xref: /openjdk10/hotspot/src/cpu/sparc/vm/sparc.ad

//
// Copyright (c) 1998, 2010, 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.
//
//

// SPARC 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, vm name );
// 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.


// ----------------------------
// Integer/Long Registers
// ----------------------------

// Need to expose the hi/lo aspect of 64-bit registers
// This register set is used for both the 64-bit build and
// the 32-bit build with 1-register longs.

// Global Registers 0-7
reg_def R_G0H( NS,  NS, Op_RegI,128, G0->as_VMReg()->next());
reg_def R_G0 ( NS,  NS, Op_RegI,  0, G0->as_VMReg());
reg_def R_G1H(SOC, SOC, Op_RegI,129, G1->as_VMReg()->next());
reg_def R_G1 (SOC, SOC, Op_RegI,  1, G1->as_VMReg());
reg_def R_G2H( NS,  NS, Op_RegI,130, G2->as_VMReg()->next());
reg_def R_G2 ( NS,  NS, Op_RegI,  2, G2->as_VMReg());
reg_def R_G3H(SOC, SOC, Op_RegI,131, G3->as_VMReg()->next());
reg_def R_G3 (SOC, SOC, Op_RegI,  3, G3->as_VMReg());
reg_def R_G4H(SOC, SOC, Op_RegI,132, G4->as_VMReg()->next());
reg_def R_G4 (SOC, SOC, Op_RegI,  4, G4->as_VMReg());
reg_def R_G5H(SOC, SOC, Op_RegI,133, G5->as_VMReg()->next());
reg_def R_G5 (SOC, SOC, Op_RegI,  5, G5->as_VMReg());
reg_def R_G6H( NS,  NS, Op_RegI,134, G6->as_VMReg()->next());
reg_def R_G6 ( NS,  NS, Op_RegI,  6, G6->as_VMReg());
reg_def R_G7H( NS,  NS, Op_RegI,135, G7->as_VMReg()->next());
reg_def R_G7 ( NS,  NS, Op_RegI,  7, G7->as_VMReg());

// Output Registers 0-7
reg_def R_O0H(SOC, SOC, Op_RegI,136, O0->as_VMReg()->next());
reg_def R_O0 (SOC, SOC, Op_RegI,  8, O0->as_VMReg());
reg_def R_O1H(SOC, SOC, Op_RegI,137, O1->as_VMReg()->next());
reg_def R_O1 (SOC, SOC, Op_RegI,  9, O1->as_VMReg());
reg_def R_O2H(SOC, SOC, Op_RegI,138, O2->as_VMReg()->next());
reg_def R_O2 (SOC, SOC, Op_RegI, 10, O2->as_VMReg());
reg_def R_O3H(SOC, SOC, Op_RegI,139, O3->as_VMReg()->next());
reg_def R_O3 (SOC, SOC, Op_RegI, 11, O3->as_VMReg());
reg_def R_O4H(SOC, SOC, Op_RegI,140, O4->as_VMReg()->next());
reg_def R_O4 (SOC, SOC, Op_RegI, 12, O4->as_VMReg());
reg_def R_O5H(SOC, SOC, Op_RegI,141, O5->as_VMReg()->next());
reg_def R_O5 (SOC, SOC, Op_RegI, 13, O5->as_VMReg());
reg_def R_SPH( NS,  NS, Op_RegI,142, SP->as_VMReg()->next());
reg_def R_SP ( NS,  NS, Op_RegI, 14, SP->as_VMReg());
reg_def R_O7H(SOC, SOC, Op_RegI,143, O7->as_VMReg()->next());
reg_def R_O7 (SOC, SOC, Op_RegI, 15, O7->as_VMReg());

// Local Registers 0-7
reg_def R_L0H( NS,  NS, Op_RegI,144, L0->as_VMReg()->next());
reg_def R_L0 ( NS,  NS, Op_RegI, 16, L0->as_VMReg());
reg_def R_L1H( NS,  NS, Op_RegI,145, L1->as_VMReg()->next());
reg_def R_L1 ( NS,  NS, Op_RegI, 17, L1->as_VMReg());
reg_def R_L2H( NS,  NS, Op_RegI,146, L2->as_VMReg()->next());
reg_def R_L2 ( NS,  NS, Op_RegI, 18, L2->as_VMReg());
reg_def R_L3H( NS,  NS, Op_RegI,147, L3->as_VMReg()->next());
reg_def R_L3 ( NS,  NS, Op_RegI, 19, L3->as_VMReg());
reg_def R_L4H( NS,  NS, Op_RegI,148, L4->as_VMReg()->next());
reg_def R_L4 ( NS,  NS, Op_RegI, 20, L4->as_VMReg());
reg_def R_L5H( NS,  NS, Op_RegI,149, L5->as_VMReg()->next());
reg_def R_L5 ( NS,  NS, Op_RegI, 21, L5->as_VMReg());
reg_def R_L6H( NS,  NS, Op_RegI,150, L6->as_VMReg()->next());
reg_def R_L6 ( NS,  NS, Op_RegI, 22, L6->as_VMReg());
reg_def R_L7H( NS,  NS, Op_RegI,151, L7->as_VMReg()->next());
reg_def R_L7 ( NS,  NS, Op_RegI, 23, L7->as_VMReg());

// Input Registers 0-7
reg_def R_I0H( NS,  NS, Op_RegI,152, I0->as_VMReg()->next());
reg_def R_I0 ( NS,  NS, Op_RegI, 24, I0->as_VMReg());
reg_def R_I1H( NS,  NS, Op_RegI,153, I1->as_VMReg()->next());
reg_def R_I1 ( NS,  NS, Op_RegI, 25, I1->as_VMReg());
reg_def R_I2H( NS,  NS, Op_RegI,154, I2->as_VMReg()->next());
reg_def R_I2 ( NS,  NS, Op_RegI, 26, I2->as_VMReg());
reg_def R_I3H( NS,  NS, Op_RegI,155, I3->as_VMReg()->next());
reg_def R_I3 ( NS,  NS, Op_RegI, 27, I3->as_VMReg());
reg_def R_I4H( NS,  NS, Op_RegI,156, I4->as_VMReg()->next());
reg_def R_I4 ( NS,  NS, Op_RegI, 28, I4->as_VMReg());
reg_def R_I5H( NS,  NS, Op_RegI,157, I5->as_VMReg()->next());
reg_def R_I5 ( NS,  NS, Op_RegI, 29, I5->as_VMReg());
reg_def R_FPH( NS,  NS, Op_RegI,158, FP->as_VMReg()->next());
reg_def R_FP ( NS,  NS, Op_RegI, 30, FP->as_VMReg());
reg_def R_I7H( NS,  NS, Op_RegI,159, I7->as_VMReg()->next());
reg_def R_I7 ( NS,  NS, Op_RegI, 31, I7->as_VMReg());

// ----------------------------
// Float/Double Registers
// ----------------------------

// Float Registers
reg_def R_F0 ( SOC, SOC, Op_RegF,  0, F0->as_VMReg());
reg_def R_F1 ( SOC, SOC, Op_RegF,  1, F1->as_VMReg());
reg_def R_F2 ( SOC, SOC, Op_RegF,  2, F2->as_VMReg());
reg_def R_F3 ( SOC, SOC, Op_RegF,  3, F3->as_VMReg());
reg_def R_F4 ( SOC, SOC, Op_RegF,  4, F4->as_VMReg());
reg_def R_F5 ( SOC, SOC, Op_RegF,  5, F5->as_VMReg());
reg_def R_F6 ( SOC, SOC, Op_RegF,  6, F6->as_VMReg());
reg_def R_F7 ( SOC, SOC, Op_RegF,  7, F7->as_VMReg());
reg_def R_F8 ( SOC, SOC, Op_RegF,  8, F8->as_VMReg());
reg_def R_F9 ( SOC, SOC, Op_RegF,  9, F9->as_VMReg());
reg_def R_F10( SOC, SOC, Op_RegF, 10, F10->as_VMReg());
reg_def R_F11( SOC, SOC, Op_RegF, 11, F11->as_VMReg());
reg_def R_F12( SOC, SOC, Op_RegF, 12, F12->as_VMReg());
reg_def R_F13( SOC, SOC, Op_RegF, 13, F13->as_VMReg());
reg_def R_F14( SOC, SOC, Op_RegF, 14, F14->as_VMReg());
reg_def R_F15( SOC, SOC, Op_RegF, 15, F15->as_VMReg());
reg_def R_F16( SOC, SOC, Op_RegF, 16, F16->as_VMReg());
reg_def R_F17( SOC, SOC, Op_RegF, 17, F17->as_VMReg());
reg_def R_F18( SOC, SOC, Op_RegF, 18, F18->as_VMReg());
reg_def R_F19( SOC, SOC, Op_RegF, 19, F19->as_VMReg());
reg_def R_F20( SOC, SOC, Op_RegF, 20, F20->as_VMReg());
reg_def R_F21( SOC, SOC, Op_RegF, 21, F21->as_VMReg());
reg_def R_F22( SOC, SOC, Op_RegF, 22, F22->as_VMReg());
reg_def R_F23( SOC, SOC, Op_RegF, 23, F23->as_VMReg());
reg_def R_F24( SOC, SOC, Op_RegF, 24, F24->as_VMReg());
reg_def R_F25( SOC, SOC, Op_RegF, 25, F25->as_VMReg());
reg_def R_F26( SOC, SOC, Op_RegF, 26, F26->as_VMReg());
reg_def R_F27( SOC, SOC, Op_RegF, 27, F27->as_VMReg());
reg_def R_F28( SOC, SOC, Op_RegF, 28, F28->as_VMReg());
reg_def R_F29( SOC, SOC, Op_RegF, 29, F29->as_VMReg());
reg_def R_F30( SOC, SOC, Op_RegF, 30, F30->as_VMReg());
reg_def R_F31( SOC, SOC, Op_RegF, 31, F31->as_VMReg());

// Double Registers
// The rules of ADL require that double registers be defined in pairs.
// Each pair must be two 32-bit values, but not necessarily a pair of
// single float registers.  In each pair, ADLC-assigned register numbers
// must be adjacent, with the lower number even.  Finally, when the
// CPU stores such a register pair to memory, the word associated with
// the lower ADLC-assigned number must be stored to the lower address.

// These definitions specify the actual bit encodings of the sparc
// double fp register numbers.  FloatRegisterImpl in register_sparc.hpp
// wants 0-63, so we have to convert every time we want to use fp regs
// with the macroassembler, using reg_to_DoubleFloatRegister_object().
// 255 is a flag meaning "don't go here".
// I believe we can't handle callee-save doubles D32 and up until
// the place in the sparc stack crawler that asserts on the 255 is
// fixed up.
reg_def R_D32 (SOC, SOC, Op_RegD,  1, F32->as_VMReg());
reg_def R_D32x(SOC, SOC, Op_RegD,255, F32->as_VMReg()->next());
reg_def R_D34 (SOC, SOC, Op_RegD,  3, F34->as_VMReg());
reg_def R_D34x(SOC, SOC, Op_RegD,255, F34->as_VMReg()->next());
reg_def R_D36 (SOC, SOC, Op_RegD,  5, F36->as_VMReg());
reg_def R_D36x(SOC, SOC, Op_RegD,255, F36->as_VMReg()->next());
reg_def R_D38 (SOC, SOC, Op_RegD,  7, F38->as_VMReg());
reg_def R_D38x(SOC, SOC, Op_RegD,255, F38->as_VMReg()->next());
reg_def R_D40 (SOC, SOC, Op_RegD,  9, F40->as_VMReg());
reg_def R_D40x(SOC, SOC, Op_RegD,255, F40->as_VMReg()->next());
reg_def R_D42 (SOC, SOC, Op_RegD, 11, F42->as_VMReg());
reg_def R_D42x(SOC, SOC, Op_RegD,255, F42->as_VMReg()->next());
reg_def R_D44 (SOC, SOC, Op_RegD, 13, F44->as_VMReg());
reg_def R_D44x(SOC, SOC, Op_RegD,255, F44->as_VMReg()->next());
reg_def R_D46 (SOC, SOC, Op_RegD, 15, F46->as_VMReg());
reg_def R_D46x(SOC, SOC, Op_RegD,255, F46->as_VMReg()->next());
reg_def R_D48 (SOC, SOC, Op_RegD, 17, F48->as_VMReg());
reg_def R_D48x(SOC, SOC, Op_RegD,255, F48->as_VMReg()->next());
reg_def R_D50 (SOC, SOC, Op_RegD, 19, F50->as_VMReg());
reg_def R_D50x(SOC, SOC, Op_RegD,255, F50->as_VMReg()->next());
reg_def R_D52 (SOC, SOC, Op_RegD, 21, F52->as_VMReg());
reg_def R_D52x(SOC, SOC, Op_RegD,255, F52->as_VMReg()->next());
reg_def R_D54 (SOC, SOC, Op_RegD, 23, F54->as_VMReg());
reg_def R_D54x(SOC, SOC, Op_RegD,255, F54->as_VMReg()->next());
reg_def R_D56 (SOC, SOC, Op_RegD, 25, F56->as_VMReg());
reg_def R_D56x(SOC, SOC, Op_RegD,255, F56->as_VMReg()->next());
reg_def R_D58 (SOC, SOC, Op_RegD, 27, F58->as_VMReg());
reg_def R_D58x(SOC, SOC, Op_RegD,255, F58->as_VMReg()->next());
reg_def R_D60 (SOC, SOC, Op_RegD, 29, F60->as_VMReg());
reg_def R_D60x(SOC, SOC, Op_RegD,255, F60->as_VMReg()->next());
reg_def R_D62 (SOC, SOC, Op_RegD, 31, F62->as_VMReg());
reg_def R_D62x(SOC, SOC, Op_RegD,255, F62->as_VMReg()->next());


// ----------------------------
// Special Registers
// Condition Codes Flag Registers
// I tried to break out ICC and XCC but it's not very pretty.
// Every Sparc instruction which defs/kills one also kills the other.
// Hence every compare instruction which defs one kind of flags ends
// up needing a kill of the other.
reg_def CCR (SOC, SOC,  Op_RegFlags, 0, VMRegImpl::Bad());

reg_def FCC0(SOC, SOC,  Op_RegFlags, 0, VMRegImpl::Bad());
reg_def FCC1(SOC, SOC,  Op_RegFlags, 1, VMRegImpl::Bad());
reg_def FCC2(SOC, SOC,  Op_RegFlags, 2, VMRegImpl::Bad());
reg_def FCC3(SOC, SOC,  Op_RegFlags, 3, VMRegImpl::Bad());

// ----------------------------
// Specify the enum values for the registers.  These enums are only used by the
// OptoReg "class". We can convert these enum values at will to VMReg when needed
// for visibility to the rest of the vm. The order of this enum influences the
// register allocator so having the freedom to set this order and not be stuck
// with the order that is natural for the rest of the vm is worth it.
alloc_class chunk0(
  R_L0,R_L0H, R_L1,R_L1H, R_L2,R_L2H, R_L3,R_L3H, R_L4,R_L4H, R_L5,R_L5H, R_L6,R_L6H, R_L7,R_L7H,
  R_G0,R_G0H, R_G1,R_G1H, R_G2,R_G2H, R_G3,R_G3H, R_G4,R_G4H, R_G5,R_G5H, R_G6,R_G6H, R_G7,R_G7H,
  R_O7,R_O7H, R_SP,R_SPH, R_O0,R_O0H, R_O1,R_O1H, R_O2,R_O2H, R_O3,R_O3H, R_O4,R_O4H, R_O5,R_O5H,
  R_I0,R_I0H, R_I1,R_I1H, R_I2,R_I2H, R_I3,R_I3H, R_I4,R_I4H, R_I5,R_I5H, R_FP,R_FPH, R_I7,R_I7H);

// Note that a register is not allocatable unless it is also mentioned
// in a widely-used reg_class below.  Thus, R_G7 and R_G0 are outside i_reg.

alloc_class chunk1(
  // The first registers listed here are those most likely to be used
  // as temporaries.  We move F0..F7 away from the front of the list,
  // to reduce the likelihood of interferences with parameters and
  // return values.  Likewise, we avoid using F0/F1 for parameters,
  // since they are used for return values.
  // This FPU fine-tuning is worth about 1% on the SPEC geomean.
  R_F8 ,R_F9 ,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
  R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,
  R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,R_F30,R_F31,
  R_F0 ,R_F1 ,R_F2 ,R_F3 ,R_F4 ,R_F5 ,R_F6 ,R_F7 , // used for arguments and return values
  R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,
  R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
  R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,
  R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x);

alloc_class chunk2(CCR, FCC0, FCC1, FCC2, FCC3);

//----------Architecture Description Register Classes--------------------------
// Several register classes are automatically defined based upon information in
// this architecture description.
// 1) reg_class inline_cache_reg           ( as defined in frame section )
// 2) reg_class interpreter_method_oop_reg ( as defined in frame section )
// 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
//

// G0 is not included in integer class since it has special meaning.
reg_class g0_reg(R_G0);

// ----------------------------
// Integer Register Classes
// ----------------------------
// Exclusions from i_reg:
// R_G0: hardwired zero
// R_G2: reserved by HotSpot to the TLS register (invariant within Java)
// R_G6: reserved by Solaris ABI to tools
// R_G7: reserved by Solaris ABI to libthread
// R_O7: Used as a temp in many encodings
reg_class int_reg(R_G1,R_G3,R_G4,R_G5,R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);

// Class for all integer registers, except the G registers.  This is used for
// encodings which use G registers as temps.  The regular inputs to such
// instructions use a "notemp_" prefix, as a hack to ensure that the allocator
// will not put an input into a temp register.
reg_class notemp_int_reg(R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);

reg_class g1_regI(R_G1);
reg_class g3_regI(R_G3);
reg_class g4_regI(R_G4);
reg_class o0_regI(R_O0);
reg_class o7_regI(R_O7);

// ----------------------------
// Pointer Register Classes
// ----------------------------
#ifdef _LP64
// 64-bit build means 64-bit pointers means hi/lo pairs
reg_class ptr_reg(            R_G1H,R_G1,             R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
                  R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
                  R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
                  R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
// Lock encodings use G3 and G4 internally
reg_class lock_ptr_reg(       R_G1H,R_G1,                                     R_G5H,R_G5,
                  R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
                  R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
                  R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
// Special class for storeP instructions, which can store SP or RPC to TLS.
// It is also used for memory addressing, allowing direct TLS addressing.
reg_class sp_ptr_reg(         R_G1H,R_G1, R_G2H,R_G2, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
                  R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5, R_SPH,R_SP,
                  R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
                  R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5, R_FPH,R_FP );
// R_L7 is the lowest-priority callee-save (i.e., NS) register
// We use it to save R_G2 across calls out of Java.
reg_class l7_regP(R_L7H,R_L7);

// Other special pointer regs
reg_class g1_regP(R_G1H,R_G1);
reg_class g2_regP(R_G2H,R_G2);
reg_class g3_regP(R_G3H,R_G3);
reg_class g4_regP(R_G4H,R_G4);
reg_class g5_regP(R_G5H,R_G5);
reg_class i0_regP(R_I0H,R_I0);
reg_class o0_regP(R_O0H,R_O0);
reg_class o1_regP(R_O1H,R_O1);
reg_class o2_regP(R_O2H,R_O2);
reg_class o7_regP(R_O7H,R_O7);

#else // _LP64
// 32-bit build means 32-bit pointers means 1 register.
reg_class ptr_reg(     R_G1,     R_G3,R_G4,R_G5,
                  R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
                  R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
                  R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
// Lock encodings use G3 and G4 internally
reg_class lock_ptr_reg(R_G1,               R_G5,
                  R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
                  R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
                  R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
// Special class for storeP instructions, which can store SP or RPC to TLS.
// It is also used for memory addressing, allowing direct TLS addressing.
reg_class sp_ptr_reg(  R_G1,R_G2,R_G3,R_G4,R_G5,
                  R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_SP,
                  R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
                  R_I0,R_I1,R_I2,R_I3,R_I4,R_I5,R_FP);
// R_L7 is the lowest-priority callee-save (i.e., NS) register
// We use it to save R_G2 across calls out of Java.
reg_class l7_regP(R_L7);

// Other special pointer regs
reg_class g1_regP(R_G1);
reg_class g2_regP(R_G2);
reg_class g3_regP(R_G3);
reg_class g4_regP(R_G4);
reg_class g5_regP(R_G5);
reg_class i0_regP(R_I0);
reg_class o0_regP(R_O0);
reg_class o1_regP(R_O1);
reg_class o2_regP(R_O2);
reg_class o7_regP(R_O7);
#endif // _LP64


// ----------------------------
// Long Register Classes
// ----------------------------
// Longs in 1 register.  Aligned adjacent hi/lo pairs.
// Note:  O7 is never in this class; it is sometimes used as an encoding temp.
reg_class long_reg(             R_G1H,R_G1,             R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5
                   ,R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5
#ifdef _LP64
// 64-bit, longs in 1 register: use all 64-bit integer registers
// 32-bit, longs in 1 register: cannot use I's and L's.  Restrict to O's and G's.
                   ,R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7
                   ,R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5
#endif // _LP64
                  );

reg_class g1_regL(R_G1H,R_G1);
reg_class g3_regL(R_G3H,R_G3);
reg_class o2_regL(R_O2H,R_O2);
reg_class o7_regL(R_O7H,R_O7);

// ----------------------------
// Special Class for Condition Code Flags Register
reg_class int_flags(CCR);
reg_class float_flags(FCC0,FCC1,FCC2,FCC3);
reg_class float_flag0(FCC0);


// ----------------------------
// Float Point Register Classes
// ----------------------------
// Skip F30/F31, they are reserved for mem-mem copies
reg_class sflt_reg(R_F0,R_F1,R_F2,R_F3,R_F4,R_F5,R_F6,R_F7,R_F8,R_F9,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);

// Paired floating point registers--they show up in the same order as the floats,
// but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
reg_class dflt_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
                   R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,
                   /* Use extra V9 double registers; this AD file does not support V8 */
                   R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
                   R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x
                   );

// Paired floating point registers--they show up in the same order as the floats,
// but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
// This class is usable for mis-aligned loads as happen in I2C adapters.
reg_class dflt_low_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
                   R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,R_F30,R_F31 );
%}

//----------DEFINITION BLOCK---------------------------------------------------
// Define name --> value mappings to inform the ADLC of an integer valued name
// Current support includes integer values in the range [0, 0x7FFFFFFF]
// Format:
//        int_def  <name>         ( <int_value>, <expression>);
// Generated Code in ad_<arch>.hpp
//        #define  <name>   (<expression>)
//        // value == <int_value>
// Generated code in ad_<arch>.cpp adlc_verification()
//        assert( <name> == <int_value>, "Expect (<expression>) to equal <int_value>");
//
definitions %{
// The default cost (of an ALU instruction).
  int_def DEFAULT_COST      (    100,     100);
  int_def HUGE_COST         (1000000, 1000000);

// Memory refs are twice as expensive as run-of-the-mill.
  int_def MEMORY_REF_COST   (    200, DEFAULT_COST * 2);

// Branches are even more expensive.
  int_def BRANCH_COST       (    300, DEFAULT_COST * 3);
  int_def CALL_COST         (    300, DEFAULT_COST * 3);
%}


//----------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_sparc.cpp
extern bool can_branch_register( Node *bol, Node *cmp );

// Macros to extract hi & lo halves from a long pair.
// G0 is not part of any long pair, so assert on that.
// Prevents accidentally using G1 instead of G0.
#define LONG_HI_REG(x) (x)
#define LONG_LO_REG(x) (x)

%}

source %{
#define __ _masm.

// Block initializing store
#define ASI_BLK_INIT_QUAD_LDD_P    0xE2

// tertiary op of a LoadP or StoreP encoding
#define REGP_OP true

static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding);
static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding);
static Register reg_to_register_object(int register_encoding);

// Used by the DFA in dfa_sparc.cpp.
// Check for being able to use a V9 branch-on-register.  Requires a
// compare-vs-zero, equal/not-equal, of a value which was zero- or sign-
// extended.  Doesn't work following an integer ADD, for example, because of
// overflow (-1 incremented yields 0 plus a carry in the high-order word).  On
// 32-bit V9 systems, interrupts currently blow away the high-order 32 bits and
// replace them with zero, which could become sign-extension in a different OS
// release.  There's no obvious reason why an interrupt will ever fill these
// bits with non-zero junk (the registers are reloaded with standard LD
// instructions which either zero-fill or sign-fill).
bool can_branch_register( Node *bol, Node *cmp ) {
  if( !BranchOnRegister ) return false;
#ifdef _LP64
  if( cmp->Opcode() == Op_CmpP )
    return true;  // No problems with pointer compares
#endif
  if( cmp->Opcode() == Op_CmpL )
    return true;  // No problems with long compares

  if( !SparcV9RegsHiBitsZero ) return false;
  if( bol->as_Bool()->_test._test != BoolTest::ne &&
      bol->as_Bool()->_test._test != BoolTest::eq )
     return false;

  // Check for comparing against a 'safe' value.  Any operation which
  // clears out the high word is safe.  Thus, loads and certain shifts
  // are safe, as are non-negative constants.  Any operation which
  // preserves zero bits in the high word is safe as long as each of its
  // inputs are safe.  Thus, phis and bitwise booleans are safe if their
  // inputs are safe.  At present, the only important case to recognize
  // seems to be loads.  Constants should fold away, and shifts &
  // logicals can use the 'cc' forms.
  Node *x = cmp->in(1);
  if( x->is_Load() ) return true;
  if( x->is_Phi() ) {
    for( uint i = 1; i < x->req(); i++ )
      if( !x->in(i)->is_Load() )
        return false;
    return true;
  }
  return false;
}

// ****************************************************************************

// REQUIRED FUNCTIONALITY

// !!!!! 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.
//       The "return address" is the address of the call instruction, plus 8.

int MachCallStaticJavaNode::ret_addr_offset() {
  int offset = NativeCall::instruction_size;  // call; delay slot
  if (_method_handle_invoke)
    offset += 4;  // restore SP
  return offset;
}

int MachCallDynamicJavaNode::ret_addr_offset() {
  int vtable_index = this->_vtable_index;
  if (vtable_index < 0) {
    // must be invalid_vtable_index, not nonvirtual_vtable_index
    assert(vtable_index == methodOopDesc::invalid_vtable_index, "correct sentinel value");
    return (NativeMovConstReg::instruction_size +
           NativeCall::instruction_size);  // sethi; setlo; call; delay slot
  } else {
    assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
    int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
    int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
    int klass_load_size;
    if (UseCompressedOops) {
      assert(Universe::heap() != NULL, "java heap should be initialized");
      if (Universe::narrow_oop_base() == NULL)
        klass_load_size = 2*BytesPerInstWord; // see MacroAssembler::load_klass()
      else
        klass_load_size = 3*BytesPerInstWord;
    } else {
      klass_load_size = 1*BytesPerInstWord;
    }
    if( Assembler::is_simm13(v_off) ) {
      return klass_load_size +
             (2*BytesPerInstWord +           // ld_ptr, ld_ptr
             NativeCall::instruction_size);  // call; delay slot
    } else {
      return klass_load_size +
             (4*BytesPerInstWord +           // set_hi, set, ld_ptr, ld_ptr
             NativeCall::instruction_size);  // call; delay slot
    }
  }
}

int MachCallRuntimeNode::ret_addr_offset() {
#ifdef _LP64
  return NativeFarCall::instruction_size;  // farcall; delay slot
#else
  return NativeCall::instruction_size;  // call; delay slot
#endif
}

// Indicate if the safepoint node needs the polling page as an input.
// Since Sparc does not have absolute addressing, it does.
bool SafePointNode::needs_polling_address_input() {
  return true;
}

// emit an interrupt that is caught by the debugger (for debugging compiler)
void emit_break(CodeBuffer &cbuf) {
  MacroAssembler _masm(&cbuf);
  __ breakpoint_trap();
}

#ifndef PRODUCT
void MachBreakpointNode::format( PhaseRegAlloc *, outputStream *st ) const {
  st->print("TA");
}
#endif

void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  emit_break(cbuf);
}

uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const {
  return MachNode::size(ra_);
}

// Traceable jump
void  emit_jmpl(CodeBuffer &cbuf, int jump_target) {
  MacroAssembler _masm(&cbuf);
  Register rdest = reg_to_register_object(jump_target);
  __ JMP(rdest, 0);
  __ delayed()->nop();
}

// Traceable jump and set exception pc
void  emit_jmpl_set_exception_pc(CodeBuffer &cbuf, int jump_target) {
  MacroAssembler _masm(&cbuf);
  Register rdest = reg_to_register_object(jump_target);
  __ JMP(rdest, 0);
  __ delayed()->add(O7, frame::pc_return_offset, Oissuing_pc );
}

void emit_nop(CodeBuffer &cbuf) {
  MacroAssembler _masm(&cbuf);
  __ nop();
}

void emit_illtrap(CodeBuffer &cbuf) {
  MacroAssembler _masm(&cbuf);
  __ illtrap(0);
}


intptr_t get_offset_from_base(const MachNode* n, const TypePtr* atype, int disp32) {
  assert(n->rule() != loadUB_rule, "");

  intptr_t offset = 0;
  const TypePtr *adr_type = TYPE_PTR_SENTINAL;  // Check for base==RegI, disp==immP
  const Node* addr = n->get_base_and_disp(offset, adr_type);
  assert(adr_type == (const TypePtr*)-1, "VerifyOops: no support for sparc operands with base==RegI, disp==immP");
  assert(addr != NULL && addr != (Node*)-1, "invalid addr");
  assert(addr->bottom_type()->isa_oopptr() == atype, "");
  atype = atype->add_offset(offset);
  assert(disp32 == offset, "wrong disp32");
  return atype->_offset;
}


intptr_t get_offset_from_base_2(const MachNode* n, const TypePtr* atype, int disp32) {
  assert(n->rule() != loadUB_rule, "");

  intptr_t offset = 0;
  Node* addr = n->in(2);
  assert(addr->bottom_type()->isa_oopptr() == atype, "");
  if (addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP) {
    Node* a = addr->in(2/*AddPNode::Address*/);
    Node* o = addr->in(3/*AddPNode::Offset*/);
    offset = o->is_Con() ? o->bottom_type()->is_intptr_t()->get_con() : Type::OffsetBot;
    atype = a->bottom_type()->is_ptr()->add_offset(offset);
    assert(atype->isa_oop_ptr(), "still an oop");
  }
  offset = atype->is_ptr()->_offset;
  if (offset != Type::OffsetBot)  offset += disp32;
  return offset;
}

// Standard Sparc opcode form2 field breakdown
static inline void emit2_19(CodeBuffer &cbuf, int f30, int f29, int f25, int f22, int f20, int f19, int f0 ) {
  f0 &= (1<<19)-1;     // Mask displacement to 19 bits
  int op = (f30 << 30) |
           (f29 << 29) |
           (f25 << 25) |
           (f22 << 22) |
           (f20 << 20) |
           (f19 << 19) |
           (f0  <<  0);
  cbuf.insts()->emit_int32(op);
}

// Standard Sparc opcode form2 field breakdown
static inline void emit2_22(CodeBuffer &cbuf, int f30, int f25, int f22, int f0 ) {
  f0 >>= 10;           // Drop 10 bits
  f0 &= (1<<22)-1;     // Mask displacement to 22 bits
  int op = (f30 << 30) |
           (f25 << 25) |
           (f22 << 22) |
           (f0  <<  0);
  cbuf.insts()->emit_int32(op);
}

// Standard Sparc opcode form3 field breakdown
static inline void emit3(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int f5, int f0 ) {
  int op = (f30 << 30) |
           (f25 << 25) |
           (f19 << 19) |
           (f14 << 14) |
           (f5  <<  5) |
           (f0  <<  0);
  cbuf.insts()->emit_int32(op);
}

// Standard Sparc opcode form3 field breakdown
static inline void emit3_simm13(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm13 ) {
  simm13 &= (1<<13)-1; // Mask to 13 bits
  int op = (f30 << 30) |
           (f25 << 25) |
           (f19 << 19) |
           (f14 << 14) |
           (1   << 13) | // bit to indicate immediate-mode
           (simm13<<0);
  cbuf.insts()->emit_int32(op);
}

static inline void emit3_simm10(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm10 ) {
  simm10 &= (1<<10)-1; // Mask to 10 bits
  emit3_simm13(cbuf,f30,f25,f19,f14,simm10);
}

#ifdef ASSERT
// Helper function for VerifyOops in emit_form3_mem_reg
void verify_oops_warning(const MachNode *n, int ideal_op, int mem_op) {
  warning("VerifyOops encountered unexpected instruction:");
  n->dump(2);
  warning("Instruction has ideal_Opcode==Op_%s and op_ld==Op_%s \n", NodeClassNames[ideal_op], NodeClassNames[mem_op]);
}
#endif


void emit_form3_mem_reg(CodeBuffer &cbuf, const MachNode* n, int primary, int tertiary,
                        int src1_enc, int disp32, int src2_enc, int dst_enc) {

#ifdef ASSERT
  // The following code implements the +VerifyOops feature.
  // It verifies oop values which are loaded into or stored out of
  // the current method activation.  +VerifyOops complements techniques
  // like ScavengeALot, because it eagerly inspects oops in transit,
  // as they enter or leave the stack, as opposed to ScavengeALot,
  // which inspects oops "at rest", in the stack or heap, at safepoints.
  // For this reason, +VerifyOops can sometimes detect bugs very close
  // to their point of creation.  It can also serve as a cross-check
  // on the validity of oop maps, when used toegether with ScavengeALot.

  // It would be good to verify oops at other points, especially
  // when an oop is used as a base pointer for a load or store.
  // This is presently difficult, because it is hard to know when
  // a base address is biased or not.  (If we had such information,
  // it would be easy and useful to make a two-argument version of
  // verify_oop which unbiases the base, and performs verification.)

  assert((uint)tertiary == 0xFFFFFFFF || tertiary == REGP_OP, "valid tertiary");
  bool is_verified_oop_base  = false;
  bool is_verified_oop_load  = false;
  bool is_verified_oop_store = false;
  int tmp_enc = -1;
  if (VerifyOops && src1_enc != R_SP_enc) {
    // classify the op, mainly for an assert check
    int st_op = 0, ld_op = 0;
    switch (primary) {
    case Assembler::stb_op3:  st_op = Op_StoreB; break;
    case Assembler::sth_op3:  st_op = Op_StoreC; break;
    case Assembler::stx_op3:  // may become StoreP or stay StoreI or StoreD0
    case Assembler::stw_op3:  st_op = Op_StoreI; break;
    case Assembler::std_op3:  st_op = Op_StoreL; break;
    case Assembler::stf_op3:  st_op = Op_StoreF; break;
    case Assembler::stdf_op3: st_op = Op_StoreD; break;

    case Assembler::ldsb_op3: ld_op = Op_LoadB; break;
    case Assembler::lduh_op3: ld_op = Op_LoadUS; break;
    case Assembler::ldsh_op3: ld_op = Op_LoadS; break;
    case Assembler::ldx_op3:  // may become LoadP or stay LoadI
    case Assembler::ldsw_op3: // may become LoadP or stay LoadI
    case Assembler::lduw_op3: ld_op = Op_LoadI; break;
    case Assembler::ldd_op3:  ld_op = Op_LoadL; break;
    case Assembler::ldf_op3:  ld_op = Op_LoadF; break;
    case Assembler::lddf_op3: ld_op = Op_LoadD; break;
    case Assembler::ldub_op3: ld_op = Op_LoadB; break;
    case Assembler::prefetch_op3: ld_op = Op_LoadI; break;

    default: ShouldNotReachHere();
    }
    if (tertiary == REGP_OP) {
      if      (st_op == Op_StoreI)  st_op = Op_StoreP;
      else if (ld_op == Op_LoadI)   ld_op = Op_LoadP;
      else                          ShouldNotReachHere();
      if (st_op) {
        // a store
        // inputs are (0:control, 1:memory, 2:address, 3:value)
        Node* n2 = n->in(3);
        if (n2 != NULL) {
          const Type* t = n2->bottom_type();
          is_verified_oop_store = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
        }
      } else {
        // a load
        const Type* t = n->bottom_type();
        is_verified_oop_load = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
      }
    }

    if (ld_op) {
      // a Load
      // inputs are (0:control, 1:memory, 2:address)
      if (!(n->ideal_Opcode()==ld_op)       && // Following are special cases
          !(n->ideal_Opcode()==Op_LoadLLocked && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_LoadPLocked && ld_op==Op_LoadP) &&
          !(n->ideal_Opcode()==Op_LoadI     && ld_op==Op_LoadF) &&
          !(n->ideal_Opcode()==Op_LoadF     && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_LoadRange && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_LoadKlass && ld_op==Op_LoadP) &&
          !(n->ideal_Opcode()==Op_LoadL     && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_LoadL_unaligned && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_LoadD_unaligned && ld_op==Op_LoadF) &&
          !(n->ideal_Opcode()==Op_ConvI2F   && ld_op==Op_LoadF) &&
          !(n->ideal_Opcode()==Op_ConvI2D   && ld_op==Op_LoadF) &&
          !(n->ideal_Opcode()==Op_PrefetchRead  && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_PrefetchWrite && ld_op==Op_LoadI) &&
          !(n->ideal_Opcode()==Op_Load2I    && ld_op==Op_LoadD) &&
          !(n->ideal_Opcode()==Op_Load4C    && ld_op==Op_LoadD) &&
          !(n->ideal_Opcode()==Op_Load4S    && ld_op==Op_LoadD) &&
          !(n->ideal_Opcode()==Op_Load8B    && ld_op==Op_LoadD) &&
          !(n->rule() == loadUB_rule)) {
        verify_oops_warning(n, n->ideal_Opcode(), ld_op);
      }
    } else if (st_op) {
      // a Store
      // inputs are (0:control, 1:memory, 2:address, 3:value)
      if (!(n->ideal_Opcode()==st_op)    && // Following are special cases
          !(n->ideal_Opcode()==Op_StoreCM && st_op==Op_StoreB) &&
          !(n->ideal_Opcode()==Op_StoreI && st_op==Op_StoreF) &&
          !(n->ideal_Opcode()==Op_StoreF && st_op==Op_StoreI) &&
          !(n->ideal_Opcode()==Op_StoreL && st_op==Op_StoreI) &&
          !(n->ideal_Opcode()==Op_Store2I && st_op==Op_StoreD) &&
          !(n->ideal_Opcode()==Op_Store4C && st_op==Op_StoreD) &&
          !(n->ideal_Opcode()==Op_Store8B && st_op==Op_StoreD) &&
          !(n->ideal_Opcode()==Op_StoreD && st_op==Op_StoreI && n->rule() == storeD0_rule)) {
        verify_oops_warning(n, n->ideal_Opcode(), st_op);
      }
    }

    if (src2_enc == R_G0_enc && n->rule() != loadUB_rule && n->ideal_Opcode() != Op_StoreCM ) {
      Node* addr = n->in(2);
      if (!(addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP)) {
        const TypeOopPtr* atype = addr->bottom_type()->isa_instptr();  // %%% oopptr?
        if (atype != NULL) {
          intptr_t offset = get_offset_from_base(n, atype, disp32);
          intptr_t offset_2 = get_offset_from_base_2(n, atype, disp32);
          if (offset != offset_2) {
            get_offset_from_base(n, atype, disp32);
            get_offset_from_base_2(n, atype, disp32);
          }
          assert(offset == offset_2, "different offsets");
          if (offset == disp32) {
            // we now know that src1 is a true oop pointer
            is_verified_oop_base = true;
            if (ld_op && src1_enc == dst_enc && ld_op != Op_LoadF && ld_op != Op_LoadD) {
              if( primary == Assembler::ldd_op3 ) {
                is_verified_oop_base = false; // Cannot 'ldd' into O7
              } else {
                tmp_enc = dst_enc;
                dst_enc = R_O7_enc; // Load into O7; preserve source oop
                assert(src1_enc != dst_enc, "");
              }
            }
          }
          if (st_op && (( offset == oopDesc::klass_offset_in_bytes())
                       || offset == oopDesc::mark_offset_in_bytes())) {
                      // loading the mark should not be allowed either, but
                      // we don't check this since it conflicts with InlineObjectHash
                      // usage of LoadINode to get the mark. We could keep the
                      // check if we create a new LoadMarkNode
            // but do not verify the object before its header is initialized
            ShouldNotReachHere();
          }
        }
      }
    }
  }
#endif

  uint instr;
  instr = (Assembler::ldst_op << 30)
        | (dst_enc        << 25)
        | (primary        << 19)
        | (src1_enc       << 14);

  uint index = src2_enc;
  int disp = disp32;

  if (src1_enc == R_SP_enc || src1_enc == R_FP_enc)
    disp += STACK_BIAS;

  // We should have a compiler bailout here rather than a guarantee.
  // Better yet would be some mechanism to handle variable-size matches correctly.
  guarantee(Assembler::is_simm13(disp), "Do not match large constant offsets" );

  if( disp == 0 ) {
    // use reg-reg form
    // bit 13 is already zero
    instr |= index;
  } else {
    // use reg-imm form
    instr |= 0x00002000;          // set bit 13 to one
    instr |= disp & 0x1FFF;
  }

  cbuf.insts()->emit_int32(instr);

#ifdef ASSERT
  {
    MacroAssembler _masm(&cbuf);
    if (is_verified_oop_base) {
      __ verify_oop(reg_to_register_object(src1_enc));
    }
    if (is_verified_oop_store) {
      __ verify_oop(reg_to_register_object(dst_enc));
    }
    if (tmp_enc != -1) {
      __ mov(O7, reg_to_register_object(tmp_enc));
    }
    if (is_verified_oop_load) {
      __ verify_oop(reg_to_register_object(dst_enc));
    }
  }
#endif
}

void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false, bool force_far_call = false) {
  // The method which records debug information at every safepoint
  // expects the call to be the first instruction in the snippet as
  // it creates a PcDesc structure which tracks the offset of a call
  // from the start of the codeBlob. This offset is computed as
  // code_end() - code_begin() of the code which has been emitted
  // so far.
  // In this particular case we have skirted around the problem by
  // putting the "mov" instruction in the delay slot but the problem
  // may bite us again at some other point and a cleaner/generic
  // solution using relocations would be needed.
  MacroAssembler _masm(&cbuf);
  __ set_inst_mark();

  // We flush the current window just so that there is a valid stack copy
  // the fact that the current window becomes active again instantly is
  // not a problem there is nothing live in it.

#ifdef ASSERT
  int startpos = __ offset();
#endif /* ASSERT */

#ifdef _LP64
  // Calls to the runtime or native may not be reachable from compiled code,
  // so we generate the far call sequence on 64 bit sparc.
  // This code sequence is relocatable to any address, even on LP64.
  if ( force_far_call ) {
    __ relocate(rtype);
    AddressLiteral dest(entry_point);
    __ jumpl_to(dest, O7, O7);
  }
  else
#endif
  {
     __ call((address)entry_point, rtype);
  }

  if (preserve_g2)   __ delayed()->mov(G2, L7);
  else __ delayed()->nop();

  if (preserve_g2)   __ mov(L7, G2);

#ifdef ASSERT
  if (preserve_g2 && (VerifyCompiledCode || VerifyOops)) {
#ifdef _LP64
    // Trash argument dump slots.
    __ set(0xb0b8ac0db0b8ac0d, G1);
    __ mov(G1, G5);
    __ stx(G1, SP, STACK_BIAS + 0x80);
    __ stx(G1, SP, STACK_BIAS + 0x88);
    __ stx(G1, SP, STACK_BIAS + 0x90);
    __ stx(G1, SP, STACK_BIAS + 0x98);
    __ stx(G1, SP, STACK_BIAS + 0xA0);
    __ stx(G1, SP, STACK_BIAS + 0xA8);
#else // _LP64
    // this is also a native call, so smash the first 7 stack locations,
    // and the various registers

    // Note:  [SP+0x40] is sp[callee_aggregate_return_pointer_sp_offset],
    // while [SP+0x44..0x58] are the argument dump slots.
    __ set((intptr_t)0xbaadf00d, G1);
    __ mov(G1, G5);
    __ sllx(G1, 32, G1);
    __ or3(G1, G5, G1);
    __ mov(G1, G5);
    __ stx(G1, SP, 0x40);
    __ stx(G1, SP, 0x48);
    __ stx(G1, SP, 0x50);
    __ stw(G1, SP, 0x58); // Do not trash [SP+0x5C] which is a usable spill slot
#endif // _LP64
  }
#endif /*ASSERT*/
}

//=============================================================================
// REQUIRED FUNCTIONALITY for encoding
void emit_lo(CodeBuffer &cbuf, int val) {  }
void emit_hi(CodeBuffer &cbuf, int val) {  }


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

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

  for (int i = 0; i < OptoPrologueNops; i++) {
    st->print_cr("NOP"); st->print("\t");
  }

  if( VerifyThread ) {
    st->print_cr("Verify_Thread"); st->print("\t");
  }

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

  // Calls to C2R adapters often do not accept exceptional returns.
  // We require that their callers must bang for them.  But be careful, because
  // some VM calls (such as call site linkage) can use several kilobytes of
  // stack.  But the stack safety zone should account for that.
  // See bugs 4446381, 4468289, 4497237.
  if (C->need_stack_bang(framesize)) {
    st->print_cr("! stack bang"); st->print("\t");
  }

  if (Assembler::is_simm13(-framesize)) {
    st->print   ("SAVE   R_SP,-%d,R_SP",framesize);
  } else {
    st->print_cr("SETHI  R_SP,hi%%(-%d),R_G3",framesize); st->print("\t");
    st->print_cr("ADD    R_G3,lo%%(-%d),R_G3",framesize); st->print("\t");
    st->print   ("SAVE   R_SP,R_G3,R_SP");
  }

}
#endif

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

  for (int i = 0; i < OptoPrologueNops; i++) {
    __ nop();
  }

  __ verify_thread();

  size_t framesize = C->frame_slots() << LogBytesPerInt;
  assert(framesize >= 16*wordSize, "must have room for reg. save area");
  assert(framesize%(2*wordSize) == 0, "must preserve 2*wordSize alignment");

  // Calls to C2R adapters often do not accept exceptional returns.
  // We require that their callers must bang for them.  But be careful, because
  // some VM calls (such as call site linkage) can use several kilobytes of
  // stack.  But the stack safety zone should account for that.
  // See bugs 4446381, 4468289, 4497237.
  if (C->need_stack_bang(framesize)) {
    __ generate_stack_overflow_check(framesize);
  }

  if (Assembler::is_simm13(-framesize)) {
    __ save(SP, -framesize, SP);
  } else {
    __ sethi(-framesize & ~0x3ff, G3);
    __ add(G3, -framesize & 0x3ff, G3);
    __ save(SP, G3, SP);
  }
  C->set_frame_complete( __ offset() );
}

uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
  return MachNode::size(ra_);
}

int MachPrologNode::reloc() const {
  return 10; // a large enough number
}

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

  if( do_polling() && ra_->C->is_method_compilation() ) {
    st->print("SETHI  #PollAddr,L0\t! Load Polling address\n\t");
#ifdef _LP64
    st->print("LDX    [L0],G0\t!Poll for Safepointing\n\t");
#else
    st->print("LDUW   [L0],G0\t!Poll for Safepointing\n\t");
#endif
  }

  if( do_polling() )
    st->print("RET\n\t");

  st->print("RESTORE");
}
#endif

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

  __ verify_thread();

  // If this does safepoint polling, then do it here
  if( do_polling() && ra_->C->is_method_compilation() ) {
    AddressLiteral polling_page(os::get_polling_page());
    __ sethi(polling_page, L0);
    __ relocate(relocInfo::poll_return_type);
    __ ld_ptr( L0, 0, G0 );
  }

  // If this is a return, then stuff the restore in the delay slot
  if( do_polling() ) {
    __ ret();
    __ delayed()->restore();
  } else {
    __ restore();
  }
}

uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
  return MachNode::size(ra_);
}

int MachEpilogNode::reloc() const {
  return 16; // a large enough number
}

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

int MachEpilogNode::safepoint_offset() const {
  assert( do_polling(), "no return for this epilog node");
  return MacroAssembler::size_of_sethi(os::get_polling_page());
}

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

// Figure out which register class each belongs in: rc_int, rc_float, rc_stack
enum RC { rc_bad, rc_int, rc_float, 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;
  assert(r->is_FloatRegister(), "must be");
  return rc_float;
}

static int impl_helper( const MachNode *mach, CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, bool is_load, int offset, int reg, int opcode, const char *op_str, int size, outputStream* st ) {
  if( cbuf ) {
    // Better yet would be some mechanism to handle variable-size matches correctly
    if (!Assembler::is_simm13(offset + STACK_BIAS)) {
      ra_->C->record_method_not_compilable("unable to handle large constant offsets");
    } else {
      emit_form3_mem_reg(*cbuf, mach, opcode, -1, R_SP_enc, offset, 0, Matcher::_regEncode[reg]);
    }
  }
#ifndef PRODUCT
  else if( !do_size ) {
    if( size != 0 ) st->print("\n\t");
    if( is_load ) st->print("%s   [R_SP + #%d],R_%s\t! spill",op_str,offset,OptoReg::regname(reg));
    else          st->print("%s   R_%s,[R_SP + #%d]\t! spill",op_str,OptoReg::regname(reg),offset);
  }
#endif
  return size+4;
}

static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int op1, int op2, const char *op_str, int size, outputStream* st ) {
  if( cbuf ) emit3( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst], op1, 0, op2, Matcher::_regEncode[src] );
#ifndef PRODUCT
  else if( !do_size ) {
    if( size != 0 ) st->print("\n\t");
    st->print("%s  R_%s,R_%s\t! spill",op_str,OptoReg::regname(src),OptoReg::regname(dst));
  }
#endif
  return size+4;
}

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

  // --------------------------------------
  // Check for mem-mem move.  Load into unused float registers and fall into
  // the float-store case.
  if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
    int offset = ra_->reg2offset(src_first);
    // Further check for aligned-adjacent pair, so we can use a double load
    if( (src_first&1)==0 && src_first+1 == src_second ) {
      src_second    = OptoReg::Name(R_F31_num);
      src_second_rc = rc_float;
      size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::lddf_op3,"LDDF",size, st);
    } else {
      size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::ldf_op3 ,"LDF ",size, st);
    }
    src_first    = OptoReg::Name(R_F30_num);
    src_first_rc = rc_float;
  }

  if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) {
    int offset = ra_->reg2offset(src_second);
    size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F31_num,Assembler::ldf_op3,"LDF ",size, st);
    src_second    = OptoReg::Name(R_F31_num);
    src_second_rc = rc_float;
  }

  // --------------------------------------
  // Check for float->int copy; requires a trip through memory
  if( src_first_rc == rc_float && dst_first_rc == rc_int ) {
    int offset = frame::register_save_words*wordSize;
    if( cbuf ) {
      emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::sub_op3, R_SP_enc, 16 );
      impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
      impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
      emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::add_op3, R_SP_enc, 16 );
    }
#ifndef PRODUCT
    else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print(  "SUB    R_SP,16,R_SP\n");
      impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
      impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
      st->print("\tADD    R_SP,16,R_SP\n");
    }
#endif
    size += 16;
  }

  // --------------------------------------
  // In the 32-bit 1-reg-longs build ONLY, I see mis-aligned long destinations.
  // In such cases, I have to do the big-endian swap.  For aligned targets, the
  // hardware does the flop for me.  Doubles are always aligned, so no problem
  // there.  Misaligned sources only come from native-long-returns (handled
  // special below).
#ifndef _LP64
  if( src_first_rc == rc_int &&     // source is already big-endian
      src_second_rc != rc_bad &&    // 64-bit move
      ((dst_first&1)!=0 || dst_second != dst_first+1) ) { // misaligned dst
    assert( (src_first&1)==0 && src_second == src_first+1, "source must be aligned" );
    // Do the big-endian flop.
    OptoReg::Name tmp    = dst_first   ; dst_first    = dst_second   ; dst_second    = tmp   ;
    enum RC       tmp_rc = dst_first_rc; dst_first_rc = dst_second_rc; dst_second_rc = tmp_rc;
  }
#endif

  // --------------------------------------
  // Check for integer reg-reg copy
  if( src_first_rc == rc_int && dst_first_rc == rc_int ) {
#ifndef _LP64
    if( src_first == R_O0_num && src_second == R_O1_num ) {  // Check for the evil O0/O1 native long-return case
      // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
      //       as stored in memory.  On a big-endian machine like SPARC, this means that the _second
      //       operand contains the least significant word of the 64-bit value and vice versa.
      OptoReg::Name tmp = OptoReg::Name(R_O7_num);
      assert( (dst_first&1)==0 && dst_second == dst_first+1, "return a native O0/O1 long to an aligned-adjacent 64-bit reg" );
      // Shift O0 left in-place, zero-extend O1, then OR them into the dst
      if( cbuf ) {
        emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tmp], Assembler::sllx_op3, Matcher::_regEncode[src_first], 0x1020 );
        emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[src_second], Assembler::srl_op3, Matcher::_regEncode[src_second], 0x0000 );
        emit3       ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler:: or_op3, Matcher::_regEncode[tmp], 0, Matcher::_regEncode[src_second] );
#ifndef PRODUCT
      } else if( !do_size ) {
        if( size != 0 ) st->print("\n\t");
        st->print("SLLX   R_%s,32,R_%s\t! Move O0-first to O7-high\n\t", OptoReg::regname(src_first), OptoReg::regname(tmp));
        st->print("SRL    R_%s, 0,R_%s\t! Zero-extend O1\n\t", OptoReg::regname(src_second), OptoReg::regname(src_second));
        st->print("OR     R_%s,R_%s,R_%s\t! spill",OptoReg::regname(tmp), OptoReg::regname(src_second), OptoReg::regname(dst_first));
#endif
      }
      return size+12;
    }
    else if( dst_first == R_I0_num && dst_second == R_I1_num ) {
      // returning a long value in I0/I1
      // a SpillCopy must be able to target a return instruction's reg_class
      // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
      //       as stored in memory.  On a big-endian machine like SPARC, this means that the _second
      //       operand contains the least significant word of the 64-bit value and vice versa.
      OptoReg::Name tdest = dst_first;

      if (src_first == dst_first) {
        tdest = OptoReg::Name(R_O7_num);
        size += 4;
      }

      if( cbuf ) {
        assert( (src_first&1) == 0 && (src_first+1) == src_second, "return value was in an aligned-adjacent 64-bit reg");
        // Shift value in upper 32-bits of src to lower 32-bits of I0; move lower 32-bits to I1
        // ShrL_reg_imm6
        emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tdest], Assembler::srlx_op3, Matcher::_regEncode[src_second], 32 | 0x1000 );
        // ShrR_reg_imm6  src, 0, dst
        emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srl_op3, Matcher::_regEncode[src_first], 0x0000 );
        if (tdest != dst_first) {
          emit3     ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler::or_op3, 0/*G0*/, 0/*op2*/, Matcher::_regEncode[tdest] );
        }
      }
#ifndef PRODUCT
      else if( !do_size ) {
        if( size != 0 ) st->print("\n\t");  // %%%%% !!!!!
        st->print("SRLX   R_%s,32,R_%s\t! Extract MSW\n\t",OptoReg::regname(src_second),OptoReg::regname(tdest));
        st->print("SRL    R_%s, 0,R_%s\t! Extract LSW\n\t",OptoReg::regname(src_first),OptoReg::regname(dst_second));
        if (tdest != dst_first) {
          st->print("MOV    R_%s,R_%s\t! spill\n\t", OptoReg::regname(tdest), OptoReg::regname(dst_first));
        }
      }
#endif // PRODUCT
      return size+8;
    }
#endif // !_LP64
    // Else normal reg-reg copy
    assert( src_second != dst_first, "smashed second before evacuating it" );
    size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::or_op3,0,"MOV  ",size, st);
    assert( (src_first&1) == 0 && (dst_first&1) == 0, "never move second-halves of int registers" );
    // This moves an aligned adjacent pair.
    // See if we are done.
    if( src_first+1 == src_second && dst_first+1 == dst_second )
      return size;
  }

  // Check for integer store
  if( src_first_rc == rc_int && dst_first_rc == rc_stack ) {
    int offset = ra_->reg2offset(dst_first);
    // Further check for aligned-adjacent pair, so we can use a double store
    if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
      return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stx_op3,"STX ",size, st);
    size  =  impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stw_op3,"STW ",size, st);
  }

  // Check for integer load
  if( dst_first_rc == rc_int && src_first_rc == rc_stack ) {
    int offset = ra_->reg2offset(src_first);
    // Further check for aligned-adjacent pair, so we can use a double load
    if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
      return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldx_op3 ,"LDX ",size, st);
    size  =  impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
  }

  // Check for float reg-reg copy
  if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
    // Further check for aligned-adjacent pair, so we can use a double move
    if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
      return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovd_opf,"FMOVD",size, st);
    size  =  impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovs_opf,"FMOVS",size, st);
  }

  // Check for float store
  if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
    int offset = ra_->reg2offset(dst_first);
    // Further check for aligned-adjacent pair, so we can use a double store
    if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
      return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stdf_op3,"STDF",size, st);
    size  =  impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
  }

  // Check for float load
  if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
    int offset = ra_->reg2offset(src_first);
    // Further check for aligned-adjacent pair, so we can use a double load
    if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
      return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lddf_op3,"LDDF",size, st);
    size  =  impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldf_op3 ,"LDF ",size, st);
  }

  // --------------------------------------------------------------------
  // Check for hi bits still needing moving.  Only happens for misaligned
  // arguments to native calls.
  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" );

#ifndef _LP64
  // In the LP64 build, all registers can be moved as aligned/adjacent
  // pairs, so there's never any need to move the high bits separately.
  // The 32-bit builds have to deal with the 32-bit ABI which can force
  // all sorts of silly alignment problems.

  // Check for integer reg-reg copy.  Hi bits are stuck up in the top
  // 32-bits of a 64-bit register, but are needed in low bits of another
  // register (else it's a hi-bits-to-hi-bits copy which should have
  // happened already as part of a 64-bit move)
  if( src_second_rc == rc_int && dst_second_rc == rc_int ) {
    assert( (src_second&1)==1, "its the evil O0/O1 native return case" );
    assert( (dst_second&1)==0, "should have moved with 1 64-bit move" );
    // Shift src_second down to dst_second's low bits.
    if( cbuf ) {
      emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print("SRLX   R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(dst_second));
#endif
    }
    return size+4;
  }

  // Check for high word integer store.  Must down-shift the hi bits
  // into a temp register, then fall into the case of storing int bits.
  if( src_second_rc == rc_int && dst_second_rc == rc_stack && (src_second&1)==1 ) {
    // Shift src_second down to dst_second's low bits.
    if( cbuf ) {
      emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[R_O7_num], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
#ifndef PRODUCT
    } else if( !do_size ) {
      if( size != 0 ) st->print("\n\t");
      st->print("SRLX   R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(R_O7_num));
#endif
    }
    size+=4;
    src_second = OptoReg::Name(R_O7_num); // Not R_O7H_num!
  }

  // Check for high word integer load
  if( dst_second_rc == rc_int && src_second_rc == rc_stack )
    return impl_helper(this,cbuf,ra_,do_size,true ,ra_->reg2offset(src_second),dst_second,Assembler::lduw_op3,"LDUW",size, st);

  // Check for high word integer store
  if( src_second_rc == rc_int && dst_second_rc == rc_stack )
    return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stw_op3 ,"STW ",size, st);

  // Check for high word float store
  if( src_second_rc == rc_float && dst_second_rc == rc_stack )
    return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stf_op3 ,"STF ",size, st);

#endif // !_LP64

  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 MachNopNode::format( PhaseRegAlloc *, outputStream *st ) const {
  st->print("NOP \t# %d bytes pad for loops and calls", 4 * _count);
}
#endif

void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc * ) const {
  MacroAssembler _masm(&cbuf);
  for(int i = 0; i < _count; i += 1) {
    __ nop();
  }
}

uint MachNopNode::size(PhaseRegAlloc *ra_) const {
  return 4 * _count;
}


//=============================================================================
#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    [R_SP+#%d+BIAS],%s",offset,Matcher::regName[reg]);
}
#endif

void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  MacroAssembler _masm(&cbuf);
  int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()) + STACK_BIAS;
  int reg = ra_->get_encode(this);

  if (Assembler::is_simm13(offset)) {
     __ add(SP, offset, reg_to_register_object(reg));
  } else {
     __ set(offset, O7);
     __ add(SP, O7, reg_to_register_object(reg));
  }
}

uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
  // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_)
  assert(ra_ == ra_->C->regalloc(), "sanity");
  return ra_->C->scratch_emit_size(this);
}

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

// emit call stub, compiled java to interpretor
void emit_java_to_interp(CodeBuffer &cbuf ) {

  // Stub is fixed up when the corresponding call is converted from calling
  // compiled code to calling interpreted code.
  // set (empty), G5
  // jmp -1

  address mark = cbuf.insts_mark();  // get mark within main instrs section

  MacroAssembler _masm(&cbuf);

  address base =
  __ start_a_stub(Compile::MAX_stubs_size);
  if (base == NULL)  return;  // CodeBuffer::expand failed

  // static stub relocation stores the instruction address of the call
  __ relocate(static_stub_Relocation::spec(mark));

  __ set_oop(NULL, reg_to_register_object(Matcher::inline_cache_reg_encode()));

  __ set_inst_mark();
  AddressLiteral addrlit(-1);
  __ JUMP(addrlit, G3, 0);

  __ delayed()->nop();

  // Update current stubs pointer and restore code_end.
  __ end_a_stub();
}

// size of call stub, compiled java to interpretor
uint size_java_to_interp() {
  // This doesn't need to be accurate but it must be larger or equal to
  // the real size of the stub.
  return (NativeMovConstReg::instruction_size +  // sethi/setlo;
          NativeJump::instruction_size + // sethi; jmp; nop
          (TraceJumps ? 20 * BytesPerInstWord : 0) );
}
// relocation entries for call stub, compiled java to interpretor
uint reloc_java_to_interp() {
  return 10;  // 4 in emit_java_to_interp + 1 in Java_Static_Call
}


//=============================================================================
#ifndef PRODUCT
void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
  st->print_cr("\nUEP:");
#ifdef    _LP64
  if (UseCompressedOops) {
    assert(Universe::heap() != NULL, "java heap should be initialized");
    st->print_cr("\tLDUW   [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass");
    st->print_cr("\tSLL    R_G5,3,R_G5");
    if (Universe::narrow_oop_base() != NULL)
      st->print_cr("\tADD    R_G5,R_G6_heap_base,R_G5");
  } else {
    st->print_cr("\tLDX    [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
  }
  st->print_cr("\tCMP    R_G5,R_G3" );
  st->print   ("\tTne    xcc,R_G0+ST_RESERVED_FOR_USER_0+2");
#else  // _LP64
  st->print_cr("\tLDUW   [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
  st->print_cr("\tCMP    R_G5,R_G3" );
  st->print   ("\tTne    icc,R_G0+ST_RESERVED_FOR_USER_0+2");
#endif // _LP64
}
#endif

void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  MacroAssembler _masm(&cbuf);
  Label L;
  Register G5_ic_reg  = reg_to_register_object(Matcher::inline_cache_reg_encode());
  Register temp_reg   = G3;
  assert( G5_ic_reg != temp_reg, "conflicting registers" );

  // Load klass from receiver
  __ load_klass(O0, temp_reg);
  // Compare against expected klass
  __ cmp(temp_reg, G5_ic_reg);
  // Branch to miss code, checks xcc or icc depending
  __ trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2);
}

uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
  return MachNode::size(ra_);
}


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

uint size_exception_handler() {
  if (TraceJumps) {
    return (400); // just a guess
  }
  return ( NativeJump::instruction_size ); // sethi;jmp;nop
}

uint size_deopt_handler() {
  if (TraceJumps) {
    return (400); // just a guess
  }
  return ( 4+  NativeJump::instruction_size ); // save;sethi;jmp;restore
}

// Emit exception handler code.
int emit_exception_handler(CodeBuffer& cbuf) {
  Register temp_reg = G3;
  AddressLiteral exception_blob(OptoRuntime::exception_blob()->entry_point());
  MacroAssembler _masm(&cbuf);

  address base =
  __ start_a_stub(size_exception_handler());
  if (base == NULL)  return 0;  // CodeBuffer::expand failed

  int offset = __ offset();

  __ JUMP(exception_blob, temp_reg, 0); // sethi;jmp
  __ delayed()->nop();

  assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");

  __ end_a_stub();

  return offset;
}

int emit_deopt_handler(CodeBuffer& cbuf) {
  // Can't use any of the current frame's registers as we may have deopted
  // at a poll and everything (including G3) can be live.
  Register temp_reg = L0;
  AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
  MacroAssembler _masm(&cbuf);

  address base =
  __ start_a_stub(size_deopt_handler());
  if (base == NULL)  return 0;  // CodeBuffer::expand failed

  int offset = __ offset();
  __ save_frame(0);
  __ JUMP(deopt_blob, temp_reg, 0); // sethi;jmp
  __ delayed()->restore();

  assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");

  __ end_a_stub();
  return offset;

}

// Given a register encoding, produce a Integer Register object
static Register reg_to_register_object(int register_encoding) {
  assert(L5->encoding() == R_L5_enc && G1->encoding() == R_G1_enc, "right coding");
  return as_Register(register_encoding);
}

// Given a register encoding, produce a single-precision Float Register object
static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding) {
  assert(F5->encoding(FloatRegisterImpl::S) == R_F5_enc && F12->encoding(FloatRegisterImpl::S) == R_F12_enc, "right coding");
  return as_SingleFloatRegister(register_encoding);
}

// Given a register encoding, produce a double-precision Float Register object
static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding) {
  assert(F4->encoding(FloatRegisterImpl::D) == R_F4_enc, "right coding");
  assert(F32->encoding(FloatRegisterImpl::D) == R_D32_enc, "right coding");
  return as_DoubleFloatRegister(register_encoding);
}

const bool Matcher::match_rule_supported(int opcode) {
  if (!has_match_rule(opcode))
    return false;

  switch (opcode) {
  case Op_CountLeadingZerosI:
  case Op_CountLeadingZerosL:
  case Op_CountTrailingZerosI:
  case Op_CountTrailingZerosL:
    if (!UsePopCountInstruction)
      return false;
    break;
  }

  return true;  // Per default match rules are supported.
}

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

#ifdef ASSERT
address last_rethrow = NULL;  // debugging aid for Rethrow encoding
#endif

// Vector width in bytes
const uint Matcher::vector_width_in_bytes(void) {
  return 8;
}

// Vector ideal reg
const uint Matcher::vector_ideal_reg(void) {
  return Op_RegD;
}

// USII supports fxtof through the whole range of number, USIII doesn't
const bool Matcher::convL2FSupported(void) {
  return VM_Version::has_fast_fxtof();
}

// 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 offset) {
  return false;
}

const bool Matcher::isSimpleConstant64(jlong value) {
  // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
  // Depends on optimizations in MacroAssembler::setx.
  int hi = (int)(value >> 32);
  int lo = (int)(value & ~0);
  return (hi == 0) || (hi == -1) || (lo == 0);
}

// No scaling for the parameter the ClearArray node.
const bool Matcher::init_array_count_is_in_bytes = true;

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

// 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 = false;

bool Matcher::narrow_oop_use_complex_address() {
  NOT_LP64(ShouldNotCallThis());
  assert(UseCompressedOops, "only for compressed oops code");
  return false;
}

// 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 = false;

// 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.
#ifdef _LP64
const bool Matcher::misaligned_doubles_ok = true;
#else
const bool Matcher::misaligned_doubles_ok = false;
#endif

// No-op on SPARC.
void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
}

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

// Are floats conerted to double when stored to stack during deoptimization?
// Sparc does not handle callee-save floats.
bool Matcher::float_in_double() { return false; }

// Do ints take an entire long register or just half?
// Note that we if-def off of _LP64.
// The relevant question is how the int is callee-saved.  In _LP64
// the whole long is written but de-opt'ing will have to extract
// the relevant 32 bits, in not-_LP64 only the low 32 bits is written.
#ifdef _LP64
const bool Matcher::int_in_long = true;
#else
const bool Matcher::int_in_long = false;
#endif

// 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 ) {
  // Standard sparc 6 args in registers
  if( reg == R_I0_num ||
      reg == R_I1_num ||
      reg == R_I2_num ||
      reg == R_I3_num ||
      reg == R_I4_num ||
      reg == R_I5_num ) return true;
#ifdef _LP64
  // 64-bit builds can pass 64-bit pointers and longs in
  // the high I registers
  if( reg == R_I0H_num ||
      reg == R_I1H_num ||
      reg == R_I2H_num ||
      reg == R_I3H_num ||
      reg == R_I4H_num ||
      reg == R_I5H_num ) return true;

  if ((UseCompressedOops) && (reg == R_G6_num || reg == R_G6H_num)) {
    return true;
  }

#else
  // 32-bit builds with longs-in-one-entry pass longs in G1 & G4.
  // Longs cannot be passed in O regs, because O regs become I regs
  // after a 'save' and I regs get their high bits chopped off on
  // interrupt.
  if( reg == R_G1H_num || reg == R_G1_num ) return true;
  if( reg == R_G4H_num || reg == R_G4_num ) return true;
#endif
  // A few float args in registers
  if( reg >= R_F0_num && reg <= R_F7_num ) return true;

  return false;
}

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

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

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

// 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 L7_REGP_mask;
}

%}


// The intptr_t operand types, defined by textual substitution.
// (Cf. opto/type.hpp.  This lets us avoid many, many other ifdefs.)
#ifdef _LP64
#define immX      immL
#define immX13    immL13
#define immX13m7  immL13m7
#define iRegX     iRegL
#define g1RegX    g1RegL
#else
#define immX      immI
#define immX13    immI13
#define immX13m7  immI13m7
#define iRegX     iRegI
#define g1RegX    g1RegI
#endif

//----------ENCODING BLOCK-----------------------------------------------------
// This block specifies the encoding classes used by the compiler to output
// byte streams.  Encoding classes are parameterized macros used 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.  Again, a function
// is available to check if the constant displacement is an oop. They use the
// ins_encode keyword to specify their encoding classes (which must be
// a sequence of enc_class names, and their parameters, 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 %{
  enc_class enc_untested %{
#ifdef ASSERT
    MacroAssembler _masm(&cbuf);
    __ untested("encoding");
#endif
  %}

  enc_class form3_mem_reg( memory mem, iRegI dst ) %{
    emit_form3_mem_reg(cbuf, this, $primary, $tertiary,
                       $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
  %}

  enc_class simple_form3_mem_reg( memory mem, iRegI dst ) %{
    emit_form3_mem_reg(cbuf, this, $primary, -1,
                       $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
  %}

  enc_class form3_mem_prefetch_read( memory mem ) %{
    emit_form3_mem_reg(cbuf, this, $primary, -1,
                       $mem$$base, $mem$$disp, $mem$$index, 0/*prefetch function many-reads*/);
  %}

  enc_class form3_mem_prefetch_write( memory mem ) %{
    emit_form3_mem_reg(cbuf, this, $primary, -1,
                       $mem$$base, $mem$$disp, $mem$$index, 2/*prefetch function many-writes*/);
  %}

  enc_class form3_mem_reg_long_unaligned_marshal( memory mem, iRegL reg ) %{
    assert( Assembler::is_simm13($mem$$disp  ), "need disp and disp+4" );
    assert( Assembler::is_simm13($mem$$disp+4), "need disp and disp+4" );
    guarantee($mem$$index == R_G0_enc, "double index?");
    emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, R_O7_enc );
    emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp,   R_G0_enc, $reg$$reg );
    emit3_simm13( cbuf, Assembler::arith_op, $reg$$reg, Assembler::sllx_op3, $reg$$reg, 0x1020 );
    emit3( cbuf, Assembler::arith_op, $reg$$reg, Assembler::or_op3, $reg$$reg, 0, R_O7_enc );
  %}

  enc_class form3_mem_reg_double_unaligned( memory mem, RegD_low reg ) %{
    assert( Assembler::is_simm13($mem$$disp  ), "need disp and disp+4" );
    assert( Assembler::is_simm13($mem$$disp+4), "need disp and disp+4" );
    guarantee($mem$$index == R_G0_enc, "double index?");
    // Load long with 2 instructions
    emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp,   R_G0_enc, $reg$$reg+0 );
    emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, $reg$$reg+1 );
  %}

  //%%% form3_mem_plus_4_reg is a hack--get rid of it
  enc_class form3_mem_plus_4_reg( memory mem, iRegI dst ) %{
    guarantee($mem$$disp, "cannot offset a reg-reg operand by 4");
    emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp + 4, $mem$$index, $dst$$reg);
  %}

  enc_class form3_g0_rs2_rd_move( iRegI rs2, iRegI rd ) %{
    // Encode a reg-reg copy.  If it is useless, then empty encoding.
    if( $rs2$$reg != $rd$$reg )
      emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, $rs2$$reg );
  %}

  // Target lo half of long
  enc_class form3_g0_rs2_rd_move_lo( iRegI rs2, iRegL rd ) %{
    // Encode a reg-reg copy.  If it is useless, then empty encoding.
    if( $rs2$$reg != LONG_LO_REG($rd$$reg) )
      emit3( cbuf, Assembler::arith_op, LONG_LO_REG($rd$$reg), Assembler::or_op3, 0, 0, $rs2$$reg );
  %}

  // Source lo half of long
  enc_class form3_g0_rs2_rd_move_lo2( iRegL rs2, iRegI rd ) %{
    // Encode a reg-reg copy.  If it is useless, then empty encoding.
    if( LONG_LO_REG($rs2$$reg) != $rd$$reg )
      emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_LO_REG($rs2$$reg) );
  %}

  // Target hi half of long
  enc_class form3_rs1_rd_copysign_hi( iRegI rs1, iRegL rd ) %{
    emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 31 );
  %}

  // Source lo half of long, and leave it sign extended.
  enc_class form3_rs1_rd_signextend_lo1( iRegL rs1, iRegI rd ) %{
    // Sign extend low half
    emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 0, 0 );
  %}

  // Source hi half of long, and leave it sign extended.
  enc_class form3_rs1_rd_copy_hi1( iRegL rs1, iRegI rd ) %{
    // Shift high half to low half
    emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::srlx_op3, $rs1$$reg, 32 );
  %}

  // Source hi half of long
  enc_class form3_g0_rs2_rd_move_hi2( iRegL rs2, iRegI rd ) %{
    // Encode a reg-reg copy.  If it is useless, then empty encoding.
    if( LONG_HI_REG($rs2$$reg) != $rd$$reg )
      emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_HI_REG($rs2$$reg) );
  %}

  enc_class form3_rs1_rs2_rd( iRegI rs1, iRegI rs2, iRegI rd ) %{
    emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0, $rs2$$reg );
  %}

  enc_class enc_to_bool( iRegI src, iRegI dst ) %{
    emit3       ( cbuf, Assembler::arith_op,         0, Assembler::subcc_op3, 0, 0, $src$$reg );
    emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::addc_op3 , 0, 0 );
  %}

  enc_class enc_ltmask( iRegI p, iRegI q, iRegI dst ) %{
    emit3       ( cbuf, Assembler::arith_op,         0, Assembler::subcc_op3, $p$$reg, 0, $q$$reg );
    // clear if nothing else is happening
    emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0,  0 );
    // blt,a,pn done
    emit2_19    ( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less, Assembler::bp_op2, Assembler::icc, 0/*predict not taken*/, 2 );
    // mov dst,-1 in delay slot
    emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
  %}

  enc_class form3_rs1_imm5_rd( iRegI rs1, immU5 imm5, iRegI rd ) %{
    emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $imm5$$constant & 0x1F );
  %}

  enc_class form3_sd_rs1_imm6_rd( iRegL rs1, immU6 imm6, iRegL rd ) %{
    emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, ($imm6$$constant & 0x3F) | 0x1000 );
  %}

  enc_class form3_sd_rs1_rs2_rd( iRegL rs1, iRegI rs2, iRegL rd ) %{
    emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0x80, $rs2$$reg );
  %}

  enc_class form3_rs1_simm13_rd( iRegI rs1, immI13 simm13, iRegI rd ) %{
    emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $simm13$$constant );
  %}

  enc_class move_return_pc_to_o1() %{
    emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::add_op3, R_O7_enc, frame::pc_return_offset );
  %}

#ifdef _LP64
  /* %%% merge with enc_to_bool */
  enc_class enc_convP2B( iRegI dst, iRegP src ) %{
    MacroAssembler _masm(&cbuf);

    Register   src_reg = reg_to_register_object($src$$reg);
    Register   dst_reg = reg_to_register_object($dst$$reg);
    __ movr(Assembler::rc_nz, src_reg, 1, dst_reg);
  %}
#endif

  enc_class enc_cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp ) %{
    // (Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)))
    MacroAssembler _masm(&cbuf);

    Register   p_reg = reg_to_register_object($p$$reg);
    Register   q_reg = reg_to_register_object($q$$reg);
    Register   y_reg = reg_to_register_object($y$$reg);
    Register tmp_reg = reg_to_register_object($tmp$$reg);

    __ subcc( p_reg, q_reg,   p_reg );
    __ add  ( p_reg, y_reg, tmp_reg );
    __ movcc( Assembler::less, false, Assembler::icc, tmp_reg, p_reg );
  %}

  enc_class form_d2i_helper(regD src, regF dst) %{
    // fcmp %fcc0,$src,$src
    emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
    // branch %fcc0 not-nan, predict taken
    emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
    // fdtoi $src,$dst
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fdtoi_opf, $src$$reg );
    // fitos $dst,$dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fitos_opf, $dst$$reg );
    // clear $dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
    // carry on here...
  %}

  enc_class form_d2l_helper(regD src, regD dst) %{
    // fcmp %fcc0,$src,$src  check for NAN
    emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
    // branch %fcc0 not-nan, predict taken
    emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
    // fdtox $src,$dst   convert in delay slot
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fdtox_opf, $src$$reg );
    // fxtod $dst,$dst  (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fxtod_opf, $dst$$reg );
    // clear $dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
    // carry on here...
  %}

  enc_class form_f2i_helper(regF src, regF dst) %{
    // fcmps %fcc0,$src,$src
    emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
    // branch %fcc0 not-nan, predict taken
    emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
    // fstoi $src,$dst
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fstoi_opf, $src$$reg );
    // fitos $dst,$dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fitos_opf, $dst$$reg );
    // clear $dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
    // carry on here...
  %}

  enc_class form_f2l_helper(regF src, regD dst) %{
    // fcmps %fcc0,$src,$src
    emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
    // branch %fcc0 not-nan, predict taken
    emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
    // fstox $src,$dst
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fstox_opf, $src$$reg );
    // fxtod $dst,$dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fxtod_opf, $dst$$reg );
    // clear $dst (if nan)
    emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
    // carry on here...
  %}

  enc_class form3_opf_rs2F_rdF(regF rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
  enc_class form3_opf_rs2F_rdD(regF rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
  enc_class form3_opf_rs2D_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
  enc_class form3_opf_rs2D_rdD(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}

  enc_class form3_opf_rs2D_lo_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg+1); %}

  enc_class form3_opf_rs2D_hi_rdD_hi(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
  enc_class form3_opf_rs2D_lo_rdD_lo(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg+1,$primary,0,$tertiary,$rs2$$reg+1); %}

  enc_class form3_opf_rs1F_rs2F_rdF( regF rs1, regF rs2, regF rd ) %{
    emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
  %}

  enc_class form3_opf_rs1D_rs2D_rdD( regD rs1, regD rs2, regD rd ) %{
    emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
  %}

  enc_class form3_opf_rs1F_rs2F_fcc( regF rs1, regF rs2, flagsRegF fcc ) %{
    emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
  %}

  enc_class form3_opf_rs1D_rs2D_fcc( regD rs1, regD rs2, flagsRegF fcc ) %{
    emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
  %}

  enc_class form3_convI2F(regF rs2, regF rd) %{
    emit3(cbuf,Assembler::arith_op,$rd$$reg,Assembler::fpop1_op3,0,$secondary,$rs2$$reg);
  %}

  // Encloding class for traceable jumps
  enc_class form_jmpl(g3RegP dest) %{
    emit_jmpl(cbuf, $dest$$reg);
  %}

  enc_class form_jmpl_set_exception_pc(g1RegP dest) %{
    emit_jmpl_set_exception_pc(cbuf, $dest$$reg);
  %}

  enc_class form2_nop() %{
    emit_nop(cbuf);
  %}

  enc_class form2_illtrap() %{
    emit_illtrap(cbuf);
  %}


  // Compare longs and convert into -1, 0, 1.
  enc_class cmpl_flag( iRegL src1, iRegL src2, iRegI dst ) %{
    // CMP $src1,$src2
    emit3( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $src1$$reg, 0, $src2$$reg );
    // blt,a,pn done
    emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less   , Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 5 );
    // mov dst,-1 in delay slot
    emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
    // bgt,a,pn done
    emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::greater, Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 3 );
    // mov dst,1 in delay slot
    emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0,  1 );
    // CLR    $dst
    emit3( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3 , 0, 0, 0 );
  %}

  enc_class enc_PartialSubtypeCheck() %{
    MacroAssembler _masm(&cbuf);
    __ call(StubRoutines::Sparc::partial_subtype_check(), relocInfo::runtime_call_type);
    __ delayed()->nop();
  %}

  enc_class enc_bp( Label labl, cmpOp cmp, flagsReg cc ) %{
    MacroAssembler _masm(&cbuf);
    Label &L = *($labl$$label);
    Assembler::Predict predict_taken =
      cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn;

    __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, L);
    __ delayed()->nop();
  %}

  enc_class enc_bpl( Label labl, cmpOp cmp, flagsRegL cc ) %{
    MacroAssembler _masm(&cbuf);
    Label &L = *($labl$$label);
    Assembler::Predict predict_taken =
      cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn;

    __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, L);
    __ delayed()->nop();
  %}

  enc_class enc_bpx( Label labl, cmpOp cmp, flagsRegP cc ) %{
    MacroAssembler _masm(&cbuf);
    Label &L = *($labl$$label);
    Assembler::Predict predict_taken =
      cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn;

    __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, L);
    __ delayed()->nop();
  %}

  enc_class enc_fbp( Label labl, cmpOpF cmp, flagsRegF cc ) %{
    MacroAssembler _masm(&cbuf);
    Label &L = *($labl$$label);
    Assembler::Predict predict_taken =
      cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn;

    __ fbp( (Assembler::Condition)($cmp$$cmpcode), false, (Assembler::CC)($cc$$reg), predict_taken, L);
    __ delayed()->nop();
  %}

  enc_class jump_enc( iRegX switch_val, o7RegI table) %{
    MacroAssembler _masm(&cbuf);

    Register switch_reg       = as_Register($switch_val$$reg);
    Register table_reg        = O7;

    address table_base = __ address_table_constant(_index2label);
    RelocationHolder rspec = internal_word_Relocation::spec(table_base);

    // Move table address into a register.
    __ set(table_base, table_reg, rspec);

    // Jump to base address + switch value
    __ ld_ptr(table_reg, switch_reg, table_reg);
    __ jmp(table_reg, G0);
    __ delayed()->nop();

  %}

  enc_class enc_ba( Label labl ) %{
    MacroAssembler _masm(&cbuf);
    Label &L = *($labl$$label);
    __ ba(false, L);
    __ delayed()->nop();
  %}

  enc_class enc_bpr( Label labl, cmpOp_reg cmp, iRegI op1 ) %{
    MacroAssembler _masm(&cbuf);
    Label &L = *$labl$$label;
    Assembler::Predict predict_taken =
      cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn;

    __ bpr( (Assembler::RCondition)($cmp$$cmpcode), false, predict_taken, as_Register($op1$$reg), L);
    __ delayed()->nop();
  %}

  enc_class enc_cmov_reg( cmpOp cmp, iRegI dst, iRegI src, immI pcc) %{
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::movcc_op3 << 19) |
             (1 << 18) |                    // cc2 bit for 'icc'
             ($cmp$$cmpcode << 14) |
             (0 << 13) |                    // select register move
             ($pcc$$constant << 11) |       // cc1, cc0 bits for 'icc' or 'xcc'
             ($src$$reg << 0);
    cbuf.insts()->emit_int32(op);
  %}

  enc_class enc_cmov_imm( cmpOp cmp, iRegI dst, immI11 src, immI pcc ) %{
    int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::movcc_op3 << 19) |
             (1 << 18) |                    // cc2 bit for 'icc'
             ($cmp$$cmpcode << 14) |
             (1 << 13) |                    // select immediate move
             ($pcc$$constant << 11) |       // cc1, cc0 bits for 'icc'
             (simm11 << 0);
    cbuf.insts()->emit_int32(op);
  %}

  enc_class enc_cmov_reg_f( cmpOpF cmp, iRegI dst, iRegI src, flagsRegF fcc ) %{
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::movcc_op3 << 19) |
             (0 << 18) |                    // cc2 bit for 'fccX'
             ($cmp$$cmpcode << 14) |
             (0 << 13) |                    // select register move
             ($fcc$$reg << 11) |            // cc1, cc0 bits for fcc0-fcc3
             ($src$$reg << 0);
    cbuf.insts()->emit_int32(op);
  %}

  enc_class enc_cmov_imm_f( cmpOp cmp, iRegI dst, immI11 src, flagsRegF fcc ) %{
    int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::movcc_op3 << 19) |
             (0 << 18) |                    // cc2 bit for 'fccX'
             ($cmp$$cmpcode << 14) |
             (1 << 13) |                    // select immediate move
             ($fcc$$reg << 11) |            // cc1, cc0 bits for fcc0-fcc3
             (simm11 << 0);
    cbuf.insts()->emit_int32(op);
  %}

  enc_class enc_cmovf_reg( cmpOp cmp, regD dst, regD src, immI pcc ) %{
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::fpop2_op3 << 19) |
             (0 << 18) |
             ($cmp$$cmpcode << 14) |
             (1 << 13) |                    // select register move
             ($pcc$$constant << 11) |       // cc1-cc0 bits for 'icc' or 'xcc'
             ($primary << 5) |              // select single, double or quad
             ($src$$reg << 0);
    cbuf.insts()->emit_int32(op);
  %}

  enc_class enc_cmovff_reg( cmpOpF cmp, flagsRegF fcc, regD dst, regD src ) %{
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::fpop2_op3 << 19) |
             (0 << 18) |
             ($cmp$$cmpcode << 14) |
             ($fcc$$reg << 11) |            // cc2-cc0 bits for 'fccX'
             ($primary << 5) |              // select single, double or quad
             ($src$$reg << 0);
    cbuf.insts()->emit_int32(op);
  %}

  // Used by the MIN/MAX encodings.  Same as a CMOV, but
  // the condition comes from opcode-field instead of an argument.
  enc_class enc_cmov_reg_minmax( iRegI dst, iRegI src ) %{
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::movcc_op3 << 19) |
             (1 << 18) |                    // cc2 bit for 'icc'
             ($primary << 14) |
             (0 << 13) |                    // select register move
             (0 << 11) |                    // cc1, cc0 bits for 'icc'
             ($src$$reg << 0);
    cbuf.insts()->emit_int32(op);
  %}

  enc_class enc_cmov_reg_minmax_long( iRegL dst, iRegL src ) %{
    int op = (Assembler::arith_op << 30) |
             ($dst$$reg << 25) |
             (Assembler::movcc_op3 << 19) |
             (6 << 16) |                    // cc2 bit for 'xcc'
             ($primary << 14) |
             (0 << 13) |                    // select register move
             (0 << 11) |                    // cc1, cc0 bits for 'icc'
             ($src$$reg << 0);
    cbuf.insts()->emit_int32(op);
  %}

  // Utility encoding for loading a 64 bit Pointer into a register
  // The 64 bit pointer is stored in the generated code stream
  enc_class SetPtr( immP src, iRegP rd ) %{
    Register dest = reg_to_register_object($rd$$reg);
    MacroAssembler _masm(&cbuf);
    // [RGV] This next line should be generated from ADLC
    if ( _opnds[1]->constant_is_oop() ) {
      intptr_t val = $src$$constant;
      __ set_oop_constant((jobject)val, dest);
    } else {          // non-oop pointers, e.g. card mark base, heap top
      __ set($src$$constant, dest);
    }
  %}

  enc_class Set13( immI13 src, iRegI rd ) %{
    emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, $src$$constant );
  %}

  enc_class SetHi22( immI src, iRegI rd ) %{
    emit2_22( cbuf, Assembler::branch_op, $rd$$reg, Assembler::sethi_op2, $src$$constant );
  %}

  enc_class Set32( immI src, iRegI rd ) %{
    MacroAssembler _masm(&cbuf);
    __ set($src$$constant, reg_to_register_object($rd$$reg));
  %}

  enc_class SetNull( iRegI rd ) %{
    emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0 );
  %}

  enc_class call_epilog %{
    if( VerifyStackAtCalls ) {
      MacroAssembler _masm(&cbuf);
      int framesize = ra_->C->frame_slots() << LogBytesPerInt;
      Register temp_reg = G3;
      __ add(SP, framesize, temp_reg);
      __ cmp(temp_reg, FP);
      __ breakpoint_trap(Assembler::notEqual, Assembler::ptr_cc);
    }
  %}

  // Long values come back from native calls in O0:O1 in the 32-bit VM, copy the value
  // to G1 so the register allocator will not have to deal with the misaligned register
  // pair.
  enc_class adjust_long_from_native_call %{
#ifndef _LP64
    if (returns_long()) {
      //    sllx  O0,32,O0
      emit3_simm13( cbuf, Assembler::arith_op, R_O0_enc, Assembler::sllx_op3, R_O0_enc, 0x1020 );
      //    srl   O1,0,O1
      emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::srl_op3, R_O1_enc, 0x0000 );
      //    or    O0,O1,G1
      emit3       ( cbuf, Assembler::arith_op, R_G1_enc, Assembler:: or_op3, R_O0_enc, 0, R_O1_enc );
    }
#endif
  %}

  enc_class Java_To_Runtime (method meth) %{    // CALL Java_To_Runtime
    // CALL directly to the runtime
    // The user of this is responsible for ensuring that R_L7 is empty (killed).
    emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type,
                    /*preserve_g2=*/true, /*force far call*/true);
  %}

  enc_class preserve_SP %{
    MacroAssembler _masm(&cbuf);
    __ mov(SP, L7_mh_SP_save);
  %}

  enc_class restore_SP %{
    MacroAssembler _masm(&cbuf);
    __ mov(L7_mh_SP_save, SP);
  %}

  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.
    if ( !_method ) {
      emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type);
    } else if (_optimized_virtual) {
      emit_call_reloc(cbuf, $meth$$method, relocInfo::opt_virtual_call_type);
    } else {
      emit_call_reloc(cbuf, $meth$$method, relocInfo::static_call_type);
    }
    if( _method ) {  // Emit stub for static call
      emit_java_to_interp(cbuf);
    }
  %}

  enc_class Java_Dynamic_Call (method meth) %{    // JAVA DYNAMIC CALL
    MacroAssembler _masm(&cbuf);
    __ set_inst_mark();
    int vtable_index = this->_vtable_index;
    // MachCallDynamicJavaNode::ret_addr_offset uses this same test
    if (vtable_index < 0) {
      // must be invalid_vtable_index, not nonvirtual_vtable_index
      assert(vtable_index == methodOopDesc::invalid_vtable_index, "correct sentinel value");
      Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
      assert(G5_ic_reg == G5_inline_cache_reg, "G5_inline_cache_reg used in assemble_ic_buffer_code()");
      assert(G5_ic_reg == G5_megamorphic_method, "G5_megamorphic_method used in megamorphic call stub");
      // !!!!!
      // Generate  "set 0x01, R_G5", placeholder instruction to load oop-info
      // emit_call_dynamic_prologue( cbuf );
      __ set_oop((jobject)Universe::non_oop_word(), G5_ic_reg);

      address  virtual_call_oop_addr = __ inst_mark();
      // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
      // who we intended to call.
      __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr));
      emit_call_reloc(cbuf, $meth$$method, relocInfo::none);
    } else {
      assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
      // Just go thru the vtable
      // get receiver klass (receiver already checked for non-null)
      // If we end up going thru a c2i adapter interpreter expects method in G5
      int off = __ offset();
      __ load_klass(O0, G3_scratch);
      int klass_load_size;
      if (UseCompressedOops) {
        assert(Universe::heap() != NULL, "java heap should be initialized");
        if (Universe::narrow_oop_base() == NULL)
          klass_load_size = 2*BytesPerInstWord;
        else
          klass_load_size = 3*BytesPerInstWord;
      } else {
        klass_load_size = 1*BytesPerInstWord;
      }
      int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
      int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
      if( __ is_simm13(v_off) ) {
        __ ld_ptr(G3, v_off, G5_method);
      } else {
        // Generate 2 instructions
        __ Assembler::sethi(v_off & ~0x3ff, G5_method);
        __ or3(G5_method, v_off & 0x3ff, G5_method);
        // ld_ptr, set_hi, set
        assert(__ offset() - off == klass_load_size + 2*BytesPerInstWord,
               "Unexpected instruction size(s)");
        __ ld_ptr(G3, G5_method, G5_method);
      }
      // NOTE: for vtable dispatches, the vtable entry will never be null.
      // However it may very well end up in handle_wrong_method if the
      // method is abstract for the particular class.
      __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch);
      // jump to target (either compiled code or c2iadapter)
      __ jmpl(G3_scratch, G0, O7);
      __ delayed()->nop();
    }
  %}

  enc_class Java_Compiled_Call (method meth) %{    // JAVA COMPILED CALL
    MacroAssembler _masm(&cbuf);

    Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
    Register temp_reg = G3;   // caller must kill G3!  We cannot reuse G5_ic_reg here because
                              // we might be calling a C2I adapter which needs it.

    assert(temp_reg != G5_ic_reg, "conflicting registers");
    // Load nmethod
    __ ld_ptr(G5_ic_reg, in_bytes(methodOopDesc::from_compiled_offset()), temp_reg);

    // CALL to compiled java, indirect the contents of G3
    __ set_inst_mark();
    __ callr(temp_reg, G0);
    __ delayed()->nop();
  %}

enc_class idiv_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst) %{
    MacroAssembler _masm(&cbuf);
    Register Rdividend = reg_to_register_object($src1$$reg);
    Register Rdivisor = reg_to_register_object($src2$$reg);
    Register Rresult = reg_to_register_object($dst$$reg);

    __ sra(Rdivisor, 0, Rdivisor);
    __ sra(Rdividend, 0, Rdividend);
    __ sdivx(Rdividend, Rdivisor, Rresult);
%}

enc_class idiv_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst) %{
    MacroAssembler _masm(&cbuf);

    Register Rdividend = reg_to_register_object($src1$$reg);
    int divisor = $imm$$constant;
    Register Rresult = reg_to_register_object($dst$$reg);

    __ sra(Rdividend, 0, Rdividend);
    __ sdivx(Rdividend, divisor, Rresult);
%}

enc_class enc_mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2) %{
    MacroAssembler _masm(&cbuf);
    Register Rsrc1 = reg_to_register_object($src1$$reg);
    Register Rsrc2 = reg_to_register_object($src2$$reg);
    Register Rdst  = reg_to_register_object($dst$$reg);

    __ sra( Rsrc1, 0, Rsrc1 );
    __ sra( Rsrc2, 0, Rsrc2 );
    __ mulx( Rsrc1, Rsrc2, Rdst );
    __ srlx( Rdst, 32, Rdst );
%}

enc_class irem_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst, o7RegL scratch) %{
    MacroAssembler _masm(&cbuf);
    Register Rdividend = reg_to_register_object($src1$$reg);
    Register Rdivisor = reg_to_register_object($src2$$reg);
    Register Rresult = reg_to_register_object($dst$$reg);
    Register Rscratch = reg_to_register_object($scratch$$reg);

    assert(Rdividend != Rscratch, "");
    assert(Rdivisor  != Rscratch, "");

    __ sra(Rdividend, 0, Rdividend);
    __ sra(Rdivisor, 0, Rdivisor);
    __ sdivx(Rdividend, Rdivisor, Rscratch);
    __ mulx(Rscratch, Rdivisor, Rscratch);
    __ sub(Rdividend, Rscratch, Rresult);
%}

enc_class irem_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst, o7RegL scratch) %{
    MacroAssembler _masm(&cbuf);

    Register Rdividend = reg_to_register_object($src1$$reg);
    int divisor = $imm$$constant;
    Register Rresult = reg_to_register_object($dst$$reg);
    Register Rscratch = reg_to_register_object($scratch$$reg);

    assert(Rdividend != Rscratch, "");

    __ sra(Rdividend, 0, Rdividend);
    __ sdivx(Rdividend, divisor, Rscratch);
    __ mulx(Rscratch, divisor, Rscratch);
    __ sub(Rdividend, Rscratch, Rresult);
%}

enc_class fabss (sflt_reg dst, sflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);

    __ fabs(FloatRegisterImpl::S, Fsrc, Fdst);
%}

enc_class fabsd (dflt_reg dst, dflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);

    __ fabs(FloatRegisterImpl::D, Fsrc, Fdst);
%}

enc_class fnegd (dflt_reg dst, dflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);

    __ fneg(FloatRegisterImpl::D, Fsrc, Fdst);
%}

enc_class fsqrts (sflt_reg dst, sflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);

    __ fsqrt(FloatRegisterImpl::S, Fsrc, Fdst);
%}

enc_class fsqrtd (dflt_reg dst, dflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);

    __ fsqrt(FloatRegisterImpl::D, Fsrc, Fdst);
%}

enc_class fmovs (dflt_reg dst, dflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);

    __ fmov(FloatRegisterImpl::S, Fsrc, Fdst);
%}

enc_class fmovd (dflt_reg dst, dflt_reg src) %{
    MacroAssembler _masm(&cbuf);

    FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
    FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);

    __ fmov(FloatRegisterImpl::D, Fsrc, Fdst);
%}

enc_class Fast_Lock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
    MacroAssembler _masm(&cbuf);

    Register Roop  = reg_to_register_object($oop$$reg);
    Register Rbox  = reg_to_register_object($box$$reg);
    Register Rscratch = reg_to_register_object($scratch$$reg);
    Register Rmark =    reg_to_register_object($scratch2$$reg);

    assert(Roop  != Rscratch, "");
    assert(Roop  != Rmark, "");
    assert(Rbox  != Rscratch, "");
    assert(Rbox  != Rmark, "");

    __ compiler_lock_object(Roop, Rmark, Rbox, Rscratch, _counters, UseBiasedLocking && !UseOptoBiasInlining);
%}

enc_class Fast_Unlock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
    MacroAssembler _masm(&cbuf);

    Register Roop  = reg_to_register_object($oop$$reg);
    Register Rbox  = reg_to_register_object($box$$reg);
    Register Rscratch = reg_to_register_object($scratch$$reg);
    Register Rmark =    reg_to_register_object($scratch2$$reg);

    assert(Roop  != Rscratch, "");
    assert(Roop  != Rmark, "");
    assert(Rbox  != Rscratch, "");
    assert(Rbox  != Rmark, "");

    __ compiler_unlock_object(Roop, Rmark, Rbox, Rscratch, UseBiasedLocking && !UseOptoBiasInlining);
  %}

  enc_class enc_cas( iRegP mem, iRegP old, iRegP new ) %{
    MacroAssembler _masm(&cbuf);
    Register Rmem = reg_to_register_object($mem$$reg);
    Register Rold = reg_to_register_object($old$$reg);
    Register Rnew = reg_to_register_object($new$$reg);

    // casx_under_lock picks 1 of 3 encodings:
    // For 32-bit pointers you get a 32-bit CAS
    // For 64-bit pointers you get a 64-bit CASX
    __ casn(Rmem, Rold, Rnew); // Swap(*Rmem,Rnew) if *Rmem == Rold
    __ cmp( Rold, Rnew );
  %}

  enc_class enc_casx( iRegP mem, iRegL old, iRegL new) %{
    Register Rmem = reg_to_register_object($mem$$reg);
    Register Rold = reg_to_register_object($old$$reg);
    Register Rnew = reg_to_register_object($new$$reg);

    MacroAssembler _masm(&cbuf);
    __ mov(Rnew, O7);
    __ casx(Rmem, Rold, O7);
    __ cmp( Rold, O7 );
  %}

  // raw int cas, used for compareAndSwap
  enc_class enc_casi( iRegP mem, iRegL old, iRegL new) %{
    Register Rmem = reg_to_register_object($mem$$reg);
    Register Rold = reg_to_register_object($old$$reg);
    Register Rnew = reg_to_register_object($new$$reg);

    MacroAssembler _masm(&cbuf);
    __ mov(Rnew, O7);
    __ cas(Rmem, Rold, O7);
    __ cmp( Rold, O7 );
  %}

  enc_class enc_lflags_ne_to_boolean( iRegI res ) %{
    Register Rres = reg_to_register_object($res$$reg);

    MacroAssembler _masm(&cbuf);
    __ mov(1, Rres);
    __ movcc( Assembler::notEqual, false, Assembler::xcc, G0, Rres );
  %}

  enc_class enc_iflags_ne_to_boolean( iRegI res ) %{
    Register Rres = reg_to_register_object($res$$reg);

    MacroAssembler _masm(&cbuf);
    __ mov(1, Rres);
    __ movcc( Assembler::notEqual, false, Assembler::icc, G0, Rres );
  %}

  enc_class floating_cmp ( iRegP dst, regF src1, regF src2 ) %{
    MacroAssembler _masm(&cbuf);
    Register Rdst = reg_to_register_object($dst$$reg);
    FloatRegister Fsrc1 = $primary ? reg_to_SingleFloatRegister_object($src1$$reg)
                                     : reg_to_DoubleFloatRegister_object($src1$$reg);
    FloatRegister Fsrc2 = $primary ? reg_to_SingleFloatRegister_object($src2$$reg)
                                     : reg_to_DoubleFloatRegister_object($src2$$reg);

    // Convert condition code fcc0 into -1,0,1; unordered reports less-than (-1)
    __ float_cmp( $primary, -1, Fsrc1, Fsrc2, Rdst);
  %}

  enc_class LdImmL (immL src, iRegL dst, o7RegL tmp) %{   // Load Immediate
    MacroAssembler _masm(&cbuf);
    Register dest = reg_to_register_object($dst$$reg);
    Register temp = reg_to_register_object($tmp$$reg);
    __ set64( $src$$constant, dest, temp );
  %}

  enc_class LdReplImmI(immI src, regD dst, o7RegP tmp, int count, int width) %{
    // Load a constant replicated "count" times with width "width"
    int bit_width = $width$$constant * 8;
    jlong elt_val = $src$$constant;
    elt_val  &= (((jlong)1) << bit_width) - 1; // mask off sign bits
    jlong val = elt_val;
    for (int i = 0; i < $count$$constant - 1; i++) {
        val <<= bit_width;
        val |= elt_val;
    }
    jdouble dval = *(jdouble*)&val; // coerce to double type
    MacroAssembler _masm(&cbuf);
    address double_address = __ double_constant(dval);
    RelocationHolder rspec = internal_word_Relocation::spec(double_address);
    AddressLiteral addrlit(double_address, rspec);

    __ sethi(addrlit, $tmp$$Register);
    // XXX This is a quick fix for 6833573.
    //__ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), $dst$$FloatRegister, rspec);
    __ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), as_DoubleFloatRegister($dst$$reg), rspec);
  %}

  // Compiler ensures base is doubleword aligned and cnt is count of doublewords
  enc_class enc_Clear_Array(iRegX cnt, iRegP base, iRegX temp) %{
    MacroAssembler _masm(&cbuf);
    Register    nof_bytes_arg   = reg_to_register_object($cnt$$reg);
    Register    nof_bytes_tmp    = reg_to_register_object($temp$$reg);
    Register    base_pointer_arg = reg_to_register_object($base$$reg);

    Label loop;
    __ mov(nof_bytes_arg, nof_bytes_tmp);

    // Loop and clear, walking backwards through the array.
    // nof_bytes_tmp (if >0) is always the number of bytes to zero
    __ bind(loop);
    __ deccc(nof_bytes_tmp, 8);
    __ br(Assembler::greaterEqual, true, Assembler::pt, loop);
    __ delayed()-> stx(G0, base_pointer_arg, nof_bytes_tmp);
    // %%%% this mini-loop must not cross a cache boundary!
  %}


  enc_class enc_String_Compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result) %{
    Label Ldone, Lloop;
    MacroAssembler _masm(&cbuf);

    Register   str1_reg = reg_to_register_object($str1$$reg);
    Register   str2_reg = reg_to_register_object($str2$$reg);
    Register   cnt1_reg = reg_to_register_object($cnt1$$reg);
    Register   cnt2_reg = reg_to_register_object($cnt2$$reg);
    Register result_reg = reg_to_register_object($result$$reg);

    assert(result_reg != str1_reg &&
           result_reg != str2_reg &&
           result_reg != cnt1_reg &&
           result_reg != cnt2_reg ,
           "need different registers");

    // Compute the minimum of the string lengths(str1_reg) and the
    // difference of the string lengths (stack)

    // See if the lengths are different, and calculate min in str1_reg.
    // Stash diff in O7 in case we need it for a tie-breaker.
    Label Lskip;
    __ subcc(cnt1_reg, cnt2_reg, O7);
    __ sll(cnt1_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
    __ br(Assembler::greater, true, Assembler::pt, Lskip);
    // cnt2 is shorter, so use its count:
    __ delayed()->sll(cnt2_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
    __ bind(Lskip);

    // reallocate cnt1_reg, cnt2_reg, result_reg
    // Note:  limit_reg holds the string length pre-scaled by 2
    Register limit_reg =   cnt1_reg;
    Register  chr2_reg =   cnt2_reg;
    Register  chr1_reg = result_reg;
    // str{12} are the base pointers

    // Is the minimum length zero?
    __ cmp(limit_reg, (int)(0 * sizeof(jchar))); // use cast to resolve overloading ambiguity
    __ br(Assembler::equal, true, Assembler::pn, Ldone);
    __ delayed()->mov(O7, result_reg);  // result is difference in lengths

    // Load first characters
    __ lduh(str1_reg, 0, chr1_reg);
    __ lduh(str2_reg, 0, chr2_reg);

    // Compare first characters
    __ subcc(chr1_reg, chr2_reg, chr1_reg);
    __ br(Assembler::notZero, false, Assembler::pt,  Ldone);
    assert(chr1_reg == result_reg, "result must be pre-placed");
    __ delayed()->nop();

    {
      // Check after comparing first character to see if strings are equivalent
      Label LSkip2;
      // Check if the strings start at same location
      __ cmp(str1_reg, str2_reg);
      __ brx(Assembler::notEqual, true, Assembler::pt, LSkip2);
      __ delayed()->nop();

      // Check if the length difference is zero (in O7)
      __ cmp(G0, O7);
      __ br(Assembler::equal, true, Assembler::pn, Ldone);
      __ delayed()->mov(G0, result_reg);  // result is zero

      // Strings might not be equal
      __ bind(LSkip2);
    }

    __ subcc(limit_reg, 1 * sizeof(jchar), chr1_reg);
    __ br(Assembler::equal, true, Assembler::pn, Ldone);
    __ delayed()->mov(O7, result_reg);  // result is difference in lengths

    // Shift str1_reg and str2_reg to the end of the arrays, negate limit
    __ add(str1_reg, limit_reg, str1_reg);
    __ add(str2_reg, limit_reg, str2_reg);
    __ neg(chr1_reg, limit_reg);  // limit = -(limit-2)

    // Compare the rest of the characters
    __ lduh(str1_reg, limit_reg, chr1_reg);
    __ bind(Lloop);
    // __ lduh(str1_reg, limit_reg, chr1_reg); // hoisted
    __ lduh(str2_reg, limit_reg, chr2_reg);
    __ subcc(chr1_reg, chr2_reg, chr1_reg);
    __ br(Assembler::notZero, false, Assembler::pt, Ldone);
    assert(chr1_reg == result_reg, "result must be pre-placed");
    __ delayed()->inccc(limit_reg, sizeof(jchar));
    // annul LDUH if branch is not taken to prevent access past end of string
    __ br(Assembler::notZero, true, Assembler::pt, Lloop);
    __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted

    // If strings are equal up to min length, return the length difference.
    __ mov(O7, result_reg);

    // Otherwise, return the difference between the first mismatched chars.
    __ bind(Ldone);
  %}

enc_class enc_String_Equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result) %{
    Label Lword_loop, Lpost_word, Lchar, Lchar_loop, Ldone;
    MacroAssembler _masm(&cbuf);

    Register   str1_reg = reg_to_register_object($str1$$reg);
    Register   str2_reg = reg_to_register_object($str2$$reg);
    Register    cnt_reg = reg_to_register_object($cnt$$reg);
    Register   tmp1_reg = O7;
    Register result_reg = reg_to_register_object($result$$reg);

    assert(result_reg != str1_reg &&
           result_reg != str2_reg &&
           result_reg !=  cnt_reg &&
           result_reg != tmp1_reg ,
           "need different registers");

    __ cmp(str1_reg, str2_reg); //same char[] ?
    __ brx(Assembler::equal, true, Assembler::pn, Ldone);
    __ delayed()->add(G0, 1, result_reg);

    __ br_on_reg_cond(Assembler::rc_z, true, Assembler::pn, cnt_reg, Ldone);
    __ delayed()->add(G0, 1, result_reg); // count == 0

    //rename registers
    Register limit_reg =    cnt_reg;
    Register  chr1_reg = result_reg;
    Register  chr2_reg =   tmp1_reg;

    //check for alignment and position the pointers to the ends
    __ or3(str1_reg, str2_reg, chr1_reg);
    __ andcc(chr1_reg, 0x3, chr1_reg);
    // notZero means at least one not 4-byte aligned.
    // We could optimize the case when both arrays are not aligned
    // but it is not frequent case and it requires additional checks.
    __ br(Assembler::notZero, false, Assembler::pn, Lchar); // char by char compare
    __ delayed()->sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg); // set byte count

    // Compare char[] arrays aligned to 4 bytes.
    __ char_arrays_equals(str1_reg, str2_reg, limit_reg, result_reg,
                          chr1_reg, chr2_reg, Ldone);
    __ ba(false,Ldone);
    __ delayed()->add(G0, 1, result_reg);

    // char by char compare
    __ bind(Lchar);
    __ add(str1_reg, limit_reg, str1_reg);
    __ add(str2_reg, limit_reg, str2_reg);
    __ neg(limit_reg); //negate count

    __ lduh(str1_reg, limit_reg, chr1_reg);
    // Lchar_loop
    __ bind(Lchar_loop);
    __ lduh(str2_reg, limit_reg, chr2_reg);
    __ cmp(chr1_reg, chr2_reg);
    __ br(Assembler::notEqual, true, Assembler::pt, Ldone);
    __ delayed()->mov(G0, result_reg); //not equal
    __ inccc(limit_reg, sizeof(jchar));
    // annul LDUH if branch is not taken to prevent access past end of string
    __ br(Assembler::notZero, true, Assembler::pt, Lchar_loop);
    __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted

    __ add(G0, 1, result_reg);  //equal

    __ bind(Ldone);
  %}

enc_class enc_Array_Equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, notemp_iRegI result) %{
    Label Lvector, Ldone, Lloop;
    MacroAssembler _masm(&cbuf);

    Register   ary1_reg = reg_to_register_object($ary1$$reg);
    Register   ary2_reg = reg_to_register_object($ary2$$reg);
    Register   tmp1_reg = reg_to_register_object($tmp1$$reg);
    Register   tmp2_reg = O7;
    Register result_reg = reg_to_register_object($result$$reg);

    int length_offset  = arrayOopDesc::length_offset_in_bytes();
    int base_offset    = arrayOopDesc::base_offset_in_bytes(T_CHAR);

    // return true if the same array
    __ cmp(ary1_reg, ary2_reg);
    __ brx(Assembler::equal, true, Assembler::pn, Ldone);
    __ delayed()->add(G0, 1, result_reg); // equal

    __ br_null(ary1_reg, true, Assembler::pn, Ldone);
    __ delayed()->mov(G0, result_reg);    // not equal

    __ br_null(ary2_reg, true, Assembler::pn, Ldone);
    __ delayed()->mov(G0, result_reg);    // not equal

    //load the lengths of arrays
    __ ld(Address(ary1_reg, length_offset), tmp1_reg);
    __ ld(Address(ary2_reg, length_offset), tmp2_reg);

    // return false if the two arrays are not equal length
    __ cmp(tmp1_reg, tmp2_reg);
    __ br(Assembler::notEqual, true, Assembler::pn, Ldone);
    __ delayed()->mov(G0, result_reg);     // not equal

    __ br_on_reg_cond(Assembler::rc_z, true, Assembler::pn, tmp1_reg, Ldone);
    __ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal

    // load array addresses
    __ add(ary1_reg, base_offset, ary1_reg);
    __ add(ary2_reg, base_offset, ary2_reg);

    // renaming registers
    Register chr1_reg  =  result_reg; // for characters in ary1
    Register chr2_reg  =  tmp2_reg;   // for characters in ary2
    Register limit_reg =  tmp1_reg;   // length

    // set byte count
    __ sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg);

    // Compare char[] arrays aligned to 4 bytes.
    __ char_arrays_equals(ary1_reg, ary2_reg, limit_reg, result_reg,
                          chr1_reg, chr2_reg, Ldone);
    __ add(G0, 1, result_reg); // equals

    __ bind(Ldone);
  %}

  enc_class enc_rethrow() %{
    cbuf.set_insts_mark();
    Register temp_reg = G3;
    AddressLiteral rethrow_stub(OptoRuntime::rethrow_stub());
    assert(temp_reg != reg_to_register_object(R_I0_num), "temp must not break oop_reg");
    MacroAssembler _masm(&cbuf);
#ifdef ASSERT
    __ save_frame(0);
    AddressLiteral last_rethrow_addrlit(&last_rethrow);
    __ sethi(last_rethrow_addrlit, L1);
    Address addr(L1, last_rethrow_addrlit.low10());
    __ get_pc(L2);
    __ inc(L2, 3 * BytesPerInstWord);  // skip this & 2 more insns to point at jump_to
    __ st_ptr(L2, addr);
    __ restore();
#endif
    __ JUMP(rethrow_stub, temp_reg, 0); // sethi;jmp
    __ delayed()->nop();
  %}

  enc_class emit_mem_nop() %{
    // Generates the instruction LDUXA [o6,g0],#0x82,g0
    cbuf.insts()->emit_int32((unsigned int) 0xc0839040);
  %}

  enc_class emit_fadd_nop() %{
    // Generates the instruction FMOVS f31,f31
    cbuf.insts()->emit_int32((unsigned int) 0xbfa0003f);
  %}

  enc_class emit_br_nop() %{
    // Generates the instruction BPN,PN .
    cbuf.insts()->emit_int32((unsigned int) 0x00400000);
  %}

  enc_class enc_membar_acquire %{
    MacroAssembler _masm(&cbuf);
    __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::LoadLoad) );
  %}

  enc_class enc_membar_release %{
    MacroAssembler _masm(&cbuf);
    __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore) );
  %}

  enc_class enc_membar_volatile %{
    MacroAssembler _masm(&cbuf);
    __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
  %}

  enc_class enc_repl8b( iRegI src, iRegL dst ) %{
    MacroAssembler _masm(&cbuf);
    Register src_reg = reg_to_register_object($src$$reg);
    Register dst_reg = reg_to_register_object($dst$$reg);
    __ sllx(src_reg, 56, dst_reg);
    __ srlx(dst_reg,  8, O7);
    __ or3 (dst_reg, O7, dst_reg);
    __ srlx(dst_reg, 16, O7);
    __ or3 (dst_reg, O7, dst_reg);
    __ srlx(dst_reg, 32, O7);
    __ or3 (dst_reg, O7, dst_reg);
  %}

  enc_class enc_repl4b( iRegI src, iRegL dst ) %{
    MacroAssembler _masm(&cbuf);
    Register src_reg = reg_to_register_object($src$$reg);
    Register dst_reg = reg_to_register_object($dst$$reg);
    __ sll(src_reg, 24, dst_reg);
    __ srl(dst_reg,  8, O7);
    __ or3(dst_reg, O7, dst_reg);
    __ srl(dst_reg, 16, O7);
    __ or3(dst_reg, O7, dst_reg);
  %}

  enc_class enc_repl4s( iRegI src, iRegL dst ) %{
    MacroAssembler _masm(&cbuf);
    Register src_reg = reg_to_register_object($src$$reg);
    Register dst_reg = reg_to_register_object($dst$$reg);
    __ sllx(src_reg, 48, dst_reg);
    __ srlx(dst_reg, 16, O7);
    __ or3 (dst_reg, O7, dst_reg);
    __ srlx(dst_reg, 32, O7);
    __ or3 (dst_reg, O7, dst_reg);
  %}

  enc_class enc_repl2i( iRegI src, iRegL dst ) %{
    MacroAssembler _masm(&cbuf);
    Register src_reg = reg_to_register_object($src$$reg);
    Register dst_reg = reg_to_register_object($dst$$reg);
    __ sllx(src_reg, 32, dst_reg);
    __ srlx(dst_reg, 32, O7);
    __ or3 (dst_reg, O7, dst_reg);
  %}

%}

//----------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 VMRegImpl::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, 8 (or 16 in LP64)-byte 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
//        |        +--------+----> VMRegImpl::stack0, 8 (or 16 in LP64)-byte 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 native & Java)
  stack_direction(TOWARDS_LOW);

  // These two registers define part of the calling convention
  // between compiled code and the interpreter.
  inline_cache_reg(R_G5);                // Inline Cache Register or methodOop for I2C
  interpreter_method_oop_reg(R_G5);      // Method Oop Register when calling interpreter

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

  // Number of stack slots consumed by a Monitor enter
#ifdef _LP64
  sync_stack_slots(2);
#else
  sync_stack_slots(1);
#endif

  // Compiled code's Frame Pointer
  frame_pointer(R_SP);

  // Stack alignment requirement
  stack_alignment(StackAlignmentInBytes);
  //  LP64: Alignment size in bytes (128-bit -> 16 bytes)
  // !LP64: Alignment size in bytes (64-bit  ->  8 bytes)

  // 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.
  in_preserve_stack_slots(0);

  // 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.
  // ADLC doesn't support parsing expressions, so I folded the math by hand.
#ifdef _LP64
  // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (0)) * 2-stack-slots-per-word
  varargs_C_out_slots_killed(12);
#else
  // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (1)) * 1-stack-slots-per-word
  varargs_C_out_slots_killed( 7);
#endif

  // 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.
  return_addr(REG R_I7);          // Ret Addr is in register I7

  // Body of function which returns an OptoRegs array locating
  // arguments either in registers or in stack slots for calling
  // java
  calling_convention %{
    (void) SharedRuntime::java_calling_convention(sig_bt, regs, length, is_outgoing);

  %}

  // Body of function which returns an OptoRegs array locating
  // arguments either in registers or in stack slots for callin
  // C.
  c_calling_convention %{
    // This is obviously always outgoing
    (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
  %}

  // Location of native (C/C++) and interpreter return values.  This is specified to
  // be the  same as Java.  In the 32-bit VM, long values are actually returned from
  // native calls in O0:O1 and returned to the interpreter in I0:I1.  The copying
  // to and from the register pairs is done by the appropriate call and epilog
  // opcodes.  This simplifies the register allocator.
  c_return_value %{
    assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
#ifdef     _LP64
    static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_O0_num };
    static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num,    OptoReg::Bad, R_F1_num, R_O0H_num};
    static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_I0_num };
    static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num,    OptoReg::Bad, R_F1_num, R_I0H_num};
#else  // !_LP64
    static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_G1_num };
    static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
    static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_G1_num };
    static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
#endif
    return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
                        (is_outgoing?lo_out:lo_in)[ideal_reg] );
  %}

  // Location of compiled Java return values.  Same as C
  return_value %{
    assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
#ifdef     _LP64
    static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_O0_num };
    static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num,    OptoReg::Bad, R_F1_num, R_O0H_num};
    static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_I0_num };
    static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num,    OptoReg::Bad, R_F1_num, R_I0H_num};
#else  // !_LP64
    static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_G1_num };
    static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
    static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_G1_num };
    static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
#endif
    return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
                        (is_outgoing?lo_out:lo_in)[ideal_reg] );
  %}

%}


//----------ATTRIBUTES---------------------------------------------------------
//----------Operand Attributes-------------------------------------------------
op_attrib op_cost(1);          // Required cost attribute

//----------Instruction Attributes---------------------------------------------
ins_attrib ins_cost(DEFAULT_COST); // Required cost attribute
ins_attrib ins_size(32);       // Required size attribute (in bits)
ins_attrib ins_pc_relative(0); // Required PC Relative flag
ins_attrib ins_short_branch(0); // Required flag: is this instruction a
                                // non-matching short branch variant of some
                                                            // long branch?

//----------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: 32-bit
operand immI() %{
  match(ConI);

  op_cost(0);
  // formats are generated automatically for constants and base registers
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 8-bit
operand immI8() %{
  predicate(Assembler::is_simm(n->get_int(), 8));
  match(ConI);
  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 13-bit
operand immI13() %{
  predicate(Assembler::is_simm13(n->get_int()));
  match(ConI);
  op_cost(0);

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

// Integer Immediate: 13-bit minus 7
operand immI13m7() %{
  predicate((-4096 < n->get_int()) && ((n->get_int() + 7) <= 4095));
  match(ConI);
  op_cost(0);

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

// Integer Immediate: 16-bit
operand immI16() %{
  predicate(Assembler::is_simm(n->get_int(), 16));
  match(ConI);
  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned (positive) Integer Immediate: 13-bit
operand immU13() %{
  predicate((0 <= n->get_int()) && Assembler::is_simm13(n->get_int()));
  match(ConI);
  op_cost(0);

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

// Integer Immediate: 6-bit
operand immU6() %{
  predicate(n->get_int() >= 0 && n->get_int() <= 63);
  match(ConI);
  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 11-bit
operand immI11() %{
  predicate(Assembler::is_simm(n->get_int(),11));
  match(ConI);
  op_cost(0);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 0-bit
operand immI0() %{
  predicate(n->get_int() == 0);
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the value 10
operand immI10() %{
  predicate(n->get_int() == 10);
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the values 0-31
operand immU5() %{
  predicate(n->get_int() >= 0 && n->get_int() <= 31);
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the values 1-31
operand immI_1_31() %{
  predicate(n->get_int() >= 1 && n->get_int() <= 31);
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the values 32-63
operand immI_32_63() %{
  predicate(n->get_int() >= 32 && n->get_int() <= 63);
  match(ConI);
  op_cost(0);

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

// Immediates for special shifts (sign extend)

// Integer Immediate: the value 16
operand immI_16() %{
  predicate(n->get_int() == 16);
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the value 24
operand immI_24() %{
  predicate(n->get_int() == 24);
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the value 255
operand immI_255() %{
  predicate( n->get_int() == 255 );
  match(ConI);
  op_cost(0);

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

// Integer Immediate: the value 65535
operand immI_65535() %{
  predicate(n->get_int() == 65535);
  match(ConI);
  op_cost(0);

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

// Long Immediate: the value FF
operand immL_FF() %{
  predicate( n->get_long() == 0xFFL );
  match(ConL);
  op_cost(0);

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

// Long Immediate: the value FFFF
operand immL_FFFF() %{
  predicate( n->get_long() == 0xFFFFL );
  match(ConL);
  op_cost(0);

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

// Pointer Immediate: 32 or 64-bit
operand immP() %{
  match(ConP);

  op_cost(5);
  // formats are generated automatically for constants and base registers
  format %{ %}
  interface(CONST_INTER);
%}

operand immP13() %{
  predicate((-4096 < n->get_ptr()) && (n->get_ptr() <= 4095));
  match(ConP);
  op_cost(0);

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

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

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

operand immP_poll() %{
  predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page());
  match(ConP);

  // formats are generated automatically for constants and base registers
  format %{ %}
  interface(CONST_INTER);
%}

// Pointer Immediate
operand immN()
%{
  match(ConN);

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

// NULL Pointer Immediate
operand immN0()
%{
  predicate(n->get_narrowcon() == 0);
  match(ConN);

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

operand immL() %{
  match(ConL);
  op_cost(40);
  // formats are generated automatically for constants and base registers
  format %{ %}
  interface(CONST_INTER);
%}

operand immL0() %{
  predicate(n->get_long() == 0L);
  match(ConL);
  op_cost(0);
  // formats are generated automatically for constants and base registers
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: 13-bit
operand immL13() %{
  predicate((-4096L < n->get_long()) && (n->get_long() <= 4095L));
  match(ConL);
  op_cost(0);

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

// Long Immediate: 13-bit minus 7
operand immL13m7() %{
  predicate((-4096L < n->get_long()) && ((n->get_long() + 7L) <= 4095L));
  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);
%}

// Double Immediate
operand immD() %{
  match(ConD);

  op_cost(40);
  format %{ %}
  interface(CONST_INTER);
%}

operand immD0() %{
#ifdef _LP64
  // on 64-bit architectures this comparision is faster
  predicate(jlong_cast(n->getd()) == 0);
#else
  predicate((n->getd() == 0) && (fpclass(n->getd()) == FP_PZERO));
#endif
  match(ConD);

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

// Float Immediate
operand immF() %{
  match(ConF);

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

// Float Immediate: 0
operand immF0() %{
  predicate((n->getf() == 0) && (fpclass(n->getf()) == FP_PZERO));
  match(ConF);

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

// Integer Register Operands
// Integer Register
operand iRegI() %{
  constraint(ALLOC_IN_RC(int_reg));
  match(RegI);

  match(notemp_iRegI);
  match(g1RegI);
  match(o0RegI);
  match(iRegIsafe);

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

operand notemp_iRegI() %{
  constraint(ALLOC_IN_RC(notemp_int_reg));
  match(RegI);

  match(o0RegI);

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

operand o0RegI() %{
  constraint(ALLOC_IN_RC(o0_regI));
  match(iRegI);

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

// Pointer Register
operand iRegP() %{
  constraint(ALLOC_IN_RC(ptr_reg));
  match(RegP);

  match(lock_ptr_RegP);
  match(g1RegP);
  match(g2RegP);
  match(g3RegP);
  match(g4RegP);
  match(i0RegP);
  match(o0RegP);
  match(o1RegP);
  match(l7RegP);

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

operand sp_ptr_RegP() %{
  constraint(ALLOC_IN_RC(sp_ptr_reg));
  match(RegP);
  match(iRegP);

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

operand lock_ptr_RegP() %{
  constraint(ALLOC_IN_RC(lock_ptr_reg));
  match(RegP);
  match(i0RegP);
  match(o0RegP);
  match(o1RegP);
  match(l7RegP);

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

operand g1RegP() %{
  constraint(ALLOC_IN_RC(g1_regP));
  match(iRegP);

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

operand g2RegP() %{
  constraint(ALLOC_IN_RC(g2_regP));
  match(iRegP);

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

operand g3RegP() %{
  constraint(ALLOC_IN_RC(g3_regP));
  match(iRegP);

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

operand g1RegI() %{
  constraint(ALLOC_IN_RC(g1_regI));
  match(iRegI);

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

operand g3RegI() %{
  constraint(ALLOC_IN_RC(g3_regI));
  match(iRegI);

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

operand g4RegI() %{
  constraint(ALLOC_IN_RC(g4_regI));
  match(iRegI);

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

operand g4RegP() %{
  constraint(ALLOC_IN_RC(g4_regP));
  match(iRegP);

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

operand i0RegP() %{
  constraint(ALLOC_IN_RC(i0_regP));
  match(iRegP);

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

operand o0RegP() %{
  constraint(ALLOC_IN_RC(o0_regP));
  match(iRegP);

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

operand o1RegP() %{
  constraint(ALLOC_IN_RC(o1_regP));
  match(iRegP);

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

operand o2RegP() %{
  constraint(ALLOC_IN_RC(o2_regP));
  match(iRegP);

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

operand o7RegP() %{
  constraint(ALLOC_IN_RC(o7_regP));
  match(iRegP);

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

operand l7RegP() %{
  constraint(ALLOC_IN_RC(l7_regP));
  match(iRegP);

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

operand o7RegI() %{
  constraint(ALLOC_IN_RC(o7_regI));
  match(iRegI);

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

operand iRegN() %{
  constraint(ALLOC_IN_RC(int_reg));
  match(RegN);

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

// Long Register
operand iRegL() %{
  constraint(ALLOC_IN_RC(long_reg));
  match(RegL);

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

operand o2RegL() %{
  constraint(ALLOC_IN_RC(o2_regL));
  match(iRegL);

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

operand o7RegL() %{
  constraint(ALLOC_IN_RC(o7_regL));
  match(iRegL);

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

operand g1RegL() %{
  constraint(ALLOC_IN_RC(g1_regL));
  match(iRegL);

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

operand g3RegL() %{
  constraint(ALLOC_IN_RC(g3_regL));
  match(iRegL);

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

// Int Register safe
// This is 64bit safe
operand iRegIsafe() %{
  constraint(ALLOC_IN_RC(long_reg));

  match(iRegI);

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

// Condition Code Flag Register
operand flagsReg() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);

  format %{ "ccr" %} // both ICC and XCC
  interface(REG_INTER);
%}

// Condition Code Register, unsigned comparisons.
operand flagsRegU() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);

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

// Condition Code Register, pointer comparisons.
operand flagsRegP() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);

#ifdef _LP64
  format %{ "xcc_P" %}
#else
  format %{ "icc_P" %}
#endif
  interface(REG_INTER);
%}

// Condition Code Register, long comparisons.
operand flagsRegL() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);

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

// Condition Code Register, floating comparisons, unordered same as "less".
operand flagsRegF() %{
  constraint(ALLOC_IN_RC(float_flags));
  match(RegFlags);
  match(flagsRegF0);

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

operand flagsRegF0() %{
  constraint(ALLOC_IN_RC(float_flag0));
  match(RegFlags);

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


// Condition Code Flag Register used by long compare
operand flagsReg_long_LTGE() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  format %{ "icc_LTGE" %}
  interface(REG_INTER);
%}
operand flagsReg_long_EQNE() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  format %{ "icc_EQNE" %}
  interface(REG_INTER);
%}
operand flagsReg_long_LEGT() %{
  constraint(ALLOC_IN_RC(int_flags));
  match(RegFlags);
  format %{ "icc_LEGT" %}
  interface(REG_INTER);
%}


operand regD() %{
  constraint(ALLOC_IN_RC(dflt_reg));
  match(RegD);

  match(regD_low);

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

operand regF() %{
  constraint(ALLOC_IN_RC(sflt_reg));
  match(RegF);

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

operand regD_low() %{
  constraint(ALLOC_IN_RC(dflt_low_reg));
  match(regD);

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

// Special Registers

// Method Register
operand inline_cache_regP(iRegP reg) %{
  constraint(ALLOC_IN_RC(g5_regP)); // G5=inline_cache_reg but uses 2 bits instead of 1
  match(reg);
  format %{ %}
  interface(REG_INTER);
%}

operand interpreter_method_oop_regP(iRegP reg) %{
  constraint(ALLOC_IN_RC(g5_regP)); // G5=interpreter_method_oop_reg but uses 2 bits instead of 1
  match(reg);
  format %{ %}
  interface(REG_INTER);
%}


//----------Complex Operands---------------------------------------------------
// Indirect Memory Reference
operand indirect(sp_ptr_RegP reg) %{
  constraint(ALLOC_IN_RC(sp_ptr_reg));
  match(reg);

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

// Indirect with simm13 Offset
operand indOffset13(sp_ptr_RegP reg, immX13 offset) %{
  constraint(ALLOC_IN_RC(sp_ptr_reg));
  match(AddP reg offset);

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

// Indirect with simm13 Offset minus 7
operand indOffset13m7(sp_ptr_RegP reg, immX13m7 offset) %{
  constraint(ALLOC_IN_RC(sp_ptr_reg));
  match(AddP reg offset);

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

// Note:  Intel has a swapped version also, like this:
//operand indOffsetX(iRegI reg, immP offset) %{
//  constraint(ALLOC_IN_RC(int_reg));
//  match(AddP offset reg);
//
//  op_cost(100);
//  format %{ "[$reg + $offset]" %}
//  interface(MEMORY_INTER) %{
//    base($reg);
//    index(0x0);
//    scale(0x0);
//    disp($offset);
//  %}
//%}
//// However, it doesn't make sense for SPARC, since
// we have no particularly good way to embed oops in
// single instructions.

// Indirect with Register Index
operand indIndex(iRegP addr, iRegX index) %{
  constraint(ALLOC_IN_RC(ptr_reg));
  match(AddP addr index);

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

//----------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 stackSlotI(sRegI reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(100);
  //match(RegI);
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0xE);   // R_SP
    index(0x0);
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotP(sRegP reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(100);
  //match(RegP);
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0xE);   // R_SP
    index(0x0);
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotF(sRegF reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(100);
  //match(RegF);
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0xE);   // R_SP
    index(0x0);
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}
operand stackSlotD(sRegD reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(100);
  //match(RegD);
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0xE);   // R_SP
    index(0x0);
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}
operand stackSlotL(sRegL reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(100);
  //match(RegL);
  format %{ "[$reg]" %}
  interface(MEMORY_INTER) %{
    base(0xE);   // R_SP
    index(0x0);
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}

// Operands for expressing Control Flow
// NOTE:  Label is a predefined operand which should not be redefined in
//        the AD file.  It is generically handled within the ADLC.

//----------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.

operand cmpOp() %{
  match(Bool);

  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x1);
    not_equal(0x9);
    less(0x3);
    greater_equal(0xB);
    less_equal(0x2);
    greater(0xA);
  %}
%}

// Comparison Op, unsigned
operand cmpOpU() %{
  match(Bool);

  format %{ "u" %}
  interface(COND_INTER) %{
    equal(0x1);
    not_equal(0x9);
    less(0x5);
    greater_equal(0xD);
    less_equal(0x4);
    greater(0xC);
  %}
%}

// Comparison Op, pointer (same as unsigned)
operand cmpOpP() %{
  match(Bool);

  format %{ "p" %}
  interface(COND_INTER) %{
    equal(0x1);
    not_equal(0x9);
    less(0x5);
    greater_equal(0xD);
    less_equal(0x4);
    greater(0xC);
  %}
%}

// Comparison Op, branch-register encoding
operand cmpOp_reg() %{
  match(Bool);

  format %{ "" %}
  interface(COND_INTER) %{
    equal        (0x1);
    not_equal    (0x5);
    less         (0x3);
    greater_equal(0x7);
    less_equal   (0x2);
    greater      (0x6);
  %}
%}

// Comparison Code, floating, unordered same as less
operand cmpOpF() %{
  match(Bool);

  format %{ "fl" %}
  interface(COND_INTER) %{
    equal(0x9);
    not_equal(0x1);
    less(0x3);
    greater_equal(0xB);
    less_equal(0xE);
    greater(0x6);
  %}
%}

// Used by long compare
operand cmpOp_commute() %{
  match(Bool);

  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x1);
    not_equal(0x9);
    less(0xA);
    greater_equal(0x2);
    less_equal(0xB);
    greater(0x3);
  %}
%}

//----------OPERAND CLASSES----------------------------------------------------
// Operand Classes are groups of operands that are used 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( indirect, indOffset13, indIndex );
opclass indIndexMemory( indIndex );

//----------PIPELINE-----------------------------------------------------------
pipeline %{

//----------ATTRIBUTES---------------------------------------------------------
attributes %{
  fixed_size_instructions;           // Fixed size instructions
  branch_has_delay_slot;             // Branch has delay slot following
  max_instructions_per_bundle = 4;   // Up to 4 instructions per bundle
  instruction_unit_size = 4;         // An instruction is 4 bytes long
  instruction_fetch_unit_size = 16;  // The processor fetches one line
  instruction_fetch_units = 1;       // of 16 bytes

  // List of nop instructions
  nops( Nop_A0, Nop_A1, Nop_MS, Nop_FA, Nop_BR );
%}

//----------RESOURCES----------------------------------------------------------
// Resources are the functional units available to the machine
resources(A0, A1, MS, BR, FA, FM, IDIV, FDIV, IALU = A0 | A1);

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

pipe_desc(A, P, F, B, I, J, S, R, E, C, M, W, X, T, D);

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

// Integer ALU reg-reg operation
pipe_class ialu_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
    single_instruction;
    dst   : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU reg-reg long operation
pipe_class ialu_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
    instruction_count(2);
    dst   : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
    IALU  : R;
%}

// Integer ALU reg-reg long dependent operation
pipe_class ialu_reg_reg_2_dep(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write);
    src1  : R(read);
    src2  : R(read);
    cr    : E(write);
    IALU  : R(2);
%}

// Integer ALU reg-imm operaion
pipe_class ialu_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
    single_instruction;
    dst   : E(write);
    src1  : R(read);
    IALU  : R;
%}

// Integer ALU reg-reg operation with condition code
pipe_class ialu_cc_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
    single_instruction;
    dst   : E(write);
    cr    : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU reg-imm operation with condition code
pipe_class ialu_cc_reg_imm(iRegI dst, iRegI src1, immI13 src2, flagsReg cr) %{
    single_instruction;
    dst   : E(write);
    cr    : E(write);
    src1  : R(read);
    IALU  : R;
%}

// Integer ALU zero-reg operation
pipe_class ialu_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
    single_instruction;
    dst   : E(write);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU zero-reg operation with condition code only
pipe_class ialu_cconly_zero_reg(flagsReg cr, iRegI src) %{
    single_instruction;
    cr    : E(write);
    src   : R(read);
    IALU  : R;
%}

// Integer ALU reg-reg operation with condition code only
pipe_class ialu_cconly_reg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
    single_instruction;
    cr    : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU reg-imm operation with condition code only
pipe_class ialu_cconly_reg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
    single_instruction;
    cr    : E(write);
    src1  : R(read);
    IALU  : R;
%}

// Integer ALU reg-reg-zero operation with condition code only
pipe_class ialu_cconly_reg_reg_zero(flagsReg cr, iRegI src1, iRegI src2, immI0 zero) %{
    single_instruction;
    cr    : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU reg-imm-zero operation with condition code only
pipe_class ialu_cconly_reg_imm_zero(flagsReg cr, iRegI src1, immI13 src2, immI0 zero) %{
    single_instruction;
    cr    : E(write);
    src1  : R(read);
    IALU  : R;
%}

// Integer ALU reg-reg operation with condition code, src1 modified
pipe_class ialu_cc_rwreg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
    single_instruction;
    cr    : E(write);
    src1  : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU reg-imm operation with condition code, src1 modified
pipe_class ialu_cc_rwreg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
    single_instruction;
    cr    : E(write);
    src1  : E(write);
    src1  : R(read);
    IALU  : R;
%}

pipe_class cmpL_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg cr ) %{
    multiple_bundles;
    dst   : E(write)+4;
    cr    : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R(3);
    BR    : R(2);
%}

// Integer ALU operation
pipe_class ialu_none(iRegI dst) %{
    single_instruction;
    dst   : E(write);
    IALU  : R;
%}

// Integer ALU reg operation
pipe_class ialu_reg(iRegI dst, iRegI src) %{
    single_instruction; may_have_no_code;
    dst   : E(write);
    src   : R(read);
    IALU  : R;
%}

// Integer ALU reg conditional operation
// This instruction has a 1 cycle stall, and cannot execute
// in the same cycle as the instruction setting the condition
// code. We kludge this by pretending to read the condition code
// 1 cycle earlier, and by marking the functional units as busy
// for 2 cycles with the result available 1 cycle later than
// is really the case.
pipe_class ialu_reg_flags( iRegI op2_out, iRegI op2_in, iRegI op1, flagsReg cr ) %{
    single_instruction;
    op2_out : C(write);
    op1     : R(read);
    cr      : R(read);       // This is really E, with a 1 cycle stall
    BR      : R(2);
    MS      : R(2);
%}

#ifdef _LP64
pipe_class ialu_clr_and_mover( iRegI dst, iRegP src ) %{
    instruction_count(1); multiple_bundles;
    dst     : C(write)+1;
    src     : R(read)+1;
    IALU    : R(1);
    BR      : E(2);
    MS      : E(2);
%}
#endif

// Integer ALU reg operation
pipe_class ialu_move_reg_L_to_I(iRegI dst, iRegL src) %{
    single_instruction; may_have_no_code;
    dst   : E(write);
    src   : R(read);
    IALU  : R;
%}
pipe_class ialu_move_reg_I_to_L(iRegL dst, iRegI src) %{
    single_instruction; may_have_no_code;
    dst   : E(write);
    src   : R(read);
    IALU  : R;
%}

// Two integer ALU reg operations
pipe_class ialu_reg_2(iRegL dst, iRegL src) %{
    instruction_count(2);
    dst   : E(write);
    src   : R(read);
    A0    : R;
    A1    : R;
%}

// Two integer ALU reg operations
pipe_class ialu_move_reg_L_to_L(iRegL dst, iRegL src) %{
    instruction_count(2); may_have_no_code;
    dst   : E(write);
    src   : R(read);
    A0    : R;
    A1    : R;
%}

// Integer ALU imm operation
pipe_class ialu_imm(iRegI dst, immI13 src) %{
    single_instruction;
    dst   : E(write);
    IALU  : R;
%}

// Integer ALU reg-reg with carry operation
pipe_class ialu_reg_reg_cy(iRegI dst, iRegI src1, iRegI src2, iRegI cy) %{
    single_instruction;
    dst   : E(write);
    src1  : R(read);
    src2  : R(read);
    IALU  : R;
%}

// Integer ALU cc operation
pipe_class ialu_cc(iRegI dst, flagsReg cc) %{
    single_instruction;
    dst   : E(write);
    cc    : R(read);
    IALU  : R;
%}

// Integer ALU cc / second IALU operation
pipe_class ialu_reg_ialu( iRegI dst, iRegI src ) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write)+1;
    src   : R(read);
    IALU  : R;
%}

// Integer ALU cc / second IALU operation
pipe_class ialu_reg_reg_ialu( iRegI dst, iRegI p, iRegI q ) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write)+1;
    p     : R(read);
    q     : R(read);
    IALU  : R;
%}

// Integer ALU hi-lo-reg operation
pipe_class ialu_hi_lo_reg(iRegI dst, immI src) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write)+1;
    IALU  : R(2);
%}

// Float ALU hi-lo-reg operation (with temp)
pipe_class ialu_hi_lo_reg_temp(regF dst, immF src, g3RegP tmp) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write)+1;
    IALU  : R(2);
%}

// Long Constant
pipe_class loadConL( iRegL dst, immL src ) %{
    instruction_count(2); multiple_bundles;
    dst   : E(write)+1;
    IALU  : R(2);
    IALU  : R(2);
%}

// Pointer Constant
pipe_class loadConP( iRegP dst, immP src ) %{
    instruction_count(0); multiple_bundles;
    fixed_latency(6);
%}

// Polling Address
pipe_class loadConP_poll( iRegP dst, immP_poll src ) %{
#ifdef _LP64
    instruction_count(0); multiple_bundles;
    fixed_latency(6);
#else
    dst   : E(write);
    IALU  : R;
#endif
%}

// Long Constant small
pipe_class loadConLlo( iRegL dst, immL src ) %{
    instruction_count(2);
    dst   : E(write);
    IALU  : R;
    IALU  : R;
%}

// [PHH] This is wrong for 64-bit.  See LdImmF/D.
pipe_class loadConFD(regF dst, immF src, g3RegP tmp) %{
    instruction_count(1); multiple_bundles;
    src   : R(read);
    dst   : M(write)+1;
    IALU  : R;
    MS    : E;
%}

// Integer ALU nop operation
pipe_class ialu_nop() %{
    single_instruction;
    IALU  : R;
%}

// Integer ALU nop operation
pipe_class ialu_nop_A0() %{
    single_instruction;
    A0    : R;
%}

// Integer ALU nop operation
pipe_class ialu_nop_A1() %{
    single_instruction;
    A1    : R;
%}

// Integer Multiply reg-reg operation
pipe_class imul_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
    single_instruction;
    dst   : E(write);
    src1  : R(read);
    src2  : R(read);
    MS    : R(5);
%}

// Integer Multiply reg-imm operation
pipe_class imul_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
    single_instruction;
    dst   : E(write);
    src1  : R(read);
    MS    : R(5);
%}

pipe_class mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
    single_instruction;
    dst   : E(write)+4;
    src1  : R(read);
    src2  : R(read);
    MS    : R(6);
%}

pipe_class mulL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
    single_instruction;
    dst   : E(write)+4;
    src1  : R(read);
    MS    : R(6);
%}

// Integer Divide reg-reg
pipe_class sdiv_reg_reg(iRegI dst, iRegI src1, iRegI src2, iRegI temp, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write);
    temp  : E(write);
    src1  : R(read);
    src2  : R(read);
    temp  : R(read);
    MS    : R(38);
%}

// Integer Divide reg-imm
pipe_class sdiv_reg_imm(iRegI dst, iRegI src1, immI13 src2, iRegI temp, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : E(write);
    temp  : E(write);
    src1  : R(read);
    temp  : R(read);
    MS    : R(38);
%}

// Long Divide
pipe_class divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
    dst  : E(write)+71;
    src1 : R(read);
    src2 : R(read)+1;
    MS   : R(70);
%}

pipe_class divL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
    dst  : E(write)+71;
    src1 : R(read);
    MS   : R(70);
%}

// Floating Point Add Float
pipe_class faddF_reg_reg(regF dst, regF src1, regF src2) %{
    single_instruction;
    dst   : X(write);
    src1  : E(read);
    src2  : E(read);
    FA    : R;
%}

// Floating Point Add Double
pipe_class faddD_reg_reg(regD dst, regD src1, regD src2) %{
    single_instruction;
    dst   : X(write);
    src1  : E(read);
    src2  : E(read);
    FA    : R;
%}

// Floating Point Conditional Move based on integer flags
pipe_class int_conditional_float_move (cmpOp cmp, flagsReg cr, regF dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    cr    : R(read);
    FA    : R(2);
    BR    : R(2);
%}

// Floating Point Conditional Move based on integer flags
pipe_class int_conditional_double_move (cmpOp cmp, flagsReg cr, regD dst, regD src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    cr    : R(read);
    FA    : R(2);
    BR    : R(2);
%}

// Floating Point Multiply Float
pipe_class fmulF_reg_reg(regF dst, regF src1, regF src2) %{
    single_instruction;
    dst   : X(write);
    src1  : E(read);
    src2  : E(read);
    FM    : R;
%}

// Floating Point Multiply Double
pipe_class fmulD_reg_reg(regD dst, regD src1, regD src2) %{
    single_instruction;
    dst   : X(write);
    src1  : E(read);
    src2  : E(read);
    FM    : R;
%}

// Floating Point Divide Float
pipe_class fdivF_reg_reg(regF dst, regF src1, regF src2) %{
    single_instruction;
    dst   : X(write);
    src1  : E(read);
    src2  : E(read);
    FM    : R;
    FDIV  : C(14);
%}

// Floating Point Divide Double
pipe_class fdivD_reg_reg(regD dst, regD src1, regD src2) %{
    single_instruction;
    dst   : X(write);
    src1  : E(read);
    src2  : E(read);
    FM    : R;
    FDIV  : C(17);
%}

// Floating Point Move/Negate/Abs Float
pipe_class faddF_reg(regF dst, regF src) %{
    single_instruction;
    dst   : W(write);
    src   : E(read);
    FA    : R(1);
%}

// Floating Point Move/Negate/Abs Double
pipe_class faddD_reg(regD dst, regD src) %{
    single_instruction;
    dst   : W(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert F->D
pipe_class fcvtF2D(regD dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert I->D
pipe_class fcvtI2D(regD dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert LHi->D
pipe_class fcvtLHi2D(regD dst, regD src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert L->D
pipe_class fcvtL2D(regD dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert L->F
pipe_class fcvtL2F(regD dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert D->F
pipe_class fcvtD2F(regD dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert I->L
pipe_class fcvtI2L(regD dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert D->F
pipe_class fcvtD2I(regF dst, regD src, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : X(write)+6;
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert D->L
pipe_class fcvtD2L(regD dst, regD src, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : X(write)+6;
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert F->I
pipe_class fcvtF2I(regF dst, regF src, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : X(write)+6;
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert F->L
pipe_class fcvtF2L(regD dst, regF src, flagsReg cr) %{
    instruction_count(1); multiple_bundles;
    dst   : X(write)+6;
    src   : E(read);
    FA    : R;
%}

// Floating Point Convert I->F
pipe_class fcvtI2F(regF dst, regF src) %{
    single_instruction;
    dst   : X(write);
    src   : E(read);
    FA    : R;
%}

// Floating Point Compare
pipe_class faddF_fcc_reg_reg_zero(flagsRegF cr, regF src1, regF src2, immI0 zero) %{
    single_instruction;
    cr    : X(write);
    src1  : E(read);
    src2  : E(read);
    FA    : R;
%}

// Floating Point Compare
pipe_class faddD_fcc_reg_reg_zero(flagsRegF cr, regD src1, regD src2, immI0 zero) %{
    single_instruction;
    cr    : X(write);
    src1  : E(read);
    src2  : E(read);
    FA    : R;
%}

// Floating Add Nop
pipe_class fadd_nop() %{
    single_instruction;
    FA  : R;
%}

// Integer Store to Memory
pipe_class istore_mem_reg(memory mem, iRegI src) %{
    single_instruction;
    mem   : R(read);
    src   : C(read);
    MS    : R;
%}

// Integer Store to Memory
pipe_class istore_mem_spORreg(memory mem, sp_ptr_RegP src) %{
    single_instruction;
    mem   : R(read);
    src   : C(read);
    MS    : R;
%}

// Integer Store Zero to Memory
pipe_class istore_mem_zero(memory mem, immI0 src) %{
    single_instruction;
    mem   : R(read);
    MS    : R;
%}

// Special Stack Slot Store
pipe_class istore_stk_reg(stackSlotI stkSlot, iRegI src) %{
    single_instruction;
    stkSlot : R(read);
    src     : C(read);
    MS      : R;
%}

// Special Stack Slot Store
pipe_class lstoreI_stk_reg(stackSlotL stkSlot, iRegI src) %{
    instruction_count(2); multiple_bundles;
    stkSlot : R(read);
    src     : C(read);
    MS      : R(2);
%}

// Float Store
pipe_class fstoreF_mem_reg(memory mem, RegF src) %{
    single_instruction;
    mem : R(read);
    src : C(read);
    MS  : R;
%}

// Float Store
pipe_class fstoreF_mem_zero(memory mem, immF0 src) %{
    single_instruction;
    mem : R(read);
    MS  : R;
%}

// Double Store
pipe_class fstoreD_mem_reg(memory mem, RegD src) %{
    instruction_count(1);
    mem : R(read);
    src : C(read);
    MS  : R;
%}

// Double Store
pipe_class fstoreD_mem_zero(memory mem, immD0 src) %{
    single_instruction;
    mem : R(read);
    MS  : R;
%}

// Special Stack Slot Float Store
pipe_class fstoreF_stk_reg(stackSlotI stkSlot, RegF src) %{
    single_instruction;
    stkSlot : R(read);
    src     : C(read);
    MS      : R;
%}

// Special Stack Slot Double Store
pipe_class fstoreD_stk_reg(stackSlotI stkSlot, RegD src) %{
    single_instruction;
    stkSlot : R(read);
    src     : C(read);
    MS      : R;
%}

// Integer Load (when sign bit propagation not needed)
pipe_class iload_mem(iRegI dst, memory mem) %{
    single_instruction;
    mem : R(read);
    dst : C(write);
    MS  : R;
%}

// Integer Load from stack operand
pipe_class iload_stkD(iRegI dst, stackSlotD mem ) %{
    single_instruction;
    mem : R(read);
    dst : C(write);
    MS  : R;
%}

// Integer Load (when sign bit propagation or masking is needed)
pipe_class iload_mask_mem(iRegI dst, memory mem) %{
    single_instruction;
    mem : R(read);
    dst : M(write);
    MS  : R;
%}

// Float Load
pipe_class floadF_mem(regF dst, memory mem) %{
    single_instruction;
    mem : R(read);
    dst : M(write);
    MS  : R;
%}

// Float Load
pipe_class floadD_mem(regD dst, memory mem) %{
    instruction_count(1); multiple_bundles; // Again, unaligned argument is only multiple case
    mem : R(read);
    dst : M(write);
    MS  : R;
%}

// Float Load
pipe_class floadF_stk(regF dst, stackSlotI stkSlot) %{
    single_instruction;
    stkSlot : R(read);
    dst : M(write);
    MS  : R;
%}

// Float Load
pipe_class floadD_stk(regD dst, stackSlotI stkSlot) %{
    single_instruction;
    stkSlot : R(read);
    dst : M(write);
    MS  : R;
%}

// Memory Nop
pipe_class mem_nop() %{
    single_instruction;
    MS  : R;
%}

pipe_class sethi(iRegP dst, immI src) %{
    single_instruction;
    dst  : E(write);
    IALU : R;
%}

pipe_class loadPollP(iRegP poll) %{
    single_instruction;
    poll : R(read);
    MS   : R;
%}

pipe_class br(Universe br, label labl) %{
    single_instruction_with_delay_slot;
    BR  : R;
%}

pipe_class br_cc(Universe br, cmpOp cmp, flagsReg cr, label labl) %{
    single_instruction_with_delay_slot;
    cr    : E(read);
    BR    : R;
%}

pipe_class br_reg(Universe br, cmpOp cmp, iRegI op1, label labl) %{
    single_instruction_with_delay_slot;
    op1 : E(read);
    BR  : R;
    MS  : R;
%}

pipe_class br_fcc(Universe br, cmpOpF cc, flagsReg cr, label labl) %{
    single_instruction_with_delay_slot;
    cr    : E(read);
    BR    : R;
%}

pipe_class br_nop() %{
    single_instruction;
    BR  : R;
%}

pipe_class simple_call(method meth) %{
    instruction_count(2); multiple_bundles; force_serialization;
    fixed_latency(100);
    BR  : R(1);
    MS  : R(1);
    A0  : R(1);
%}

pipe_class compiled_call(method meth) %{
    instruction_count(1); multiple_bundles; force_serialization;
    fixed_latency(100);
    MS  : R(1);
%}

pipe_class call(method meth) %{
    instruction_count(0); multiple_bundles; force_serialization;
    fixed_latency(100);
%}

pipe_class tail_call(Universe ignore, label labl) %{
    single_instruction; has_delay_slot;
    fixed_latency(100);
    BR  : R(1);
    MS  : R(1);
%}

pipe_class ret(Universe ignore) %{
    single_instruction; has_delay_slot;
    BR  : R(1);
    MS  : R(1);
%}

pipe_class ret_poll(g3RegP poll) %{
    instruction_count(3); has_delay_slot;
    poll : E(read);
    MS   : R;
%}

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

pipe_class long_memory_op() %{
    instruction_count(0); multiple_bundles; force_serialization;
    fixed_latency(25);
    MS  : R(1);
%}

// Check-cast
pipe_class partial_subtype_check_pipe(Universe ignore, iRegP array, iRegP match ) %{
    array : R(read);
    match  : R(read);
    IALU   : R(2);
    BR     : R(2);
    MS     : R;
%}

// Convert FPU flags into +1,0,-1
pipe_class floating_cmp( iRegI dst, regF src1, regF src2 ) %{
    src1  : E(read);
    src2  : E(read);
    dst   : E(write);
    FA    : R;
    MS    : R(2);
    BR    : R(2);
%}

// Compare for p < q, and conditionally add y
pipe_class cadd_cmpltmask( iRegI p, iRegI q, iRegI y ) %{
    p     : E(read);
    q     : E(read);
    y     : E(read);
    IALU  : R(3)
%}

// Perform a compare, then move conditionally in a branch delay slot.
pipe_class min_max( iRegI src2, iRegI srcdst ) %{
    src2   : E(read);
    srcdst : E(read);
    IALU   : R;
    BR     : R;
%}

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

%}

//----------INSTRUCTIONS-------------------------------------------------------

//------------Special Stack Slot instructions - no match rules-----------------
instruct stkI_to_regF(regF dst, stackSlotI src) %{
  // No match rule to avoid chain rule match.
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDF    $src,$dst\t! stkI to regF" %}
  opcode(Assembler::ldf_op3);
  ins_encode(simple_form3_mem_reg(src, dst));
  ins_pipe(floadF_stk);
%}

instruct stkL_to_regD(regD dst, stackSlotL src) %{
  // No match rule to avoid chain rule match.
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDDF   $src,$dst\t! stkL to regD" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg(src, dst));
  ins_pipe(floadD_stk);
%}

instruct regF_to_stkI(stackSlotI dst, regF src) %{
  // No match rule to avoid chain rule match.
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STF    $src,$dst\t! regF to stkI" %}
  opcode(Assembler::stf_op3);
  ins_encode(simple_form3_mem_reg(dst, src));
  ins_pipe(fstoreF_stk_reg);
%}

instruct regD_to_stkL(stackSlotL dst, regD src) %{
  // No match rule to avoid chain rule match.
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STDF   $src,$dst\t! regD to stkL" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg(dst, src));
  ins_pipe(fstoreD_stk_reg);
%}

instruct regI_to_stkLHi(stackSlotL dst, iRegI src) %{
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST*2);
  size(8);
  format %{ "STW    $src,$dst.hi\t! long\n\t"
            "STW    R_G0,$dst.lo" %}
  opcode(Assembler::stw_op3);
  ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, R_G0));
  ins_pipe(lstoreI_stk_reg);
%}

instruct regL_to_stkD(stackSlotD dst, iRegL src) %{
  // No match rule to avoid chain rule match.
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $src,$dst\t! regL to stkD" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_stk_reg);
%}

//---------- Chain stack slots between similar types --------

// Load integer from stack slot
instruct stkI_to_regI( iRegI dst, stackSlotI src ) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUW   $src,$dst\t!stk" %}
  opcode(Assembler::lduw_op3);
  ins_encode(simple_form3_mem_reg( src, dst ) );
  ins_pipe(iload_mem);
%}

// Store integer to stack slot
instruct regI_to_stkI( stackSlotI dst, iRegI src ) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STW    $src,$dst\t!stk" %}
  opcode(Assembler::stw_op3);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_reg);
%}

// Load long from stack slot
instruct stkL_to_regL( iRegL dst, stackSlotL src ) %{
  match(Set dst src);

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDX    $src,$dst\t! long" %}
  opcode(Assembler::ldx_op3);
  ins_encode(simple_form3_mem_reg( src, dst ) );
  ins_pipe(iload_mem);
%}

// Store long to stack slot
instruct regL_to_stkL(stackSlotL dst, iRegL src) %{
  match(Set dst src);

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $src,$dst\t! long" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_reg);
%}

#ifdef _LP64
// Load pointer from stack slot, 64-bit encoding
instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDX    $src,$dst\t!ptr" %}
  opcode(Assembler::ldx_op3);
  ins_encode(simple_form3_mem_reg( src, dst ) );
  ins_pipe(iload_mem);
%}

// Store pointer to stack slot
instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $src,$dst\t!ptr" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_reg);
%}
#else // _LP64
// Load pointer from stack slot, 32-bit encoding
instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  format %{ "LDUW   $src,$dst\t!ptr" %}
  opcode(Assembler::lduw_op3, Assembler::ldst_op);
  ins_encode(simple_form3_mem_reg( src, dst ) );
  ins_pipe(iload_mem);
%}

// Store pointer to stack slot
instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  format %{ "STW    $src,$dst\t!ptr" %}
  opcode(Assembler::stw_op3, Assembler::ldst_op);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_reg);
%}
#endif // _LP64

//------------Special Nop instructions for bundling - no match rules-----------
// Nop using the A0 functional unit
instruct Nop_A0() %{
  ins_cost(0);

  format %{ "NOP    ! Alu Pipeline" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form2_nop() );
  ins_pipe(ialu_nop_A0);
%}

// Nop using the A1 functional unit
instruct Nop_A1( ) %{
  ins_cost(0);

  format %{ "NOP    ! Alu Pipeline" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form2_nop() );
  ins_pipe(ialu_nop_A1);
%}

// Nop using the memory functional unit
instruct Nop_MS( ) %{
  ins_cost(0);

  format %{ "NOP    ! Memory Pipeline" %}
  ins_encode( emit_mem_nop );
  ins_pipe(mem_nop);
%}

// Nop using the floating add functional unit
instruct Nop_FA( ) %{
  ins_cost(0);

  format %{ "NOP    ! Floating Add Pipeline" %}
  ins_encode( emit_fadd_nop );
  ins_pipe(fadd_nop);
%}

// Nop using the branch functional unit
instruct Nop_BR( ) %{
  ins_cost(0);

  format %{ "NOP    ! Branch Pipeline" %}
  ins_encode( emit_br_nop );
  ins_pipe(br_nop);
%}

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

  size(4);
  format %{ "LDSB   $mem,$dst\t! byte" %}
  ins_encode %{
    __ ldsb($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Byte (8bit signed) into a Long Register
instruct loadB2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadB mem)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDSB   $mem,$dst\t! byte -> long" %}
  ins_encode %{
    __ ldsb($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Unsigned Byte (8bit UNsigned) into an int reg
instruct loadUB(iRegI dst, memory mem) %{
  match(Set dst (LoadUB mem));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUB   $mem,$dst\t! ubyte" %}
  ins_encode %{
    __ ldub($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Byte (8bit UNsigned) into a Long Register
instruct loadUB2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadUB mem)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUB   $mem,$dst\t! ubyte -> long" %}
  ins_encode %{
    __ ldub($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Byte (8 bit UNsigned) with 8-bit mask into Long Register
instruct loadUB2L_immI8(iRegL dst, memory mem, immI8 mask) %{
  match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
  ins_cost(MEMORY_REF_COST + DEFAULT_COST);

  size(2*4);
  format %{ "LDUB   $mem,$dst\t# ubyte & 8-bit mask -> long\n\t"
            "AND    $dst,$mask,$dst" %}
  ins_encode %{
    __ ldub($mem$$Address, $dst$$Register);
    __ and3($dst$$Register, $mask$$constant, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

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

  size(4);
  format %{ "LDSH   $mem,$dst\t! short" %}
  ins_encode %{
    __ ldsh($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Short (16 bit signed) to Byte (8 bit signed)
instruct loadS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
  ins_cost(MEMORY_REF_COST);

  size(4);

  format %{ "LDSB   $mem+1,$dst\t! short -> byte" %}
  ins_encode %{
    __ ldsb($mem$$Address, $dst$$Register, 1);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Short (16bit signed) into a Long Register
instruct loadS2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadS mem)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDSH   $mem,$dst\t! short -> long" %}
  ins_encode %{
    __ ldsh($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Unsigned Short/Char (16bit UNsigned)
instruct loadUS(iRegI dst, memory mem) %{
  match(Set dst (LoadUS mem));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUH   $mem,$dst\t! ushort/char" %}
  ins_encode %{
    __ lduh($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
instruct loadUS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDSB   $mem+1,$dst\t! ushort -> byte" %}
  ins_encode %{
    __ ldsb($mem$$Address, $dst$$Register, 1);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Unsigned Short/Char (16bit UNsigned) into a Long Register
instruct loadUS2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadUS mem)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUH   $mem,$dst\t! ushort/char -> long" %}
  ins_encode %{
    __ lduh($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Short/Char (16bit UNsigned) with mask 0xFF into a Long Register
instruct loadUS2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
  match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUB   $mem+1,$dst\t! ushort/char & 0xFF -> long" %}
  ins_encode %{
    __ ldub($mem$$Address, $dst$$Register, 1);  // LSB is index+1 on BE
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Short/Char (16bit UNsigned) with a 13-bit mask into a Long Register
instruct loadUS2L_immI13(iRegL dst, memory mem, immI13 mask) %{
  match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
  ins_cost(MEMORY_REF_COST + DEFAULT_COST);

  size(2*4);
  format %{ "LDUH   $mem,$dst\t! ushort/char & 13-bit mask -> long\n\t"
            "AND    $dst,$mask,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ lduh($mem$$Address, Rdst);
    __ and3(Rdst, $mask$$constant, Rdst);
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Short/Char (16bit UNsigned) with a 16-bit mask into a Long Register
instruct loadUS2L_immI16(iRegL dst, memory mem, immI16 mask, iRegL tmp) %{
  match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
  effect(TEMP dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);

  size((3+1)*4);  // set may use two instructions.
  format %{ "LDUH   $mem,$dst\t! ushort/char & 16-bit mask -> long\n\t"
            "SET    $mask,$tmp\n\t"
            "AND    $dst,$tmp,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    __ lduh($mem$$Address, Rdst);
    __ set($mask$$constant, Rtmp);
    __ and3(Rdst, Rtmp, Rdst);
  %}
  ins_pipe(iload_mem);
%}

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

  size(4);
  format %{ "LDUW   $mem,$dst\t! int" %}
  ins_encode %{
    __ lduw($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Integer to Byte (8 bit signed)
instruct loadI2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
  match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
  ins_cost(MEMORY_REF_COST);

  size(4);

  format %{ "LDSB   $mem+3,$dst\t! int -> byte" %}
  ins_encode %{
    __ ldsb($mem$$Address, $dst$$Register, 3);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Integer to Unsigned Byte (8 bit UNsigned)
instruct loadI2UB(iRegI dst, indOffset13m7 mem, immI_255 mask) %{
  match(Set dst (AndI (LoadI mem) mask));
  ins_cost(MEMORY_REF_COST);

  size(4);

  format %{ "LDUB   $mem+3,$dst\t! int -> ubyte" %}
  ins_encode %{
    __ ldub($mem$$Address, $dst$$Register, 3);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Integer to Short (16 bit signed)
instruct loadI2S(iRegI dst, indOffset13m7 mem, immI_16 sixteen) %{
  match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
  ins_cost(MEMORY_REF_COST);

  size(4);

  format %{ "LDSH   $mem+2,$dst\t! int -> short" %}
  ins_encode %{
    __ ldsh($mem$$Address, $dst$$Register, 2);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Integer to Unsigned Short (16 bit UNsigned)
instruct loadI2US(iRegI dst, indOffset13m7 mem, immI_65535 mask) %{
  match(Set dst (AndI (LoadI mem) mask));
  ins_cost(MEMORY_REF_COST);

  size(4);

  format %{ "LDUH   $mem+2,$dst\t! int -> ushort/char" %}
  ins_encode %{
    __ lduh($mem$$Address, $dst$$Register, 2);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Integer into a Long Register
instruct loadI2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadI mem)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDSW   $mem,$dst\t! int -> long" %}
  ins_encode %{
    __ ldsw($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mask_mem);
%}

// Load Integer with mask 0xFF into a Long Register
instruct loadI2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUB   $mem+3,$dst\t! int & 0xFF -> long" %}
  ins_encode %{
    __ ldub($mem$$Address, $dst$$Register, 3);  // LSB is index+3 on BE
  %}
  ins_pipe(iload_mem);
%}

// Load Integer with mask 0xFFFF into a Long Register
instruct loadI2L_immI_65535(iRegL dst, indOffset13m7 mem, immI_65535 mask) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUH   $mem+2,$dst\t! int & 0xFFFF -> long" %}
  ins_encode %{
    __ lduh($mem$$Address, $dst$$Register, 2);  // LSW is index+2 on BE
  %}
  ins_pipe(iload_mem);
%}

// Load Integer with a 13-bit mask into a Long Register
instruct loadI2L_immI13(iRegL dst, memory mem, immI13 mask) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  ins_cost(MEMORY_REF_COST + DEFAULT_COST);

  size(2*4);
  format %{ "LDUW   $mem,$dst\t! int & 13-bit mask -> long\n\t"
            "AND    $dst,$mask,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    __ lduw($mem$$Address, Rdst);
    __ and3(Rdst, $mask$$constant, Rdst);
  %}
  ins_pipe(iload_mem);
%}

// Load Integer with a 32-bit mask into a Long Register
instruct loadI2L_immI(iRegL dst, memory mem, immI mask, iRegL tmp) %{
  match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
  effect(TEMP dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);

  size((3+1)*4);  // set may use two instructions.
  format %{ "LDUW   $mem,$dst\t! int & 32-bit mask -> long\n\t"
            "SET    $mask,$tmp\n\t"
            "AND    $dst,$tmp,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    __ lduw($mem$$Address, Rdst);
    __ set($mask$$constant, Rtmp);
    __ and3(Rdst, Rtmp, Rdst);
  %}
  ins_pipe(iload_mem);
%}

// Load Unsigned Integer into a Long Register
instruct loadUI2L(iRegL dst, memory mem) %{
  match(Set dst (LoadUI2L mem));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUW   $mem,$dst\t! uint -> long" %}
  ins_encode %{
    __ lduw($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Long - aligned
instruct loadL(iRegL dst, memory mem ) %{
  match(Set dst (LoadL mem));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDX    $mem,$dst\t! long" %}
  ins_encode %{
    __ ldx($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Long - UNaligned
instruct loadL_unaligned(iRegL dst, memory mem, o7RegI tmp) %{
  match(Set dst (LoadL_unaligned mem));
  effect(KILL tmp);
  ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
  size(16);
  format %{ "LDUW   $mem+4,R_O7\t! misaligned long\n"
          "\tLDUW   $mem  ,$dst\n"
          "\tSLLX   #32, $dst, $dst\n"
          "\tOR     $dst, R_O7, $dst" %}
  opcode(Assembler::lduw_op3);
  ins_encode(form3_mem_reg_long_unaligned_marshal( mem, dst ));
  ins_pipe(iload_mem);
%}

// Load Aligned Packed Byte into a Double Register
instruct loadA8B(regD dst, memory mem) %{
  match(Set dst (Load8B mem));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDDF   $mem,$dst\t! packed8B" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadD_mem);
%}

// Load Aligned Packed Char into a Double Register
instruct loadA4C(regD dst, memory mem) %{
  match(Set dst (Load4C mem));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDDF   $mem,$dst\t! packed4C" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadD_mem);
%}

// Load Aligned Packed Short into a Double Register
instruct loadA4S(regD dst, memory mem) %{
  match(Set dst (Load4S mem));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDDF   $mem,$dst\t! packed4S" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadD_mem);
%}

// Load Aligned Packed Int into a Double Register
instruct loadA2I(regD dst, memory mem) %{
  match(Set dst (Load2I mem));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDDF   $mem,$dst\t! packed2I" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadD_mem);
%}

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

  size(4);
  format %{ "LDUW   $mem,$dst\t! range" %}
  opcode(Assembler::lduw_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(iload_mem);
%}

// Load Integer into %f register (for fitos/fitod)
instruct loadI_freg(regF dst, memory mem) %{
  match(Set dst (LoadI mem));
  ins_cost(MEMORY_REF_COST);
  size(4);

  format %{ "LDF    $mem,$dst\t! for fitos/fitod" %}
  opcode(Assembler::ldf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadF_mem);
%}

// Load Pointer
instruct loadP(iRegP dst, memory mem) %{
  match(Set dst (LoadP mem));
  ins_cost(MEMORY_REF_COST);
  size(4);

#ifndef _LP64
  format %{ "LDUW   $mem,$dst\t! ptr" %}
  ins_encode %{
    __ lduw($mem$$Address, $dst$$Register);
  %}
#else
  format %{ "LDX    $mem,$dst\t! ptr" %}
  ins_encode %{
    __ ldx($mem$$Address, $dst$$Register);
  %}
#endif
  ins_pipe(iload_mem);
%}

// Load Compressed Pointer
instruct loadN(iRegN dst, memory mem) %{
  match(Set dst (LoadN mem));
  ins_cost(MEMORY_REF_COST);
  size(4);

  format %{ "LDUW   $mem,$dst\t! compressed ptr" %}
  ins_encode %{
    __ lduw($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Klass Pointer
instruct loadKlass(iRegP dst, memory mem) %{
  match(Set dst (LoadKlass mem));
  ins_cost(MEMORY_REF_COST);
  size(4);

#ifndef _LP64
  format %{ "LDUW   $mem,$dst\t! klass ptr" %}
  ins_encode %{
    __ lduw($mem$$Address, $dst$$Register);
  %}
#else
  format %{ "LDX    $mem,$dst\t! klass ptr" %}
  ins_encode %{
    __ ldx($mem$$Address, $dst$$Register);
  %}
#endif
  ins_pipe(iload_mem);
%}

// Load narrow Klass Pointer
instruct loadNKlass(iRegN dst, memory mem) %{
  match(Set dst (LoadNKlass mem));
  ins_cost(MEMORY_REF_COST);
  size(4);

  format %{ "LDUW   $mem,$dst\t! compressed klass ptr" %}
  ins_encode %{
    __ lduw($mem$$Address, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Double
instruct loadD(regD dst, memory mem) %{
  match(Set dst (LoadD mem));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDDF   $mem,$dst" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadD_mem);
%}

// Load Double - UNaligned
instruct loadD_unaligned(regD_low dst, memory mem ) %{
  match(Set dst (LoadD_unaligned mem));
  ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
  size(8);
  format %{ "LDF    $mem  ,$dst.hi\t! misaligned double\n"
          "\tLDF    $mem+4,$dst.lo\t!" %}
  opcode(Assembler::ldf_op3);
  ins_encode( form3_mem_reg_double_unaligned( mem, dst ));
  ins_pipe(iload_mem);
%}

// Load Float
instruct loadF(regF dst, memory mem) %{
  match(Set dst (LoadF mem));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDF    $mem,$dst" %}
  opcode(Assembler::ldf_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(floadF_mem);
%}

// Load Constant
instruct loadConI( iRegI dst, immI src ) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST * 3/2);
  format %{ "SET    $src,$dst" %}
  ins_encode( Set32(src, dst) );
  ins_pipe(ialu_hi_lo_reg);
%}

instruct loadConI13( iRegI dst, immI13 src ) %{
  match(Set dst src);

  size(4);
  format %{ "MOV    $src,$dst" %}
  ins_encode( Set13( src, dst ) );
  ins_pipe(ialu_imm);
%}

instruct loadConP(iRegP dst, immP src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST * 3/2);
  format %{ "SET    $src,$dst\t!ptr" %}
  // This rule does not use "expand" unlike loadConI because then
  // the result type is not known to be an Oop.  An ADLC
  // enhancement will be needed to make that work - not worth it!

  ins_encode( SetPtr( src, dst ) );
  ins_pipe(loadConP);

%}

instruct loadConP0(iRegP dst, immP0 src) %{
  match(Set dst src);

  size(4);
  format %{ "CLR    $dst\t!ptr" %}
  ins_encode( SetNull( dst ) );
  ins_pipe(ialu_imm);
%}

instruct loadConP_poll(iRegP dst, immP_poll src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST);
  format %{ "SET    $src,$dst\t!ptr" %}
  ins_encode %{
    AddressLiteral polling_page(os::get_polling_page());
    __ sethi(polling_page, reg_to_register_object($dst$$reg));
  %}
  ins_pipe(loadConP_poll);
%}

instruct loadConN0(iRegN dst, immN0 src) %{
  match(Set dst src);

  size(4);
  format %{ "CLR    $dst\t! compressed NULL ptr" %}
  ins_encode( SetNull( dst ) );
  ins_pipe(ialu_imm);
%}

instruct loadConN(iRegN dst, immN src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST * 3/2);
  format %{ "SET    $src,$dst\t! compressed ptr" %}
  ins_encode %{
    Register dst = $dst$$Register;
    __ set_narrow_oop((jobject)$src$$constant, dst);
  %}
  ins_pipe(ialu_hi_lo_reg);
%}

instruct loadConL(iRegL dst, immL src, o7RegL tmp) %{
  // %%% maybe this should work like loadConD
  match(Set dst src);
  effect(KILL tmp);
  ins_cost(DEFAULT_COST * 4);
  format %{ "SET64   $src,$dst KILL $tmp\t! long" %}
  ins_encode( LdImmL(src, dst, tmp) );
  ins_pipe(loadConL);
%}

instruct loadConL0( iRegL dst, immL0 src ) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "CLR    $dst\t! long" %}
  ins_encode( Set13( src, dst ) );
  ins_pipe(ialu_imm);
%}

instruct loadConL13( iRegL dst, immL13 src ) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST * 2);

  size(4);
  format %{ "MOV    $src,$dst\t! long" %}
  ins_encode( Set13( src, dst ) );
  ins_pipe(ialu_imm);
%}

instruct loadConF(regF dst, immF src, o7RegP tmp) %{
  match(Set dst src);
  effect(KILL tmp);

#ifdef _LP64
  size(8*4);
#else
  size(2*4);
#endif

  format %{ "SETHI  hi(&$src),$tmp\t!get float $src from table\n\t"
            "LDF    [$tmp+lo(&$src)],$dst" %}
  ins_encode %{
    address float_address = __ float_constant($src$$constant);
    RelocationHolder rspec = internal_word_Relocation::spec(float_address);
    AddressLiteral addrlit(float_address, rspec);

    __ sethi(addrlit, $tmp$$Register);
    __ ldf(FloatRegisterImpl::S, $tmp$$Register, addrlit.low10(), $dst$$FloatRegister, rspec);
  %}
  ins_pipe(loadConFD);
%}

instruct loadConD(regD dst, immD src, o7RegP tmp) %{
  match(Set dst src);
  effect(KILL tmp);

#ifdef _LP64
  size(8*4);
#else
  size(2*4);
#endif

  format %{ "SETHI  hi(&$src),$tmp\t!get double $src from table\n\t"
            "LDDF   [$tmp+lo(&$src)],$dst" %}
  ins_encode %{
    address double_address = __ double_constant($src$$constant);
    RelocationHolder rspec = internal_word_Relocation::spec(double_address);
    AddressLiteral addrlit(double_address, rspec);

    __ sethi(addrlit, $tmp$$Register);
    // XXX This is a quick fix for 6833573.
    //__ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), $dst$$FloatRegister, rspec);
    __ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), as_DoubleFloatRegister($dst$$reg), rspec);
  %}
  ins_pipe(loadConFD);
%}

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

instruct prefetchr( memory mem ) %{
  match( PrefetchRead mem );
  ins_cost(MEMORY_REF_COST);

  format %{ "PREFETCH $mem,0\t! Prefetch read-many" %}
  opcode(Assembler::prefetch_op3);
  ins_encode( form3_mem_prefetch_read( mem ) );
  ins_pipe(iload_mem);
%}

instruct prefetchw( memory mem ) %{
  predicate(AllocatePrefetchStyle != 3 );
  match( PrefetchWrite mem );
  ins_cost(MEMORY_REF_COST);

  format %{ "PREFETCH $mem,2\t! Prefetch write-many (and read)" %}
  opcode(Assembler::prefetch_op3);
  ins_encode( form3_mem_prefetch_write( mem ) );
  ins_pipe(iload_mem);
%}

// Use BIS instruction to prefetch.
instruct prefetchw_bis( memory mem ) %{
  predicate(AllocatePrefetchStyle == 3);
  match( PrefetchWrite mem );
  ins_cost(MEMORY_REF_COST);

  format %{ "STXA   G0,$mem\t! // Block initializing store" %}
  ins_encode %{
     Register base = as_Register($mem$$base);
     int disp = $mem$$disp;
     if (disp != 0) {
       __ add(base, AllocatePrefetchStepSize, base);
     }
     __ stxa(G0, base, G0, ASI_BLK_INIT_QUAD_LDD_P);
  %}
  ins_pipe(istore_mem_reg);
%}

//----------Store Instructions-------------------------------------------------
// Store Byte
instruct storeB(memory mem, iRegI src) %{
  match(Set mem (StoreB mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STB    $src,$mem\t! byte" %}
  opcode(Assembler::stb_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(istore_mem_reg);
%}

instruct storeB0(memory mem, immI0 src) %{
  match(Set mem (StoreB mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STB    $src,$mem\t! byte" %}
  opcode(Assembler::stb_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(istore_mem_zero);
%}

instruct storeCM0(memory mem, immI0 src) %{
  match(Set mem (StoreCM mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STB    $src,$mem\t! CMS card-mark byte 0" %}
  opcode(Assembler::stb_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(istore_mem_zero);
%}

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

  size(4);
  format %{ "STH    $src,$mem\t! short" %}
  opcode(Assembler::sth_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(istore_mem_reg);
%}

instruct storeC0(memory mem, immI0 src) %{
  match(Set mem (StoreC mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STH    $src,$mem\t! short" %}
  opcode(Assembler::sth_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(istore_mem_zero);
%}

// Store Integer
instruct storeI(memory mem, iRegI src) %{
  match(Set mem (StoreI mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STW    $src,$mem" %}
  opcode(Assembler::stw_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(istore_mem_reg);
%}

// Store Long
instruct storeL(memory mem, iRegL src) %{
  match(Set mem (StoreL mem src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $src,$mem\t! long" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(istore_mem_reg);
%}

instruct storeI0(memory mem, immI0 src) %{
  match(Set mem (StoreI mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STW    $src,$mem" %}
  opcode(Assembler::stw_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(istore_mem_zero);
%}

instruct storeL0(memory mem, immL0 src) %{
  match(Set mem (StoreL mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STX    $src,$mem" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(istore_mem_zero);
%}

// Store Integer from float register (used after fstoi)
instruct storeI_Freg(memory mem, regF src) %{
  match(Set mem (StoreI mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STF    $src,$mem\t! after fstoi/fdtoi" %}
  opcode(Assembler::stf_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(fstoreF_mem_reg);
%}

// Store Pointer
instruct storeP(memory dst, sp_ptr_RegP src) %{
  match(Set dst (StoreP dst src));
  ins_cost(MEMORY_REF_COST);
  size(4);

#ifndef _LP64
  format %{ "STW    $src,$dst\t! ptr" %}
  opcode(Assembler::stw_op3, 0, REGP_OP);
#else
  format %{ "STX    $src,$dst\t! ptr" %}
  opcode(Assembler::stx_op3, 0, REGP_OP);
#endif
  ins_encode( form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_spORreg);
%}

instruct storeP0(memory dst, immP0 src) %{
  match(Set dst (StoreP dst src));
  ins_cost(MEMORY_REF_COST);
  size(4);

#ifndef _LP64
  format %{ "STW    $src,$dst\t! ptr" %}
  opcode(Assembler::stw_op3, 0, REGP_OP);
#else
  format %{ "STX    $src,$dst\t! ptr" %}
  opcode(Assembler::stx_op3, 0, REGP_OP);
#endif
  ins_encode( form3_mem_reg( dst, R_G0 ) );
  ins_pipe(istore_mem_zero);
%}

// Store Compressed Pointer
instruct storeN(memory dst, iRegN src) %{
   match(Set dst (StoreN dst src));
   ins_cost(MEMORY_REF_COST);
   size(4);

   format %{ "STW    $src,$dst\t! compressed ptr" %}
   ins_encode %{
     Register base = as_Register($dst$$base);
     Register index = as_Register($dst$$index);
     Register src = $src$$Register;
     if (index != G0) {
       __ stw(src, base, index);
     } else {
       __ stw(src, base, $dst$$disp);
     }
   %}
   ins_pipe(istore_mem_spORreg);
%}

instruct storeN0(memory dst, immN0 src) %{
   match(Set dst (StoreN dst src));
   ins_cost(MEMORY_REF_COST);
   size(4);

   format %{ "STW    $src,$dst\t! compressed ptr" %}
   ins_encode %{
     Register base = as_Register($dst$$base);
     Register index = as_Register($dst$$index);
     if (index != G0) {
       __ stw(0, base, index);
     } else {
       __ stw(0, base, $dst$$disp);
     }
   %}
   ins_pipe(istore_mem_zero);
%}

// Store Double
instruct storeD( memory mem, regD src) %{
  match(Set mem (StoreD mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STDF   $src,$mem" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(fstoreD_mem_reg);
%}

instruct storeD0( memory mem, immD0 src) %{
  match(Set mem (StoreD mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STX    $src,$mem" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(fstoreD_mem_zero);
%}

// Store Float
instruct storeF( memory mem, regF src) %{
  match(Set mem (StoreF mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STF    $src,$mem" %}
  opcode(Assembler::stf_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(fstoreF_mem_reg);
%}

instruct storeF0( memory mem, immF0 src) %{
  match(Set mem (StoreF mem src));
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STW    $src,$mem\t! storeF0" %}
  opcode(Assembler::stw_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(fstoreF_mem_zero);
%}

// Store Aligned Packed Bytes in Double register to memory
instruct storeA8B(memory mem, regD src) %{
  match(Set mem (Store8B mem src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STDF   $src,$mem\t! packed8B" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(fstoreD_mem_reg);
%}

// Convert oop pointer into compressed form
instruct encodeHeapOop(iRegN dst, iRegP src) %{
  predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
  match(Set dst (EncodeP src));
  format %{ "encode_heap_oop $src, $dst" %}
  ins_encode %{
    __ encode_heap_oop($src$$Register, $dst$$Register);
  %}
  ins_pipe(ialu_reg);
%}

instruct encodeHeapOop_not_null(iRegN dst, iRegP src) %{
  predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
  match(Set dst (EncodeP src));
  format %{ "encode_heap_oop_not_null $src, $dst" %}
  ins_encode %{
    __ encode_heap_oop_not_null($src$$Register, $dst$$Register);
  %}
  ins_pipe(ialu_reg);
%}

instruct decodeHeapOop(iRegP dst, iRegN src) %{
  predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
            n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
  match(Set dst (DecodeN src));
  format %{ "decode_heap_oop $src, $dst" %}
  ins_encode %{
    __ decode_heap_oop($src$$Register, $dst$$Register);
  %}
  ins_pipe(ialu_reg);
%}

instruct decodeHeapOop_not_null(iRegP dst, iRegN src) %{
  predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
            n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
  match(Set dst (DecodeN src));
  format %{ "decode_heap_oop_not_null $src, $dst" %}
  ins_encode %{
    __ decode_heap_oop_not_null($src$$Register, $dst$$Register);
  %}
  ins_pipe(ialu_reg);
%}


// Store Zero into Aligned Packed Bytes
instruct storeA8B0(memory mem, immI0 zero) %{
  match(Set mem (Store8B mem zero));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $zero,$mem\t! packed8B" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(fstoreD_mem_zero);
%}

// Store Aligned Packed Chars/Shorts in Double register to memory
instruct storeA4C(memory mem, regD src) %{
  match(Set mem (Store4C mem src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STDF   $src,$mem\t! packed4C" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(fstoreD_mem_reg);
%}

// Store Zero into Aligned Packed Chars/Shorts
instruct storeA4C0(memory mem, immI0 zero) %{
  match(Set mem (Store4C mem (Replicate4C zero)));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $zero,$mem\t! packed4C" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(fstoreD_mem_zero);
%}

// Store Aligned Packed Ints in Double register to memory
instruct storeA2I(memory mem, regD src) %{
  match(Set mem (Store2I mem src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STDF   $src,$mem\t! packed2I" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg( mem, src ) );
  ins_pipe(fstoreD_mem_reg);
%}

// Store Zero into Aligned Packed Ints
instruct storeA2I0(memory mem, immI0 zero) %{
  match(Set mem (Store2I mem zero));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STX    $zero,$mem\t! packed2I" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
  ins_pipe(fstoreD_mem_zero);
%}


//----------MemBar Instructions-----------------------------------------------
// Memory barrier flavors

instruct membar_acquire() %{
  match(MemBarAcquire);
  ins_cost(4*MEMORY_REF_COST);

  size(0);
  format %{ "MEMBAR-acquire" %}
  ins_encode( enc_membar_acquire );
  ins_pipe(long_memory_op);
%}

instruct membar_acquire_lock() %{
  match(MemBarAcquire);
  predicate(Matcher::prior_fast_lock(n));
  ins_cost(0);

  size(0);
  format %{ "!MEMBAR-acquire (CAS in prior FastLock so empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

instruct membar_release() %{
  match(MemBarRelease);
  ins_cost(4*MEMORY_REF_COST);

  size(0);
  format %{ "MEMBAR-release" %}
  ins_encode( enc_membar_release );
  ins_pipe(long_memory_op);
%}

instruct membar_release_lock() %{
  match(MemBarRelease);
  predicate(Matcher::post_fast_unlock(n));
  ins_cost(0);

  size(0);
  format %{ "!MEMBAR-release (CAS in succeeding FastUnlock so empty encoding)" %}
  ins_encode( );
  ins_pipe(empty);
%}

instruct membar_volatile() %{
  match(MemBarVolatile);
  ins_cost(4*MEMORY_REF_COST);

  size(4);
  format %{ "MEMBAR-volatile" %}
  ins_encode( enc_membar_volatile );
  ins_pipe(long_memory_op);
%}

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);
%}

//----------Register Move Instructions-----------------------------------------
instruct roundDouble_nop(regD dst) %{
  match(Set dst (RoundDouble dst));
  ins_cost(0);
  // SPARC results are already "rounded" (i.e., normal-format IEEE)
  ins_encode( );
  ins_pipe(empty);
%}


instruct roundFloat_nop(regF dst) %{
  match(Set dst (RoundFloat dst));
  ins_cost(0);
  // SPARC results are already "rounded" (i.e., normal-format IEEE)
  ins_encode( );
  ins_pipe(empty);
%}


// Cast Index to Pointer for unsafe natives
instruct castX2P(iRegX src, iRegP dst) %{
  match(Set dst (CastX2P src));

  format %{ "MOV    $src,$dst\t! IntX->Ptr" %}
  ins_encode( form3_g0_rs2_rd_move( src, dst ) );
  ins_pipe(ialu_reg);
%}

// Cast Pointer to Index for unsafe natives
instruct castP2X(iRegP src, iRegX dst) %{
  match(Set dst (CastP2X src));

  format %{ "MOV    $src,$dst\t! Ptr->IntX" %}
  ins_encode( form3_g0_rs2_rd_move( src, dst ) );
  ins_pipe(ialu_reg);
%}

instruct stfSSD(stackSlotD stkSlot, regD src) %{
  // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
  match(Set stkSlot src);   // chain rule
  ins_cost(MEMORY_REF_COST);
  format %{ "STDF   $src,$stkSlot\t!stk" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg(stkSlot, src));
  ins_pipe(fstoreD_stk_reg);
%}

instruct ldfSSD(regD dst, stackSlotD stkSlot) %{
  // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
  match(Set dst stkSlot);   // chain rule
  ins_cost(MEMORY_REF_COST);
  format %{ "LDDF   $stkSlot,$dst\t!stk" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg(stkSlot, dst));
  ins_pipe(floadD_stk);
%}

instruct stfSSF(stackSlotF stkSlot, regF src) %{
  // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
  match(Set stkSlot src);   // chain rule
  ins_cost(MEMORY_REF_COST);
  format %{ "STF   $src,$stkSlot\t!stk" %}
  opcode(Assembler::stf_op3);
  ins_encode(simple_form3_mem_reg(stkSlot, src));
  ins_pipe(fstoreF_stk_reg);
%}

//----------Conditional Move---------------------------------------------------
// Conditional move
instruct cmovIP_reg(cmpOpP cmp, flagsRegP pcc, iRegI dst, iRegI src) %{
  match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp $pcc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovIP_imm(cmpOpP cmp, flagsRegP pcc, iRegI dst, immI11 src) %{
  match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
  ins_cost(140);
  format %{ "MOV$cmp $pcc,$src,$dst" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovII_reg(cmpOp cmp, flagsReg icc, iRegI dst, iRegI src) %{
  match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovII_imm(cmpOp cmp, flagsReg icc, iRegI dst, immI11 src) %{
  match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
  ins_cost(140);
  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovIIu_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{
  match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovIIu_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{
  match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
  ins_cost(140);
  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovIF_reg(cmpOpF cmp, flagsRegF fcc, iRegI dst, iRegI src) %{
  match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp $fcc,$src,$dst" %}
  ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
  ins_pipe(ialu_reg);
%}

instruct cmovIF_imm(cmpOpF cmp, flagsRegF fcc, iRegI dst, immI11 src) %{
  match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
  ins_cost(140);
  size(4);
  format %{ "MOV$cmp $fcc,$src,$dst" %}
  ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
  ins_pipe(ialu_imm);
%}

// Conditional move for RegN. Only cmov(reg,reg).
instruct cmovNP_reg(cmpOpP cmp, flagsRegP pcc, iRegN dst, iRegN src) %{
  match(Set dst (CMoveN (Binary cmp pcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp $pcc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_reg);
%}

// This instruction also works with CmpN so we don't need cmovNN_reg.
instruct cmovNI_reg(cmpOp cmp, flagsReg icc, iRegN dst, iRegN src) %{
  match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}

// This instruction also works with CmpN so we don't need cmovNN_reg.
instruct cmovNIu_reg(cmpOpU cmp, flagsRegU icc, iRegN dst, iRegN src) %{
  match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovNF_reg(cmpOpF cmp, flagsRegF fcc, iRegN dst, iRegN src) %{
  match(Set dst (CMoveN (Binary cmp fcc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp $fcc,$src,$dst" %}
  ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
  ins_pipe(ialu_reg);
%}

// Conditional move
instruct cmovPP_reg(cmpOpP cmp, flagsRegP pcc, iRegP dst, iRegP src) %{
  match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovPP_imm(cmpOpP cmp, flagsRegP pcc, iRegP dst, immP0 src) %{
  match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
  ins_cost(140);
  format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_imm);
%}

// This instruction also works with CmpN so we don't need cmovPN_reg.
instruct cmovPI_reg(cmpOp cmp, flagsReg icc, iRegP dst, iRegP src) %{
  match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovPIu_reg(cmpOpU cmp, flagsRegU icc, iRegP dst, iRegP src) %{
  match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovPI_imm(cmpOp cmp, flagsReg icc, iRegP dst, immP0 src) %{
  match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
  ins_cost(140);

  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovPIu_imm(cmpOpU cmp, flagsRegU icc, iRegP dst, immP0 src) %{
  match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
  ins_cost(140);

  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovPF_reg(cmpOpF cmp, flagsRegF fcc, iRegP dst, iRegP src) %{
  match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "MOV$cmp $fcc,$src,$dst" %}
  ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
  ins_pipe(ialu_imm);
%}

instruct cmovPF_imm(cmpOpF cmp, flagsRegF fcc, iRegP dst, immP0 src) %{
  match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
  ins_cost(140);
  size(4);
  format %{ "MOV$cmp $fcc,$src,$dst" %}
  ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
  ins_pipe(ialu_imm);
%}

// Conditional move
instruct cmovFP_reg(cmpOpP cmp, flagsRegP pcc, regF dst, regF src) %{
  match(Set dst (CMoveF (Binary cmp pcc) (Binary dst src)));
  ins_cost(150);
  opcode(0x101);
  format %{ "FMOVD$cmp $pcc,$src,$dst" %}
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(int_conditional_float_move);
%}

instruct cmovFI_reg(cmpOp cmp, flagsReg icc, regF dst, regF src) %{
  match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "FMOVS$cmp $icc,$src,$dst" %}
  opcode(0x101);
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(int_conditional_float_move);
%}

instruct cmovFIu_reg(cmpOpU cmp, flagsRegU icc, regF dst, regF src) %{
  match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "FMOVS$cmp $icc,$src,$dst" %}
  opcode(0x101);
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(int_conditional_float_move);
%}

// Conditional move,
instruct cmovFF_reg(cmpOpF cmp, flagsRegF fcc, regF dst, regF src) %{
  match(Set dst (CMoveF (Binary cmp fcc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "FMOVF$cmp $fcc,$src,$dst" %}
  opcode(0x1);
  ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
  ins_pipe(int_conditional_double_move);
%}

// Conditional move
instruct cmovDP_reg(cmpOpP cmp, flagsRegP pcc, regD dst, regD src) %{
  match(Set dst (CMoveD (Binary cmp pcc) (Binary dst src)));
  ins_cost(150);
  size(4);
  opcode(0x102);
  format %{ "FMOVD$cmp $pcc,$src,$dst" %}
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(int_conditional_double_move);
%}

instruct cmovDI_reg(cmpOp cmp, flagsReg icc, regD dst, regD src) %{
  match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "FMOVD$cmp $icc,$src,$dst" %}
  opcode(0x102);
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(int_conditional_double_move);
%}

instruct cmovDIu_reg(cmpOpU cmp, flagsRegU icc, regD dst, regD src) %{
  match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "FMOVD$cmp $icc,$src,$dst" %}
  opcode(0x102);
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(int_conditional_double_move);
%}

// Conditional move,
instruct cmovDF_reg(cmpOpF cmp, flagsRegF fcc, regD dst, regD src) %{
  match(Set dst (CMoveD (Binary cmp fcc) (Binary dst src)));
  ins_cost(150);
  size(4);
  format %{ "FMOVD$cmp $fcc,$src,$dst" %}
  opcode(0x2);
  ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
  ins_pipe(int_conditional_double_move);
%}

// Conditional move
instruct cmovLP_reg(cmpOpP cmp, flagsRegP pcc, iRegL dst, iRegL src) %{
  match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovLP_imm(cmpOpP cmp, flagsRegP pcc, iRegL dst, immI11 src) %{
  match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
  ins_cost(140);
  format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovLI_reg(cmpOp cmp, flagsReg icc, iRegL dst, iRegL src) %{
  match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}


instruct cmovLIu_reg(cmpOpU cmp, flagsRegU icc, iRegL dst, iRegL src) %{
  match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "MOV$cmp  $icc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
  ins_pipe(ialu_reg);
%}


instruct cmovLF_reg(cmpOpF cmp, flagsRegF fcc, iRegL dst, iRegL src) %{
  match(Set dst (CMoveL (Binary cmp fcc) (Binary dst src)));
  ins_cost(150);

  size(4);
  format %{ "MOV$cmp  $fcc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
  ins_pipe(ialu_reg);
%}



//----------OS and Locking Instructions----------------------------------------

// This name is KNOWN by the ADLC and cannot be changed.
// The ADLC forces a 'TypeRawPtr::BOTTOM' output type
// for this guy.
instruct tlsLoadP(g2RegP dst) %{
  match(Set dst (ThreadLocal));

  size(0);
  ins_cost(0);
  format %{ "# TLS is in G2" %}
  ins_encode( /*empty encoding*/ );
  ins_pipe(ialu_none);
%}

instruct checkCastPP( iRegP dst ) %{
  match(Set dst (CheckCastPP dst));

  size(0);
  format %{ "# checkcastPP of $dst" %}
  ins_encode( /*empty encoding*/ );
  ins_pipe(empty);
%}


instruct castPP( iRegP dst ) %{
  match(Set dst (CastPP dst));
  format %{ "# castPP of $dst" %}
  ins_encode( /*empty encoding*/ );
  ins_pipe(empty);
%}

instruct castII( iRegI dst ) %{
  match(Set dst (CastII dst));
  format %{ "# castII of $dst" %}
  ins_encode( /*empty encoding*/ );
  ins_cost(0);
  ins_pipe(empty);
%}

//----------Arithmetic Instructions--------------------------------------------
// Addition Instructions
// Register Addition
instruct addI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (AddI src1 src2));

  size(4);
  format %{ "ADD    $src1,$src2,$dst" %}
  ins_encode %{
    __ add($src1$$Register, $src2$$Register, $dst$$Register);
  %}
  ins_pipe(ialu_reg_reg);
%}

// Immediate Addition
instruct addI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
  match(Set dst (AddI src1 src2));

  size(4);
  format %{ "ADD    $src1,$src2,$dst" %}
  opcode(Assembler::add_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Pointer Register Addition
instruct addP_reg_reg(iRegP dst, iRegP src1, iRegX src2) %{
  match(Set dst (AddP src1 src2));

  size(4);
  format %{ "ADD    $src1,$src2,$dst" %}
  opcode(Assembler::add_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Pointer Immediate Addition
instruct addP_reg_imm13(iRegP dst, iRegP src1, immX13 src2) %{
  match(Set dst (AddP src1 src2));

  size(4);
  format %{ "ADD    $src1,$src2,$dst" %}
  opcode(Assembler::add_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Long Addition
instruct addL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (AddL src1 src2));

  size(4);
  format %{ "ADD    $src1,$src2,$dst\t! long" %}
  opcode(Assembler::add_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

instruct addL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
  match(Set dst (AddL src1 con));

  size(4);
  format %{ "ADD    $src1,$con,$dst" %}
  opcode(Assembler::add_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
  ins_pipe(ialu_reg_imm);
%}

//----------Conditional_store--------------------------------------------------
// Conditional-store of the updated heap-top.
// Used during allocation of the shared heap.
// Sets flags (EQ) on success.  Implemented with a CASA on Sparc.

// LoadP-locked.  Same as a regular pointer load when used with a compare-swap
instruct loadPLocked(iRegP dst, memory mem) %{
  match(Set dst (LoadPLocked mem));
  ins_cost(MEMORY_REF_COST);

#ifndef _LP64
  size(4);
  format %{ "LDUW   $mem,$dst\t! ptr" %}
  opcode(Assembler::lduw_op3, 0, REGP_OP);
#else
  format %{ "LDX    $mem,$dst\t! ptr" %}
  opcode(Assembler::ldx_op3, 0, REGP_OP);
#endif
  ins_encode( form3_mem_reg( mem, dst ) );
  ins_pipe(iload_mem);
%}

// LoadL-locked.  Same as a regular long load when used with a compare-swap
instruct loadLLocked(iRegL dst, memory mem) %{
  match(Set dst (LoadLLocked mem));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDX    $mem,$dst\t! long" %}
  opcode(Assembler::ldx_op3);
  ins_encode(simple_form3_mem_reg( mem, dst ) );
  ins_pipe(iload_mem);
%}

instruct storePConditional( iRegP heap_top_ptr, iRegP oldval, g3RegP newval, flagsRegP pcc ) %{
  match(Set pcc (StorePConditional heap_top_ptr (Binary oldval newval)));
  effect( KILL newval );
  format %{ "CASA   [$heap_top_ptr],$oldval,R_G3\t! If $oldval==[$heap_top_ptr] Then store R_G3 into [$heap_top_ptr], set R_G3=[$heap_top_ptr] in any case\n\t"
            "CMP    R_G3,$oldval\t\t! See if we made progress"  %}
  ins_encode( enc_cas(heap_top_ptr,oldval,newval) );
  ins_pipe( long_memory_op );
%}

// Conditional-store of an int value.
instruct storeIConditional( iRegP mem_ptr, iRegI oldval, g3RegI newval, flagsReg icc ) %{
  match(Set icc (StoreIConditional mem_ptr (Binary oldval newval)));
  effect( KILL newval );
  format %{ "CASA   [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
            "CMP    $oldval,$newval\t\t! See if we made progress"  %}
  ins_encode( enc_cas(mem_ptr,oldval,newval) );
  ins_pipe( long_memory_op );
%}

// Conditional-store of a long value.
instruct storeLConditional( iRegP mem_ptr, iRegL oldval, g3RegL newval, flagsRegL xcc ) %{
  match(Set xcc (StoreLConditional mem_ptr (Binary oldval newval)));
  effect( KILL newval );
  format %{ "CASXA  [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
            "CMP    $oldval,$newval\t\t! See if we made progress"  %}
  ins_encode( enc_cas(mem_ptr,oldval,newval) );
  ins_pipe( long_memory_op );
%}

// No flag versions for CompareAndSwap{P,I,L} because matcher can't match them

instruct compareAndSwapL_bool(iRegP mem_ptr, iRegL oldval, iRegL newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
  match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
  effect( USE mem_ptr, KILL ccr, KILL tmp1);
  format %{
            "MOV    $newval,O7\n\t"
            "CASXA  [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
            "CMP    $oldval,O7\t\t! See if we made progress\n\t"
            "MOV    1,$res\n\t"
            "MOVne  xcc,R_G0,$res"
  %}
  ins_encode( enc_casx(mem_ptr, oldval, newval),
              enc_lflags_ne_to_boolean(res) );
  ins_pipe( long_memory_op );
%}


instruct compareAndSwapI_bool(iRegP mem_ptr, iRegI oldval, iRegI newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
  match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
  effect( USE mem_ptr, KILL ccr, KILL tmp1);
  format %{
            "MOV    $newval,O7\n\t"
            "CASA   [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
            "CMP    $oldval,O7\t\t! See if we made progress\n\t"
            "MOV    1,$res\n\t"
            "MOVne  icc,R_G0,$res"
  %}
  ins_encode( enc_casi(mem_ptr, oldval, newval),
              enc_iflags_ne_to_boolean(res) );
  ins_pipe( long_memory_op );
%}

instruct compareAndSwapP_bool(iRegP mem_ptr, iRegP oldval, iRegP newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
  match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
  effect( USE mem_ptr, KILL ccr, KILL tmp1);
  format %{
            "MOV    $newval,O7\n\t"
            "CASA_PTR  [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
            "CMP    $oldval,O7\t\t! See if we made progress\n\t"
            "MOV    1,$res\n\t"
            "MOVne  xcc,R_G0,$res"
  %}
#ifdef _LP64
  ins_encode( enc_casx(mem_ptr, oldval, newval),
              enc_lflags_ne_to_boolean(res) );
#else
  ins_encode( enc_casi(mem_ptr, oldval, newval),
              enc_iflags_ne_to_boolean(res) );
#endif
  ins_pipe( long_memory_op );
%}

instruct compareAndSwapN_bool(iRegP mem_ptr, iRegN oldval, iRegN newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
  match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
  effect( USE mem_ptr, KILL ccr, KILL tmp1);
  format %{
            "MOV    $newval,O7\n\t"
            "CASA   [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
            "CMP    $oldval,O7\t\t! See if we made progress\n\t"
            "MOV    1,$res\n\t"
            "MOVne  icc,R_G0,$res"
  %}
  ins_encode( enc_casi(mem_ptr, oldval, newval),
              enc_iflags_ne_to_boolean(res) );
  ins_pipe( long_memory_op );
%}

//---------------------
// Subtraction Instructions
// Register Subtraction
instruct subI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (SubI src1 src2));

  size(4);
  format %{ "SUB    $src1,$src2,$dst" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Immediate Subtraction
instruct subI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
  match(Set dst (SubI src1 src2));

  size(4);
  format %{ "SUB    $src1,$src2,$dst" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

instruct subI_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
  match(Set dst (SubI zero src2));

  size(4);
  format %{ "NEG    $src2,$dst" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
  ins_pipe(ialu_zero_reg);
%}

// Long subtraction
instruct subL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (SubL src1 src2));

  size(4);
  format %{ "SUB    $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Immediate Subtraction
instruct subL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
  match(Set dst (SubL src1 con));

  size(4);
  format %{ "SUB    $src1,$con,$dst\t! long" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Long negation
instruct negL_reg_reg(iRegL dst, immL0 zero, iRegL src2) %{
  match(Set dst (SubL zero src2));

  size(4);
  format %{ "NEG    $src2,$dst\t! long" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
  ins_pipe(ialu_zero_reg);
%}

// Multiplication Instructions
// Integer Multiplication
// Register Multiplication
instruct mulI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (MulI src1 src2));

  size(4);
  format %{ "MULX   $src1,$src2,$dst" %}
  opcode(Assembler::mulx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(imul_reg_reg);
%}

// Immediate Multiplication
instruct mulI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
  match(Set dst (MulI src1 src2));

  size(4);
  format %{ "MULX   $src1,$src2,$dst" %}
  opcode(Assembler::mulx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(imul_reg_imm);
%}

instruct mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (MulL src1 src2));
  ins_cost(DEFAULT_COST * 5);
  size(4);
  format %{ "MULX   $src1,$src2,$dst\t! long" %}
  opcode(Assembler::mulx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(mulL_reg_reg);
%}

// Immediate Multiplication
instruct mulL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
  match(Set dst (MulL src1 src2));
  ins_cost(DEFAULT_COST * 5);
  size(4);
  format %{ "MULX   $src1,$src2,$dst" %}
  opcode(Assembler::mulx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(mulL_reg_imm);
%}

// Integer Division
// Register Division
instruct divI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2) %{
  match(Set dst (DivI src1 src2));
  ins_cost((2+71)*DEFAULT_COST);

  format %{ "SRA     $src2,0,$src2\n\t"
            "SRA     $src1,0,$src1\n\t"
            "SDIVX   $src1,$src2,$dst" %}
  ins_encode( idiv_reg( src1, src2, dst ) );
  ins_pipe(sdiv_reg_reg);
%}

// Immediate Division
instruct divI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2) %{
  match(Set dst (DivI src1 src2));
  ins_cost((2+71)*DEFAULT_COST);

  format %{ "SRA     $src1,0,$src1\n\t"
            "SDIVX   $src1,$src2,$dst" %}
  ins_encode( idiv_imm( src1, src2, dst ) );
  ins_pipe(sdiv_reg_imm);
%}

//----------Div-By-10-Expansion------------------------------------------------
// Extract hi bits of a 32x32->64 bit multiply.
// Expand rule only, not matched
instruct mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2 ) %{
  effect( DEF dst, USE src1, USE src2 );
  format %{ "MULX   $src1,$src2,$dst\t! Used in div-by-10\n\t"
            "SRLX   $dst,#32,$dst\t\t! Extract only hi word of result" %}
  ins_encode( enc_mul_hi(dst,src1,src2));
  ins_pipe(sdiv_reg_reg);
%}

// Magic constant, reciprocal of 10
instruct loadConI_x66666667(iRegIsafe dst) %{
  effect( DEF dst );

  size(8);
  format %{ "SET    0x66666667,$dst\t! Used in div-by-10" %}
  ins_encode( Set32(0x66666667, dst) );
  ins_pipe(ialu_hi_lo_reg);
%}

// Register Shift Right Arithmetic Long by 32-63
instruct sra_31( iRegI dst, iRegI src ) %{
  effect( DEF dst, USE src );
  format %{ "SRA    $src,31,$dst\t! Used in div-by-10" %}
  ins_encode( form3_rs1_rd_copysign_hi(src,dst) );
  ins_pipe(ialu_reg_reg);
%}

// Arithmetic Shift Right by 8-bit immediate
instruct sra_reg_2( iRegI dst, iRegI src ) %{
  effect( DEF dst, USE src );
  format %{ "SRA    $src,2,$dst\t! Used in div-by-10" %}
  opcode(Assembler::sra_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src, 0x2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Integer DIV with 10
instruct divI_10( iRegI dst, iRegIsafe src, immI10 div ) %{
  match(Set dst (DivI src div));
  ins_cost((6+6)*DEFAULT_COST);
  expand %{
    iRegIsafe tmp1;               // Killed temps;
    iRegIsafe tmp2;               // Killed temps;
    iRegI tmp3;                   // Killed temps;
    iRegI tmp4;                   // Killed temps;
    loadConI_x66666667( tmp1 );   // SET  0x66666667 -> tmp1
    mul_hi( tmp2, src, tmp1 );    // MUL  hibits(src * tmp1) -> tmp2
    sra_31( tmp3, src );          // SRA  src,31 -> tmp3
    sra_reg_2( tmp4, tmp2 );      // SRA  tmp2,2 -> tmp4
    subI_reg_reg( dst,tmp4,tmp3); // SUB  tmp4 - tmp3 -> dst
  %}
%}

// Register Long Division
instruct divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (DivL src1 src2));
  ins_cost(DEFAULT_COST*71);
  size(4);
  format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sdivx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(divL_reg_reg);
%}

// Register Long Division
instruct divL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
  match(Set dst (DivL src1 src2));
  ins_cost(DEFAULT_COST*71);
  size(4);
  format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sdivx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(divL_reg_imm);
%}

// Integer Remainder
// Register Remainder
instruct modI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2, o7RegP temp, flagsReg ccr ) %{
  match(Set dst (ModI src1 src2));
  effect( KILL ccr, KILL temp);

  format %{ "SREM   $src1,$src2,$dst" %}
  ins_encode( irem_reg(src1, src2, dst, temp) );
  ins_pipe(sdiv_reg_reg);
%}

// Immediate Remainder
instruct modI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2, o7RegP temp, flagsReg ccr ) %{
  match(Set dst (ModI src1 src2));
  effect( KILL ccr, KILL temp);

  format %{ "SREM   $src1,$src2,$dst" %}
  ins_encode( irem_imm(src1, src2, dst, temp) );
  ins_pipe(sdiv_reg_imm);
%}

// Register Long Remainder
instruct divL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sdivx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(divL_reg_reg);
%}

// Register Long Division
instruct divL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sdivx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(divL_reg_imm);
%}

instruct mulL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "MULX   $src1,$src2,$dst\t! long" %}
  opcode(Assembler::mulx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(mulL_reg_reg);
%}

// Immediate Multiplication
instruct mulL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "MULX   $src1,$src2,$dst" %}
  opcode(Assembler::mulx_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(mulL_reg_imm);
%}

instruct subL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "SUB    $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

instruct subL_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "SUB    $src1,$src2,$dst\t! long" %}
  opcode(Assembler::sub_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Long Remainder
instruct modL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (ModL src1 src2));
  ins_cost(DEFAULT_COST*(71 + 6 + 1));
  expand %{
    iRegL tmp1;
    iRegL tmp2;
    divL_reg_reg_1(tmp1, src1, src2);
    mulL_reg_reg_1(tmp2, tmp1, src2);
    subL_reg_reg_1(dst,  src1, tmp2);
  %}
%}

// Register Long Remainder
instruct modL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
  match(Set dst (ModL src1 src2));
  ins_cost(DEFAULT_COST*(71 + 6 + 1));
  expand %{
    iRegL tmp1;
    iRegL tmp2;
    divL_reg_imm13_1(tmp1, src1, src2);
    mulL_reg_imm13_1(tmp2, tmp1, src2);
    subL_reg_reg_2  (dst,  src1, tmp2);
  %}
%}

// Integer Shift Instructions
// Register Shift Left
instruct shlI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (LShiftI src1 src2));

  size(4);
  format %{ "SLL    $src1,$src2,$dst" %}
  opcode(Assembler::sll_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Shift Left Immediate
instruct shlI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
  match(Set dst (LShiftI src1 src2));

  size(4);
  format %{ "SLL    $src1,$src2,$dst" %}
  opcode(Assembler::sll_op3, Assembler::arith_op);
  ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Shift Left
instruct shlL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
  match(Set dst (LShiftL src1 src2));

  size(4);
  format %{ "SLLX   $src1,$src2,$dst" %}
  opcode(Assembler::sllx_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Shift Left Immediate
instruct shlL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
  match(Set dst (LShiftL src1 src2));

  size(4);
  format %{ "SLLX   $src1,$src2,$dst" %}
  opcode(Assembler::sllx_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Arithmetic Shift Right
instruct sarI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (RShiftI src1 src2));
  size(4);
  format %{ "SRA    $src1,$src2,$dst" %}
  opcode(Assembler::sra_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Arithmetic Shift Right Immediate
instruct sarI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
  match(Set dst (RShiftI src1 src2));

  size(4);
  format %{ "SRA    $src1,$src2,$dst" %}
  opcode(Assembler::sra_op3, Assembler::arith_op);
  ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Shift Right Arithmatic Long
instruct sarL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
  match(Set dst (RShiftL src1 src2));

  size(4);
  format %{ "SRAX   $src1,$src2,$dst" %}
  opcode(Assembler::srax_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Shift Left Immediate
instruct sarL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
  match(Set dst (RShiftL src1 src2));

  size(4);
  format %{ "SRAX   $src1,$src2,$dst" %}
  opcode(Assembler::srax_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Shift Right
instruct shrI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (URShiftI src1 src2));

  size(4);
  format %{ "SRL    $src1,$src2,$dst" %}
  opcode(Assembler::srl_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Shift Right Immediate
instruct shrI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
  match(Set dst (URShiftI src1 src2));

  size(4);
  format %{ "SRL    $src1,$src2,$dst" %}
  opcode(Assembler::srl_op3, Assembler::arith_op);
  ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Shift Right
instruct shrL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
  match(Set dst (URShiftL src1 src2));

  size(4);
  format %{ "SRLX   $src1,$src2,$dst" %}
  opcode(Assembler::srlx_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Register Shift Right Immediate
instruct shrL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
  match(Set dst (URShiftL src1 src2));

  size(4);
  format %{ "SRLX   $src1,$src2,$dst" %}
  opcode(Assembler::srlx_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Shift Right Immediate with a CastP2X
#ifdef _LP64
instruct shrP_reg_imm6(iRegL dst, iRegP src1, immU6 src2) %{
  match(Set dst (URShiftL (CastP2X src1) src2));
  size(4);
  format %{ "SRLX   $src1,$src2,$dst\t! Cast ptr $src1 to long and shift" %}
  opcode(Assembler::srlx_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}
#else
instruct shrP_reg_imm5(iRegI dst, iRegP src1, immU5 src2) %{
  match(Set dst (URShiftI (CastP2X src1) src2));
  size(4);
  format %{ "SRL    $src1,$src2,$dst\t! Cast ptr $src1 to int and shift" %}
  opcode(Assembler::srl_op3, Assembler::arith_op);
  ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}
#endif


//----------Floating Point Arithmetic Instructions-----------------------------

//  Add float single precision
instruct addF_reg_reg(regF dst, regF src1, regF src2) %{
  match(Set dst (AddF src1 src2));

  size(4);
  format %{ "FADDS  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fadds_opf);
  ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
  ins_pipe(faddF_reg_reg);
%}

//  Add float double precision
instruct addD_reg_reg(regD dst, regD src1, regD src2) %{
  match(Set dst (AddD src1 src2));

  size(4);
  format %{ "FADDD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(faddD_reg_reg);
%}

//  Sub float single precision
instruct subF_reg_reg(regF dst, regF src1, regF src2) %{
  match(Set dst (SubF src1 src2));

  size(4);
  format %{ "FSUBS  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubs_opf);
  ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
  ins_pipe(faddF_reg_reg);
%}

//  Sub float double precision
instruct subD_reg_reg(regD dst, regD src1, regD src2) %{
  match(Set dst (SubD src1 src2));

  size(4);
  format %{ "FSUBD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(faddD_reg_reg);
%}

//  Mul float single precision
instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{
  match(Set dst (MulF src1 src2));

  size(4);
  format %{ "FMULS  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuls_opf);
  ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
  ins_pipe(fmulF_reg_reg);
%}

//  Mul float double precision
instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{
  match(Set dst (MulD src1 src2));

  size(4);
  format %{ "FMULD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(fmulD_reg_reg);
%}

//  Div float single precision
instruct divF_reg_reg(regF dst, regF src1, regF src2) %{
  match(Set dst (DivF src1 src2));

  size(4);
  format %{ "FDIVS  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivs_opf);
  ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
  ins_pipe(fdivF_reg_reg);
%}

//  Div float double precision
instruct divD_reg_reg(regD dst, regD src1, regD src2) %{
  match(Set dst (DivD src1 src2));

  size(4);
  format %{ "FDIVD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivd_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(fdivD_reg_reg);
%}

//  Absolute float double precision
instruct absD_reg(regD dst, regD src) %{
  match(Set dst (AbsD src));

  format %{ "FABSd  $src,$dst" %}
  ins_encode(fabsd(dst, src));
  ins_pipe(faddD_reg);
%}

//  Absolute float single precision
instruct absF_reg(regF dst, regF src) %{
  match(Set dst (AbsF src));

  format %{ "FABSs  $src,$dst" %}
  ins_encode(fabss(dst, src));
  ins_pipe(faddF_reg);
%}

instruct negF_reg(regF dst, regF src) %{
  match(Set dst (NegF src));

  size(4);
  format %{ "FNEGs  $src,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fnegs_opf);
  ins_encode(form3_opf_rs2F_rdF(src, dst));
  ins_pipe(faddF_reg);
%}

instruct negD_reg(regD dst, regD src) %{
  match(Set dst (NegD src));

  format %{ "FNEGd  $src,$dst" %}
  ins_encode(fnegd(dst, src));
  ins_pipe(faddD_reg);
%}

//  Sqrt float double precision
instruct sqrtF_reg_reg(regF dst, regF src) %{
  match(Set dst (ConvD2F (SqrtD (ConvF2D src))));

  size(4);
  format %{ "FSQRTS $src,$dst" %}
  ins_encode(fsqrts(dst, src));
  ins_pipe(fdivF_reg_reg);
%}

//  Sqrt float double precision
instruct sqrtD_reg_reg(regD dst, regD src) %{
  match(Set dst (SqrtD src));

  size(4);
  format %{ "FSQRTD $src,$dst" %}
  ins_encode(fsqrtd(dst, src));
  ins_pipe(fdivD_reg_reg);
%}

//----------Logical Instructions-----------------------------------------------
// And Instructions
// Register And
instruct andI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (AndI src1 src2));

  size(4);
  format %{ "AND    $src1,$src2,$dst" %}
  opcode(Assembler::and_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Immediate And
instruct andI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
  match(Set dst (AndI src1 src2));

  size(4);
  format %{ "AND    $src1,$src2,$dst" %}
  opcode(Assembler::and_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register And Long
instruct andL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (AndL src1 src2));

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "AND    $src1,$src2,$dst\t! long" %}
  opcode(Assembler::and_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

instruct andL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
  match(Set dst (AndL src1 con));

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "AND    $src1,$con,$dst\t! long" %}
  opcode(Assembler::and_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Or Instructions
// Register Or
instruct orI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (OrI src1 src2));

  size(4);
  format %{ "OR     $src1,$src2,$dst" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Immediate Or
instruct orI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
  match(Set dst (OrI src1 src2));

  size(4);
  format %{ "OR     $src1,$src2,$dst" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Or Long
instruct orL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (OrL src1 src2));

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "OR     $src1,$src2,$dst\t! long" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

instruct orL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
  match(Set dst (OrL src1 con));
  ins_cost(DEFAULT_COST*2);

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "OR     $src1,$con,$dst\t! long" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
  ins_pipe(ialu_reg_imm);
%}

#ifndef _LP64

// Use sp_ptr_RegP to match G2 (TLS register) without spilling.
instruct orI_reg_castP2X(iRegI dst, iRegI src1, sp_ptr_RegP src2) %{
  match(Set dst (OrI src1 (CastP2X src2)));

  size(4);
  format %{ "OR     $src1,$src2,$dst" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

#else

instruct orL_reg_castP2X(iRegL dst, iRegL src1, sp_ptr_RegP src2) %{
  match(Set dst (OrL src1 (CastP2X src2)));

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "OR     $src1,$src2,$dst\t! long" %}
  opcode(Assembler::or_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

#endif

// Xor Instructions
// Register Xor
instruct xorI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  match(Set dst (XorI src1 src2));

  size(4);
  format %{ "XOR    $src1,$src2,$dst" %}
  opcode(Assembler::xor_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Immediate Xor
instruct xorI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
  match(Set dst (XorI src1 src2));

  size(4);
  format %{ "XOR    $src1,$src2,$dst" %}
  opcode(Assembler::xor_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Register Xor Long
instruct xorL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
  match(Set dst (XorL src1 src2));

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "XOR    $src1,$src2,$dst\t! long" %}
  opcode(Assembler::xor_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
  ins_pipe(ialu_reg_reg);
%}

instruct xorL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
  match(Set dst (XorL src1 con));

  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "XOR    $src1,$con,$dst\t! long" %}
  opcode(Assembler::xor_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
  ins_pipe(ialu_reg_imm);
%}

//----------Convert to Boolean-------------------------------------------------
// Nice hack for 32-bit tests but doesn't work for
// 64-bit pointers.
instruct convI2B( iRegI dst, iRegI src, flagsReg ccr ) %{
  match(Set dst (Conv2B src));
  effect( KILL ccr );
  ins_cost(DEFAULT_COST*2);
  format %{ "CMP    R_G0,$src\n\t"
            "ADDX   R_G0,0,$dst" %}
  ins_encode( enc_to_bool( src, dst ) );
  ins_pipe(ialu_reg_ialu);
%}

#ifndef _LP64
instruct convP2B( iRegI dst, iRegP src, flagsReg ccr ) %{
  match(Set dst (Conv2B src));
  effect( KILL ccr );
  ins_cost(DEFAULT_COST*2);
  format %{ "CMP    R_G0,$src\n\t"
            "ADDX   R_G0,0,$dst" %}
  ins_encode( enc_to_bool( src, dst ) );
  ins_pipe(ialu_reg_ialu);
%}
#else
instruct convP2B( iRegI dst, iRegP src ) %{
  match(Set dst (Conv2B src));
  ins_cost(DEFAULT_COST*2);
  format %{ "MOV    $src,$dst\n\t"
            "MOVRNZ $src,1,$dst" %}
  ins_encode( form3_g0_rs2_rd_move( src, dst ), enc_convP2B( dst, src ) );
  ins_pipe(ialu_clr_and_mover);
%}
#endif

instruct cmpLTMask_reg_reg( iRegI dst, iRegI p, iRegI q, flagsReg ccr ) %{
  match(Set dst (CmpLTMask p q));
  effect( KILL ccr );
  ins_cost(DEFAULT_COST*4);
  format %{ "CMP    $p,$q\n\t"
            "MOV    #0,$dst\n\t"
            "BLT,a  .+8\n\t"
            "MOV    #-1,$dst" %}
  ins_encode( enc_ltmask(p,q,dst) );
  ins_pipe(ialu_reg_reg_ialu);
%}

instruct cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{
  match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
  effect(KILL ccr, TEMP tmp);
  ins_cost(DEFAULT_COST*3);

  format %{ "SUBcc  $p,$q,$p\t! p' = p-q\n\t"
            "ADD    $p,$y,$tmp\t! g3=p-q+y\n\t"
            "MOVl   $tmp,$p\t! p' < 0 ? p'+y : p'" %}
  ins_encode( enc_cadd_cmpLTMask(p, q, y, tmp) );
  ins_pipe( cadd_cmpltmask );
%}

instruct cadd_cmpLTMask2( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{
  match(Set p (AddI (SubI p q) (AndI (CmpLTMask p q) y)));
  effect( KILL ccr, TEMP tmp);
  ins_cost(DEFAULT_COST*3);

  format %{ "SUBcc  $p,$q,$p\t! p' = p-q\n\t"
            "ADD    $p,$y,$tmp\t! g3=p-q+y\n\t"
            "MOVl   $tmp,$p\t! p' < 0 ? p'+y : p'" %}
  ins_encode( enc_cadd_cmpLTMask(p, q, y, tmp) );
  ins_pipe( cadd_cmpltmask );
%}

//----------Arithmetic Conversion Instructions---------------------------------
// The conversions operations are all Alpha sorted.  Please keep it that way!

instruct convD2F_reg(regF dst, regD src) %{
  match(Set dst (ConvD2F src));
  size(4);
  format %{ "FDTOS  $src,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdtos_opf);
  ins_encode(form3_opf_rs2D_rdF(src, dst));
  ins_pipe(fcvtD2F);
%}


// Convert a double to an int in a float register.
// If the double is a NAN, stuff a zero in instead.
instruct convD2I_helper(regF dst, regD src, flagsRegF0 fcc0) %{
  effect(DEF dst, USE src, KILL fcc0);
  format %{ "FCMPd  fcc0,$src,$src\t! check for NAN\n\t"
            "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
            "FDTOI  $src,$dst\t! convert in delay slot\n\t"
            "FITOS  $dst,$dst\t! change NaN/max-int to valid float\n\t"
            "FSUBs  $dst,$dst,$dst\t! cleared only if nan\n"
      "skip:" %}
  ins_encode(form_d2i_helper(src,dst));
  ins_pipe(fcvtD2I);
%}

instruct convD2I_reg(stackSlotI dst, regD src) %{
  match(Set dst (ConvD2I src));
  ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
  expand %{
    regF tmp;
    convD2I_helper(tmp, src);
    regF_to_stkI(dst, tmp);
  %}
%}

// Convert a double to a long in a double register.
// If the double is a NAN, stuff a zero in instead.
instruct convD2L_helper(regD dst, regD src, flagsRegF0 fcc0) %{
  effect(DEF dst, USE src, KILL fcc0);
  format %{ "FCMPd  fcc0,$src,$src\t! check for NAN\n\t"
            "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
            "FDTOX  $src,$dst\t! convert in delay slot\n\t"
            "FXTOD  $dst,$dst\t! change NaN/max-long to valid double\n\t"
            "FSUBd  $dst,$dst,$dst\t! cleared only if nan\n"
      "skip:" %}
  ins_encode(form_d2l_helper(src,dst));
  ins_pipe(fcvtD2L);
%}


// Double to Long conversion
instruct convD2L_reg(stackSlotL dst, regD src) %{
  match(Set dst (ConvD2L src));
  ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
  expand %{
    regD tmp;
    convD2L_helper(tmp, src);
    regD_to_stkL(dst, tmp);
  %}
%}


instruct convF2D_reg(regD dst, regF src) %{
  match(Set dst (ConvF2D src));
  format %{ "FSTOD  $src,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fstod_opf);
  ins_encode(form3_opf_rs2F_rdD(src, dst));
  ins_pipe(fcvtF2D);
%}


instruct convF2I_helper(regF dst, regF src, flagsRegF0 fcc0) %{
  effect(DEF dst, USE src, KILL fcc0);
  format %{ "FCMPs  fcc0,$src,$src\t! check for NAN\n\t"
            "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
            "FSTOI  $src,$dst\t! convert in delay slot\n\t"
            "FITOS  $dst,$dst\t! change NaN/max-int to valid float\n\t"
            "FSUBs  $dst,$dst,$dst\t! cleared only if nan\n"
      "skip:" %}
  ins_encode(form_f2i_helper(src,dst));
  ins_pipe(fcvtF2I);
%}

instruct convF2I_reg(stackSlotI dst, regF src) %{
  match(Set dst (ConvF2I src));
  ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
  expand %{
    regF tmp;
    convF2I_helper(tmp, src);
    regF_to_stkI(dst, tmp);
  %}
%}


instruct convF2L_helper(regD dst, regF src, flagsRegF0 fcc0) %{
  effect(DEF dst, USE src, KILL fcc0);
  format %{ "FCMPs  fcc0,$src,$src\t! check for NAN\n\t"
            "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
            "FSTOX  $src,$dst\t! convert in delay slot\n\t"
            "FXTOD  $dst,$dst\t! change NaN/max-long to valid double\n\t"
            "FSUBd  $dst,$dst,$dst\t! cleared only if nan\n"
      "skip:" %}
  ins_encode(form_f2l_helper(src,dst));
  ins_pipe(fcvtF2L);
%}

// Float to Long conversion
instruct convF2L_reg(stackSlotL dst, regF src) %{
  match(Set dst (ConvF2L src));
  ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
  expand %{
    regD tmp;
    convF2L_helper(tmp, src);
    regD_to_stkL(dst, tmp);
  %}
%}


instruct convI2D_helper(regD dst, regF tmp) %{
  effect(USE tmp, DEF dst);
  format %{ "FITOD  $tmp,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
  ins_encode(form3_opf_rs2F_rdD(tmp, dst));
  ins_pipe(fcvtI2D);
%}

instruct convI2D_reg(stackSlotI src, regD dst) %{
  match(Set dst (ConvI2D src));
  ins_cost(DEFAULT_COST + MEMORY_REF_COST);
  expand %{
    regF tmp;
    stkI_to_regF( tmp, src);
    convI2D_helper( dst, tmp);
  %}
%}

instruct convI2D_mem( regD_low dst, memory mem ) %{
  match(Set dst (ConvI2D (LoadI mem)));
  ins_cost(DEFAULT_COST + MEMORY_REF_COST);
  size(8);
  format %{ "LDF    $mem,$dst\n\t"
            "FITOD  $dst,$dst" %}
  opcode(Assembler::ldf_op3, Assembler::fitod_opf);
  ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
  ins_pipe(floadF_mem);
%}


instruct convI2F_helper(regF dst, regF tmp) %{
  effect(DEF dst, USE tmp);
  format %{ "FITOS  $tmp,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitos_opf);
  ins_encode(form3_opf_rs2F_rdF(tmp, dst));
  ins_pipe(fcvtI2F);
%}

instruct convI2F_reg( regF dst, stackSlotI src ) %{
  match(Set dst (ConvI2F src));
  ins_cost(DEFAULT_COST + MEMORY_REF_COST);
  expand %{
    regF tmp;
    stkI_to_regF(tmp,src);
    convI2F_helper(dst, tmp);
  %}
%}

instruct convI2F_mem( regF dst, memory mem ) %{
  match(Set dst (ConvI2F (LoadI mem)));
  ins_cost(DEFAULT_COST + MEMORY_REF_COST);
  size(8);
  format %{ "LDF    $mem,$dst\n\t"
            "FITOS  $dst,$dst" %}
  opcode(Assembler::ldf_op3, Assembler::fitos_opf);
  ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
  ins_pipe(floadF_mem);
%}


instruct convI2L_reg(iRegL dst, iRegI src) %{
  match(Set dst (ConvI2L src));
  size(4);
  format %{ "SRA    $src,0,$dst\t! int->long" %}
  opcode(Assembler::sra_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Zero-extend convert int to long
instruct convI2L_reg_zex(iRegL dst, iRegI src, immL_32bits mask ) %{
  match(Set dst (AndL (ConvI2L src) mask) );
  size(4);
  format %{ "SRL    $src,0,$dst\t! zero-extend int to long" %}
  opcode(Assembler::srl_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
  ins_pipe(ialu_reg_reg);
%}

// Zero-extend long
instruct zerox_long(iRegL dst, iRegL src, immL_32bits mask ) %{
  match(Set dst (AndL src mask) );
  size(4);
  format %{ "SRL    $src,0,$dst\t! zero-extend long" %}
  opcode(Assembler::srl_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
  ins_pipe(ialu_reg_reg);
%}

instruct MoveF2I_stack_reg(iRegI dst, stackSlotF src) %{
  match(Set dst (MoveF2I src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDUW   $src,$dst\t! MoveF2I" %}
  opcode(Assembler::lduw_op3);
  ins_encode(simple_form3_mem_reg( src, dst ) );
  ins_pipe(iload_mem);
%}

instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
  match(Set dst (MoveI2F src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDF    $src,$dst\t! MoveI2F" %}
  opcode(Assembler::ldf_op3);
  ins_encode(simple_form3_mem_reg(src, dst));
  ins_pipe(floadF_stk);
%}

instruct MoveD2L_stack_reg(iRegL dst, stackSlotD src) %{
  match(Set dst (MoveD2L src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDX    $src,$dst\t! MoveD2L" %}
  opcode(Assembler::ldx_op3);
  ins_encode(simple_form3_mem_reg( src, dst ) );
  ins_pipe(iload_mem);
%}

instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
  match(Set dst (MoveL2D src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "LDDF   $src,$dst\t! MoveL2D" %}
  opcode(Assembler::lddf_op3);
  ins_encode(simple_form3_mem_reg(src, dst));
  ins_pipe(floadD_stk);
%}

instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
  match(Set dst (MoveF2I src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STF   $src,$dst\t!MoveF2I" %}
  opcode(Assembler::stf_op3);
  ins_encode(simple_form3_mem_reg(dst, src));
  ins_pipe(fstoreF_stk_reg);
%}

instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{
  match(Set dst (MoveI2F src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STW    $src,$dst\t!MoveI2F" %}
  opcode(Assembler::stw_op3);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_reg);
%}

instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
  match(Set dst (MoveD2L src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STDF   $src,$dst\t!MoveD2L" %}
  opcode(Assembler::stdf_op3);
  ins_encode(simple_form3_mem_reg(dst, src));
  ins_pipe(fstoreD_stk_reg);
%}

instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{
  match(Set dst (MoveL2D src));
  effect(DEF dst, USE src);
  ins_cost(MEMORY_REF_COST);

  size(4);
  format %{ "STX    $src,$dst\t!MoveL2D" %}
  opcode(Assembler::stx_op3);
  ins_encode(simple_form3_mem_reg( dst, src ) );
  ins_pipe(istore_mem_reg);
%}


//-----------
// Long to Double conversion using V8 opcodes.
// Still useful because cheetah traps and becomes
// amazingly slow for some common numbers.

// Magic constant, 0x43300000
instruct loadConI_x43300000(iRegI dst) %{
  effect(DEF dst);
  size(4);
  format %{ "SETHI  HI(0x43300000),$dst\t! 2^52" %}
  ins_encode(SetHi22(0x43300000, dst));
  ins_pipe(ialu_none);
%}

// Magic constant, 0x41f00000
instruct loadConI_x41f00000(iRegI dst) %{
  effect(DEF dst);
  size(4);
  format %{ "SETHI  HI(0x41f00000),$dst\t! 2^32" %}
  ins_encode(SetHi22(0x41f00000, dst));
  ins_pipe(ialu_none);
%}

// Construct a double from two float halves
instruct regDHi_regDLo_to_regD(regD_low dst, regD_low src1, regD_low src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(8);
  format %{ "FMOVS  $src1.hi,$dst.hi\n\t"
            "FMOVS  $src2.lo,$dst.lo" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmovs_opf);
  ins_encode(form3_opf_rs2D_hi_rdD_hi(src1, dst), form3_opf_rs2D_lo_rdD_lo(src2, dst));
  ins_pipe(faddD_reg_reg);
%}

// Convert integer in high half of a double register (in the lower half of
// the double register file) to double
instruct convI2D_regDHi_regD(regD dst, regD_low src) %{
  effect(DEF dst, USE src);
  size(4);
  format %{ "FITOD  $src,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
  ins_encode(form3_opf_rs2D_rdD(src, dst));
  ins_pipe(fcvtLHi2D);
%}

// Add float double precision
instruct addD_regD_regD(regD dst, regD src1, regD src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "FADDD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(faddD_reg_reg);
%}

// Sub float double precision
instruct subD_regD_regD(regD dst, regD src1, regD src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "FSUBD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(faddD_reg_reg);
%}

// Mul float double precision
instruct mulD_regD_regD(regD dst, regD src1, regD src2) %{
  effect(DEF dst, USE src1, USE src2);
  size(4);
  format %{ "FMULD  $src1,$src2,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
  ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
  ins_pipe(fmulD_reg_reg);
%}

instruct convL2D_reg_slow_fxtof(regD dst, stackSlotL src) %{
  match(Set dst (ConvL2D src));
  ins_cost(DEFAULT_COST*8 + MEMORY_REF_COST*6);

  expand %{
    regD_low   tmpsrc;
    iRegI      ix43300000;
    iRegI      ix41f00000;
    stackSlotL lx43300000;
    stackSlotL lx41f00000;
    regD_low   dx43300000;
    regD       dx41f00000;
    regD       tmp1;
    regD_low   tmp2;
    regD       tmp3;
    regD       tmp4;

    stkL_to_regD(tmpsrc, src);

    loadConI_x43300000(ix43300000);
    loadConI_x41f00000(ix41f00000);
    regI_to_stkLHi(lx43300000, ix43300000);
    regI_to_stkLHi(lx41f00000, ix41f00000);
    stkL_to_regD(dx43300000, lx43300000);
    stkL_to_regD(dx41f00000, lx41f00000);

    convI2D_regDHi_regD(tmp1, tmpsrc);
    regDHi_regDLo_to_regD(tmp2, dx43300000, tmpsrc);
    subD_regD_regD(tmp3, tmp2, dx43300000);
    mulD_regD_regD(tmp4, tmp1, dx41f00000);
    addD_regD_regD(dst, tmp3, tmp4);
  %}
%}

// Long to Double conversion using fast fxtof
instruct convL2D_helper(regD dst, regD tmp) %{
  effect(DEF dst, USE tmp);
  size(4);
  format %{ "FXTOD  $tmp,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtod_opf);
  ins_encode(form3_opf_rs2D_rdD(tmp, dst));
  ins_pipe(fcvtL2D);
%}

instruct convL2D_reg_fast_fxtof(regD dst, stackSlotL src) %{
  predicate(VM_Version::has_fast_fxtof());
  match(Set dst (ConvL2D src));
  ins_cost(DEFAULT_COST + 3 * MEMORY_REF_COST);
  expand %{
    regD tmp;
    stkL_to_regD(tmp, src);
    convL2D_helper(dst, tmp);
  %}
%}

//-----------
// Long to Float conversion using V8 opcodes.
// Still useful because cheetah traps and becomes
// amazingly slow for some common numbers.

// Long to Float conversion using fast fxtof
instruct convL2F_helper(regF dst, regD tmp) %{
  effect(DEF dst, USE tmp);
  size(4);
  format %{ "FXTOS  $tmp,$dst" %}
  opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtos_opf);
  ins_encode(form3_opf_rs2D_rdF(tmp, dst));
  ins_pipe(fcvtL2F);
%}

instruct convL2F_reg_fast_fxtof(regF dst, stackSlotL src) %{
  match(Set dst (ConvL2F src));
  ins_cost(DEFAULT_COST + MEMORY_REF_COST);
  expand %{
    regD tmp;
    stkL_to_regD(tmp, src);
    convL2F_helper(dst, tmp);
  %}
%}
//-----------

instruct convL2I_reg(iRegI dst, iRegL src) %{
  match(Set dst (ConvL2I src));
#ifndef _LP64
  format %{ "MOV    $src.lo,$dst\t! long->int" %}
  ins_encode( form3_g0_rs2_rd_move_lo2( src, dst ) );
  ins_pipe(ialu_move_reg_I_to_L);
#else
  size(4);
  format %{ "SRA    $src,R_G0,$dst\t! long->int" %}
  ins_encode( form3_rs1_rd_signextend_lo1( src, dst ) );
  ins_pipe(ialu_reg);
#endif
%}

// Register Shift Right Immediate
instruct shrL_reg_imm6_L2I(iRegI dst, iRegL src, immI_32_63 cnt) %{
  match(Set dst (ConvL2I (RShiftL src cnt)));

  size(4);
  format %{ "SRAX   $src,$cnt,$dst" %}
  opcode(Assembler::srax_op3, Assembler::arith_op);
  ins_encode( form3_sd_rs1_imm6_rd( src, cnt, dst ) );
  ins_pipe(ialu_reg_imm);
%}

// Replicate scalar to packed byte values in Double register
instruct Repl8B_reg_helper(iRegL dst, iRegI src) %{
  effect(DEF dst, USE src);
  format %{ "SLLX  $src,56,$dst\n\t"
            "SRLX  $dst, 8,O7\n\t"
            "OR    $dst,O7,$dst\n\t"
            "SRLX  $dst,16,O7\n\t"
            "OR    $dst,O7,$dst\n\t"
            "SRLX  $dst,32,O7\n\t"
            "OR    $dst,O7,$dst\t! replicate8B" %}
  ins_encode( enc_repl8b(src, dst));
  ins_pipe(ialu_reg);
%}

// Replicate scalar to packed byte values in Double register
instruct Repl8B_reg(stackSlotD dst, iRegI src) %{
  match(Set dst (Replicate8B src));
  expand %{
    iRegL tmp;
    Repl8B_reg_helper(tmp, src);
    regL_to_stkD(dst, tmp);
  %}
%}

// Replicate scalar constant to packed byte values in Double register
instruct Repl8B_immI(regD dst, immI13 src, o7RegP tmp) %{
  match(Set dst (Replicate8B src));
#ifdef _LP64
  size(36);
#else
  size(8);
#endif
  format %{ "SETHI  hi(&Repl8($src)),$tmp\t!get Repl8B($src) from table\n\t"
            "LDDF   [$tmp+lo(&Repl8($src))],$dst" %}
  ins_encode( LdReplImmI(src, dst, tmp, (8), (1)) );
  ins_pipe(loadConFD);
%}

// Replicate scalar to packed char values into stack slot
instruct Repl4C_reg_helper(iRegL dst, iRegI src) %{
  effect(DEF dst, USE src);
  format %{ "SLLX  $src,48,$dst\n\t"
            "SRLX  $dst,16,O7\n\t"
            "OR    $dst,O7,$dst\n\t"
            "SRLX  $dst,32,O7\n\t"
            "OR    $dst,O7,$dst\t! replicate4C" %}
  ins_encode( enc_repl4s(src, dst) );
  ins_pipe(ialu_reg);
%}

// Replicate scalar to packed char values into stack slot
instruct Repl4C_reg(stackSlotD dst, iRegI src) %{
  match(Set dst (Replicate4C src));
  expand %{
    iRegL tmp;
    Repl4C_reg_helper(tmp, src);
    regL_to_stkD(dst, tmp);
  %}
%}

// Replicate scalar constant to packed char values in Double register
instruct Repl4C_immI(regD dst, immI src, o7RegP tmp) %{
  match(Set dst (Replicate4C src));
#ifdef _LP64
  size(36);
#else
  size(8);
#endif
  format %{ "SETHI  hi(&Repl4($src)),$tmp\t!get Repl4C($src) from table\n\t"
            "LDDF   [$tmp+lo(&Repl4($src))],$dst" %}
  ins_encode( LdReplImmI(src, dst, tmp, (4), (2)) );
  ins_pipe(loadConFD);
%}

// Replicate scalar to packed short values into stack slot
instruct Repl4S_reg_helper(iRegL dst, iRegI src) %{
  effect(DEF dst, USE src);
  format %{ "SLLX  $src,48,$dst\n\t"
            "SRLX  $dst,16,O7\n\t"
            "OR    $dst,O7,$dst\n\t"
            "SRLX  $dst,32,O7\n\t"
            "OR    $dst,O7,$dst\t! replicate4S" %}
  ins_encode( enc_repl4s(src, dst) );
  ins_pipe(ialu_reg);
%}

// Replicate scalar to packed short values into stack slot
instruct Repl4S_reg(stackSlotD dst, iRegI src) %{
  match(Set dst (Replicate4S src));
  expand %{
    iRegL tmp;
    Repl4S_reg_helper(tmp, src);
    regL_to_stkD(dst, tmp);
  %}
%}

// Replicate scalar constant to packed short values in Double register
instruct Repl4S_immI(regD dst, immI src, o7RegP tmp) %{
  match(Set dst (Replicate4S src));
#ifdef _LP64
  size(36);
#else
  size(8);
#endif
  format %{ "SETHI  hi(&Repl4($src)),$tmp\t!get Repl4S($src) from table\n\t"
            "LDDF   [$tmp+lo(&Repl4($src))],$dst" %}
  ins_encode( LdReplImmI(src, dst, tmp, (4), (2)) );
  ins_pipe(loadConFD);
%}

// Replicate scalar to packed int values in Double register
instruct Repl2I_reg_helper(iRegL dst, iRegI src) %{
  effect(DEF dst, USE src);
  format %{ "SLLX  $src,32,$dst\n\t"
            "SRLX  $dst,32,O7\n\t"
            "OR    $dst,O7,$dst\t! replicate2I" %}
  ins_encode( enc_repl2i(src, dst));
  ins_pipe(ialu_reg);
%}

// Replicate scalar to packed int values in Double register
instruct Repl2I_reg(stackSlotD dst, iRegI src) %{
  match(Set dst (Replicate2I src));
  expand %{
    iRegL tmp;
    Repl2I_reg_helper(tmp, src);
    regL_to_stkD(dst, tmp);
  %}
%}

// Replicate scalar zero constant to packed int values in Double register
instruct Repl2I_immI(regD dst, immI src, o7RegP tmp) %{
  match(Set dst (Replicate2I src));
#ifdef _LP64
  size(36);
#else
  size(8);
#endif
  format %{ "SETHI  hi(&Repl2($src)),$tmp\t!get Repl2I($src) from table\n\t"
            "LDDF   [$tmp+lo(&Repl2($src))],$dst" %}
  ins_encode( LdReplImmI(src, dst, tmp, (2), (4)) );
  ins_pipe(loadConFD);
%}

//----------Control Flow Instructions------------------------------------------
// Compare Instructions
// Compare Integers
instruct compI_iReg(flagsReg icc, iRegI op1, iRegI op2) %{
  match(Set icc (CmpI op1 op2));
  effect( DEF icc, USE op1, USE op2 );

  size(4);
  format %{ "CMP    $op1,$op2" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct compU_iReg(flagsRegU icc, iRegI op1, iRegI op2) %{
  match(Set icc (CmpU op1 op2));

  size(4);
  format %{ "CMP    $op1,$op2\t! unsigned" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct compI_iReg_imm13(flagsReg icc, iRegI op1, immI13 op2) %{
  match(Set icc (CmpI op1 op2));
  effect( DEF icc, USE op1 );

  size(4);
  format %{ "CMP    $op1,$op2" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_imm);
%}

instruct testI_reg_reg( flagsReg icc, iRegI op1, iRegI op2, immI0 zero ) %{
  match(Set icc (CmpI (AndI op1 op2) zero));

  size(4);
  format %{ "BTST   $op2,$op1" %}
  opcode(Assembler::andcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg_zero);
%}

instruct testI_reg_imm( flagsReg icc, iRegI op1, immI13 op2, immI0 zero ) %{
  match(Set icc (CmpI (AndI op1 op2) zero));

  size(4);
  format %{ "BTST   $op2,$op1" %}
  opcode(Assembler::andcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_imm_zero);
%}

instruct compL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2 ) %{
  match(Set xcc (CmpL op1 op2));
  effect( DEF xcc, USE op1, USE op2 );

  size(4);
  format %{ "CMP    $op1,$op2\t\t! long" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct compL_reg_con(flagsRegL xcc, iRegL op1, immL13 con) %{
  match(Set xcc (CmpL op1 con));
  effect( DEF xcc, USE op1, USE con );

  size(4);
  format %{ "CMP    $op1,$con\t\t! long" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct testL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2, immL0 zero) %{
  match(Set xcc (CmpL (AndL op1 op2) zero));
  effect( DEF xcc, USE op1, USE op2 );

  size(4);
  format %{ "BTST   $op1,$op2\t\t! long" %}
  opcode(Assembler::andcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

// useful for checking the alignment of a pointer:
instruct testL_reg_con(flagsRegL xcc, iRegL op1, immL13 con, immL0 zero) %{
  match(Set xcc (CmpL (AndL op1 con) zero));
  effect( DEF xcc, USE op1, USE con );

  size(4);
  format %{ "BTST   $op1,$con\t\t! long" %}
  opcode(Assembler::andcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct compU_iReg_imm13(flagsRegU icc, iRegI op1, immU13 op2 ) %{
  match(Set icc (CmpU op1 op2));

  size(4);
  format %{ "CMP    $op1,$op2\t! unsigned" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_imm);
%}

// Compare Pointers
instruct compP_iRegP(flagsRegP pcc, iRegP op1, iRegP op2 ) %{
  match(Set pcc (CmpP op1 op2));

  size(4);
  format %{ "CMP    $op1,$op2\t! ptr" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct compP_iRegP_imm13(flagsRegP pcc, iRegP op1, immP13 op2 ) %{
  match(Set pcc (CmpP op1 op2));

  size(4);
  format %{ "CMP    $op1,$op2\t! ptr" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_imm);
%}

// Compare Narrow oops
instruct compN_iRegN(flagsReg icc, iRegN op1, iRegN op2 ) %{
  match(Set icc (CmpN op1 op2));

  size(4);
  format %{ "CMP    $op1,$op2\t! compressed ptr" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_reg);
%}

instruct compN_iRegN_immN0(flagsReg icc, iRegN op1, immN0 op2 ) %{
  match(Set icc (CmpN op1 op2));

  size(4);
  format %{ "CMP    $op1,$op2\t! compressed ptr" %}
  opcode(Assembler::subcc_op3, Assembler::arith_op);
  ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
  ins_pipe(ialu_cconly_reg_imm);
%}

//----------Max and Min--------------------------------------------------------
// Min Instructions
// Conditional move for min
instruct cmovI_reg_lt( iRegI op2, iRegI op1, flagsReg icc ) %{
  effect( USE_DEF op2, USE op1, USE icc );

  size(4);
  format %{ "MOVlt  icc,$op1,$op2\t! min" %}
  opcode(Assembler::less);
  ins_encode( enc_cmov_reg_minmax(op2,op1) );
  ins_pipe(ialu_reg_flags);
%}

// Min Register with Register.
instruct minI_eReg(iRegI op1, iRegI op2) %{
  match(Set op2 (MinI op1 op2));
  ins_cost(DEFAULT_COST*2);
  expand %{
    flagsReg icc;
    compI_iReg(icc,op1,op2);
    cmovI_reg_lt(op2,op1,icc);
  %}
%}

// Max Instructions
// Conditional move for max
instruct cmovI_reg_gt( iRegI op2, iRegI op1, flagsReg icc ) %{
  effect( USE_DEF op2, USE op1, USE icc );
  format %{ "MOVgt  icc,$op1,$op2\t! max" %}
  opcode(Assembler::greater);
  ins_encode( enc_cmov_reg_minmax(op2,op1) );
  ins_pipe(ialu_reg_flags);
%}

// Max Register with Register
instruct maxI_eReg(iRegI op1, iRegI op2) %{
  match(Set op2 (MaxI op1 op2));
  ins_cost(DEFAULT_COST*2);
  expand %{
    flagsReg icc;
    compI_iReg(icc,op1,op2);
    cmovI_reg_gt(op2,op1,icc);
  %}
%}


//----------Float Compares----------------------------------------------------
// Compare floating, generate condition code
instruct cmpF_cc(flagsRegF fcc, regF src1, regF src2) %{
  match(Set fcc (CmpF src1 src2));

  size(4);
  format %{ "FCMPs  $fcc,$src1,$src2" %}
  opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmps_opf);
  ins_encode( form3_opf_rs1F_rs2F_fcc( src1, src2, fcc ) );
  ins_pipe(faddF_fcc_reg_reg_zero);
%}

instruct cmpD_cc(flagsRegF fcc, regD src1, regD src2) %{
  match(Set fcc (CmpD src1 src2));

  size(4);
  format %{ "FCMPd  $fcc,$src1,$src2" %}
  opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmpd_opf);
  ins_encode( form3_opf_rs1D_rs2D_fcc( src1, src2, fcc ) );
  ins_pipe(faddD_fcc_reg_reg_zero);
%}


// Compare floating, generate -1,0,1
instruct cmpF_reg(iRegI dst, regF src1, regF src2, flagsRegF0 fcc0) %{
  match(Set dst (CmpF3 src1 src2));
  effect(KILL fcc0);
  ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
  format %{ "fcmpl  $dst,$src1,$src2" %}
  // Primary = float
  opcode( true );
  ins_encode( floating_cmp( dst, src1, src2 ) );
  ins_pipe( floating_cmp );
%}

instruct cmpD_reg(iRegI dst, regD src1, regD src2, flagsRegF0 fcc0) %{
  match(Set dst (CmpD3 src1 src2));
  effect(KILL fcc0);
  ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
  format %{ "dcmpl  $dst,$src1,$src2" %}
  // Primary = double (not float)
  opcode( false );
  ins_encode( floating_cmp( dst, src1, src2 ) );
  ins_pipe( floating_cmp );
%}

//----------Branches---------------------------------------------------------
// Jump
// (compare 'operand indIndex' and 'instruct addP_reg_reg' above)
instruct jumpXtnd(iRegX switch_val, o7RegI table) %{
  match(Jump switch_val);

  ins_cost(350);

  format %{  "SETHI  [hi(table_base)],O7\n\t"
             "ADD    O7, lo(table_base), O7\n\t"
             "LD     [O7+$switch_val], O7\n\t"
             "JUMP   O7"
         %}
  ins_encode( jump_enc( switch_val, table) );
  ins_pc_relative(1);
  ins_pipe(ialu_reg_reg);
%}

// Direct Branch.  Use V8 version with longer range.
instruct branch(label labl) %{
  match(Goto);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BA     $labl" %}
  // Prim = bits 24-22, Secnd = bits 31-30, Tert = cond
  opcode(Assembler::br_op2, Assembler::branch_op, Assembler::always);
  ins_encode( enc_ba( labl ) );
  ins_pc_relative(1);
  ins_pipe(br);
%}

// Conditional Direct Branch
instruct branchCon(cmpOp cmp, flagsReg icc, label labl) %{
  match(If cmp icc);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BP$cmp   $icc,$labl" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_bp( labl, cmp, icc ) );
  ins_pc_relative(1);
  ins_pipe(br_cc);
%}

// Branch-on-register tests all 64 bits.  We assume that values
// in 64-bit registers always remains zero or sign extended
// unless our code munges the high bits.  Interrupts can chop
// the high order bits to zero or sign at any time.
instruct branchCon_regI(cmpOp_reg cmp, iRegI op1, immI0 zero, label labl) %{
  match(If cmp (CmpI op1 zero));
  predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BR$cmp   $op1,$labl" %}
  ins_encode( enc_bpr( labl, cmp, op1 ) );
  ins_pc_relative(1);
  ins_pipe(br_reg);
%}

instruct branchCon_regP(cmpOp_reg cmp, iRegP op1, immP0 null, label labl) %{
  match(If cmp (CmpP op1 null));
  predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BR$cmp   $op1,$labl" %}
  ins_encode( enc_bpr( labl, cmp, op1 ) );
  ins_pc_relative(1);
  ins_pipe(br_reg);
%}

instruct branchCon_regL(cmpOp_reg cmp, iRegL op1, immL0 zero, label labl) %{
  match(If cmp (CmpL op1 zero));
  predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BR$cmp   $op1,$labl" %}
  ins_encode( enc_bpr( labl, cmp, op1 ) );
  ins_pc_relative(1);
  ins_pipe(br_reg);
%}

instruct branchConU(cmpOpU cmp, flagsRegU icc, label labl) %{
  match(If cmp icc);
  effect(USE labl);

  format %{ "BP$cmp  $icc,$labl" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_bp( labl, cmp, icc ) );
  ins_pc_relative(1);
  ins_pipe(br_cc);
%}

instruct branchConP(cmpOpP cmp, flagsRegP pcc, label labl) %{
  match(If cmp pcc);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BP$cmp  $pcc,$labl" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_bpx( labl, cmp, pcc ) );
  ins_pc_relative(1);
  ins_pipe(br_cc);
%}

instruct branchConF(cmpOpF cmp, flagsRegF fcc, label labl) %{
  match(If cmp fcc);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "FBP$cmp $fcc,$labl" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_fbp( labl, cmp, fcc ) );
  ins_pc_relative(1);
  ins_pipe(br_fcc);
%}

instruct branchLoopEnd(cmpOp cmp, flagsReg icc, label labl) %{
  match(CountedLoopEnd cmp icc);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BP$cmp   $icc,$labl\t! Loop end" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_bp( labl, cmp, icc ) );
  ins_pc_relative(1);
  ins_pipe(br_cc);
%}

instruct branchLoopEndU(cmpOpU cmp, flagsRegU icc, label labl) %{
  match(CountedLoopEnd cmp icc);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BP$cmp  $icc,$labl\t! Loop end" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_bp( labl, cmp, icc ) );
  ins_pc_relative(1);
  ins_pipe(br_cc);
%}

// ============================================================================
// 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.

instruct branchCon_long(cmpOp cmp, flagsRegL xcc, label labl) %{
  match(If cmp xcc);
  effect(USE labl);

  size(8);
  ins_cost(BRANCH_COST);
  format %{ "BP$cmp   $xcc,$labl" %}
  // Prim = bits 24-22, Secnd = bits 31-30
  ins_encode( enc_bpl( labl, cmp, xcc ) );
  ins_pc_relative(1);
  ins_pipe(br_cc);
%}

// Manifest a CmpL3 result in an integer register.  Very painful.
// This is the test to avoid.
instruct cmpL3_reg_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg ccr ) %{
  match(Set dst (CmpL3 src1 src2) );
  effect( KILL ccr );
  ins_cost(6*DEFAULT_COST);
  size(24);
  format %{ "CMP    $src1,$src2\t\t! long\n"
          "\tBLT,a,pn done\n"
          "\tMOV    -1,$dst\t! delay slot\n"
          "\tBGT,a,pn done\n"
          "\tMOV    1,$dst\t! delay slot\n"
          "\tCLR    $dst\n"
    "done:"     %}
  ins_encode( cmpl_flag(src1,src2,dst) );
  ins_pipe(cmpL_reg);
%}

// Conditional move
instruct cmovLL_reg(cmpOp cmp, flagsRegL xcc, iRegL dst, iRegL src) %{
  match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp  $xcc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovLL_imm(cmpOp cmp, flagsRegL xcc, iRegL dst, immL0 src) %{
  match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
  ins_cost(140);
  format %{ "MOV$cmp  $xcc,$src,$dst\t! long" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovIL_reg(cmpOp cmp, flagsRegL xcc, iRegI dst, iRegI src) %{
  match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp  $xcc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovIL_imm(cmpOp cmp, flagsRegL xcc, iRegI dst, immI11 src) %{
  match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
  ins_cost(140);
  format %{ "MOV$cmp  $xcc,$src,$dst" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovNL_reg(cmpOp cmp, flagsRegL xcc, iRegN dst, iRegN src) %{
  match(Set dst (CMoveN (Binary cmp xcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp  $xcc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovPL_reg(cmpOp cmp, flagsRegL xcc, iRegP dst, iRegP src) %{
  match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
  ins_cost(150);
  format %{ "MOV$cmp  $xcc,$src,$dst" %}
  ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_reg);
%}

instruct cmovPL_imm(cmpOp cmp, flagsRegL xcc, iRegP dst, immP0 src) %{
  match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
  ins_cost(140);
  format %{ "MOV$cmp  $xcc,$src,$dst" %}
  ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(ialu_imm);
%}

instruct cmovFL_reg(cmpOp cmp, flagsRegL xcc, regF dst, regF src) %{
  match(Set dst (CMoveF (Binary cmp xcc) (Binary dst src)));
  ins_cost(150);
  opcode(0x101);
  format %{ "FMOVS$cmp $xcc,$src,$dst" %}
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(int_conditional_float_move);
%}

instruct cmovDL_reg(cmpOp cmp, flagsRegL xcc, regD dst, regD src) %{
  match(Set dst (CMoveD (Binary cmp xcc) (Binary dst src)));
  ins_cost(150);
  opcode(0x102);
  format %{ "FMOVD$cmp $xcc,$src,$dst" %}
  ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
  ins_pipe(int_conditional_float_move);
%}

// ============================================================================
// Safepoint Instruction
instruct safePoint_poll(iRegP poll) %{
  match(SafePoint poll);
  effect(USE poll);

  size(4);
#ifdef _LP64
  format %{ "LDX    [$poll],R_G0\t! Safepoint: poll for GC" %}
#else
  format %{ "LDUW   [$poll],R_G0\t! Safepoint: poll for GC" %}
#endif
  ins_encode %{
    __ relocate(relocInfo::poll_type);
    __ ld_ptr($poll$$Register, 0, G0);
  %}
  ins_pipe(loadPollP);
%}

// ============================================================================
// Call Instructions
// Call Java Static Instruction
instruct CallStaticJavaDirect( method meth ) %{
  match(CallStaticJava);
  predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
  effect(USE meth);

  size(8);
  ins_cost(CALL_COST);
  format %{ "CALL,static  ; NOP ==> " %}
  ins_encode( Java_Static_Call( meth ), call_epilog );
  ins_pc_relative(1);
  ins_pipe(simple_call);
%}

// Call Java Static Instruction (method handle version)
instruct CallStaticJavaHandle(method meth, l7RegP l7_mh_SP_save) %{
  match(CallStaticJava);
  predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
  effect(USE meth, KILL l7_mh_SP_save);

  size(8);
  ins_cost(CALL_COST);
  format %{ "CALL,static/MethodHandle" %}
  ins_encode(preserve_SP, Java_Static_Call(meth), restore_SP, call_epilog);
  ins_pc_relative(1);
  ins_pipe(simple_call);
%}

// Call Java Dynamic Instruction
instruct CallDynamicJavaDirect( method meth ) %{
  match(CallDynamicJava);
  effect(USE meth);

  ins_cost(CALL_COST);
  format %{ "SET    (empty),R_G5\n\t"
            "CALL,dynamic  ; NOP ==> " %}
  ins_encode( Java_Dynamic_Call( meth ), call_epilog );
  ins_pc_relative(1);
  ins_pipe(call);
%}

// Call Runtime Instruction
instruct CallRuntimeDirect(method meth, l7RegP l7) %{
  match(CallRuntime);
  effect(USE meth, KILL l7);
  ins_cost(CALL_COST);
  format %{ "CALL,runtime" %}
  ins_encode( Java_To_Runtime( meth ),
              call_epilog, adjust_long_from_native_call );
  ins_pc_relative(1);
  ins_pipe(simple_call);
%}

// Call runtime without safepoint - same as CallRuntime
instruct CallLeafDirect(method meth, l7RegP l7) %{
  match(CallLeaf);
  effect(USE meth, KILL l7);
  ins_cost(CALL_COST);
  format %{ "CALL,runtime leaf" %}
  ins_encode( Java_To_Runtime( meth ),
              call_epilog,
              adjust_long_from_native_call );
  ins_pc_relative(1);
  ins_pipe(simple_call);
%}

// Call runtime without safepoint - same as CallLeaf
instruct CallLeafNoFPDirect(method meth, l7RegP l7) %{
  match(CallLeafNoFP);
  effect(USE meth, KILL l7);
  ins_cost(CALL_COST);
  format %{ "CALL,runtime leaf nofp" %}
  ins_encode( Java_To_Runtime( meth ),
              call_epilog,
              adjust_long_from_native_call );
  ins_pc_relative(1);
  ins_pipe(simple_call);
%}

// 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(g3RegP jump_target, inline_cache_regP method_oop) %{
  match(TailCall jump_target method_oop );

  ins_cost(CALL_COST);
  format %{ "Jmp     $jump_target  ; NOP \t! $method_oop holds method oop" %}
  ins_encode(form_jmpl(jump_target));
  ins_pipe(tail_call);
%}


// Return Instruction
instruct Ret() %{
  match(Return);

  // The epilogue node did the ret already.
  size(0);
  format %{ "! return" %}
  ins_encode();
  ins_pipe(empty);
%}


// Tail Jump; remove the return address; jump to target.
// TailCall above leaves the return address around.
// TailJump is used in only one place, the rethrow_Java stub (fancy_jump=2).
// ex_oop (Exception Oop) is needed in %o0 at the jump. As there would be a
// "restore" before this instruction (in Epilogue), we need to materialize it
// in %i0.
instruct tailjmpInd(g1RegP jump_target, i0RegP ex_oop) %{
  match( TailJump jump_target ex_oop );
  ins_cost(CALL_COST);
  format %{ "! discard R_O7\n\t"
            "Jmp     $jump_target  ; ADD O7,8,O1 \t! $ex_oop holds exc. oop" %}
  ins_encode(form_jmpl_set_exception_pc(jump_target));
  // opcode(Assembler::jmpl_op3, Assembler::arith_op);
  // The hack duplicates the exception oop into G3, so that CreateEx can use it there.
  // ins_encode( form3_rs1_simm13_rd( jump_target, 0x00, R_G0 ), move_return_pc_to_o1() );
  ins_pipe(tail_call);
%}

// 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( o0RegP ex_oop )
%{
  match(Set ex_oop (CreateEx));
  ins_cost(0);

  size(0);
  // use the following format syntax
  format %{ "! exception oop is in R_O0; 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);
  ins_cost(CALL_COST);

  // use the following format syntax
  format %{ "Jmp    rethrow_stub" %}
  ins_encode(enc_rethrow);
  ins_pipe(tail_call);
%}


// Die now
instruct ShouldNotReachHere( )
%{
  match(Halt);
  ins_cost(CALL_COST);

  size(4);
  // Use the following format syntax
  format %{ "ILLTRAP   ; ShouldNotReachHere" %}
  ins_encode( form2_illtrap() );
  ins_pipe(tail_call);
%}

// ============================================================================
// 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
// not zero for a miss or zero for a hit.  The encoding ALSO sets flags.
instruct partialSubtypeCheck( o0RegP index, o1RegP sub, o2RegP super, flagsRegP pcc, o7RegP o7 ) %{
  match(Set index (PartialSubtypeCheck sub super));
  effect( KILL pcc, KILL o7 );
  ins_cost(DEFAULT_COST*10);
  format %{ "CALL   PartialSubtypeCheck\n\tNOP" %}
  ins_encode( enc_PartialSubtypeCheck() );
  ins_pipe(partial_subtype_check_pipe);
%}

instruct partialSubtypeCheck_vs_zero( flagsRegP pcc, o1RegP sub, o2RegP super, immP0 zero, o0RegP idx, o7RegP o7 ) %{
  match(Set pcc (CmpP (PartialSubtypeCheck sub super) zero));
  effect( KILL idx, KILL o7 );
  ins_cost(DEFAULT_COST*10);
  format %{ "CALL   PartialSubtypeCheck\n\tNOP\t# (sets condition codes)" %}
  ins_encode( enc_PartialSubtypeCheck() );
  ins_pipe(partial_subtype_check_pipe);
%}


// ============================================================================
// inlined locking and unlocking

instruct cmpFastLock(flagsRegP pcc, iRegP object, iRegP box, iRegP scratch2, o7RegP scratch ) %{
  match(Set pcc (FastLock object box));

  effect(KILL scratch, TEMP scratch2);
  ins_cost(100);

  size(4*112);       // conservative overestimation ...
  format %{ "FASTLOCK  $object, $box; KILL $scratch, $scratch2, $box" %}
  ins_encode( Fast_Lock(object, box, scratch, scratch2) );
  ins_pipe(long_memory_op);
%}


instruct cmpFastUnlock(flagsRegP pcc, iRegP object, iRegP box, iRegP scratch2, o7RegP scratch ) %{
  match(Set pcc (FastUnlock object box));
  effect(KILL scratch, TEMP scratch2);
  ins_cost(100);

  size(4*120);       // conservative overestimation ...
  format %{ "FASTUNLOCK  $object, $box; KILL $scratch, $scratch2, $box" %}
  ins_encode( Fast_Unlock(object, box, scratch, scratch2) );
  ins_pipe(long_memory_op);
%}

// Count and Base registers are fixed because the allocator cannot
// kill unknown registers.  The encodings are generic.
instruct clear_array(iRegX cnt, iRegP base, iRegX temp, Universe dummy, flagsReg ccr) %{
  match(Set dummy (ClearArray cnt base));
  effect(TEMP temp, KILL ccr);
  ins_cost(300);
  format %{ "MOV    $cnt,$temp\n"
    "loop:   SUBcc  $temp,8,$temp\t! Count down a dword of bytes\n"
    "        BRge   loop\t\t! Clearing loop\n"
    "        STX    G0,[$base+$temp]\t! delay slot" %}
  ins_encode( enc_Clear_Array(cnt, base, temp) );
  ins_pipe(long_memory_op);
%}

instruct string_compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result,
                        o7RegI tmp, flagsReg ccr) %{
  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ccr, KILL tmp);
  ins_cost(300);
  format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result   // KILL $tmp" %}
  ins_encode( enc_String_Compare(str1, str2, cnt1, cnt2, result) );
  ins_pipe(long_memory_op);
%}

instruct string_equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result,
                       o7RegI tmp, flagsReg ccr) %{
  match(Set result (StrEquals (Binary str1 str2) cnt));
  effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp, KILL ccr);
  ins_cost(300);
  format %{ "String Equals $str1,$str2,$cnt -> $result   // KILL $tmp" %}
  ins_encode( enc_String_Equals(str1, str2, cnt, result) );
  ins_pipe(long_memory_op);
%}

instruct array_equals(o0RegP ary1, o1RegP ary2, g3RegI tmp1, notemp_iRegI result,
                      o7RegI tmp2, flagsReg ccr) %{
  match(Set result (AryEq ary1 ary2));
  effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL ccr);
  ins_cost(300);
  format %{ "Array Equals $ary1,$ary2 -> $result   // KILL $tmp1,$tmp2" %}
  ins_encode( enc_Array_Equals(ary1, ary2, tmp1, result));
  ins_pipe(long_memory_op);
%}


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

instruct countLeadingZerosI(iRegI dst, iRegI src, iRegI tmp, flagsReg cr) %{
  predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
  match(Set dst (CountLeadingZerosI src));
  effect(TEMP dst, TEMP tmp, KILL cr);

  // x |= (x >> 1);
  // x |= (x >> 2);
  // x |= (x >> 4);
  // x |= (x >> 8);
  // x |= (x >> 16);
  // return (WORDBITS - popc(x));
  format %{ "SRL     $src,1,$tmp\t! count leading zeros (int)\n\t"
            "SRL     $src,0,$dst\t! 32-bit zero extend\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRL     $dst,2,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRL     $dst,4,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRL     $dst,8,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRL     $dst,16,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "POPC    $dst,$dst\n\t"
            "MOV     32,$tmp\n\t"
            "SUB     $tmp,$dst,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Register Rtmp = $tmp$$Register;
    __ srl(Rsrc, 1, Rtmp);
    __ srl(Rsrc, 0, Rdst);
    __ or3(Rdst, Rtmp, Rdst);
    __ srl(Rdst, 2, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srl(Rdst, 4, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srl(Rdst, 8, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srl(Rdst, 16, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ popc(Rdst, Rdst);
    __ mov(BitsPerInt, Rtmp);
    __ sub(Rtmp, Rdst, Rdst);
  %}
  ins_pipe(ialu_reg);
%}

instruct countLeadingZerosL(iRegI dst, iRegL src, iRegL tmp, flagsReg cr) %{
  predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
  match(Set dst (CountLeadingZerosL src));
  effect(TEMP dst, TEMP tmp, KILL cr);

  // x |= (x >> 1);
  // x |= (x >> 2);
  // x |= (x >> 4);
  // x |= (x >> 8);
  // x |= (x >> 16);
  // x |= (x >> 32);
  // return (WORDBITS - popc(x));
  format %{ "SRLX    $src,1,$tmp\t! count leading zeros (long)\n\t"
            "OR      $src,$tmp,$dst\n\t"
            "SRLX    $dst,2,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRLX    $dst,4,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRLX    $dst,8,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRLX    $dst,16,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "SRLX    $dst,32,$tmp\n\t"
            "OR      $dst,$tmp,$dst\n\t"
            "POPC    $dst,$dst\n\t"
            "MOV     64,$tmp\n\t"
            "SUB     $tmp,$dst,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Register Rtmp = $tmp$$Register;
    __ srlx(Rsrc, 1, Rtmp);
    __ or3(Rsrc, Rtmp, Rdst);
    __ srlx(Rdst, 2, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srlx(Rdst, 4, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srlx(Rdst, 8, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srlx(Rdst, 16, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ srlx(Rdst, 32, Rtmp);
    __ or3(Rdst, Rtmp, Rdst);
    __ popc(Rdst, Rdst);
    __ mov(BitsPerLong, Rtmp);
    __ sub(Rtmp, Rdst, Rdst);
  %}
  ins_pipe(ialu_reg);
%}

instruct countTrailingZerosI(iRegI dst, iRegI src, flagsReg cr) %{
  predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
  match(Set dst (CountTrailingZerosI src));
  effect(TEMP dst, KILL cr);

  // return popc(~x & (x - 1));
  format %{ "SUB     $src,1,$dst\t! count trailing zeros (int)\n\t"
            "ANDN    $dst,$src,$dst\n\t"
            "SRL     $dst,R_G0,$dst\n\t"
            "POPC    $dst,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    __ sub(Rsrc, 1, Rdst);
    __ andn(Rdst, Rsrc, Rdst);
    __ srl(Rdst, G0, Rdst);
    __ popc(Rdst, Rdst);
  %}
  ins_pipe(ialu_reg);
%}

instruct countTrailingZerosL(iRegI dst, iRegL src, flagsReg cr) %{
  predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
  match(Set dst (CountTrailingZerosL src));
  effect(TEMP dst, KILL cr);

  // return popc(~x & (x - 1));
  format %{ "SUB     $src,1,$dst\t! count trailing zeros (long)\n\t"
            "ANDN    $dst,$src,$dst\n\t"
            "POPC    $dst,$dst" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    __ sub(Rsrc, 1, Rdst);
    __ andn(Rdst, Rsrc, Rdst);
    __ popc(Rdst, Rdst);
  %}
  ins_pipe(ialu_reg);
%}


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

instruct popCountI(iRegI dst, iRegI src) %{
  predicate(UsePopCountInstruction);
  match(Set dst (PopCountI src));

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

// Note: Long.bitCount(long) returns an int.
instruct popCountL(iRegI dst, iRegL src) %{
  predicate(UsePopCountInstruction);
  match(Set dst (PopCountL src));

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


// ============================================================================
//------------Bytes reverse--------------------------------------------------

instruct bytes_reverse_int(iRegI dst, stackSlotI src) %{
  match(Set dst (ReverseBytesI src));

  // Op cost is artificially doubled to make sure that load or store
  // instructions are preferred over this one which requires a spill
  // onto a stack slot.
  ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
  format %{ "LDUWA  $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ set($src$$disp + STACK_BIAS, O7);
    __ lduwa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe( iload_mem );
%}

instruct bytes_reverse_long(iRegL dst, stackSlotL src) %{
  match(Set dst (ReverseBytesL src));

  // Op cost is artificially doubled to make sure that load or store
  // instructions are preferred over this one which requires a spill
  // onto a stack slot.
  ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
  format %{ "LDXA   $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ set($src$$disp + STACK_BIAS, O7);
    __ ldxa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe( iload_mem );
%}

instruct bytes_reverse_unsigned_short(iRegI dst, stackSlotI src) %{
  match(Set dst (ReverseBytesUS src));

  // Op cost is artificially doubled to make sure that load or store
  // instructions are preferred over this one which requires a spill
  // onto a stack slot.
  ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
  format %{ "LDUHA  $src, $dst\t!asi=primary_little\n\t" %}

  ins_encode %{
    // the value was spilled as an int so bias the load
    __ set($src$$disp + STACK_BIAS + 2, O7);
    __ lduha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe( iload_mem );
%}

instruct bytes_reverse_short(iRegI dst, stackSlotI src) %{
  match(Set dst (ReverseBytesS src));

  // Op cost is artificially doubled to make sure that load or store
  // instructions are preferred over this one which requires a spill
  // onto a stack slot.
  ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
  format %{ "LDSHA  $src, $dst\t!asi=primary_little\n\t" %}

  ins_encode %{
    // the value was spilled as an int so bias the load
    __ set($src$$disp + STACK_BIAS + 2, O7);
    __ ldsha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe( iload_mem );
%}

// Load Integer reversed byte order
instruct loadI_reversed(iRegI dst, indIndexMemory src) %{
  match(Set dst (ReverseBytesI (LoadI src)));

  ins_cost(DEFAULT_COST + MEMORY_REF_COST);
  size(4);
  format %{ "LDUWA  $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ lduwa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load Long - aligned and reversed
instruct loadL_reversed(iRegL dst, indIndexMemory src) %{
  match(Set dst (ReverseBytesL (LoadL src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDXA   $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ ldxa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load unsigned short / char reversed byte order
instruct loadUS_reversed(iRegI dst, indIndexMemory src) %{
  match(Set dst (ReverseBytesUS (LoadUS src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDUHA  $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ lduha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Load short reversed byte order
instruct loadS_reversed(iRegI dst, indIndexMemory src) %{
  match(Set dst (ReverseBytesS (LoadS src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LDSHA  $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ ldsha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
  %}
  ins_pipe(iload_mem);
%}

// Store Integer reversed byte order
instruct storeI_reversed(indIndexMemory dst, iRegI src) %{
  match(Set dst (StoreI dst (ReverseBytesI src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STWA   $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ stwa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
  %}
  ins_pipe(istore_mem_reg);
%}

// Store Long reversed byte order
instruct storeL_reversed(indIndexMemory dst, iRegL src) %{
  match(Set dst (StoreL dst (ReverseBytesL src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STXA   $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ stxa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
  %}
  ins_pipe(istore_mem_reg);
%}

// Store unsighed short/char reversed byte order
instruct storeUS_reversed(indIndexMemory dst, iRegI src) %{
  match(Set dst (StoreC dst (ReverseBytesUS src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STHA   $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
  %}
  ins_pipe(istore_mem_reg);
%}

// Store short reversed byte order
instruct storeS_reversed(indIndexMemory dst, iRegI src) %{
  match(Set dst (StoreC dst (ReverseBytesS src)));

  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STHA   $src, $dst\t!asi=primary_little" %}

  ins_encode %{
    __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
  %}
  ins_pipe(istore_mem_reg);
%}

//----------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(eRegI dst, eRegI src) %{
//   match(Set dst (CopyI src));
// %}
//
// instruct incI_eReg(eRegI 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 ( incI_eReg_immI1( 0.dst 1.src 0.src ) );
// %}
//

// // Change load of spilled value to only a spill
// instruct storeI(memory mem, eRegI src) %{
//   match(Set mem (StoreI mem src));
// %}
//
// instruct loadI(eRegI 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.
//
// SPARC will probably not have any of these rules due to RISC instruction set.

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