();
}

// Returns true if the high 32 bits of the value is known to be zero.
bool is_operand_hi32_zero(Node* n) {
  int opc = n->Opcode();
  if (opc == Op_AndL) {
    Node* o2 = n->in(2);
    if (o2->is_Con() && (o2->get_long() & 0xFFFFFFFF00000000LL) == 0LL) {
      return true;
    }
  }
  if (opc == Op_ConL && (n->get_long() & 0xFFFFFFFF00000000LL) == 0LL) {
    return true;
  }
  return false;
}

%}

//----------ENCODING BLOCK-----------------------------------------------------
// This block specifies the encoding classes used by the compiler to output
// byte streams.  Encoding classes generate functions which are called by
// Machine Instruction Nodes in order to generate the bit encoding of the
// instruction.  Operands specify their base encoding interface with the
// interface keyword.  There are currently supported four interfaces,
// REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER.  REG_INTER causes an
// operand to generate a function which returns its register number when
// queried.   CONST_INTER causes an operand to generate a function which
// returns the value of the constant when queried.  MEMORY_INTER causes an
// operand to generate four functions which return the Base Register, the
// Index Register, the Scale Value, and the Offset Value of the operand when
// queried.  COND_INTER causes an operand to generate six functions which
// return the encoding code (ie - encoding bits for the instruction)
// associated with each basic boolean condition for a conditional instruction.
// Instructions specify two basic values for encoding.  They use the
// ins_encode keyword to specify their encoding class (which must be one of
// the class names specified in the encoding block), and they use the
// opcode keyword to specify, in order, their primary, secondary, and
// tertiary opcode.  Only the opcode sections which a particular instruction
// needs for encoding need to be specified.
encode %{
  // Build emit functions for each basic byte or larger field in the intel
  // encoding scheme (opcode, rm, sib, immediate), and call them from C++
  // code in the enc_class source block.  Emit functions will live in the
  // main source block for now.  In future, we can generalize this by
  // adding a syntax that specifies the sizes of fields in an order,
  // so that the adlc can build the emit functions automagically

  // Emit primary opcode
  enc_class OpcP %{
    emit_opcode(cbuf, $primary);
  %}

  // Emit secondary opcode
  enc_class OpcS %{
    emit_opcode(cbuf, $secondary);
  %}

  // Emit opcode directly
  enc_class Opcode(immI d8) %{
    emit_opcode(cbuf, $d8$$constant);
  %}

  enc_class SizePrefix %{
    emit_opcode(cbuf,0x66);
  %}

  enc_class RegReg (rRegI dst, rRegI src) %{    // RegReg(Many)
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class OpcRegReg (immI opcode, rRegI dst, rRegI src) %{    // OpcRegReg(Many)
    emit_opcode(cbuf,$opcode$$constant);
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class mov_r32_imm0( rRegI dst ) %{
    emit_opcode( cbuf, 0xB8 + $dst$$reg ); // 0xB8+ rd   -- MOV r32  ,imm32
    emit_d32   ( cbuf, 0x0  );             //                         imm32==0x0
  %}

  enc_class cdq_enc %{
    // Full implementation of Java idiv and irem; checks for
    // special case as described in JVM spec., p.243 & p.271.
    //
    //         normal case                           special case
    //
    // input : rax,: dividend                         min_int
    //         reg: divisor                          -1
    //
    // output: rax,: quotient  (= rax, idiv reg)       min_int
    //         rdx: remainder (= rax, irem reg)       0
    //
    //  Code sequnce:
    //
    //  81 F8 00 00 00 80    cmp         rax,80000000h
    //  0F 85 0B 00 00 00    jne         normal_case
    //  33 D2                xor         rdx,edx
    //  83 F9 FF             cmp         rcx,0FFh
    //  0F 84 03 00 00 00    je          done
    //                  normal_case:
    //  99                   cdq
    //  F7 F9                idiv        rax,ecx
    //                  done:
    //
    emit_opcode(cbuf,0x81); emit_d8(cbuf,0xF8);
    emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);
    emit_opcode(cbuf,0x00); emit_d8(cbuf,0x80);                     // cmp rax,80000000h
    emit_opcode(cbuf,0x0F); emit_d8(cbuf,0x85);
    emit_opcode(cbuf,0x0B); emit_d8(cbuf,0x00);
    emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);                     // jne normal_case
    emit_opcode(cbuf,0x33); emit_d8(cbuf,0xD2);                     // xor rdx,edx
    emit_opcode(cbuf,0x83); emit_d8(cbuf,0xF9); emit_d8(cbuf,0xFF); // cmp rcx,0FFh
    emit_opcode(cbuf,0x0F); emit_d8(cbuf,0x84);
    emit_opcode(cbuf,0x03); emit_d8(cbuf,0x00);
    emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);                     // je done
    // normal_case:
    emit_opcode(cbuf,0x99);                                         // cdq
    // idiv (note: must be emitted by the user of this rule)
    // normal:
  %}

  // Dense encoding for older common ops
  enc_class Opc_plus(immI opcode, rRegI reg) %{
    emit_opcode(cbuf, $opcode$$constant + $reg$$reg);
  %}


  // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
  enc_class OpcSE (immI imm) %{ // Emit primary opcode and set sign-extend bit
    // Check for 8-bit immediate, and set sign extend bit in opcode
    if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
      emit_opcode(cbuf, $primary | 0x02);
    }
    else {                          // If 32-bit immediate
      emit_opcode(cbuf, $primary);
    }
  %}

  enc_class OpcSErm (rRegI dst, immI imm) %{    // OpcSEr/m
    // Emit primary opcode and set sign-extend bit
    // Check for 8-bit immediate, and set sign extend bit in opcode
    if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
      emit_opcode(cbuf, $primary | 0x02);    }
    else {                          // If 32-bit immediate
      emit_opcode(cbuf, $primary);
    }
    // Emit r/m byte with secondary opcode, after primary opcode.
    emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
  %}

  enc_class Con8or32 (immI imm) %{    // Con8or32(storeImmI), 8 or 32 bits
    // Check for 8-bit immediate, and set sign extend bit in opcode
    if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
      $$$emit8$imm$$constant;
    }
    else {                          // If 32-bit immediate
      // Output immediate
      $$$emit32$imm$$constant;
    }
  %}

  enc_class Long_OpcSErm_Lo(eRegL dst, immL imm) %{
    // Emit primary opcode and set sign-extend bit
    // Check for 8-bit immediate, and set sign extend bit in opcode
    int con = (int)$imm$$constant; // Throw away top bits
    emit_opcode(cbuf, ((con >= -128) && (con <= 127)) ? ($primary | 0x02) : $primary);
    // Emit r/m byte with secondary opcode, after primary opcode.
    emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
    if ((con >= -128) && (con <= 127)) emit_d8 (cbuf,con);
    else                               emit_d32(cbuf,con);
  %}

  enc_class Long_OpcSErm_Hi(eRegL dst, immL imm) %{
    // Emit primary opcode and set sign-extend bit
    // Check for 8-bit immediate, and set sign extend bit in opcode
    int con = (int)($imm$$constant >> 32); // Throw away bottom bits
    emit_opcode(cbuf, ((con >= -128) && (con <= 127)) ? ($primary | 0x02) : $primary);
    // Emit r/m byte with tertiary opcode, after primary opcode.
    emit_rm(cbuf, 0x3, $tertiary, HIGH_FROM_LOW($dst$$reg));
    if ((con >= -128) && (con <= 127)) emit_d8 (cbuf,con);
    else                               emit_d32(cbuf,con);
  %}

  enc_class OpcSReg (rRegI dst) %{    // BSWAP
    emit_cc(cbuf, $secondary, $dst$$reg );
  %}

  enc_class bswap_long_bytes(eRegL dst) %{ // BSWAP
    int destlo = $dst$$reg;
    int desthi = HIGH_FROM_LOW(destlo);
    // bswap lo
    emit_opcode(cbuf, 0x0F);
    emit_cc(cbuf, 0xC8, destlo);
    // bswap hi
    emit_opcode(cbuf, 0x0F);
    emit_cc(cbuf, 0xC8, desthi);
    // xchg lo and hi
    emit_opcode(cbuf, 0x87);
    emit_rm(cbuf, 0x3, destlo, desthi);
  %}

  enc_class RegOpc (rRegI div) %{    // IDIV, IMOD, JMP indirect, ...
    emit_rm(cbuf, 0x3, $secondary, $div$$reg );
  %}

  enc_class enc_cmov(cmpOp cop ) %{ // CMOV
    $$$emit8$primary;
    emit_cc(cbuf, $secondary, $cop$$cmpcode);
  %}

  enc_class enc_cmov_dpr(cmpOp cop, regDPR src ) %{ // CMOV
    int op = 0xDA00 + $cop$$cmpcode + ($src$$reg-1);
    emit_d8(cbuf, op >> 8 );
    emit_d8(cbuf, op & 255);
  %}

  // emulate a CMOV with a conditional branch around a MOV
  enc_class enc_cmov_branch( cmpOp cop, immI brOffs ) %{ // CMOV
    // Invert sense of branch from sense of CMOV
    emit_cc( cbuf, 0x70, ($cop$$cmpcode^1) );
    emit_d8( cbuf, $brOffs$$constant );
  %}

  enc_class enc_PartialSubtypeCheck( ) %{
    Register Redi = as_Register(EDI_enc); // result register
    Register Reax = as_Register(EAX_enc); // super class
    Register Recx = as_Register(ECX_enc); // killed
    Register Resi = as_Register(ESI_enc); // sub class
    Label miss;

    MacroAssembler _masm(&cbuf);
    __ check_klass_subtype_slow_path(Resi, Reax, Recx, Redi,
                                     NULL, &miss,
                                     /*set_cond_codes:*/ true);
    if ($primary) {
      __ xorptr(Redi, Redi);
    }
    __ bind(miss);
  %}

  enc_class FFree_Float_Stack_All %{    // Free_Float_Stack_All
    MacroAssembler masm(&cbuf);
    int start = masm.offset();
    if (UseSSE >= 2) {
      if (VerifyFPU) {
        masm.verify_FPU(0, "must be empty in SSE2+ mode");
      }
    } else {
      // External c_calling_convention expects the FPU stack to be 'clean'.
      // Compiled code leaves it dirty.  Do cleanup now.
      masm.empty_FPU_stack();
    }
    if (sizeof_FFree_Float_Stack_All == -1) {
      sizeof_FFree_Float_Stack_All = masm.offset() - start;
    } else {
      assert(masm.offset() - start == sizeof_FFree_Float_Stack_All, "wrong size");
    }
  %}

  enc_class Verify_FPU_For_Leaf %{
    if( VerifyFPU ) {
      MacroAssembler masm(&cbuf);
      masm.verify_FPU( -3, "Returning from Runtime Leaf call");
    }
  %}

  enc_class Java_To_Runtime (method meth) %{    // CALL Java_To_Runtime, Java_To_Runtime_Leaf
    // This is the instruction starting address for relocation info.
    cbuf.set_insts_mark();
    $$$emit8$primary;
    // CALL directly to the runtime
    emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
                runtime_call_Relocation::spec(), RELOC_IMM32 );

    if (UseSSE >= 2) {
      MacroAssembler _masm(&cbuf);
      BasicType rt = tf()->return_type();

      if ((rt == T_FLOAT || rt == T_DOUBLE) && !return_value_is_used()) {
        // A C runtime call where the return value is unused.  In SSE2+
        // mode the result needs to be removed from the FPU stack.  It's
        // likely that this function call could be removed by the
        // optimizer if the C function is a pure function.
        __ ffree(0);
      } else if (rt == T_FLOAT) {
        __ lea(rsp, Address(rsp, -4));
        __ fstp_s(Address(rsp, 0));
        __ movflt(xmm0, Address(rsp, 0));
        __ lea(rsp, Address(rsp,  4));
      } else if (rt == T_DOUBLE) {
        __ lea(rsp, Address(rsp, -8));
        __ fstp_d(Address(rsp, 0));
        __ movdbl(xmm0, Address(rsp, 0));
        __ lea(rsp, Address(rsp,  8));
      }
    }
  %}


  enc_class pre_call_resets %{
    // If method sets FPU control word restore it here
    debug_only(int off0 = cbuf.insts_size());
    if (ra_->C->in_24_bit_fp_mode()) {
      MacroAssembler _masm(&cbuf);
      __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
    }
    if (ra_->C->max_vector_size() > 16) {
      // Clear upper bits of YMM registers when current compiled code uses
      // wide vectors to avoid AVX <-> SSE transition penalty during call.
      MacroAssembler _masm(&cbuf);
      __ vzeroupper();
    }
    debug_only(int off1 = cbuf.insts_size());
    assert(off1 - off0 == pre_call_resets_size(), "correct size prediction");
  %}

  enc_class post_call_FPU %{
    // If method sets FPU control word do it here also
    if (Compile::current()->in_24_bit_fp_mode()) {
      MacroAssembler masm(&cbuf);
      masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
    }
  %}

  enc_class Java_Static_Call (method meth) %{    // JAVA STATIC CALL
    // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
    // who we intended to call.
    cbuf.set_insts_mark();
    $$$emit8$primary;
    if (!_method) {
      emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
                     runtime_call_Relocation::spec(), RELOC_IMM32 );
    } else if (_optimized_virtual) {
      emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
                     opt_virtual_call_Relocation::spec(), RELOC_IMM32 );
    } else {
      emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
                     static_call_Relocation::spec(), RELOC_IMM32 );
    }
    if (_method) {  // Emit stub for static call.
      CompiledStaticCall::emit_to_interp_stub(cbuf);
    }
  %}

  enc_class Java_Dynamic_Call (method meth) %{    // JAVA DYNAMIC CALL
    MacroAssembler _masm(&cbuf);
    __ ic_call((address)$meth$$method);
  %}

  enc_class Java_Compiled_Call (method meth) %{    // JAVA COMPILED CALL
    int disp = in_bytes(Method::from_compiled_offset());
    assert( -128 <= disp && disp <= 127, "compiled_code_offset isn't small");

    // CALL *[EAX+in_bytes(Method::from_compiled_code_entry_point_offset())]
    cbuf.set_insts_mark();
    $$$emit8$primary;
    emit_rm(cbuf, 0x01, $secondary, EAX_enc );  // R/M byte
    emit_d8(cbuf, disp);             // Displacement

  %}

//   Following encoding is no longer used, but may be restored if calling
//   convention changes significantly.
//   Became: Xor_Reg(EBP), Java_To_Runtime( labl )
//
//   enc_class Java_Interpreter_Call (label labl) %{    // JAVA INTERPRETER CALL
//     // int ic_reg     = Matcher::inline_cache_reg();
//     // int ic_encode  = Matcher::_regEncode[ic_reg];
//     // int imo_reg    = Matcher::interpreter_method_oop_reg();
//     // int imo_encode = Matcher::_regEncode[imo_reg];
//
//     // // Interpreter expects method_oop in EBX, currently a callee-saved register,
//     // // so we load it immediately before the call
//     // emit_opcode(cbuf, 0x8B);                     // MOV    imo_reg,ic_reg  # method_oop
//     // emit_rm(cbuf, 0x03, imo_encode, ic_encode ); // R/M byte
//
//     // xor rbp,ebp
//     emit_opcode(cbuf, 0x33);
//     emit_rm(cbuf, 0x3, EBP_enc, EBP_enc);
//
//     // CALL to interpreter.
//     cbuf.set_insts_mark();
//     $$$emit8$primary;
//     emit_d32_reloc(cbuf, ($labl$$label - (int)(cbuf.insts_end()) - 4),
//                 runtime_call_Relocation::spec(), RELOC_IMM32 );
//   %}

  enc_class RegOpcImm (rRegI dst, immI8 shift) %{    // SHL, SAR, SHR
    $$$emit8$primary;
    emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
    $$$emit8$shift$$constant;
  %}

  enc_class LdImmI (rRegI dst, immI src) %{    // Load Immediate
    // Load immediate does not have a zero or sign extended version
    // for 8-bit immediates
    emit_opcode(cbuf, 0xB8 + $dst$$reg);
    $$$emit32$src$$constant;
  %}

  enc_class LdImmP (rRegI dst, immI src) %{    // Load Immediate
    // Load immediate does not have a zero or sign extended version
    // for 8-bit immediates
    emit_opcode(cbuf, $primary + $dst$$reg);
    $$$emit32$src$$constant;
  %}

  enc_class LdImmL_Lo( eRegL dst, immL src) %{    // Load Immediate
    // Load immediate does not have a zero or sign extended version
    // for 8-bit immediates
    int dst_enc = $dst$$reg;
    int src_con = $src$$constant & 0x0FFFFFFFFL;
    if (src_con == 0) {
      // xor dst, dst
      emit_opcode(cbuf, 0x33);
      emit_rm(cbuf, 0x3, dst_enc, dst_enc);
    } else {
      emit_opcode(cbuf, $primary + dst_enc);
      emit_d32(cbuf, src_con);
    }
  %}

  enc_class LdImmL_Hi( eRegL dst, immL src) %{    // Load Immediate
    // Load immediate does not have a zero or sign extended version
    // for 8-bit immediates
    int dst_enc = $dst$$reg + 2;
    int src_con = ((julong)($src$$constant)) >> 32;
    if (src_con == 0) {
      // xor dst, dst
      emit_opcode(cbuf, 0x33);
      emit_rm(cbuf, 0x3, dst_enc, dst_enc);
    } else {
      emit_opcode(cbuf, $primary + dst_enc);
      emit_d32(cbuf, src_con);
    }
  %}


  // Encode a reg-reg copy.  If it is useless, then empty encoding.
  enc_class enc_Copy( rRegI dst, rRegI src ) %{
    encode_Copy( cbuf, $dst$$reg, $src$$reg );
  %}

  enc_class enc_CopyL_Lo( rRegI dst, eRegL src ) %{
    encode_Copy( cbuf, $dst$$reg, $src$$reg );
  %}

  enc_class RegReg (rRegI dst, rRegI src) %{    // RegReg(Many)
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class RegReg_Lo(eRegL dst, eRegL src) %{    // RegReg(Many)
    $$$emit8$primary;
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class RegReg_Hi(eRegL dst, eRegL src) %{    // RegReg(Many)
    $$$emit8$secondary;
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg));
  %}

  enc_class RegReg_Lo2(eRegL dst, eRegL src) %{    // RegReg(Many)
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class RegReg_Hi2(eRegL dst, eRegL src) %{    // RegReg(Many)
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg));
  %}

  enc_class RegReg_HiLo( eRegL src, rRegI dst ) %{
    emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($src$$reg));
  %}

  enc_class Con32 (immI src) %{    // Con32(storeImmI)
    // Output immediate
    $$$emit32$src$$constant;
  %}

  enc_class Con32FPR_as_bits(immFPR src) %{        // storeF_imm
    // Output Float immediate bits
    jfloat jf = $src$$constant;
    int    jf_as_bits = jint_cast( jf );
    emit_d32(cbuf, jf_as_bits);
  %}

  enc_class Con32F_as_bits(immF src) %{      // storeX_imm
    // Output Float immediate bits
    jfloat jf = $src$$constant;
    int    jf_as_bits = jint_cast( jf );
    emit_d32(cbuf, jf_as_bits);
  %}

  enc_class Con16 (immI src) %{    // Con16(storeImmI)
    // Output immediate
    $$$emit16$src$$constant;
  %}

  enc_class Con_d32(immI src) %{
    emit_d32(cbuf,$src$$constant);
  %}

  enc_class conmemref (eRegP t1) %{    // Con32(storeImmI)
    // Output immediate memory reference
    emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
    emit_d32(cbuf, 0x00);
  %}

  enc_class lock_prefix( ) %{
    if( os::is_MP() )
      emit_opcode(cbuf,0xF0);         // [Lock]
  %}

  // Cmp-xchg long value.
  // Note: we need to swap rbx, and rcx before and after the
  //       cmpxchg8 instruction because the instruction uses
  //       rcx as the high order word of the new value to store but
  //       our register encoding uses rbx,.
  enc_class enc_cmpxchg8(eSIRegP mem_ptr) %{

    // XCHG  rbx,ecx
    emit_opcode(cbuf,0x87);
    emit_opcode(cbuf,0xD9);
    // [Lock]
    if( os::is_MP() )
      emit_opcode(cbuf,0xF0);
    // CMPXCHG8 [Eptr]
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,0xC7);
    emit_rm( cbuf, 0x0, 1, $mem_ptr$$reg );
    // XCHG  rbx,ecx
    emit_opcode(cbuf,0x87);
    emit_opcode(cbuf,0xD9);
  %}

  enc_class enc_cmpxchg(eSIRegP mem_ptr) %{
    // [Lock]
    if( os::is_MP() )
      emit_opcode(cbuf,0xF0);

    // CMPXCHG [Eptr]
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,0xB1);
    emit_rm( cbuf, 0x0, 1, $mem_ptr$$reg );
  %}

  enc_class enc_flags_ne_to_boolean( iRegI res ) %{
    int res_encoding = $res$$reg;

    // MOV  res,0
    emit_opcode( cbuf, 0xB8 + res_encoding);
    emit_d32( cbuf, 0 );
    // JNE,s  fail
    emit_opcode(cbuf,0x75);
    emit_d8(cbuf, 5 );
    // MOV  res,1
    emit_opcode( cbuf, 0xB8 + res_encoding);
    emit_d32( cbuf, 1 );
    // fail:
  %}

  enc_class set_instruction_start( ) %{
    cbuf.set_insts_mark();            // Mark start of opcode for reloc info in mem operand
  %}

  enc_class RegMem (rRegI ereg, memory mem) %{    // emit_reg_mem
    int reg_encoding = $ereg$$reg;
    int base  = $mem$$base;
    int index = $mem$$index;
    int scale = $mem$$scale;
    int displace = $mem$$disp;
    relocInfo::relocType disp_reloc = $mem->disp_reloc();
    encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
  %}

  enc_class RegMem_Hi(eRegL ereg, memory mem) %{    // emit_reg_mem
    int reg_encoding = HIGH_FROM_LOW($ereg$$reg);  // Hi register of pair, computed from lo
    int base  = $mem$$base;
    int index = $mem$$index;
    int scale = $mem$$scale;
    int displace = $mem$$disp + 4;      // Offset is 4 further in memory
    assert( $mem->disp_reloc() == relocInfo::none, "Cannot add 4 to oop" );
    encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, relocInfo::none);
  %}

  enc_class move_long_small_shift( eRegL dst, immI_1_31 cnt ) %{
    int r1, r2;
    if( $tertiary == 0xA4 ) { r1 = $dst$$reg;  r2 = HIGH_FROM_LOW($dst$$reg); }
    else                    { r2 = $dst$$reg;  r1 = HIGH_FROM_LOW($dst$$reg); }
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,$tertiary);
    emit_rm(cbuf, 0x3, r1, r2);
    emit_d8(cbuf,$cnt$$constant);
    emit_d8(cbuf,$primary);
    emit_rm(cbuf, 0x3, $secondary, r1);
    emit_d8(cbuf,$cnt$$constant);
  %}

  enc_class move_long_big_shift_sign( eRegL dst, immI_32_63 cnt ) %{
    emit_opcode( cbuf, 0x8B ); // Move
    emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg));
    if( $cnt$$constant > 32 ) { // Shift, if not by zero
      emit_d8(cbuf,$primary);
      emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
      emit_d8(cbuf,$cnt$$constant-32);
    }
    emit_d8(cbuf,$primary);
    emit_rm(cbuf, 0x3, $secondary, HIGH_FROM_LOW($dst$$reg));
    emit_d8(cbuf,31);
  %}

  enc_class move_long_big_shift_clr( eRegL dst, immI_32_63 cnt ) %{
    int r1, r2;
    if( $secondary == 0x5 ) { r1 = $dst$$reg;  r2 = HIGH_FROM_LOW($dst$$reg); }
    else                    { r2 = $dst$$reg;  r1 = HIGH_FROM_LOW($dst$$reg); }

    emit_opcode( cbuf, 0x8B ); // Move r1,r2
    emit_rm(cbuf, 0x3, r1, r2);
    if( $cnt$$constant > 32 ) { // Shift, if not by zero
      emit_opcode(cbuf,$primary);
      emit_rm(cbuf, 0x3, $secondary, r1);
      emit_d8(cbuf,$cnt$$constant-32);
    }
    emit_opcode(cbuf,0x33);  // XOR r2,r2
    emit_rm(cbuf, 0x3, r2, r2);
  %}

  // Clone of RegMem but accepts an extra parameter to access each
  // half of a double in memory; it never needs relocation info.
  enc_class Mov_MemD_half_to_Reg (immI opcode, memory mem, immI disp_for_half, rRegI rm_reg) %{
    emit_opcode(cbuf,$opcode$$constant);
    int reg_encoding = $rm_reg$$reg;
    int base     = $mem$$base;
    int index    = $mem$$index;
    int scale    = $mem$$scale;
    int displace = $mem$$disp + $disp_for_half$$constant;
    relocInfo::relocType disp_reloc = relocInfo::none;
    encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
  %}

  // !!!!! Special Custom Code used by MemMove, and stack access instructions !!!!!
  //
  // Clone of RegMem except the RM-byte's reg/opcode field is an ADLC-time constant
  // and it never needs relocation information.
  // Frequently used to move data between FPU's Stack Top and memory.
  enc_class RMopc_Mem_no_oop (immI rm_opcode, memory mem) %{
    int rm_byte_opcode = $rm_opcode$$constant;
    int base     = $mem$$base;
    int index    = $mem$$index;
    int scale    = $mem$$scale;
    int displace = $mem$$disp;
    assert( $mem->disp_reloc() == relocInfo::none, "No oops here because no reloc info allowed" );
    encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, relocInfo::none);
  %}

  enc_class RMopc_Mem (immI rm_opcode, memory mem) %{
    int rm_byte_opcode = $rm_opcode$$constant;
    int base     = $mem$$base;
    int index    = $mem$$index;
    int scale    = $mem$$scale;
    int displace = $mem$$disp;
    relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
    encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_reloc);
  %}

  enc_class RegLea (rRegI dst, rRegI src0, immI src1 ) %{    // emit_reg_lea
    int reg_encoding = $dst$$reg;
    int base         = $src0$$reg;      // 0xFFFFFFFF indicates no base
    int index        = 0x04;            // 0x04 indicates no index
    int scale        = 0x00;            // 0x00 indicates no scale
    int displace     = $src1$$constant; // 0x00 indicates no displacement
    relocInfo::relocType disp_reloc = relocInfo::none;
    encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
  %}

  enc_class min_enc (rRegI dst, rRegI src) %{    // MIN
    // Compare dst,src
    emit_opcode(cbuf,0x3B);
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
    // jmp dst < src around move
    emit_opcode(cbuf,0x7C);
    emit_d8(cbuf,2);
    // move dst,src
    emit_opcode(cbuf,0x8B);
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class max_enc (rRegI dst, rRegI src) %{    // MAX
    // Compare dst,src
    emit_opcode(cbuf,0x3B);
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
    // jmp dst > src around move
    emit_opcode(cbuf,0x7F);
    emit_d8(cbuf,2);
    // move dst,src
    emit_opcode(cbuf,0x8B);
    emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
  %}

  enc_class enc_FPR_store(memory mem, regDPR src) %{
    // If src is FPR1, we can just FST to store it.
    // Else we need to FLD it to FPR1, then FSTP to store/pop it.
    int reg_encoding = 0x2; // Just store
    int base  = $mem$$base;
    int index = $mem$$index;
    int scale = $mem$$scale;
    int displace = $mem$$disp;
    relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
    if( $src$$reg != FPR1L_enc ) {
      reg_encoding = 0x3;  // Store & pop
      emit_opcode( cbuf, 0xD9 ); // FLD (i.e., push it)
      emit_d8( cbuf, 0xC0-1+$src$$reg );
    }
    cbuf.set_insts_mark();       // Mark start of opcode for reloc info in mem operand
    emit_opcode(cbuf,$primary);
    encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
  %}

  enc_class neg_reg(rRegI dst) %{
    // NEG $dst
    emit_opcode(cbuf,0xF7);
    emit_rm(cbuf, 0x3, 0x03, $dst$$reg );
  %}

  enc_class setLT_reg(eCXRegI dst) %{
    // SETLT $dst
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,0x9C);
    emit_rm( cbuf, 0x3, 0x4, $dst$$reg );
  %}

  enc_class enc_cmpLTP(ncxRegI p, ncxRegI q, ncxRegI y, eCXRegI tmp) %{    // cadd_cmpLT
    int tmpReg = $tmp$$reg;

    // SUB $p,$q
    emit_opcode(cbuf,0x2B);
    emit_rm(cbuf, 0x3, $p$$reg, $q$$reg);
    // SBB $tmp,$tmp
    emit_opcode(cbuf,0x1B);
    emit_rm(cbuf, 0x3, tmpReg, tmpReg);
    // AND $tmp,$y
    emit_opcode(cbuf,0x23);
    emit_rm(cbuf, 0x3, tmpReg, $y$$reg);
    // ADD $p,$tmp
    emit_opcode(cbuf,0x03);
    emit_rm(cbuf, 0x3, $p$$reg, tmpReg);
  %}

  enc_class shift_left_long( eRegL dst, eCXRegI shift ) %{
    // TEST shift,32
    emit_opcode(cbuf,0xF7);
    emit_rm(cbuf, 0x3, 0, ECX_enc);
    emit_d32(cbuf,0x20);
    // JEQ,s small
    emit_opcode(cbuf, 0x74);
    emit_d8(cbuf, 0x04);
    // MOV    $dst.hi,$dst.lo
    emit_opcode( cbuf, 0x8B );
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg );
    // CLR    $dst.lo
    emit_opcode(cbuf, 0x33);
    emit_rm(cbuf, 0x3, $dst$$reg, $dst$$reg);
// small:
    // SHLD   $dst.hi,$dst.lo,$shift
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,0xA5);
    emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg));
    // SHL    $dst.lo,$shift"
    emit_opcode(cbuf,0xD3);
    emit_rm(cbuf, 0x3, 0x4, $dst$$reg );
  %}

  enc_class shift_right_long( eRegL dst, eCXRegI shift ) %{
    // TEST shift,32
    emit_opcode(cbuf,0xF7);
    emit_rm(cbuf, 0x3, 0, ECX_enc);
    emit_d32(cbuf,0x20);
    // JEQ,s small
    emit_opcode(cbuf, 0x74);
    emit_d8(cbuf, 0x04);
    // MOV    $dst.lo,$dst.hi
    emit_opcode( cbuf, 0x8B );
    emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg) );
    // CLR    $dst.hi
    emit_opcode(cbuf, 0x33);
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($dst$$reg));
// small:
    // SHRD   $dst.lo,$dst.hi,$shift
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,0xAD);
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg);
    // SHR    $dst.hi,$shift"
    emit_opcode(cbuf,0xD3);
    emit_rm(cbuf, 0x3, 0x5, HIGH_FROM_LOW($dst$$reg) );
  %}

  enc_class shift_right_arith_long( eRegL dst, eCXRegI shift ) %{
    // TEST shift,32
    emit_opcode(cbuf,0xF7);
    emit_rm(cbuf, 0x3, 0, ECX_enc);
    emit_d32(cbuf,0x20);
    // JEQ,s small
    emit_opcode(cbuf, 0x74);
    emit_d8(cbuf, 0x05);
    // MOV    $dst.lo,$dst.hi
    emit_opcode( cbuf, 0x8B );
    emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg) );
    // SAR    $dst.hi,31
    emit_opcode(cbuf, 0xC1);
    emit_rm(cbuf, 0x3, 7, HIGH_FROM_LOW($dst$$reg) );
    emit_d8(cbuf, 0x1F );
// small:
    // SHRD   $dst.lo,$dst.hi,$shift
    emit_opcode(cbuf,0x0F);
    emit_opcode(cbuf,0xAD);
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg);
    // SAR    $dst.hi,$shift"
    emit_opcode(cbuf,0xD3);
    emit_rm(cbuf, 0x3, 0x7, HIGH_FROM_LOW($dst$$reg) );
  %}


  // ----------------- Encodings for floating point unit -----------------
  // May leave result in FPU-TOS or FPU reg depending on opcodes
  enc_class OpcReg_FPR(regFPR src) %{    // FMUL, FDIV
    $$$emit8$primary;
    emit_rm(cbuf, 0x3, $secondary, $src$$reg );
  %}

  // Pop argument in FPR0 with FSTP ST(0)
  enc_class PopFPU() %{
    emit_opcode( cbuf, 0xDD );
    emit_d8( cbuf, 0xD8 );
  %}

  // !!!!! equivalent to Pop_Reg_F
  enc_class Pop_Reg_DPR( regDPR dst ) %{
    emit_opcode( cbuf, 0xDD );           // FSTP   ST(i)
    emit_d8( cbuf, 0xD8+$dst$$reg );
  %}

  enc_class Push_Reg_DPR( regDPR dst ) %{
    emit_opcode( cbuf, 0xD9 );
    emit_d8( cbuf, 0xC0-1+$dst$$reg );   // FLD ST(i-1)
  %}

  enc_class strictfp_bias1( regDPR dst ) %{
    emit_opcode( cbuf, 0xDB );           // FLD m80real
    emit_opcode( cbuf, 0x2D );
    emit_d32( cbuf, (int)StubRoutines::addr_fpu_subnormal_bias1() );
    emit_opcode( cbuf, 0xDE );           // FMULP ST(dst), ST0
    emit_opcode( cbuf, 0xC8+$dst$$reg );
  %}

  enc_class strictfp_bias2( regDPR dst ) %{
    emit_opcode( cbuf, 0xDB );           // FLD m80real
    emit_opcode( cbuf, 0x2D );
    emit_d32( cbuf, (int)StubRoutines::addr_fpu_subnormal_bias2() );
    emit_opcode( cbuf, 0xDE );           // FMULP ST(dst), ST0
    emit_opcode( cbuf, 0xC8+$dst$$reg );
  %}

  // Special case for moving an integer register to a stack slot.
  enc_class OpcPRegSS( stackSlotI dst, rRegI src ) %{ // RegSS
    store_to_stackslot( cbuf, $primary, $src$$reg, $dst$$disp );
  %}

  // Special case for moving a register to a stack slot.
  enc_class RegSS( stackSlotI dst, rRegI src ) %{ // RegSS
    // Opcode already emitted
    emit_rm( cbuf, 0x02, $src$$reg, ESP_enc );   // R/M byte
    emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);          // SIB byte
    emit_d32(cbuf, $dst$$disp);   // Displacement
  %}

  // Push the integer in stackSlot 'src' onto FP-stack
  enc_class Push_Mem_I( memory src ) %{    // FILD   [ESP+src]
    store_to_stackslot( cbuf, $primary, $secondary, $src$$disp );
  %}

  // Push FPU's TOS float to a stack-slot, and pop FPU-stack
  enc_class Pop_Mem_FPR( stackSlotF dst ) %{ // FSTP_S [ESP+dst]
    store_to_stackslot( cbuf, 0xD9, 0x03, $dst$$disp );
  %}

  // Same as Pop_Mem_F except for opcode
  // Push FPU's TOS double to a stack-slot, and pop FPU-stack
  enc_class Pop_Mem_DPR( stackSlotD dst ) %{ // FSTP_D [ESP+dst]
    store_to_stackslot( cbuf, 0xDD, 0x03, $dst$$disp );
  %}

  enc_class Pop_Reg_FPR( regFPR dst ) %{
    emit_opcode( cbuf, 0xDD );           // FSTP   ST(i)
    emit_d8( cbuf, 0xD8+$dst$$reg );
  %}

  enc_class Push_Reg_FPR( regFPR dst ) %{
    emit_opcode( cbuf, 0xD9 );           // FLD    ST(i-1)
    emit_d8( cbuf, 0xC0-1+$dst$$reg );
  %}

  // Push FPU's float to a stack-slot, and pop FPU-stack
  enc_class Pop_Mem_Reg_FPR( stackSlotF dst, regFPR src ) %{
    int pop = 0x02;
    if ($src$$reg != FPR1L_enc) {
      emit_opcode( cbuf, 0xD9 );         // FLD    ST(i-1)
      emit_d8( cbuf, 0xC0-1+$src$$reg );
      pop = 0x03;
    }
    store_to_stackslot( cbuf, 0xD9, pop, $dst$$disp ); // FST<P>_S  [ESP+dst]
  %}

  // Push FPU's double to a stack-slot, and pop FPU-stack
  enc_class Pop_Mem_Reg_DPR( stackSlotD dst, regDPR src ) %{
    int pop = 0x02;
    if ($src$$reg != FPR1L_enc) {
      emit_opcode( cbuf, 0xD9 );         // FLD    ST(i-1)
      emit_d8( cbuf, 0xC0-1+$src$$reg );
      pop = 0x03;
    }
    store_to_stackslot( cbuf, 0xDD, pop, $dst$$disp ); // FST<P>_D  [ESP+dst]
  %}

  // Push FPU's double to a FPU-stack-slot, and pop FPU-stack
  enc_class Pop_Reg_Reg_DPR( regDPR dst, regFPR src ) %{
    int pop = 0xD0 - 1; // -1 since we skip FLD
    if ($src$$reg != FPR1L_enc) {
      emit_opcode( cbuf, 0xD9 );         // FLD    ST(src-1)
      emit_d8( cbuf, 0xC0-1+$src$$reg );
      pop = 0xD8;
    }
    emit_opcode( cbuf, 0xDD );
    emit_d8( cbuf, pop+$dst$$reg );      // FST<P> ST(i)
  %}


  enc_class Push_Reg_Mod_DPR( regDPR dst, regDPR src) %{
    // load dst in FPR0
    emit_opcode( cbuf, 0xD9 );
    emit_d8( cbuf, 0xC0-1+$dst$$reg );
    if ($src$$reg != FPR1L_enc) {
      // fincstp
      emit_opcode (cbuf, 0xD9);
      emit_opcode (cbuf, 0xF7);
      // swap src with FPR1:
      // FXCH FPR1 with src
      emit_opcode(cbuf, 0xD9);
      emit_d8(cbuf, 0xC8-1+$src$$reg );
      // fdecstp
      emit_opcode (cbuf, 0xD9);
      emit_opcode (cbuf, 0xF6);
    }
  %}

  enc_class Push_ModD_encoding(regD src0, regD src1) %{
    MacroAssembler _masm(&cbuf);
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src1$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ movdbl(Address(rsp, 0), $src0$$XMMRegister);
    __ fld_d(Address(rsp, 0));
  %}

  enc_class Push_ModF_encoding(regF src0, regF src1) %{
    MacroAssembler _masm(&cbuf);
    __ subptr(rsp, 4);
    __ movflt(Address(rsp, 0), $src1$$XMMRegister);
    __ fld_s(Address(rsp, 0));
    __ movflt(Address(rsp, 0), $src0$$XMMRegister);
    __ fld_s(Address(rsp, 0));
  %}

  enc_class Push_ResultD(regD dst) %{
    MacroAssembler _masm(&cbuf);
    __ fstp_d(Address(rsp, 0));
    __ movdbl($dst$$XMMRegister, Address(rsp, 0));
    __ addptr(rsp, 8);
  %}

  enc_class Push_ResultF(regF dst, immI d8) %{
    MacroAssembler _masm(&cbuf);
    __ fstp_s(Address(rsp, 0));
    __ movflt($dst$$XMMRegister, Address(rsp, 0));
    __ addptr(rsp, $d8$$constant);
  %}

  enc_class Push_SrcD(regD src) %{
    MacroAssembler _masm(&cbuf);
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src$$XMMRegister);
    __ fld_d(Address(rsp, 0));
  %}

  enc_class push_stack_temp_qword() %{
    MacroAssembler _masm(&cbuf);
    __ subptr(rsp, 8);
  %}

  enc_class pop_stack_temp_qword() %{
    MacroAssembler _masm(&cbuf);
    __ addptr(rsp, 8);
  %}

  enc_class push_xmm_to_fpr1(regD src) %{
    MacroAssembler _masm(&cbuf);
    __ movdbl(Address(rsp, 0), $src$$XMMRegister);
    __ fld_d(Address(rsp, 0));
  %}

  enc_class Push_Result_Mod_DPR( regDPR src) %{
    if ($src$$reg != FPR1L_enc) {
      // fincstp
      emit_opcode (cbuf, 0xD9);
      emit_opcode (cbuf, 0xF7);
      // FXCH FPR1 with src
      emit_opcode(cbuf, 0xD9);
      emit_d8(cbuf, 0xC8-1+$src$$reg );
      // fdecstp
      emit_opcode (cbuf, 0xD9);
      emit_opcode (cbuf, 0xF6);
    }
    // // following asm replaced with Pop_Reg_F or Pop_Mem_F
    // // FSTP   FPR$dst$$reg
    // emit_opcode( cbuf, 0xDD );
    // emit_d8( cbuf, 0xD8+$dst$$reg );
  %}

  enc_class fnstsw_sahf_skip_parity() %{
    // fnstsw ax
    emit_opcode( cbuf, 0xDF );
    emit_opcode( cbuf, 0xE0 );
    // sahf
    emit_opcode( cbuf, 0x9E );
    // jnp  ::skip
    emit_opcode( cbuf, 0x7B );
    emit_opcode( cbuf, 0x05 );
  %}

  enc_class emitModDPR() %{
    // fprem must be iterative
    // :: loop
    // fprem
    emit_opcode( cbuf, 0xD9 );
    emit_opcode( cbuf, 0xF8 );
    // wait
    emit_opcode( cbuf, 0x9b );
    // fnstsw ax
    emit_opcode( cbuf, 0xDF );
    emit_opcode( cbuf, 0xE0 );
    // sahf
    emit_opcode( cbuf, 0x9E );
    // jp  ::loop
    emit_opcode( cbuf, 0x0F );
    emit_opcode( cbuf, 0x8A );
    emit_opcode( cbuf, 0xF4 );
    emit_opcode( cbuf, 0xFF );
    emit_opcode( cbuf, 0xFF );
    emit_opcode( cbuf, 0xFF );
  %}

  enc_class fpu_flags() %{
    // fnstsw_ax
    emit_opcode( cbuf, 0xDF);
    emit_opcode( cbuf, 0xE0);
    // test ax,0x0400
    emit_opcode( cbuf, 0x66 );   // operand-size prefix for 16-bit immediate
    emit_opcode( cbuf, 0xA9 );
    emit_d16   ( cbuf, 0x0400 );
    // // // This sequence works, but stalls for 12-16 cycles on PPro
    // // test rax,0x0400
    // emit_opcode( cbuf, 0xA9 );
    // emit_d32   ( cbuf, 0x00000400 );
    //
    // jz exit (no unordered comparison)
    emit_opcode( cbuf, 0x74 );
    emit_d8    ( cbuf, 0x02 );
    // mov ah,1 - treat as LT case (set carry flag)
    emit_opcode( cbuf, 0xB4 );
    emit_d8    ( cbuf, 0x01 );
    // sahf
    emit_opcode( cbuf, 0x9E);
  %}

  enc_class cmpF_P6_fixup() %{
    // Fixup the integer flags in case comparison involved a NaN
    //
    // JNP exit (no unordered comparison, P-flag is set by NaN)
    emit_opcode( cbuf, 0x7B );
    emit_d8    ( cbuf, 0x03 );
    // MOV AH,1 - treat as LT case (set carry flag)
    emit_opcode( cbuf, 0xB4 );
    emit_d8    ( cbuf, 0x01 );
    // SAHF
    emit_opcode( cbuf, 0x9E);
    // NOP     // target for branch to avoid branch to branch
    emit_opcode( cbuf, 0x90);
  %}

//     fnstsw_ax();
//     sahf();
//     movl(dst, nan_result);
//     jcc(Assembler::parity, exit);
//     movl(dst, less_result);
//     jcc(Assembler::below, exit);
//     movl(dst, equal_result);
//     jcc(Assembler::equal, exit);
//     movl(dst, greater_result);

// less_result     =  1;
// greater_result  = -1;
// equal_result    = 0;
// nan_result      = -1;

  enc_class CmpF_Result(rRegI dst) %{
    // fnstsw_ax();
    emit_opcode( cbuf, 0xDF);
    emit_opcode( cbuf, 0xE0);
    // sahf
    emit_opcode( cbuf, 0x9E);
    // movl(dst, nan_result);
    emit_opcode( cbuf, 0xB8 + $dst$$reg);
    emit_d32( cbuf, -1 );
    // jcc(Assembler::parity, exit);
    emit_opcode( cbuf, 0x7A );
    emit_d8    ( cbuf, 0x13 );
    // movl(dst, less_result);
    emit_opcode( cbuf, 0xB8 + $dst$$reg);
    emit_d32( cbuf, -1 );
    // jcc(Assembler::below, exit);
    emit_opcode( cbuf, 0x72 );
    emit_d8    ( cbuf, 0x0C );
    // movl(dst, equal_result);
    emit_opcode( cbuf, 0xB8 + $dst$$reg);
    emit_d32( cbuf, 0 );
    // jcc(Assembler::equal, exit);
    emit_opcode( cbuf, 0x74 );
    emit_d8    ( cbuf, 0x05 );
    // movl(dst, greater_result);
    emit_opcode( cbuf, 0xB8 + $dst$$reg);
    emit_d32( cbuf, 1 );
  %}


  // Compare the longs and set flags
  // BROKEN!  Do Not use as-is
  enc_class cmpl_test( eRegL src1, eRegL src2 ) %{
    // CMP    $src1.hi,$src2.hi
    emit_opcode( cbuf, 0x3B );
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($src1$$reg), HIGH_FROM_LOW($src2$$reg) );
    // JNE,s  done
    emit_opcode(cbuf,0x75);
    emit_d8(cbuf, 2 );
    // CMP    $src1.lo,$src2.lo
    emit_opcode( cbuf, 0x3B );
    emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
// done:
  %}

  enc_class convert_int_long( regL dst, rRegI src ) %{
    // mov $dst.lo,$src
    int dst_encoding = $dst$$reg;
    int src_encoding = $src$$reg;
    encode_Copy( cbuf, dst_encoding  , src_encoding );
    // mov $dst.hi,$src
    encode_Copy( cbuf, HIGH_FROM_LOW(dst_encoding), src_encoding );
    // sar $dst.hi,31
    emit_opcode( cbuf, 0xC1 );
    emit_rm(cbuf, 0x3, 7, HIGH_FROM_LOW(dst_encoding) );
    emit_d8(cbuf, 0x1F );
  %}

  enc_class convert_long_double( eRegL src ) %{
    // push $src.hi
    emit_opcode(cbuf, 0x50+HIGH_FROM_LOW($src$$reg));
    // push $src.lo
    emit_opcode(cbuf, 0x50+$src$$reg  );
    // fild 64-bits at [SP]
    emit_opcode(cbuf,0xdf);
    emit_d8(cbuf, 0x6C);
    emit_d8(cbuf, 0x24);
    emit_d8(cbuf, 0x00);
    // pop stack
    emit_opcode(cbuf, 0x83); // add  SP, #8
    emit_rm(cbuf, 0x3, 0x00, ESP_enc);
    emit_d8(cbuf, 0x8);
  %}

  enc_class multiply_con_and_shift_high( eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32_63 cnt, eFlagsReg cr ) %{
    // IMUL   EDX:EAX,$src1
    emit_opcode( cbuf, 0xF7 );
    emit_rm( cbuf, 0x3, 0x5, $src1$$reg );
    // SAR    EDX,$cnt-32
    int shift_count = ((int)$cnt$$constant) - 32;
    if (shift_count > 0) {
      emit_opcode(cbuf, 0xC1);
      emit_rm(cbuf, 0x3, 7, $dst$$reg );
      emit_d8(cbuf, shift_count);
    }
  %}

  // this version doesn't have add sp, 8
  enc_class convert_long_double2( eRegL src ) %{
    // push $src.hi
    emit_opcode(cbuf, 0x50+HIGH_FROM_LOW($src$$reg));
    // push $src.lo
    emit_opcode(cbuf, 0x50+$src$$reg  );
    // fild 64-bits at [SP]
    emit_opcode(cbuf,0xdf);
    emit_d8(cbuf, 0x6C);
    emit_d8(cbuf, 0x24);
    emit_d8(cbuf, 0x00);
  %}

  enc_class long_int_multiply( eADXRegL dst, nadxRegI src) %{
    // Basic idea: long = (long)int * (long)int
    // IMUL EDX:EAX, src
    emit_opcode( cbuf, 0xF7 );
    emit_rm( cbuf, 0x3, 0x5, $src$$reg);
  %}

  enc_class long_uint_multiply( eADXRegL dst, nadxRegI src) %{
    // Basic Idea:  long = (int & 0xffffffffL) * (int & 0xffffffffL)
    // MUL EDX:EAX, src
    emit_opcode( cbuf, 0xF7 );
    emit_rm( cbuf, 0x3, 0x4, $src$$reg);
  %}

  enc_class long_multiply( eADXRegL dst, eRegL src, rRegI tmp ) %{
    // Basic idea: lo(result) = lo(x_lo * y_lo)
    //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
    // MOV    $tmp,$src.lo
    encode_Copy( cbuf, $tmp$$reg, $src$$reg );
    // IMUL   $tmp,EDX
    emit_opcode( cbuf, 0x0F );
    emit_opcode( cbuf, 0xAF );
    emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
    // MOV    EDX,$src.hi
    encode_Copy( cbuf, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg) );
    // IMUL   EDX,EAX
    emit_opcode( cbuf, 0x0F );
    emit_opcode( cbuf, 0xAF );
    emit_rm( cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg );
    // ADD    $tmp,EDX
    emit_opcode( cbuf, 0x03 );
    emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
    // MUL   EDX:EAX,$src.lo
    emit_opcode( cbuf, 0xF7 );
    emit_rm( cbuf, 0x3, 0x4, $src$$reg );
    // ADD    EDX,ESI
    emit_opcode( cbuf, 0x03 );
    emit_rm( cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $tmp$$reg );
  %}

  enc_class long_multiply_con( eADXRegL dst, immL_127 src, rRegI tmp ) %{
    // Basic idea: lo(result) = lo(src * y_lo)
    //             hi(result) = hi(src * y_lo) + lo(src * y_hi)
    // IMUL   $tmp,EDX,$src
    emit_opcode( cbuf, 0x6B );
    emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
    emit_d8( cbuf, (int)$src$$constant );
    // MOV    EDX,$src
    emit_opcode(cbuf, 0xB8 + EDX_enc);
    emit_d32( cbuf, (int)$src$$constant );
    // MUL   EDX:EAX,EDX
    emit_opcode( cbuf, 0xF7 );
    emit_rm( cbuf, 0x3, 0x4, EDX_enc );
    // ADD    EDX,ESI
    emit_opcode( cbuf, 0x03 );
    emit_rm( cbuf, 0x3, EDX_enc, $tmp$$reg );
  %}

  enc_class long_div( eRegL src1, eRegL src2 ) %{
    // PUSH src1.hi
    emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src1$$reg) );
    // PUSH src1.lo
    emit_opcode(cbuf,               0x50+$src1$$reg  );
    // PUSH src2.hi
    emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src2$$reg) );
    // PUSH src2.lo
    emit_opcode(cbuf,               0x50+$src2$$reg  );
    // CALL directly to the runtime
    cbuf.set_insts_mark();
    emit_opcode(cbuf,0xE8);       // Call into runtime
    emit_d32_reloc(cbuf, (CAST_FROM_FN_PTR(address, SharedRuntime::ldiv) - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
    // Restore stack
    emit_opcode(cbuf, 0x83); // add  SP, #framesize
    emit_rm(cbuf, 0x3, 0x00, ESP_enc);
    emit_d8(cbuf, 4*4);
  %}

  enc_class long_mod( eRegL src1, eRegL src2 ) %{
    // PUSH src1.hi
    emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src1$$reg) );
    // PUSH src1.lo
    emit_opcode(cbuf,               0x50+$src1$$reg  );
    // PUSH src2.hi
    emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src2$$reg) );
    // PUSH src2.lo
    emit_opcode(cbuf,               0x50+$src2$$reg  );
    // CALL directly to the runtime
    cbuf.set_insts_mark();
    emit_opcode(cbuf,0xE8);       // Call into runtime
    emit_d32_reloc(cbuf, (CAST_FROM_FN_PTR(address, SharedRuntime::lrem ) - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
    // Restore stack
    emit_opcode(cbuf, 0x83); // add  SP, #framesize
    emit_rm(cbuf, 0x3, 0x00, ESP_enc);
    emit_d8(cbuf, 4*4);
  %}

  enc_class long_cmp_flags0( eRegL src, rRegI tmp ) %{
    // MOV   $tmp,$src.lo
    emit_opcode(cbuf, 0x8B);
    emit_rm(cbuf, 0x3, $tmp$$reg, $src$$reg);
    // OR    $tmp,$src.hi
    emit_opcode(cbuf, 0x0B);
    emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src$$reg));
  %}

  enc_class long_cmp_flags1( eRegL src1, eRegL src2 ) %{
    // CMP    $src1.lo,$src2.lo
    emit_opcode( cbuf, 0x3B );
    emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
    // JNE,s  skip
    emit_cc(cbuf, 0x70, 0x5);
    emit_d8(cbuf,2);
    // CMP    $src1.hi,$src2.hi
    emit_opcode( cbuf, 0x3B );
    emit_rm(cbuf, 0x3, HIGH_FROM_LOW($src1$$reg), HIGH_FROM_LOW($src2$$reg) );
  %}

  enc_class long_cmp_flags2( eRegL src1, eRegL src2, rRegI tmp ) %{
    // CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits
    emit_opcode( cbuf, 0x3B );
    emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
    // MOV    $tmp,$src1.hi
    emit_opcode( cbuf, 0x8B );
    emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src1$$reg) );
    // SBB   $tmp,$src2.hi\t! Compute flags for long compare
    emit_opcode( cbuf, 0x1B );
    emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src2$$reg) );
  %}

  enc_class long_cmp_flags3( eRegL src, rRegI tmp ) %{
    // XOR    $tmp,$tmp
    emit_opcode(cbuf,0x33);  // XOR
    emit_rm(cbuf,0x3, $tmp$$reg, $tmp$$reg);
    // CMP    $tmp,$src.lo
    emit_opcode( cbuf, 0x3B );
    emit_rm(cbuf, 0x3, $tmp$$reg, $src$$reg );
    // SBB    $tmp,$src.hi
    emit_opcode( cbuf, 0x1B );
    emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src$$reg) );
  %}

 // Sniff, sniff... smells like Gnu Superoptimizer
  enc_class neg_long( eRegL dst ) %{
    emit_opcode(cbuf,0xF7);    // NEG hi
    emit_rm    (cbuf,0x3, 0x3, HIGH_FROM_LOW($dst$$reg));
    emit_opcode(cbuf,0xF7);    // NEG lo
    emit_rm    (cbuf,0x3, 0x3,               $dst$$reg );
    emit_opcode(cbuf,0x83);    // SBB hi,0
    emit_rm    (cbuf,0x3, 0x3, HIGH_FROM_LOW($dst$$reg));
    emit_d8    (cbuf,0 );
  %}


  // Because the transitions from emitted code to the runtime
  // monitorenter/exit helper stubs are so slow it's critical that
  // we inline both the stack-locking fast-path and the inflated fast path.
  //
  // See also: cmpFastLock and cmpFastUnlock.
  //
  // What follows is a specialized inline transliteration of the code
  // in slow_enter() and slow_exit().  If we're concerned about I$ bloat
  // another option would be to emit TrySlowEnter and TrySlowExit methods
  // at startup-time.  These methods would accept arguments as
  // (rax,=Obj, rbx=Self, rcx=box, rdx=Scratch) and return success-failure
  // indications in the icc.ZFlag.  Fast_Lock and Fast_Unlock would simply
  // marshal the arguments and emit calls to TrySlowEnter and TrySlowExit.
  // In practice, however, the # of lock sites is bounded and is usually small.
  // Besides the call overhead, TrySlowEnter and TrySlowExit might suffer
  // if the processor uses simple bimodal branch predictors keyed by EIP
  // Since the helper routines would be called from multiple synchronization
  // sites.
  //
  // An even better approach would be write "MonitorEnter()" and "MonitorExit()"
  // in java - using j.u.c and unsafe - and just bind the lock and unlock sites
  // to those specialized methods.  That'd give us a mostly platform-independent
  // implementation that the JITs could optimize and inline at their pleasure.
  // Done correctly, the only time we'd need to cross to native could would be
  // to park() or unpark() threads.  We'd also need a few more unsafe operators
  // to (a) prevent compiler-JIT reordering of non-volatile accesses, and
  // (b) explicit barriers or fence operations.
  //
  // TODO:
  //
  // *  Arrange for C2 to pass "Self" into Fast_Lock and Fast_Unlock in one of the registers (scr).
  //    This avoids manifesting the Self pointer in the Fast_Lock and Fast_Unlock terminals.
  //    Given TLAB allocation, Self is usually manifested in a register, so passing it into
  //    the lock operators would typically be faster than reifying Self.
  //
  // *  Ideally I'd define the primitives as:
  //       fast_lock   (nax Obj, nax box, EAX tmp, nax scr) where box, tmp and scr are KILLED.
  //       fast_unlock (nax Obj, EAX box, nax tmp) where box and tmp are KILLED
  //    Unfortunately ADLC bugs prevent us from expressing the ideal form.
  //    Instead, we're stuck with a rather awkward and brittle register assignments below.
  //    Furthermore the register assignments are overconstrained, possibly resulting in
  //    sub-optimal code near the synchronization site.
  //
  // *  Eliminate the sp-proximity tests and just use "== Self" tests instead.
  //    Alternately, use a better sp-proximity test.
  //
  // *  Currently ObjectMonitor._Owner can hold either an sp value or a (THREAD *) value.
  //    Either one is sufficient to uniquely identify a thread.
  //    TODO: eliminate use of sp in _owner and use get_thread(tr) instead.
  //
  // *  Intrinsify notify() and notifyAll() for the common cases where the
  //    object is locked by the calling thread but the waitlist is empty.
  //    avoid the expensive JNI call to JVM_Notify() and JVM_NotifyAll().
  //
  // *  use jccb and jmpb instead of jcc and jmp to improve code density.
  //    But beware of excessive branch density on AMD Opterons.
  //
  // *  Both Fast_Lock and Fast_Unlock set the ICC.ZF to indicate success
  //    or failure of the fast-path.  If the fast-path fails then we pass
  //    control to the slow-path, typically in C.  In Fast_Lock and
  //    Fast_Unlock we often branch to DONE_LABEL, just to find that C2
  //    will emit a conditional branch immediately after the node.
  //    So we have branches to branches and lots of ICC.ZF games.
  //    Instead, it might be better to have C2 pass a "FailureLabel"
  //    into Fast_Lock and Fast_Unlock.  In the case of success, control
  //    will drop through the node.  ICC.ZF is undefined at exit.
  //    In the case of failure, the node will branch directly to the
  //    FailureLabel


  // obj: object to lock
  // box: on-stack box address (displaced header location) - KILLED
  // rax,: tmp -- KILLED
  // scr: tmp -- KILLED
  enc_class Fast_Lock( eRegP obj, eRegP box, eAXRegI tmp, eRegP scr ) %{

    Register objReg = as_Register($obj$$reg);
    Register boxReg = as_Register($box$$reg);
    Register tmpReg = as_Register($tmp$$reg);
    Register scrReg = as_Register($scr$$reg);

    // Ensure the register assignents are disjoint
    guarantee (objReg != boxReg, "") ;
    guarantee (objReg != tmpReg, "") ;
    guarantee (objReg != scrReg, "") ;
    guarantee (boxReg != tmpReg, "") ;
    guarantee (boxReg != scrReg, "") ;
    guarantee (tmpReg == as_Register(EAX_enc), "") ;

    MacroAssembler masm(&cbuf);

    if (_counters != NULL) {
      masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
    }
    if (EmitSync & 1) {
        // set box->dhw = unused_mark (3)
        // Force all sync thru slow-path: slow_enter() and slow_exit() 
        masm.movptr (Address(boxReg, 0), int32_t(markOopDesc::unused_mark())) ;             
        masm.cmpptr (rsp, (int32_t)0) ;                        
    } else 
    if (EmitSync & 2) { 
        Label DONE_LABEL ;           
        if (UseBiasedLocking) {
           // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
           masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
        }

        masm.movptr(tmpReg, Address(objReg, 0)) ;          // fetch markword 
        masm.orptr (tmpReg, 0x1);
        masm.movptr(Address(boxReg, 0), tmpReg);           // Anticipate successful CAS 
        if (os::is_MP()) { masm.lock();  }
        masm.cmpxchgptr(boxReg, Address(objReg, 0));          // Updates tmpReg
        masm.jcc(Assembler::equal, DONE_LABEL);
        // Recursive locking
        masm.subptr(tmpReg, rsp);
        masm.andptr(tmpReg, (int32_t) 0xFFFFF003 );
        masm.movptr(Address(boxReg, 0), tmpReg);
        masm.bind(DONE_LABEL) ; 
    } else {  
      // Possible cases that we'll encounter in fast_lock 
      // ------------------------------------------------
      // * Inflated
      //    -- unlocked
      //    -- Locked
      //       = by self
      //       = by other
      // * biased
      //    -- by Self
      //    -- by other
      // * neutral
      // * stack-locked
      //    -- by self
      //       = sp-proximity test hits
      //       = sp-proximity test generates false-negative
      //    -- by other
      //

      Label IsInflated, DONE_LABEL, PopDone ;

      // TODO: optimize away redundant LDs of obj->mark and improve the markword triage
      // order to reduce the number of conditional branches in the most common cases.
      // Beware -- there's a subtle invariant that fetch of the markword
      // at [FETCH], below, will never observe a biased encoding (*101b).
      // If this invariant is not held we risk exclusion (safety) failure.
      if (UseBiasedLocking && !UseOptoBiasInlining) {
        masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
      }

      masm.movptr(tmpReg, Address(objReg, 0)) ;         // [FETCH]
      masm.testptr(tmpReg, 0x02) ;                      // Inflated v (Stack-locked or neutral)
      masm.jccb  (Assembler::notZero, IsInflated) ;

      // Attempt stack-locking ...
      masm.orptr (tmpReg, 0x1);
      masm.movptr(Address(boxReg, 0), tmpReg);          // Anticipate successful CAS
      if (os::is_MP()) { masm.lock();  }
      masm.cmpxchgptr(boxReg, Address(objReg, 0));           // Updates tmpReg
      if (_counters != NULL) {
        masm.cond_inc32(Assembler::equal,
                        ExternalAddress((address)_counters->fast_path_entry_count_addr()));
      }
      masm.jccb (Assembler::equal, DONE_LABEL);

      // Recursive locking
      masm.subptr(tmpReg, rsp);
      masm.andptr(tmpReg, 0xFFFFF003 );
      masm.movptr(Address(boxReg, 0), tmpReg);
      if (_counters != NULL) {
        masm.cond_inc32(Assembler::equal,
                        ExternalAddress((address)_counters->fast_path_entry_count_addr()));
      }
      masm.jmp  (DONE_LABEL) ;

      masm.bind (IsInflated) ;

      // The object is inflated.
      //
      // TODO-FIXME: eliminate the ugly use of manifest constants:
      //   Use markOopDesc::monitor_value instead of "2".
      //   use markOop::unused_mark() instead of "3".
      // The tmpReg value is an objectMonitor reference ORed with
      // markOopDesc::monitor_value (2).   We can either convert tmpReg to an
      // objectmonitor pointer by masking off the "2" bit or we can just
      // use tmpReg as an objectmonitor pointer but bias the objectmonitor
      // field offsets with "-2" to compensate for and annul the low-order tag bit.
      //
      // I use the latter as it avoids AGI stalls.
      // As such, we write "mov r, [tmpReg+OFFSETOF(Owner)-2]"
      // instead of "mov r, [tmpReg+OFFSETOF(Owner)]".
      //
      #define OFFSET_SKEWED(f) ((ObjectMonitor::f ## _offset_in_bytes())-2)

      // boxReg refers to the on-stack BasicLock in the current frame.
      // We'd like to write:
      //   set box->_displaced_header = markOop::unused_mark().  Any non-0 value suffices.
      // This is convenient but results a ST-before-CAS penalty.  The following CAS suffers
      // additional latency as we have another ST in the store buffer that must drain.

      if (EmitSync & 8192) { 
         masm.movptr(Address(boxReg, 0), 3) ;            // results in ST-before-CAS penalty
         masm.get_thread (scrReg) ; 
         masm.movptr(boxReg, tmpReg);                    // consider: LEA box, [tmp-2] 
         masm.movptr(tmpReg, NULL_WORD);                 // consider: xor vs mov
         if (os::is_MP()) { masm.lock(); } 
         masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
      } else 
      if ((EmitSync & 128) == 0) {                      // avoid ST-before-CAS
         masm.movptr(scrReg, boxReg) ; 
         masm.movptr(boxReg, tmpReg);                   // consider: LEA box, [tmp-2] 

         // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
         if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
            // prefetchw [eax + Offset(_owner)-2]
            masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
         }

         if ((EmitSync & 64) == 0) {
           // Optimistic form: consider XORL tmpReg,tmpReg
           masm.movptr(tmpReg, NULL_WORD) ; 
         } else { 
           // Can suffer RTS->RTO upgrades on shared or cold $ lines
           // Test-And-CAS instead of CAS
           masm.movptr(tmpReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;   // rax, = m->_owner
           masm.testptr(tmpReg, tmpReg) ;                   // Locked ? 
           masm.jccb  (Assembler::notZero, DONE_LABEL) ;                   
         }

         // Appears unlocked - try to swing _owner from null to non-null.
         // Ideally, I'd manifest "Self" with get_thread and then attempt
         // to CAS the register containing Self into m->Owner.
         // But we don't have enough registers, so instead we can either try to CAS
         // rsp or the address of the box (in scr) into &m->owner.  If the CAS succeeds
         // we later store "Self" into m->Owner.  Transiently storing a stack address
         // (rsp or the address of the box) into  m->owner is harmless.
         // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
         if (os::is_MP()) { masm.lock();  }
         masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
         masm.movptr(Address(scrReg, 0), 3) ;          // box->_displaced_header = 3
         masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
         masm.get_thread (scrReg) ;                    // beware: clobbers ICCs
         masm.movptr(Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2), scrReg) ; 
         masm.xorptr(boxReg, boxReg) ;                 // set icc.ZFlag = 1 to indicate success
                       
         // If the CAS fails we can either retry or pass control to the slow-path.  
         // We use the latter tactic.  
         // Pass the CAS result in the icc.ZFlag into DONE_LABEL
         // If the CAS was successful ...
         //   Self has acquired the lock
         //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
         // Intentional fall-through into DONE_LABEL ...
      } else {
         masm.movptr(Address(boxReg, 0), 3) ;       // results in ST-before-CAS penalty
         masm.movptr(boxReg, tmpReg) ; 

         // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
         if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
            // prefetchw [eax + Offset(_owner)-2]
            masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
         }

         if ((EmitSync & 64) == 0) {
           // Optimistic form
           masm.xorptr  (tmpReg, tmpReg) ; 
         } else { 
           // Can suffer RTS->RTO upgrades on shared or cold $ lines
           masm.movptr(tmpReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;   // rax, = m->_owner
           masm.testptr(tmpReg, tmpReg) ;                   // Locked ? 
           masm.jccb  (Assembler::notZero, DONE_LABEL) ;                   
         }

         // Appears unlocked - try to swing _owner from null to non-null.
         // Use either "Self" (in scr) or rsp as thread identity in _owner.
         // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
         masm.get_thread (scrReg) ;
         if (os::is_MP()) { masm.lock(); }
         masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;

         // If the CAS fails we can either retry or pass control to the slow-path.
         // We use the latter tactic.
         // Pass the CAS result in the icc.ZFlag into DONE_LABEL
         // If the CAS was successful ...
         //   Self has acquired the lock
         //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
         // Intentional fall-through into DONE_LABEL ...
      }

      // DONE_LABEL is a hot target - we'd really like to place it at the
      // start of cache line by padding with NOPs.
      // See the AMD and Intel software optimization manuals for the
      // most efficient "long" NOP encodings.
      // Unfortunately none of our alignment mechanisms suffice.
      masm.bind(DONE_LABEL);

      // Avoid branch-to-branch on AMD processors
      // This appears to be superstition.
      if (EmitSync & 32) masm.nop() ;


      // At DONE_LABEL the icc ZFlag is set as follows ...
      // Fast_Unlock uses the same protocol.
      // ZFlag == 1 -> Success
      // ZFlag == 0 -> Failure - force control through the slow-path
    }
  %}

  // obj: object to unlock
  // box: box address (displaced header location), killed.  Must be EAX.
  // rbx,: killed tmp; cannot be obj nor box.
  //
  // Some commentary on balanced locking:
  //
  // Fast_Lock and Fast_Unlock are emitted only for provably balanced lock sites.
  // Methods that don't have provably balanced locking are forced to run in the
  // interpreter - such methods won't be compiled to use fast_lock and fast_unlock.
  // The interpreter provides two properties:
  // I1:  At return-time the interpreter automatically and quietly unlocks any
  //      objects acquired the current activation (frame).  Recall that the
  //      interpreter maintains an on-stack list of locks currently held by
  //      a frame.
  // I2:  If a method attempts to unlock an object that is not held by the
  //      the frame the interpreter throws IMSX.
  //
  // Lets say A(), which has provably balanced locking, acquires O and then calls B().
  // B() doesn't have provably balanced locking so it runs in the interpreter.
  // Control returns to A() and A() unlocks O.  By I1 and I2, above, we know that O
  // is still locked by A().
  //
  // The only other source of unbalanced locking would be JNI.  The "Java Native Interface:
  // Programmer's Guide and Specification" claims that an object locked by jni_monitorenter
  // should not be unlocked by "normal" java-level locking and vice-versa.  The specification
  // doesn't specify what will occur if a program engages in such mixed-mode locking, however.

  enc_class Fast_Unlock( nabxRegP obj, eAXRegP box, eRegP tmp) %{

    Register objReg = as_Register($obj$$reg);
    Register boxReg = as_Register($box$$reg);
    Register tmpReg = as_Register($tmp$$reg);

    guarantee (objReg != boxReg, "") ;
    guarantee (objReg != tmpReg, "") ;
    guarantee (boxReg != tmpReg, "") ;
    guarantee (boxReg == as_Register(EAX_enc), "") ;
    MacroAssembler masm(&cbuf);

    if (EmitSync & 4) {
      // Disable - inhibit all inlining.  Force control through the slow-path
      masm.cmpptr (rsp, 0) ; 
    } else 
    if (EmitSync & 8) {
      Label DONE_LABEL ;
      if (UseBiasedLocking) {
         masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
      }
      // classic stack-locking code ...
      masm.movptr(tmpReg, Address(boxReg, 0)) ;
      masm.testptr(tmpReg, tmpReg) ;
      masm.jcc   (Assembler::zero, DONE_LABEL) ;
      if (os::is_MP()) { masm.lock(); }
      masm.cmpxchgptr(tmpReg, Address(objReg, 0));          // Uses EAX which is box
      masm.bind(DONE_LABEL);
    } else {
      Label DONE_LABEL, Stacked, CheckSucc, Inflated ;

      // Critically, the biased locking test must have precedence over
      // and appear before the (box->dhw == 0) recursive stack-lock test.
      if (UseBiasedLocking && !UseOptoBiasInlining) {
         masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
      }
      
      masm.cmpptr(Address(boxReg, 0), 0) ;            // Examine the displaced header
      masm.movptr(tmpReg, Address(objReg, 0)) ;       // Examine the object's markword
      masm.jccb  (Assembler::zero, DONE_LABEL) ;      // 0 indicates recursive stack-lock

      masm.testptr(tmpReg, 0x02) ;                     // Inflated? 
      masm.jccb  (Assembler::zero, Stacked) ;

      masm.bind  (Inflated) ;
      // It's inflated.
      // Despite our balanced locking property we still check that m->_owner == Self
      // as java routines or native JNI code called by this thread might
      // have released the lock.
      // Refer to the comments in synchronizer.cpp for how we might encode extra
      // state in _succ so we can avoid fetching EntryList|cxq.
      //
      // I'd like to add more cases in fast_lock() and fast_unlock() --
      // such as recursive enter and exit -- but we have to be wary of
      // I$ bloat, T$ effects and BP$ effects.
      //
      // If there's no contention try a 1-0 exit.  That is, exit without
      // a costly MEMBAR or CAS.  See synchronizer.cpp for details on how
      // we detect and recover from the race that the 1-0 exit admits.
      //
      // Conceptually Fast_Unlock() must execute a STST|LDST "release" barrier
      // before it STs null into _owner, releasing the lock.  Updates
      // to data protected by the critical section must be visible before
      // we drop the lock (and thus before any other thread could acquire
      // the lock and observe the fields protected by the lock).
      // IA32's memory-model is SPO, so STs are ordered with respect to
      // each other and there's no need for an explicit barrier (fence).
      // See also http://gee.cs.oswego.edu/dl/jmm/cookbook.html.

      masm.get_thread (boxReg) ;
      if ((EmitSync & 4096) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
        // prefetchw [ebx + Offset(_owner)-2]
        masm.prefetchw(Address(rbx, ObjectMonitor::owner_offset_in_bytes()-2));
      }

      // Note that we could employ various encoding schemes to reduce
      // the number of loads below (currently 4) to just 2 or 3.
      // Refer to the comments in synchronizer.cpp.
      // In practice the chain of fetches doesn't seem to impact performance, however.
      if ((EmitSync & 65536) == 0 && (EmitSync & 256)) {
         // Attempt to reduce branch density - AMD's branch predictor.
         masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;  
         masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
         masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
         masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
         masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
         masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
         masm.jmpb  (DONE_LABEL) ; 
      } else { 
         masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;  
         masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
         masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
         masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
         masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
         masm.jccb  (Assembler::notZero, CheckSucc) ; 
         masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
         masm.jmpb  (DONE_LABEL) ; 
      }

      // The Following code fragment (EmitSync & 65536) improves the performance of
      // contended applications and contended synchronization microbenchmarks.
      // Unfortunately the emission of the code - even though not executed - causes regressions
      // in scimark and jetstream, evidently because of $ effects.  Replacing the code
      // with an equal number of never-executed NOPs results in the same regression.
      // We leave it off by default.

      if ((EmitSync & 65536) != 0) {
         Label LSuccess, LGoSlowPath ;

         masm.bind  (CheckSucc) ;

         // Optional pre-test ... it's safe to elide this
         if ((EmitSync & 16) == 0) { 
            masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), 0) ; 
            masm.jccb  (Assembler::zero, LGoSlowPath) ; 
         }

         // We have a classic Dekker-style idiom:
         //    ST m->_owner = 0 ; MEMBAR; LD m->_succ
         // There are a number of ways to implement the barrier:
         // (1) lock:andl &m->_owner, 0
         //     is fast, but mask doesn't currently support the "ANDL M,IMM32" form.
         //     LOCK: ANDL [ebx+Offset(_Owner)-2], 0
         //     Encodes as 81 31 OFF32 IMM32 or 83 63 OFF8 IMM8
         // (2) If supported, an explicit MFENCE is appealing.
         //     In older IA32 processors MFENCE is slower than lock:add or xchg
         //     particularly if the write-buffer is full as might be the case if
         //     if stores closely precede the fence or fence-equivalent instruction.
         //     In more modern implementations MFENCE appears faster, however.
         // (3) In lieu of an explicit fence, use lock:addl to the top-of-stack
         //     The $lines underlying the top-of-stack should be in M-state.
         //     The locked add instruction is serializing, of course.
         // (4) Use xchg, which is serializing
         //     mov boxReg, 0; xchgl boxReg, [tmpReg + Offset(_owner)-2] also works
         // (5) ST m->_owner = 0 and then execute lock:orl &m->_succ, 0.
         //     The integer condition codes will tell us if succ was 0.
         //     Since _succ and _owner should reside in the same $line and
         //     we just stored into _owner, it's likely that the $line
         //     remains in M-state for the lock:orl.
         //
         // We currently use (3), although it's likely that switching to (2)
         // is correct for the future.
            
         masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
         if (os::is_MP()) { 
            if (VM_Version::supports_sse2() && 1 == FenceInstruction) { 
              masm.mfence();
            } else { 
              masm.lock () ; masm.addptr(Address(rsp, 0), 0) ; 
            }
         }
         // Ratify _succ remains non-null
         masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), 0) ; 
         masm.jccb  (Assembler::notZero, LSuccess) ; 

         masm.xorptr(boxReg, boxReg) ;                  // box is really EAX
         if (os::is_MP()) { masm.lock(); }
         masm.cmpxchgptr(rsp, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
         masm.jccb  (Assembler::notEqual, LSuccess) ;
         // Since we're low on registers we installed rsp as a placeholding in _owner.
         // Now install Self over rsp.  This is safe as we're transitioning from
         // non-null to non=null
         masm.get_thread (boxReg) ;
         masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), boxReg) ;
         // Intentional fall-through into LGoSlowPath ...

         masm.bind  (LGoSlowPath) ; 
         masm.orptr(boxReg, 1) ;                      // set ICC.ZF=0 to indicate failure
         masm.jmpb  (DONE_LABEL) ; 

         masm.bind  (LSuccess) ; 
         masm.xorptr(boxReg, boxReg) ;                 // set ICC.ZF=1 to indicate success
         masm.jmpb  (DONE_LABEL) ; 
      }

      masm.bind (Stacked) ;
      // It's not inflated and it's not recursively stack-locked and it's not biased.
      // It must be stack-locked.
      // Try to reset the header to displaced header.
      // The "box" value on the stack is stable, so we can reload
      // and be assured we observe the same value as above.
      masm.movptr(tmpReg, Address(boxReg, 0)) ;
      if (os::is_MP()) {   masm.lock();    }
      masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses EAX which is box
      // Intention fall-thru into DONE_LABEL


      // DONE_LABEL is a hot target - we'd really like to place it at the
      // start of cache line by padding with NOPs.
      // See the AMD and Intel software optimization manuals for the
      // most efficient "long" NOP encodings.
      // Unfortunately none of our alignment mechanisms suffice.
      if ((EmitSync & 65536) == 0) {
         masm.bind (CheckSucc) ;
      }
      masm.bind(DONE_LABEL);

      // Avoid branch to branch on AMD processors
      if (EmitSync & 32768) { masm.nop() ; }
    }
  %}


  enc_class enc_pop_rdx() %{
    emit_opcode(cbuf,0x5A);
  %}

  enc_class enc_rethrow() %{
    cbuf.set_insts_mark();
    emit_opcode(cbuf, 0xE9);        // jmp    entry
    emit_d32_reloc(cbuf, (int)OptoRuntime::rethrow_stub() - ((int)cbuf.insts_end())-4,
                   runtime_call_Relocation::spec(), RELOC_IMM32 );
  %}


  // Convert a double to an int.  Java semantics require we do complex
  // manglelations in the corner cases.  So we set the rounding mode to
  // 'zero', store the darned double down as an int, and reset the
  // rounding mode to 'nearest'.  The hardware throws an exception which
  // patches up the correct value directly to the stack.
  enc_class DPR2I_encoding( regDPR src ) %{
    // Flip to round-to-zero mode.  We attempted to allow invalid-op
    // exceptions here, so that a NAN or other corner-case value will
    // thrown an exception (but normal values get converted at full speed).
    // However, I2C adapters and other float-stack manglers leave pending
    // invalid-op exceptions hanging.  We would have to clear them before
    // enabling them and that is more expensive than just testing for the
    // invalid value Intel stores down in the corner cases.
    emit_opcode(cbuf,0xD9);            // FLDCW  trunc
    emit_opcode(cbuf,0x2D);
    emit_d32(cbuf,(int)StubRoutines::addr_fpu_cntrl_wrd_trunc());
    // Allocate a word
    emit_opcode(cbuf,0x83);            // SUB ESP,4
    emit_opcode(cbuf,0xEC);
    emit_d8(cbuf,0x04);
    // Encoding assumes a double has been pushed into FPR0.
    // Store down the double as an int, popping the FPU stack
    emit_opcode(cbuf,0xDB);            // FISTP [ESP]
    emit_opcode(cbuf,0x1C);
    emit_d8(cbuf,0x24);
    // Restore the rounding mode; mask the exception
    emit_opcode(cbuf,0xD9);            // FLDCW   std/24-bit mode
    emit_opcode(cbuf,0x2D);
    emit_d32( cbuf, Compile::current()->in_24_bit_fp_mode()
        ? (int)StubRoutines::addr_fpu_cntrl_wrd_24()
        : (int)StubRoutines::addr_fpu_cntrl_wrd_std());

    // Load the converted int; adjust CPU stack
    emit_opcode(cbuf,0x58);       // POP EAX
    emit_opcode(cbuf,0x3D);       // CMP EAX,imm
    emit_d32   (cbuf,0x80000000); //         0x80000000
    emit_opcode(cbuf,0x75);       // JNE around_slow_call
    emit_d8    (cbuf,0x07);       // Size of slow_call
    // Push src onto stack slow-path
    emit_opcode(cbuf,0xD9 );      // FLD     ST(i)
    emit_d8    (cbuf,0xC0-1+$src$$reg );
    // CALL directly to the runtime
    cbuf.set_insts_mark();
    emit_opcode(cbuf,0xE8);       // Call into runtime
    emit_d32_reloc(cbuf, (StubRoutines::d2i_wrapper() - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
    // Carry on here...
  %}

  enc_class DPR2L_encoding( regDPR src ) %{
    emit_opcode(cbuf,0xD9);            // FLDCW  trunc
    emit_opcode(cbuf,0x2D);
    emit_d32(cbuf,(int)StubRoutines::addr_fpu_cntrl_wrd_trunc());
    // Allocate a word
    emit_opcode(cbuf,0x83);            // SUB ESP,8
    emit_opcode(cbuf,0xEC);
    emit_d8(cbuf,0x08);
    // Encoding assumes a double has been pushed into FPR0.
    // Store down the double as a long, popping the FPU stack
    emit_opcode(cbuf,0xDF);            // FISTP [ESP]
    emit_opcode(cbuf,0x3C);
    emit_d8(cbuf,0x24);
    // Restore the rounding mode; mask the exception
    emit_opcode(cbuf,0xD9);            // FLDCW   std/24-bit mode
    emit_opcode(cbuf,0x2D);
    emit_d32( cbuf, Compile::current()->in_24_bit_fp_mode()
        ? (int)StubRoutines::addr_fpu_cntrl_wrd_24()
        : (int)StubRoutines::addr_fpu_cntrl_wrd_std());

    // Load the converted int; adjust CPU stack
    emit_opcode(cbuf,0x58);       // POP EAX
    emit_opcode(cbuf,0x5A);       // POP EDX
    emit_opcode(cbuf,0x81);       // CMP EDX,imm
    emit_d8    (cbuf,0xFA);       // rdx
    emit_d32   (cbuf,0x80000000); //         0x80000000
    emit_opcode(cbuf,0x75);       // JNE around_slow_call
    emit_d8    (cbuf,0x07+4);     // Size of slow_call
    emit_opcode(cbuf,0x85);       // TEST EAX,EAX
    emit_opcode(cbuf,0xC0);       // 2/rax,/rax,
    emit_opcode(cbuf,0x75);       // JNE around_slow_call
    emit_d8    (cbuf,0x07);       // Size of slow_call
    // Push src onto stack slow-path
    emit_opcode(cbuf,0xD9 );      // FLD     ST(i)
    emit_d8    (cbuf,0xC0-1+$src$$reg );
    // CALL directly to the runtime
    cbuf.set_insts_mark();
    emit_opcode(cbuf,0xE8);       // Call into runtime
    emit_d32_reloc(cbuf, (StubRoutines::d2l_wrapper() - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
    // Carry on here...
  %}

  enc_class FMul_ST_reg( eRegFPR src1 ) %{
    // Operand was loaded from memory into fp ST (stack top)
    // FMUL   ST,$src  /* D8 C8+i */
    emit_opcode(cbuf, 0xD8);
    emit_opcode(cbuf, 0xC8 + $src1$$reg);
  %}

  enc_class FAdd_ST_reg( eRegFPR src2 ) %{
    // FADDP  ST,src2  /* D8 C0+i */
    emit_opcode(cbuf, 0xD8);
    emit_opcode(cbuf, 0xC0 + $src2$$reg);
    //could use FADDP  src2,fpST  /* DE C0+i */
  %}

  enc_class FAddP_reg_ST( eRegFPR src2 ) %{
    // FADDP  src2,ST  /* DE C0+i */
    emit_opcode(cbuf, 0xDE);
    emit_opcode(cbuf, 0xC0 + $src2$$reg);
  %}

  enc_class subFPR_divFPR_encode( eRegFPR src1, eRegFPR src2) %{
    // Operand has been loaded into fp ST (stack top)
      // FSUB   ST,$src1
      emit_opcode(cbuf, 0xD8);
      emit_opcode(cbuf, 0xE0 + $src1$$reg);

      // FDIV
      emit_opcode(cbuf, 0xD8);
      emit_opcode(cbuf, 0xF0 + $src2$$reg);
  %}

  enc_class MulFAddF (eRegFPR src1, eRegFPR src2) %{
    // Operand was loaded from memory into fp ST (stack top)
    // FADD   ST,$src  /* D8 C0+i */
    emit_opcode(cbuf, 0xD8);
    emit_opcode(cbuf, 0xC0 + $src1$$reg);

    // FMUL  ST,src2  /* D8 C*+i */
    emit_opcode(cbuf, 0xD8);
    emit_opcode(cbuf, 0xC8 + $src2$$reg);
  %}


  enc_class MulFAddFreverse (eRegFPR src1, eRegFPR src2) %{
    // Operand was loaded from memory into fp ST (stack top)
    // FADD   ST,$src  /* D8 C0+i */
    emit_opcode(cbuf, 0xD8);
    emit_opcode(cbuf, 0xC0 + $src1$$reg);

    // FMULP  src2,ST  /* DE C8+i */
    emit_opcode(cbuf, 0xDE);
    emit_opcode(cbuf, 0xC8 + $src2$$reg);
  %}

  // Atomically load the volatile long
  enc_class enc_loadL_volatile( memory mem, stackSlotL dst ) %{
    emit_opcode(cbuf,0xDF);
    int rm_byte_opcode = 0x05;
    int base     = $mem$$base;
    int index    = $mem$$index;
    int scale    = $mem$$scale;
    int displace = $mem$$disp;
    relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
    encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_reloc);
    store_to_stackslot( cbuf, 0x0DF, 0x07, $dst$$disp );
  %}

  // Volatile Store Long.  Must be atomic, so move it into
  // the FP TOS and then do a 64-bit FIST.  Has to probe the
  // target address before the store (for null-ptr checks)
  // so the memory operand is used twice in the encoding.
  enc_class enc_storeL_volatile( memory mem, stackSlotL src ) %{
    store_to_stackslot( cbuf, 0x0DF, 0x05, $src$$disp );
    cbuf.set_insts_mark();            // Mark start of FIST in case $mem has an oop
    emit_opcode(cbuf,0xDF);
    int rm_byte_opcode = 0x07;
    int base     = $mem$$base;
    int index    = $mem$$index;
    int scale    = $mem$$scale;
    int displace = $mem$$disp;
    relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
    encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_reloc);
  %}

  // Safepoint Poll.  This polls the safepoint page, and causes an
  // exception if it is not readable. Unfortunately, it kills the condition code
  // in the process
  // We current use TESTL [spp],EDI
  // A better choice might be TESTB [spp + pagesize() - CacheLineSize()],0

  enc_class Safepoint_Poll() %{
    cbuf.relocate(cbuf.insts_mark(), relocInfo::poll_type, 0);
    emit_opcode(cbuf,0x85);
    emit_rm (cbuf, 0x0, 0x7, 0x5);
    emit_d32(cbuf, (intptr_t)os::get_polling_page());
  %}
%}


//----------FRAME--------------------------------------------------------------
// Definition of frame structure and management information.
//
//  S T A C K   L A Y O U T    Allocators stack-slot number
//                             |   (to get allocators register number
//  G  Owned by    |        |  v    add OptoReg::stack0())
//  r   CALLER     |        |
//  o     |        +--------+      pad to even-align allocators stack-slot
//  w     V        |  pad0  |        numbers; owned by CALLER
//  t   -----------+--------+----> Matcher::_in_arg_limit, unaligned
//  h     ^        |   in   |  5
//        |        |  args  |  4   Holes in incoming args owned by SELF
//  |     |        |        |  3
//  |     |        +--------+
//  V     |        | old out|      Empty on Intel, window on Sparc
//        |    old |preserve|      Must be even aligned.
//        |     SP-+--------+----> Matcher::_old_SP, even aligned
//        |        |   in   |  3   area for Intel ret address
//     Owned by    |preserve|      Empty on Sparc.
//       SELF      +--------+
//        |        |  pad2  |  2   pad to align old SP
//        |        +--------+  1
//        |        | locks  |  0
//        |        +--------+----> OptoReg::stack0(), even aligned
//        |        |  pad1  | 11   pad to align new SP
//        |        +--------+
//        |        |        | 10
//        |        | spills |  9   spills
//        V        |        |  8   (pad0 slot for callee)
//      -----------+--------+----> Matcher::_out_arg_limit, unaligned
//        ^        |  out   |  7
//        |        |  args  |  6   Holes in outgoing args owned by CALLEE
//     Owned by    +--------+
//      CALLEE     | new out|  6   Empty on Intel, window on Sparc
//        |    new |preserve|      Must be even-aligned.
//        |     SP-+--------+----> Matcher::_new_SP, even aligned
//        |        |        |
//
// Note 1: Only region 8-11 is determined by the allocator.  Region 0-5 is
//         known from SELF's arguments and the Java calling convention.
//         Region 6-7 is determined per call site.
// Note 2: If the calling convention leaves holes in the incoming argument
//         area, those holes are owned by SELF.  Holes in the outgoing area
//         are owned by the CALLEE.  Holes should not be nessecary in the
//         incoming area, as the Java calling convention is completely under
//         the control of the AD file.  Doubles can be sorted and packed to
//         avoid holes.  Holes in the outgoing arguments may be nessecary for
//         varargs C calling conventions.
// Note 3: Region 0-3 is even aligned, with pad2 as needed.  Region 3-5 is
//         even aligned with pad0 as needed.
//         Region 6 is even aligned.  Region 6-7 is NOT even aligned;
//         region 6-11 is even aligned; it may be padded out more so that
//         the region from SP to FP meets the minimum stack alignment.

frame %{
  // What direction does stack grow in (assumed to be same for C & Java)
  stack_direction(TOWARDS_LOW);

  // These three registers define part of the calling convention
  // between compiled code and the interpreter.
  inline_cache_reg(EAX);                // Inline Cache Register
  interpreter_method_oop_reg(EBX);      // Method Oop Register when calling interpreter

  // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
  cisc_spilling_operand_name(indOffset32);

  // Number of stack slots consumed by locking an object
  sync_stack_slots(1);

  // Compiled code's Frame Pointer
  frame_pointer(ESP);
  // Interpreter stores its frame pointer in a register which is
  // stored to the stack by I2CAdaptors.
  // I2CAdaptors convert from interpreted java to compiled java.
  interpreter_frame_pointer(EBP);

  // Stack alignment requirement
  // Alignment size in bytes (128-bit -> 16 bytes)
  stack_alignment(StackAlignmentInBytes);

  // Number of stack slots between incoming argument block and the start of
  // a new frame.  The PROLOG must add this many slots to the stack.  The
  // EPILOG must remove this many slots.  Intel needs one slot for
  // return address and one for rbp, (must save rbp)
  in_preserve_stack_slots(2+VerifyStackAtCalls);

  // Number of outgoing stack slots killed above the out_preserve_stack_slots
  // for calls to C.  Supports the var-args backing area for register parms.
  varargs_C_out_slots_killed(0);

  // The after-PROLOG location of the return address.  Location of
  // return address specifies a type (REG or STACK) and a number
  // representing the register number (i.e. - use a register name) or
  // stack slot.
  // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
  // Otherwise, it is above the locks and verification slot and alignment word
  return_addr(STACK - 1 +
              round_to((Compile::current()->in_preserve_stack_slots() +
                        Compile::current()->fixed_slots()),
                       stack_alignment_in_slots()));

  // Body of function which returns an integer array locating
  // arguments either in registers or in stack slots.  Passed an array
  // of ideal registers called "sig" and a "length" count.  Stack-slot
  // offsets are based on outgoing arguments, i.e. a CALLER setting up
  // arguments for a CALLEE.  Incoming stack arguments are
  // automatically biased by the preserve_stack_slots field above.
  calling_convention %{
    // No difference between ingoing/outgoing just pass false
    SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
  %}


  // Body of function which returns an integer array locating
  // arguments either in registers or in stack slots.  Passed an array
  // of ideal registers called "sig" and a "length" count.  Stack-slot
  // offsets are based on outgoing arguments, i.e. a CALLER setting up
  // arguments for a CALLEE.  Incoming stack arguments are
  // automatically biased by the preserve_stack_slots field above.
  c_calling_convention %{
    // This is obviously always outgoing
    (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
  %}

  // Location of C & interpreter return values
  c_return_value %{
    assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
    static int lo[Op_RegL+1] = { 0, 0, OptoReg::Bad, EAX_num,      EAX_num,      FPR1L_num,    FPR1L_num, EAX_num };
    static int hi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, FPR1H_num, EDX_num };

    // in SSE2+ mode we want to keep the FPU stack clean so pretend
    // that C functions return float and double results in XMM0.
    if( ideal_reg == Op_RegD && UseSSE>=2 )
      return OptoRegPair(XMM0b_num,XMM0_num);
    if( ideal_reg == Op_RegF && UseSSE>=2 )
      return OptoRegPair(OptoReg::Bad,XMM0_num);

    return OptoRegPair(hi[ideal_reg],lo[ideal_reg]);
  %}

  // Location of return values
  return_value %{
    assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
    static int lo[Op_RegL+1] = { 0, 0, OptoReg::Bad, EAX_num,      EAX_num,      FPR1L_num,    FPR1L_num, EAX_num };
    static int hi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, FPR1H_num, EDX_num };
    if( ideal_reg == Op_RegD && UseSSE>=2 )
      return OptoRegPair(XMM0b_num,XMM0_num);
    if( ideal_reg == Op_RegF && UseSSE>=1 )
      return OptoRegPair(OptoReg::Bad,XMM0_num);
    return OptoRegPair(hi[ideal_reg],lo[ideal_reg]);
  %}

%}

//----------ATTRIBUTES---------------------------------------------------------
//----------Operand Attributes-------------------------------------------------
op_attrib op_cost(0);        // Required cost attribute

//----------Instruction Attributes---------------------------------------------
ins_attrib ins_cost(100);       // Required cost attribute
ins_attrib ins_size(8);         // Required size attribute (in bits)
ins_attrib ins_short_branch(0); // Required flag: is this instruction a
                                // non-matching short branch variant of some
                                                            // long branch?
ins_attrib ins_alignment(1);    // Required alignment attribute (must be a power of 2)
                                // specifies the alignment that some part of the instruction (not
                                // necessarily the start) requires.  If > 1, a compute_padding()
                                // function must be provided for the instruction

//----------OPERANDS-----------------------------------------------------------
// Operand definitions must precede instruction definitions for correct parsing
// in the ADLC because operands constitute user defined types which are used in
// instruction definitions.

//----------Simple Operands----------------------------------------------------
// Immediate Operands
// Integer Immediate
operand immI() %{
  match(ConI);

  op_cost(10);
  format %{ %}
  interface(CONST_INTER);
%}

// Constant for test vs zero
operand immI0() %{
  predicate(n->get_int() == 0);
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Constant for increment
operand immI1() %{
  predicate(n->get_int() == 1);
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Constant for decrement
operand immI_M1() %{
  predicate(n->get_int() == -1);
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Valid scale values for addressing modes
operand immI2() %{
  predicate(0 <= n->get_int() && (n->get_int() <= 3));
  match(ConI);

  format %{ %}
  interface(CONST_INTER);
%}

operand immI8() %{
  predicate((-128 <= n->get_int()) && (n->get_int() <= 127));
  match(ConI);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

operand immI16() %{
  predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
  match(ConI);

  op_cost(10);
  format %{ %}
  interface(CONST_INTER);
%}

// Constant for long shifts
operand immI_32() %{
  predicate( n->get_int() == 32 );
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

operand immI_1_31() %{
  predicate( n->get_int() >= 1 && n->get_int() <= 31 );
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

operand immI_32_63() %{
  predicate( n->get_int() >= 32 && n->get_int() <= 63 );
  match(ConI);
  op_cost(0);

  format %{ %}
  interface(CONST_INTER);
%}

operand immI_1() %{
  predicate( n->get_int() == 1 );
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

operand immI_2() %{
  predicate( n->get_int() == 2 );
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

operand immI_3() %{
  predicate( n->get_int() == 3 );
  match(ConI);

  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Pointer Immediate
operand immP() %{
  match(ConP);

  op_cost(10);
  format %{ %}
  interface(CONST_INTER);
%}

// NULL Pointer Immediate
operand immP0() %{
  predicate( n->get_ptr() == 0 );
  match(ConP);
  op_cost(0);

  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate
operand immL() %{
  match(ConL);

  op_cost(20);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate zero
operand immL0() %{
  predicate( n->get_long() == 0L );
  match(ConL);
  op_cost(0);

  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate zero
operand immL_M1() %{
  predicate( n->get_long() == -1L );
  match(ConL);
  op_cost(0);

  format %{ %}
  interface(CONST_INTER);
%}

// Long immediate from 0 to 127.
// Used for a shorter form of long mul by 10.
operand immL_127() %{
  predicate((0 <= n->get_long()) && (n->get_long() <= 127));
  match(ConL);
  op_cost(0);

  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: low 32-bit mask
operand immL_32bits() %{
  predicate(n->get_long() == 0xFFFFFFFFL);
  match(ConL);
  op_cost(0);

  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: low 32-bit mask
operand immL32() %{
  predicate(n->get_long() == (int)(n->get_long()));
  match(ConL);
  op_cost(20);

  format %{ %}
  interface(CONST_INTER);
%}

//Double Immediate zero
operand immDPR0() %{
  // Do additional (and counter-intuitive) test against NaN to work around VC++
  // bug that generates code such that NaNs compare equal to 0.0
  predicate( UseSSE<=1 && n->getd() == 0.0 && !g_isnan(n->getd()) );
  match(ConD);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Double Immediate one
operand immDPR1() %{
  predicate( UseSSE<=1 && n->getd() == 1.0 );
  match(ConD);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Double Immediate
operand immDPR() %{
  predicate(UseSSE<=1);
  match(ConD);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

operand immD() %{
  predicate(UseSSE>=2);
  match(ConD);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Double Immediate zero
operand immD0() %{
  // Do additional (and counter-intuitive) test against NaN to work around VC++
  // bug that generates code such that NaNs compare equal to 0.0 AND do not
  // compare equal to -0.0.
  predicate( UseSSE>=2 && jlong_cast(n->getd()) == 0 );
  match(ConD);

  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate zero
operand immFPR0() %{
  predicate(UseSSE == 0 && n->getf() == 0.0F);
  match(ConF);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate one
operand immFPR1() %{
  predicate(UseSSE == 0 && n->getf() == 1.0F);
  match(ConF);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate
operand immFPR() %{
  predicate( UseSSE == 0 );
  match(ConF);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate
operand immF() %{
  predicate(UseSSE >= 1);
  match(ConF);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate zero.  Zero and not -0.0
operand immF0() %{
  predicate( UseSSE >= 1 && jint_cast(n->getf()) == 0 );
  match(ConF);

  op_cost(5);
  format %{ %}
  interface(CONST_INTER);
%}

// Immediates for special shifts (sign extend)

// Constants for increment
operand immI_16() %{
  predicate( n->get_int() == 16 );
  match(ConI);

  format %{ %}
  interface(CONST_INTER);
%}

operand immI_24() %{
  predicate( n->get_int() == 24 );
  match(ConI);

  format %{ %}
  interface(CONST_INTER);
%}

// Constant for byte-wide masking
operand immI_255() %{
  predicate( n->get_int() == 255 );
  match(ConI);

  format %{ %}
  interface(CONST_INTER);
%}

// Constant for short-wide masking
operand immI_65535() %{
  predicate(n->get_int() == 65535);
  match(ConI);

  format %{ %}
  interface(CONST_INTER);
%}

// Register Operands
// Integer Register
operand rRegI() %{
  constraint(ALLOC_IN_RC(int_reg));
  match(RegI);
  match(xRegI);
  match(eAXRegI);
  match(eBXRegI);
  match(eCXRegI);
  match(eDXRegI);
  match(eDIRegI);
  match(eSIRegI);

  format %{ %}
  interface(REG_INTER);
%}

// Subset of Integer Register
operand xRegI(rRegI reg) %{
  constraint(ALLOC_IN_RC(int_x_reg));
  match(reg);
  match(eAXRegI);
  match(eBXRegI);
  match(eCXRegI);
  match(eDXRegI);

  format %{ %}
  interface(REG_INTER);
%}

// Special Registers
operand eAXRegI(xRegI reg) %{
  constraint(ALLOC_IN_RC(eax_reg));
  match(reg);
  match(rRegI);

  format %{ "EAX" %}
  interface(REG_INTER);
%}

// Special Registers
operand eBXRegI(xRegI reg) %{
  constraint(ALLOC_IN_RC(ebx_reg));
  match(reg);
  match(rRegI);

  format %{ "EBX" %}
  interface(REG_INTER);
%}

operand eCXRegI(xRegI reg) %{
  constraint(ALLOC_IN_RC(ecx_reg));
  match(reg);
  match(rRegI);

  format %{ "ECX" %}
  interface(REG_INTER);
%}

operand eDXRegI(xRegI reg) %{
  constraint(ALLOC_IN_RC(edx_reg));
  match(reg);
  match(rRegI);

  format %{ "EDX" %}
  interface(REG_INTER);
%}

operand eDIRegI(xRegI reg) %{
  constraint(ALLOC_IN_RC(edi_reg));
  match(reg);
  match(rRegI);

  format %{ "EDI" %}
  interface(REG_INTER);
%}

operand naxRegI() %{
  constraint(ALLOC_IN_RC(nax_reg));
  match(RegI);
  match(eCXRegI);
  match(eDXRegI);
  match(eSIRegI);
  match(eDIRegI);

  format %{ %}
  interface(REG_INTER);
%}

operand nadxRegI() %{
  constraint(ALLOC_IN_RC(nadx_reg));
  match(RegI);
  match(eBXRegI);
  match(eCXRegI);
  match(eSIRegI);
  match(eDIRegI);

  format %{ %}
  interface(REG_INTER);
%}

operand ncxRegI() %{
  constraint(ALLOC_IN_RC(ncx_reg));
  match(RegI);
  match(eAXRegI);
  match(eDXRegI);
  match(eSIRegI);
  match(eDIRegI);

  format %{ %}
  interface(REG_INTER);
%}

// // This operand was used by cmpFastUnlock, but conflicted with 'object' reg
// //
operand eSIRegI(xRegI reg) %{
   constraint(ALLOC_IN_RC(esi_reg));
   match(reg);
   match(rRegI);

   format %{ "ESI" %}
   interface(REG_INTER);
%}

// Pointer Register
operand anyRegP() %{
  constraint(ALLOC_IN_RC(any_reg));
  match(RegP);
  match(eAXRegP);
  match(eBXRegP);
  match(eCXRegP);
  match(eDIRegP);
  match(eRegP);

  format %{ %}
  interface(REG_INTER);
%}

operand eRegP() %{
  constraint(ALLOC_IN_RC(int_reg));
  match(RegP);
  match(eAXRegP);
  match(eBXRegP);
  match(eCXRegP);
  match(eDIRegP);

  format %{ %}
  interface(REG_INTER);
%}

// On windows95, EBP is not safe to use for implicit null tests.
operand eRegP_no_EBP() %{
  constraint(ALLOC_IN_RC(int_reg_no_rbp));
  match(RegP);
  match(eAXRegP);
  match(eBXRegP);
  match(eCXRegP);
  match(eDIRegP);

  op_cost(100);
  format %{ %}
  interface(REG_INTER);
%}

operand naxRegP() %{
  constraint(ALLOC_IN_RC(nax_reg));
  match(RegP);
  match(eBXRegP);
  match(eDXRegP);
  match(eCXRegP);
  match(eSIRegP);
  match(eDIRegP);

  format %{ %}
  interface(REG_INTER);
%}

operand nabxRegP() %{
  constraint(ALLOC_IN_RC(nabx_reg));
  match(RegP);
  match(eCXRegP);
  match(eDXRegP);
  match(eSIRegP);
  match(eDIRegP);

  format %{ %}
  interface(REG_INTER);
%}

operand pRegP() %{
  constraint(ALLOC_IN_RC(p_reg));
  match(RegP);
  match(eBXRegP);
  match(eDXRegP);
  match(eSIRegP);
  match(eDIRegP);

  format %{ %}
  interface(REG_INTER);
%}

// Special Registers
// Return a pointer value
operand eAXRegP(eRegP reg) %{
  constraint(ALLOC_IN_RC(eax_reg));
  match(reg);
  format %{ "EAX" %}
  interface(REG_INTER);
%}

// Used in AtomicAdd
operand eBXRegP(eRegP reg) %{
  constraint(ALLOC_IN_RC(ebx_reg));
  match(reg);
  format %{ "EBX" %}
  interface(REG_INTER);
%}

// Tail-call (interprocedural jump) to interpreter
operand eCXRegP(eRegP reg) %{
  constraint(ALLOC_IN_RC(ecx_reg));
  match(reg);
  format %{ "ECX" %}
  interface(REG_INTER);
%}

operand eSIRegP(eRegP reg) %{
  constraint(ALLOC_IN_RC(esi_reg));
  match(reg);
  format %{ "ESI" %}
  interface(REG_INTER);
%}

// Used in rep stosw
operand eDIRegP(eRegP reg) %{
  constraint(ALLOC_IN_RC(edi_reg));
  match(reg);
  format %{ "EDI" %}
  interface(REG_INTER);
%}

operand eBPRegP() %{
  constraint(ALLOC_IN_RC(ebp_reg));
  match(RegP);
  format %{ "EBP" %}
  interface(REG_INTER);
%}

operand eRegL() %{
  constraint(ALLOC_IN_RC(long_reg));
  match(RegL);
  match(eADXRegL);

  format %{ %}
  interface(REG_INTER);
%}

operand eADXRegL( eRegL reg ) %{
  constraint(ALLOC_IN_RC(eadx_reg));
  match(reg);

  format %{ "EDX:EAX" %}
  interface(REG_INTER);
%}

operand eBCXRegL( eRegL reg ) %{
  constraint(ALLOC_IN_RC(ebcx_reg));
  match(reg);

  format %{ "EBX:ECX" %}
  interface(REG_INTER);
%}

// Special case for integer high multiply
operand eADXRegL_low_only() %{
  constraint(ALLOC_IN_RC(eadx_reg));
  match(RegL);

  format %{ "EAX" %}
  interface(REG_INTER);
%}

// Flags register, used as output of compare instructions
operand eFlagsReg() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);

  format %{ "EFLAGS" %}
  interface(REG_INTER);
%}

// Flags register, used as output of FLOATING POINT compare instructions
operand eFlagsRegU() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);

  format %{ "EFLAGS_U" %}
  interface(REG_INTER);
%}

operand eFlagsRegUCF() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  predicate(false);

  format %{ "EFLAGS_U_CF" %}
  interface(REG_INTER);
%}

// Condition Code Register used by long compare
operand flagsReg_long_LTGE() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  format %{ "FLAGS_LTGE" %}
  interface(REG_INTER);
%}
operand flagsReg_long_EQNE() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  format %{ "FLAGS_EQNE" %}
  interface(REG_INTER);
%}
operand flagsReg_long_LEGT() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  format %{ "FLAGS_LEGT" %}
  interface(REG_INTER);
%}

// Float register operands
operand regDPR() %{
  predicate( UseSSE < 2 );
  constraint(ALLOC_IN_RC(fp_dbl_reg));
  match(RegD);
  match(regDPR1);
  match(regDPR2);
  format %{ %}
  interface(REG_INTER);
%}

operand regDPR1(regDPR reg) %{
  predicate( UseSSE < 2 );
  constraint(ALLOC_IN_RC(fp_dbl_reg0));
  match(reg);
  format %{ "FPR1" %}
  interface(REG_INTER);
%}

operand regDPR2(regDPR reg) %{
  predicate( UseSSE < 2 );
  constraint(ALLOC_IN_RC(fp_dbl_reg1));
  match(reg);
  format %{ "FPR2" %}
  interface(REG_INTER);
%}

operand regnotDPR1(regDPR reg) %{
  predicate( UseSSE < 2 );
  constraint(ALLOC_IN_RC(fp_dbl_notreg0));
  match(reg);
  format %{ %}
  interface(REG_INTER);
%}

// Float register operands
operand regFPR() %{
  predicate( UseSSE < 2 );
  constraint(ALLOC_IN_RC(fp_flt_reg));
  match(RegF);
  match(regFPR1);
  format %{ %}
  interface(REG_INTER);
%}

// Float register operands
operand regFPR1(regFPR reg) %{
  predicate( UseSSE < 2 );
  constraint(ALLOC_IN_RC(fp_flt_reg0));
  match(reg);
  format %{ "FPR1" %}
  interface(REG_INTER);
%}

// XMM Float register operands
operand regF() %{
  predicate( UseSSE>=1 );
  constraint(ALLOC_IN_RC(float_reg));
  match(RegF);
  format %{ %}
  interface(REG_INTER);
%}

// XMM Double register operands
operand regD() %{
  predicate( UseSSE>=2 );
  constraint(ALLOC_IN_RC(double_reg));
  match(RegD);
  format %{ %}
  interface(REG_INTER);
%}


//----------Memory Operands----------------------------------------------------
// Direct Memory Operand
operand direct(immP addr) %{
  match(addr);

  format %{ "[$addr]" %}
  interface(MEMORY_INTER) %{
    base(0xFFFFFFFF);
    index(0x4);
    scale(0x0);
    disp($addr);
  %}
%}

// Indirect Memory Operand
operand indirect(eRegP reg) %{
  constraint(ALLOC_IN_RC(int_reg));
  match(reg);

  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp(0x0);
  %}
%}

// Indirect Memory Plus Short Offset Operand
operand indOffset8(eRegP reg, immI8 off) %{
  match(AddP reg off);

  format %{ "[$reg + $off]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Plus Long Offset Operand
operand indOffset32(eRegP reg, immI off) %{
  match(AddP reg off);

  format %{ "[$reg + $off]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Plus Long Offset Operand
operand indOffset32X(rRegI reg, immP off) %{
  match(AddP off reg);

  format %{ "[$reg + $off]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Plus Index Register Plus Offset Operand
operand indIndexOffset(eRegP reg, rRegI ireg, immI off) %{
  match(AddP (AddP reg ireg) off);

  op_cost(10);
  format %{"[$reg + $off + $ireg]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Plus Index Register Plus Offset Operand
operand indIndex(eRegP reg, rRegI ireg) %{
  match(AddP reg ireg);

  op_cost(10);
  format %{"[$reg + $ireg]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale(0x0);
    disp(0x0);
  %}
%}

// // -------------------------------------------------------------------------
// // 486 architecture doesn't support "scale * index + offset" with out a base
// // -------------------------------------------------------------------------
// // Scaled Memory Operands
// // Indirect Memory Times Scale Plus Offset Operand
// operand indScaleOffset(immP off, rRegI ireg, immI2 scale) %{
//   match(AddP off (LShiftI ireg scale));
//
//   op_cost(10);
//   format %{"[$off + $ireg << $scale]" %}
//   interface(MEMORY_INTER) %{
//     base(0x4);
//     index($ireg);
//     scale($scale);
//     disp($off);
//   %}
// %}

// Indirect Memory Times Scale Plus Index Register
operand indIndexScale(eRegP reg, rRegI ireg, immI2 scale) %{
  match(AddP reg (LShiftI ireg scale));

  op_cost(10);
  format %{"[$reg + $ireg << $scale]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale($scale);
    disp(0x0);
  %}
%}

// Indirect Memory Times Scale Plus Index Register Plus Offset Operand
operand indIndexScaleOffset(eRegP reg, immI off, rRegI ireg, immI2 scale) %{
  match(AddP (AddP reg (LShiftI ireg scale)) off);

  op_cost(10);
  format %{"[$reg + $off + $ireg << $scale]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale($scale);
    disp($off);
  %}
%}

//----------Load Long Memory Operands------------------------------------------
// The load-long idiom will use it's address expression again after loading
// the first word of the long.  If the load-long destination overlaps with
// registers used in the addressing expression, the 2nd half will be loaded
// from a clobbered address.  Fix this by requiring that load-long use
// address registers that do not overlap with the load-long target.

// load-long support
operand load_long_RegP() %{
  constraint(ALLOC_IN_RC(esi_reg));
  match(RegP);
  match(eSIRegP);
  op_cost(100);
  format %{  %}
  interface(REG_INTER);
%}

// Indirect Memory Operand Long
operand load_long_indirect(load_long_RegP reg) %{
  constraint(ALLOC_IN_RC(esi_reg));
  match(reg);

  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp(0x0);
  %}
%}

// Indirect Memory Plus Long Offset Operand
operand load_long_indOffset32(load_long_RegP reg, immI off) %{
  match(AddP reg off);

  format %{ "[$reg + $off]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp($off);
  %}
%}

opclass load_long_memory(load_long_indirect, load_long_indOffset32);


//----------Special Memory Operands--------------------------------------------
// Stack Slot Operand - This operand is used for loading and storing temporary
//                      values on the stack where a match requires a value to
//                      flow through memory.
operand stackSlotP(sRegP reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  // No match rule because this operand is only generated in matching
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0x4);   // ESP
    index(0x4);  // No Index
    scale(0x0);  // No Scale
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotI(sRegI reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  // No match rule because this operand is only generated in matching
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0x4);   // ESP
    index(0x4);  // No Index
    scale(0x0);  // No Scale
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotF(sRegF reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  // No match rule because this operand is only generated in matching
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0x4);   // ESP
    index(0x4);  // No Index
    scale(0x0);  // No Scale
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotD(sRegD reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  // No match rule because this operand is only generated in matching
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0x4);   // ESP
    index(0x4);  // No Index
    scale(0x0);  // No Scale
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotL(sRegL reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  // No match rule because this operand is only generated in matching
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0x4);   // ESP
    index(0x4);  // No Index
    scale(0x0);  // No Scale
    disp($reg);  // Stack Offset
  %}
%}

//----------Memory Operands - Win95 Implicit Null Variants----------------
// Indirect Memory Operand
operand indirect_win95_safe(eRegP_no_EBP reg)
%{
  constraint(ALLOC_IN_RC(int_reg));
  match(reg);

  op_cost(100);
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp(0x0);
  %}
%}

// Indirect Memory Plus Short Offset Operand
operand indOffset8_win95_safe(eRegP_no_EBP reg, immI8 off)
%{
  match(AddP reg off);

  op_cost(100);
  format %{ "[$reg + $off]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Plus Long Offset Operand
operand indOffset32_win95_safe(eRegP_no_EBP reg, immI off)
%{
  match(AddP reg off);

  op_cost(100);
  format %{ "[$reg + $off]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index(0x4);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Plus Index Register Plus Offset Operand
operand indIndexOffset_win95_safe(eRegP_no_EBP reg, rRegI ireg, immI off)
%{
  match(AddP (AddP reg ireg) off);

  op_cost(100);
  format %{"[$reg + $off + $ireg]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale(0x0);
    disp($off);
  %}
%}

// Indirect Memory Times Scale Plus Index Register
operand indIndexScale_win95_safe(eRegP_no_EBP reg, rRegI ireg, immI2 scale)
%{
  match(AddP reg (LShiftI ireg scale));

  op_cost(100);
  format %{"[$reg + $ireg << $scale]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale($scale);
    disp(0x0);
  %}
%}

// Indirect Memory Times Scale Plus Index Register Plus Offset Operand
operand indIndexScaleOffset_win95_safe(eRegP_no_EBP reg, immI off, rRegI ireg, immI2 scale)
%{
  match(AddP (AddP reg (LShiftI ireg scale)) off);

  op_cost(100);
  format %{"[$reg + $off + $ireg << $scale]" %}
  interface(MEMORY_INTER) %{
    base($reg);
    index($ireg);
    scale($scale);
    disp($off);
  %}
%}

//----------Conditional Branch Operands----------------------------------------
// Comparison Op  - This is the operation of the comparison, and is limited to
//                  the following set of codes:
//                  L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
//
// Other attributes of the comparison, such as unsignedness, are specified
// by the comparison instruction that sets a condition code flags register.
// That result is represented by a flags operand whose subtype is appropriate
// to the unsignedness (etc.) of the comparison.
//
// Later, the instruction which matches both the Comparison Op (a Bool) and
// the flags (produced by the Cmp) specifies the coding of the comparison op
// by matching a specific subtype of Bool operand below, such as cmpOpU.

// Comparision Code
operand cmpOp() %{
  match(Bool);

  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x4, "e");
    not_equal(0x5, "ne");
    less(0xC, "l");
    greater_equal(0xD, "ge");
    less_equal(0xE, "le");
    greater(0xF, "g");
    overflow(0x0, "o");
    no_overflow(0x1, "no");
  %}
%}

// Comparison Code, unsigned compare.  Used by FP also, with
// C2 (unordered) turned into GT or LT already.  The other bits
// C0 and C3 are turned into Carry & Zero flags.
operand cmpOpU() %{
  match(Bool);

  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x4, "e");
    not_equal(0x5, "ne");
    less(0x2, "b");
    greater_equal(0x3, "nb");
    less_equal(0x6, "be");
    greater(0x7, "nbe");
    overflow(0x0, "o");
    no_overflow(0x1, "no");
  %}
%}

// Floating comparisons that don't require any fixup for the unordered case
operand cmpOpUCF() %{
  match(Bool);
  predicate(n->as_Bool()->_test._test == BoolTest::lt ||
            n->as_Bool()->_test._test == BoolTest::ge ||
            n->as_Bool()->_test._test == BoolTest::le ||
            n->as_Bool()->_test._test == BoolTest::gt);
  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x4, "e");
    not_equal(0x5, "ne");
    less(0x2, "b");
    greater_equal(0x3, "nb");
    less_equal(0x6, "be");
    greater(0x7, "nbe");
    overflow(0x0, "o");
    no_overflow(0x1, "no");
  %}
%}


// Floating comparisons that can be fixed up with extra conditional jumps
operand cmpOpUCF2() %{
  match(Bool);
  predicate(n->as_Bool()->_test._test == BoolTest::ne ||
            n->as_Bool()->_test._test == BoolTest::eq);
  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x4, "e");
    not_equal(0x5, "ne");
    less(0x2, "b");
    greater_equal(0x3, "nb");
    less_equal(0x6, "be");
    greater(0x7, "nbe");
    overflow(0x0, "o");
    no_overflow(0x1, "no");
  %}
%}

// Comparison Code for FP conditional move
operand cmpOp_fcmov() %{
  match(Bool);

  predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
            n->as_Bool()->_test._test != BoolTest::no_overflow);
  format %{ "" %}
  interface(COND_INTER) %{
    equal        (0x0C8);
    not_equal    (0x1C8);
    less         (0x0C0);
    greater_equal(0x1C0);
    less_equal   (0x0D0);
    greater      (0x1D0);
    overflow(0x0, "o"); // not really supported by the instruction
    no_overflow(0x1, "no"); // not really supported by the instruction
  %}
%}

// Comparision Code used in long compares
operand cmpOp_commute() %{
  match(Bool);

  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x4, "e");
    not_equal(0x5, "ne");
    less(0xF, "g");
    greater_equal(0xE, "le");
    less_equal(0xD, "ge");
    greater(0xC, "l");
    overflow(0x0, "o");
    no_overflow(0x1, "no");
  %}
%}

//----------OPERAND CLASSES----------------------------------------------------
// Operand Classes are groups of operands that are used as to simplify
// instruction definitions by not requiring the AD writer to specify separate
// instructions for every form of operand when the instruction accepts
// multiple operand types with the same basic encoding and format.  The classic
// case of this is memory operands.

opclass memory(direct, indirect, indOffset8, indOffset32, indOffset32X, indIndexOffset,
               indIndex, indIndexScale, indIndexScaleOffset);

// Long memory operations are encoded in 2 instructions and a +4 offset.
// This means some kind of offset is always required and you cannot use
// an oop as the offset (done when working on static globals).
opclass long_memory(direct, indirect, indOffset8, indOffset32, indIndexOffset,
                    indIndex, indIndexScale, indIndexScaleOffset);


//----------PIPELINE-----------------------------------------------------------
// Rules which define the behavior of the target architectures pipeline.
pipeline %{

//----------ATTRIBUTES---------------------------------------------------------
attributes %{
  variable_size_instructions;        // Fixed size instructions
  max_instructions_per_bundle = 3;   // Up to 3 instructions per bundle
  instruction_unit_size = 1;         // An instruction is 1 bytes long
  instruction_fetch_unit_size = 16;  // The processor fetches one line
  instruction_fetch_units = 1;       // of 16 bytes

  // List of nop instructions
  nops( MachNop );
%}

//----------RESOURCES----------------------------------------------------------
// Resources are the functional units available to the machine

// Generic P2/P3 pipeline
// 3 decoders, only D0 handles big operands; a "bundle" is the limit of
// 3 instructions decoded per cycle.
// 2 load/store ops per cycle, 1 branch, 1 FPU,
// 2 ALU op, only ALU0 handles mul/div instructions.
resources( D0, D1, D2, DECODE = D0 | D1 | D2,
           MS0, MS1, MEM = MS0 | MS1,
           BR, FPU,
           ALU0, ALU1, ALU = ALU0 | ALU1 );

//----------PIPELINE DESCRIPTION-----------------------------------------------
// Pipeline Description specifies the stages in the machine's pipeline

// Generic P2/P3 pipeline
pipe_desc(S0, S1, S2, S3, S4, S5);

//----------PIPELINE CLASSES---------------------------------------------------
// Pipeline Classes describe the stages in which input and output are
// referenced by the hardware pipeline.

// Naming convention: ialu or fpu
// Then: _reg
// Then: _reg if there is a 2nd register
// Then: _long if it's a pair of instructions implementing a long
// Then: _fat if it requires the big decoder
//   Or: _mem if it requires the big decoder and a memory unit.

// Integer ALU reg operation
pipe_class ialu_reg(rRegI dst) %{
    single_instruction;
    dst    : S4(write);
    dst    : S3(read);
    DECODE : S0;        // any decoder
    ALU    : S3;        // any alu
%}

// Long ALU reg operation
pipe_class ialu_reg_long(eRegL dst) %{
    instruction_count(2);
    dst    : S4(write);
    dst    : S3(read);
    DECODE : S0(2);     // any 2 decoders
    ALU    : S3(2);     // both alus
%}

// Integer ALU reg operation using big decoder
pipe_class ialu_reg_fat(rRegI dst) %{
    single_instruction;
    dst    : S4(write);
    dst    : S3(read);
    D0     : S0;        // big decoder only
    ALU    : S3;        // any alu
%}

// Long ALU reg operation using big decoder
pipe_class ialu_reg_long_fat(eRegL dst) %{
    instruction_count(2);
    dst    : S4(write);
    dst    : S3(read);
    D0     : S0(2);     // big decoder only; twice
    ALU    : S3(2);     // any 2 alus
%}

// Integer ALU reg-reg operation
pipe_class ialu_reg_reg(rRegI dst, rRegI src) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    DECODE : S0;        // any decoder
    ALU    : S3;        // any alu
%}

// Long ALU reg-reg operation
pipe_class ialu_reg_reg_long(eRegL dst, eRegL src) %{
    instruction_count(2);
    dst    : S4(write);
    src    : S3(read);
    DECODE : S0(2);     // any 2 decoders
    ALU    : S3(2);     // both alus
%}

// Integer ALU reg-reg operation
pipe_class ialu_reg_reg_fat(rRegI dst, memory src) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    D0     : S0;        // big decoder only
    ALU    : S3;        // any alu
%}

// Long ALU reg-reg operation
pipe_class ialu_reg_reg_long_fat(eRegL dst, eRegL src) %{
    instruction_count(2);
    dst    : S4(write);
    src    : S3(read);
    D0     : S0(2);     // big decoder only; twice
    ALU    : S3(2);     // both alus
%}

// Integer ALU reg-mem operation
pipe_class ialu_reg_mem(rRegI dst, memory mem) %{
    single_instruction;
    dst    : S5(write);
    mem    : S3(read);
    D0     : S0;        // big decoder only
    ALU    : S4;        // any alu
    MEM    : S3;        // any mem
%}

// Long ALU reg-mem operation
pipe_class ialu_reg_long_mem(eRegL dst, load_long_memory mem) %{
    instruction_count(2);
    dst    : S5(write);
    mem    : S3(read);
    D0     : S0(2);     // big decoder only; twice
    ALU    : S4(2);     // any 2 alus
    MEM    : S3(2);     // both mems
%}

// Integer mem operation (prefetch)
pipe_class ialu_mem(memory mem)
%{
    single_instruction;
    mem    : S3(read);
    D0     : S0;        // big decoder only
    MEM    : S3;        // any mem
%}

// Integer Store to Memory
pipe_class ialu_mem_reg(memory mem, rRegI src) %{
    single_instruction;
    mem    : S3(read);
    src    : S5(read);
    D0     : S0;        // big decoder only
    ALU    : S4;        // any alu
    MEM    : S3;
%}

// Long Store to Memory
pipe_class ialu_mem_long_reg(memory mem, eRegL src) %{
    instruction_count(2);
    mem    : S3(read);
    src    : S5(read);
    D0     : S0(2);     // big decoder only; twice
    ALU    : S4(2);     // any 2 alus
    MEM    : S3(2);     // Both mems
%}

// Integer Store to Memory
pipe_class ialu_mem_imm(memory mem) %{
    single_instruction;
    mem    : S3(read);
    D0     : S0;        // big decoder only
    ALU    : S4;        // any alu
    MEM    : S3;
%}

// Integer ALU0 reg-reg operation
pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    D0     : S0;        // Big decoder only
    ALU0   : S3;        // only alu0
%}

// Integer ALU0 reg-mem operation
pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem) %{
    single_instruction;
    dst    : S5(write);
    mem    : S3(read);
    D0     : S0;        // big decoder only
    ALU0   : S4;        // ALU0 only
    MEM    : S3;        // any mem
%}

// Integer ALU reg-reg operation
pipe_class ialu_cr_reg_reg(eFlagsReg cr, rRegI src1, rRegI src2) %{
    single_instruction;
    cr     : S4(write);
    src1   : S3(read);
    src2   : S3(read);
    DECODE : S0;        // any decoder
    ALU    : S3;        // any alu
%}

// Integer ALU reg-imm operation
pipe_class ialu_cr_reg_imm(eFlagsReg cr, rRegI src1) %{
    single_instruction;
    cr     : S4(write);
    src1   : S3(read);
    DECODE : S0;        // any decoder
    ALU    : S3;        // any alu
%}

// Integer ALU reg-mem operation
pipe_class ialu_cr_reg_mem(eFlagsReg cr, rRegI src1, memory src2) %{
    single_instruction;
    cr     : S4(write);
    src1   : S3(read);
    src2   : S3(read);
    D0     : S0;        // big decoder only
    ALU    : S4;        // any alu
    MEM    : S3;
%}

// Conditional move reg-reg
pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y ) %{
    instruction_count(4);
    y      : S4(read);
    q      : S3(read);
    p      : S3(read);
    DECODE : S0(4);     // any decoder
%}

// Conditional move reg-reg
pipe_class pipe_cmov_reg( rRegI dst, rRegI src, eFlagsReg cr ) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    cr     : S3(read);
    DECODE : S0;        // any decoder
%}

// Conditional move reg-mem
pipe_class pipe_cmov_mem( eFlagsReg cr, rRegI dst, memory src) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    cr     : S3(read);
    DECODE : S0;        // any decoder
    MEM    : S3;
%}

// Conditional move reg-reg long
pipe_class pipe_cmov_reg_long( eFlagsReg cr, eRegL dst, eRegL src) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    cr     : S3(read);
    DECODE : S0(2);     // any 2 decoders
%}

// Conditional move double reg-reg
pipe_class pipe_cmovDPR_reg( eFlagsReg cr, regDPR1 dst, regDPR src) %{
    single_instruction;
    dst    : S4(write);
    src    : S3(read);
    cr     : S3(read);
    DECODE : S0;        // any decoder
%}

// Float reg-reg operation
pipe_class fpu_reg(regDPR dst) %{
    instruction_count(2);
    dst    : S3(read);
    DECODE : S0(2);     // any 2 decoders
    FPU    : S3;
%}

// Float reg-reg operation
pipe_class fpu_reg_reg(regDPR dst, regDPR src) %{
    instruction_count(2);
    dst    : S4(write);
    src    : S3(read);
    DECODE : S0(2);     // any 2 decoders
    FPU    : S3;
%}

// Float reg-reg operation
pipe_class fpu_reg_reg_reg(regDPR dst, regDPR src1, regDPR src2) %{
    instruction_count(3);
    dst    : S4(write);
    src1   : S3(read);
    src2   : S3(read);
    DECODE : S0(3);     // any 3 decoders
    FPU    : S3(2);
%}

// Float reg-reg operation
pipe_class fpu_reg_reg_reg_reg(regDPR dst, regDPR src1, regDPR src2, regDPR src3) %{
    instruction_count(4);
    dst    : S4(write);
    src1   : S3(read);
    src2   : S3(read);
    src3   : S3(read);
    DECODE : S0(4);     // any 3 decoders
    FPU    : S3(2);
%}

// Float reg-reg operation
pipe_class fpu_reg_mem_reg_reg(regDPR dst, memory src1, regDPR src2, regDPR src3) %{
    instruction_count(4);
    dst    : S4(write);
    src1   : S3(read);
    src2   : S3(read);
    src3   : S3(read);
    DECODE : S1(3);     // any 3 decoders
    D0     : S0;        // Big decoder only
    FPU    : S3(2);
    MEM    : S3;
%}

// Float reg-mem operation
pipe_class fpu_reg_mem(regDPR dst, memory mem) %{
    instruction_count(2);
    dst    : S5(write);
    mem    : S3(read);
    D0     : S0;        // big decoder only
    DECODE : S1;        // any decoder for FPU POP
    FPU    : S4;
    MEM    : S3;        // any mem
%}

// Float reg-mem operation
pipe_class fpu_reg_reg_mem(regDPR dst, regDPR src1, memory mem) %{
    instruction_count(3);
    dst    : S5(write);
    src1   : S3(read);
    mem    : S3(read);
    D0     : S0;        // big decoder only
    DECODE : S1(2);     // any decoder for FPU POP
    FPU    : S4;
    MEM    : S3;        // any mem
%}

// Float mem-reg operation
pipe_class fpu_mem_reg(memory mem, regDPR src) %{
    instruction_count(2);
    src    : S5(read);
    mem    : S3(read);
    DECODE : S0;        // any decoder for FPU PUSH
    D0     : S1;        // big decoder only
    FPU    : S4;
    MEM    : S3;        // any mem
%}

pipe_class fpu_mem_reg_reg(memory mem, regDPR src1, regDPR src2) %{
    instruction_count(3);
    src1   : S3(read);
    src2   : S3(read);
    mem    : S3(read);
    DECODE : S0(2);     // any decoder for FPU PUSH
    D0     : S1;        // big decoder only
    FPU    : S4;
    MEM    : S3;        // any mem
%}

pipe_class fpu_mem_reg_mem(memory mem, regDPR src1, memory src2) %{
    instruction_count(3);
    src1   : S3(read);
    src2   : S3(read);
    mem    : S4(read);
    DECODE : S0;        // any decoder for FPU PUSH
    D0     : S0(2);     // big decoder only
    FPU    : S4;
    MEM    : S3(2);     // any mem
%}

pipe_class fpu_mem_mem(memory dst, memory src1) %{
    instruction_count(2);
    src1   : S3(read);
    dst    : S4(read);
    D0     : S0(2);     // big decoder only
    MEM    : S3(2);     // any mem
%}

pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2) %{
    instruction_count(3);
    src1   : S3(read);
    src2   : S3(read);
    dst    : S4(read);
    D0     : S0(3);     // big decoder only
    FPU    : S4;
    MEM    : S3(3);     // any mem
%}

pipe_class fpu_mem_reg_con(memory mem, regDPR src1) %{
    instruction_count(3);
    src1   : S4(read);
    mem    : S4(read);
    DECODE : S0;        // any decoder for FPU PUSH
    D0     : S0(2);     // big decoder only
    FPU    : S4;
    MEM    : S3(2);     // any mem
%}

// Float load constant
pipe_class fpu_reg_con(regDPR dst) %{
    instruction_count(2);
    dst    : S5(write);
    D0     : S0;        // big decoder only for the load
    DECODE : S1;        // any decoder for FPU POP
    FPU    : S4;
    MEM    : S3;        // any mem
%}

// Float load constant
pipe_class fpu_reg_reg_con(regDPR dst, regDPR src) %{
    instruction_count(3);
    dst    : S5(write);
    src    : S3(read);
    D0     : S0;        // big decoder only for the load
    DECODE : S1(2);     // any decoder for FPU POP
    FPU    : S4;
    MEM    : S3;        // any mem
%}

// UnConditional branch
pipe_class pipe_jmp( label labl ) %{
    single_instruction;
    BR   : S3;
%}

// Conditional branch
pipe_class pipe_jcc( cmpOp cmp, eFlagsReg cr, label labl ) %{
    single_instruction;
    cr    : S1(read);
    BR    : S3;
%}

// Allocation idiom
pipe_class pipe_cmpxchg( eRegP dst, eRegP heap_ptr ) %{
    instruction_count(1); force_serialization;
    fixed_latency(6);
    heap_ptr : S3(read);
    DECODE   : S0(3);
    D0       : S2;
    MEM      : S3;
    ALU      : S3(2);
    dst      : S5(write);
    BR       : S5;
%}

// Generic big/slow expanded idiom
pipe_class pipe_slow(  ) %{
    instruction_count(10); multiple_bundles; force_serialization;
    fixed_latency(100);
    D0  : S0(2);
    MEM : S3(2);
%}

// The real do-nothing guy
pipe_class empty( ) %{
    instruction_count(0);
%}

// Define the class for the Nop node
define %{
   MachNop = empty;
%}

%}

//----------INSTRUCTIONS-------------------------------------------------------
//
// match      -- States which machine-independent subtree may be replaced
//               by this instruction.
// ins_cost   -- The estimated cost of this instruction is used by instruction
//               selection to identify a minimum cost tree of machine
//               instructions that matches a tree of machine-independent
//               instructions.
// format     -- A string providing the disassembly for this instruction.
//               The value of an instruction's operand may be inserted
//               by referring to it with a '$' prefix.
// opcode     -- Three instruction opcodes may be provided.  These are referred
//               to within an encode class as $primary, $secondary, and $tertiary
//               respectively.  The primary opcode is commonly used to
//               indicate the type of machine instruction, while secondary
//               and tertiary are often used for prefix options or addressing
//               modes.
// ins_encode -- A list of encode classes with parameters. The encode class
//               name must have been defined in an 'enc_class' specification
//               in the encode section of the architecture description.

//----------BSWAP-Instruction--------------------------------------------------
instruct bytes_reverse_int(rRegI dst) %{
  match(Set dst (ReverseBytesI dst));

  format %{ "BSWAP  $dst" %}
  opcode(0x0F, 0xC8);
  ins_encode( OpcP, OpcSReg(dst) );
  ins_pipe( ialu_reg );
%}

instruct bytes_reverse_long(eRegL dst) %{
  match(Set dst (ReverseBytesL dst));

  format %{ "BSWAP  $dst.lo\n\t"
            "BSWAP  $dst.hi\n\t"
            "XCHG   $dst.lo $dst.hi" %}

  ins_cost(125);
  ins_encode( bswap_long_bytes(dst) );
  ins_pipe( ialu_reg_reg);
%}

instruct bytes_reverse_unsigned_short(rRegI dst, eFlagsReg cr) %{
  match(Set dst (ReverseBytesUS dst));
  effect(KILL cr);

  format %{ "BSWAP  $dst\n\t" 
            "SHR    $dst,16\n\t" %}
  ins_encode %{
    __ bswapl($dst$$Register);
    __ shrl($dst$$Register, 16); 
  %}
  ins_pipe( ialu_reg );
%}

instruct bytes_reverse_short(rRegI dst, eFlagsReg cr) %{
  match(Set dst (ReverseBytesS dst));
  effect(KILL cr);

  format %{ "BSWAP  $dst\n\t" 
            "SAR    $dst,16\n\t" %}
  ins_encode %{
    __ bswapl($dst$$Register);
    __ sarl($dst$$Register, 16); 
  %}
  ins_pipe( ialu_reg );
%}


//---------- Zeros Count Instructions ------------------------------------------

instruct countLeadingZerosI(rRegI dst, rRegI src, eFlagsReg cr) %{
  predicate(UseCountLeadingZerosInstruction);
  match(Set dst (CountLeadingZerosI src));
  effect(KILL cr);

  format %{ "LZCNT  $dst, $src\t# count leading zeros (int)" %}
  ins_encode %{
    __ lzcntl($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg);
%}

instruct countLeadingZerosI_bsr(rRegI dst, rRegI src, eFlagsReg cr) %{
  predicate(!UseCountLeadingZerosInstruction);
  match(Set dst (CountLeadingZerosI src));
  effect(KILL cr);

  format %{ "BSR    $dst, $src\t# count leading zeros (int)\n\t"
            "JNZ    skip\n\t"
            "MOV    $dst, -1\n"
      "skip:\n\t"
            "NEG    $dst\n\t"
            "ADD    $dst, 31" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Label skip;
    __ bsrl(Rdst, Rsrc);
    __ jccb(Assembler::notZero, skip);
    __ movl(Rdst, -1);
    __ bind(skip);
    __ negl(Rdst);
    __ addl(Rdst, BitsPerInt - 1);
  %}
  ins_pipe(ialu_reg);
%}

instruct countLeadingZerosL(rRegI dst, eRegL src, eFlagsReg cr) %{
  predicate(UseCountLeadingZerosInstruction);
  match(Set dst (CountLeadingZerosL src));
  effect(TEMP dst, KILL cr);

  format %{ "LZCNT  $dst, $src.hi\t# count leading zeros (long)\n\t"
            "JNC    done\n\t"
            "LZCNT  $dst, $src.lo\n\t"
            "ADD    $dst, 32\n"
      "done:" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Label done;
    __ lzcntl(Rdst, HIGH_FROM_LOW(Rsrc));
    __ jccb(Assembler::carryClear, done);
    __ lzcntl(Rdst, Rsrc);
    __ addl(Rdst, BitsPerInt);
    __ bind(done);
  %}
  ins_pipe(ialu_reg);
%}

instruct countLeadingZerosL_bsr(rRegI dst, eRegL src, eFlagsReg cr) %{
  predicate(!UseCountLeadingZerosInstruction);
  match(Set dst (CountLeadingZerosL src));
  effect(TEMP dst, KILL cr);

  format %{ "BSR    $dst, $src.hi\t# count leading zeros (long)\n\t"
            "JZ     msw_is_zero\n\t"
            "ADD    $dst, 32\n\t"
            "JMP    not_zero\n"
      "msw_is_zero:\n\t"
            "BSR    $dst, $src.lo\n\t"
            "JNZ    not_zero\n\t"
            "MOV    $dst, -1\n"
      "not_zero:\n\t"
            "NEG    $dst\n\t"
            "ADD    $dst, 63\n" %}
 ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Label msw_is_zero;
    Label not_zero;
    __ bsrl(Rdst, HIGH_FROM_LOW(Rsrc));
    __ jccb(Assembler::zero, msw_is_zero);
    __ addl(Rdst, BitsPerInt);
    __ jmpb(not_zero);
    __ bind(msw_is_zero);
    __ bsrl(Rdst, Rsrc);
    __ jccb(Assembler::notZero, not_zero);
    __ movl(Rdst, -1);
    __ bind(not_zero);
    __ negl(Rdst);
    __ addl(Rdst, BitsPerLong - 1);
  %}
  ins_pipe(ialu_reg);
%}

instruct countTrailingZerosI(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (CountTrailingZerosI src));
  effect(KILL cr);

  format %{ "BSF    $dst, $src\t# count trailing zeros (int)\n\t"
            "JNZ    done\n\t"
            "MOV    $dst, 32\n"
      "done:" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Label done;
    __ bsfl(Rdst, $src$$Register);
    __ jccb(Assembler::notZero, done);
    __ movl(Rdst, BitsPerInt);
    __ bind(done);
  %}
  ins_pipe(ialu_reg);
%}

instruct countTrailingZerosL(rRegI dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (CountTrailingZerosL src));
  effect(TEMP dst, KILL cr);

  format %{ "BSF    $dst, $src.lo\t# count trailing zeros (long)\n\t"
            "JNZ    done\n\t"
            "BSF    $dst, $src.hi\n\t"
            "JNZ    msw_not_zero\n\t"
            "MOV    $dst, 32\n"
      "msw_not_zero:\n\t"
            "ADD    $dst, 32\n"
      "done:" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Label msw_not_zero;
    Label done;
    __ bsfl(Rdst, Rsrc);
    __ jccb(Assembler::notZero, done);
    __ bsfl(Rdst, HIGH_FROM_LOW(Rsrc));
    __ jccb(Assembler::notZero, msw_not_zero);
    __ movl(Rdst, BitsPerInt);
    __ bind(msw_not_zero);
    __ addl(Rdst, BitsPerInt);
    __ bind(done);
  %}
  ins_pipe(ialu_reg);
%}


//---------- Population Count Instructions -------------------------------------

instruct popCountI(rRegI dst, rRegI src, eFlagsReg cr) %{
  predicate(UsePopCountInstruction);
  match(Set dst (PopCountI src));
  effect(KILL cr);

  format %{ "POPCNT $dst, $src" %}
  ins_encode %{
    __ popcntl($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg);
%}

instruct popCountI_mem(rRegI dst, memory mem, eFlagsReg cr) %{
  predicate(UsePopCountInstruction);
  match(Set dst (PopCountI (LoadI mem)));
  effect(KILL cr);

  format %{ "POPCNT $dst, $mem" %}
  ins_encode %{
    __ popcntl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg);
%}

// Note: Long.bitCount(long) returns an int.
instruct popCountL(rRegI dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
  predicate(UsePopCountInstruction);
  match(Set dst (PopCountL src));
  effect(KILL cr, TEMP tmp, TEMP dst);

  format %{ "POPCNT $dst, $src.lo\n\t"
            "POPCNT $tmp, $src.hi\n\t"
            "ADD    $dst, $tmp" %}
  ins_encode %{
    __ popcntl($dst$$Register, $src$$Register);
    __ popcntl($tmp$$Register, HIGH_FROM_LOW($src$$Register));
    __ addl($dst$$Register, $tmp$$Register);
  %}
  ins_pipe(ialu_reg);
%}

// Note: Long.bitCount(long) returns an int.
instruct popCountL_mem(rRegI dst, memory mem, rRegI tmp, eFlagsReg cr) %{
  predicate(UsePopCountInstruction);
  match(Set dst (PopCountL (LoadL mem)));
  effect(KILL cr, TEMP tmp, TEMP dst);

  format %{ "POPCNT $dst, $mem\n\t"
            "POPCNT $tmp, $mem+4\n\t"
            "ADD    $dst, $tmp" %}
  ins_encode %{
    //__ popcntl($dst$$Register, $mem$$Address$$first);
    //__ popcntl($tmp$$Register, $mem$$Address$$second);
    __ popcntl($dst$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, relocInfo::none));
    __ popcntl($tmp$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, relocInfo::none));
    __ addl($dst$$Register, $tmp$$Register);
  %}
  ins_pipe(ialu_reg);
%}


//----------Load/Store/Move Instructions---------------------------------------
//----------Load Instructions--------------------------------------------------
// Load Byte (8bit signed)
instruct loadB(xRegI dst, memory mem) %{
  match(Set dst (LoadB mem));

  ins_cost(125);
  format %{ "MOVSX8 $dst,$mem\t# byte" %}

  ins_encode %{
    __ movsbl($dst$$Register, $mem$$Address);
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Byte (8bit signed) into Long Register
instruct loadB2L(eRegL dst, memory mem, eFlagsReg cr) %{
  match(Set dst (ConvI2L (LoadB mem)));
  effect(KILL cr);

  ins_cost(375);
  format %{ "MOVSX8 $dst.lo,$mem\t# byte -> long\n\t"
            "MOV    $dst.hi,$dst.lo\n\t"
            "SAR    $dst.hi,7" %}

  ins_encode %{
    __ movsbl($dst$$Register, $mem$$Address);
    __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
    __ sarl(HIGH_FROM_LOW($dst$$Register), 7); // 24+1 MSB are already signed extended.
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Byte (8bit UNsigned)
instruct loadUB(xRegI dst, memory mem) %{
  match(Set dst (LoadUB mem));

  ins_cost(125);
  format %{ "MOVZX8 $dst,$mem\t# ubyte -> int" %}

  ins_encode %{
    __ movzbl($dst$$Register, $mem$$Address);
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Byte (8 bit UNsigned) into Long Register
instruct loadUB2L(eRegL dst, memory mem, eFlagsReg cr) %{
  match(Set dst (ConvI2L (LoadUB mem)));
  effect(KILL cr);

  ins_cost(250);
  format %{ "MOVZX8 $dst.lo,$mem\t# ubyte -> long\n\t"
            "XOR    $dst.hi,$dst.hi" %}

  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movzbl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Byte (8 bit UNsigned) with mask into Long Register
instruct loadUB2L_immI8(eRegL dst, memory mem, immI8 mask, eFlagsReg cr) %{
  match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
  effect(KILL cr);

  format %{ "MOVZX8 $dst.lo,$mem\t# ubyte & 8-bit mask -> long\n\t"
            "XOR    $dst.hi,$dst.hi\n\t"
            "AND    $dst.lo,$mask" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movzbl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
    __ andl(Rdst, $mask$$constant);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Short (16bit signed)
instruct loadS(rRegI dst, memory mem) %{
  match(Set dst (LoadS mem));

  ins_cost(125);
  format %{ "MOVSX  $dst,$mem\t# short" %}

  ins_encode %{
    __ movswl($dst$$Register, $mem$$Address);
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Short (16 bit signed) to Byte (8 bit signed)
instruct loadS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));

  ins_cost(125);
  format %{ "MOVSX  $dst, $mem\t# short -> byte" %}
  ins_encode %{
    __ movsbl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Short (16bit signed) into Long Register
instruct loadS2L(eRegL dst, memory mem, eFlagsReg cr) %{
  match(Set dst (ConvI2L (LoadS mem)));
  effect(KILL cr);

  ins_cost(375);
  format %{ "MOVSX  $dst.lo,$mem\t# short -> long\n\t"
            "MOV    $dst.hi,$dst.lo\n\t"
            "SAR    $dst.hi,15" %}

  ins_encode %{
    __ movswl($dst$$Register, $mem$$Address);
    __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
    __ sarl(HIGH_FROM_LOW($dst$$Register), 15); // 16+1 MSB are already signed extended.
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Short/Char (16bit unsigned)
instruct loadUS(rRegI dst, memory mem) %{
  match(Set dst (LoadUS mem));

  ins_cost(125);
  format %{ "MOVZX  $dst,$mem\t# ushort/char -> int" %}

  ins_encode %{
    __ movzwl($dst$$Register, $mem$$Address);
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
instruct loadUS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));

  ins_cost(125);
  format %{ "MOVSX  $dst, $mem\t# ushort -> byte" %}
  ins_encode %{
    __ movsbl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Short/Char (16 bit UNsigned) into Long Register
instruct loadUS2L(eRegL dst, memory mem, eFlagsReg cr) %{
  match(Set dst (ConvI2L (LoadUS mem)));
  effect(KILL cr);

  ins_cost(250);
  format %{ "MOVZX  $dst.lo,$mem\t# ushort/char -> long\n\t"
            "XOR    $dst.hi,$dst.hi" %}

  ins_encode %{
    __ movzwl($dst$$Register, $mem$$Address);
    __ xorl(HIGH_FROM_LOW($dst$$Register), HIGH_FROM_LOW($dst$$Register));
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Short/Char (16 bit UNsigned) with mask 0xFF into Long Register
instruct loadUS2L_immI_255(eRegL dst, memory mem, immI_255 mask, eFlagsReg cr) %{
  match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
  effect(KILL cr);

  format %{ "MOVZX8 $dst.lo,$mem\t# ushort/char & 0xFF -> long\n\t"
            "XOR    $dst.hi,$dst.hi" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movzbl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Short/Char (16 bit UNsigned) with a 16-bit mask into Long Register
instruct loadUS2L_immI16(eRegL dst, memory mem, immI16 mask, eFlagsReg cr) %{
  match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
  effect(KILL cr);

  format %{ "MOVZX  $dst.lo, $mem\t# ushort/char & 16-bit mask -> long\n\t"
            "XOR    $dst.hi,$dst.hi\n\t"
            "AND    $dst.lo,$mask" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movzwl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
    __ andl(Rdst, $mask$$constant);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer
instruct loadI(rRegI dst, memory mem) %{
  match(Set dst (LoadI mem));

  ins_cost(125);
  format %{ "MOV    $dst,$mem\t# int" %}

  ins_encode %{
    __ movl($dst$$Register, $mem$$Address);
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Integer (32 bit signed) to Byte (8 bit signed)
instruct loadI2B(rRegI dst, memory mem, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));

  ins_cost(125);
  format %{ "MOVSX  $dst, $mem\t# int -> byte" %}
  ins_encode %{
    __ movsbl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer (32 bit signed) to Unsigned Byte (8 bit UNsigned)
instruct loadI2UB(rRegI dst, memory mem, immI_255 mask) %{
  match(Set dst (AndI (LoadI mem) mask));

  ins_cost(125);
  format %{ "MOVZX  $dst, $mem\t# int -> ubyte" %}
  ins_encode %{
    __ movzbl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer (32 bit signed) to Short (16 bit signed)
instruct loadI2S(rRegI dst, memory mem, immI_16 sixteen) %{
  match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));

  ins_cost(125);
  format %{ "MOVSX  $dst, $mem\t# int -> short" %}
  ins_encode %{
    __ movswl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer (32 bit signed) to Unsigned Short/Char (16 bit UNsigned)
instruct loadI2US(rRegI dst, memory mem, immI_65535 mask) %{
  match(Set dst (AndI (LoadI mem) mask));

  ins_cost(125);
  format %{ "MOVZX  $dst, $mem\t# int -> ushort/char" %}
  ins_encode %{
    __ movzwl($dst$$Register, $mem$$Address);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer into Long Register
instruct loadI2L(eRegL dst, memory mem, eFlagsReg cr) %{
  match(Set dst (ConvI2L (LoadI mem)));
  effect(KILL cr);

  ins_cost(375);
  format %{ "MOV    $dst.lo,$mem\t# int -> long\n\t"
            "MOV    $dst.hi,$dst.lo\n\t"
            "SAR    $dst.hi,31" %}

  ins_encode %{
    __ movl($dst$$Register, $mem$$Address);
    __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
    __ sarl(HIGH_FROM_LOW($dst$$Register), 31);
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Integer with mask 0xFF into Long Register
instruct loadI2L_immI_255(eRegL dst, memory mem, immI_255 mask, eFlagsReg cr) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  effect(KILL cr);

  format %{ "MOVZX8 $dst.lo,$mem\t# int & 0xFF -> long\n\t"
            "XOR    $dst.hi,$dst.hi" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movzbl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer with mask 0xFFFF into Long Register
instruct loadI2L_immI_65535(eRegL dst, memory mem, immI_65535 mask, eFlagsReg cr) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  effect(KILL cr);

  format %{ "MOVZX  $dst.lo,$mem\t# int & 0xFFFF -> long\n\t"
            "XOR    $dst.hi,$dst.hi" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movzwl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Integer with 32-bit mask into Long Register
instruct loadI2L_immI(eRegL dst, memory mem, immI mask, eFlagsReg cr) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  effect(KILL cr);

  format %{ "MOV    $dst.lo,$mem\t# int & 32-bit mask -> long\n\t"
            "XOR    $dst.hi,$dst.hi\n\t"
            "AND    $dst.lo,$mask" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ movl(Rdst, $mem$$Address);
    __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
    __ andl(Rdst, $mask$$constant);
  %}
  ins_pipe(ialu_reg_mem);
%}

// Load Unsigned Integer into Long Register
instruct loadUI2L(eRegL dst, memory mem, immL_32bits mask, eFlagsReg cr) %{
  match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
  effect(KILL cr);

  ins_cost(250);
  format %{ "MOV    $dst.lo,$mem\t# uint -> long\n\t"
            "XOR    $dst.hi,$dst.hi" %}

  ins_encode %{
    __ movl($dst$$Register, $mem$$Address);
    __ xorl(HIGH_FROM_LOW($dst$$Register), HIGH_FROM_LOW($dst$$Register));
  %}

  ins_pipe(ialu_reg_mem);
%}

// Load Long.  Cannot clobber address while loading, so restrict address
// register to ESI
instruct loadL(eRegL dst, load_long_memory mem) %{
  predicate(!((LoadLNode*)n)->require_atomic_access());
  match(Set dst (LoadL mem));

  ins_cost(250);
  format %{ "MOV    $dst.lo,$mem\t# long\n\t"
            "MOV    $dst.hi,$mem+4" %}

  ins_encode %{
    Address Amemlo = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, relocInfo::none);
    Address Amemhi = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, relocInfo::none);
    __ movl($dst$$Register, Amemlo);
    __ movl(HIGH_FROM_LOW($dst$$Register), Amemhi);
  %}

  ins_pipe(ialu_reg_long_mem);
%}

// Volatile Load Long.  Must be atomic, so do 64-bit FILD
// then store it down to the stack and reload on the int
// side.
instruct loadL_volatile(stackSlotL dst, memory mem) %{
  predicate(UseSSE<=1 && ((LoadLNode*)n)->require_atomic_access());
  match(Set dst (LoadL mem));

  ins_cost(200);
  format %{ "FILD   $mem\t# Atomic volatile long load\n\t"
            "FISTp  $dst" %}
  ins_encode(enc_loadL_volatile(mem,dst));
  ins_pipe( fpu_reg_mem );
%}

instruct loadLX_volatile(stackSlotL dst, memory mem, regD tmp) %{
  predicate(UseSSE>=2 && ((LoadLNode*)n)->require_atomic_access());
  match(Set dst (LoadL mem));
  effect(TEMP tmp);
  ins_cost(180);
  format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
            "MOVSD  $dst,$tmp" %}
  ins_encode %{
    __ movdbl($tmp$$XMMRegister, $mem$$Address);
    __ movdbl(Address(rsp, $dst$$disp), $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct loadLX_reg_volatile(eRegL dst, memory mem, regD tmp) %{
  predicate(UseSSE>=2 && ((LoadLNode*)n)->require_atomic_access());
  match(Set dst (LoadL mem));
  effect(TEMP tmp);
  ins_cost(160);
  format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
            "MOVD   $dst.lo,$tmp\n\t"
            "PSRLQ  $tmp,32\n\t"
            "MOVD   $dst.hi,$tmp" %}
  ins_encode %{
    __ movdbl($tmp$$XMMRegister, $mem$$Address);
    __ movdl($dst$$Register, $tmp$$XMMRegister);
    __ psrlq($tmp$$XMMRegister, 32);
    __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Load Range
instruct loadRange(rRegI dst, memory mem) %{
  match(Set dst (LoadRange mem));

  ins_cost(125);
  format %{ "MOV    $dst,$mem" %}
  opcode(0x8B);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_mem );
%}


// Load Pointer
instruct loadP(eRegP dst, memory mem) %{
  match(Set dst (LoadP mem));

  ins_cost(125);
  format %{ "MOV    $dst,$mem" %}
  opcode(0x8B);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_mem );
%}

// Load Klass Pointer
instruct loadKlass(eRegP dst, memory mem) %{
  match(Set dst (LoadKlass mem));

  ins_cost(125);
  format %{ "MOV    $dst,$mem" %}
  opcode(0x8B);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_mem );
%}

// Load Double
instruct loadDPR(regDPR dst, memory mem) %{
  predicate(UseSSE<=1);
  match(Set dst (LoadD mem));

  ins_cost(150);
  format %{ "FLD_D  ST,$mem\n\t"
            "FSTP   $dst" %}
  opcode(0xDD);               /* DD /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),
              Pop_Reg_DPR(dst) );
  ins_pipe( fpu_reg_mem );
%}

// Load Double to XMM
instruct loadD(regD dst, memory mem) %{
  predicate(UseSSE>=2 && UseXmmLoadAndClearUpper);
  match(Set dst (LoadD mem));
  ins_cost(145);
  format %{ "MOVSD  $dst,$mem" %}
  ins_encode %{
    __ movdbl ($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

instruct loadD_partial(regD dst, memory mem) %{
  predicate(UseSSE>=2 && !UseXmmLoadAndClearUpper);
  match(Set dst (LoadD mem));
  ins_cost(145);
  format %{ "MOVLPD $dst,$mem" %}
  ins_encode %{
    __ movdbl ($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Load to XMM register (single-precision floating point)
// MOVSS instruction
instruct loadF(regF dst, memory mem) %{
  predicate(UseSSE>=1);
  match(Set dst (LoadF mem));
  ins_cost(145);
  format %{ "MOVSS  $dst,$mem" %}
  ins_encode %{
    __ movflt ($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Load Float
instruct loadFPR(regFPR dst, memory mem) %{
  predicate(UseSSE==0);
  match(Set dst (LoadF mem));

  ins_cost(150);
  format %{ "FLD_S  ST,$mem\n\t"
            "FSTP   $dst" %}
  opcode(0xD9);               /* D9 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_mem );
%}

// Load Effective Address
instruct leaP8(eRegP dst, indOffset8 mem) %{
  match(Set dst mem);

  ins_cost(110);
  format %{ "LEA    $dst,$mem" %}
  opcode(0x8D);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_reg_fat );
%}

instruct leaP32(eRegP dst, indOffset32 mem) %{
  match(Set dst mem);

  ins_cost(110);
  format %{ "LEA    $dst,$mem" %}
  opcode(0x8D);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_reg_fat );
%}

instruct leaPIdxOff(eRegP dst, indIndexOffset mem) %{
  match(Set dst mem);

  ins_cost(110);
  format %{ "LEA    $dst,$mem" %}
  opcode(0x8D);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_reg_fat );
%}

instruct leaPIdxScale(eRegP dst, indIndexScale mem) %{
  match(Set dst mem);

  ins_cost(110);
  format %{ "LEA    $dst,$mem" %}
  opcode(0x8D);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_reg_fat );
%}

instruct leaPIdxScaleOff(eRegP dst, indIndexScaleOffset mem) %{
  match(Set dst mem);

  ins_cost(110);
  format %{ "LEA    $dst,$mem" %}
  opcode(0x8D);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_reg_fat );
%}

// Load Constant
instruct loadConI(rRegI dst, immI src) %{
  match(Set dst src);

  format %{ "MOV    $dst,$src" %}
  ins_encode( LdImmI(dst, src) );
  ins_pipe( ialu_reg_fat );
%}

// Load Constant zero
instruct loadConI0(rRegI dst, immI0 src, eFlagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);

  ins_cost(50);
  format %{ "XOR    $dst,$dst" %}
  opcode(0x33);  /* + rd */
  ins_encode( OpcP, RegReg( dst, dst ) );
  ins_pipe( ialu_reg );
%}

instruct loadConP(eRegP dst, immP src) %{
  match(Set dst src);

  format %{ "MOV    $dst,$src" %}
  opcode(0xB8);  /* + rd */
  ins_encode( LdImmP(dst, src) );
  ins_pipe( ialu_reg_fat );
%}

instruct loadConL(eRegL dst, immL src, eFlagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(200);
  format %{ "MOV    $dst.lo,$src.lo\n\t"
            "MOV    $dst.hi,$src.hi" %}
  opcode(0xB8);
  ins_encode( LdImmL_Lo(dst, src), LdImmL_Hi(dst, src) );
  ins_pipe( ialu_reg_long_fat );
%}

instruct loadConL0(eRegL dst, immL0 src, eFlagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(150);
  format %{ "XOR    $dst.lo,$dst.lo\n\t"
            "XOR    $dst.hi,$dst.hi" %}
  opcode(0x33,0x33);
  ins_encode( RegReg_Lo(dst,dst), RegReg_Hi(dst, dst) );
  ins_pipe( ialu_reg_long );
%}

// The instruction usage is guarded by predicate in operand immFPR().
instruct loadConFPR(regFPR dst, immFPR con) %{
  match(Set dst con);
  ins_cost(125);
  format %{ "FLD_S  ST,[$constantaddress]\t# load from constant table: float=$con\n\t"
            "FSTP   $dst" %}
  ins_encode %{
    __ fld_s($constantaddress($con));
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_con);
%}

// The instruction usage is guarded by predicate in operand immFPR0().
instruct loadConFPR0(regFPR dst, immFPR0 con) %{
  match(Set dst con);
  ins_cost(125);
  format %{ "FLDZ   ST\n\t"
            "FSTP   $dst" %}
  ins_encode %{
    __ fldz();
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_con);
%}

// The instruction usage is guarded by predicate in operand immFPR1().
instruct loadConFPR1(regFPR dst, immFPR1 con) %{
  match(Set dst con);
  ins_cost(125);
  format %{ "FLD1   ST\n\t"
            "FSTP   $dst" %}
  ins_encode %{
    __ fld1();
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_con);
%}

// The instruction usage is guarded by predicate in operand immF().
instruct loadConF(regF dst, immF con) %{
  match(Set dst con);
  ins_cost(125);
  format %{ "MOVSS  $dst,[$constantaddress]\t# load from constant table: float=$con" %}
  ins_encode %{
    __ movflt($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

// The instruction usage is guarded by predicate in operand immF0().
instruct loadConF0(regF dst, immF0 src) %{
  match(Set dst src);
  ins_cost(100);
  format %{ "XORPS  $dst,$dst\t# float 0.0" %}
  ins_encode %{
    __ xorps($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

// The instruction usage is guarded by predicate in operand immDPR().
instruct loadConDPR(regDPR dst, immDPR con) %{
  match(Set dst con);
  ins_cost(125);

  format %{ "FLD_D  ST,[$constantaddress]\t# load from constant table: double=$con\n\t"
            "FSTP   $dst" %}
  ins_encode %{
    __ fld_d($constantaddress($con));
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_con);
%}

// The instruction usage is guarded by predicate in operand immDPR0().
instruct loadConDPR0(regDPR dst, immDPR0 con) %{
  match(Set dst con);
  ins_cost(125);

  format %{ "FLDZ   ST\n\t"
            "FSTP   $dst" %}
  ins_encode %{
    __ fldz();
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_con);
%}

// The instruction usage is guarded by predicate in operand immDPR1().
instruct loadConDPR1(regDPR dst, immDPR1 con) %{
  match(Set dst con);
  ins_cost(125);

  format %{ "FLD1   ST\n\t"
            "FSTP   $dst" %}
  ins_encode %{
    __ fld1();
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_con);
%}

// The instruction usage is guarded by predicate in operand immD().
instruct loadConD(regD dst, immD con) %{
  match(Set dst con);
  ins_cost(125);
  format %{ "MOVSD  $dst,[$constantaddress]\t# load from constant table: double=$con" %}
  ins_encode %{
    __ movdbl($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

// The instruction usage is guarded by predicate in operand immD0().
instruct loadConD0(regD dst, immD0 src) %{
  match(Set dst src);
  ins_cost(100);
  format %{ "XORPD  $dst,$dst\t# double 0.0" %}
  ins_encode %{
    __ xorpd ($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Load Stack Slot
instruct loadSSI(rRegI dst, stackSlotI src) %{
  match(Set dst src);
  ins_cost(125);

  format %{ "MOV    $dst,$src" %}
  opcode(0x8B);
  ins_encode( OpcP, RegMem(dst,src));
  ins_pipe( ialu_reg_mem );
%}

instruct loadSSL(eRegL dst, stackSlotL src) %{
  match(Set dst src);

  ins_cost(200);
  format %{ "MOV    $dst,$src.lo\n\t"
            "MOV    $dst+4,$src.hi" %}
  opcode(0x8B, 0x8B);
  ins_encode( OpcP, RegMem( dst, src ), OpcS, RegMem_Hi( dst, src ) );
  ins_pipe( ialu_mem_long_reg );
%}

// Load Stack Slot
instruct loadSSP(eRegP dst, stackSlotP src) %{
  match(Set dst src);
  ins_cost(125);

  format %{ "MOV    $dst,$src" %}
  opcode(0x8B);
  ins_encode( OpcP, RegMem(dst,src));
  ins_pipe( ialu_reg_mem );
%}

// Load Stack Slot
instruct loadSSF(regFPR dst, stackSlotF src) %{
  match(Set dst src);
  ins_cost(125);

  format %{ "FLD_S  $src\n\t"
            "FSTP   $dst" %}
  opcode(0xD9);               /* D9 /0, FLD m32real */
  ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_mem );
%}

// Load Stack Slot
instruct loadSSD(regDPR dst, stackSlotD src) %{
  match(Set dst src);
  ins_cost(125);

  format %{ "FLD_D  $src\n\t"
            "FSTP   $dst" %}
  opcode(0xDD);               /* DD /0, FLD m64real */
  ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
              Pop_Reg_DPR(dst) );
  ins_pipe( fpu_reg_mem );
%}

// Prefetch instructions.
// Must be safe to execute with invalid address (cannot fault).

instruct prefetchr0( memory mem ) %{
  predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
  match(PrefetchRead mem);
  ins_cost(0);
  size(0);
  format %{ "PREFETCHR (non-SSE is empty encoding)" %}
  ins_encode();
  ins_pipe(empty);
%}

instruct prefetchr( memory mem ) %{
  predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch() || ReadPrefetchInstr==3);
  match(PrefetchRead mem);
  ins_cost(100);

  format %{ "PREFETCHR $mem\t! Prefetch into level 1 cache for read" %}
  ins_encode %{
    __ prefetchr($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchrNTA( memory mem ) %{
  predicate(UseSSE>=1 && ReadPrefetchInstr==0);
  match(PrefetchRead mem);
  ins_cost(100);

  format %{ "PREFETCHNTA $mem\t! Prefetch into non-temporal cache for read" %}
  ins_encode %{
    __ prefetchnta($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchrT0( memory mem ) %{
  predicate(UseSSE>=1 && ReadPrefetchInstr==1);
  match(PrefetchRead mem);
  ins_cost(100);

  format %{ "PREFETCHT0 $mem\t! Prefetch into L1 and L2 caches for read" %}
  ins_encode %{
    __ prefetcht0($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchrT2( memory mem ) %{
  predicate(UseSSE>=1 && ReadPrefetchInstr==2);
  match(PrefetchRead mem);
  ins_cost(100);

  format %{ "PREFETCHT2 $mem\t! Prefetch into L2 cache for read" %}
  ins_encode %{
    __ prefetcht2($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchw0( memory mem ) %{
  predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
  match(PrefetchWrite mem);
  ins_cost(0);
  size(0);
  format %{ "Prefetch (non-SSE is empty encoding)" %}
  ins_encode();
  ins_pipe(empty);
%}

instruct prefetchw( memory mem ) %{
  predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch());
  match( PrefetchWrite mem );
  ins_cost(100);

  format %{ "PREFETCHW $mem\t! Prefetch into L1 cache and mark modified" %}
  ins_encode %{
    __ prefetchw($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchwNTA( memory mem ) %{
  predicate(UseSSE>=1);
  match(PrefetchWrite mem);
  ins_cost(100);

  format %{ "PREFETCHNTA $mem\t! Prefetch into non-temporal cache for write" %}
  ins_encode %{
    __ prefetchnta($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

// Prefetch instructions for allocation.

instruct prefetchAlloc0( memory mem ) %{
  predicate(UseSSE==0 && AllocatePrefetchInstr!=3);
  match(PrefetchAllocation mem);
  ins_cost(0);
  size(0);
  format %{ "Prefetch allocation (non-SSE is empty encoding)" %}
  ins_encode();
  ins_pipe(empty);
%}

instruct prefetchAlloc( memory mem ) %{
  predicate(AllocatePrefetchInstr==3);
  match( PrefetchAllocation mem );
  ins_cost(100);

  format %{ "PREFETCHW $mem\t! Prefetch allocation into L1 cache and mark modified" %}
  ins_encode %{
    __ prefetchw($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchAllocNTA( memory mem ) %{
  predicate(UseSSE>=1 && AllocatePrefetchInstr==0);
  match(PrefetchAllocation mem);
  ins_cost(100);

  format %{ "PREFETCHNTA $mem\t! Prefetch allocation into non-temporal cache for write" %}
  ins_encode %{
    __ prefetchnta($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchAllocT0( memory mem ) %{
  predicate(UseSSE>=1 && AllocatePrefetchInstr==1);
  match(PrefetchAllocation mem);
  ins_cost(100);

  format %{ "PREFETCHT0 $mem\t! Prefetch allocation into L1 and L2 caches for write" %}
  ins_encode %{
    __ prefetcht0($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

instruct prefetchAllocT2( memory mem ) %{
  predicate(UseSSE>=1 && AllocatePrefetchInstr==2);
  match(PrefetchAllocation mem);
  ins_cost(100);

  format %{ "PREFETCHT2 $mem\t! Prefetch allocation into L2 cache for write" %}
  ins_encode %{
    __ prefetcht2($mem$$Address);
  %}
  ins_pipe(ialu_mem);
%}

//----------Store Instructions-------------------------------------------------

// Store Byte
instruct storeB(memory mem, xRegI src) %{
  match(Set mem (StoreB mem src));

  ins_cost(125);
  format %{ "MOV8   $mem,$src" %}
  opcode(0x88);
  ins_encode( OpcP, RegMem( src, mem ) );
  ins_pipe( ialu_mem_reg );
%}

// Store Char/Short
instruct storeC(memory mem, rRegI src) %{
  match(Set mem (StoreC mem src));

  ins_cost(125);
  format %{ "MOV16  $mem,$src" %}
  opcode(0x89, 0x66);
  ins_encode( OpcS, OpcP, RegMem( src, mem ) );
  ins_pipe( ialu_mem_reg );
%}

// Store Integer
instruct storeI(memory mem, rRegI src) %{
  match(Set mem (StoreI mem src));

  ins_cost(125);
  format %{ "MOV    $mem,$src" %}
  opcode(0x89);
  ins_encode( OpcP, RegMem( src, mem ) );
  ins_pipe( ialu_mem_reg );
%}

// Store Long
instruct storeL(long_memory mem, eRegL src) %{
  predicate(!((StoreLNode*)n)->require_atomic_access());
  match(Set mem (StoreL mem src));

  ins_cost(200);
  format %{ "MOV    $mem,$src.lo\n\t"
            "MOV    $mem+4,$src.hi" %}
  opcode(0x89, 0x89);
  ins_encode( OpcP, RegMem( src, mem ), OpcS, RegMem_Hi( src, mem ) );
  ins_pipe( ialu_mem_long_reg );
%}

// Store Long to Integer
instruct storeL2I(memory mem, eRegL src) %{
  match(Set mem (StoreI mem (ConvL2I src)));

  format %{ "MOV    $mem,$src.lo\t# long -> int" %}
  ins_encode %{
    __ movl($mem$$Address, $src$$Register);
  %}
  ins_pipe(ialu_mem_reg);
%}

// Volatile Store Long.  Must be atomic, so move it into
// the FP TOS and then do a 64-bit FIST.  Has to probe the
// target address before the store (for null-ptr checks)
// so the memory operand is used twice in the encoding.
instruct storeL_volatile(memory mem, stackSlotL src, eFlagsReg cr ) %{
  predicate(UseSSE<=1 && ((StoreLNode*)n)->require_atomic_access());
  match(Set mem (StoreL mem src));
  effect( KILL cr );
  ins_cost(400);
  format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
            "FILD   $src\n\t"
            "FISTp  $mem\t # 64-bit atomic volatile long store" %}
  opcode(0x3B);
  ins_encode( OpcP, RegMem( EAX, mem ), enc_storeL_volatile(mem,src));
  ins_pipe( fpu_reg_mem );
%}

instruct storeLX_volatile(memory mem, stackSlotL src, regD tmp, eFlagsReg cr) %{
  predicate(UseSSE>=2 && ((StoreLNode*)n)->require_atomic_access());
  match(Set mem (StoreL mem src));
  effect( TEMP tmp, KILL cr );
  ins_cost(380);
  format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
            "MOVSD  $tmp,$src\n\t"
            "MOVSD  $mem,$tmp\t # 64-bit atomic volatile long store" %}
  ins_encode %{
    __ cmpl(rax, $mem$$Address);
    __ movdbl($tmp$$XMMRegister, Address(rsp, $src$$disp));
    __ movdbl($mem$$Address, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct storeLX_reg_volatile(memory mem, eRegL src, regD tmp2, regD tmp, eFlagsReg cr) %{
  predicate(UseSSE>=2 && ((StoreLNode*)n)->require_atomic_access());
  match(Set mem (StoreL mem src));
  effect( TEMP tmp2 , TEMP tmp, KILL cr );
  ins_cost(360);
  format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
            "MOVD   $tmp,$src.lo\n\t"
            "MOVD   $tmp2,$src.hi\n\t"
            "PUNPCKLDQ $tmp,$tmp2\n\t"
            "MOVSD  $mem,$tmp\t # 64-bit atomic volatile long store" %}
  ins_encode %{
    __ cmpl(rax, $mem$$Address);
    __ movdl($tmp$$XMMRegister, $src$$Register);
    __ movdl($tmp2$$XMMRegister, HIGH_FROM_LOW($src$$Register));
    __ punpckldq($tmp$$XMMRegister, $tmp2$$XMMRegister);
    __ movdbl($mem$$Address, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Store Pointer; for storing unknown oops and raw pointers
instruct storeP(memory mem, anyRegP src) %{
  match(Set mem (StoreP mem src));

  ins_cost(125);
  format %{ "MOV    $mem,$src" %}
  opcode(0x89);
  ins_encode( OpcP, RegMem( src, mem ) );
  ins_pipe( ialu_mem_reg );
%}

// Store Integer Immediate
instruct storeImmI(memory mem, immI src) %{
  match(Set mem (StoreI mem src));

  ins_cost(150);
  format %{ "MOV    $mem,$src" %}
  opcode(0xC7);               /* C7 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store Short/Char Immediate
instruct storeImmI16(memory mem, immI16 src) %{
  predicate(UseStoreImmI16);
  match(Set mem (StoreC mem src));

  ins_cost(150);
  format %{ "MOV16  $mem,$src" %}
  opcode(0xC7);     /* C7 /0 Same as 32 store immediate with prefix */
  ins_encode( SizePrefix, OpcP, RMopc_Mem(0x00,mem),  Con16( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store Pointer Immediate; null pointers or constant oops that do not
// need card-mark barriers.
instruct storeImmP(memory mem, immP src) %{
  match(Set mem (StoreP mem src));

  ins_cost(150);
  format %{ "MOV    $mem,$src" %}
  opcode(0xC7);               /* C7 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store Byte Immediate
instruct storeImmB(memory mem, immI8 src) %{
  match(Set mem (StoreB mem src));

  ins_cost(150);
  format %{ "MOV8   $mem,$src" %}
  opcode(0xC6);               /* C6 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con8or32( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store CMS card-mark Immediate
instruct storeImmCM(memory mem, immI8 src) %{
  match(Set mem (StoreCM mem src));

  ins_cost(150);
  format %{ "MOV8   $mem,$src\t! CMS card-mark imm0" %}
  opcode(0xC6);               /* C6 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con8or32( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store Double
instruct storeDPR( memory mem, regDPR1 src) %{
  predicate(UseSSE<=1);
  match(Set mem (StoreD mem src));

  ins_cost(100);
  format %{ "FST_D  $mem,$src" %}
  opcode(0xDD);       /* DD /2 */
  ins_encode( enc_FPR_store(mem,src) );
  ins_pipe( fpu_mem_reg );
%}

// Store double does rounding on x86
instruct storeDPR_rounded( memory mem, regDPR1 src) %{
  predicate(UseSSE<=1);
  match(Set mem (StoreD mem (RoundDouble src)));

  ins_cost(100);
  format %{ "FST_D  $mem,$src\t# round" %}
  opcode(0xDD);       /* DD /2 */
  ins_encode( enc_FPR_store(mem,src) );
  ins_pipe( fpu_mem_reg );
%}

// Store XMM register to memory (double-precision floating points)
// MOVSD instruction
instruct storeD(memory mem, regD src) %{
  predicate(UseSSE>=2);
  match(Set mem (StoreD mem src));
  ins_cost(95);
  format %{ "MOVSD  $mem,$src" %}
  ins_encode %{
    __ movdbl($mem$$Address, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Store XMM register to memory (single-precision floating point)
// MOVSS instruction
instruct storeF(memory mem, regF src) %{
  predicate(UseSSE>=1);
  match(Set mem (StoreF mem src));
  ins_cost(95);
  format %{ "MOVSS  $mem,$src" %}
  ins_encode %{
    __ movflt($mem$$Address, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Store Float
instruct storeFPR( memory mem, regFPR1 src) %{
  predicate(UseSSE==0);
  match(Set mem (StoreF mem src));

  ins_cost(100);
  format %{ "FST_S  $mem,$src" %}
  opcode(0xD9);       /* D9 /2 */
  ins_encode( enc_FPR_store(mem,src) );
  ins_pipe( fpu_mem_reg );
%}

// Store Float does rounding on x86
instruct storeFPR_rounded( memory mem, regFPR1 src) %{
  predicate(UseSSE==0);
  match(Set mem (StoreF mem (RoundFloat src)));

  ins_cost(100);
  format %{ "FST_S  $mem,$src\t# round" %}
  opcode(0xD9);       /* D9 /2 */
  ins_encode( enc_FPR_store(mem,src) );
  ins_pipe( fpu_mem_reg );
%}

// Store Float does rounding on x86
instruct storeFPR_Drounded( memory mem, regDPR1 src) %{
  predicate(UseSSE<=1);
  match(Set mem (StoreF mem (ConvD2F src)));

  ins_cost(100);
  format %{ "FST_S  $mem,$src\t# D-round" %}
  opcode(0xD9);       /* D9 /2 */
  ins_encode( enc_FPR_store(mem,src) );
  ins_pipe( fpu_mem_reg );
%}

// Store immediate Float value (it is faster than store from FPU register)
// The instruction usage is guarded by predicate in operand immFPR().
instruct storeFPR_imm( memory mem, immFPR src) %{
  match(Set mem (StoreF mem src));

  ins_cost(50);
  format %{ "MOV    $mem,$src\t# store float" %}
  opcode(0xC7);               /* C7 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32FPR_as_bits( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store immediate Float value (it is faster than store from XMM register)
// The instruction usage is guarded by predicate in operand immF().
instruct storeF_imm( memory mem, immF src) %{
  match(Set mem (StoreF mem src));

  ins_cost(50);
  format %{ "MOV    $mem,$src\t# store float" %}
  opcode(0xC7);               /* C7 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32F_as_bits( src ));
  ins_pipe( ialu_mem_imm );
%}

// Store Integer to stack slot
instruct storeSSI(stackSlotI dst, rRegI src) %{
  match(Set dst src);

  ins_cost(100);
  format %{ "MOV    $dst,$src" %}
  opcode(0x89);
  ins_encode( OpcPRegSS( dst, src ) );
  ins_pipe( ialu_mem_reg );
%}

// Store Integer to stack slot
instruct storeSSP(stackSlotP dst, eRegP src) %{
  match(Set dst src);

  ins_cost(100);
  format %{ "MOV    $dst,$src" %}
  opcode(0x89);
  ins_encode( OpcPRegSS( dst, src ) );
  ins_pipe( ialu_mem_reg );
%}

// Store Long to stack slot
instruct storeSSL(stackSlotL dst, eRegL src) %{
  match(Set dst src);

  ins_cost(200);
  format %{ "MOV    $dst,$src.lo\n\t"
            "MOV    $dst+4,$src.hi" %}
  opcode(0x89, 0x89);
  ins_encode( OpcP, RegMem( src, dst ), OpcS, RegMem_Hi( src, dst ) );
  ins_pipe( ialu_mem_long_reg );
%}

//----------MemBar Instructions-----------------------------------------------
// Memory barrier flavors

instruct membar_acquire() %{
  match(MemBarAcquire);
  ins_cost(400);

  size(0);
  format %{ "MEMBAR-acquire ! (empty encoding)" %}
  ins_encode();
  ins_pipe(empty);
%}

instruct membar_acquire_lock() %{
  match(MemBarAcquireLock);
  ins_cost(0);

  size(0);
  format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

instruct membar_release() %{
  match(MemBarRelease);
  ins_cost(400);

  size(0);
  format %{ "MEMBAR-release ! (empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

instruct membar_release_lock() %{
  match(MemBarReleaseLock);
  ins_cost(0);

  size(0);
  format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

instruct membar_volatile(eFlagsReg cr) %{
  match(MemBarVolatile);
  effect(KILL cr);
  ins_cost(400);

  format %{ 
    $$template
    if (os::is_MP()) {
      $$emit$$"LOCK ADDL [ESP + #0], 0\t! membar_volatile"
    } else {
      $$emit$$"MEMBAR-volatile ! (empty encoding)"
    }
  %}
  ins_encode %{
    __ membar(Assembler::StoreLoad);
  %}
  ins_pipe(pipe_slow);
%}

instruct unnecessary_membar_volatile() %{
  match(MemBarVolatile);
  predicate(Matcher::post_store_load_barrier(n));
  ins_cost(0);

  size(0);
  format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

instruct membar_storestore() %{
  match(MemBarStoreStore);
  ins_cost(0);

  size(0);
  format %{ "MEMBAR-storestore (empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

//----------Move Instructions--------------------------------------------------
instruct castX2P(eAXRegP dst, eAXRegI src) %{
  match(Set dst (CastX2P src));
  format %{ "# X2P  $dst, $src" %}
  ins_encode( /*empty encoding*/ );
  ins_cost(0);
  ins_pipe(empty);
%}

instruct castP2X(rRegI dst, eRegP src ) %{
  match(Set dst (CastP2X src));
  ins_cost(50);
  format %{ "MOV    $dst, $src\t# CastP2X" %}
  ins_encode( enc_Copy( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

//----------Conditional Move---------------------------------------------------
// Conditional move
instruct jmovI_reg(cmpOp cop, eFlagsReg cr, rRegI dst, rRegI src) %{
  predicate(!VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "J$cop,us skip\t# signed cmove\n\t"
            "MOV    $dst,$src\n"
      "skip:" %}
  ins_encode %{
    Label Lskip;
    // Invert sense of branch from sense of CMOV
    __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
    __ movl($dst$$Register, $src$$Register);
    __ bind(Lskip);
  %}
  ins_pipe( pipe_cmov_reg );
%}

instruct jmovI_regU(cmpOpU cop, eFlagsRegU cr, rRegI dst, rRegI src) %{
  predicate(!VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "J$cop,us skip\t# unsigned cmove\n\t"
            "MOV    $dst,$src\n"
      "skip:" %}
  ins_encode %{
    Label Lskip;
    // Invert sense of branch from sense of CMOV
    __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
    __ movl($dst$$Register, $src$$Register);
    __ bind(Lskip);
  %}
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovI_reg(rRegI dst, rRegI src, eFlagsReg cr, cmpOp cop ) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cop $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovI_regU( cmpOpU cop, eFlagsRegU cr, rRegI dst, rRegI src ) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cop $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovI_regUCF( cmpOpUCF cop, eFlagsRegUCF cr, rRegI dst, rRegI src ) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  expand %{
    cmovI_regU(cop, cr, dst, src);
  %}
%}

// Conditional move
instruct cmovI_mem(cmpOp cop, eFlagsReg cr, rRegI dst, memory src) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
  ins_cost(250);
  format %{ "CMOV$cop $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
  ins_pipe( pipe_cmov_mem );
%}

// Conditional move
instruct cmovI_memU(cmpOpU cop, eFlagsRegU cr, rRegI dst, memory src) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
  ins_cost(250);
  format %{ "CMOV$cop $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
  ins_pipe( pipe_cmov_mem );
%}

instruct cmovI_memUCF(cmpOpUCF cop, eFlagsRegUCF cr, rRegI dst, memory src) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
  ins_cost(250);
  expand %{
    cmovI_memU(cop, cr, dst, src);
  %}
%}

// Conditional move
instruct cmovP_reg(eRegP dst, eRegP src, eFlagsReg cr, cmpOp cop ) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cop $dst,$src\t# ptr" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

// Conditional move (non-P6 version)
// Note:  a CMoveP is generated for  stubs and native wrappers
//        regardless of whether we are on a P6, so we
//        emulate a cmov here
instruct cmovP_reg_nonP6(eRegP dst, eRegP src, eFlagsReg cr, cmpOp cop ) %{
  match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  ins_cost(300);
  format %{ "Jn$cop   skip\n\t"
          "MOV    $dst,$src\t# pointer\n"
      "skip:" %}
  opcode(0x8b);
  ins_encode( enc_cmov_branch(cop, 0x2), OpcP, RegReg(dst, src));
  ins_pipe( pipe_cmov_reg );
%}

// Conditional move
instruct cmovP_regU(cmpOpU cop, eFlagsRegU cr, eRegP dst, eRegP src ) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cop $dst,$src\t# ptr" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovP_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegP dst, eRegP src ) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  expand %{
    cmovP_regU(cop, cr, dst, src);
  %}
%}

// DISABLED: Requires the ADLC to emit a bottom_type call that
// correctly meets the two pointer arguments; one is an incoming
// register but the other is a memory operand.  ALSO appears to
// be buggy with implicit null checks.
//
//// Conditional move
//instruct cmovP_mem(cmpOp cop, eFlagsReg cr, eRegP dst, memory src) %{
//  predicate(VM_Version::supports_cmov() );
//  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
//  ins_cost(250);
//  format %{ "CMOV$cop $dst,$src\t# ptr" %}
//  opcode(0x0F,0x40);
//  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
//  ins_pipe( pipe_cmov_mem );
//%}
//
//// Conditional move
//instruct cmovP_memU(cmpOpU cop, eFlagsRegU cr, eRegP dst, memory src) %{
//  predicate(VM_Version::supports_cmov() );
//  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
//  ins_cost(250);
//  format %{ "CMOV$cop $dst,$src\t# ptr" %}
//  opcode(0x0F,0x40);
//  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
//  ins_pipe( pipe_cmov_mem );
//%}

// Conditional move
instruct fcmovDPR_regU(cmpOp_fcmov cop, eFlagsRegU cr, regDPR1 dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "FCMOV$cop $dst,$src\t# double" %}
  opcode(0xDA);
  ins_encode( enc_cmov_dpr(cop,src) );
  ins_pipe( pipe_cmovDPR_reg );
%}

// Conditional move
instruct fcmovFPR_regU(cmpOp_fcmov cop, eFlagsRegU cr, regFPR1 dst, regFPR src) %{
  predicate(UseSSE==0);
  match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "FCMOV$cop $dst,$src\t# float" %}
  opcode(0xDA);
  ins_encode( enc_cmov_dpr(cop,src) );
  ins_pipe( pipe_cmovDPR_reg );
%}

// Float CMOV on Intel doesn't handle *signed* compares, only unsigned.
instruct fcmovDPR_regS(cmpOp cop, eFlagsReg cr, regDPR dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "Jn$cop   skip\n\t"
            "MOV    $dst,$src\t# double\n"
      "skip:" %}
  opcode (0xdd, 0x3);     /* DD D8+i or DD /3 */
  ins_encode( enc_cmov_branch( cop, 0x4 ), Push_Reg_DPR(src), OpcP, RegOpc(dst) );
  ins_pipe( pipe_cmovDPR_reg );
%}

// Float CMOV on Intel doesn't handle *signed* compares, only unsigned.
instruct fcmovFPR_regS(cmpOp cop, eFlagsReg cr, regFPR dst, regFPR src) %{
  predicate(UseSSE==0);
  match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "Jn$cop    skip\n\t"
            "MOV    $dst,$src\t# float\n"
      "skip:" %}
  opcode (0xdd, 0x3);     /* DD D8+i or DD /3 */
  ins_encode( enc_cmov_branch( cop, 0x4 ), Push_Reg_FPR(src), OpcP, RegOpc(dst) );
  ins_pipe( pipe_cmovDPR_reg );
%}

// No CMOVE with SSE/SSE2
instruct fcmovF_regS(cmpOp cop, eFlagsReg cr, regF dst, regF src) %{
  predicate (UseSSE>=1);
  match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "Jn$cop   skip\n\t"
            "MOVSS  $dst,$src\t# float\n"
      "skip:" %}
  ins_encode %{
    Label skip;
    // Invert sense of branch from sense of CMOV
    __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
    __ movflt($dst$$XMMRegister, $src$$XMMRegister);
    __ bind(skip);
  %}
  ins_pipe( pipe_slow );
%}

// No CMOVE with SSE/SSE2
instruct fcmovD_regS(cmpOp cop, eFlagsReg cr, regD dst, regD src) %{
  predicate (UseSSE>=2);
  match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "Jn$cop   skip\n\t"
            "MOVSD  $dst,$src\t# float\n"
      "skip:" %}
  ins_encode %{
    Label skip;
    // Invert sense of branch from sense of CMOV
    __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
    __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
    __ bind(skip);
  %}
  ins_pipe( pipe_slow );
%}

// unsigned version
instruct fcmovF_regU(cmpOpU cop, eFlagsRegU cr, regF dst, regF src) %{
  predicate (UseSSE>=1);
  match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "Jn$cop   skip\n\t"
            "MOVSS  $dst,$src\t# float\n"
      "skip:" %}
  ins_encode %{
    Label skip;
    // Invert sense of branch from sense of CMOV
    __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
    __ movflt($dst$$XMMRegister, $src$$XMMRegister);
    __ bind(skip);
  %}
  ins_pipe( pipe_slow );
%}

instruct fcmovF_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, regF dst, regF src) %{
  predicate (UseSSE>=1);
  match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovF_regU(cop, cr, dst, src);
  %}
%}

// unsigned version
instruct fcmovD_regU(cmpOpU cop, eFlagsRegU cr, regD dst, regD src) %{
  predicate (UseSSE>=2);
  match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "Jn$cop   skip\n\t"
            "MOVSD  $dst,$src\t# float\n"
      "skip:" %}
  ins_encode %{
    Label skip;
    // Invert sense of branch from sense of CMOV
    __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
    __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
    __ bind(skip);
  %}
  ins_pipe( pipe_slow );
%}

instruct fcmovD_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, regD dst, regD src) %{
  predicate (UseSSE>=2);
  match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovD_regU(cop, cr, dst, src);
  %}
%}

instruct cmovL_reg(cmpOp cop, eFlagsReg cr, eRegL dst, eRegL src) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cop $dst.lo,$src.lo\n\t"
            "CMOV$cop $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegReg_Lo2( dst, src ), enc_cmov(cop), RegReg_Hi2( dst, src ) );
  ins_pipe( pipe_cmov_reg_long );
%}

instruct cmovL_regU(cmpOpU cop, eFlagsRegU cr, eRegL dst, eRegL src) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cop $dst.lo,$src.lo\n\t"
            "CMOV$cop $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cop), RegReg_Lo2( dst, src ), enc_cmov(cop), RegReg_Hi2( dst, src ) );
  ins_pipe( pipe_cmov_reg_long );
%}

instruct cmovL_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegL dst, eRegL src) %{
  predicate(VM_Version::supports_cmov() );
  match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
  ins_cost(200);
  expand %{
    cmovL_regU(cop, cr, dst, src);
  %}
%}

//----------Arithmetic Instructions--------------------------------------------
//----------Addition Instructions----------------------------------------------

instruct addExactI_eReg(eAXRegI dst, rRegI src, eFlagsReg cr)
%{
  match(AddExactI dst src);
  effect(DEF cr);

  format %{ "ADD    $dst, $src\t# addExact int" %}
  ins_encode %{
    __ addl($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg_reg);
%}

instruct addExactI_eReg_imm(eAXRegI dst, immI src, eFlagsReg cr)
%{
  match(AddExactI dst src);
  effect(DEF cr);

  format %{ "ADD    $dst, $src\t# addExact int" %}
  ins_encode %{
    __ addl($dst$$Register, $src$$constant);
  %}
  ins_pipe(ialu_reg_reg);
%}

instruct addExactI_eReg_mem(eAXRegI dst, memory src, eFlagsReg cr)
%{
  match(AddExactI dst (LoadI src));
  effect(DEF cr);

  ins_cost(125);
  format %{ "ADD    $dst,$src\t# addExact int" %}
  ins_encode %{
    __ addl($dst$$Register, $src$$Address);
  %}
  ins_pipe( ialu_reg_mem );
%}


// Integer Addition Instructions
instruct addI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (AddI dst src));
  effect(KILL cr);

  size(2);
  format %{ "ADD    $dst,$src" %}
  opcode(0x03);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct addI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
  match(Set dst (AddI dst src));
  effect(KILL cr);

  format %{ "ADD    $dst,$src" %}
  opcode(0x81, 0x00); /* /0 id */
  ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
  ins_pipe( ialu_reg );
%}

instruct incI_eReg(rRegI dst, immI1 src, eFlagsReg cr) %{
  predicate(UseIncDec);
  match(Set dst (AddI dst src));
  effect(KILL cr);

  size(1);
  format %{ "INC    $dst" %}
  opcode(0x40); /*  */
  ins_encode( Opc_plus( primary, dst ) );
  ins_pipe( ialu_reg );
%}

instruct leaI_eReg_immI(rRegI dst, rRegI src0, immI src1) %{
  match(Set dst (AddI src0 src1));
  ins_cost(110);

  format %{ "LEA    $dst,[$src0 + $src1]" %}
  opcode(0x8D); /* 0x8D /r */
  ins_encode( OpcP, RegLea( dst, src0, src1 ) );
  ins_pipe( ialu_reg_reg );
%}

instruct leaP_eReg_immI(eRegP dst, eRegP src0, immI src1) %{
  match(Set dst (AddP src0 src1));
  ins_cost(110);

  format %{ "LEA    $dst,[$src0 + $src1]\t# ptr" %}
  opcode(0x8D); /* 0x8D /r */
  ins_encode( OpcP, RegLea( dst, src0, src1 ) );
  ins_pipe( ialu_reg_reg );
%}

instruct decI_eReg(rRegI dst, immI_M1 src, eFlagsReg cr) %{
  predicate(UseIncDec);
  match(Set dst (AddI dst src));
  effect(KILL cr);

  size(1);
  format %{ "DEC    $dst" %}
  opcode(0x48); /*  */
  ins_encode( Opc_plus( primary, dst ) );
  ins_pipe( ialu_reg );
%}

instruct addP_eReg(eRegP dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (AddP dst src));
  effect(KILL cr);

  size(2);
  format %{ "ADD    $dst,$src" %}
  opcode(0x03);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct addP_eReg_imm(eRegP dst, immI src, eFlagsReg cr) %{
  match(Set dst (AddP dst src));
  effect(KILL cr);

  format %{ "ADD    $dst,$src" %}
  opcode(0x81,0x00); /* Opcode 81 /0 id */
  // ins_encode( RegImm( dst, src) );
  ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
  ins_pipe( ialu_reg );
%}

instruct addI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
  match(Set dst (AddI dst (LoadI src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "ADD    $dst,$src" %}
  opcode(0x03);
  ins_encode( OpcP, RegMem( dst, src) );
  ins_pipe( ialu_reg_mem );
%}

instruct addI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(150);
  format %{ "ADD    $dst,$src" %}
  opcode(0x01);  /* Opcode 01 /r */
  ins_encode( OpcP, RegMem( src, dst ) );
  ins_pipe( ialu_mem_reg );
%}

// Add Memory with Immediate
instruct addI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "ADD    $dst,$src" %}
  opcode(0x81);               /* Opcode 81 /0 id */
  ins_encode( OpcSE( src ), RMopc_Mem(0x00,dst), Con8or32( src ) );
  ins_pipe( ialu_mem_imm );
%}

instruct incI_mem(memory dst, immI1 src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "INC    $dst" %}
  opcode(0xFF);               /* Opcode FF /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,dst));
  ins_pipe( ialu_mem_imm );
%}

instruct decI_mem(memory dst, immI_M1 src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "DEC    $dst" %}
  opcode(0xFF);               /* Opcode FF /1 */
  ins_encode( OpcP, RMopc_Mem(0x01,dst));
  ins_pipe( ialu_mem_imm );
%}


instruct checkCastPP( eRegP dst ) %{
  match(Set dst (CheckCastPP dst));

  size(0);
  format %{ "#checkcastPP of $dst" %}
  ins_encode( /*empty encoding*/ );
  ins_pipe( empty );
%}

instruct castPP( eRegP dst ) %{
  match(Set dst (CastPP dst));
  format %{ "#castPP of $dst" %}
  ins_encode( /*empty encoding*/ );
  ins_pipe( empty );
%}

instruct castII( rRegI dst ) %{
  match(Set dst (CastII dst));
  format %{ "#castII of $dst" %}
  ins_encode( /*empty encoding*/ );
  ins_cost(0);
  ins_pipe( empty );
%}


// Load-locked - same as a regular pointer load when used with compare-swap
instruct loadPLocked(eRegP dst, memory mem) %{
  match(Set dst (LoadPLocked mem));

  ins_cost(125);
  format %{ "MOV    $dst,$mem\t# Load ptr. locked" %}
  opcode(0x8B);
  ins_encode( OpcP, RegMem(dst,mem));
  ins_pipe( ialu_reg_mem );
%}

// Conditional-store of the updated heap-top.
// Used during allocation of the shared heap.
// Sets flags (EQ) on success.  Implemented with a CMPXCHG on Intel.
instruct storePConditional( memory heap_top_ptr, eAXRegP oldval, eRegP newval, eFlagsReg cr ) %{
  match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
  // EAX is killed if there is contention, but then it's also unused.
  // In the common case of no contention, EAX holds the new oop address.
  format %{ "CMPXCHG $heap_top_ptr,$newval\t# If EAX==$heap_top_ptr Then store $newval into $heap_top_ptr" %}
  ins_encode( lock_prefix, Opcode(0x0F), Opcode(0xB1), RegMem(newval,heap_top_ptr) );
  ins_pipe( pipe_cmpxchg );
%}

// Conditional-store of an int value.
// ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG on Intel.
instruct storeIConditional( memory mem, eAXRegI oldval, rRegI newval, eFlagsReg cr ) %{
  match(Set cr (StoreIConditional mem (Binary oldval newval)));
  effect(KILL oldval);
  format %{ "CMPXCHG $mem,$newval\t# If EAX==$mem Then store $newval into $mem" %}
  ins_encode( lock_prefix, Opcode(0x0F), Opcode(0xB1), RegMem(newval, mem) );
  ins_pipe( pipe_cmpxchg );
%}

// Conditional-store of a long value.
// ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG8 on Intel.
instruct storeLConditional( memory mem, eADXRegL oldval, eBCXRegL newval, eFlagsReg cr ) %{
  match(Set cr (StoreLConditional mem (Binary oldval newval)));
  effect(KILL oldval);
  format %{ "XCHG   EBX,ECX\t# correct order for CMPXCHG8 instruction\n\t"
            "CMPXCHG8 $mem,ECX:EBX\t# If EDX:EAX==$mem Then store ECX:EBX into $mem\n\t"
            "XCHG   EBX,ECX"
  %}
  ins_encode %{
    // Note: we need to swap rbx, and rcx before and after the
    //       cmpxchg8 instruction because the instruction uses
    //       rcx as the high order word of the new value to store but
    //       our register encoding uses rbx.
    __ xchgl(as_Register(EBX_enc), as_Register(ECX_enc));
    if( os::is_MP() )
      __ lock();
    __ cmpxchg8($mem$$Address);
    __ xchgl(as_Register(EBX_enc), as_Register(ECX_enc));
  %}
  ins_pipe( pipe_cmpxchg );
%}

// No flag versions for CompareAndSwap{P,I,L} because matcher can't match them

instruct compareAndSwapL( rRegI res, eSIRegP mem_ptr, eADXRegL oldval, eBCXRegL newval, eFlagsReg cr ) %{
  predicate(VM_Version::supports_cx8());
  match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
  effect(KILL cr, KILL oldval);
  format %{ "CMPXCHG8 [$mem_ptr],$newval\t# If EDX:EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
            "MOV    $res,0\n\t"
            "JNE,s  fail\n\t"
            "MOV    $res,1\n"
          "fail:" %}
  ins_encode( enc_cmpxchg8(mem_ptr),
              enc_flags_ne_to_boolean(res) );
  ins_pipe( pipe_cmpxchg );
%}

instruct compareAndSwapP( rRegI res,  pRegP mem_ptr, eAXRegP oldval, eCXRegP newval, eFlagsReg cr) %{
  match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
  effect(KILL cr, KILL oldval);
  format %{ "CMPXCHG [$mem_ptr],$newval\t# If EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
            "MOV    $res,0\n\t"
            "JNE,s  fail\n\t"
            "MOV    $res,1\n"
          "fail:" %}
  ins_encode( enc_cmpxchg(mem_ptr), enc_flags_ne_to_boolean(res) );
  ins_pipe( pipe_cmpxchg );
%}

instruct compareAndSwapI( rRegI res, pRegP mem_ptr, eAXRegI oldval, eCXRegI newval, eFlagsReg cr) %{
  match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
  effect(KILL cr, KILL oldval);
  format %{ "CMPXCHG [$mem_ptr],$newval\t# If EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
            "MOV    $res,0\n\t"
            "JNE,s  fail\n\t"
            "MOV    $res,1\n"
          "fail:" %}
  ins_encode( enc_cmpxchg(mem_ptr), enc_flags_ne_to_boolean(res) );
  ins_pipe( pipe_cmpxchg );
%}

instruct xaddI_no_res( memory mem, Universe dummy, immI add, eFlagsReg cr) %{
  predicate(n->as_LoadStore()->result_not_used());
  match(Set dummy (GetAndAddI mem add));
  effect(KILL cr);
  format %{ "ADDL  [$mem],$add" %}
  ins_encode %{
    if (os::is_MP()) { __ lock(); }
    __ addl($mem$$Address, $add$$constant);
  %}
  ins_pipe( pipe_cmpxchg );
%}

instruct xaddI( memory mem, rRegI newval, eFlagsReg cr) %{
  match(Set newval (GetAndAddI mem newval));
  effect(KILL cr);
  format %{ "XADDL  [$mem],$newval" %}
  ins_encode %{
    if (os::is_MP()) { __ lock(); }
    __ xaddl($mem$$Address, $newval$$Register);
  %}
  ins_pipe( pipe_cmpxchg );
%}

instruct xchgI( memory mem, rRegI newval) %{
  match(Set newval (GetAndSetI mem newval));
  format %{ "XCHGL  $newval,[$mem]" %}
  ins_encode %{
    __ xchgl($newval$$Register, $mem$$Address);
  %}
  ins_pipe( pipe_cmpxchg );
%}

instruct xchgP( memory mem, pRegP newval) %{
  match(Set newval (GetAndSetP mem newval));
  format %{ "XCHGL  $newval,[$mem]" %}
  ins_encode %{
    __ xchgl($newval$$Register, $mem$$Address);
  %}
  ins_pipe( pipe_cmpxchg );
%}

//----------Subtraction Instructions-------------------------------------------

instruct subExactI_eReg(eAXRegI dst, rRegI src, eFlagsReg cr)
%{
  match(SubExactI dst src);
  effect(DEF cr);

  format %{ "SUB    $dst, $src\t# subExact int" %}
  ins_encode %{
    __ subl($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg_reg);
%}

instruct subExactI_eReg_imm(eAXRegI dst, immI src, eFlagsReg cr)
%{
  match(SubExactI dst src);
  effect(DEF cr);

  format %{ "SUB    $dst, $src\t# subExact int" %}
  ins_encode %{
    __ subl($dst$$Register, $src$$constant);
  %}
  ins_pipe(ialu_reg_reg);
%}

instruct subExactI_eReg_mem(eAXRegI dst, memory src, eFlagsReg cr)
%{
  match(SubExactI dst (LoadI src));
  effect(DEF cr);

  ins_cost(125);
  format %{ "SUB    $dst,$src\t# subExact int" %}
  ins_encode %{
    __ subl($dst$$Register, $src$$Address);
  %}
  ins_pipe( ialu_reg_mem );
%}

// Integer Subtraction Instructions
instruct subI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (SubI dst src));
  effect(KILL cr);

  size(2);
  format %{ "SUB    $dst,$src" %}
  opcode(0x2B);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct subI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
  match(Set dst (SubI dst src));
  effect(KILL cr);

  format %{ "SUB    $dst,$src" %}
  opcode(0x81,0x05);  /* Opcode 81 /5 */
  // ins_encode( RegImm( dst, src) );
  ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
  ins_pipe( ialu_reg );
%}

instruct subI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
  match(Set dst (SubI dst (LoadI src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "SUB    $dst,$src" %}
  opcode(0x2B);
  ins_encode( OpcP, RegMem( dst, src) );
  ins_pipe( ialu_reg_mem );
%}

instruct subI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (SubI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(150);
  format %{ "SUB    $dst,$src" %}
  opcode(0x29);  /* Opcode 29 /r */
  ins_encode( OpcP, RegMem( src, dst ) );
  ins_pipe( ialu_mem_reg );
%}

// Subtract from a pointer
instruct subP_eReg(eRegP dst, rRegI src, immI0 zero, eFlagsReg cr) %{
  match(Set dst (AddP dst (SubI zero src)));
  effect(KILL cr);

  size(2);
  format %{ "SUB    $dst,$src" %}
  opcode(0x2B);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct negI_eReg(rRegI dst, immI0 zero, eFlagsReg cr) %{
  match(Set dst (SubI zero dst));
  effect(KILL cr);

  size(2);
  format %{ "NEG    $dst" %}
  opcode(0xF7,0x03);  // Opcode F7 /3
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg );
%}

instruct negExactI_eReg(eAXRegI dst, eFlagsReg cr) %{
  match(NegExactI dst);
  effect(DEF cr);

  format %{ "NEG    $dst\t# negExact int"%}
  ins_encode %{
    __ negl($dst$$Register);
  %}
  ins_pipe(ialu_reg);
%}

//----------Multiplication/Division Instructions-------------------------------
// Integer Multiplication Instructions
// Multiply Register
instruct mulI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (MulI dst src));
  effect(KILL cr);

  size(3);
  ins_cost(300);
  format %{ "IMUL   $dst,$src" %}
  opcode(0xAF, 0x0F);
  ins_encode( OpcS, OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg_alu0 );
%}

// Multiply 32-bit Immediate
instruct mulI_eReg_imm(rRegI dst, rRegI src, immI imm, eFlagsReg cr) %{
  match(Set dst (MulI src imm));
  effect(KILL cr);

  ins_cost(300);
  format %{ "IMUL   $dst,$src,$imm" %}
  opcode(0x69);  /* 69 /r id */
  ins_encode( OpcSE(imm), RegReg( dst, src ), Con8or32( imm ) );
  ins_pipe( ialu_reg_reg_alu0 );
%}

instruct loadConL_low_only(eADXRegL_low_only dst, immL32 src, eFlagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);

  // Note that this is artificially increased to make it more expensive than loadConL
  ins_cost(250);
  format %{ "MOV    EAX,$src\t// low word only" %}
  opcode(0xB8);
  ins_encode( LdImmL_Lo(dst, src) );
  ins_pipe( ialu_reg_fat );
%}

// Multiply by 32-bit Immediate, taking the shifted high order results
//  (special case for shift by 32)
instruct mulI_imm_high(eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32 cnt, eFlagsReg cr) %{
  match(Set dst (ConvL2I (RShiftL (MulL (ConvI2L src1) src2) cnt)));
  predicate( _kids[0]->_kids[0]->_kids[1]->_leaf->Opcode() == Op_ConL &&
             _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() >= min_jint &&
             _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() <= max_jint );
  effect(USE src1, KILL cr);

  // Note that this is adjusted by 150 to compensate for the overcosting of loadConL_low_only
  ins_cost(0*100 + 1*400 - 150);
  format %{ "IMUL   EDX:EAX,$src1" %}
  ins_encode( multiply_con_and_shift_high( dst, src1, src2, cnt, cr ) );
  ins_pipe( pipe_slow );
%}

// Multiply by 32-bit Immediate, taking the shifted high order results
instruct mulI_imm_RShift_high(eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32_63 cnt, eFlagsReg cr) %{
  match(Set dst (ConvL2I (RShiftL (MulL (ConvI2L src1) src2) cnt)));
  predicate( _kids[0]->_kids[0]->_kids[1]->_leaf->Opcode() == Op_ConL &&
             _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() >= min_jint &&
             _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() <= max_jint );
  effect(USE src1, KILL cr);

  // Note that this is adjusted by 150 to compensate for the overcosting of loadConL_low_only
  ins_cost(1*100 + 1*400 - 150);
  format %{ "IMUL   EDX:EAX,$src1\n\t"
            "SAR    EDX,$cnt-32" %}
  ins_encode( multiply_con_and_shift_high( dst, src1, src2, cnt, cr ) );
  ins_pipe( pipe_slow );
%}

// Multiply Memory 32-bit Immediate
instruct mulI_mem_imm(rRegI dst, memory src, immI imm, eFlagsReg cr) %{
  match(Set dst (MulI (LoadI src) imm));
  effect(KILL cr);

  ins_cost(300);
  format %{ "IMUL   $dst,$src,$imm" %}
  opcode(0x69);  /* 69 /r id */
  ins_encode( OpcSE(imm), RegMem( dst, src ), Con8or32( imm ) );
  ins_pipe( ialu_reg_mem_alu0 );
%}

// Multiply Memory
instruct mulI(rRegI dst, memory src, eFlagsReg cr) %{
  match(Set dst (MulI dst (LoadI src)));
  effect(KILL cr);

  ins_cost(350);
  format %{ "IMUL   $dst,$src" %}
  opcode(0xAF, 0x0F);
  ins_encode( OpcS, OpcP, RegMem( dst, src) );
  ins_pipe( ialu_reg_mem_alu0 );
%}

// Multiply Register Int to Long
instruct mulI2L(eADXRegL dst, eAXRegI src, nadxRegI src1, eFlagsReg flags) %{
  // Basic Idea: long = (long)int * (long)int
  match(Set dst (MulL (ConvI2L src) (ConvI2L src1)));
  effect(DEF dst, USE src, USE src1, KILL flags);

  ins_cost(300);
  format %{ "IMUL   $dst,$src1" %}

  ins_encode( long_int_multiply( dst, src1 ) );
  ins_pipe( ialu_reg_reg_alu0 );
%}

instruct mulIS_eReg(eADXRegL dst, immL_32bits mask, eFlagsReg flags, eAXRegI src, nadxRegI src1) %{
  // Basic Idea:  long = (int & 0xffffffffL) * (int & 0xffffffffL)
  match(Set dst (MulL (AndL (ConvI2L src) mask) (AndL (ConvI2L src1) mask)));
  effect(KILL flags);

  ins_cost(300);
  format %{ "MUL    $dst,$src1" %}

  ins_encode( long_uint_multiply(dst, src1) );
  ins_pipe( ialu_reg_reg_alu0 );
%}

// Multiply Register Long
instruct mulL_eReg(eADXRegL dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
  match(Set dst (MulL dst src));
  effect(KILL cr, TEMP tmp);
  ins_cost(4*100+3*400);
// Basic idea: lo(result) = lo(x_lo * y_lo)
//             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
  format %{ "MOV    $tmp,$src.lo\n\t"
            "IMUL   $tmp,EDX\n\t"
            "MOV    EDX,$src.hi\n\t"
            "IMUL   EDX,EAX\n\t"
            "ADD    $tmp,EDX\n\t"
            "MUL    EDX:EAX,$src.lo\n\t"
            "ADD    EDX,$tmp" %}
  ins_encode( long_multiply( dst, src, tmp ) );
  ins_pipe( pipe_slow );
%}

// Multiply Register Long where the left operand's high 32 bits are zero
instruct mulL_eReg_lhi0(eADXRegL dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
  predicate(is_operand_hi32_zero(n->in(1)));
  match(Set dst (MulL dst src));
  effect(KILL cr, TEMP tmp);
  ins_cost(2*100+2*400);
// Basic idea: lo(result) = lo(x_lo * y_lo)
//             hi(result) = hi(x_lo * y_lo) + lo(x_lo * y_hi) where lo(x_hi * y_lo) = 0 because x_hi = 0
  format %{ "MOV    $tmp,$src.hi\n\t"
            "IMUL   $tmp,EAX\n\t"
            "MUL    EDX:EAX,$src.lo\n\t"
            "ADD    EDX,$tmp" %}
  ins_encode %{
    __ movl($tmp$$Register, HIGH_FROM_LOW($src$$Register));
    __ imull($tmp$$Register, rax);
    __ mull($src$$Register);
    __ addl(rdx, $tmp$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// Multiply Register Long where the right operand's high 32 bits are zero
instruct mulL_eReg_rhi0(eADXRegL dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
  predicate(is_operand_hi32_zero(n->in(2)));
  match(Set dst (MulL dst src));
  effect(KILL cr, TEMP tmp);
  ins_cost(2*100+2*400);
// Basic idea: lo(result) = lo(x_lo * y_lo)
//             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) where lo(x_lo * y_hi) = 0 because y_hi = 0
  format %{ "MOV    $tmp,$src.lo\n\t"
            "IMUL   $tmp,EDX\n\t"
            "MUL    EDX:EAX,$src.lo\n\t"
            "ADD    EDX,$tmp" %}
  ins_encode %{
    __ movl($tmp$$Register, $src$$Register);
    __ imull($tmp$$Register, rdx);
    __ mull($src$$Register);
    __ addl(rdx, $tmp$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// Multiply Register Long where the left and the right operands' high 32 bits are zero
instruct mulL_eReg_hi0(eADXRegL dst, eRegL src, eFlagsReg cr) %{
  predicate(is_operand_hi32_zero(n->in(1)) && is_operand_hi32_zero(n->in(2)));
  match(Set dst (MulL dst src));
  effect(KILL cr);
  ins_cost(1*400);
// Basic idea: lo(result) = lo(x_lo * y_lo)
//             hi(result) = hi(x_lo * y_lo) where lo(x_hi * y_lo) = 0 and lo(x_lo * y_hi) = 0 because x_hi = 0 and y_hi = 0
  format %{ "MUL    EDX:EAX,$src.lo\n\t" %}
  ins_encode %{
    __ mull($src$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// Multiply Register Long by small constant
instruct mulL_eReg_con(eADXRegL dst, immL_127 src, rRegI tmp, eFlagsReg cr) %{
  match(Set dst (MulL dst src));
  effect(KILL cr, TEMP tmp);
  ins_cost(2*100+2*400);
  size(12);
// Basic idea: lo(result) = lo(src * EAX)
//             hi(result) = hi(src * EAX) + lo(src * EDX)
  format %{ "IMUL   $tmp,EDX,$src\n\t"
            "MOV    EDX,$src\n\t"
            "MUL    EDX\t# EDX*EAX -> EDX:EAX\n\t"
            "ADD    EDX,$tmp" %}
  ins_encode( long_multiply_con( dst, src, tmp ) );
  ins_pipe( pipe_slow );
%}

instruct mulExactI_eReg(eAXRegI dst, rRegI src, eFlagsReg cr)
%{
  match(MulExactI dst src);
  effect(DEF cr);

  ins_cost(300);
  format %{ "IMUL   $dst, $src\t# mulExact int" %}
  ins_encode %{
    __ imull($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg_reg_alu0);
%}

instruct mulExactI_eReg_imm(eAXRegI dst, rRegI src, immI imm, eFlagsReg cr)
%{
  match(MulExactI src imm);
  effect(DEF cr);

  ins_cost(300);
  format %{ "IMUL   $dst, $src, $imm\t# mulExact int" %}
  ins_encode %{
    __ imull($dst$$Register, $src$$Register, $imm$$constant);
  %}
  ins_pipe(ialu_reg_reg_alu0);
%}

instruct mulExactI_eReg_mem(eAXRegI dst, memory src, eFlagsReg cr)
%{
  match(MulExactI dst (LoadI src));
  effect(DEF cr);

  ins_cost(350);
  format %{ "IMUL   $dst, $src\t# mulExact int" %}
  ins_encode %{
    __ imull($dst$$Register, $src$$Address);
  %}
  ins_pipe(ialu_reg_mem_alu0);
%}


// Integer DIV with Register
instruct divI_eReg(eAXRegI rax, eDXRegI rdx, eCXRegI div, eFlagsReg cr) %{
  match(Set rax (DivI rax div));
  effect(KILL rdx, KILL cr);
  size(26);
  ins_cost(30*100+10*100);
  format %{ "CMP    EAX,0x80000000\n\t"
            "JNE,s  normal\n\t"
            "XOR    EDX,EDX\n\t"
            "CMP    ECX,-1\n\t"
            "JE,s   done\n"
    "normal: CDQ\n\t"
            "IDIV   $div\n\t"
    "done:"        %}
  opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  ins_encode( cdq_enc, OpcP, RegOpc(div) );
  ins_pipe( ialu_reg_reg_alu0 );
%}

// Divide Register Long
instruct divL_eReg( eADXRegL dst, eRegL src1, eRegL src2, eFlagsReg cr, eCXRegI cx, eBXRegI bx ) %{
  match(Set dst (DivL src1 src2));
  effect( KILL cr, KILL cx, KILL bx );
  ins_cost(10000);
  format %{ "PUSH   $src1.hi\n\t"
            "PUSH   $src1.lo\n\t"
            "PUSH   $src2.hi\n\t"
            "PUSH   $src2.lo\n\t"
            "CALL   SharedRuntime::ldiv\n\t"
            "ADD    ESP,16" %}
  ins_encode( long_div(src1,src2) );
  ins_pipe( pipe_slow );
%}

// Integer DIVMOD with Register, both quotient and mod results
instruct divModI_eReg_divmod(eAXRegI rax, eDXRegI rdx, eCXRegI div, eFlagsReg cr) %{
  match(DivModI rax div);
  effect(KILL cr);
  size(26);
  ins_cost(30*100+10*100);
  format %{ "CMP    EAX,0x80000000\n\t"
            "JNE,s  normal\n\t"
            "XOR    EDX,EDX\n\t"
            "CMP    ECX,-1\n\t"
            "JE,s   done\n"
    "normal: CDQ\n\t"
            "IDIV   $div\n\t"
    "done:"        %}
  opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  ins_encode( cdq_enc, OpcP, RegOpc(div) );
  ins_pipe( pipe_slow );
%}

// Integer MOD with Register
instruct modI_eReg(eDXRegI rdx, eAXRegI rax, eCXRegI div, eFlagsReg cr) %{
  match(Set rdx (ModI rax div));
  effect(KILL rax, KILL cr);

  size(26);
  ins_cost(300);
  format %{ "CDQ\n\t"
            "IDIV   $div" %}
  opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  ins_encode( cdq_enc, OpcP, RegOpc(div) );
  ins_pipe( ialu_reg_reg_alu0 );
%}

// Remainder Register Long
instruct modL_eReg( eADXRegL dst, eRegL src1, eRegL src2, eFlagsReg cr, eCXRegI cx, eBXRegI bx ) %{
  match(Set dst (ModL src1 src2));
  effect( KILL cr, KILL cx, KILL bx );
  ins_cost(10000);
  format %{ "PUSH   $src1.hi\n\t"
            "PUSH   $src1.lo\n\t"
            "PUSH   $src2.hi\n\t"
            "PUSH   $src2.lo\n\t"
            "CALL   SharedRuntime::lrem\n\t"
            "ADD    ESP,16" %}
  ins_encode( long_mod(src1,src2) );
  ins_pipe( pipe_slow );
%}

// Divide Register Long (no special case since divisor != -1)
instruct divL_eReg_imm32( eADXRegL dst, immL32 imm, rRegI tmp, rRegI tmp2, eFlagsReg cr ) %{
  match(Set dst (DivL dst imm));
  effect( TEMP tmp, TEMP tmp2, KILL cr );
  ins_cost(1000);
  format %{ "MOV    $tmp,abs($imm) # ldiv EDX:EAX,$imm\n\t"
            "XOR    $tmp2,$tmp2\n\t"
            "CMP    $tmp,EDX\n\t"
            "JA,s   fast\n\t"
            "MOV    $tmp2,EAX\n\t"
            "MOV    EAX,EDX\n\t"
            "MOV    EDX,0\n\t"
            "JLE,s  pos\n\t"
            "LNEG   EAX : $tmp2\n\t"
            "DIV    $tmp # unsigned division\n\t"
            "XCHG   EAX,$tmp2\n\t"
            "DIV    $tmp\n\t"
            "LNEG   $tmp2 : EAX\n\t"
            "JMP,s  done\n"
    "pos:\n\t"
            "DIV    $tmp\n\t"
            "XCHG   EAX,$tmp2\n"
    "fast:\n\t"
            "DIV    $tmp\n"
    "done:\n\t"
            "MOV    EDX,$tmp2\n\t"
            "NEG    EDX:EAX # if $imm < 0" %}
  ins_encode %{
    int con = (int)$imm$$constant;
    assert(con != 0 && con != -1 && con != min_jint, "wrong divisor");
    int pcon = (con > 0) ? con : -con;
    Label Lfast, Lpos, Ldone;

    __ movl($tmp$$Register, pcon);
    __ xorl($tmp2$$Register,$tmp2$$Register);
    __ cmpl($tmp$$Register, HIGH_FROM_LOW($dst$$Register));
    __ jccb(Assembler::above, Lfast); // result fits into 32 bit

    __ movl($tmp2$$Register, $dst$$Register); // save
    __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
    __ movl(HIGH_FROM_LOW($dst$$Register),0); // preserve flags
    __ jccb(Assembler::lessEqual, Lpos); // result is positive

    // Negative dividend.
    // convert value to positive to use unsigned division
    __ lneg($dst$$Register, $tmp2$$Register);
    __ divl($tmp$$Register);
    __ xchgl($dst$$Register, $tmp2$$Register);
    __ divl($tmp$$Register);
    // revert result back to negative
    __ lneg($tmp2$$Register, $dst$$Register);
    __ jmpb(Ldone);

    __ bind(Lpos);
    __ divl($tmp$$Register); // Use unsigned division
    __ xchgl($dst$$Register, $tmp2$$Register);
    // Fallthrow for final divide, tmp2 has 32 bit hi result

    __ bind(Lfast);
    // fast path: src is positive
    __ divl($tmp$$Register); // Use unsigned division

    __ bind(Ldone);
    __ movl(HIGH_FROM_LOW($dst$$Register),$tmp2$$Register);
    if (con < 0) {
      __ lneg(HIGH_FROM_LOW($dst$$Register), $dst$$Register);
    }
  %}
  ins_pipe( pipe_slow );
%}

// Remainder Register Long (remainder fit into 32 bits)
instruct modL_eReg_imm32( eADXRegL dst, immL32 imm, rRegI tmp, rRegI tmp2, eFlagsReg cr ) %{
  match(Set dst (ModL dst imm));
  effect( TEMP tmp, TEMP tmp2, KILL cr );
  ins_cost(1000);
  format %{ "MOV    $tmp,abs($imm) # lrem EDX:EAX,$imm\n\t"
            "CMP    $tmp,EDX\n\t"
            "JA,s   fast\n\t"
            "MOV    $tmp2,EAX\n\t"
            "MOV    EAX,EDX\n\t"
            "MOV    EDX,0\n\t"
            "JLE,s  pos\n\t"
            "LNEG   EAX : $tmp2\n\t"
            "DIV    $tmp # unsigned division\n\t"
            "MOV    EAX,$tmp2\n\t"
            "DIV    $tmp\n\t"
            "NEG    EDX\n\t"
            "JMP,s  done\n"
    "pos:\n\t"
            "DIV    $tmp\n\t"
            "MOV    EAX,$tmp2\n"
    "fast:\n\t"
            "DIV    $tmp\n"
    "done:\n\t"
            "MOV    EAX,EDX\n\t"
            "SAR    EDX,31\n\t" %}
  ins_encode %{
    int con = (int)$imm$$constant;
    assert(con != 0 && con != -1 && con != min_jint, "wrong divisor");
    int pcon = (con > 0) ? con : -con;
    Label  Lfast, Lpos, Ldone;

    __ movl($tmp$$Register, pcon);
    __ cmpl($tmp$$Register, HIGH_FROM_LOW($dst$$Register));
    __ jccb(Assembler::above, Lfast); // src is positive and result fits into 32 bit

    __ movl($tmp2$$Register, $dst$$Register); // save
    __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
    __ movl(HIGH_FROM_LOW($dst$$Register),0); // preserve flags
    __ jccb(Assembler::lessEqual, Lpos); // result is positive

    // Negative dividend.
    // convert value to positive to use unsigned division
    __ lneg($dst$$Register, $tmp2$$Register);
    __ divl($tmp$$Register);
    __ movl($dst$$Register, $tmp2$$Register);
    __ divl($tmp$$Register);
    // revert remainder back to negative
    __ negl(HIGH_FROM_LOW($dst$$Register));
    __ jmpb(Ldone);

    __ bind(Lpos);
    __ divl($tmp$$Register);
    __ movl($dst$$Register, $tmp2$$Register);

    __ bind(Lfast);
    // fast path: src is positive
    __ divl($tmp$$Register);

    __ bind(Ldone);
    __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
    __ sarl(HIGH_FROM_LOW($dst$$Register), 31); // result sign

  %}
  ins_pipe( pipe_slow );
%}

// Integer Shift Instructions
// Shift Left by one
instruct shlI_eReg_1(rRegI dst, immI1 shift, eFlagsReg cr) %{
  match(Set dst (LShiftI dst shift));
  effect(KILL cr);

  size(2);
  format %{ "SHL    $dst,$shift" %}
  opcode(0xD1, 0x4);  /* D1 /4 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg );
%}

// Shift Left by 8-bit immediate
instruct salI_eReg_imm(rRegI dst, immI8 shift, eFlagsReg cr) %{
  match(Set dst (LShiftI dst shift));
  effect(KILL cr);

  size(3);
  format %{ "SHL    $dst,$shift" %}
  opcode(0xC1, 0x4);  /* C1 /4 ib */
  ins_encode( RegOpcImm( dst, shift) );
  ins_pipe( ialu_reg );
%}

// Shift Left by variable
instruct salI_eReg_CL(rRegI dst, eCXRegI shift, eFlagsReg cr) %{
  match(Set dst (LShiftI dst shift));
  effect(KILL cr);

  size(2);
  format %{ "SHL    $dst,$shift" %}
  opcode(0xD3, 0x4);  /* D3 /4 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg_reg );
%}

// Arithmetic shift right by one
instruct sarI_eReg_1(rRegI dst, immI1 shift, eFlagsReg cr) %{
  match(Set dst (RShiftI dst shift));
  effect(KILL cr);

  size(2);
  format %{ "SAR    $dst,$shift" %}
  opcode(0xD1, 0x7);  /* D1 /7 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg );
%}

// Arithmetic shift right by one
instruct sarI_mem_1(memory dst, immI1 shift, eFlagsReg cr) %{
  match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
  effect(KILL cr);
  format %{ "SAR    $dst,$shift" %}
  opcode(0xD1, 0x7);  /* D1 /7 */
  ins_encode( OpcP, RMopc_Mem(secondary,dst) );
  ins_pipe( ialu_mem_imm );
%}

// Arithmetic Shift Right by 8-bit immediate
instruct sarI_eReg_imm(rRegI dst, immI8 shift, eFlagsReg cr) %{
  match(Set dst (RShiftI dst shift));
  effect(KILL cr);

  size(3);
  format %{ "SAR    $dst,$shift" %}
  opcode(0xC1, 0x7);  /* C1 /7 ib */
  ins_encode( RegOpcImm( dst, shift ) );
  ins_pipe( ialu_mem_imm );
%}

// Arithmetic Shift Right by 8-bit immediate
instruct sarI_mem_imm(memory dst, immI8 shift, eFlagsReg cr) %{
  match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
  effect(KILL cr);

  format %{ "SAR    $dst,$shift" %}
  opcode(0xC1, 0x7);  /* C1 /7 ib */
  ins_encode( OpcP, RMopc_Mem(secondary, dst ), Con8or32( shift ) );
  ins_pipe( ialu_mem_imm );
%}

// Arithmetic Shift Right by variable
instruct sarI_eReg_CL(rRegI dst, eCXRegI shift, eFlagsReg cr) %{
  match(Set dst (RShiftI dst shift));
  effect(KILL cr);

  size(2);
  format %{ "SAR    $dst,$shift" %}
  opcode(0xD3, 0x7);  /* D3 /7 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg_reg );
%}

// Logical shift right by one
instruct shrI_eReg_1(rRegI dst, immI1 shift, eFlagsReg cr) %{
  match(Set dst (URShiftI dst shift));
  effect(KILL cr);

  size(2);
  format %{ "SHR    $dst,$shift" %}
  opcode(0xD1, 0x5);  /* D1 /5 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg );
%}

// Logical Shift Right by 8-bit immediate
instruct shrI_eReg_imm(rRegI dst, immI8 shift, eFlagsReg cr) %{
  match(Set dst (URShiftI dst shift));
  effect(KILL cr);

  size(3);
  format %{ "SHR    $dst,$shift" %}
  opcode(0xC1, 0x5);  /* C1 /5 ib */
  ins_encode( RegOpcImm( dst, shift) );
  ins_pipe( ialu_reg );
%}


// Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
// This idiom is used by the compiler for the i2b bytecode.
instruct i2b(rRegI dst, xRegI src, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));

  size(3);
  format %{ "MOVSX  $dst,$src :8" %}
  ins_encode %{
    __ movsbl($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg_reg);
%}

// Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
// This idiom is used by the compiler the i2s bytecode.
instruct i2s(rRegI dst, xRegI src, immI_16 sixteen) %{
  match(Set dst (RShiftI (LShiftI src sixteen) sixteen));

  size(3);
  format %{ "MOVSX  $dst,$src :16" %}
  ins_encode %{
    __ movswl($dst$$Register, $src$$Register);
  %}
  ins_pipe(ialu_reg_reg);
%}


// Logical Shift Right by variable
instruct shrI_eReg_CL(rRegI dst, eCXRegI shift, eFlagsReg cr) %{
  match(Set dst (URShiftI dst shift));
  effect(KILL cr);

  size(2);
  format %{ "SHR    $dst,$shift" %}
  opcode(0xD3, 0x5);  /* D3 /5 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg_reg );
%}


//----------Logical Instructions-----------------------------------------------
//----------Integer Logical Instructions---------------------------------------
// And Instructions
// And Register with Register
instruct andI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (AndI dst src));
  effect(KILL cr);

  size(2);
  format %{ "AND    $dst,$src" %}
  opcode(0x23);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

// And Register with Immediate
instruct andI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
  match(Set dst (AndI dst src));
  effect(KILL cr);

  format %{ "AND    $dst,$src" %}
  opcode(0x81,0x04);  /* Opcode 81 /4 */
  // ins_encode( RegImm( dst, src) );
  ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
  ins_pipe( ialu_reg );
%}

// And Register with Memory
instruct andI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
  match(Set dst (AndI dst (LoadI src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "AND    $dst,$src" %}
  opcode(0x23);
  ins_encode( OpcP, RegMem( dst, src) );
  ins_pipe( ialu_reg_mem );
%}

// And Memory with Register
instruct andI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (AndI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(150);
  format %{ "AND    $dst,$src" %}
  opcode(0x21);  /* Opcode 21 /r */
  ins_encode( OpcP, RegMem( src, dst ) );
  ins_pipe( ialu_mem_reg );
%}

// And Memory with Immediate
instruct andI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (AndI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "AND    $dst,$src" %}
  opcode(0x81, 0x4);  /* Opcode 81 /4 id */
  // ins_encode( MemImm( dst, src) );
  ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
  ins_pipe( ialu_mem_imm );
%}

// Or Instructions
// Or Register with Register
instruct orI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (OrI dst src));
  effect(KILL cr);

  size(2);
  format %{ "OR     $dst,$src" %}
  opcode(0x0B);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct orI_eReg_castP2X(rRegI dst, eRegP src, eFlagsReg cr) %{
  match(Set dst (OrI dst (CastP2X src)));
  effect(KILL cr);

  size(2);
  format %{ "OR     $dst,$src" %}
  opcode(0x0B);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}


// Or Register with Immediate
instruct orI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
  match(Set dst (OrI dst src));
  effect(KILL cr);

  format %{ "OR     $dst,$src" %}
  opcode(0x81,0x01);  /* Opcode 81 /1 id */
  // ins_encode( RegImm( dst, src) );
  ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
  ins_pipe( ialu_reg );
%}

// Or Register with Memory
instruct orI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
  match(Set dst (OrI dst (LoadI src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "OR     $dst,$src" %}
  opcode(0x0B);
  ins_encode( OpcP, RegMem( dst, src) );
  ins_pipe( ialu_reg_mem );
%}

// Or Memory with Register
instruct orI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (OrI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(150);
  format %{ "OR     $dst,$src" %}
  opcode(0x09);  /* Opcode 09 /r */
  ins_encode( OpcP, RegMem( src, dst ) );
  ins_pipe( ialu_mem_reg );
%}

// Or Memory with Immediate
instruct orI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (OrI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "OR     $dst,$src" %}
  opcode(0x81,0x1);  /* Opcode 81 /1 id */
  // ins_encode( MemImm( dst, src) );
  ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
  ins_pipe( ialu_mem_imm );
%}

// ROL/ROR
// ROL expand
instruct rolI_eReg_imm1(rRegI dst, immI1 shift, eFlagsReg cr) %{
  effect(USE_DEF dst, USE shift, KILL cr);

  format %{ "ROL    $dst, $shift" %}
  opcode(0xD1, 0x0); /* Opcode D1 /0 */
  ins_encode( OpcP, RegOpc( dst ));
  ins_pipe( ialu_reg );
%}

instruct rolI_eReg_imm8(rRegI dst, immI8 shift, eFlagsReg cr) %{
  effect(USE_DEF dst, USE shift, KILL cr);

  format %{ "ROL    $dst, $shift" %}
  opcode(0xC1, 0x0); /*Opcode /C1  /0  */
  ins_encode( RegOpcImm(dst, shift) );
  ins_pipe(ialu_reg);
%}

instruct rolI_eReg_CL(ncxRegI dst, eCXRegI shift, eFlagsReg cr) %{
  effect(USE_DEF dst, USE shift, KILL cr);

  format %{ "ROL    $dst, $shift" %}
  opcode(0xD3, 0x0);    /* Opcode D3 /0 */
  ins_encode(OpcP, RegOpc(dst));
  ins_pipe( ialu_reg_reg );
%}
// end of ROL expand

// ROL 32bit by one once
instruct rolI_eReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, eFlagsReg cr) %{
  match(Set dst ( OrI (LShiftI dst lshift) (URShiftI dst rshift)));

  expand %{
    rolI_eReg_imm1(dst, lshift, cr);
  %}
%}

// ROL 32bit var by imm8 once
instruct rolI_eReg_i8(rRegI dst, immI8 lshift, immI8 rshift, eFlagsReg cr) %{
  predicate(  0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
  match(Set dst ( OrI (LShiftI dst lshift) (URShiftI dst rshift)));

  expand %{
    rolI_eReg_imm8(dst, lshift, cr);
  %}
%}

// ROL 32bit var by var once
instruct rolI_eReg_Var_C0(ncxRegI dst, eCXRegI shift, immI0 zero, eFlagsReg cr) %{
  match(Set dst ( OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));

  expand %{
    rolI_eReg_CL(dst, shift, cr);
  %}
%}

// ROL 32bit var by var once
instruct rolI_eReg_Var_C32(ncxRegI dst, eCXRegI shift, immI_32 c32, eFlagsReg cr) %{
  match(Set dst ( OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));

  expand %{
    rolI_eReg_CL(dst, shift, cr);
  %}
%}

// ROR expand
instruct rorI_eReg_imm1(rRegI dst, immI1 shift, eFlagsReg cr) %{
  effect(USE_DEF dst, USE shift, KILL cr);

  format %{ "ROR    $dst, $shift" %}
  opcode(0xD1,0x1);  /* Opcode D1 /1 */
  ins_encode( OpcP, RegOpc( dst ) );
  ins_pipe( ialu_reg );
%}

instruct rorI_eReg_imm8(rRegI dst, immI8 shift, eFlagsReg cr) %{
  effect (USE_DEF dst, USE shift, KILL cr);

  format %{ "ROR    $dst, $shift" %}
  opcode(0xC1, 0x1); /* Opcode /C1 /1 ib */
  ins_encode( RegOpcImm(dst, shift) );
  ins_pipe( ialu_reg );
%}

instruct rorI_eReg_CL(ncxRegI dst, eCXRegI shift, eFlagsReg cr)%{
  effect(USE_DEF dst, USE shift, KILL cr);

  format %{ "ROR    $dst, $shift" %}
  opcode(0xD3, 0x1);    /* Opcode D3 /1 */
  ins_encode(OpcP, RegOpc(dst));
  ins_pipe( ialu_reg_reg );
%}
// end of ROR expand

// ROR right once
instruct rorI_eReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, eFlagsReg cr) %{
  match(Set dst ( OrI (URShiftI dst rshift) (LShiftI dst lshift)));

  expand %{
    rorI_eReg_imm1(dst, rshift, cr);
  %}
%}

// ROR 32bit by immI8 once
instruct rorI_eReg_i8(rRegI dst, immI8 rshift, immI8 lshift, eFlagsReg cr) %{
  predicate(  0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
  match(Set dst ( OrI (URShiftI dst rshift) (LShiftI dst lshift)));

  expand %{
    rorI_eReg_imm8(dst, rshift, cr);
  %}
%}

// ROR 32bit var by var once
instruct rorI_eReg_Var_C0(ncxRegI dst, eCXRegI shift, immI0 zero, eFlagsReg cr) %{
  match(Set dst ( OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));

  expand %{
    rorI_eReg_CL(dst, shift, cr);
  %}
%}

// ROR 32bit var by var once
instruct rorI_eReg_Var_C32(ncxRegI dst, eCXRegI shift, immI_32 c32, eFlagsReg cr) %{
  match(Set dst ( OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));

  expand %{
    rorI_eReg_CL(dst, shift, cr);
  %}
%}

// Xor Instructions
// Xor Register with Register
instruct xorI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (XorI dst src));
  effect(KILL cr);

  size(2);
  format %{ "XOR    $dst,$src" %}
  opcode(0x33);
  ins_encode( OpcP, RegReg( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

// Xor Register with Immediate -1
instruct xorI_eReg_im1(rRegI dst, immI_M1 imm) %{
  match(Set dst (XorI dst imm));  

  size(2);
  format %{ "NOT    $dst" %}  
  ins_encode %{
     __ notl($dst$$Register);
  %}
  ins_pipe( ialu_reg );
%}

// Xor Register with Immediate
instruct xorI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
  match(Set dst (XorI dst src));
  effect(KILL cr);

  format %{ "XOR    $dst,$src" %}
  opcode(0x81,0x06);  /* Opcode 81 /6 id */
  // ins_encode( RegImm( dst, src) );
  ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
  ins_pipe( ialu_reg );
%}

// Xor Register with Memory
instruct xorI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
  match(Set dst (XorI dst (LoadI src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "XOR    $dst,$src" %}
  opcode(0x33);
  ins_encode( OpcP, RegMem(dst, src) );
  ins_pipe( ialu_reg_mem );
%}

// Xor Memory with Register
instruct xorI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (XorI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(150);
  format %{ "XOR    $dst,$src" %}
  opcode(0x31);  /* Opcode 31 /r */
  ins_encode( OpcP, RegMem( src, dst ) );
  ins_pipe( ialu_mem_reg );
%}

// Xor Memory with Immediate
instruct xorI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
  match(Set dst (StoreI dst (XorI (LoadI dst) src)));
  effect(KILL cr);

  ins_cost(125);
  format %{ "XOR    $dst,$src" %}
  opcode(0x81,0x6);  /* Opcode 81 /6 id */
  ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
  ins_pipe( ialu_mem_imm );
%}

//----------Convert Int to Boolean---------------------------------------------

instruct movI_nocopy(rRegI dst, rRegI src) %{
  effect( DEF dst, USE src );
  format %{ "MOV    $dst,$src" %}
  ins_encode( enc_Copy( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct ci2b( rRegI dst, rRegI src, eFlagsReg cr ) %{
  effect( USE_DEF dst, USE src, KILL cr );

  size(4);
  format %{ "NEG    $dst\n\t"
            "ADC    $dst,$src" %}
  ins_encode( neg_reg(dst),
              OpcRegReg(0x13,dst,src) );
  ins_pipe( ialu_reg_reg_long );
%}

instruct convI2B( rRegI dst, rRegI src, eFlagsReg cr ) %{
  match(Set dst (Conv2B src));

  expand %{
    movI_nocopy(dst,src);
    ci2b(dst,src,cr);
  %}
%}

instruct movP_nocopy(rRegI dst, eRegP src) %{
  effect( DEF dst, USE src );
  format %{ "MOV    $dst,$src" %}
  ins_encode( enc_Copy( dst, src) );
  ins_pipe( ialu_reg_reg );
%}

instruct cp2b( rRegI dst, eRegP src, eFlagsReg cr ) %{
  effect( USE_DEF dst, USE src, KILL cr );
  format %{ "NEG    $dst\n\t"
            "ADC    $dst,$src" %}
  ins_encode( neg_reg(dst),
              OpcRegReg(0x13,dst,src) );
  ins_pipe( ialu_reg_reg_long );
%}

instruct convP2B( rRegI dst, eRegP src, eFlagsReg cr ) %{
  match(Set dst (Conv2B src));

  expand %{
    movP_nocopy(dst,src);
    cp2b(dst,src,cr);
  %}
%}

instruct cmpLTMask(eCXRegI dst, ncxRegI p, ncxRegI q, eFlagsReg cr) %{
  match(Set dst (CmpLTMask p q));
  effect(KILL cr);
  ins_cost(400);

  // SETlt can only use low byte of EAX,EBX, ECX, or EDX as destination
  format %{ "XOR    $dst,$dst\n\t"
            "CMP    $p,$q\n\t"
            "SETlt  $dst\n\t"
            "NEG    $dst" %}
  ins_encode %{
    Register Rp = $p$$Register;
    Register Rq = $q$$Register;
    Register Rd = $dst$$Register;
    Label done;
    __ xorl(Rd, Rd);
    __ cmpl(Rp, Rq);
    __ setb(Assembler::less, Rd);
    __ negl(Rd);
  %}

  ins_pipe(pipe_slow);
%}

instruct cmpLTMask0(rRegI dst, immI0 zero, eFlagsReg cr) %{
  match(Set dst (CmpLTMask dst zero));
  effect(DEF dst, KILL cr);
  ins_cost(100);

  format %{ "SAR    $dst,31\t# cmpLTMask0" %}
  ins_encode %{
  __ sarl($dst$$Register, 31);
  %}
  ins_pipe(ialu_reg);
%}

/* better to save a register than avoid a branch */
instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y, eFlagsReg cr) %{
  match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
  effect(KILL cr);
  ins_cost(400);
  format %{ "SUB    $p,$q\t# cadd_cmpLTMask\n\t"
            "JGE    done\n\t"
            "ADD    $p,$y\n"
            "done:  " %}
  ins_encode %{
    Register Rp = $p$$Register;
    Register Rq = $q$$Register;
    Register Ry = $y$$Register;
    Label done;
    __ subl(Rp, Rq);
    __ jccb(Assembler::greaterEqual, done);
    __ addl(Rp, Ry);
    __ bind(done);
  %}

  ins_pipe(pipe_cmplt);
%}

/* better to save a register than avoid a branch */
instruct and_cmpLTMask(rRegI p, rRegI q, rRegI y, eFlagsReg cr) %{
  match(Set y (AndI (CmpLTMask p q) y));
  effect(KILL cr);

  ins_cost(300);

  format %{ "CMPL     $p, $q\t# and_cmpLTMask\n\t"
            "JLT      done\n\t"
            "XORL     $y, $y\n"
            "done:  " %}
  ins_encode %{
    Register Rp = $p$$Register;
    Register Rq = $q$$Register;
    Register Ry = $y$$Register;
    Label done;
    __ cmpl(Rp, Rq);
    __ jccb(Assembler::less, done);
    __ xorl(Ry, Ry);
    __ bind(done);
  %}

  ins_pipe(pipe_cmplt);
%}

/* If I enable this, I encourage spilling in the inner loop of compress.
instruct cadd_cmpLTMask_mem(ncxRegI p, ncxRegI q, memory y, eCXRegI tmp, eFlagsReg cr) %{
  match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
*/

//----------Long Instructions------------------------------------------------
// Add Long Register with Register
instruct addL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (AddL dst src));
  effect(KILL cr);
  ins_cost(200);
  format %{ "ADD    $dst.lo,$src.lo\n\t"
            "ADC    $dst.hi,$src.hi" %}
  opcode(0x03, 0x13);
  ins_encode( RegReg_Lo(dst, src), RegReg_Hi(dst,src) );
  ins_pipe( ialu_reg_reg_long );
%}

// Add Long Register with Immediate
instruct addL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
  match(Set dst (AddL dst src));
  effect(KILL cr);
  format %{ "ADD    $dst.lo,$src.lo\n\t"
            "ADC    $dst.hi,$src.hi" %}
  opcode(0x81,0x00,0x02);  /* Opcode 81 /0, 81 /2 */
  ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
  ins_pipe( ialu_reg_long );
%}

// Add Long Register with Memory
instruct addL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
  match(Set dst (AddL dst (LoadL mem)));
  effect(KILL cr);
  ins_cost(125);
  format %{ "ADD    $dst.lo,$mem\n\t"
            "ADC    $dst.hi,$mem+4" %}
  opcode(0x03, 0x13);
  ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
  ins_pipe( ialu_reg_long_mem );
%}

// Subtract Long Register with Register.
instruct subL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (SubL dst src));
  effect(KILL cr);
  ins_cost(200);
  format %{ "SUB    $dst.lo,$src.lo\n\t"
            "SBB    $dst.hi,$src.hi" %}
  opcode(0x2B, 0x1B);
  ins_encode( RegReg_Lo(dst, src), RegReg_Hi(dst,src) );
  ins_pipe( ialu_reg_reg_long );
%}

// Subtract Long Register with Immediate
instruct subL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
  match(Set dst (SubL dst src));
  effect(KILL cr);
  format %{ "SUB    $dst.lo,$src.lo\n\t"
            "SBB    $dst.hi,$src.hi" %}
  opcode(0x81,0x05,0x03);  /* Opcode 81 /5, 81 /3 */
  ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
  ins_pipe( ialu_reg_long );
%}

// Subtract Long Register with Memory
instruct subL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
  match(Set dst (SubL dst (LoadL mem)));
  effect(KILL cr);
  ins_cost(125);
  format %{ "SUB    $dst.lo,$mem\n\t"
            "SBB    $dst.hi,$mem+4" %}
  opcode(0x2B, 0x1B);
  ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
  ins_pipe( ialu_reg_long_mem );
%}

instruct negL_eReg(eRegL dst, immL0 zero, eFlagsReg cr) %{
  match(Set dst (SubL zero dst));
  effect(KILL cr);
  ins_cost(300);
  format %{ "NEG    $dst.hi\n\tNEG    $dst.lo\n\tSBB    $dst.hi,0" %}
  ins_encode( neg_long(dst) );
  ins_pipe( ialu_reg_reg_long );
%}

// And Long Register with Register
instruct andL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  format %{ "AND    $dst.lo,$src.lo\n\t"
            "AND    $dst.hi,$src.hi" %}
  opcode(0x23,0x23);
  ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
  ins_pipe( ialu_reg_reg_long );
%}

// And Long Register with Immediate
instruct andL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  format %{ "AND    $dst.lo,$src.lo\n\t"
            "AND    $dst.hi,$src.hi" %}
  opcode(0x81,0x04,0x04);  /* Opcode 81 /4, 81 /4 */
  ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
  ins_pipe( ialu_reg_long );
%}

// And Long Register with Memory
instruct andL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
  match(Set dst (AndL dst (LoadL mem)));
  effect(KILL cr);
  ins_cost(125);
  format %{ "AND    $dst.lo,$mem\n\t"
            "AND    $dst.hi,$mem+4" %}
  opcode(0x23, 0x23);
  ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
  ins_pipe( ialu_reg_long_mem );
%}

// Or Long Register with Register
instruct orl_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (OrL dst src));
  effect(KILL cr);
  format %{ "OR     $dst.lo,$src.lo\n\t"
            "OR     $dst.hi,$src.hi" %}
  opcode(0x0B,0x0B);
  ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
  ins_pipe( ialu_reg_reg_long );
%}

// Or Long Register with Immediate
instruct orl_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
  match(Set dst (OrL dst src));
  effect(KILL cr);
  format %{ "OR     $dst.lo,$src.lo\n\t"
            "OR     $dst.hi,$src.hi" %}
  opcode(0x81,0x01,0x01);  /* Opcode 81 /1, 81 /1 */
  ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
  ins_pipe( ialu_reg_long );
%}

// Or Long Register with Memory
instruct orl_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
  match(Set dst (OrL dst (LoadL mem)));
  effect(KILL cr);
  ins_cost(125);
  format %{ "OR     $dst.lo,$mem\n\t"
            "OR     $dst.hi,$mem+4" %}
  opcode(0x0B,0x0B);
  ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
  ins_pipe( ialu_reg_long_mem );
%}

// Xor Long Register with Register
instruct xorl_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (XorL dst src));
  effect(KILL cr);
  format %{ "XOR    $dst.lo,$src.lo\n\t"
            "XOR    $dst.hi,$src.hi" %}
  opcode(0x33,0x33);
  ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
  ins_pipe( ialu_reg_reg_long );
%}

// Xor Long Register with Immediate -1
instruct xorl_eReg_im1(eRegL dst, immL_M1 imm) %{
  match(Set dst (XorL dst imm));  
  format %{ "NOT    $dst.lo\n\t"
            "NOT    $dst.hi" %}
  ins_encode %{
     __ notl($dst$$Register);
     __ notl(HIGH_FROM_LOW($dst$$Register));
  %}
  ins_pipe( ialu_reg_long );
%}

// Xor Long Register with Immediate
instruct xorl_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
  match(Set dst (XorL dst src));
  effect(KILL cr);
  format %{ "XOR    $dst.lo,$src.lo\n\t"
            "XOR    $dst.hi,$src.hi" %}
  opcode(0x81,0x06,0x06);  /* Opcode 81 /6, 81 /6 */
  ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
  ins_pipe( ialu_reg_long );
%}

// Xor Long Register with Memory
instruct xorl_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
  match(Set dst (XorL dst (LoadL mem)));
  effect(KILL cr);
  ins_cost(125);
  format %{ "XOR    $dst.lo,$mem\n\t"
            "XOR    $dst.hi,$mem+4" %}
  opcode(0x33,0x33);
  ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
  ins_pipe( ialu_reg_long_mem );
%}

// Shift Left Long by 1
instruct shlL_eReg_1(eRegL dst, immI_1 cnt, eFlagsReg cr) %{
  predicate(UseNewLongLShift);
  match(Set dst (LShiftL dst cnt));
  effect(KILL cr);
  ins_cost(100);
  format %{ "ADD    $dst.lo,$dst.lo\n\t"
            "ADC    $dst.hi,$dst.hi" %}
  ins_encode %{
    __ addl($dst$$Register,$dst$$Register);
    __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
  %}
  ins_pipe( ialu_reg_long );
%}

// Shift Left Long by 2
instruct shlL_eReg_2(eRegL dst, immI_2 cnt, eFlagsReg cr) %{
  predicate(UseNewLongLShift);
  match(Set dst (LShiftL dst cnt));
  effect(KILL cr);
  ins_cost(100);
  format %{ "ADD    $dst.lo,$dst.lo\n\t"
            "ADC    $dst.hi,$dst.hi\n\t" 
            "ADD    $dst.lo,$dst.lo\n\t"
            "ADC    $dst.hi,$dst.hi" %}
  ins_encode %{
    __ addl($dst$$Register,$dst$$Register);
    __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
    __ addl($dst$$Register,$dst$$Register);
    __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
  %}
  ins_pipe( ialu_reg_long );
%}

// Shift Left Long by 3
instruct shlL_eReg_3(eRegL dst, immI_3 cnt, eFlagsReg cr) %{
  predicate(UseNewLongLShift);
  match(Set dst (LShiftL dst cnt));
  effect(KILL cr);
  ins_cost(100);
  format %{ "ADD    $dst.lo,$dst.lo\n\t"
            "ADC    $dst.hi,$dst.hi\n\t" 
            "ADD    $dst.lo,$dst.lo\n\t"
            "ADC    $dst.hi,$dst.hi\n\t" 
            "ADD    $dst.lo,$dst.lo\n\t"
            "ADC    $dst.hi,$dst.hi" %}
  ins_encode %{
    __ addl($dst$$Register,$dst$$Register);
    __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
    __ addl($dst$$Register,$dst$$Register);
    __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
    __ addl($dst$$Register,$dst$$Register);
    __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
  %}
  ins_pipe( ialu_reg_long );
%}

// Shift Left Long by 1-31
instruct shlL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
  match(Set dst (LShiftL dst cnt));
  effect(KILL cr);
  ins_cost(200);
  format %{ "SHLD   $dst.hi,$dst.lo,$cnt\n\t"
            "SHL    $dst.lo,$cnt" %}
  opcode(0xC1, 0x4, 0xA4);  /* 0F/A4, then C1 /4 ib */
  ins_encode( move_long_small_shift(dst,cnt) );
  ins_pipe( ialu_reg_long );
%}

// Shift Left Long by 32-63
instruct shlL_eReg_32_63(eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
  match(Set dst (LShiftL dst cnt));
  effect(KILL cr);
  ins_cost(300);
  format %{ "MOV    $dst.hi,$dst.lo\n"
          "\tSHL    $dst.hi,$cnt-32\n"
          "\tXOR    $dst.lo,$dst.lo" %}
  opcode(0xC1, 0x4);  /* C1 /4 ib */
  ins_encode( move_long_big_shift_clr(dst,cnt) );
  ins_pipe( ialu_reg_long );
%}

// Shift Left Long by variable
instruct salL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
  match(Set dst (LShiftL dst shift));
  effect(KILL cr);
  ins_cost(500+200);
  size(17);
  format %{ "TEST   $shift,32\n\t"
            "JEQ,s  small\n\t"
            "MOV    $dst.hi,$dst.lo\n\t"
            "XOR    $dst.lo,$dst.lo\n"
    "small:\tSHLD   $dst.hi,$dst.lo,$shift\n\t"
            "SHL    $dst.lo,$shift" %}
  ins_encode( shift_left_long( dst, shift ) );
  ins_pipe( pipe_slow );
%}

// Shift Right Long by 1-31
instruct shrL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
  match(Set dst (URShiftL dst cnt));
  effect(KILL cr);
  ins_cost(200);
  format %{ "SHRD   $dst.lo,$dst.hi,$cnt\n\t"
            "SHR    $dst.hi,$cnt" %}
  opcode(0xC1, 0x5, 0xAC);  /* 0F/AC, then C1 /5 ib */
  ins_encode( move_long_small_shift(dst,cnt) );
  ins_pipe( ialu_reg_long );
%}

// Shift Right Long by 32-63
instruct shrL_eReg_32_63(eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
  match(Set dst (URShiftL dst cnt));
  effect(KILL cr);
  ins_cost(300);
  format %{ "MOV    $dst.lo,$dst.hi\n"
          "\tSHR    $dst.lo,$cnt-32\n"
          "\tXOR    $dst.hi,$dst.hi" %}
  opcode(0xC1, 0x5);  /* C1 /5 ib */
  ins_encode( move_long_big_shift_clr(dst,cnt) );
  ins_pipe( ialu_reg_long );
%}

// Shift Right Long by variable
instruct shrL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
  match(Set dst (URShiftL dst shift));
  effect(KILL cr);
  ins_cost(600);
  size(17);
  format %{ "TEST   $shift,32\n\t"
            "JEQ,s  small\n\t"
            "MOV    $dst.lo,$dst.hi\n\t"
            "XOR    $dst.hi,$dst.hi\n"
    "small:\tSHRD   $dst.lo,$dst.hi,$shift\n\t"
            "SHR    $dst.hi,$shift" %}
  ins_encode( shift_right_long( dst, shift ) );
  ins_pipe( pipe_slow );
%}

// Shift Right Long by 1-31
instruct sarL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
  match(Set dst (RShiftL dst cnt));
  effect(KILL cr);
  ins_cost(200);
  format %{ "SHRD   $dst.lo,$dst.hi,$cnt\n\t"
            "SAR    $dst.hi,$cnt" %}
  opcode(0xC1, 0x7, 0xAC);  /* 0F/AC, then C1 /7 ib */
  ins_encode( move_long_small_shift(dst,cnt) );
  ins_pipe( ialu_reg_long );
%}

// Shift Right Long by 32-63
instruct sarL_eReg_32_63( eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
  match(Set dst (RShiftL dst cnt));
  effect(KILL cr);
  ins_cost(300);
  format %{ "MOV    $dst.lo,$dst.hi\n"
          "\tSAR    $dst.lo,$cnt-32\n"
          "\tSAR    $dst.hi,31" %}
  opcode(0xC1, 0x7);  /* C1 /7 ib */
  ins_encode( move_long_big_shift_sign(dst,cnt) );
  ins_pipe( ialu_reg_long );
%}

// Shift Right arithmetic Long by variable
instruct sarL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
  match(Set dst (RShiftL dst shift));
  effect(KILL cr);
  ins_cost(600);
  size(18);
  format %{ "TEST   $shift,32\n\t"
            "JEQ,s  small\n\t"
            "MOV    $dst.lo,$dst.hi\n\t"
            "SAR    $dst.hi,31\n"
    "small:\tSHRD   $dst.lo,$dst.hi,$shift\n\t"
            "SAR    $dst.hi,$shift" %}
  ins_encode( shift_right_arith_long( dst, shift ) );
  ins_pipe( pipe_slow );
%}


//----------Double Instructions------------------------------------------------
// Double Math

// Compare & branch

// P6 version of float compare, sets condition codes in EFLAGS
instruct cmpDPR_cc_P6(eFlagsRegU cr, regDPR src1, regDPR src2, eAXRegI rax) %{
  predicate(VM_Version::supports_cmov() && UseSSE <=1);
  match(Set cr (CmpD src1 src2));
  effect(KILL rax);
  ins_cost(150);
  format %{ "FLD    $src1\n\t"
            "FUCOMIP ST,$src2  // P6 instruction\n\t"
            "JNP    exit\n\t"
            "MOV    ah,1       // saw a NaN, set CF\n\t"
            "SAHF\n"
     "exit:\tNOP               // avoid branch to branch" %}
  opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2),
              cmpF_P6_fixup );
  ins_pipe( pipe_slow );
%}

instruct cmpDPR_cc_P6CF(eFlagsRegUCF cr, regDPR src1, regDPR src2) %{
  predicate(VM_Version::supports_cmov() && UseSSE <=1);
  match(Set cr (CmpD src1 src2));
  ins_cost(150);
  format %{ "FLD    $src1\n\t"
            "FUCOMIP ST,$src2  // P6 instruction" %}
  opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2));
  ins_pipe( pipe_slow );
%}

// Compare & branch
instruct cmpDPR_cc(eFlagsRegU cr, regDPR src1, regDPR src2, eAXRegI rax) %{
  predicate(UseSSE<=1);
  match(Set cr (CmpD src1 src2));
  effect(KILL rax);
  ins_cost(200);
  format %{ "FLD    $src1\n\t"
            "FCOMp  $src2\n\t"
            "FNSTSW AX\n\t"
            "TEST   AX,0x400\n\t"
            "JZ,s   flags\n\t"
            "MOV    AH,1\t# unordered treat as LT\n"
    "flags:\tSAHF" %}
  opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2),
              fpu_flags);
  ins_pipe( pipe_slow );
%}

// Compare vs zero into -1,0,1
instruct cmpDPR_0(rRegI dst, regDPR src1, immDPR0 zero, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE<=1);
  match(Set dst (CmpD3 src1 zero));
  effect(KILL cr, KILL rax);
  ins_cost(280);
  format %{ "FTSTD  $dst,$src1" %}
  opcode(0xE4, 0xD9);
  ins_encode( Push_Reg_DPR(src1),
              OpcS, OpcP, PopFPU,
              CmpF_Result(dst));
  ins_pipe( pipe_slow );
%}

// Compare into -1,0,1
instruct cmpDPR_reg(rRegI dst, regDPR src1, regDPR src2, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE<=1);
  match(Set dst (CmpD3 src1 src2));
  effect(KILL cr, KILL rax);
  ins_cost(300);
  format %{ "FCMPD  $dst,$src1,$src2" %}
  opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2),
              CmpF_Result(dst));
  ins_pipe( pipe_slow );
%}

// float compare and set condition codes in EFLAGS by XMM regs
instruct cmpD_cc(eFlagsRegU cr, regD src1, regD src2) %{
  predicate(UseSSE>=2);
  match(Set cr (CmpD src1 src2));
  ins_cost(145);
  format %{ "UCOMISD $src1,$src2\n\t"
            "JNP,s   exit\n\t"
            "PUSHF\t# saw NaN, set CF\n\t"
            "AND     [rsp], #0xffffff2b\n\t"
            "POPF\n"
    "exit:" %}
  ins_encode %{
    __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
    emit_cmpfp_fixup(_masm);
  %}
  ins_pipe( pipe_slow );
%}

instruct cmpD_ccCF(eFlagsRegUCF cr, regD src1, regD src2) %{
  predicate(UseSSE>=2);
  match(Set cr (CmpD src1 src2));
  ins_cost(100);
  format %{ "UCOMISD $src1,$src2" %}
  ins_encode %{
    __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// float compare and set condition codes in EFLAGS by XMM regs
instruct cmpD_ccmem(eFlagsRegU cr, regD src1, memory src2) %{
  predicate(UseSSE>=2);
  match(Set cr (CmpD src1 (LoadD src2)));
  ins_cost(145);
  format %{ "UCOMISD $src1,$src2\n\t"
            "JNP,s   exit\n\t"
            "PUSHF\t# saw NaN, set CF\n\t"
            "AND     [rsp], #0xffffff2b\n\t"
            "POPF\n"
    "exit:" %}
  ins_encode %{
    __ ucomisd($src1$$XMMRegister, $src2$$Address);
    emit_cmpfp_fixup(_masm);
  %}
  ins_pipe( pipe_slow );
%}

instruct cmpD_ccmemCF(eFlagsRegUCF cr, regD src1, memory src2) %{
  predicate(UseSSE>=2);
  match(Set cr (CmpD src1 (LoadD src2)));
  ins_cost(100);
  format %{ "UCOMISD $src1,$src2" %}
  ins_encode %{
    __ ucomisd($src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Compare into -1,0,1 in XMM
instruct cmpD_reg(xRegI dst, regD src1, regD src2, eFlagsReg cr) %{
  predicate(UseSSE>=2);
  match(Set dst (CmpD3 src1 src2));
  effect(KILL cr);
  ins_cost(255);
  format %{ "UCOMISD $src1, $src2\n\t"
            "MOV     $dst, #-1\n\t"
            "JP,s    done\n\t"
            "JB,s    done\n\t"
            "SETNE   $dst\n\t"
            "MOVZB   $dst, $dst\n"
    "done:" %}
  ins_encode %{
    __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
    emit_cmpfp3(_masm, $dst$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// Compare into -1,0,1 in XMM and memory
instruct cmpD_regmem(xRegI dst, regD src1, memory src2, eFlagsReg cr) %{
  predicate(UseSSE>=2);
  match(Set dst (CmpD3 src1 (LoadD src2)));
  effect(KILL cr);
  ins_cost(275);
  format %{ "UCOMISD $src1, $src2\n\t"
            "MOV     $dst, #-1\n\t"
            "JP,s    done\n\t"
            "JB,s    done\n\t"
            "SETNE   $dst\n\t"
            "MOVZB   $dst, $dst\n"
    "done:" %}
  ins_encode %{
    __ ucomisd($src1$$XMMRegister, $src2$$Address);
    emit_cmpfp3(_masm, $dst$$Register);
  %}
  ins_pipe( pipe_slow );
%}


instruct subDPR_reg(regDPR dst, regDPR src) %{
  predicate (UseSSE <=1);
  match(Set dst (SubD dst src));

  format %{ "FLD    $src\n\t"
            "DSUBp  $dst,ST" %}
  opcode(0xDE, 0x5); /* DE E8+i  or DE /5 */
  ins_cost(150);
  ins_encode( Push_Reg_DPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}

instruct subDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
  predicate (UseSSE <=1);
  match(Set dst (RoundDouble (SubD src1 src2)));
  ins_cost(250);

  format %{ "FLD    $src2\n\t"
            "DSUB   ST,$src1\n\t"
            "FSTP_D $dst\t# D-round" %}
  opcode(0xD8, 0x5);
  ins_encode( Push_Reg_DPR(src2),
              OpcP, RegOpc(src1), Pop_Mem_DPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}


instruct subDPR_reg_mem(regDPR dst, memory src) %{
  predicate (UseSSE <=1);
  match(Set dst (SubD dst (LoadD src)));
  ins_cost(150);

  format %{ "FLD    $src\n\t"
            "DSUBp  $dst,ST" %}
  opcode(0xDE, 0x5, 0xDD); /* DE C0+i */  /* LoadD  DD /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_mem );
%}

instruct absDPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate (UseSSE<=1);
  match(Set dst (AbsD src));
  ins_cost(100);
  format %{ "FABS" %}
  opcode(0xE1, 0xD9);
  ins_encode( OpcS, OpcP );
  ins_pipe( fpu_reg_reg );
%}

instruct negDPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate(UseSSE<=1);
  match(Set dst (NegD src));
  ins_cost(100);
  format %{ "FCHS" %}
  opcode(0xE0, 0xD9);
  ins_encode( OpcS, OpcP );
  ins_pipe( fpu_reg_reg );
%}

instruct addDPR_reg(regDPR dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (AddD dst src));
  format %{ "FLD    $src\n\t"
            "DADD   $dst,ST" %}
  size(4);
  ins_cost(150);
  opcode(0xDE, 0x0); /* DE C0+i or DE /0*/
  ins_encode( Push_Reg_DPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}


instruct addDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
  predicate(UseSSE<=1);
  match(Set dst (RoundDouble (AddD src1 src2)));
  ins_cost(250);

  format %{ "FLD    $src2\n\t"
            "DADD   ST,$src1\n\t"
            "FSTP_D $dst\t# D-round" %}
  opcode(0xD8, 0x0); /* D8 C0+i or D8 /0*/
  ins_encode( Push_Reg_DPR(src2),
              OpcP, RegOpc(src1), Pop_Mem_DPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}


instruct addDPR_reg_mem(regDPR dst, memory src) %{
  predicate(UseSSE<=1);
  match(Set dst (AddD dst (LoadD src)));
  ins_cost(150);

  format %{ "FLD    $src\n\t"
            "DADDp  $dst,ST" %}
  opcode(0xDE, 0x0, 0xDD); /* DE C0+i */  /* LoadD  DD /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_mem );
%}

// add-to-memory
instruct addDPR_mem_reg(memory dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (StoreD dst (RoundDouble (AddD (LoadD dst) src))));
  ins_cost(150);

  format %{ "FLD_D  $dst\n\t"
            "DADD   ST,$src\n\t"
            "FST_D  $dst" %}
  opcode(0xDD, 0x0);
  ins_encode( Opcode(0xDD), RMopc_Mem(0x00,dst),
              Opcode(0xD8), RegOpc(src),
              set_instruction_start,
              Opcode(0xDD), RMopc_Mem(0x03,dst) );
  ins_pipe( fpu_reg_mem );
%}

instruct addDPR_reg_imm1(regDPR dst, immDPR1 con) %{
  predicate(UseSSE<=1);
  match(Set dst (AddD dst con));
  ins_cost(125);
  format %{ "FLD1\n\t"
            "DADDp  $dst,ST" %}
  ins_encode %{
    __ fld1();
    __ faddp($dst$$reg);
  %}
  ins_pipe(fpu_reg);
%}

instruct addDPR_reg_imm(regDPR dst, immDPR con) %{
  predicate(UseSSE<=1 && _kids[1]->_leaf->getd() != 0.0 && _kids[1]->_leaf->getd() != 1.0 );
  match(Set dst (AddD dst con));
  ins_cost(200);
  format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
            "DADDp  $dst,ST" %}
  ins_encode %{
    __ fld_d($constantaddress($con));
    __ faddp($dst$$reg);
  %}
  ins_pipe(fpu_reg_mem);
%}

instruct addDPR_reg_imm_round(stackSlotD dst, regDPR src, immDPR con) %{
  predicate(UseSSE<=1 && _kids[0]->_kids[1]->_leaf->getd() != 0.0 && _kids[0]->_kids[1]->_leaf->getd() != 1.0 );
  match(Set dst (RoundDouble (AddD src con)));
  ins_cost(200);
  format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
            "DADD   ST,$src\n\t"
            "FSTP_D $dst\t# D-round" %}
  ins_encode %{
    __ fld_d($constantaddress($con));
    __ fadd($src$$reg);
    __ fstp_d(Address(rsp, $dst$$disp));
  %}
  ins_pipe(fpu_mem_reg_con);
%}

instruct mulDPR_reg(regDPR dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (MulD dst src));
  format %{ "FLD    $src\n\t"
            "DMULp  $dst,ST" %}
  opcode(0xDE, 0x1); /* DE C8+i or DE /1*/
  ins_cost(150);
  ins_encode( Push_Reg_DPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}

// Strict FP instruction biases argument before multiply then
// biases result to avoid double rounding of subnormals.
//
// scale arg1 by multiplying arg1 by 2^(-15360)
// load arg2
// multiply scaled arg1 by arg2
// rescale product by 2^(15360)
//
instruct strictfp_mulDPR_reg(regDPR1 dst, regnotDPR1 src) %{
  predicate( UseSSE<=1 && Compile::current()->has_method() && Compile::current()->method()->is_strict() );
  match(Set dst (MulD dst src));
  ins_cost(1);   // Select this instruction for all strict FP double multiplies

  format %{ "FLD    StubRoutines::_fpu_subnormal_bias1\n\t"
            "DMULp  $dst,ST\n\t"
            "FLD    $src\n\t"
            "DMULp  $dst,ST\n\t"
            "FLD    StubRoutines::_fpu_subnormal_bias2\n\t"
            "DMULp  $dst,ST\n\t" %}
  opcode(0xDE, 0x1); /* DE C8+i or DE /1*/
  ins_encode( strictfp_bias1(dst),
              Push_Reg_DPR(src),
              OpcP, RegOpc(dst),
              strictfp_bias2(dst) );
  ins_pipe( fpu_reg_reg );
%}

instruct mulDPR_reg_imm(regDPR dst, immDPR con) %{
  predicate( UseSSE<=1 && _kids[1]->_leaf->getd() != 0.0 && _kids[1]->_leaf->getd() != 1.0 );
  match(Set dst (MulD dst con));
  ins_cost(200);
  format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
            "DMULp  $dst,ST" %}
  ins_encode %{
    __ fld_d($constantaddress($con));
    __ fmulp($dst$$reg);
  %}
  ins_pipe(fpu_reg_mem);
%}


instruct mulDPR_reg_mem(regDPR dst, memory src) %{
  predicate( UseSSE<=1 );
  match(Set dst (MulD dst (LoadD src)));
  ins_cost(200);
  format %{ "FLD_D  $src\n\t"
            "DMULp  $dst,ST" %}
  opcode(0xDE, 0x1, 0xDD); /* DE C8+i or DE /1*/  /* LoadD  DD /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_mem );
%}

//
// Cisc-alternate to reg-reg multiply
instruct mulDPR_reg_mem_cisc(regDPR dst, regDPR src, memory mem) %{
  predicate( UseSSE<=1 );
  match(Set dst (MulD src (LoadD mem)));
  ins_cost(250);
  format %{ "FLD_D  $mem\n\t"
            "DMUL   ST,$src\n\t"
            "FSTP_D $dst" %}
  opcode(0xD8, 0x1, 0xD9); /* D8 C8+i */  /* LoadD D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,mem),
              OpcReg_FPR(src),
              Pop_Reg_DPR(dst) );
  ins_pipe( fpu_reg_reg_mem );
%}


// MACRO3 -- addDPR a mulDPR
// This instruction is a '2-address' instruction in that the result goes
// back to src2.  This eliminates a move from the macro; possibly the
// register allocator will have to add it back (and maybe not).
instruct addDPR_mulDPR_reg(regDPR src2, regDPR src1, regDPR src0) %{
  predicate( UseSSE<=1 );
  match(Set src2 (AddD (MulD src0 src1) src2));
  format %{ "FLD    $src0\t# ===MACRO3d===\n\t"
            "DMUL   ST,$src1\n\t"
            "DADDp  $src2,ST" %}
  ins_cost(250);
  opcode(0xDD); /* LoadD DD /0 */
  ins_encode( Push_Reg_FPR(src0),
              FMul_ST_reg(src1),
              FAddP_reg_ST(src2) );
  ins_pipe( fpu_reg_reg_reg );
%}


// MACRO3 -- subDPR a mulDPR
instruct subDPR_mulDPR_reg(regDPR src2, regDPR src1, regDPR src0) %{
  predicate( UseSSE<=1 );
  match(Set src2 (SubD (MulD src0 src1) src2));
  format %{ "FLD    $src0\t# ===MACRO3d===\n\t"
            "DMUL   ST,$src1\n\t"
            "DSUBRp $src2,ST" %}
  ins_cost(250);
  ins_encode( Push_Reg_FPR(src0),
              FMul_ST_reg(src1),
              Opcode(0xDE), Opc_plus(0xE0,src2));
  ins_pipe( fpu_reg_reg_reg );
%}


instruct divDPR_reg(regDPR dst, regDPR src) %{
  predicate( UseSSE<=1 );
  match(Set dst (DivD dst src));

  format %{ "FLD    $src\n\t"
            "FDIVp  $dst,ST" %}
  opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
  ins_cost(150);
  ins_encode( Push_Reg_DPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}

// Strict FP instruction biases argument before division then
// biases result, to avoid double rounding of subnormals.
//
// scale dividend by multiplying dividend by 2^(-15360)
// load divisor
// divide scaled dividend by divisor
// rescale quotient by 2^(15360)
//
instruct strictfp_divDPR_reg(regDPR1 dst, regnotDPR1 src) %{
  predicate (UseSSE<=1);
  match(Set dst (DivD dst src));
  predicate( UseSSE<=1 && Compile::current()->has_method() && Compile::current()->method()->is_strict() );
  ins_cost(01);

  format %{ "FLD    StubRoutines::_fpu_subnormal_bias1\n\t"
            "DMULp  $dst,ST\n\t"
            "FLD    $src\n\t"
            "FDIVp  $dst,ST\n\t"
            "FLD    StubRoutines::_fpu_subnormal_bias2\n\t"
            "DMULp  $dst,ST\n\t" %}
  opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
  ins_encode( strictfp_bias1(dst),
              Push_Reg_DPR(src),
              OpcP, RegOpc(dst),
              strictfp_bias2(dst) );
  ins_pipe( fpu_reg_reg );
%}

instruct divDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
  predicate( UseSSE<=1 && !(Compile::current()->has_method() && Compile::current()->method()->is_strict()) );
  match(Set dst (RoundDouble (DivD src1 src2)));

  format %{ "FLD    $src1\n\t"
            "FDIV   ST,$src2\n\t"
            "FSTP_D $dst\t# D-round" %}
  opcode(0xD8, 0x6); /* D8 F0+i or D8 /6 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2), Pop_Mem_DPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}


instruct modDPR_reg(regDPR dst, regDPR src, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE<=1);
  match(Set dst (ModD dst src));
  effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS

  format %{ "DMOD   $dst,$src" %}
  ins_cost(250);
  ins_encode(Push_Reg_Mod_DPR(dst, src),
              emitModDPR(),
              Push_Result_Mod_DPR(src),
              Pop_Reg_DPR(dst));
  ins_pipe( pipe_slow );
%}

instruct modD_reg(regD dst, regD src0, regD src1, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE>=2);
  match(Set dst (ModD src0 src1));
  effect(KILL rax, KILL cr);

  format %{ "SUB    ESP,8\t # DMOD\n"
          "\tMOVSD  [ESP+0],$src1\n"
          "\tFLD_D  [ESP+0]\n"
          "\tMOVSD  [ESP+0],$src0\n"
          "\tFLD_D  [ESP+0]\n"
     "loop:\tFPREM\n"
          "\tFWAIT\n"
          "\tFNSTSW AX\n"
          "\tSAHF\n"
          "\tJP     loop\n"
          "\tFSTP_D [ESP+0]\n"
          "\tMOVSD  $dst,[ESP+0]\n"
          "\tADD    ESP,8\n"
          "\tFSTP   ST0\t # Restore FPU Stack"
    %}
  ins_cost(250);
  ins_encode( Push_ModD_encoding(src0, src1), emitModDPR(), Push_ResultD(dst), PopFPU);
  ins_pipe( pipe_slow );
%}

instruct sinDPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate (UseSSE<=1);
  match(Set dst (SinD src));
  ins_cost(1800);
  format %{ "DSIN   $dst" %}
  opcode(0xD9, 0xFE);
  ins_encode( OpcP, OpcS );
  ins_pipe( pipe_slow );
%}

instruct sinD_reg(regD dst, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst (SinD dst));
  effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
  ins_cost(1800);
  format %{ "DSIN   $dst" %}
  opcode(0xD9, 0xFE);
  ins_encode( Push_SrcD(dst), OpcP, OpcS, Push_ResultD(dst) );
  ins_pipe( pipe_slow );
%}

instruct cosDPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate (UseSSE<=1);
  match(Set dst (CosD src));
  ins_cost(1800);
  format %{ "DCOS   $dst" %}
  opcode(0xD9, 0xFF);
  ins_encode( OpcP, OpcS );
  ins_pipe( pipe_slow );
%}

instruct cosD_reg(regD dst, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst (CosD dst));
  effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
  ins_cost(1800);
  format %{ "DCOS   $dst" %}
  opcode(0xD9, 0xFF);
  ins_encode( Push_SrcD(dst), OpcP, OpcS, Push_ResultD(dst) );
  ins_pipe( pipe_slow );
%}

instruct tanDPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate (UseSSE<=1);
  match(Set dst(TanD src));
  format %{ "DTAN   $dst" %}
  ins_encode( Opcode(0xD9), Opcode(0xF2),    // fptan
              Opcode(0xDD), Opcode(0xD8));   // fstp st
  ins_pipe( pipe_slow );
%}

instruct tanD_reg(regD dst, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst(TanD dst));
  effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
  format %{ "DTAN   $dst" %}
  ins_encode( Push_SrcD(dst),
              Opcode(0xD9), Opcode(0xF2),    // fptan
              Opcode(0xDD), Opcode(0xD8),   // fstp st
              Push_ResultD(dst) );
  ins_pipe( pipe_slow );
%}

instruct atanDPR_reg(regDPR dst, regDPR src) %{
  predicate (UseSSE<=1);
  match(Set dst(AtanD dst src));
  format %{ "DATA   $dst,$src" %}
  opcode(0xD9, 0xF3);
  ins_encode( Push_Reg_DPR(src),
              OpcP, OpcS, RegOpc(dst) );
  ins_pipe( pipe_slow );
%}

instruct atanD_reg(regD dst, regD src, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst(AtanD dst src));
  effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
  format %{ "DATA   $dst,$src" %}
  opcode(0xD9, 0xF3);
  ins_encode( Push_SrcD(src),
              OpcP, OpcS, Push_ResultD(dst) );
  ins_pipe( pipe_slow );
%}

instruct sqrtDPR_reg(regDPR dst, regDPR src) %{
  predicate (UseSSE<=1);
  match(Set dst (SqrtD src));
  format %{ "DSQRT  $dst,$src" %}
  opcode(0xFA, 0xD9);
  ins_encode( Push_Reg_DPR(src),
              OpcS, OpcP, Pop_Reg_DPR(dst) );
  ins_pipe( pipe_slow );
%}

instruct powDPR_reg(regDPR X, regDPR1 Y, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
  predicate (UseSSE<=1);
  match(Set Y (PowD X Y));  // Raise X to the Yth power
  effect(KILL rax, KILL rdx, KILL rcx, KILL cr);
  format %{ "fast_pow $X $Y -> $Y  // KILL $rax, $rcx, $rdx" %}
  ins_encode %{
    __ subptr(rsp, 8);
    __ fld_s($X$$reg - 1);
    __ fast_pow();
    __ addptr(rsp, 8);
  %}
  ins_pipe( pipe_slow );
%}

instruct powD_reg(regD dst, regD src0, regD src1, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst (PowD src0 src1));  // Raise src0 to the src1'th power
  effect(KILL rax, KILL rdx, KILL rcx, KILL cr);
  format %{ "fast_pow $src0 $src1 -> $dst  // KILL $rax, $rcx, $rdx" %}
  ins_encode %{
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src1$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ movdbl(Address(rsp, 0), $src0$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ fast_pow();
    __ fstp_d(Address(rsp, 0));
    __ movdbl($dst$$XMMRegister, Address(rsp, 0));
    __ addptr(rsp, 8);
  %}
  ins_pipe( pipe_slow );
%}


instruct expDPR_reg(regDPR1 dpr1, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
  predicate (UseSSE<=1);
  match(Set dpr1 (ExpD dpr1));
  effect(KILL rax, KILL rcx, KILL rdx, KILL cr);
  format %{ "fast_exp $dpr1 -> $dpr1  // KILL $rax, $rcx, $rdx" %}
  ins_encode %{
    __ fast_exp();
  %}
  ins_pipe( pipe_slow );
%}

instruct expD_reg(regD dst, regD src, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst (ExpD src));
  effect(KILL rax, KILL rcx, KILL rdx, KILL cr);
  format %{ "fast_exp $dst -> $src  // KILL $rax, $rcx, $rdx" %}
  ins_encode %{
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ fast_exp();
    __ fstp_d(Address(rsp, 0));
    __ movdbl($dst$$XMMRegister, Address(rsp, 0));
    __ addptr(rsp, 8);
  %}
  ins_pipe( pipe_slow );
%}

instruct log10DPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate (UseSSE<=1);
  // The source Double operand on FPU stack
  match(Set dst (Log10D src));
  // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
  // fxch         ; swap ST(0) with ST(1)
  // fyl2x        ; compute log_10(2) * log_2(x)
  format %{ "FLDLG2 \t\t\t#Log10\n\t"
            "FXCH   \n\t"
            "FYL2X  \t\t\t# Q=Log10*Log_2(x)"
         %}
  ins_encode( Opcode(0xD9), Opcode(0xEC),   // fldlg2
              Opcode(0xD9), Opcode(0xC9),   // fxch
              Opcode(0xD9), Opcode(0xF1));  // fyl2x

  ins_pipe( pipe_slow );
%}

instruct log10D_reg(regD dst, regD src, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  effect(KILL cr);
  match(Set dst (Log10D src));
  // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
  // fyl2x        ; compute log_10(2) * log_2(x)
  format %{ "FLDLG2 \t\t\t#Log10\n\t"
            "FYL2X  \t\t\t# Q=Log10*Log_2(x)"
         %}
  ins_encode( Opcode(0xD9), Opcode(0xEC),   // fldlg2
              Push_SrcD(src),
              Opcode(0xD9), Opcode(0xF1),   // fyl2x
              Push_ResultD(dst));

  ins_pipe( pipe_slow );
%}

instruct logDPR_reg(regDPR1 dst, regDPR1 src) %{
  predicate (UseSSE<=1);
  // The source Double operand on FPU stack
  match(Set dst (LogD src));
  // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
  // fxch         ; swap ST(0) with ST(1)
  // fyl2x        ; compute log_e(2) * log_2(x)
  format %{ "FLDLN2 \t\t\t#Log_e\n\t"
            "FXCH   \n\t"
            "FYL2X  \t\t\t# Q=Log_e*Log_2(x)"
         %}
  ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
              Opcode(0xD9), Opcode(0xC9),   // fxch
              Opcode(0xD9), Opcode(0xF1));  // fyl2x

  ins_pipe( pipe_slow );
%}

instruct logD_reg(regD dst, regD src, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  effect(KILL cr);
  // The source and result Double operands in XMM registers
  match(Set dst (LogD src));
  // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
  // fyl2x        ; compute log_e(2) * log_2(x)
  format %{ "FLDLN2 \t\t\t#Log_e\n\t"
            "FYL2X  \t\t\t# Q=Log_e*Log_2(x)"
         %}
  ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
              Push_SrcD(src),
              Opcode(0xD9), Opcode(0xF1),   // fyl2x
              Push_ResultD(dst));
  ins_pipe( pipe_slow );
%}

//-------------Float Instructions-------------------------------
// Float Math

// Code for float compare:
//     fcompp();
//     fwait(); fnstsw_ax();
//     sahf();
//     movl(dst, unordered_result);
//     jcc(Assembler::parity, exit);
//     movl(dst, less_result);
//     jcc(Assembler::below, exit);
//     movl(dst, equal_result);
//     jcc(Assembler::equal, exit);
//     movl(dst, greater_result);
//   exit:

// P6 version of float compare, sets condition codes in EFLAGS
instruct cmpFPR_cc_P6(eFlagsRegU cr, regFPR src1, regFPR src2, eAXRegI rax) %{
  predicate(VM_Version::supports_cmov() && UseSSE == 0);
  match(Set cr (CmpF src1 src2));
  effect(KILL rax);
  ins_cost(150);
  format %{ "FLD    $src1\n\t"
            "FUCOMIP ST,$src2  // P6 instruction\n\t"
            "JNP    exit\n\t"
            "MOV    ah,1       // saw a NaN, set CF (treat as LT)\n\t"
            "SAHF\n"
     "exit:\tNOP               // avoid branch to branch" %}
  opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2),
              cmpF_P6_fixup );
  ins_pipe( pipe_slow );
%}

instruct cmpFPR_cc_P6CF(eFlagsRegUCF cr, regFPR src1, regFPR src2) %{
  predicate(VM_Version::supports_cmov() && UseSSE == 0);
  match(Set cr (CmpF src1 src2));
  ins_cost(100);
  format %{ "FLD    $src1\n\t"
            "FUCOMIP ST,$src2  // P6 instruction" %}
  opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2));
  ins_pipe( pipe_slow );
%}


// Compare & branch
instruct cmpFPR_cc(eFlagsRegU cr, regFPR src1, regFPR src2, eAXRegI rax) %{
  predicate(UseSSE == 0);
  match(Set cr (CmpF src1 src2));
  effect(KILL rax);
  ins_cost(200);
  format %{ "FLD    $src1\n\t"
            "FCOMp  $src2\n\t"
            "FNSTSW AX\n\t"
            "TEST   AX,0x400\n\t"
            "JZ,s   flags\n\t"
            "MOV    AH,1\t# unordered treat as LT\n"
    "flags:\tSAHF" %}
  opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2),
              fpu_flags);
  ins_pipe( pipe_slow );
%}

// Compare vs zero into -1,0,1
instruct cmpFPR_0(rRegI dst, regFPR src1, immFPR0 zero, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE == 0);
  match(Set dst (CmpF3 src1 zero));
  effect(KILL cr, KILL rax);
  ins_cost(280);
  format %{ "FTSTF  $dst,$src1" %}
  opcode(0xE4, 0xD9);
  ins_encode( Push_Reg_DPR(src1),
              OpcS, OpcP, PopFPU,
              CmpF_Result(dst));
  ins_pipe( pipe_slow );
%}

// Compare into -1,0,1
instruct cmpFPR_reg(rRegI dst, regFPR src1, regFPR src2, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE == 0);
  match(Set dst (CmpF3 src1 src2));
  effect(KILL cr, KILL rax);
  ins_cost(300);
  format %{ "FCMPF  $dst,$src1,$src2" %}
  opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
  ins_encode( Push_Reg_DPR(src1),
              OpcP, RegOpc(src2),
              CmpF_Result(dst));
  ins_pipe( pipe_slow );
%}

// float compare and set condition codes in EFLAGS by XMM regs
instruct cmpF_cc(eFlagsRegU cr, regF src1, regF src2) %{
  predicate(UseSSE>=1);
  match(Set cr (CmpF src1 src2));
  ins_cost(145);
  format %{ "UCOMISS $src1,$src2\n\t"
            "JNP,s   exit\n\t"
            "PUSHF\t# saw NaN, set CF\n\t"
            "AND     [rsp], #0xffffff2b\n\t"
            "POPF\n"
    "exit:" %}
  ins_encode %{
    __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
    emit_cmpfp_fixup(_masm);
  %}
  ins_pipe( pipe_slow );
%}

instruct cmpF_ccCF(eFlagsRegUCF cr, regF src1, regF src2) %{
  predicate(UseSSE>=1);
  match(Set cr (CmpF src1 src2));
  ins_cost(100);
  format %{ "UCOMISS $src1,$src2" %}
  ins_encode %{
    __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// float compare and set condition codes in EFLAGS by XMM regs
instruct cmpF_ccmem(eFlagsRegU cr, regF src1, memory src2) %{
  predicate(UseSSE>=1);
  match(Set cr (CmpF src1 (LoadF src2)));
  ins_cost(165);
  format %{ "UCOMISS $src1,$src2\n\t"
            "JNP,s   exit\n\t"
            "PUSHF\t# saw NaN, set CF\n\t"
            "AND     [rsp], #0xffffff2b\n\t"
            "POPF\n"
    "exit:" %}
  ins_encode %{
    __ ucomiss($src1$$XMMRegister, $src2$$Address);
    emit_cmpfp_fixup(_masm);
  %}
  ins_pipe( pipe_slow );
%}

instruct cmpF_ccmemCF(eFlagsRegUCF cr, regF src1, memory src2) %{
  predicate(UseSSE>=1);
  match(Set cr (CmpF src1 (LoadF src2)));
  ins_cost(100);
  format %{ "UCOMISS $src1,$src2" %}
  ins_encode %{
    __ ucomiss($src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Compare into -1,0,1 in XMM
instruct cmpF_reg(xRegI dst, regF src1, regF src2, eFlagsReg cr) %{
  predicate(UseSSE>=1);
  match(Set dst (CmpF3 src1 src2));
  effect(KILL cr);
  ins_cost(255);
  format %{ "UCOMISS $src1, $src2\n\t"
            "MOV     $dst, #-1\n\t"
            "JP,s    done\n\t"
            "JB,s    done\n\t"
            "SETNE   $dst\n\t"
            "MOVZB   $dst, $dst\n"
    "done:" %}
  ins_encode %{
    __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
    emit_cmpfp3(_masm, $dst$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// Compare into -1,0,1 in XMM and memory
instruct cmpF_regmem(xRegI dst, regF src1, memory src2, eFlagsReg cr) %{
  predicate(UseSSE>=1);
  match(Set dst (CmpF3 src1 (LoadF src2)));
  effect(KILL cr);
  ins_cost(275);
  format %{ "UCOMISS $src1, $src2\n\t"
            "MOV     $dst, #-1\n\t"
            "JP,s    done\n\t"
            "JB,s    done\n\t"
            "SETNE   $dst\n\t"
            "MOVZB   $dst, $dst\n"
    "done:" %}
  ins_encode %{
    __ ucomiss($src1$$XMMRegister, $src2$$Address);
    emit_cmpfp3(_masm, $dst$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// Spill to obtain 24-bit precision
instruct subFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (SubF src1 src2));

  format %{ "FSUB   $dst,$src1 - $src2" %}
  opcode(0xD8, 0x4); /* D8 E0+i or D8 /4 mod==0x3 ;; result in TOS */
  ins_encode( Push_Reg_FPR(src1),
              OpcReg_FPR(src2),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}
//
// This instruction does not round to 24-bits
instruct subFPR_reg(regFPR dst, regFPR src) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (SubF dst src));

  format %{ "FSUB   $dst,$src" %}
  opcode(0xDE, 0x5); /* DE E8+i  or DE /5 */
  ins_encode( Push_Reg_FPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}

// Spill to obtain 24-bit precision
instruct addFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src1 src2));

  format %{ "FADD   $dst,$src1,$src2" %}
  opcode(0xD8, 0x0); /* D8 C0+i */
  ins_encode( Push_Reg_FPR(src2),
              OpcReg_FPR(src1),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}
//
// This instruction does not round to 24-bits
instruct addFPR_reg(regFPR dst, regFPR src) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (AddF dst src));

  format %{ "FLD    $src\n\t"
            "FADDp  $dst,ST" %}
  opcode(0xDE, 0x0); /* DE C0+i or DE /0*/
  ins_encode( Push_Reg_FPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}

instruct absFPR_reg(regFPR1 dst, regFPR1 src) %{
  predicate(UseSSE==0);
  match(Set dst (AbsF src));
  ins_cost(100);
  format %{ "FABS" %}
  opcode(0xE1, 0xD9);
  ins_encode( OpcS, OpcP );
  ins_pipe( fpu_reg_reg );
%}

instruct negFPR_reg(regFPR1 dst, regFPR1 src) %{
  predicate(UseSSE==0);
  match(Set dst (NegF src));
  ins_cost(100);
  format %{ "FCHS" %}
  opcode(0xE0, 0xD9);
  ins_encode( OpcS, OpcP );
  ins_pipe( fpu_reg_reg );
%}

// Cisc-alternate to addFPR_reg
// Spill to obtain 24-bit precision
instruct addFPR24_reg_mem(stackSlotF dst, regFPR src1, memory src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src1 (LoadF src2)));

  format %{ "FLD    $src2\n\t"
            "FADD   ST,$src1\n\t"
            "FSTP_S $dst" %}
  opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
              OpcReg_FPR(src1),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_mem );
%}
//
// Cisc-alternate to addFPR_reg
// This instruction does not round to 24-bits
instruct addFPR_reg_mem(regFPR dst, memory src) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (AddF dst (LoadF src)));

  format %{ "FADD   $dst,$src" %}
  opcode(0xDE, 0x0, 0xD9); /* DE C0+i or DE /0*/  /* LoadF  D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_mem );
%}

// // Following two instructions for _222_mpegaudio
// Spill to obtain 24-bit precision
instruct addFPR24_mem_reg(stackSlotF dst, regFPR src2, memory src1 ) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src1 src2));

  format %{ "FADD   $dst,$src1,$src2" %}
  opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src1),
              OpcReg_FPR(src2),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_mem );
%}

// Cisc-spill variant
// Spill to obtain 24-bit precision
instruct addFPR24_mem_cisc(stackSlotF dst, memory src1, memory src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src1 (LoadF src2)));

  format %{ "FADD   $dst,$src1,$src2 cisc" %}
  opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
              set_instruction_start,
              OpcP, RMopc_Mem(secondary,src1),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_mem_mem );
%}

// Spill to obtain 24-bit precision
instruct addFPR24_mem_mem(stackSlotF dst, memory src1, memory src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src1 src2));

  format %{ "FADD   $dst,$src1,$src2" %}
  opcode(0xD8, 0x0, 0xD9); /* D8 /0 */  /* LoadF  D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
              set_instruction_start,
              OpcP, RMopc_Mem(secondary,src1),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_mem_mem );
%}


// Spill to obtain 24-bit precision
instruct addFPR24_reg_imm(stackSlotF dst, regFPR src, immFPR con) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src con));
  format %{ "FLD    $src\n\t"
            "FADD_S [$constantaddress]\t# load from constant table: float=$con\n\t"
            "FSTP_S $dst"  %}
  ins_encode %{
    __ fld_s($src$$reg - 1);  // FLD ST(i-1)
    __ fadd_s($constantaddress($con));
    __ fstp_s(Address(rsp, $dst$$disp));
  %}
  ins_pipe(fpu_mem_reg_con);
%}
//
// This instruction does not round to 24-bits
instruct addFPR_reg_imm(regFPR dst, regFPR src, immFPR con) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (AddF src con));
  format %{ "FLD    $src\n\t"
            "FADD_S [$constantaddress]\t# load from constant table: float=$con\n\t"
            "FSTP   $dst"  %}
  ins_encode %{
    __ fld_s($src$$reg - 1);  // FLD ST(i-1)
    __ fadd_s($constantaddress($con));
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_reg_con);
%}

// Spill to obtain 24-bit precision
instruct mulFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src1 src2));

  format %{ "FLD    $src1\n\t"
            "FMUL   $src2\n\t"
            "FSTP_S $dst"  %}
  opcode(0xD8, 0x1); /* D8 C8+i or D8 /1 ;; result in TOS */
  ins_encode( Push_Reg_FPR(src1),
              OpcReg_FPR(src2),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}
//
// This instruction does not round to 24-bits
instruct mulFPR_reg(regFPR dst, regFPR src1, regFPR src2) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src1 src2));

  format %{ "FLD    $src1\n\t"
            "FMUL   $src2\n\t"
            "FSTP_S $dst"  %}
  opcode(0xD8, 0x1); /* D8 C8+i */
  ins_encode( Push_Reg_FPR(src2),
              OpcReg_FPR(src1),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_reg_reg );
%}


// Spill to obtain 24-bit precision
// Cisc-alternate to reg-reg multiply
instruct mulFPR24_reg_mem(stackSlotF dst, regFPR src1, memory src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src1 (LoadF src2)));

  format %{ "FLD_S  $src2\n\t"
            "FMUL   $src1\n\t"
            "FSTP_S $dst"  %}
  opcode(0xD8, 0x1, 0xD9); /* D8 C8+i or DE /1*/  /* LoadF D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
              OpcReg_FPR(src1),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_mem );
%}
//
// This instruction does not round to 24-bits
// Cisc-alternate to reg-reg multiply
instruct mulFPR_reg_mem(regFPR dst, regFPR src1, memory src2) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src1 (LoadF src2)));

  format %{ "FMUL   $dst,$src1,$src2" %}
  opcode(0xD8, 0x1, 0xD9); /* D8 C8+i */  /* LoadF D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
              OpcReg_FPR(src1),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_reg_mem );
%}

// Spill to obtain 24-bit precision
instruct mulFPR24_mem_mem(stackSlotF dst, memory src1, memory src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src1 src2));

  format %{ "FMUL   $dst,$src1,$src2" %}
  opcode(0xD8, 0x1, 0xD9); /* D8 /1 */  /* LoadF D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
              set_instruction_start,
              OpcP, RMopc_Mem(secondary,src1),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_mem_mem );
%}

// Spill to obtain 24-bit precision
instruct mulFPR24_reg_imm(stackSlotF dst, regFPR src, immFPR con) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src con));

  format %{ "FLD    $src\n\t"
            "FMUL_S [$constantaddress]\t# load from constant table: float=$con\n\t"
            "FSTP_S $dst"  %}
  ins_encode %{
    __ fld_s($src$$reg - 1);  // FLD ST(i-1)
    __ fmul_s($constantaddress($con));
    __ fstp_s(Address(rsp, $dst$$disp));
  %}
  ins_pipe(fpu_mem_reg_con);
%}
//
// This instruction does not round to 24-bits
instruct mulFPR_reg_imm(regFPR dst, regFPR src, immFPR con) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (MulF src con));

  format %{ "FLD    $src\n\t"
            "FMUL_S [$constantaddress]\t# load from constant table: float=$con\n\t"
            "FSTP   $dst"  %}
  ins_encode %{
    __ fld_s($src$$reg - 1);  // FLD ST(i-1)
    __ fmul_s($constantaddress($con));
    __ fstp_d($dst$$reg);
  %}
  ins_pipe(fpu_reg_reg_con);
%}


//
// MACRO1 -- subsume unshared load into mulFPR
// This instruction does not round to 24-bits
instruct mulFPR_reg_load1(regFPR dst, regFPR src, memory mem1 ) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (MulF (LoadF mem1) src));

  format %{ "FLD    $mem1    ===MACRO1===\n\t"
            "FMUL   ST,$src\n\t"
            "FSTP   $dst" %}
  opcode(0xD8, 0x1, 0xD9); /* D8 C8+i or D8 /1 */  /* LoadF D9 /0 */
  ins_encode( Opcode(tertiary), RMopc_Mem(0x00,mem1),
              OpcReg_FPR(src),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_reg_mem );
%}
//
// MACRO2 -- addFPR a mulFPR which subsumed an unshared load
// This instruction does not round to 24-bits
instruct addFPR_mulFPR_reg_load1(regFPR dst, memory mem1, regFPR src1, regFPR src2) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (AddF (MulF (LoadF mem1) src1) src2));
  ins_cost(95);

  format %{ "FLD    $mem1     ===MACRO2===\n\t"
            "FMUL   ST,$src1  subsume mulFPR left load\n\t"
            "FADD   ST,$src2\n\t"
            "FSTP   $dst" %}
  opcode(0xD9); /* LoadF D9 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem1),
              FMul_ST_reg(src1),
              FAdd_ST_reg(src2),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_mem_reg_reg );
%}

// MACRO3 -- addFPR a mulFPR
// This instruction does not round to 24-bits.  It is a '2-address'
// instruction in that the result goes back to src2.  This eliminates
// a move from the macro; possibly the register allocator will have
// to add it back (and maybe not).
instruct addFPR_mulFPR_reg(regFPR src2, regFPR src1, regFPR src0) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set src2 (AddF (MulF src0 src1) src2));

  format %{ "FLD    $src0     ===MACRO3===\n\t"
            "FMUL   ST,$src1\n\t"
            "FADDP  $src2,ST" %}
  opcode(0xD9); /* LoadF D9 /0 */
  ins_encode( Push_Reg_FPR(src0),
              FMul_ST_reg(src1),
              FAddP_reg_ST(src2) );
  ins_pipe( fpu_reg_reg_reg );
%}

// MACRO4 -- divFPR subFPR
// This instruction does not round to 24-bits
instruct subFPR_divFPR_reg(regFPR dst, regFPR src1, regFPR src2, regFPR src3) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (DivF (SubF src2 src1) src3));

  format %{ "FLD    $src2   ===MACRO4===\n\t"
            "FSUB   ST,$src1\n\t"
            "FDIV   ST,$src3\n\t"
            "FSTP  $dst" %}
  opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
  ins_encode( Push_Reg_FPR(src2),
              subFPR_divFPR_encode(src1,src3),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_reg_reg_reg );
%}

// Spill to obtain 24-bit precision
instruct divFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
  predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (DivF src1 src2));

  format %{ "FDIV   $dst,$src1,$src2" %}
  opcode(0xD8, 0x6); /* D8 F0+i or DE /6*/
  ins_encode( Push_Reg_FPR(src1),
              OpcReg_FPR(src2),
              Pop_Mem_FPR(dst) );
  ins_pipe( fpu_mem_reg_reg );
%}
//
// This instruction does not round to 24-bits
instruct divFPR_reg(regFPR dst, regFPR src) %{
  predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (DivF dst src));

  format %{ "FDIV   $dst,$src" %}
  opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
  ins_encode( Push_Reg_FPR(src),
              OpcP, RegOpc(dst) );
  ins_pipe( fpu_reg_reg );
%}


// Spill to obtain 24-bit precision
instruct modFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2, eAXRegI rax, eFlagsReg cr) %{
  predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (ModF src1 src2));
  effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS

  format %{ "FMOD   $dst,$src1,$src2" %}
  ins_encode( Push_Reg_Mod_DPR(src1, src2),
              emitModDPR(),
              Push_Result_Mod_DPR(src2),
              Pop_Mem_FPR(dst));
  ins_pipe( pipe_slow );
%}
//
// This instruction does not round to 24-bits
instruct modFPR_reg(regFPR dst, regFPR src, eAXRegI rax, eFlagsReg cr) %{
  predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (ModF dst src));
  effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS

  format %{ "FMOD   $dst,$src" %}
  ins_encode(Push_Reg_Mod_DPR(dst, src),
              emitModDPR(),
              Push_Result_Mod_DPR(src),
              Pop_Reg_FPR(dst));
  ins_pipe( pipe_slow );
%}

instruct modF_reg(regF dst, regF src0, regF src1, eAXRegI rax, eFlagsReg cr) %{
  predicate(UseSSE>=1);
  match(Set dst (ModF src0 src1));
  effect(KILL rax, KILL cr);
  format %{ "SUB    ESP,4\t # FMOD\n"
          "\tMOVSS  [ESP+0],$src1\n"
          "\tFLD_S  [ESP+0]\n"
          "\tMOVSS  [ESP+0],$src0\n"
          "\tFLD_S  [ESP+0]\n"
     "loop:\tFPREM\n"
          "\tFWAIT\n"
          "\tFNSTSW AX\n"
          "\tSAHF\n"
          "\tJP     loop\n"
          "\tFSTP_S [ESP+0]\n"
          "\tMOVSS  $dst,[ESP+0]\n"
          "\tADD    ESP,4\n"
          "\tFSTP   ST0\t # Restore FPU Stack"
    %}
  ins_cost(250);
  ins_encode( Push_ModF_encoding(src0, src1), emitModDPR(), Push_ResultF(dst,0x4), PopFPU);
  ins_pipe( pipe_slow );
%}


//----------Arithmetic Conversion Instructions---------------------------------
// The conversions operations are all Alpha sorted.  Please keep it that way!

instruct roundFloat_mem_reg(stackSlotF dst, regFPR src) %{
  predicate(UseSSE==0);
  match(Set dst (RoundFloat src));
  ins_cost(125);
  format %{ "FST_S  $dst,$src\t# F-round" %}
  ins_encode( Pop_Mem_Reg_FPR(dst, src) );
  ins_pipe( fpu_mem_reg );
%}

instruct roundDouble_mem_reg(stackSlotD dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (RoundDouble src));
  ins_cost(125);
  format %{ "FST_D  $dst,$src\t# D-round" %}
  ins_encode( Pop_Mem_Reg_DPR(dst, src) );
  ins_pipe( fpu_mem_reg );
%}

// Force rounding to 24-bit precision and 6-bit exponent
instruct convDPR2FPR_reg(stackSlotF dst, regDPR src) %{
  predicate(UseSSE==0);
  match(Set dst (ConvD2F src));
  format %{ "FST_S  $dst,$src\t# F-round" %}
  expand %{
    roundFloat_mem_reg(dst,src);
  %}
%}

// Force rounding to 24-bit precision and 6-bit exponent
instruct convDPR2F_reg(regF dst, regDPR src, eFlagsReg cr) %{
  predicate(UseSSE==1);
  match(Set dst (ConvD2F src));
  effect( KILL cr );
  format %{ "SUB    ESP,4\n\t"
            "FST_S  [ESP],$src\t# F-round\n\t"
            "MOVSS  $dst,[ESP]\n\t"
            "ADD ESP,4" %}
  ins_encode %{
    __ subptr(rsp, 4);
    if ($src$$reg != FPR1L_enc) {
      __ fld_s($src$$reg-1);
      __ fstp_s(Address(rsp, 0));
    } else {
      __ fst_s(Address(rsp, 0));
    }
    __ movflt($dst$$XMMRegister, Address(rsp, 0));
    __ addptr(rsp, 4);
  %}
  ins_pipe( pipe_slow );
%}

// Force rounding double precision to single precision
instruct convD2F_reg(regF dst, regD src) %{
  predicate(UseSSE>=2);
  match(Set dst (ConvD2F src));
  format %{ "CVTSD2SS $dst,$src\t# F-round" %}
  ins_encode %{
    __ cvtsd2ss ($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct convFPR2DPR_reg_reg(regDPR dst, regFPR src) %{
  predicate(UseSSE==0);
  match(Set dst (ConvF2D src));
  format %{ "FST_S  $dst,$src\t# D-round" %}
  ins_encode( Pop_Reg_Reg_DPR(dst, src));
  ins_pipe( fpu_reg_reg );
%}

instruct convFPR2D_reg(stackSlotD dst, regFPR src) %{
  predicate(UseSSE==1);
  match(Set dst (ConvF2D src));
  format %{ "FST_D  $dst,$src\t# D-round" %}
  expand %{
    roundDouble_mem_reg(dst,src);
  %}
%}

instruct convF2DPR_reg(regDPR dst, regF src, eFlagsReg cr) %{
  predicate(UseSSE==1);
  match(Set dst (ConvF2D src));
  effect( KILL cr );
  format %{ "SUB    ESP,4\n\t"
            "MOVSS  [ESP] $src\n\t"
            "FLD_S  [ESP]\n\t"
            "ADD    ESP,4\n\t"
            "FSTP   $dst\t# D-round" %}
  ins_encode %{
    __ subptr(rsp, 4);
    __ movflt(Address(rsp, 0), $src$$XMMRegister);
    __ fld_s(Address(rsp, 0));
    __ addptr(rsp, 4);
    __ fstp_d($dst$$reg);
  %}
  ins_pipe( pipe_slow );
%}

instruct convF2D_reg(regD dst, regF src) %{
  predicate(UseSSE>=2);
  match(Set dst (ConvF2D src));
  format %{ "CVTSS2SD $dst,$src\t# D-round" %}
  ins_encode %{
    __ cvtss2sd ($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Convert a double to an int.  If the double is a NAN, stuff a zero in instead.
instruct convDPR2I_reg_reg( eAXRegI dst, eDXRegI tmp, regDPR src, eFlagsReg cr ) %{
  predicate(UseSSE<=1);
  match(Set dst (ConvD2I src));
  effect( KILL tmp, KILL cr );
  format %{ "FLD    $src\t# Convert double to int \n\t"
            "FLDCW  trunc mode\n\t"
            "SUB    ESP,4\n\t"
            "FISTp  [ESP + #0]\n\t"
            "FLDCW  std/24-bit mode\n\t"
            "POP    EAX\n\t"
            "CMP    EAX,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "FLD_D  $src\n\t"
            "CALL   d2i_wrapper\n"
      "fast:" %}
  ins_encode( Push_Reg_DPR(src), DPR2I_encoding(src) );
  ins_pipe( pipe_slow );
%}

// Convert a double to an int.  If the double is a NAN, stuff a zero in instead.
instruct convD2I_reg_reg( eAXRegI dst, eDXRegI tmp, regD src, eFlagsReg cr ) %{
  predicate(UseSSE>=2);
  match(Set dst (ConvD2I src));
  effect( KILL tmp, KILL cr );
  format %{ "CVTTSD2SI $dst, $src\n\t"
            "CMP    $dst,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "SUB    ESP, 8\n\t"
            "MOVSD  [ESP], $src\n\t"
            "FLD_D  [ESP]\n\t"
            "ADD    ESP, 8\n\t"
            "CALL   d2i_wrapper\n"
      "fast:" %}
  ins_encode %{
    Label fast;
    __ cvttsd2sil($dst$$Register, $src$$XMMRegister);
    __ cmpl($dst$$Register, 0x80000000);
    __ jccb(Assembler::notEqual, fast);
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ addptr(rsp, 8);
    __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2i_wrapper())));
    __ bind(fast);
  %}
  ins_pipe( pipe_slow );
%}

instruct convDPR2L_reg_reg( eADXRegL dst, regDPR src, eFlagsReg cr ) %{
  predicate(UseSSE<=1);
  match(Set dst (ConvD2L src));
  effect( KILL cr );
  format %{ "FLD    $src\t# Convert double to long\n\t"
            "FLDCW  trunc mode\n\t"
            "SUB    ESP,8\n\t"
            "FISTp  [ESP + #0]\n\t"
            "FLDCW  std/24-bit mode\n\t"
            "POP    EAX\n\t"
            "POP    EDX\n\t"
            "CMP    EDX,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "TEST   EAX,EAX\n\t"
            "JNE,s  fast\n\t"
            "FLD    $src\n\t"
            "CALL   d2l_wrapper\n"
      "fast:" %}
  ins_encode( Push_Reg_DPR(src),  DPR2L_encoding(src) );
  ins_pipe( pipe_slow );
%}

// XMM lacks a float/double->long conversion, so use the old FPU stack.
instruct convD2L_reg_reg( eADXRegL dst, regD src, eFlagsReg cr ) %{
  predicate (UseSSE>=2);
  match(Set dst (ConvD2L src));
  effect( KILL cr );
  format %{ "SUB    ESP,8\t# Convert double to long\n\t"
            "MOVSD  [ESP],$src\n\t"
            "FLD_D  [ESP]\n\t"
            "FLDCW  trunc mode\n\t"
            "FISTp  [ESP + #0]\n\t"
            "FLDCW  std/24-bit mode\n\t"
            "POP    EAX\n\t"
            "POP    EDX\n\t"
            "CMP    EDX,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "TEST   EAX,EAX\n\t"
            "JNE,s  fast\n\t"
            "SUB    ESP,8\n\t"
            "MOVSD  [ESP],$src\n\t"
            "FLD_D  [ESP]\n\t"
            "ADD    ESP,8\n\t"
            "CALL   d2l_wrapper\n"
      "fast:" %}
  ins_encode %{
    Label fast;
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
    __ fistp_d(Address(rsp, 0));
    // Restore the rounding mode, mask the exception
    if (Compile::current()->in_24_bit_fp_mode()) {
      __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
    } else {
      __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
    }
    // Load the converted long, adjust CPU stack
    __ pop(rax);
    __ pop(rdx);
    __ cmpl(rdx, 0x80000000);
    __ jccb(Assembler::notEqual, fast);
    __ testl(rax, rax);
    __ jccb(Assembler::notEqual, fast);
    __ subptr(rsp, 8);
    __ movdbl(Address(rsp, 0), $src$$XMMRegister);
    __ fld_d(Address(rsp, 0));
    __ addptr(rsp, 8);
    __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2l_wrapper())));
    __ bind(fast);
  %}
  ins_pipe( pipe_slow );
%}

// Convert a double to an int.  Java semantics require we do complex
// manglations in the corner cases.  So we set the rounding mode to
// 'zero', store the darned double down as an int, and reset the
// rounding mode to 'nearest'.  The hardware stores a flag value down
// if we would overflow or converted a NAN; we check for this and
// and go the slow path if needed.
instruct convFPR2I_reg_reg(eAXRegI dst, eDXRegI tmp, regFPR src, eFlagsReg cr ) %{
  predicate(UseSSE==0);
  match(Set dst (ConvF2I src));
  effect( KILL tmp, KILL cr );
  format %{ "FLD    $src\t# Convert float to int \n\t"
            "FLDCW  trunc mode\n\t"
            "SUB    ESP,4\n\t"
            "FISTp  [ESP + #0]\n\t"
            "FLDCW  std/24-bit mode\n\t"
            "POP    EAX\n\t"
            "CMP    EAX,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "FLD    $src\n\t"
            "CALL   d2i_wrapper\n"
      "fast:" %}
  // DPR2I_encoding works for FPR2I
  ins_encode( Push_Reg_FPR(src), DPR2I_encoding(src) );
  ins_pipe( pipe_slow );
%}

// Convert a float in xmm to an int reg.
instruct convF2I_reg(eAXRegI dst, eDXRegI tmp, regF src, eFlagsReg cr ) %{
  predicate(UseSSE>=1);
  match(Set dst (ConvF2I src));
  effect( KILL tmp, KILL cr );
  format %{ "CVTTSS2SI $dst, $src\n\t"
            "CMP    $dst,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "SUB    ESP, 4\n\t"
            "MOVSS  [ESP], $src\n\t"
            "FLD    [ESP]\n\t"
            "ADD    ESP, 4\n\t"
            "CALL   d2i_wrapper\n"
      "fast:" %}
  ins_encode %{
    Label fast;
    __ cvttss2sil($dst$$Register, $src$$XMMRegister);
    __ cmpl($dst$$Register, 0x80000000);
    __ jccb(Assembler::notEqual, fast);
    __ subptr(rsp, 4);
    __ movflt(Address(rsp, 0), $src$$XMMRegister);
    __ fld_s(Address(rsp, 0));
    __ addptr(rsp, 4);
    __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2i_wrapper())));
    __ bind(fast);
  %}
  ins_pipe( pipe_slow );
%}

instruct convFPR2L_reg_reg( eADXRegL dst, regFPR src, eFlagsReg cr ) %{
  predicate(UseSSE==0);
  match(Set dst (ConvF2L src));
  effect( KILL cr );
  format %{ "FLD    $src\t# Convert float to long\n\t"
            "FLDCW  trunc mode\n\t"
            "SUB    ESP,8\n\t"
            "FISTp  [ESP + #0]\n\t"
            "FLDCW  std/24-bit mode\n\t"
            "POP    EAX\n\t"
            "POP    EDX\n\t"
            "CMP    EDX,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "TEST   EAX,EAX\n\t"
            "JNE,s  fast\n\t"
            "FLD    $src\n\t"
            "CALL   d2l_wrapper\n"
      "fast:" %}
  // DPR2L_encoding works for FPR2L
  ins_encode( Push_Reg_FPR(src), DPR2L_encoding(src) );
  ins_pipe( pipe_slow );
%}

// XMM lacks a float/double->long conversion, so use the old FPU stack.
instruct convF2L_reg_reg( eADXRegL dst, regF src, eFlagsReg cr ) %{
  predicate (UseSSE>=1);
  match(Set dst (ConvF2L src));
  effect( KILL cr );
  format %{ "SUB    ESP,8\t# Convert float to long\n\t"
            "MOVSS  [ESP],$src\n\t"
            "FLD_S  [ESP]\n\t"
            "FLDCW  trunc mode\n\t"
            "FISTp  [ESP + #0]\n\t"
            "FLDCW  std/24-bit mode\n\t"
            "POP    EAX\n\t"
            "POP    EDX\n\t"
            "CMP    EDX,0x80000000\n\t"
            "JNE,s  fast\n\t"
            "TEST   EAX,EAX\n\t"
            "JNE,s  fast\n\t"
            "SUB    ESP,4\t# Convert float to long\n\t"
            "MOVSS  [ESP],$src\n\t"
            "FLD_S  [ESP]\n\t"
            "ADD    ESP,4\n\t"
            "CALL   d2l_wrapper\n"
      "fast:" %}
  ins_encode %{
    Label fast;
    __ subptr(rsp, 8);
    __ movflt(Address(rsp, 0), $src$$XMMRegister);
    __ fld_s(Address(rsp, 0));
    __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
    __ fistp_d(Address(rsp, 0));
    // Restore the rounding mode, mask the exception
    if (Compile::current()->in_24_bit_fp_mode()) {
      __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
    } else {
      __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
    }
    // Load the converted long, adjust CPU stack
    __ pop(rax);
    __ pop(rdx);
    __ cmpl(rdx, 0x80000000);
    __ jccb(Assembler::notEqual, fast);
    __ testl(rax, rax);
    __ jccb(Assembler::notEqual, fast);
    __ subptr(rsp, 4);
    __ movflt(Address(rsp, 0), $src$$XMMRegister);
    __ fld_s(Address(rsp, 0));
    __ addptr(rsp, 4);
    __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2l_wrapper())));
    __ bind(fast);
  %}
  ins_pipe( pipe_slow );
%}

instruct convI2DPR_reg(regDPR dst, stackSlotI src) %{
  predicate( UseSSE<=1 );
  match(Set dst (ConvI2D src));
  format %{ "FILD   $src\n\t"
            "FSTP   $dst" %}
  opcode(0xDB, 0x0);  /* DB /0 */
  ins_encode(Push_Mem_I(src), Pop_Reg_DPR(dst));
  ins_pipe( fpu_reg_mem );
%}

instruct convI2D_reg(regD dst, rRegI src) %{
  predicate( UseSSE>=2 && !UseXmmI2D );
  match(Set dst (ConvI2D src));
  format %{ "CVTSI2SD $dst,$src" %}
  ins_encode %{
    __ cvtsi2sdl ($dst$$XMMRegister, $src$$Register);
  %}
  ins_pipe( pipe_slow );
%}

instruct convI2D_mem(regD dst, memory mem) %{
  predicate( UseSSE>=2 );
  match(Set dst (ConvI2D (LoadI mem)));
  format %{ "CVTSI2SD $dst,$mem" %}
  ins_encode %{
    __ cvtsi2sdl ($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

instruct convXI2D_reg(regD dst, rRegI src)
%{
  predicate( UseSSE>=2 && UseXmmI2D );
  match(Set dst (ConvI2D src));

  format %{ "MOVD  $dst,$src\n\t"
            "CVTDQ2PD $dst,$dst\t# i2d" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe(pipe_slow); // XXX
%}

instruct convI2DPR_mem(regDPR dst, memory mem) %{
  predicate( UseSSE<=1 && !Compile::current()->select_24_bit_instr());
  match(Set dst (ConvI2D (LoadI mem)));
  format %{ "FILD   $mem\n\t"
            "FSTP   $dst" %}
  opcode(0xDB);      /* DB /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),
              Pop_Reg_DPR(dst));
  ins_pipe( fpu_reg_mem );
%}

// Convert a byte to a float; no rounding step needed.
instruct conv24I2FPR_reg(regFPR dst, stackSlotI src) %{
  predicate( UseSSE==0 && n->in(1)->Opcode() == Op_AndI && n->in(1)->in(2)->is_Con() && n->in(1)->in(2)->get_int() == 255 );
  match(Set dst (ConvI2F src));
  format %{ "FILD   $src\n\t"
            "FSTP   $dst" %}

  opcode(0xDB, 0x0);  /* DB /0 */
  ins_encode(Push_Mem_I(src), Pop_Reg_FPR(dst));
  ins_pipe( fpu_reg_mem );
%}

// In 24-bit mode, force exponent rounding by storing back out
instruct convI2FPR_SSF(stackSlotF dst, stackSlotI src) %{
  predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (ConvI2F src));
  ins_cost(200);
  format %{ "FILD   $src\n\t"
            "FSTP_S $dst" %}
  opcode(0xDB, 0x0);  /* DB /0 */
  ins_encode( Push_Mem_I(src),
              Pop_Mem_FPR(dst));
  ins_pipe( fpu_mem_mem );
%}

// In 24-bit mode, force exponent rounding by storing back out
instruct convI2FPR_SSF_mem(stackSlotF dst, memory mem) %{
  predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
  match(Set dst (ConvI2F (LoadI mem)));
  ins_cost(200);
  format %{ "FILD   $mem\n\t"
            "FSTP_S $dst" %}
  opcode(0xDB);  /* DB /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),
              Pop_Mem_FPR(dst));
  ins_pipe( fpu_mem_mem );
%}

// This instruction does not round to 24-bits
instruct convI2FPR_reg(regFPR dst, stackSlotI src) %{
  predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (ConvI2F src));
  format %{ "FILD   $src\n\t"
            "FSTP   $dst" %}
  opcode(0xDB, 0x0);  /* DB /0 */
  ins_encode( Push_Mem_I(src),
              Pop_Reg_FPR(dst));
  ins_pipe( fpu_reg_mem );
%}

// This instruction does not round to 24-bits
instruct convI2FPR_mem(regFPR dst, memory mem) %{
  predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
  match(Set dst (ConvI2F (LoadI mem)));
  format %{ "FILD   $mem\n\t"
            "FSTP   $dst" %}
  opcode(0xDB);      /* DB /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,mem),
              Pop_Reg_FPR(dst));
  ins_pipe( fpu_reg_mem );
%}

// Convert an int to a float in xmm; no rounding step needed.
instruct convI2F_reg(regF dst, rRegI src) %{
  predicate( UseSSE==1 || UseSSE>=2 && !UseXmmI2F );
  match(Set dst (ConvI2F src));
  format %{ "CVTSI2SS $dst, $src" %}
  ins_encode %{
    __ cvtsi2ssl ($dst$$XMMRegister, $src$$Register);
  %}
  ins_pipe( pipe_slow );
%}

 instruct convXI2F_reg(regF dst, rRegI src)
%{
  predicate( UseSSE>=2 && UseXmmI2F );
  match(Set dst (ConvI2F src));

  format %{ "MOVD  $dst,$src\n\t"
            "CVTDQ2PS $dst,$dst\t# i2f" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe(pipe_slow); // XXX
%}

instruct convI2L_reg( eRegL dst, rRegI src, eFlagsReg cr) %{
  match(Set dst (ConvI2L src));
  effect(KILL cr);
  ins_cost(375);
  format %{ "MOV    $dst.lo,$src\n\t"
            "MOV    $dst.hi,$src\n\t"
            "SAR    $dst.hi,31" %}
  ins_encode(convert_int_long(dst,src));
  ins_pipe( ialu_reg_reg_long );
%}

// Zero-extend convert int to long
instruct convI2L_reg_zex(eRegL dst, rRegI src, immL_32bits mask, eFlagsReg flags ) %{
  match(Set dst (AndL (ConvI2L src) mask) );
  effect( KILL flags );
  ins_cost(250);
  format %{ "MOV    $dst.lo,$src\n\t"
            "XOR    $dst.hi,$dst.hi" %}
  opcode(0x33); // XOR
  ins_encode(enc_Copy(dst,src), OpcP, RegReg_Hi2(dst,dst) );
  ins_pipe( ialu_reg_reg_long );
%}

// Zero-extend long
instruct zerox_long(eRegL dst, eRegL src, immL_32bits mask, eFlagsReg flags ) %{
  match(Set dst (AndL src mask) );
  effect( KILL flags );
  ins_cost(250);
  format %{ "MOV    $dst.lo,$src.lo\n\t"
            "XOR    $dst.hi,$dst.hi\n\t" %}
  opcode(0x33); // XOR
  ins_encode(enc_Copy(dst,src), OpcP, RegReg_Hi2(dst,dst) );
  ins_pipe( ialu_reg_reg_long );
%}

instruct convL2DPR_reg( stackSlotD dst, eRegL src, eFlagsReg cr) %{
  predicate (UseSSE<=1);
  match(Set dst (ConvL2D src));
  effect( KILL cr );
  format %{ "PUSH   $src.hi\t# Convert long to double\n\t"
            "PUSH   $src.lo\n\t"
            "FILD   ST,[ESP + #0]\n\t"
            "ADD    ESP,8\n\t"
            "FSTP_D $dst\t# D-round" %}
  opcode(0xDF, 0x5);  /* DF /5 */
  ins_encode(convert_long_double(src), Pop_Mem_DPR(dst));
  ins_pipe( pipe_slow );
%}

instruct convL2D_reg( regD dst, eRegL src, eFlagsReg cr) %{
  predicate (UseSSE>=2);
  match(Set dst (ConvL2D src));
  effect( KILL cr );
  format %{ "PUSH   $src.hi\t# Convert long to double\n\t"
            "PUSH   $src.lo\n\t"
            "FILD_D [ESP]\n\t"
            "FSTP_D [ESP]\n\t"
            "MOVSD  $dst,[ESP]\n\t"
            "ADD    ESP,8" %}
  opcode(0xDF, 0x5);  /* DF /5 */
  ins_encode(convert_long_double2(src), Push_ResultD(dst));
  ins_pipe( pipe_slow );
%}

instruct convL2F_reg( regF dst, eRegL src, eFlagsReg cr) %{
  predicate (UseSSE>=1);
  match(Set dst (ConvL2F src));
  effect( KILL cr );
  format %{ "PUSH   $src.hi\t# Convert long to single float\n\t"
            "PUSH   $src.lo\n\t"
            "FILD_D [ESP]\n\t"
            "FSTP_S [ESP]\n\t"
            "MOVSS  $dst,[ESP]\n\t"
            "ADD    ESP,8" %}
  opcode(0xDF, 0x5);  /* DF /5 */
  ins_encode(convert_long_double2(src), Push_ResultF(dst,0x8));
  ins_pipe( pipe_slow );
%}

instruct convL2FPR_reg( stackSlotF dst, eRegL src, eFlagsReg cr) %{
  match(Set dst (ConvL2F src));
  effect( KILL cr );
  format %{ "PUSH   $src.hi\t# Convert long to single float\n\t"
            "PUSH   $src.lo\n\t"
            "FILD   ST,[ESP + #0]\n\t"
            "ADD    ESP,8\n\t"
            "FSTP_S $dst\t# F-round" %}
  opcode(0xDF, 0x5);  /* DF /5 */
  ins_encode(convert_long_double(src), Pop_Mem_FPR(dst));
  ins_pipe( pipe_slow );
%}

instruct convL2I_reg( rRegI dst, eRegL src ) %{
  match(Set dst (ConvL2I src));
  effect( DEF dst, USE src );
  format %{ "MOV    $dst,$src.lo" %}
  ins_encode(enc_CopyL_Lo(dst,src));
  ins_pipe( ialu_reg_reg );
%}


instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
  match(Set dst (MoveF2I src));
  effect( DEF dst, USE src );
  ins_cost(100);
  format %{ "MOV    $dst,$src\t# MoveF2I_stack_reg" %}
  ins_encode %{
    __ movl($dst$$Register, Address(rsp, $src$$disp));
  %}
  ins_pipe( ialu_reg_mem );
%}

instruct MoveFPR2I_reg_stack(stackSlotI dst, regFPR src) %{
  predicate(UseSSE==0);
  match(Set dst (MoveF2I src));
  effect( DEF dst, USE src );

  ins_cost(125);
  format %{ "FST_S  $dst,$src\t# MoveF2I_reg_stack" %}
  ins_encode( Pop_Mem_Reg_FPR(dst, src) );
  ins_pipe( fpu_mem_reg );
%}

instruct MoveF2I_reg_stack_sse(stackSlotI dst, regF src) %{
  predicate(UseSSE>=1);
  match(Set dst (MoveF2I src));
  effect( DEF dst, USE src );

  ins_cost(95);
  format %{ "MOVSS  $dst,$src\t# MoveF2I_reg_stack_sse" %}
  ins_encode %{
    __ movflt(Address(rsp, $dst$$disp), $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveF2I_reg_reg_sse(rRegI dst, regF src) %{
  predicate(UseSSE>=2);
  match(Set dst (MoveF2I src));
  effect( DEF dst, USE src );
  ins_cost(85);
  format %{ "MOVD   $dst,$src\t# MoveF2I_reg_reg_sse" %}
  ins_encode %{
    __ movdl($dst$$Register, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
  match(Set dst (MoveI2F src));
  effect( DEF dst, USE src );

  ins_cost(100);
  format %{ "MOV    $dst,$src\t# MoveI2F_reg_stack" %}
  ins_encode %{
    __ movl(Address(rsp, $dst$$disp), $src$$Register);
  %}
  ins_pipe( ialu_mem_reg );
%}


instruct MoveI2FPR_stack_reg(regFPR dst, stackSlotI src) %{
  predicate(UseSSE==0);
  match(Set dst (MoveI2F src));
  effect(DEF dst, USE src);

  ins_cost(125);
  format %{ "FLD_S  $src\n\t"
            "FSTP   $dst\t# MoveI2F_stack_reg" %}
  opcode(0xD9);               /* D9 /0, FLD m32real */
  ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
              Pop_Reg_FPR(dst) );
  ins_pipe( fpu_reg_mem );
%}

instruct MoveI2F_stack_reg_sse(regF dst, stackSlotI src) %{
  predicate(UseSSE>=1);
  match(Set dst (MoveI2F src));
  effect( DEF dst, USE src );

  ins_cost(95);
  format %{ "MOVSS  $dst,$src\t# MoveI2F_stack_reg_sse" %}
  ins_encode %{
    __ movflt($dst$$XMMRegister, Address(rsp, $src$$disp));
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveI2F_reg_reg_sse(regF dst, rRegI src) %{
  predicate(UseSSE>=2);
  match(Set dst (MoveI2F src));
  effect( DEF dst, USE src );

  ins_cost(85);
  format %{ "MOVD   $dst,$src\t# MoveI2F_reg_reg_sse" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveD2L_stack_reg(eRegL dst, stackSlotD src) %{
  match(Set dst (MoveD2L src));
  effect(DEF dst, USE src);

  ins_cost(250);
  format %{ "MOV    $dst.lo,$src\n\t"
            "MOV    $dst.hi,$src+4\t# MoveD2L_stack_reg" %}
  opcode(0x8B, 0x8B);
  ins_encode( OpcP, RegMem(dst,src), OpcS, RegMem_Hi(dst,src));
  ins_pipe( ialu_mem_long_reg );
%}

instruct MoveDPR2L_reg_stack(stackSlotL dst, regDPR src) %{
  predicate(UseSSE<=1);
  match(Set dst (MoveD2L src));
  effect(DEF dst, USE src);

  ins_cost(125);
  format %{ "FST_D  $dst,$src\t# MoveD2L_reg_stack" %}
  ins_encode( Pop_Mem_Reg_DPR(dst, src) );
  ins_pipe( fpu_mem_reg );
%}

instruct MoveD2L_reg_stack_sse(stackSlotL dst, regD src) %{
  predicate(UseSSE>=2);
  match(Set dst (MoveD2L src));
  effect(DEF dst, USE src);
  ins_cost(95);
  format %{ "MOVSD  $dst,$src\t# MoveD2L_reg_stack_sse" %}
  ins_encode %{
    __ movdbl(Address(rsp, $dst$$disp), $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveD2L_reg_reg_sse(eRegL dst, regD src, regD tmp) %{
  predicate(UseSSE>=2);
  match(Set dst (MoveD2L src));
  effect(DEF dst, USE src, TEMP tmp);
  ins_cost(85);
  format %{ "MOVD   $dst.lo,$src\n\t"
            "PSHUFLW $tmp,$src,0x4E\n\t"
            "MOVD   $dst.hi,$tmp\t# MoveD2L_reg_reg_sse" %}
  ins_encode %{
    __ movdl($dst$$Register, $src$$XMMRegister);
    __ pshuflw($tmp$$XMMRegister, $src$$XMMRegister, 0x4e);
    __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveL2D_reg_stack(stackSlotD dst, eRegL src) %{
  match(Set dst (MoveL2D src));
  effect(DEF dst, USE src);

  ins_cost(200);
  format %{ "MOV    $dst,$src.lo\n\t"
            "MOV    $dst+4,$src.hi\t# MoveL2D_reg_stack" %}
  opcode(0x89, 0x89);
  ins_encode( OpcP, RegMem( src, dst ), OpcS, RegMem_Hi( src, dst ) );
  ins_pipe( ialu_mem_long_reg );
%}


instruct MoveL2DPR_stack_reg(regDPR dst, stackSlotL src) %{
  predicate(UseSSE<=1);
  match(Set dst (MoveL2D src));
  effect(DEF dst, USE src);
  ins_cost(125);

  format %{ "FLD_D  $src\n\t"
            "FSTP   $dst\t# MoveL2D_stack_reg" %}
  opcode(0xDD);               /* DD /0, FLD m64real */
  ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
              Pop_Reg_DPR(dst) );
  ins_pipe( fpu_reg_mem );
%}


instruct MoveL2D_stack_reg_sse(regD dst, stackSlotL src) %{
  predicate(UseSSE>=2 && UseXmmLoadAndClearUpper);
  match(Set dst (MoveL2D src));
  effect(DEF dst, USE src);

  ins_cost(95);
  format %{ "MOVSD  $dst,$src\t# MoveL2D_stack_reg_sse" %}
  ins_encode %{
    __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveL2D_stack_reg_sse_partial(regD dst, stackSlotL src) %{
  predicate(UseSSE>=2 && !UseXmmLoadAndClearUpper);
  match(Set dst (MoveL2D src));
  effect(DEF dst, USE src);

  ins_cost(95);
  format %{ "MOVLPD $dst,$src\t# MoveL2D_stack_reg_sse" %}
  ins_encode %{
    __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
  %}
  ins_pipe( pipe_slow );
%}

instruct MoveL2D_reg_reg_sse(regD dst, eRegL src, regD tmp) %{
  predicate(UseSSE>=2);
  match(Set dst (MoveL2D src));
  effect(TEMP dst, USE src, TEMP tmp);
  ins_cost(85);
  format %{ "MOVD   $dst,$src.lo\n\t"
            "MOVD   $tmp,$src.hi\n\t"
            "PUNPCKLDQ $dst,$tmp\t# MoveL2D_reg_reg_sse" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}


// =======================================================================
// fast clearing of an array
instruct rep_stos(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
  predicate(!UseFastStosb);
  match(Set dummy (ClearArray cnt base));
  effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
  format %{ "XOR    EAX,EAX\t# ClearArray:\n\t"
            "SHL    ECX,1\t# Convert doublewords to words\n\t"
            "REP STOS\t# store EAX into [EDI++] while ECX--" %}
  ins_encode %{ 
    __ clear_mem($base$$Register, $cnt$$Register, $zero$$Register);
  %}
  ins_pipe( pipe_slow );
%}

instruct rep_fast_stosb(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
  predicate(UseFastStosb);
  match(Set dummy (ClearArray cnt base));
  effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
  format %{ "XOR    EAX,EAX\t# ClearArray:\n\t"
            "SHL    ECX,3\t# Convert doublewords to bytes\n\t"
            "REP STOSB\t# store EAX into [EDI++] while ECX--" %}
  ins_encode %{ 
    __ clear_mem($base$$Register, $cnt$$Register, $zero$$Register);
  %}
  ins_pipe( pipe_slow );
%}

instruct string_compare(eDIRegP str1, eCXRegI cnt1, eSIRegP str2, eDXRegI cnt2,
                        eAXRegI result, regD tmp1, eFlagsReg cr) %{
  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);

  format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result   // KILL $tmp1" %}
  ins_encode %{
    __ string_compare($str1$$Register, $str2$$Register,
                      $cnt1$$Register, $cnt2$$Register, $result$$Register,
                      $tmp1$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// fast string equals
instruct string_equals(eDIRegP str1, eSIRegP str2, eCXRegI cnt, eAXRegI result,
                       regD tmp1, regD tmp2, eBXRegI tmp3, eFlagsReg cr) %{
  match(Set result (StrEquals (Binary str1 str2) cnt));
  effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp3, KILL cr);

  format %{ "String Equals $str1,$str2,$cnt -> $result    // KILL $tmp1, $tmp2, $tmp3" %}
  ins_encode %{
    __ char_arrays_equals(false, $str1$$Register, $str2$$Register,
                          $cnt$$Register, $result$$Register, $tmp3$$Register,
                          $tmp1$$XMMRegister, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// fast search of substring with known size.
instruct string_indexof_con(eDIRegP str1, eDXRegI cnt1, eSIRegP str2, immI int_cnt2,
                            eBXRegI result, regD vec, eAXRegI cnt2, eCXRegI tmp, eFlagsReg cr) %{
  predicate(UseSSE42Intrinsics);
  match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
  effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, KILL cnt2, KILL tmp, KILL cr);

  format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result   // KILL $vec, $cnt1, $cnt2, $tmp" %}
  ins_encode %{
    int icnt2 = (int)$int_cnt2$$constant;
    if (icnt2 >= 8) {
      // IndexOf for constant substrings with size >= 8 elements
      // which don't need to be loaded through stack.
      __ string_indexofC8($str1$$Register, $str2$$Register,
                          $cnt1$$Register, $cnt2$$Register,
                          icnt2, $result$$Register,
                          $vec$$XMMRegister, $tmp$$Register);
    } else {
      // Small strings are loaded through stack if they cross page boundary.
      __ string_indexof($str1$$Register, $str2$$Register,
                        $cnt1$$Register, $cnt2$$Register,
                        icnt2, $result$$Register,
                        $vec$$XMMRegister, $tmp$$Register);
    }
  %}
  ins_pipe( pipe_slow );
%}

instruct string_indexof(eDIRegP str1, eDXRegI cnt1, eSIRegP str2, eAXRegI cnt2,
                        eBXRegI result, regD vec, eCXRegI tmp, eFlagsReg cr) %{
  predicate(UseSSE42Intrinsics);
  match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL tmp, KILL cr);

  format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result   // KILL all" %}
  ins_encode %{
    __ string_indexof($str1$$Register, $str2$$Register,
                      $cnt1$$Register, $cnt2$$Register,
                      (-1), $result$$Register,
                      $vec$$XMMRegister, $tmp$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// fast array equals
instruct array_equals(eDIRegP ary1, eSIRegP ary2, eAXRegI result,
                      regD tmp1, regD tmp2, eCXRegI tmp3, eBXRegI tmp4, eFlagsReg cr)
%{
  match(Set result (AryEq ary1 ary2));
  effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
  //ins_cost(300);

  format %{ "Array Equals $ary1,$ary2 -> $result   // KILL $tmp1, $tmp2, $tmp3, $tmp4" %}
  ins_encode %{
    __ char_arrays_equals(true, $ary1$$Register, $ary2$$Register,
                          $tmp3$$Register, $result$$Register, $tmp4$$Register,
                          $tmp1$$XMMRegister, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// encode char[] to byte[] in ISO_8859_1
instruct encode_iso_array(eSIRegP src, eDIRegP dst, eDXRegI len,
                          regD tmp1, regD tmp2, regD tmp3, regD tmp4,
                          eCXRegI tmp5, eAXRegI result, eFlagsReg cr) %{
  match(Set result (EncodeISOArray src (Binary dst len)));
  effect(TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, USE_KILL src, USE_KILL dst, USE_KILL len, KILL tmp5, KILL cr);

  format %{ "Encode array $src,$dst,$len -> $result    // KILL ECX, EDX, $tmp1, $tmp2, $tmp3, $tmp4, ESI, EDI " %}
  ins_encode %{
    __ encode_iso_array($src$$Register, $dst$$Register, $len$$Register,
                        $tmp1$$XMMRegister, $tmp2$$XMMRegister, $tmp3$$XMMRegister,
                        $tmp4$$XMMRegister, $tmp5$$Register, $result$$Register);
  %}
  ins_pipe( pipe_slow );
%}


//----------Control Flow Instructions------------------------------------------
// Signed compare Instructions
instruct compI_eReg(eFlagsReg cr, rRegI op1, rRegI op2) %{
  match(Set cr (CmpI op1 op2));
  effect( DEF cr, USE op1, USE op2 );
  format %{ "CMP    $op1,$op2" %}
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegReg( op1, op2) );
  ins_pipe( ialu_cr_reg_reg );
%}

instruct compI_eReg_imm(eFlagsReg cr, rRegI op1, immI op2) %{
  match(Set cr (CmpI op1 op2));
  effect( DEF cr, USE op1 );
  format %{ "CMP    $op1,$op2" %}
  opcode(0x81,0x07);  /* Opcode 81 /7 */
  // ins_encode( RegImm( op1, op2) );  /* Was CmpImm */
  ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
  ins_pipe( ialu_cr_reg_imm );
%}

// Cisc-spilled version of cmpI_eReg
instruct compI_eReg_mem(eFlagsReg cr, rRegI op1, memory op2) %{
  match(Set cr (CmpI op1 (LoadI op2)));

  format %{ "CMP    $op1,$op2" %}
  ins_cost(500);
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegMem( op1, op2) );
  ins_pipe( ialu_cr_reg_mem );
%}

instruct testI_reg( eFlagsReg cr, rRegI src, immI0 zero ) %{
  match(Set cr (CmpI src zero));
  effect( DEF cr, USE src );

  format %{ "TEST   $src,$src" %}
  opcode(0x85);
  ins_encode( OpcP, RegReg( src, src ) );
  ins_pipe( ialu_cr_reg_imm );
%}

instruct testI_reg_imm( eFlagsReg cr, rRegI src, immI con, immI0 zero ) %{
  match(Set cr (CmpI (AndI src con) zero));

  format %{ "TEST   $src,$con" %}
  opcode(0xF7,0x00);
  ins_encode( OpcP, RegOpc(src), Con32(con) );
  ins_pipe( ialu_cr_reg_imm );
%}

instruct testI_reg_mem( eFlagsReg cr, rRegI src, memory mem, immI0 zero ) %{
  match(Set cr (CmpI (AndI src mem) zero));

  format %{ "TEST   $src,$mem" %}
  opcode(0x85);
  ins_encode( OpcP, RegMem( src, mem ) );
  ins_pipe( ialu_cr_reg_mem );
%}

// Unsigned compare Instructions; really, same as signed except they
// produce an eFlagsRegU instead of eFlagsReg.
instruct compU_eReg(eFlagsRegU cr, rRegI op1, rRegI op2) %{
  match(Set cr (CmpU op1 op2));

  format %{ "CMPu   $op1,$op2" %}
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegReg( op1, op2) );
  ins_pipe( ialu_cr_reg_reg );
%}

instruct compU_eReg_imm(eFlagsRegU cr, rRegI op1, immI op2) %{
  match(Set cr (CmpU op1 op2));

  format %{ "CMPu   $op1,$op2" %}
  opcode(0x81,0x07);  /* Opcode 81 /7 */
  ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
  ins_pipe( ialu_cr_reg_imm );
%}

// // Cisc-spilled version of cmpU_eReg
instruct compU_eReg_mem(eFlagsRegU cr, rRegI op1, memory op2) %{
  match(Set cr (CmpU op1 (LoadI op2)));

  format %{ "CMPu   $op1,$op2" %}
  ins_cost(500);
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegMem( op1, op2) );
  ins_pipe( ialu_cr_reg_mem );
%}

// // Cisc-spilled version of cmpU_eReg
//instruct compU_mem_eReg(eFlagsRegU cr, memory op1, rRegI op2) %{
//  match(Set cr (CmpU (LoadI op1) op2));
//
//  format %{ "CMPu   $op1,$op2" %}
//  ins_cost(500);
//  opcode(0x39);  /* Opcode 39 /r */
//  ins_encode( OpcP, RegMem( op1, op2) );
//%}

instruct testU_reg( eFlagsRegU cr, rRegI src, immI0 zero ) %{
  match(Set cr (CmpU src zero));

  format %{ "TESTu  $src,$src" %}
  opcode(0x85);
  ins_encode( OpcP, RegReg( src, src ) );
  ins_pipe( ialu_cr_reg_imm );
%}

// Unsigned pointer compare Instructions
instruct compP_eReg(eFlagsRegU cr, eRegP op1, eRegP op2) %{
  match(Set cr (CmpP op1 op2));

  format %{ "CMPu   $op1,$op2" %}
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegReg( op1, op2) );
  ins_pipe( ialu_cr_reg_reg );
%}

instruct compP_eReg_imm(eFlagsRegU cr, eRegP op1, immP op2) %{
  match(Set cr (CmpP op1 op2));

  format %{ "CMPu   $op1,$op2" %}
  opcode(0x81,0x07);  /* Opcode 81 /7 */
  ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
  ins_pipe( ialu_cr_reg_imm );
%}

// // Cisc-spilled version of cmpP_eReg
instruct compP_eReg_mem(eFlagsRegU cr, eRegP op1, memory op2) %{
  match(Set cr (CmpP op1 (LoadP op2)));

  format %{ "CMPu   $op1,$op2" %}
  ins_cost(500);
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegMem( op1, op2) );
  ins_pipe( ialu_cr_reg_mem );
%}

// // Cisc-spilled version of cmpP_eReg
//instruct compP_mem_eReg(eFlagsRegU cr, memory op1, eRegP op2) %{
//  match(Set cr (CmpP (LoadP op1) op2));
//
//  format %{ "CMPu   $op1,$op2" %}
//  ins_cost(500);
//  opcode(0x39);  /* Opcode 39 /r */
//  ins_encode( OpcP, RegMem( op1, op2) );
//%}

// Compare raw pointer (used in out-of-heap check).
// Only works because non-oop pointers must be raw pointers
// and raw pointers have no anti-dependencies.
instruct compP_mem_eReg( eFlagsRegU cr, eRegP op1, memory op2 ) %{
  predicate( n->in(2)->in(2)->bottom_type()->reloc() == relocInfo::none );
  match(Set cr (CmpP op1 (LoadP op2)));

  format %{ "CMPu   $op1,$op2" %}
  opcode(0x3B);  /* Opcode 3B /r */
  ins_encode( OpcP, RegMem( op1, op2) );
  ins_pipe( ialu_cr_reg_mem );
%}

//
// This will generate a signed flags result. This should be ok
// since any compare to a zero should be eq/neq.
instruct testP_reg( eFlagsReg cr, eRegP src, immP0 zero ) %{
  match(Set cr (CmpP src zero));

  format %{ "TEST   $src,$src" %}
  opcode(0x85);
  ins_encode( OpcP, RegReg( src, src ) );
  ins_pipe( ialu_cr_reg_imm );
%}

// Cisc-spilled version of testP_reg
// This will generate a signed flags result. This should be ok
// since any compare to a zero should be eq/neq.
instruct testP_Reg_mem( eFlagsReg cr, memory op, immI0 zero ) %{
  match(Set cr (CmpP (LoadP op) zero));

  format %{ "TEST   $op,0xFFFFFFFF" %}
  ins_cost(500);
  opcode(0xF7);               /* Opcode F7 /0 */
  ins_encode( OpcP, RMopc_Mem(0x00,op), Con_d32(0xFFFFFFFF) );
  ins_pipe( ialu_cr_reg_imm );
%}

// Yanked all unsigned pointer compare operations.
// Pointer compares are done with CmpP which is already unsigned.

//----------Max and Min--------------------------------------------------------
// Min Instructions
////
//   *** Min and Max using the conditional move are slower than the
//   *** branch version on a Pentium III.
// // Conditional move for min
//instruct cmovI_reg_lt( rRegI op2, rRegI op1, eFlagsReg cr ) %{
//  effect( USE_DEF op2, USE op1, USE cr );
//  format %{ "CMOVlt $op2,$op1\t! min" %}
//  opcode(0x4C,0x0F);
//  ins_encode( OpcS, OpcP, RegReg( op2, op1 ) );
//  ins_pipe( pipe_cmov_reg );
//%}
//
//// Min Register with Register (P6 version)
//instruct minI_eReg_p6( rRegI op1, rRegI op2 ) %{
//  predicate(VM_Version::supports_cmov() );
//  match(Set op2 (MinI op1 op2));
//  ins_cost(200);
//  expand %{
//    eFlagsReg cr;
//    compI_eReg(cr,op1,op2);
//    cmovI_reg_lt(op2,op1,cr);
//  %}
//%}

// Min Register with Register (generic version)
instruct minI_eReg(rRegI dst, rRegI src, eFlagsReg flags) %{
  match(Set dst (MinI dst src));
  effect(KILL flags);
  ins_cost(300);

  format %{ "MIN    $dst,$src" %}
  opcode(0xCC);
  ins_encode( min_enc(dst,src) );
  ins_pipe( pipe_slow );
%}

// Max Register with Register
//   *** Min and Max using the conditional move are slower than the
//   *** branch version on a Pentium III.
// // Conditional move for max
//instruct cmovI_reg_gt( rRegI op2, rRegI op1, eFlagsReg cr ) %{
//  effect( USE_DEF op2, USE op1, USE cr );
//  format %{ "CMOVgt $op2,$op1\t! max" %}
//  opcode(0x4F,0x0F);
//  ins_encode( OpcS, OpcP, RegReg( op2, op1 ) );
//  ins_pipe( pipe_cmov_reg );
//%}
//
// // Max Register with Register (P6 version)
//instruct maxI_eReg_p6( rRegI op1, rRegI op2 ) %{
//  predicate(VM_Version::supports_cmov() );
//  match(Set op2 (MaxI op1 op2));
//  ins_cost(200);
//  expand %{
//    eFlagsReg cr;
//    compI_eReg(cr,op1,op2);
//    cmovI_reg_gt(op2,op1,cr);
//  %}
//%}

// Max Register with Register (generic version)
instruct maxI_eReg(rRegI dst, rRegI src, eFlagsReg flags) %{
  match(Set dst (MaxI dst src));
  effect(KILL flags);
  ins_cost(300);

  format %{ "MAX    $dst,$src" %}
  opcode(0xCC);
  ins_encode( max_enc(dst,src) );
  ins_pipe( pipe_slow );
%}

// ============================================================================
// Counted Loop limit node which represents exact final iterator value.
// Note: the resulting value should fit into integer range since
// counted loops have limit check on overflow.
instruct loopLimit_eReg(eAXRegI limit, nadxRegI init, immI stride, eDXRegI limit_hi, nadxRegI tmp, eFlagsReg flags) %{
  match(Set limit (LoopLimit (Binary init limit) stride));
  effect(TEMP limit_hi, TEMP tmp, KILL flags);
  ins_cost(300);

  format %{ "loopLimit $init,$limit,$stride  # $limit = $init + $stride *( $limit - $init + $stride -1)/ $stride, kills $limit_hi" %}
  ins_encode %{
    int strd = (int)$stride$$constant;
    assert(strd != 1 && strd != -1, "sanity");
    int m1 = (strd > 0) ? 1 : -1;
    // Convert limit to long (EAX:EDX)
    __ cdql();
    // Convert init to long (init:tmp)
    __ movl($tmp$$Register, $init$$Register);
    __ sarl($tmp$$Register, 31);
    // $limit - $init
    __ subl($limit$$Register, $init$$Register);
    __ sbbl($limit_hi$$Register, $tmp$$Register);
    // + ($stride - 1)
    if (strd > 0) {
      __ addl($limit$$Register, (strd - 1));
      __ adcl($limit_hi$$Register, 0);
      __ movl($tmp$$Register, strd);
    } else {
      __ addl($limit$$Register, (strd + 1));
      __ adcl($limit_hi$$Register, -1);
      __ lneg($limit_hi$$Register, $limit$$Register);
      __ movl($tmp$$Register, -strd);
    }
    // signed devision: (EAX:EDX) / pos_stride
    __ idivl($tmp$$Register);
    if (strd < 0) {
      // restore sign
      __ negl($tmp$$Register);
    }
    // (EAX) * stride
    __ mull($tmp$$Register);
    // + init (ignore upper bits)
    __ addl($limit$$Register, $init$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// ============================================================================
// Branch Instructions
// Jump Table
instruct jumpXtnd(rRegI switch_val) %{
  match(Jump switch_val);
  ins_cost(350);
  format %{  "JMP    [$constantaddress](,$switch_val,1)\n\t" %}
  ins_encode %{
    // Jump to Address(table_base + switch_reg)
    Address index(noreg, $switch_val$$Register, Address::times_1);
    __ jump(ArrayAddress($constantaddress, index));
  %}
  ins_pipe(pipe_jmp);
%}

// Jump Direct - Label defines a relative address from JMP+1
instruct jmpDir(label labl) %{
  match(Goto);
  effect(USE labl);

  ins_cost(300);
  format %{ "JMP    $labl" %}
  size(5);
  ins_encode %{
    Label* L = $labl$$label;
    __ jmp(*L, false); // Always long jump
  %}
  ins_pipe( pipe_jmp );
%}

// Jump Direct Conditional - Label defines a relative address from Jcc+1
instruct jmpCon(cmpOp cop, eFlagsReg cr, label labl) %{
  match(If cop cr);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop    $labl" %}
  size(6);
  ins_encode %{
    Label* L = $labl$$label;
    __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
  %}
  ins_pipe( pipe_jcc );
%}

// Jump Direct Conditional - Label defines a relative address from Jcc+1
instruct jmpLoopEnd(cmpOp cop, eFlagsReg cr, label labl) %{
  match(CountedLoopEnd cop cr);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop    $labl\t# Loop end" %}
  size(6);
  ins_encode %{
    Label* L = $labl$$label;
    __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
  %}
  ins_pipe( pipe_jcc );
%}

// Jump Direct Conditional - Label defines a relative address from Jcc+1
instruct jmpLoopEndU(cmpOpU cop, eFlagsRegU cmp, label labl) %{
  match(CountedLoopEnd cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,u  $labl\t# Loop end" %}
  size(6);
  ins_encode %{
    Label* L = $labl$$label;
    __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
  %}
  ins_pipe( pipe_jcc );
%}

instruct jmpLoopEndUCF(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
  match(CountedLoopEnd cop cmp);
  effect(USE labl);

  ins_cost(200);
  format %{ "J$cop,u  $labl\t# Loop end" %}
  size(6);
  ins_encode %{
    Label* L = $labl$$label;
    __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
  %}
  ins_pipe( pipe_jcc );
%}

// Jump Direct Conditional - using unsigned comparison
instruct jmpConU(cmpOpU cop, eFlagsRegU cmp, label labl) %{
  match(If cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,u  $labl" %}
  size(6);
  ins_encode %{
    Label* L = $labl$$label;
    __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
  %}
  ins_pipe(pipe_jcc);
%}

instruct jmpConUCF(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
  match(If cop cmp);
  effect(USE labl);

  ins_cost(200);
  format %{ "J$cop,u  $labl" %}
  size(6);
  ins_encode %{
    Label* L = $labl$$label;
    __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
  %}
  ins_pipe(pipe_jcc);
%}

instruct jmpConUCF2(cmpOpUCF2 cop, eFlagsRegUCF cmp, label labl) %{
  match(If cop cmp);
  effect(USE labl);

  ins_cost(200);
  format %{ $$template
    if ($cop$$cmpcode == Assembler::notEqual) {
      $$emit$$"JP,u   $labl\n\t"
      $$emit$$"J$cop,u   $labl"
    } else {
      $$emit$$"JP,u   done\n\t"
      $$emit$$"J$cop,u   $labl\n\t"
      $$emit$$"done:"
    }
  %}
  ins_encode %{
    Label* l = $labl$$label;
    if ($cop$$cmpcode == Assembler::notEqual) {
      __ jcc(Assembler::parity, *l, false);
      __ jcc(Assembler::notEqual, *l, false);
    } else if ($cop$$cmpcode == Assembler::equal) {
      Label done;
      __ jccb(Assembler::parity, done);
      __ jcc(Assembler::equal, *l, false);
      __ bind(done);
    } else {
       ShouldNotReachHere();
    }
  %}
  ins_pipe(pipe_jcc);
%}

// ============================================================================
// The 2nd slow-half of a subtype check.  Scan the subklass's 2ndary superklass
// array for an instance of the superklass.  Set a hidden internal cache on a
// hit (cache is checked with exposed code in gen_subtype_check()).  Return
// NZ for a miss or zero for a hit.  The encoding ALSO sets flags.
instruct partialSubtypeCheck( eDIRegP result, eSIRegP sub, eAXRegP super, eCXRegI rcx, eFlagsReg cr ) %{
  match(Set result (PartialSubtypeCheck sub super));
  effect( KILL rcx, KILL cr );

  ins_cost(1100);  // slightly larger than the next version
  format %{ "MOV    EDI,[$sub+Klass::secondary_supers]\n\t"
            "MOV    ECX,[EDI+ArrayKlass::length]\t# length to scan\n\t"
            "ADD    EDI,ArrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t"
            "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
            "JNE,s  miss\t\t# Missed: EDI not-zero\n\t"
            "MOV    [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache\n\t"
            "XOR    $result,$result\t\t Hit: EDI zero\n\t"
     "miss:\t" %}

  opcode(0x1); // Force a XOR of EDI
  ins_encode( enc_PartialSubtypeCheck() );
  ins_pipe( pipe_slow );
%}

instruct partialSubtypeCheck_vs_Zero( eFlagsReg cr, eSIRegP sub, eAXRegP super, eCXRegI rcx, eDIRegP result, immP0 zero ) %{
  match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
  effect( KILL rcx, KILL result );

  ins_cost(1000);
  format %{ "MOV    EDI,[$sub+Klass::secondary_supers]\n\t"
            "MOV    ECX,[EDI+ArrayKlass::length]\t# length to scan\n\t"
            "ADD    EDI,ArrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t"
            "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
            "JNE,s  miss\t\t# Missed: flags NZ\n\t"
            "MOV    [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache, flags Z\n\t"
     "miss:\t" %}

  opcode(0x0);  // No need to XOR EDI
  ins_encode( enc_PartialSubtypeCheck() );
  ins_pipe( pipe_slow );
%}

// ============================================================================
// Branch Instructions -- short offset versions
//
// These instructions are used to replace jumps of a long offset (the default
// match) with jumps of a shorter offset.  These instructions are all tagged
// with the ins_short_branch attribute, which causes the ADLC to suppress the
// match rules in general matching.  Instead, the ADLC generates a conversion
// method in the MachNode which can be used to do in-place replacement of the
// long variant with the shorter variant.  The compiler will determine if a
// branch can be taken by the is_short_branch_offset() predicate in the machine
// specific code section of the file.

// Jump Direct - Label defines a relative address from JMP+1
instruct jmpDir_short(label labl) %{
  match(Goto);
  effect(USE labl);

  ins_cost(300);
  format %{ "JMP,s  $labl" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jmpb(*L);
  %}
  ins_pipe( pipe_jmp );
  ins_short_branch(1);
%}

// Jump Direct Conditional - Label defines a relative address from Jcc+1
instruct jmpCon_short(cmpOp cop, eFlagsReg cr, label labl) %{
  match(If cop cr);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,s  $labl" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
  %}
  ins_pipe( pipe_jcc );
  ins_short_branch(1);
%}

// Jump Direct Conditional - Label defines a relative address from Jcc+1
instruct jmpLoopEnd_short(cmpOp cop, eFlagsReg cr, label labl) %{
  match(CountedLoopEnd cop cr);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,s  $labl\t# Loop end" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
  %}
  ins_pipe( pipe_jcc );
  ins_short_branch(1);
%}

// Jump Direct Conditional - Label defines a relative address from Jcc+1
instruct jmpLoopEndU_short(cmpOpU cop, eFlagsRegU cmp, label labl) %{
  match(CountedLoopEnd cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,us $labl\t# Loop end" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
  %}
  ins_pipe( pipe_jcc );
  ins_short_branch(1);
%}

instruct jmpLoopEndUCF_short(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
  match(CountedLoopEnd cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,us $labl\t# Loop end" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
  %}
  ins_pipe( pipe_jcc );
  ins_short_branch(1);
%}

// Jump Direct Conditional - using unsigned comparison
instruct jmpConU_short(cmpOpU cop, eFlagsRegU cmp, label labl) %{
  match(If cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,us $labl" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
  %}
  ins_pipe( pipe_jcc );
  ins_short_branch(1);
%}

instruct jmpConUCF_short(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
  match(If cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ "J$cop,us $labl" %}
  size(2);
  ins_encode %{
    Label* L = $labl$$label;
    __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
  %}
  ins_pipe( pipe_jcc );
  ins_short_branch(1);
%}

instruct jmpConUCF2_short(cmpOpUCF2 cop, eFlagsRegUCF cmp, label labl) %{
  match(If cop cmp);
  effect(USE labl);

  ins_cost(300);
  format %{ $$template
    if ($cop$$cmpcode == Assembler::notEqual) {
      $$emit$$"JP,u,s   $labl\n\t"
      $$emit$$"J$cop,u,s   $labl"
    } else {
      $$emit$$"JP,u,s   done\n\t"
      $$emit$$"J$cop,u,s  $labl\n\t"
      $$emit$$"done:"
    }
  %}
  size(4);
  ins_encode %{
    Label* l = $labl$$label;
    if ($cop$$cmpcode == Assembler::notEqual) {
      __ jccb(Assembler::parity, *l);
      __ jccb(Assembler::notEqual, *l);
    } else if ($cop$$cmpcode == Assembler::equal) {
      Label done;
      __ jccb(Assembler::parity, done);
      __ jccb(Assembler::equal, *l);
      __ bind(done);
    } else {
       ShouldNotReachHere();
    }
  %}
  ins_pipe(pipe_jcc);
  ins_short_branch(1);
%}

// ============================================================================
// Long Compare
//
// Currently we hold longs in 2 registers.  Comparing such values efficiently
// is tricky.  The flavor of compare used depends on whether we are testing
// for LT, LE, or EQ.  For a simple LT test we can check just the sign bit.
// The GE test is the negated LT test.  The LE test can be had by commuting
// the operands (yielding a GE test) and then negating; negate again for the
// GT test.  The EQ test is done by ORcc'ing the high and low halves, and the
// NE test is negated from that.

// Due to a shortcoming in the ADLC, it mixes up expressions like:
// (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)).  Note the
// difference between 'Y' and '0L'.  The tree-matches for the CmpI sections
// are collapsed internally in the ADLC's dfa-gen code.  The match for
// (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the
// foo match ends up with the wrong leaf.  One fix is to not match both
// reg-reg and reg-zero forms of long-compare.  This is unfortunate because
// both forms beat the trinary form of long-compare and both are very useful
// on Intel which has so few registers.

// Manifest a CmpL result in an integer register.  Very painful.
// This is the test to avoid.
instruct cmpL3_reg_reg(eSIRegI dst, eRegL src1, eRegL src2, eFlagsReg flags ) %{
  match(Set dst (CmpL3 src1 src2));
  effect( KILL flags );
  ins_cost(1000);
  format %{ "XOR    $dst,$dst\n\t"
            "CMP    $src1.hi,$src2.hi\n\t"
            "JLT,s  m_one\n\t"
            "JGT,s  p_one\n\t"
            "CMP    $src1.lo,$src2.lo\n\t"
            "JB,s   m_one\n\t"
            "JEQ,s  done\n"
    "p_one:\tINC    $dst\n\t"
            "JMP,s  done\n"
    "m_one:\tDEC    $dst\n"
     "done:" %}
  ins_encode %{
    Label p_one, m_one, done;
    __ xorptr($dst$$Register, $dst$$Register);
    __ cmpl(HIGH_FROM_LOW($src1$$Register), HIGH_FROM_LOW($src2$$Register));
    __ jccb(Assembler::less,    m_one);
    __ jccb(Assembler::greater, p_one);
    __ cmpl($src1$$Register, $src2$$Register);
    __ jccb(Assembler::below,   m_one);
    __ jccb(Assembler::equal,   done);
    __ bind(p_one);
    __ incrementl($dst$$Register);
    __ jmpb(done);
    __ bind(m_one);
    __ decrementl($dst$$Register);
    __ bind(done);
  %}
  ins_pipe( pipe_slow );
%}

//======
// Manifest a CmpL result in the normal flags.  Only good for LT or GE
// compares.  Can be used for LE or GT compares by reversing arguments.
// NOT GOOD FOR EQ/NE tests.
instruct cmpL_zero_flags_LTGE( flagsReg_long_LTGE flags, eRegL src, immL0 zero ) %{
  match( Set flags (CmpL src zero ));
  ins_cost(100);
  format %{ "TEST   $src.hi,$src.hi" %}
  opcode(0x85);
  ins_encode( OpcP, RegReg_Hi2( src, src ) );
  ins_pipe( ialu_cr_reg_reg );
%}

// Manifest a CmpL result in the normal flags.  Only good for LT or GE
// compares.  Can be used for LE or GT compares by reversing arguments.
// NOT GOOD FOR EQ/NE tests.
instruct cmpL_reg_flags_LTGE( flagsReg_long_LTGE flags, eRegL src1, eRegL src2, rRegI tmp ) %{
  match( Set flags (CmpL src1 src2 ));
  effect( TEMP tmp );
  ins_cost(300);
  format %{ "CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits\n\t"
            "MOV    $tmp,$src1.hi\n\t"
            "SBB    $tmp,$src2.hi\t! Compute flags for long compare" %}
  ins_encode( long_cmp_flags2( src1, src2, tmp ) );
  ins_pipe( ialu_cr_reg_reg );
%}

// Long compares reg < zero/req OR reg >= zero/req.
// Just a wrapper for a normal branch, plus the predicate test.
instruct cmpL_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, label labl) %{
  match(If cmp flags);
  effect(USE labl);
  predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
  expand %{
    jmpCon(cmp,flags,labl);    // JLT or JGE...
  %}
%}

// Compare 2 longs and CMOVE longs.
instruct cmovLL_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegL dst, eRegL src) %{
  match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
  ins_cost(400);
  format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
            "CMOV$cmp $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
  ins_pipe( pipe_cmov_reg_long );
%}

instruct cmovLL_mem_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegL dst, load_long_memory src) %{
  match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
  ins_cost(500);
  format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
            "CMOV$cmp $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
  ins_pipe( pipe_cmov_reg_long );
%}

// Compare 2 longs and CMOVE ints.
instruct cmovII_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, rRegI dst, rRegI src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
  match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovII_mem_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, rRegI dst, memory src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
  match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
  ins_cost(250);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
  ins_pipe( pipe_cmov_mem );
%}

// Compare 2 longs and CMOVE ints.
instruct cmovPP_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegP dst, eRegP src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
  match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

// Compare 2 longs and CMOVE doubles
instruct cmovDDPR_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regDPR dst, regDPR src) %{
  predicate( UseSSE<=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
  match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovDPR_regS(cmp,flags,dst,src);
  %}
%}

// Compare 2 longs and CMOVE doubles
instruct cmovDD_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regD dst, regD src) %{
  predicate( UseSSE>=2 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
  match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovD_regS(cmp,flags,dst,src);
  %}
%}

instruct cmovFFPR_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regFPR dst, regFPR src) %{
  predicate( UseSSE==0 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
  match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovFPR_regS(cmp,flags,dst,src);
  %}
%}

instruct cmovFF_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regF dst, regF src) %{
  predicate( UseSSE>=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
  match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovF_regS(cmp,flags,dst,src);
  %}
%}

//======
// Manifest a CmpL result in the normal flags.  Only good for EQ/NE compares.
instruct cmpL_zero_flags_EQNE( flagsReg_long_EQNE flags, eRegL src, immL0 zero, rRegI tmp ) %{
  match( Set flags (CmpL src zero ));
  effect(TEMP tmp);
  ins_cost(200);
  format %{ "MOV    $tmp,$src.lo\n\t"
            "OR     $tmp,$src.hi\t! Long is EQ/NE 0?" %}
  ins_encode( long_cmp_flags0( src, tmp ) );
  ins_pipe( ialu_reg_reg_long );
%}

// Manifest a CmpL result in the normal flags.  Only good for EQ/NE compares.
instruct cmpL_reg_flags_EQNE( flagsReg_long_EQNE flags, eRegL src1, eRegL src2 ) %{
  match( Set flags (CmpL src1 src2 ));
  ins_cost(200+300);
  format %{ "CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits\n\t"
            "JNE,s  skip\n\t"
            "CMP    $src1.hi,$src2.hi\n\t"
     "skip:\t" %}
  ins_encode( long_cmp_flags1( src1, src2 ) );
  ins_pipe( ialu_cr_reg_reg );
%}

// Long compare reg == zero/reg OR reg != zero/reg
// Just a wrapper for a normal branch, plus the predicate test.
instruct cmpL_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, label labl) %{
  match(If cmp flags);
  effect(USE labl);
  predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
  expand %{
    jmpCon(cmp,flags,labl);    // JEQ or JNE...
  %}
%}

// Compare 2 longs and CMOVE longs.
instruct cmovLL_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegL dst, eRegL src) %{
  match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
  ins_cost(400);
  format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
            "CMOV$cmp $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
  ins_pipe( pipe_cmov_reg_long );
%}

instruct cmovLL_mem_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegL dst, load_long_memory src) %{
  match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
  ins_cost(500);
  format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
            "CMOV$cmp $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
  ins_pipe( pipe_cmov_reg_long );
%}

// Compare 2 longs and CMOVE ints.
instruct cmovII_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, rRegI dst, rRegI src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
  match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovII_mem_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, rRegI dst, memory src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
  match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
  ins_cost(250);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
  ins_pipe( pipe_cmov_mem );
%}

// Compare 2 longs and CMOVE ints.
instruct cmovPP_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegP dst, eRegP src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
  match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

// Compare 2 longs and CMOVE doubles
instruct cmovDDPR_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regDPR dst, regDPR src) %{
  predicate( UseSSE<=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
  match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovDPR_regS(cmp,flags,dst,src);
  %}
%}

// Compare 2 longs and CMOVE doubles
instruct cmovDD_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regD dst, regD src) %{
  predicate( UseSSE>=2 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
  match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovD_regS(cmp,flags,dst,src);
  %}
%}

instruct cmovFFPR_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regFPR dst, regFPR src) %{
  predicate( UseSSE==0 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
  match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovFPR_regS(cmp,flags,dst,src);
  %}
%}

instruct cmovFF_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regF dst, regF src) %{
  predicate( UseSSE>=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
  match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovF_regS(cmp,flags,dst,src);
  %}
%}

//======
// Manifest a CmpL result in the normal flags.  Only good for LE or GT compares.
// Same as cmpL_reg_flags_LEGT except must negate src
instruct cmpL_zero_flags_LEGT( flagsReg_long_LEGT flags, eRegL src, immL0 zero, rRegI tmp ) %{
  match( Set flags (CmpL src zero ));
  effect( TEMP tmp );
  ins_cost(300);
  format %{ "XOR    $tmp,$tmp\t# Long compare for -$src < 0, use commuted test\n\t"
            "CMP    $tmp,$src.lo\n\t"
            "SBB    $tmp,$src.hi\n\t" %}
  ins_encode( long_cmp_flags3(src, tmp) );
  ins_pipe( ialu_reg_reg_long );
%}

// Manifest a CmpL result in the normal flags.  Only good for LE or GT compares.
// Same as cmpL_reg_flags_LTGE except operands swapped.  Swapping operands
// requires a commuted test to get the same result.
instruct cmpL_reg_flags_LEGT( flagsReg_long_LEGT flags, eRegL src1, eRegL src2, rRegI tmp ) %{
  match( Set flags (CmpL src1 src2 ));
  effect( TEMP tmp );
  ins_cost(300);
  format %{ "CMP    $src2.lo,$src1.lo\t! Long compare, swapped operands, use with commuted test\n\t"
            "MOV    $tmp,$src2.hi\n\t"
            "SBB    $tmp,$src1.hi\t! Compute flags for long compare" %}
  ins_encode( long_cmp_flags2( src2, src1, tmp ) );
  ins_pipe( ialu_cr_reg_reg );
%}

// Long compares reg < zero/req OR reg >= zero/req.
// Just a wrapper for a normal branch, plus the predicate test
instruct cmpL_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, label labl) %{
  match(If cmp flags);
  effect(USE labl);
  predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le );
  ins_cost(300);
  expand %{
    jmpCon(cmp,flags,labl);    // JGT or JLE...
  %}
%}

// Compare 2 longs and CMOVE longs.
instruct cmovLL_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegL dst, eRegL src) %{
  match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
  ins_cost(400);
  format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
            "CMOV$cmp $dst.hi,$src.hi" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
  ins_pipe( pipe_cmov_reg_long );
%}

instruct cmovLL_mem_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegL dst, load_long_memory src) %{
  match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
  ins_cost(500);
  format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
            "CMOV$cmp $dst.hi,$src.hi+4" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
  ins_pipe( pipe_cmov_reg_long );
%}

// Compare 2 longs and CMOVE ints.
instruct cmovII_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, rRegI dst, rRegI src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
  match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

instruct cmovII_mem_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, rRegI dst, memory src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
  match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
  ins_cost(250);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
  ins_pipe( pipe_cmov_mem );
%}

// Compare 2 longs and CMOVE ptrs.
instruct cmovPP_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegP dst, eRegP src) %{
  predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
  match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  format %{ "CMOV$cmp $dst,$src" %}
  opcode(0x0F,0x40);
  ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
  ins_pipe( pipe_cmov_reg );
%}

// Compare 2 longs and CMOVE doubles
instruct cmovDDPR_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regDPR dst, regDPR src) %{
  predicate( UseSSE<=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
  match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovDPR_regS(cmp,flags,dst,src);
  %}
%}

// Compare 2 longs and CMOVE doubles
instruct cmovDD_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regD dst, regD src) %{
  predicate( UseSSE>=2 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
  match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovD_regS(cmp,flags,dst,src);
  %}
%}

instruct cmovFFPR_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regFPR dst, regFPR src) %{
  predicate( UseSSE==0 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
  match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovFPR_regS(cmp,flags,dst,src);
  %}
%}


instruct cmovFF_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regF dst, regF src) %{
  predicate( UseSSE>=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
  match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
  ins_cost(200);
  expand %{
    fcmovF_regS(cmp,flags,dst,src);
  %}
%}


// ============================================================================
// Procedure Call/Return Instructions
// Call Java Static Instruction
// Note: If this code changes, the corresponding ret_addr_offset() and
//       compute_padding() functions will have to be adjusted.
instruct CallStaticJavaDirect(method meth) %{
  match(CallStaticJava);
  predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
  effect(USE meth);

  ins_cost(300);
  format %{ "CALL,static " %}
  opcode(0xE8); /* E8 cd */
  ins_encode( pre_call_resets,
              Java_Static_Call( meth ),
              call_epilog,
              post_call_FPU );
  ins_pipe( pipe_slow );
  ins_alignment(4);
%}

// Call Java Static Instruction (method handle version)
// Note: If this code changes, the corresponding ret_addr_offset() and
//       compute_padding() functions will have to be adjusted.
instruct CallStaticJavaHandle(method meth, eBPRegP ebp_mh_SP_save) %{
  match(CallStaticJava);
  predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
  effect(USE meth);
  // EBP is saved by all callees (for interpreter stack correction).
  // We use it here for a similar purpose, in {preserve,restore}_SP.

  ins_cost(300);
  format %{ "CALL,static/MethodHandle " %}
  opcode(0xE8); /* E8 cd */
  ins_encode( pre_call_resets,
              preserve_SP,
              Java_Static_Call( meth ),
              restore_SP,
              call_epilog,
              post_call_FPU );
  ins_pipe( pipe_slow );
  ins_alignment(4);
%}

// Call Java Dynamic Instruction
// Note: If this code changes, the corresponding ret_addr_offset() and
//       compute_padding() functions will have to be adjusted.
instruct CallDynamicJavaDirect(method meth) %{
  match(CallDynamicJava);
  effect(USE meth);

  ins_cost(300);
  format %{ "MOV    EAX,(oop)-1\n\t"
            "CALL,dynamic" %}
  opcode(0xE8); /* E8 cd */
  ins_encode( pre_call_resets,
              Java_Dynamic_Call( meth ),
              call_epilog,
              post_call_FPU );
  ins_pipe( pipe_slow );
  ins_alignment(4);
%}

// Call Runtime Instruction
instruct CallRuntimeDirect(method meth) %{
  match(CallRuntime );
  effect(USE meth);

  ins_cost(300);
  format %{ "CALL,runtime " %}
  opcode(0xE8); /* E8 cd */
  // Use FFREEs to clear entries in float stack
  ins_encode( pre_call_resets,
              FFree_Float_Stack_All,
              Java_To_Runtime( meth ),
              post_call_FPU );
  ins_pipe( pipe_slow );
%}

// Call runtime without safepoint
instruct CallLeafDirect(method meth) %{
  match(CallLeaf);
  effect(USE meth);

  ins_cost(300);
  format %{ "CALL_LEAF,runtime " %}
  opcode(0xE8); /* E8 cd */
  ins_encode( pre_call_resets,
              FFree_Float_Stack_All,
              Java_To_Runtime( meth ),
              Verify_FPU_For_Leaf, post_call_FPU );
  ins_pipe( pipe_slow );
%}

instruct CallLeafNoFPDirect(method meth) %{
  match(CallLeafNoFP);
  effect(USE meth);

  ins_cost(300);
  format %{ "CALL_LEAF_NOFP,runtime " %}
  opcode(0xE8); /* E8 cd */
  ins_encode(Java_To_Runtime(meth));
  ins_pipe( pipe_slow );
%}


// Return Instruction
// Remove the return address & jump to it.
instruct Ret() %{
  match(Return);
  format %{ "RET" %}
  opcode(0xC3);
  ins_encode(OpcP);
  ins_pipe( pipe_jmp );
%}

// Tail Call; Jump from runtime stub to Java code.
// Also known as an 'interprocedural jump'.
// Target of jump will eventually return to caller.
// TailJump below removes the return address.
instruct TailCalljmpInd(eRegP_no_EBP jump_target, eBXRegP method_oop) %{
  match(TailCall jump_target method_oop );
  ins_cost(300);
  format %{ "JMP    $jump_target \t# EBX holds method oop" %}
  opcode(0xFF, 0x4);  /* Opcode FF /4 */
  ins_encode( OpcP, RegOpc(jump_target) );
  ins_pipe( pipe_jmp );
%}


// Tail Jump; remove the return address; jump to target.
// TailCall above leaves the return address around.
instruct tailjmpInd(eRegP_no_EBP jump_target, eAXRegP ex_oop) %{
  match( TailJump jump_target ex_oop );
  ins_cost(300);
  format %{ "POP    EDX\t# pop return address into dummy\n\t"
            "JMP    $jump_target " %}
  opcode(0xFF, 0x4);  /* Opcode FF /4 */
  ins_encode( enc_pop_rdx,
              OpcP, RegOpc(jump_target) );
  ins_pipe( pipe_jmp );
%}

// Create exception oop: created by stack-crawling runtime code.
// Created exception is now available to this handler, and is setup
// just prior to jumping to this handler.  No code emitted.
instruct CreateException( eAXRegP ex_oop )
%{
  match(Set ex_oop (CreateEx));

  size(0);
  // use the following format syntax
  format %{ "# exception oop is in EAX; no code emitted" %}
  ins_encode();
  ins_pipe( empty );
%}


// Rethrow exception:
// The exception oop will come in the first argument position.
// Then JUMP (not call) to the rethrow stub code.
instruct RethrowException()
%{
  match(Rethrow);

  // use the following format syntax
  format %{ "JMP    rethrow_stub" %}
  ins_encode(enc_rethrow);
  ins_pipe( pipe_jmp );
%}

// inlined locking and unlocking


instruct cmpFastLock( eFlagsReg cr, eRegP object, eBXRegP box, eAXRegI tmp, eRegP scr) %{
  match( Set cr (FastLock object box) );
  effect( TEMP tmp, TEMP scr, USE_KILL box );
  ins_cost(300);
  format %{ "FASTLOCK $object,$box\t! kills $box,$tmp,$scr" %}
  ins_encode( Fast_Lock(object,box,tmp,scr) );
  ins_pipe( pipe_slow );
%}

instruct cmpFastUnlock( eFlagsReg cr, eRegP object, eAXRegP box, eRegP tmp ) %{
  match( Set cr (FastUnlock object box) );
  effect( TEMP tmp, USE_KILL box );
  ins_cost(300);
  format %{ "FASTUNLOCK $object,$box\t! kills $box,$tmp" %}
  ins_encode( Fast_Unlock(object,box,tmp) );
  ins_pipe( pipe_slow );
%}



// ============================================================================
// Safepoint Instruction
instruct safePoint_poll(eFlagsReg cr) %{
  match(SafePoint);
  effect(KILL cr);

  // TODO-FIXME: we currently poll at offset 0 of the safepoint polling page.
  // On SPARC that might be acceptable as we can generate the address with
  // just a sethi, saving an or.  By polling at offset 0 we can end up
  // putting additional pressure on the index-0 in the D$.  Because of
  // alignment (just like the situation at hand) the lower indices tend
  // to see more traffic.  It'd be better to change the polling address
  // to offset 0 of the last $line in the polling page.

  format %{ "TSTL   #polladdr,EAX\t! Safepoint: poll for GC" %}
  ins_cost(125);
  size(6) ;
  ins_encode( Safepoint_Poll() );
  ins_pipe( ialu_reg_mem );
%}


// ============================================================================
// This name is KNOWN by the ADLC and cannot be changed.
// The ADLC forces a 'TypeRawPtr::BOTTOM' output type
// for this guy.
instruct tlsLoadP(eRegP dst, eFlagsReg cr) %{
  match(Set dst (ThreadLocal));
  effect(DEF dst, KILL cr);

  format %{ "MOV    $dst, Thread::current()" %}
  ins_encode %{
    Register dstReg = as_Register($dst$$reg);
    __ get_thread(dstReg);
  %}
  ins_pipe( ialu_reg_fat );
%}



//----------PEEPHOLE RULES-----------------------------------------------------
// These must follow all instruction definitions as they use the names
// defined in the instructions definitions.
//
// peepmatch ( root_instr_name [preceding_instruction]* );
//
// peepconstraint %{
// (instruction_number.operand_name relational_op instruction_number.operand_name
//  [, ...] );
// // instruction numbers are zero-based using left to right order in peepmatch
//
// peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
// // provide an instruction_number.operand_name for each operand that appears
// // in the replacement instruction's match rule
//
// ---------VM FLAGS---------------------------------------------------------
//
// All peephole optimizations can be turned off using -XX:-OptoPeephole
//
// Each peephole rule is given an identifying number starting with zero and
// increasing by one in the order seen by the parser.  An individual peephole
// can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
// on the command-line.
//
// ---------CURRENT LIMITATIONS----------------------------------------------
//
// Only match adjacent instructions in same basic block
// Only equality constraints
// Only constraints between operands, not (0.dest_reg == EAX_enc)
// Only one replacement instruction
//
// ---------EXAMPLE----------------------------------------------------------
//
// // pertinent parts of existing instructions in architecture description
// instruct movI(rRegI dst, rRegI src) %{
//   match(Set dst (CopyI src));
// %}
//
// instruct incI_eReg(rRegI dst, immI1 src, eFlagsReg cr) %{
//   match(Set dst (AddI dst src));
//   effect(KILL cr);
// %}
//
// // Change (inc mov) to lea
// peephole %{
//   // increment preceeded by register-register move
//   peepmatch ( incI_eReg movI );
//   // require that the destination register of the increment
//   // match the destination register of the move
//   peepconstraint ( 0.dst == 1.dst );
//   // construct a replacement instruction that sets
//   // the destination to ( move's source register + one )
//   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
// %}
//
// Implementation no longer uses movX instructions since
// machine-independent system no longer uses CopyX nodes.
//
// peephole %{
//   peepmatch ( incI_eReg movI );
//   peepconstraint ( 0.dst == 1.dst );
//   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
// %}
//
// peephole %{
//   peepmatch ( decI_eReg movI );
//   peepconstraint ( 0.dst == 1.dst );
//   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
// %}
//
// peephole %{
//   peepmatch ( addI_eReg_imm movI );
//   peepconstraint ( 0.dst == 1.dst );
//   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
// %}
//
// peephole %{
//   peepmatch ( addP_eReg_imm movP );
//   peepconstraint ( 0.dst == 1.dst );
//   peepreplace ( leaP_eReg_immI( 0.dst 1.src 0.src ) );
// %}

// // Change load of spilled value to only a spill
// instruct storeI(memory mem, rRegI src) %{
//   match(Set mem (StoreI mem src));
// %}
//
// instruct loadI(rRegI dst, memory mem) %{
//   match(Set dst (LoadI mem));
// %}
//
peephole %{
  peepmatch ( loadI storeI );
  peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
  peepreplace ( storeI( 1.mem 1.mem 1.src ) );
%}

//----------SMARTSPILL RULES---------------------------------------------------
// These must follow all instruction definitions as they use the names
// defined in the instructions definitions.