s390.ad

//
// Copyright (c) 2017, 2021, Oracle and/or its affiliates. All rights reserved.
// Copyright (c) 2017, 2020 SAP SE. 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.
//

// z/Architecture Architecture Description File

// Major contributions by AS, JL, LS.

//
// Following information is derived from private mail communication
// (Oct. 2011).
//
// General branch target alignment considerations
//
// z/Architecture does not imply a general branch target alignment requirement.
// There are side effects and side considerations, though, which may
// provide some performance benefit. These are:
//  - Align branch target on octoword (32-byte) boundary
//    On more recent models (from z9 on), I-fetch is done on a Octoword
//    (32 bytes at a time) basis. To avoid I-fetching unnecessary
//    instructions, branch targets should be 32-byte aligend. If this
//    exact alingment cannot be achieved, having the branch target in
//    the first doubleword still provides some benefit.
//  - Avoid branch targets at the end of cache lines (> 64 bytes distance).
//    Sequential instruction prefetching after the branch target starts
//    immediately after having fetched the octoword containing the
//    branch target. When I-fetching crosses a cache line, there may be
//    a small stall. The worst case: the branch target (at the end of
//    a cache line) is a L1 I-cache miss and the next line as well.
//    Then, the entire target line must be filled first (to contine at the
//    branch target). Only then can the next sequential line be filled.
//  - Avoid multiple poorly predicted branches in a row.
//

//----------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
// architecture.

register %{

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

// z/Architecture register definitions, based on the z/Architecture Principles
// of Operation, 5th Edition, September 2005, and z/Linux Elf ABI Supplement,
// 5th Edition, March 2001.
//
// For each 64-bit register we must define two registers: the register
// itself, e.g. Z_R3, and a corresponding virtual other (32-bit-)'half',
// e.g. Z_R3_H, which is needed by the allocator, but is not used
// for stores, loads, etc.

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

  // z/Architecture has 16 64-bit integer registers.

  // types: v = volatile, nv = non-volatile, s = system
  reg_def Z_R0   (SOC, SOC, Op_RegI,  0, Z_R0->as_VMReg());   // v   scratch1
  reg_def Z_R0_H (SOC, SOC, Op_RegI, 99, Z_R0->as_VMReg()->next());
  reg_def Z_R1   (SOC, SOC, Op_RegI,  1, Z_R1->as_VMReg());   // v   scratch2
  reg_def Z_R1_H (SOC, SOC, Op_RegI, 99, Z_R1->as_VMReg()->next());
  reg_def Z_R2   (SOC, SOC, Op_RegI,  2, Z_R2->as_VMReg());   // v   iarg1 & iret
  reg_def Z_R2_H (SOC, SOC, Op_RegI, 99, Z_R2->as_VMReg()->next());
  reg_def Z_R3   (SOC, SOC, Op_RegI,  3, Z_R3->as_VMReg());   // v   iarg2
  reg_def Z_R3_H (SOC, SOC, Op_RegI, 99, Z_R3->as_VMReg()->next());
  reg_def Z_R4   (SOC, SOC, Op_RegI,  4, Z_R4->as_VMReg());   // v   iarg3
  reg_def Z_R4_H (SOC, SOC, Op_RegI, 99, Z_R4->as_VMReg()->next());
  reg_def Z_R5   (SOC, SOC, Op_RegI,  5, Z_R5->as_VMReg());   // v   iarg4
  reg_def Z_R5_H (SOC, SOC, Op_RegI, 99, Z_R5->as_VMReg()->next());
  reg_def Z_R6   (SOC, SOE, Op_RegI,  6, Z_R6->as_VMReg());   // v   iarg5
  reg_def Z_R6_H (SOC, SOE, Op_RegI, 99, Z_R6->as_VMReg()->next());
  reg_def Z_R7   (SOC, SOE, Op_RegI,  7, Z_R7->as_VMReg());
  reg_def Z_R7_H (SOC, SOE, Op_RegI, 99, Z_R7->as_VMReg()->next());
  reg_def Z_R8   (SOC, SOE, Op_RegI,  8, Z_R8->as_VMReg());
  reg_def Z_R8_H (SOC, SOE, Op_RegI, 99, Z_R8->as_VMReg()->next());
  reg_def Z_R9   (SOC, SOE, Op_RegI,  9, Z_R9->as_VMReg());
  reg_def Z_R9_H (SOC, SOE, Op_RegI, 99, Z_R9->as_VMReg()->next());
  reg_def Z_R10  (SOC, SOE, Op_RegI, 10, Z_R10->as_VMReg());
  reg_def Z_R10_H(SOC, SOE, Op_RegI, 99, Z_R10->as_VMReg()->next());
  reg_def Z_R11  (SOC, SOE, Op_RegI, 11, Z_R11->as_VMReg());
  reg_def Z_R11_H(SOC, SOE, Op_RegI, 99, Z_R11->as_VMReg()->next());
  reg_def Z_R12  (SOC, SOE, Op_RegI, 12, Z_R12->as_VMReg());
  reg_def Z_R12_H(SOC, SOE, Op_RegI, 99, Z_R12->as_VMReg()->next());
  reg_def Z_R13  (SOC, SOE, Op_RegI, 13, Z_R13->as_VMReg());
  reg_def Z_R13_H(SOC, SOE, Op_RegI, 99, Z_R13->as_VMReg()->next());
  reg_def Z_R14  (NS,  NS,  Op_RegI, 14, Z_R14->as_VMReg());   // s  return_pc
  reg_def Z_R14_H(NS,  NS,  Op_RegI, 99, Z_R14->as_VMReg()->next());
  reg_def Z_R15  (NS,  NS,  Op_RegI, 15, Z_R15->as_VMReg());   // s  SP
  reg_def Z_R15_H(NS,  NS,  Op_RegI, 99, Z_R15->as_VMReg()->next());

  // Float/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.

  // z/Architecture has 16 64-bit floating-point registers. Each can store a single
  // or double precision floating-point value.

  // types: v = volatile, nv = non-volatile, s = system
  reg_def Z_F0   (SOC, SOC, Op_RegF,  0, Z_F0->as_VMReg());   // v   farg1 & fret
  reg_def Z_F0_H (SOC, SOC, Op_RegF, 99, Z_F0->as_VMReg()->next());
  reg_def Z_F1   (SOC, SOC, Op_RegF,  1, Z_F1->as_VMReg());
  reg_def Z_F1_H (SOC, SOC, Op_RegF, 99, Z_F1->as_VMReg()->next());
  reg_def Z_F2   (SOC, SOC, Op_RegF,  2, Z_F2->as_VMReg());   // v   farg2
  reg_def Z_F2_H (SOC, SOC, Op_RegF, 99, Z_F2->as_VMReg()->next());
  reg_def Z_F3   (SOC, SOC, Op_RegF,  3, Z_F3->as_VMReg());
  reg_def Z_F3_H (SOC, SOC, Op_RegF, 99, Z_F3->as_VMReg()->next());
  reg_def Z_F4   (SOC, SOC, Op_RegF,  4, Z_F4->as_VMReg());   // v   farg3
  reg_def Z_F4_H (SOC, SOC, Op_RegF, 99, Z_F4->as_VMReg()->next());
  reg_def Z_F5   (SOC, SOC, Op_RegF,  5, Z_F5->as_VMReg());
  reg_def Z_F5_H (SOC, SOC, Op_RegF, 99, Z_F5->as_VMReg()->next());
  reg_def Z_F6   (SOC, SOC, Op_RegF,  6, Z_F6->as_VMReg());
  reg_def Z_F6_H (SOC, SOC, Op_RegF, 99, Z_F6->as_VMReg()->next());
  reg_def Z_F7   (SOC, SOC, Op_RegF,  7, Z_F7->as_VMReg());
  reg_def Z_F7_H (SOC, SOC, Op_RegF, 99, Z_F7->as_VMReg()->next());
  reg_def Z_F8   (SOC, SOE, Op_RegF,  8, Z_F8->as_VMReg());
  reg_def Z_F8_H (SOC, SOE, Op_RegF, 99, Z_F8->as_VMReg()->next());
  reg_def Z_F9   (SOC, SOE, Op_RegF,  9, Z_F9->as_VMReg());
  reg_def Z_F9_H (SOC, SOE, Op_RegF, 99, Z_F9->as_VMReg()->next());
  reg_def Z_F10  (SOC, SOE, Op_RegF, 10, Z_F10->as_VMReg());
  reg_def Z_F10_H(SOC, SOE, Op_RegF, 99, Z_F10->as_VMReg()->next());
  reg_def Z_F11  (SOC, SOE, Op_RegF, 11, Z_F11->as_VMReg());
  reg_def Z_F11_H(SOC, SOE, Op_RegF, 99, Z_F11->as_VMReg()->next());
  reg_def Z_F12  (SOC, SOE, Op_RegF, 12, Z_F12->as_VMReg());
  reg_def Z_F12_H(SOC, SOE, Op_RegF, 99, Z_F12->as_VMReg()->next());
  reg_def Z_F13  (SOC, SOE, Op_RegF, 13, Z_F13->as_VMReg());
  reg_def Z_F13_H(SOC, SOE, Op_RegF, 99, Z_F13->as_VMReg()->next());
  reg_def Z_F14  (SOC, SOE, Op_RegF, 14, Z_F14->as_VMReg());
  reg_def Z_F14_H(SOC, SOE, Op_RegF, 99, Z_F14->as_VMReg()->next());
  reg_def Z_F15  (SOC, SOE, Op_RegF, 15, Z_F15->as_VMReg());
  reg_def Z_F15_H(SOC, SOE, Op_RegF, 99, Z_F15->as_VMReg()->next());


  // Special Registers

  // Condition Codes Flag Registers

  // z/Architecture has the PSW (program status word) that contains
  // (among other information) the condition code. We treat this
  // part of the PSW as a condition register CR. It consists of 4
  // bits. Floating point instructions influence the same condition register CR.

  reg_def Z_CR(SOC, SOC, Op_RegFlags, 0, Z_CR->as_VMReg());   // volatile


// Specify priority of register selection within phases of register
// allocation. Highest priority is first. A useful heuristic is to
// give registers a low priority when they are required by machine
// instructions, and choose no-save registers before save-on-call, and
// save-on-call before save-on-entry. Registers which participate in
// fix calling sequences should come last. Registers which are used
// as pairs must fall on an even boundary.

// It's worth about 1% on SPEC geomean to get this right.

// Chunk0, chunk1, and chunk2 form the MachRegisterNumbers enumeration
// in adGlobals_s390.hpp which defines the <register>_num values, e.g.
// Z_R3_num. Therefore, Z_R3_num may not be (and in reality is not)
// the same as Z_R3->encoding()! Furthermore, we cannot make any
// assumptions on ordering, e.g. Z_R3_num may be less than Z_R2_num.
// Additionally, the function
//   static enum RC rc_class(OptoReg::Name reg)
// maps a given <register>_num value to its chunk type (except for flags)
// and its current implementation relies on chunk0 and chunk1 having a
// size of 64 each.

alloc_class chunk0(
  // chunk0 contains *all* 32 integer registers halves.

  // potential SOE regs
  Z_R13,Z_R13_H,
  Z_R12,Z_R12_H,
  Z_R11,Z_R11_H,
  Z_R10,Z_R10_H,

  Z_R9,Z_R9_H,
  Z_R8,Z_R8_H,
  Z_R7,Z_R7_H,

  Z_R1,Z_R1_H,
  Z_R0,Z_R0_H,

  // argument registers
  Z_R6,Z_R6_H,
  Z_R5,Z_R5_H,
  Z_R4,Z_R4_H,
  Z_R3,Z_R3_H,
  Z_R2,Z_R2_H,

  // special registers
  Z_R14,Z_R14_H,
  Z_R15,Z_R15_H
);

alloc_class chunk1(
  // Chunk1 contains *all* 64 floating-point registers halves.

  Z_F15,Z_F15_H,
  Z_F14,Z_F14_H,
  Z_F13,Z_F13_H,
  Z_F12,Z_F12_H,
  Z_F11,Z_F11_H,
  Z_F10,Z_F10_H,
  Z_F9,Z_F9_H,
  Z_F8,Z_F8_H,
  // scratch register
  Z_F7,Z_F7_H,
  Z_F5,Z_F5_H,
  Z_F3,Z_F3_H,
  Z_F1,Z_F1_H,
  // argument registers
  Z_F6,Z_F6_H,
  Z_F4,Z_F4_H,
  Z_F2,Z_F2_H,
  Z_F0,Z_F0_H
);

alloc_class chunk2(
  Z_CR
);


//-------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 stack_slots(/* one chunk of stack-based "registers" */)

// Integer Register Classes
reg_class z_int_reg(
/*Z_R0*/              // R0
/*Z_R1*/
  Z_R2,
  Z_R3,
  Z_R4,
  Z_R5,
  Z_R6,
  Z_R7,
/*Z_R8,*/             // Z_thread
  Z_R9,
  Z_R10,
  Z_R11,
  Z_R12,
  Z_R13
/*Z_R14*/             // return_pc
/*Z_R15*/             // SP
);

reg_class z_no_odd_int_reg(
/*Z_R0*/              // R0
/*Z_R1*/
  Z_R2,
  Z_R3,
  Z_R4,
/*Z_R5,*/             // odd part of fix register pair
  Z_R6,
  Z_R7,
/*Z_R8,*/             // Z_thread
  Z_R9,
  Z_R10,
  Z_R11,
  Z_R12,
  Z_R13
/*Z_R14*/             // return_pc
/*Z_R15*/             // SP
);

reg_class z_no_arg_int_reg(
/*Z_R0*/              // R0
/*Z_R1*/              // scratch
/*Z_R2*/
/*Z_R3*/
/*Z_R4*/
/*Z_R5*/
/*Z_R6*/
  Z_R7,
/*Z_R8*/              // Z_thread
  Z_R9,
  Z_R10,
  Z_R11,
  Z_R12,
  Z_R13
/*Z_R14*/             // return_pc
/*Z_R15*/             // SP
);

reg_class z_rarg1_int_reg(Z_R2);
reg_class z_rarg2_int_reg(Z_R3);
reg_class z_rarg3_int_reg(Z_R4);
reg_class z_rarg4_int_reg(Z_R5);
reg_class z_rarg5_int_reg(Z_R6);

// Pointer Register Classes

// 64-bit build means 64-bit pointers means hi/lo pairs.

reg_class z_rarg5_ptrN_reg(Z_R6);

reg_class z_rarg1_ptr_reg(Z_R2_H,Z_R2);
reg_class z_rarg2_ptr_reg(Z_R3_H,Z_R3);
reg_class z_rarg3_ptr_reg(Z_R4_H,Z_R4);
reg_class z_rarg4_ptr_reg(Z_R5_H,Z_R5);
reg_class z_rarg5_ptr_reg(Z_R6_H,Z_R6);
reg_class z_thread_ptr_reg(Z_R8_H,Z_R8);

reg_class z_ptr_reg(
/*Z_R0_H,Z_R0*/     // R0
/*Z_R1_H,Z_R1*/
  Z_R2_H,Z_R2,
  Z_R3_H,Z_R3,
  Z_R4_H,Z_R4,
  Z_R5_H,Z_R5,
  Z_R6_H,Z_R6,
  Z_R7_H,Z_R7,
/*Z_R8_H,Z_R8,*/    // Z_thread
  Z_R9_H,Z_R9,
  Z_R10_H,Z_R10,
  Z_R11_H,Z_R11,
  Z_R12_H,Z_R12,
  Z_R13_H,Z_R13
/*Z_R14_H,Z_R14*/   // return_pc
/*Z_R15_H,Z_R15*/   // SP
);

reg_class z_lock_ptr_reg(
/*Z_R0_H,Z_R0*/     // R0
/*Z_R1_H,Z_R1*/
  Z_R2_H,Z_R2,
  Z_R3_H,Z_R3,
  Z_R4_H,Z_R4,
/*Z_R5_H,Z_R5,*/
/*Z_R6_H,Z_R6,*/
  Z_R7_H,Z_R7,
/*Z_R8_H,Z_R8,*/    // Z_thread
  Z_R9_H,Z_R9,
  Z_R10_H,Z_R10,
  Z_R11_H,Z_R11,
  Z_R12_H,Z_R12,
  Z_R13_H,Z_R13
/*Z_R14_H,Z_R14*/   // return_pc
/*Z_R15_H,Z_R15*/   // SP
);

reg_class z_no_arg_ptr_reg(
/*Z_R0_H,Z_R0*/        // R0
/*Z_R1_H,Z_R1*/        // scratch
/*Z_R2_H,Z_R2*/
/*Z_R3_H,Z_R3*/
/*Z_R4_H,Z_R4*/
/*Z_R5_H,Z_R5*/
/*Z_R6_H,Z_R6*/
  Z_R7_H, Z_R7,
/*Z_R8_H,Z_R8*/        // Z_thread
  Z_R9_H,Z_R9,
  Z_R10_H,Z_R10,
  Z_R11_H,Z_R11,
  Z_R12_H,Z_R12,
  Z_R13_H,Z_R13
/*Z_R14_H,Z_R14*/      // return_pc
/*Z_R15_H,Z_R15*/      // SP
);

// Special class for storeP instructions, which can store SP or RPC to
// TLS. (Note: Do not generalize this to "any_reg". If you add
// another register, such as FP, to this mask, the allocator may try
// to put a temp in it.)
// Register class for memory access base registers,
// This class is a superset of z_ptr_reg including Z_thread.
reg_class z_memory_ptr_reg(
/*Z_R0_H,Z_R0*/     // R0
/*Z_R1_H,Z_R1*/
  Z_R2_H,Z_R2,
  Z_R3_H,Z_R3,
  Z_R4_H,Z_R4,
  Z_R5_H,Z_R5,
  Z_R6_H,Z_R6,
  Z_R7_H,Z_R7,
  Z_R8_H,Z_R8,      // Z_thread
  Z_R9_H,Z_R9,
  Z_R10_H,Z_R10,
  Z_R11_H,Z_R11,
  Z_R12_H,Z_R12,
  Z_R13_H,Z_R13
/*Z_R14_H,Z_R14*/   // return_pc
/*Z_R15_H,Z_R15*/   // SP
);

// Other special pointer regs.
reg_class z_r1_regP(Z_R1_H,Z_R1);
reg_class z_r9_regP(Z_R9_H,Z_R9);


// Long Register Classes

reg_class z_rarg1_long_reg(Z_R2_H,Z_R2);
reg_class z_rarg2_long_reg(Z_R3_H,Z_R3);
reg_class z_rarg3_long_reg(Z_R4_H,Z_R4);
reg_class z_rarg4_long_reg(Z_R5_H,Z_R5);
reg_class z_rarg5_long_reg(Z_R6_H,Z_R6);

// Longs in 1 register. Aligned adjacent hi/lo pairs.
reg_class z_long_reg(
/*Z_R0_H,Z_R0*/     // R0
/*Z_R1_H,Z_R1*/
  Z_R2_H,Z_R2,
  Z_R3_H,Z_R3,
  Z_R4_H,Z_R4,
  Z_R5_H,Z_R5,
  Z_R6_H,Z_R6,
  Z_R7_H,Z_R7,
/*Z_R8_H,Z_R8,*/    // Z_thread
  Z_R9_H,Z_R9,
  Z_R10_H,Z_R10,
  Z_R11_H,Z_R11,
  Z_R12_H,Z_R12,
  Z_R13_H,Z_R13
/*Z_R14_H,Z_R14,*/  // return_pc
/*Z_R15_H,Z_R15*/   // SP
);

// z_long_reg without even registers
reg_class z_long_odd_reg(
/*Z_R0_H,Z_R0*/     // R0
/*Z_R1_H,Z_R1*/
  Z_R3_H,Z_R3,
  Z_R5_H,Z_R5,
  Z_R7_H,Z_R7,
  Z_R9_H,Z_R9,
  Z_R11_H,Z_R11,
  Z_R13_H,Z_R13
/*Z_R14_H,Z_R14,*/  // return_pc
/*Z_R15_H,Z_R15*/   // SP
);

// Special Class for Condition Code Flags Register

reg_class z_condition_reg(
  Z_CR
);

// Scratch register for late profiling. Callee saved.
reg_class z_rscratch2_bits64_reg(Z_R2_H, Z_R2);


// Float Register Classes

reg_class z_flt_reg(
  Z_F0,
/*Z_F1,*/ // scratch
  Z_F2,
  Z_F3,
  Z_F4,
  Z_F5,
  Z_F6,
  Z_F7,
  Z_F8,
  Z_F9,
  Z_F10,
  Z_F11,
  Z_F12,
  Z_F13,
  Z_F14,
  Z_F15
);
reg_class z_rscratch1_flt_reg(Z_F1);

// Double precision float registers have virtual `high halves' that
// are needed by the allocator.
reg_class z_dbl_reg(
  Z_F0,Z_F0_H,
/*Z_F1,Z_F1_H,*/ // scratch
  Z_F2,Z_F2_H,
  Z_F3,Z_F3_H,
  Z_F4,Z_F4_H,
  Z_F5,Z_F5_H,
  Z_F6,Z_F6_H,
  Z_F7,Z_F7_H,
  Z_F8,Z_F8_H,
  Z_F9,Z_F9_H,
  Z_F10,Z_F10_H,
  Z_F11,Z_F11_H,
  Z_F12,Z_F12_H,
  Z_F13,Z_F13_H,
  Z_F14,Z_F14_H,
  Z_F15,Z_F15_H
);
reg_class z_rscratch1_dbl_reg(Z_F1,Z_F1_H);

%}

//----------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 DEFAULT_COST_LOW  (    80,      80);
  int_def DEFAULT_COST_HIGH (   120,     120);
  int_def HUGE_COST         (1000000, 1000000);

  // Put an advantage on REG_MEM vs. MEM+REG_REG operations.
  int_def ALU_REG_COST      (   100, DEFAULT_COST);
  int_def ALU_MEMORY_COST   (   150,          150);

  // Memory refs are twice as expensive as run-of-the-mill.
  int_def MEMORY_REF_COST_HI (   220, 2 * DEFAULT_COST+20);
  int_def MEMORY_REF_COST    (   200, 2 * DEFAULT_COST);
  int_def MEMORY_REF_COST_LO (   180, 2 * DEFAULT_COST-20);

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

source %{

#ifdef PRODUCT
#define BLOCK_COMMENT(str)
#define BIND(label)        __ bind(label)
#else
#define BLOCK_COMMENT(str) __ block_comment(str)
#define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
#endif

#define __ _masm.

#define Z_DISP_SIZE Immediate::is_uimm12((long)opnd_array(1)->disp(ra_,this,2)) ?  4 : 6
#define Z_DISP3_SIZE 6

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

// Given a register encoding, produce an Integer Register object.
static Register reg_to_register_object(int register_encoding);

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

// 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.

void PhaseOutput::pd_perform_mach_node_analysis() {
}

int MachNode::pd_alignment_required() const {
  return 1;
}

int MachNode::compute_padding(int current_offset) const {
  return 0;
}

int MachCallStaticJavaNode::ret_addr_offset() {
  if (_method) {
    return 8;
  } else {
    return MacroAssembler::call_far_patchable_ret_addr_offset();
  }
}

int MachCallDynamicJavaNode::ret_addr_offset() {
  // Consider size of receiver type profiling (C2 tiers).
  int profile_receiver_type_size = 0;

  int vtable_index = this->_vtable_index;
  if (vtable_index == -4) {
    return 14 + profile_receiver_type_size;
  } else {
    assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
    return 36 + profile_receiver_type_size;
  }
}

int MachCallRuntimeNode::ret_addr_offset() {
  return 12 + MacroAssembler::call_far_patchable_ret_addr_offset();
}

int MachCallNativeNode::ret_addr_offset() {
  Unimplemented();
  return -1;
}

// Compute padding required for nodes which need alignment
//
// The addresses of the call instructions needs to be 4-byte aligned to
// ensure that they don't span a cache line so that they are atomically patchable.
// The actual calls get emitted at different offsets within the node emitters.
// ins_alignment needs to be set to 2 which means that up to 1 nop may get inserted.

int CallStaticJavaDirect_dynTOCNode::compute_padding(int current_offset) const {
  return (0 - current_offset) & 2;
}

int CallDynamicJavaDirect_dynTOCNode::compute_padding(int current_offset) const {
  return (6 - current_offset) & 2;
}

int CallRuntimeDirectNode::compute_padding(int current_offset) const {
  return (12 - current_offset) & 2;
}

int CallLeafDirectNode::compute_padding(int current_offset) const {
  return (12 - current_offset) & 2;
}

int CallLeafNoFPDirectNode::compute_padding(int current_offset) const {
  return (12 - current_offset) & 2;
}

void emit_nop(CodeBuffer &cbuf) {
  C2_MacroAssembler _masm(&cbuf);
  __ z_nop();
}

// Emit an interrupt that is caught by the debugger (for debugging compiler).
void emit_break(CodeBuffer &cbuf) {
  C2_MacroAssembler _masm(&cbuf);
  __ z_illtrap();
}

#if !defined(PRODUCT)
void MachBreakpointNode::format(PhaseRegAlloc *, outputStream *os) const {
  os->print("TA");
}
#endif

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

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

static inline void z_emit16(CodeBuffer &cbuf, long value) {
  // 32bit instructions may become sign extended.
  assert(value >= 0, "unintended sign extension (int->long)");
  assert(value < (1L << 16), "instruction too large");
  *((unsigned short*)(cbuf.insts_end())) = (unsigned short)value;
  cbuf.set_insts_end(cbuf.insts_end() + sizeof(unsigned short));
}

static inline void z_emit32(CodeBuffer &cbuf, long value) {
  // 32bit instructions may become sign extended.
  assert(value < (1L << 32), "instruction too large");
  *((unsigned int*)(cbuf.insts_end())) = (unsigned int)value;
  cbuf.set_insts_end(cbuf.insts_end() + sizeof(unsigned int));
}

static inline void z_emit48(CodeBuffer &cbuf, long value) {
  // 32bit instructions may become sign extended.
  assert(value >= 0, "unintended sign extension (int->long)");
  assert(value < (1L << 48), "instruction too large");
  value = value<<16;
  memcpy(cbuf.insts_end(), (unsigned char*)&value, 6);
  cbuf.set_insts_end(cbuf.insts_end() + 6);
}

static inline unsigned int z_emit_inst(CodeBuffer &cbuf, long value) {
  if (value < 0) {
    // There obviously has been an unintended sign extension (int->long). Revert it.
    value = (long)((unsigned long)((unsigned int)value));
  }

  if (value < (1L << 16)) { // 2-byte instruction
    z_emit16(cbuf, value);
    return 2;
  }

  if (value < (1L << 32)) { // 4-byte instruction, might be unaligned store
    z_emit32(cbuf, value);
    return 4;
  }

  // 6-byte instruction, probably unaligned store.
  z_emit48(cbuf, value);
  return 6;
}

// Check effective address (at runtime) for required alignment.
static inline void z_assert_aligned(CodeBuffer &cbuf, int disp, Register index, Register base, int alignment) {
  C2_MacroAssembler _masm(&cbuf);

  __ z_lay(Z_R0, disp, index, base);
  __ z_nill(Z_R0, alignment-1);
  __ z_brc(Assembler::bcondEqual, +3);
  __ z_illtrap();
}

int emit_call_reloc(C2_MacroAssembler &_masm, intptr_t entry_point, relocInfo::relocType rtype,
                    PhaseRegAlloc* ra_, bool is_native_call = false) {
  __ set_inst_mark(); // Used in z_enc_java_static_call() and emit_java_to_interp().
  address old_mark = __ inst_mark();
  unsigned int start_off = __ offset();

  if (is_native_call) {
    ShouldNotReachHere();
  }

  if (rtype == relocInfo::runtime_call_w_cp_type) {
    assert((__ offset() & 2) == 0, "misaligned emit_call_reloc");
    address call_addr = __ call_c_opt((address)entry_point);
    if (call_addr == NULL) {
      Compile::current()->env()->record_out_of_memory_failure();
      return -1;
    }
  } else {
    assert(rtype == relocInfo::none || rtype == relocInfo::opt_virtual_call_type ||
           rtype == relocInfo::static_call_type, "unexpected rtype");
    __ relocate(rtype);
    // BRASL must be prepended with a nop to identify it in the instruction stream.
    __ z_nop();
    __ z_brasl(Z_R14, (address)entry_point);
  }

  unsigned int ret_off = __ offset();

  return (ret_off - start_off);
}

static int emit_call_reloc(C2_MacroAssembler &_masm, intptr_t entry_point, RelocationHolder const& rspec) {
  __ set_inst_mark(); // Used in z_enc_java_static_call() and emit_java_to_interp().
  address old_mark = __ inst_mark();
  unsigned int start_off = __ offset();

  relocInfo::relocType rtype = rspec.type();
  assert(rtype == relocInfo::opt_virtual_call_type || rtype == relocInfo::static_call_type,
         "unexpected rtype");

  __ relocate(rspec);
  __ z_nop();
  __ z_brasl(Z_R14, (address)entry_point);

  unsigned int ret_off = __ offset();

  return (ret_off - start_off);
}

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

const RegMask& MachConstantBaseNode::_out_RegMask = _Z_PTR_REG_mask;
int ConstantTable::calculate_table_base_offset() const {
  return 0;  // absolute addressing, no offset
}

bool MachConstantBaseNode::requires_postalloc_expand() const { return false; }
void MachConstantBaseNode::postalloc_expand(GrowableArray <Node *> *nodes, PhaseRegAlloc *ra_) {
  ShouldNotReachHere();
}

// Even with PC-relative TOC addressing, we still need this node.
// Float loads/stores do not support PC-relative addresses.
void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
  C2_MacroAssembler _masm(&cbuf);
  Register Rtoc = as_Register(ra_->get_encode(this));
  __ load_toc(Rtoc);
}

uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
  // PCrelative TOC access.
  return 6;   // sizeof(LARL)
}

#if !defined(PRODUCT)
void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
  Register r = as_Register(ra_->get_encode(this));
  st->print("LARL    %s,&constant_pool # MachConstantBaseNode", r->name());
}
#endif

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

#if !defined(PRODUCT)
void MachPrologNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
  Compile* C = ra_->C;
  st->print_cr("--- MachPrologNode ---");
  st->print("\t");
  for (int i = 0; i < OptoPrologueNops; i++) {
    st->print_cr("NOP"); st->print("\t");
  }

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

  long framesize = C->output()->frame_size_in_bytes();
  int bangsize   = C->output()->bang_size_in_bytes();

  // 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->output()->need_stack_bang(bangsize)) {
    st->print_cr("# stack bang"); st->print("\t");
  }
  st->print_cr("push_frame %d", (int)-framesize);
  st->print("\t");
}
#endif

void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  Compile* C = ra_->C;
  C2_MacroAssembler _masm(&cbuf);

  __ verify_thread();

  size_t framesize = C->output()->frame_size_in_bytes();
  size_t bangsize  = C->output()->bang_size_in_bytes();

  assert(framesize % wordSize == 0, "must preserve wordSize alignment");

  if (C->clinit_barrier_on_entry()) {
    assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");

    Label L_skip_barrier;
    Register klass = Z_R1_scratch;

    // Notify OOP recorder (don't need the relocation)
    AddressLiteral md = __ constant_metadata_address(C->method()->holder()->constant_encoding());
    __ load_const_optimized(klass, md.value());
    __ clinit_barrier(klass, Z_thread, &L_skip_barrier /*L_fast_path*/);

    __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub());
    __ z_br(klass);

    __ bind(L_skip_barrier);
  }

  // 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->output()->need_stack_bang(bangsize)) {
    __ generate_stack_overflow_check(bangsize);
  }

  assert(Immediate::is_uimm32((long)framesize), "to do: choose suitable types!");
  __ save_return_pc();

  // The z/Architecture abi is already accounted for in `framesize' via the
  // 'out_preserve_stack_slots' declaration.
  __ push_frame((unsigned int)framesize/*includes JIT ABI*/);

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

uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
  // Variable size. Determine dynamically.
  return MachNode::size(ra_);
}

int MachPrologNode::reloc() const {
  // Return number of relocatable values contained in this instruction.
  return 1; // One reloc entry for load_const(toc).
}

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

#if !defined(PRODUCT)
void MachEpilogNode::format(PhaseRegAlloc *ra_, outputStream *os) const {
  os->print_cr("epilog");
  os->print("\t");
  if (do_polling() && ra_->C->is_method_compilation()) {
    os->print_cr("load_from_polling_page Z_R1_scratch");
    os->print("\t");
  }
}
#endif

void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  C2_MacroAssembler _masm(&cbuf);
  Compile* C = ra_->C;
  __ verify_thread();

  // If this does safepoint polling, then do it here.
  bool need_polling = do_polling() && C->is_method_compilation();

  // Pop frame, restore return_pc, and all stuff needed by interpreter.
  int frame_size_in_bytes = Assembler::align((C->output()->frame_slots() << LogBytesPerInt), frame::alignment_in_bytes);
  __ pop_frame_restore_retPC(frame_size_in_bytes);

  if (StackReservedPages > 0 && C->has_reserved_stack_access()) {
    __ reserved_stack_check(Z_R14);
  }

  // Touch the polling page.
  if (need_polling) {
    __ z_lg(Z_R1_scratch, Address(Z_thread, JavaThread::polling_page_offset()));
    // We need to mark the code position where the load from the safepoint
    // polling page was emitted as relocInfo::poll_return_type here.
    __ relocate(relocInfo::poll_return_type);
    __ load_from_polling_page(Z_R1_scratch);
  }
}

uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
  // Variable size. determine dynamically.
  return MachNode::size(ra_);
}

int MachEpilogNode::reloc() const {
  // Return number of relocatable values contained in this instruction.
  return 1; // One for load_from_polling_page.
}

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

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

// 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) {
  // Return the register class for the given register. The given register
  // reg is a <register>_num value, which is an index into the MachRegisterNumbers
  // enumeration in adGlobals_s390.hpp.

  if (reg == OptoReg::Bad) {
    return rc_bad;
  }

  // We have 32 integer register halves, starting at index 0.
  if (reg < 32) {
    return rc_int;
  }

  // We have 32 floating-point register halves, starting at index 32.
  if (reg < 32+32) {
    return rc_float;
  }

  // Between float regs & stack are the flags regs.
  assert(reg >= OptoReg::stack0(), "blow up if spilling flags");
  return rc_stack;
}

// Returns size as obtained from z_emit_instr.
static unsigned int z_ld_st_helper(CodeBuffer *cbuf, const char *op_str, unsigned long opcode,
                                   int reg, int offset, bool do_print, outputStream *os) {

  if (cbuf) {
    if (opcode > (1L<<32)) {
      return z_emit_inst(*cbuf, opcode | Assembler::reg(Matcher::_regEncode[reg], 8, 48) |
                         Assembler::simm20(offset) | Assembler::reg(Z_R0, 12, 48) | Assembler::regz(Z_SP, 16, 48));
    } else {
      return z_emit_inst(*cbuf, opcode | Assembler::reg(Matcher::_regEncode[reg], 8, 32) |
                         Assembler::uimm12(offset, 20, 32) | Assembler::reg(Z_R0, 12, 32) | Assembler::regz(Z_SP, 16, 32));
    }
  }

#if !defined(PRODUCT)
  if (do_print) {
    os->print("%s    %s,#%d[,SP]\t # MachCopy spill code",op_str, Matcher::regName[reg], offset);
  }
#endif
  return (opcode > (1L << 32)) ? 6 : 4;
}

static unsigned int z_mvc_helper(CodeBuffer *cbuf, int len, int dst_off, int src_off, bool do_print, outputStream *os) {
  if (cbuf) {
    C2_MacroAssembler _masm(cbuf);
    __ z_mvc(dst_off, len-1, Z_SP, src_off, Z_SP);
  }

#if !defined(PRODUCT)
  else if (do_print) {
    os->print("MVC     %d(%d,SP),%d(SP)\t # MachCopy spill code",dst_off, len, src_off);
  }
#endif

  return 6;
}

uint MachSpillCopyNode::implementation(CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, outputStream *os) const {
  // Get registers to move.
  OptoReg::Name src_hi = ra_->get_reg_second(in(1));
  OptoReg::Name src_lo = ra_->get_reg_first(in(1));
  OptoReg::Name dst_hi = ra_->get_reg_second(this);
  OptoReg::Name dst_lo = ra_->get_reg_first(this);

  enum RC src_hi_rc = rc_class(src_hi);
  enum RC src_lo_rc = rc_class(src_lo);
  enum RC dst_hi_rc = rc_class(dst_hi);
  enum RC dst_lo_rc = rc_class(dst_lo);

  assert(src_lo != OptoReg::Bad && dst_lo != OptoReg::Bad, "must move at least 1 register");
  bool is64 = (src_hi_rc != rc_bad);
  assert(!is64 ||
         ((src_lo&1) == 0 && src_lo+1 == src_hi && (dst_lo&1) == 0 && dst_lo+1 == dst_hi),
         "expected aligned-adjacent pairs");

  // Generate spill code!

  if (src_lo == dst_lo && src_hi == dst_hi) {
    return 0;            // Self copy, no move.
  }

  int  src_offset = ra_->reg2offset(src_lo);
  int  dst_offset = ra_->reg2offset(dst_lo);
  bool print = !do_size;
  bool src12 = Immediate::is_uimm12(src_offset);
  bool dst12 = Immediate::is_uimm12(dst_offset);

  const char   *mnemo = NULL;
  unsigned long opc = 0;

  // Memory->Memory Spill. Use Z_R0 to hold the value.
  if (src_lo_rc == rc_stack && dst_lo_rc == rc_stack) {

    assert(!is64 || (src_hi_rc==rc_stack && dst_hi_rc==rc_stack),
           "expected same type of move for high parts");

    if (src12 && dst12) {
      return z_mvc_helper(cbuf, is64 ? 8 : 4, dst_offset, src_offset, print, os);
    }

    int r0 = Z_R0_num;
    if (is64) {
      return z_ld_st_helper(cbuf, "LG  ", LG_ZOPC, r0, src_offset, print, os) +
             z_ld_st_helper(cbuf, "STG ", STG_ZOPC, r0, dst_offset, print, os);
    }

    return z_ld_st_helper(cbuf, "LY   ", LY_ZOPC, r0, src_offset, print, os) +
           z_ld_st_helper(cbuf, "STY  ", STY_ZOPC, r0, dst_offset, print, os);
  }

  // Check for float->int copy. Requires a trip through memory.
  if (src_lo_rc == rc_float && dst_lo_rc == rc_int) {
    Unimplemented();  // Unsafe, do not remove!
  }

  // Check for integer reg-reg copy.
  if (src_lo_rc == rc_int && dst_lo_rc == rc_int) {
    if (cbuf) {
      C2_MacroAssembler _masm(cbuf);
      Register Rsrc = as_Register(Matcher::_regEncode[src_lo]);
      Register Rdst = as_Register(Matcher::_regEncode[dst_lo]);
      __ z_lgr(Rdst, Rsrc);
      return 4;
    }
#if !defined(PRODUCT)
    // else
    if (print) {
      os->print("LGR     %s,%s\t # MachCopy spill code", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
    }
#endif
    return 4;
  }

  // Check for integer store.
  if (src_lo_rc == rc_int && dst_lo_rc == rc_stack) {
    assert(!is64 || (src_hi_rc==rc_int && dst_hi_rc==rc_stack),
           "expected same type of move for high parts");

    if (is64) {
      return z_ld_st_helper(cbuf, "STG ", STG_ZOPC, src_lo, dst_offset, print, os);
    }

    // else
    mnemo = dst12 ? "ST  " : "STY ";
    opc = dst12 ? ST_ZOPC : STY_ZOPC;

    return z_ld_st_helper(cbuf, mnemo, opc, src_lo, dst_offset, print, os);
  }

  // Check for integer load
  // Always load cOops zero-extended. That doesn't hurt int loads.
  if (dst_lo_rc == rc_int && src_lo_rc == rc_stack) {

    assert(!is64 || (dst_hi_rc==rc_int && src_hi_rc==rc_stack),
           "expected same type of move for high parts");

    mnemo = is64 ? "LG  " : "LLGF";
    opc = is64 ? LG_ZOPC : LLGF_ZOPC;

    return z_ld_st_helper(cbuf, mnemo, opc, dst_lo, src_offset, print, os);
  }

  // Check for float reg-reg copy.
  if (src_lo_rc == rc_float && dst_lo_rc == rc_float) {
    if (cbuf) {
      C2_MacroAssembler _masm(cbuf);
      FloatRegister Rsrc = as_FloatRegister(Matcher::_regEncode[src_lo]);
      FloatRegister Rdst = as_FloatRegister(Matcher::_regEncode[dst_lo]);
      __ z_ldr(Rdst, Rsrc);
      return 2;
    }
#if !defined(PRODUCT)
    // else
    if (print) {
      os->print("LDR      %s,%s\t # MachCopy spill code", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
    }
#endif
    return 2;
  }

  // Check for float store.
  if (src_lo_rc == rc_float && dst_lo_rc == rc_stack) {
    assert(!is64 || (src_hi_rc==rc_float && dst_hi_rc==rc_stack),
           "expected same type of move for high parts");

    if (is64) {
      mnemo = dst12 ? "STD  " : "STDY ";
      opc = dst12 ? STD_ZOPC : STDY_ZOPC;
      return z_ld_st_helper(cbuf, mnemo, opc, src_lo, dst_offset, print, os);
    }
    // else

    mnemo = dst12 ? "STE  " : "STEY ";
    opc = dst12 ? STE_ZOPC : STEY_ZOPC;
    return z_ld_st_helper(cbuf, mnemo, opc, src_lo, dst_offset, print, os);
  }

  // Check for float load.
  if (dst_lo_rc == rc_float && src_lo_rc == rc_stack) {
    assert(!is64 || (dst_hi_rc==rc_float && src_hi_rc==rc_stack),
           "expected same type of move for high parts");

    if (is64) {
      mnemo = src12 ? "LD   " : "LDY  ";
      opc = src12 ? LD_ZOPC : LDY_ZOPC;
      return z_ld_st_helper(cbuf, mnemo, opc, dst_lo, src_offset, print, os);
    }
    // else

    mnemo = src12 ? "LE   " : "LEY  ";
    opc = src12 ? LE_ZOPC : LEY_ZOPC;
    return z_ld_st_helper(cbuf, mnemo, opc, dst_lo, src_offset, print, os);
  }

  // --------------------------------------------------------------------
  // Check for hi bits still needing moving. Only happens for misaligned
  // arguments to native calls.
  if (src_hi == dst_hi) {
    return 0;               // Self copy, no move.
  }

  assert(is64 && dst_hi_rc != rc_bad, "src_hi & dst_hi cannot be Bad");
  Unimplemented();  // Unsafe, do not remove!

  return 0; // never reached, but make the compiler shut up!
}

#if !defined(PRODUCT)
void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream *os) const {
  if (ra_ && ra_->node_regs_max_index() > 0) {
    implementation(NULL, ra_, false, os);
  } else {
    if (req() == 2 && in(1)) {
      os->print("N%d = N%d\n", _idx, in(1)->_idx);
    } else {
      const char *c = "(";
      os->print("N%d = ", _idx);
      for (uint i = 1; i < req(); ++i) {
        os->print("%sN%d", c, in(i)->_idx);
        c = ", ";
      }
      os->print(")");
    }
  }
}
#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);
}

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

#if !defined(PRODUCT)
void MachNopNode::format(PhaseRegAlloc *, outputStream *os) const {
  os->print("NOP     # pad for alignment (%d nops, %d bytes)", _count, _count*MacroAssembler::nop_size());
}
#endif

void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc * ra_) const {
  C2_MacroAssembler _masm(&cbuf);

  int rem_space = 0;
  if (!(ra_->C->output()->in_scratch_emit_size())) {
    rem_space = cbuf.insts()->remaining();
    if (rem_space <= _count*2 + 8) {
      tty->print("NopNode: _count = %3.3d, remaining space before = %d", _count, rem_space);
    }
  }

  for (int i = 0; i < _count; i++) {
    __ z_nop();
  }

  if (!(ra_->C->output()->in_scratch_emit_size())) {
    if (rem_space <= _count*2 + 8) {
      int rem_space2 = cbuf.insts()->remaining();
      tty->print_cr(", after = %d", rem_space2);
    }
  }
}

uint MachNopNode::size(PhaseRegAlloc *ra_) const {
   return 2 * _count;
}

#if !defined(PRODUCT)
void BoxLockNode::format(PhaseRegAlloc *ra_, outputStream *os) const {
  int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  if (ra_ && ra_->node_regs_max_index() > 0) {
    int reg = ra_->get_reg_first(this);
    os->print("ADDHI  %s, SP, %d\t//box node", Matcher::regName[reg], offset);
  } else {
    os->print("ADDHI  N%d = SP + %d\t// box node", _idx, offset);
  }
}
#endif

// Take care of the size function, if you make changes here!
void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  C2_MacroAssembler _masm(&cbuf);

  int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  int reg = ra_->get_encode(this);
  __ z_lay(as_Register(reg), offset, Z_SP);
}

uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
  // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_)
  return 6;
}

 %} // end source section

//----------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 %{

// Header information of the source block.
// Method declarations/definitions which are used outside
// the ad-scope can conveniently be defined here.
//
// To keep related declarations/definitions/uses close together,
// we switch between source %{ }% and source_hpp %{ }% freely as needed.

#include "oops/klass.inline.hpp"

//--------------------------------------------------------------
// Used for optimization in Compile::Shorten_branches
//--------------------------------------------------------------

class CallStubImpl {
 public:

  // call trampolines
  // Size of call trampoline stub. For add'l comments, see size_java_to_interp().
  static uint size_call_trampoline() {
    return 0; // no call trampolines on this platform
  }

  // call trampolines
  // Number of relocations needed by a call trampoline stub.
  static uint reloc_call_trampoline() {
    return 0; // No call trampolines on this platform.
  }
};

%} // end source_hpp section

source %{

#if !defined(PRODUCT)
void MachUEPNode::format(PhaseRegAlloc *ra_, outputStream *os) const {
  os->print_cr("---- MachUEPNode ----");
  os->print_cr("\tTA");
  os->print_cr("\tload_const Z_R1, SharedRuntime::get_ic_miss_stub()");
  os->print_cr("\tBR(Z_R1)");
  os->print_cr("\tTA  # pad with illtraps");
  os->print_cr("\t...");
  os->print_cr("\tTA");
  os->print_cr("\tLTGR    Z_R2, Z_R2");
  os->print_cr("\tBRU     ic_miss");
}
#endif

void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
  C2_MacroAssembler _masm(&cbuf);
  const int ic_miss_offset = 2;

  // Inline_cache contains a klass.
  Register ic_klass = as_Register(Matcher::inline_cache_reg_encode());
  // ARG1 is the receiver oop.
  Register R2_receiver = Z_ARG1;
  int      klass_offset = oopDesc::klass_offset_in_bytes();
  AddressLiteral icmiss(SharedRuntime::get_ic_miss_stub());
  Register R1_ic_miss_stub_addr = Z_R1_scratch;

  // Null check of receiver.
  // This is the null check of the receiver that actually should be
  // done in the caller. It's here because in case of implicit null
  // checks we get it for free.
  assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()),
         "second word in oop should not require explicit null check.");
  if (!ImplicitNullChecks) {
    Label valid;
    if (VM_Version::has_CompareBranch()) {
      __ z_cgij(R2_receiver, 0, Assembler::bcondNotEqual, valid);
    } else {
      __ z_ltgr(R2_receiver, R2_receiver);
      __ z_bre(valid);
    }
    // The ic_miss_stub will handle the null pointer exception.
    __ load_const_optimized(R1_ic_miss_stub_addr, icmiss);
    __ z_br(R1_ic_miss_stub_addr);
    __ bind(valid);
  }

  // Check whether this method is the proper implementation for the class of
  // the receiver (ic miss check).
  {
    Label valid;
    // Compare cached class against klass from receiver.
    // This also does an implicit null check!
    __ compare_klass_ptr(ic_klass, klass_offset, R2_receiver, false);
    __ z_bre(valid);
    // The inline cache points to the wrong method. Call the
    // ic_miss_stub to find the proper method.
    __ load_const_optimized(R1_ic_miss_stub_addr, icmiss);
    __ z_br(R1_ic_miss_stub_addr);
    __ bind(valid);
  }
}

uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
  // Determine size dynamically.
  return MachNode::size(ra_);
}

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

%} // interrupt source section

source_hpp %{ // Header information of the source block.

class HandlerImpl {
 public:

  static int emit_exception_handler(CodeBuffer &cbuf);
  static int emit_deopt_handler(CodeBuffer& cbuf);

  static uint size_exception_handler() {
    return NativeJump::max_instruction_size();
  }

  static uint size_deopt_handler() {
    return NativeCall::max_instruction_size();
  }
};

class Node::PD {
public:
  enum NodeFlags {
    _last_flag = Node::_last_flag
  };
};

%} // end source_hpp section

source %{

// This exception handler code snippet is placed after the method's
// code. It is the return point if an exception occurred. it jumps to
// the exception blob.
//
// If the method gets deoptimized, the method and this code snippet
// get patched.
//
// 1) Trampoline code gets patched into the end of this exception
//   handler. the trampoline code jumps to the deoptimization blob.
//
// 2) The return address in the method's code will get patched such
//   that it jumps to the trampoline.
//
// 3) The handler will get patched such that it does not jump to the
//   exception blob, but to an entry in the deoptimization blob being
//   aware of the exception.
int HandlerImpl::emit_exception_handler(CodeBuffer &cbuf) {
  Register temp_reg = Z_R1;
  C2_MacroAssembler _masm(&cbuf);

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

  int offset = __ offset();
  // Use unconditional pc-relative jump with 32-bit range here.
  __ load_const_optimized(temp_reg, (address)OptoRuntime::exception_blob()->content_begin());
  __ z_br(temp_reg);

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

  __ end_a_stub();

  return offset;
}

// Emit deopt handler code.
int HandlerImpl::emit_deopt_handler(CodeBuffer& cbuf) {
  C2_MacroAssembler _masm(&cbuf);
  address        base = __ start_a_stub(size_deopt_handler());

  if (base == NULL) {
    return 0;  // CodeBuffer::expand failed
  }

  int offset = __ offset();

  // Size_deopt_handler() must be exact on zarch, so for simplicity
  // we do not use load_const_opt here.
  __ load_const(Z_R1, SharedRuntime::deopt_blob()->unpack());
  __ call(Z_R1);
  assert(__ offset() - offset == (int) size_deopt_handler(), "must be fixed size");

  __ end_a_stub();
  return offset;
}

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


// Given a register encoding, produce an Integer Register object.
static Register reg_to_register_object(int register_encoding) {
  assert(Z_R12->encoding() == Z_R12_enc, "wrong coding");
  return as_Register(register_encoding);
}

const bool Matcher::match_rule_supported(int opcode) {
  if (!has_match_rule(opcode)) {
    return false; // no match rule present
  }

  switch (opcode) {
    case Op_ReverseBytesI:
    case Op_ReverseBytesL:
      return UseByteReverseInstruction;
    case Op_PopCountI:
    case Op_PopCountL:
      // PopCount supported by H/W from z/Architecture G5 (z196) on.
      return (UsePopCountInstruction && VM_Version::has_PopCount());
  }

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

const bool Matcher::match_rule_supported_vector(int opcode, int vlen, BasicType bt) {
  if (!match_rule_supported(opcode) || !vector_size_supported(bt, vlen)) {
    return false;
  }
  return true; // Per default match rules are supported.
}

const RegMask* Matcher::predicate_reg_mask(void) {
  return NULL;
}

const TypeVect* Matcher::predicate_reg_type(const Type* elemTy, int length) {
  return NULL;
}

// Vector calling convention not yet implemented.
const bool Matcher::supports_vector_calling_convention(void) {
  return false;
}

OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
  Unimplemented();
  return OptoRegPair(0, 0);
}

//----------SUPERWORD HELPERS----------------------------------------

// Vector width in bytes.
const int Matcher::vector_width_in_bytes(BasicType bt) {
  assert(MaxVectorSize == 8, "");
  return 8;
}

// Vector ideal reg.
const uint Matcher::vector_ideal_reg(int size) {
  assert(MaxVectorSize == 8 && size == 8, "");
  return Op_RegL;
}

// Limits on vector size (number of elements) loaded into vector.
const int Matcher::max_vector_size(const BasicType bt) {
  assert(is_java_primitive(bt), "only primitive type vectors");
  return vector_width_in_bytes(bt)/type2aelembytes(bt);
}

const int Matcher::min_vector_size(const BasicType bt) {
  return max_vector_size(bt); // Same as max.
}

const int Matcher::scalable_vector_reg_size(const BasicType bt) {
  return -1;
}

// RETURNS: whether this branch offset is short enough that a short
// branch can be used.
//
// If the platform does not provide any short branch variants, then
// this method should return `false' for offset 0.
//
// `Compile::Fill_buffer' will decide on basis of this information
// whether to do the pass `Compile::Shorten_branches' at all.
//
// And `Compile::Shorten_branches' will decide on basis of this
// information whether to replace particular branch sites by short
// ones.
bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
  // On zarch short branches use a 16 bit signed immediate that
  // is the pc-relative offset in halfword (= 2 bytes) units.
  return Assembler::is_within_range_of_RelAddr16((address)((long)offset), (address)0);
}

MachOper* Matcher::pd_specialize_generic_vector_operand(MachOper* original_opnd, uint ideal_reg, bool is_temp) {
  ShouldNotReachHere(); // generic vector operands not supported
  return NULL;
}

bool Matcher::is_reg2reg_move(MachNode* m) {
  ShouldNotReachHere();  // generic vector operands not supported
  return false;
}

bool Matcher::is_generic_vector(MachOper* opnd)  {
  ShouldNotReachHere();  // generic vector operands not supported
  return false;
}

// Constants for c2c and c calling conventions.

const MachRegisterNumbers z_iarg_reg[5] = {
  Z_R2_num, Z_R3_num, Z_R4_num, Z_R5_num, Z_R6_num
};

const MachRegisterNumbers z_farg_reg[4] = {
  Z_F0_num, Z_F2_num, Z_F4_num, Z_F6_num
};

const int z_num_iarg_registers = sizeof(z_iarg_reg) / sizeof(z_iarg_reg[0]);

const int z_num_farg_registers = sizeof(z_farg_reg) / sizeof(z_farg_reg[0]);

// 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) {
  // We return true for all registers contained in z_iarg_reg[] and
  // z_farg_reg[] and their virtual halves.
  // We must include the virtual halves in order to get STDs and LDs
  // instead of STWs and LWs in the trampoline stubs.

  if (reg == Z_R2_num || reg == Z_R2_H_num ||
      reg == Z_R3_num || reg == Z_R3_H_num ||
      reg == Z_R4_num || reg == Z_R4_H_num ||
      reg == Z_R5_num || reg == Z_R5_H_num ||
      reg == Z_R6_num || reg == Z_R6_H_num) {
    return true;
  }

  if (reg == Z_F0_num || reg == Z_F0_H_num ||
      reg == Z_F2_num || reg == Z_F2_H_num ||
      reg == Z_F4_num || reg == Z_F4_H_num ||
      reg == Z_F6_num || reg == Z_F6_H_num) {
    return true;
  }

  return false;
}

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

uint Matcher::int_pressure_limit()
{
  // Medium size register set, 6 special purpose regs, 3 SOE regs.
  return (INTPRESSURE == -1) ? 10 : INTPRESSURE;
}

uint Matcher::float_pressure_limit()
{
  return (FLOATPRESSURE == -1) ? 15 : FLOATPRESSURE;
}

bool Matcher::use_asm_for_ldiv_by_con(jlong divisor) {
  return false;
}

// Register for DIVI projection of divmodI
RegMask Matcher::divI_proj_mask() {
  return _Z_RARG4_INT_REG_mask;
}

// Register for MODI projection of divmodI
RegMask Matcher::modI_proj_mask() {
  return _Z_RARG3_INT_REG_mask;
}

// Register for DIVL projection of divmodL
RegMask Matcher::divL_proj_mask() {
  return _Z_RARG4_LONG_REG_mask;
}

// Register for MODL projection of divmodL
RegMask Matcher::modL_proj_mask() {
  return _Z_RARG3_LONG_REG_mask;
}

// Copied from sparc.
const RegMask Matcher::method_handle_invoke_SP_save_mask() {
  return RegMask();
}

// Should the matcher clone input 'm' of node 'n'?
bool Matcher::pd_clone_node(Node* n, Node* m, Matcher::MStack& mstack) {
  return false;
}

// Should the Matcher clone shifts on addressing modes, expecting them
// to be subsumed into complex addressing expressions or compute them
// into registers?
bool Matcher::pd_clone_address_expressions(AddPNode* m, Matcher::MStack& mstack, VectorSet& address_visited) {
  return clone_base_plus_offset_address(m, mstack, address_visited);
}

%} // source

//----------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_unimplemented %{
    C2_MacroAssembler _masm(&cbuf);
    __ unimplemented("Unimplemented mach node encoding in AD file.", 13);
  %}

  enc_class enc_untested %{
#ifdef ASSERT
    C2_MacroAssembler _masm(&cbuf);
    __ untested("Untested mach node encoding in AD file.");
#endif
  %}

  enc_class z_rrform(iRegI dst, iRegI src) %{
    assert((($primary >> 14) & 0x03) == 0, "Instruction format error");
    assert( ($primary >> 16)         == 0, "Instruction format error");
    z_emit16(cbuf, $primary |
             Assembler::reg($dst$$reg,8,16) |
             Assembler::reg($src$$reg,12,16));
  %}

  enc_class z_rreform(iRegI dst1, iRegI src2) %{
    assert((($primary >> 30) & 0x03) == 2, "Instruction format error");
    z_emit32(cbuf, $primary |
             Assembler::reg($dst1$$reg,24,32) |
             Assembler::reg($src2$$reg,28,32));
  %}

  enc_class z_rrfform(iRegI dst1, iRegI src2, iRegI src3) %{
    assert((($primary >> 30) & 0x03) == 2, "Instruction format error");
    z_emit32(cbuf, $primary |
             Assembler::reg($dst1$$reg,24,32) |
             Assembler::reg($src2$$reg,28,32) |
             Assembler::reg($src3$$reg,16,32));
  %}

  enc_class z_riform_signed(iRegI dst, immI16 src) %{
    assert((($primary>>30) & 0x03) == 2, "Instruction format error");
    z_emit32(cbuf, $primary |
             Assembler::reg($dst$$reg,8,32) |
             Assembler::simm16($src$$constant,16,32));
  %}

  enc_class z_riform_unsigned(iRegI dst, uimmI16 src) %{
    assert((($primary>>30) & 0x03) == 2, "Instruction format error");
    z_emit32(cbuf, $primary |
             Assembler::reg($dst$$reg,8,32) |
             Assembler::uimm16($src$$constant,16,32));
  %}

  enc_class z_rieform_d(iRegI dst1, iRegI src3, immI src2) %{
    assert((($primary>>46) & 0x03) == 3, "Instruction format error");
    z_emit48(cbuf, $primary |
             Assembler::reg($dst1$$reg,8,48) |
             Assembler::reg($src3$$reg,12,48) |
             Assembler::simm16($src2$$constant,16,48));
  %}

  enc_class z_rilform_signed(iRegI dst, immL32 src) %{
    assert((($primary>>46) & 0x03) == 3, "Instruction format error");
    z_emit48(cbuf, $primary |
             Assembler::reg($dst$$reg,8,48) |
             Assembler::simm32($src$$constant,16,48));
  %}

  enc_class z_rilform_unsigned(iRegI dst, uimmL32 src) %{
    assert((($primary>>46) & 0x03) == 3, "Instruction format error");
    z_emit48(cbuf, $primary |
             Assembler::reg($dst$$reg,8,48) |
             Assembler::uimm32($src$$constant,16,48));
  %}

  enc_class z_rsyform_const(iRegI dst, iRegI src1, immI src2) %{
    z_emit48(cbuf, $primary |
             Assembler::reg($dst$$reg,8,48) |
             Assembler::reg($src1$$reg,12,48) |
             Assembler::simm20($src2$$constant));
  %}

  enc_class z_rsyform_reg_reg(iRegI dst, iRegI src, iRegI shft) %{
    z_emit48(cbuf, $primary |
             Assembler::reg($dst$$reg,8,48) |
             Assembler::reg($src$$reg,12,48) |
             Assembler::reg($shft$$reg,16,48) |
             Assembler::simm20(0));
  %}

  enc_class z_rxform_imm_reg_reg(iRegL dst, immL con, iRegL src1, iRegL src2) %{
    assert((($primary>>30) & 0x03) == 1, "Instruction format error");
    z_emit32(cbuf, $primary |
             Assembler::reg($dst$$reg,8,32) |
             Assembler::reg($src1$$reg,12,32) |
             Assembler::reg($src2$$reg,16,32) |
             Assembler::uimm12($con$$constant,20,32));
  %}

  enc_class z_rxform_imm_reg(iRegL dst, immL con, iRegL src) %{
    assert((($primary>>30) & 0x03) == 1, "Instruction format error");
    z_emit32(cbuf, $primary |
             Assembler::reg($dst$$reg,8,32) |
             Assembler::reg($src$$reg,16,32) |
             Assembler::uimm12($con$$constant,20,32));
  %}

  enc_class z_rxyform_imm_reg_reg(iRegL dst, immL con, iRegL src1, iRegL src2) %{
    z_emit48(cbuf, $primary |
             Assembler::reg($dst$$reg,8,48) |
             Assembler::reg($src1$$reg,12,48) |
             Assembler::reg($src2$$reg,16,48) |
             Assembler::simm20($con$$constant));
  %}

  enc_class z_rxyform_imm_reg(iRegL dst, immL con, iRegL src) %{
    z_emit48(cbuf, $primary |
             Assembler::reg($dst$$reg,8,48) |
             Assembler::reg($src$$reg,16,48) |
             Assembler::simm20($con$$constant));
  %}

  // Direct memory arithmetic.
  enc_class z_siyform(memoryRSY mem, immI8 src) %{
    int      disp = $mem$$disp;
    Register base = reg_to_register_object($mem$$base);
    int      con  = $src$$constant;

    assert(VM_Version::has_MemWithImmALUOps(), "unsupported CPU");
    z_emit_inst(cbuf, $primary |
                Assembler::regz(base,16,48) |
                Assembler::simm20(disp) |
                Assembler::simm8(con,8,48));
  %}

  enc_class z_silform(memoryRS mem, immI16 src) %{
    z_emit_inst(cbuf, $primary |
                Assembler::regz(reg_to_register_object($mem$$base),16,48) |
                Assembler::uimm12($mem$$disp,20,48) |
                Assembler::simm16($src$$constant,32,48));
  %}

  // Encoder for FP ALU reg/mem instructions (support only short displacements).
  enc_class z_form_rt_memFP(RegF dst, memoryRX mem) %{
    Register Ridx = $mem$$index$$Register;
    if (Ridx == noreg) { Ridx = Z_R0; } // Index is 0.
    if ($primary > (1L << 32)) {
      z_emit_inst(cbuf, $primary |
                  Assembler::reg($dst$$reg, 8, 48) |
                  Assembler::uimm12($mem$$disp, 20, 48) |
                  Assembler::reg(Ridx, 12, 48) |
                  Assembler::regz(reg_to_register_object($mem$$base), 16, 48));
    } else {
      z_emit_inst(cbuf, $primary |
                  Assembler::reg($dst$$reg, 8, 32) |
                  Assembler::uimm12($mem$$disp, 20, 32) |
                  Assembler::reg(Ridx, 12, 32) |
                  Assembler::regz(reg_to_register_object($mem$$base), 16, 32));
    }
  %}

  enc_class z_form_rt_mem(iRegI dst, memory mem) %{
    Register Ridx = $mem$$index$$Register;
    if (Ridx == noreg) { Ridx = Z_R0; } // Index is 0.
    if ($primary > (1L<<32)) {
      z_emit_inst(cbuf, $primary |
                  Assembler::reg($dst$$reg, 8, 48) |
                  Assembler::simm20($mem$$disp) |
                  Assembler::reg(Ridx, 12, 48) |
                  Assembler::regz(reg_to_register_object($mem$$base), 16, 48));
    } else {
      z_emit_inst(cbuf, $primary |
                  Assembler::reg($dst$$reg, 8, 32) |
                  Assembler::uimm12($mem$$disp, 20, 32) |
                  Assembler::reg(Ridx, 12, 32) |
                  Assembler::regz(reg_to_register_object($mem$$base), 16, 32));
    }
  %}

  enc_class z_form_rt_mem_opt(iRegI dst, memory mem) %{
    int isize = $secondary > 1L << 32 ? 48 : 32;
    Register Ridx = $mem$$index$$Register;
    if (Ridx == noreg) { Ridx = Z_R0; } // Index is 0.

    if (Displacement::is_shortDisp((long)$mem$$disp)) {
      z_emit_inst(cbuf, $secondary |
                  Assembler::reg($dst$$reg, 8, isize) |
                  Assembler::uimm12($mem$$disp, 20, isize) |
                  Assembler::reg(Ridx, 12, isize) |
                  Assembler::regz(reg_to_register_object($mem$$base), 16, isize));
    } else if (Displacement::is_validDisp((long)$mem$$disp)) {
      z_emit_inst(cbuf, $primary |
                  Assembler::reg($dst$$reg, 8, 48) |
                  Assembler::simm20($mem$$disp) |
                  Assembler::reg(Ridx, 12, 48) |
                  Assembler::regz(reg_to_register_object($mem$$base), 16, 48));
    } else {
        C2_MacroAssembler _masm(&cbuf);
        __ load_const_optimized(Z_R1_scratch, $mem$$disp);
        if (Ridx != Z_R0) { __ z_agr(Z_R1_scratch, Ridx); }
        z_emit_inst(cbuf, $secondary |
                    Assembler::reg($dst$$reg, 8, isize) |
                    Assembler::uimm12(0, 20, isize) |
                    Assembler::reg(Z_R1_scratch, 12, isize) |
                    Assembler::regz(reg_to_register_object($mem$$base), 16, isize));
    }
  %}

  enc_class z_enc_brul(Label lbl) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);
    __ z_brul(l);
  %}

  enc_class z_enc_bru(Label lbl) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);
    __ z_bru(l);
  %}

  enc_class z_enc_branch_con_far(cmpOp cmp, Label lbl) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);
    __ z_brcl((Assembler::branch_condition)$cmp$$cmpcode, l);
  %}

  enc_class z_enc_branch_con_short(cmpOp cmp, Label lbl) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);
    __ z_brc((Assembler::branch_condition)$cmp$$cmpcode, l);
  %}

  enc_class z_enc_cmpb_regreg(iRegI src1, iRegI src2, Label lbl, cmpOpT cmp) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);
    Assembler::branch_condition cc = (Assembler::branch_condition)$cmp$$cmpcode;
    unsigned long instr = $primary;
    if (instr == CRJ_ZOPC) {
      __ z_crj($src1$$Register, $src2$$Register, cc, l);
    } else if (instr == CLRJ_ZOPC) {
      __ z_clrj($src1$$Register, $src2$$Register, cc, l);
    } else if (instr == CGRJ_ZOPC) {
      __ z_cgrj($src1$$Register, $src2$$Register, cc, l);
    } else {
      guarantee(instr == CLGRJ_ZOPC, "opcode not implemented");
      __ z_clgrj($src1$$Register, $src2$$Register, cc, l);
    }
  %}

  enc_class z_enc_cmpb_regregFar(iRegI src1, iRegI src2, Label lbl, cmpOpT cmp) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);

    unsigned long instr = $primary;
    if (instr == CR_ZOPC) {
      __ z_cr($src1$$Register, $src2$$Register);
    } else if (instr == CLR_ZOPC) {
      __ z_clr($src1$$Register, $src2$$Register);
    } else if (instr == CGR_ZOPC) {
      __ z_cgr($src1$$Register, $src2$$Register);
    } else {
      guarantee(instr == CLGR_ZOPC, "opcode not implemented");
      __ z_clgr($src1$$Register, $src2$$Register);
    }

    __ z_brcl((Assembler::branch_condition)$cmp$$cmpcode, l);
  %}

  enc_class z_enc_cmpb_regimm(iRegI src1, immI8 src2, Label lbl, cmpOpT cmp) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);

    Assembler::branch_condition cc = (Assembler::branch_condition)$cmp$$cmpcode;
    unsigned long instr = $primary;
    if (instr == CIJ_ZOPC) {
      __ z_cij($src1$$Register, $src2$$constant, cc, l);
    } else if (instr == CLIJ_ZOPC) {
      __ z_clij($src1$$Register, $src2$$constant, cc, l);
    } else if (instr == CGIJ_ZOPC) {
      __ z_cgij($src1$$Register, $src2$$constant, cc, l);
    } else {
      guarantee(instr == CLGIJ_ZOPC, "opcode not implemented");
      __ z_clgij($src1$$Register, $src2$$constant, cc, l);
    }
  %}

  enc_class z_enc_cmpb_regimmFar(iRegI src1, immI8 src2, Label lbl, cmpOpT cmp) %{
    C2_MacroAssembler _masm(&cbuf);
    Label* p = $lbl$$label;

    // 'p' is `NULL' when this encoding class is used only to
    // determine the size of the encoded instruction.
    // Use a bound dummy label in that case.
    Label d;
    __ bind(d);
    Label& l = (NULL == p) ? d : *(p);

    unsigned long instr = $primary;
    if (instr == CHI_ZOPC) {
      __ z_chi($src1$$Register, $src2$$constant);
    } else if (instr == CLFI_ZOPC) {
      __ z_clfi($src1$$Register, $src2$$constant);
    } else if (instr == CGHI_ZOPC) {
      __ z_cghi($src1$$Register, $src2$$constant);
    } else {
      guarantee(instr == CLGFI_ZOPC, "opcode not implemented");
      __ z_clgfi($src1$$Register, $src2$$constant);
    }

    __ z_brcl((Assembler::branch_condition)$cmp$$cmpcode, l);
  %}

  // Call from Java to runtime.
  enc_class z_enc_java_to_runtime_call(method meth) %{
    C2_MacroAssembler _masm(&cbuf);

    // Save return pc before call to the place where we need it, since
    // callee doesn't.
    unsigned int start_off = __ offset();
    // Compute size of "larl + stg + call_c_opt".
    const int size_of_code = 6 + 6 + MacroAssembler::call_far_patchable_size();
    __ get_PC(Z_R14, size_of_code);
    __ save_return_pc();
    assert(__ offset() - start_off == 12, "bad prelude len: %d", __ offset() - start_off);

    assert((__ offset() & 2) == 0, "misaligned z_enc_java_to_runtime_call");
    address call_addr = __ call_c_opt((address)$meth$$method);
    if (call_addr == NULL) {
      Compile::current()->env()->record_out_of_memory_failure();
      return;
    }

#ifdef ASSERT
    // Plausibility check for size_of_code assumptions.
    unsigned int actual_ret_off = __ offset();
    assert(start_off + size_of_code == actual_ret_off, "wrong return_pc");
#endif
  %}

  enc_class z_enc_java_static_call(method meth) %{
    // Call to fixup routine. Fixup routine uses ScopeDesc info to determine
    // whom we intended to call.
    C2_MacroAssembler _masm(&cbuf);
    int ret_offset = 0;

    if (!_method) {
      ret_offset = emit_call_reloc(_masm, $meth$$method,
                                   relocInfo::runtime_call_w_cp_type, ra_);
    } else {
      int method_index = resolved_method_index(cbuf);
      if (_optimized_virtual) {
        ret_offset = emit_call_reloc(_masm, $meth$$method,
                                     opt_virtual_call_Relocation::spec(method_index));
      } else {
        ret_offset = emit_call_reloc(_masm, $meth$$method,
                                     static_call_Relocation::spec(method_index));
      }
    }
    assert(__ inst_mark() != NULL, "emit_call_reloc must set_inst_mark()");

    if (_method) { // Emit stub for static call.
      address stub = CompiledStaticCall::emit_to_interp_stub(cbuf);
      if (stub == NULL) {
        ciEnv::current()->record_failure("CodeCache is full");
        return;
      }
    }
  %}

  // Java dynamic call
  enc_class z_enc_java_dynamic_call(method meth) %{
    C2_MacroAssembler _masm(&cbuf);
    unsigned int start_off = __ offset();

    int vtable_index = this->_vtable_index;
    if (vtable_index == -4) {
      Register ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
      address virtual_call_oop_addr = NULL;

      AddressLiteral empty_ic((address) Universe::non_oop_word());
      virtual_call_oop_addr = __ pc();
      bool success = __ load_const_from_toc(ic_reg, empty_ic);
      if (!success) {
        Compile::current()->env()->record_out_of_memory_failure();
        return;
      }

      // Call to fixup routine. Fixup routine uses ScopeDesc info
      // to determine who we intended to call.
      int method_index = resolved_method_index(cbuf);
      __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr, method_index));
      unsigned int ret_off = __ offset();
      assert(__ offset() - start_off == 6, "bad prelude len: %d", __ offset() - start_off);
      ret_off += emit_call_reloc(_masm, $meth$$method, relocInfo::none, ra_);
      assert(_method, "lazy_constant may be wrong when _method==null");
    } else {
      assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
      // Go through the vtable. Get receiver klass. Receiver already
      // checked for non-null. If we'll go thru a C2I adapter, the
      // interpreter expects method in Z_method.
      // Use Z_method to temporarily hold the klass oop.
      // Z_R1_scratch is destroyed.
      __ load_klass(Z_method, Z_R2);

      int entry_offset = in_bytes(Klass::vtable_start_offset()) + vtable_index * vtableEntry::size_in_bytes();
      int v_off        = entry_offset + vtableEntry::method_offset_in_bytes();

      if (Displacement::is_validDisp(v_off) ) {
        // Can use load instruction with large offset.
        __ z_lg(Z_method, Address(Z_method /*class oop*/, v_off /*method offset*/));
      } else {
        // Worse case, must load offset into register.
        __ load_const(Z_R1_scratch, v_off);
        __ z_lg(Z_method, Address(Z_method /*class oop*/, Z_R1_scratch /*method offset*/));
      }
      // 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.
      __ z_lg(Z_R1_scratch, Address(Z_method, Method::from_compiled_offset()));
      // Call target. Either compiled code or C2I adapter.
      __ z_basr(Z_R14, Z_R1_scratch);
      unsigned int ret_off = __ offset();
    }
  %}

  enc_class z_enc_cmov_reg(cmpOp cmp, iRegI dst, iRegI src) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rdst = reg_to_register_object($dst$$reg);
    Register Rsrc = reg_to_register_object($src$$reg);

    // Don't emit code if operands are identical (same register).
    if (Rsrc != Rdst) {
      Assembler::branch_condition cc = (Assembler::branch_condition)$cmp$$cmpcode;

      if (VM_Version::has_LoadStoreConditional()) {
        __ z_locgr(Rdst, Rsrc, cc);
      } else {
        // Branch if not (cmp cr).
        Label done;
        __ z_brc(Assembler::inverse_condition(cc), done);
        __ z_lgr(Rdst, Rsrc); // Used for int and long+ptr.
        __ bind(done);
      }
    }
  %}

  enc_class z_enc_cmov_imm(cmpOp cmp, iRegI dst, immI16 src) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rdst = reg_to_register_object($dst$$reg);
    int      Csrc = $src$$constant;
    Assembler::branch_condition cc = (Assembler::branch_condition)$cmp$$cmpcode;
    Label done;
    // Branch if not (cmp cr).
    __ z_brc(Assembler::inverse_condition(cc), done);
    if (Csrc == 0) {
      // Don't set CC.
      __ clear_reg(Rdst, true, false);  // Use for int, long & ptr.
    } else {
      __ z_lghi(Rdst, Csrc); // Use for int, long & ptr.
    }
    __ bind(done);
  %}

  enc_class z_enc_cctobool(iRegI res) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rres = reg_to_register_object($res$$reg);

    if (VM_Version::has_LoadStoreConditional()) {
      __ load_const_optimized(Z_R0_scratch, 0L); // false (failed)
      __ load_const_optimized(Rres, 1L);         // true  (succeed)
      __ z_locgr(Rres, Z_R0_scratch, Assembler::bcondNotEqual);
    } else {
      Label done;
      __ load_const_optimized(Rres, 0L); // false (failed)
      __ z_brne(done);                   // Assume true to be the common case.
      __ load_const_optimized(Rres, 1L); // true  (succeed)
      __ bind(done);
    }
  %}

  enc_class z_enc_casI(iRegI compare_value, iRegI exchange_value, iRegP addr_ptr) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rcomp = reg_to_register_object($compare_value$$reg);
    Register Rnew  = reg_to_register_object($exchange_value$$reg);
    Register Raddr = reg_to_register_object($addr_ptr$$reg);

    __ z_cs(Rcomp, Rnew, 0, Raddr);
  %}

  enc_class z_enc_casL(iRegL compare_value, iRegL exchange_value, iRegP addr_ptr) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rcomp = reg_to_register_object($compare_value$$reg);
    Register Rnew  = reg_to_register_object($exchange_value$$reg);
    Register Raddr = reg_to_register_object($addr_ptr$$reg);

    __ z_csg(Rcomp, Rnew, 0, Raddr);
  %}

  enc_class z_enc_SwapI(memoryRSY mem, iRegI dst, iRegI tmp) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rdst = reg_to_register_object($dst$$reg);
    Register Rtmp = reg_to_register_object($tmp$$reg);
    guarantee(Rdst != Rtmp, "Fix match rule to use TEMP_DEF");
    Label    retry;

    // Iterate until swap succeeds.
    __ z_llgf(Rtmp, $mem$$Address);  // current contents
    __ bind(retry);
      // Calculate incremented value.
      __ z_csy(Rtmp, Rdst, $mem$$Address); // Try to store new value.
      __ z_brne(retry);                    // Yikes, concurrent update, need to retry.
    __ z_lgr(Rdst, Rtmp);                  // Exchanged value from memory is return value.
  %}

  enc_class z_enc_SwapL(memoryRSY mem, iRegL dst, iRegL tmp) %{
    C2_MacroAssembler _masm(&cbuf);
    Register Rdst = reg_to_register_object($dst$$reg);
    Register Rtmp = reg_to_register_object($tmp$$reg);
    guarantee(Rdst != Rtmp, "Fix match rule to use TEMP_DEF");
    Label    retry;

    // Iterate until swap succeeds.
    __ z_lg(Rtmp, $mem$$Address);  // current contents
    __ bind(retry);
      // Calculate incremented value.
      __ z_csg(Rtmp, Rdst, $mem$$Address); // Try to store new value.
      __ z_brne(retry);                    // Yikes, concurrent update, need to retry.
    __ z_lgr(Rdst, Rtmp);                  // Exchanged value from memory is return value.
  %}

%} // encode

source %{

  // Check whether outs are all Stores. If so, we can omit clearing the upper
  // 32 bits after encoding.
  static bool all_outs_are_Stores(const Node *n) {
    for (DUIterator_Fast imax, k = n->fast_outs(imax); k < imax; k++) {
      Node *out = n->fast_out(k);
      if (!out->is_Mach() || out->as_Mach()->ideal_Opcode() != Op_StoreN) {
        // Most other outs are SpillCopy, but there are various other.
        // jvm98 has arond 9% Encodes where we return false.
        return false;
      }
    }
    return true;
  }

%} // source


//----------FRAME--------------------------------------------------------------
// Definition of frame structure and management information.

frame %{
  // These two registers define part of the calling convention between
  // compiled code and the interpreter.

  // Inline Cache Register
  inline_cache_reg(Z_R9); // Z_inline_cache

  // Argument pointer for I2C adapters
  //
  // Tos is loaded in run_compiled_code to Z_ARG5=Z_R6.
  // interpreter_arg_ptr_reg(Z_R6);

  // Optional: name the operand used by cisc-spilling to access
  // [stack_pointer + offset].
  cisc_spilling_operand_name(indOffset12);

  // Number of stack slots consumed by a Monitor enter.
  sync_stack_slots(frame::jit_monitor_size_in_4_byte_units);

  // Compiled code's Frame Pointer
  //
  // z/Architecture stack pointer
  frame_pointer(Z_R15); // Z_SP

  // Interpreter stores its frame pointer in a register which is
  // stored to the stack by I2CAdaptors. I2CAdaptors convert from
  // interpreted java to compiled java.
  //
  // Z_state holds pointer to caller's cInterpreter.
  interpreter_frame_pointer(Z_R7); // Z_state

  // Use alignment_in_bytes instead of log_2_of_alignment_in_bits.
  stack_alignment(frame::alignment_in_bytes);

  // A `slot' is assumed 4 bytes here!
  // out_preserve_stack_slots(frame::jit_out_preserve_size_in_4_byte_units);

  // Number of outgoing stack slots killed above the
  // out_preserve_stack_slots for calls to C. Supports the var-args
  // backing area for register parms.
  varargs_C_out_slots_killed(((frame::z_abi_160_size - frame::z_jit_out_preserve_size) / VMRegImpl::stack_slot_size));

  // 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 Z_R14);

  // 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.
  //
  // Use register pair for c return value.
  c_return_value %{
    assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values");
    static int typeToRegLo[Op_RegL+1] = { 0, 0, Z_R2_num, Z_R2_num, Z_R2_num, Z_F0_num, Z_F0_num, Z_R2_num };
    static int typeToRegHi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, Z_R2_H_num, OptoReg::Bad, Z_F0_H_num, Z_R2_H_num };
    return OptoRegPair(typeToRegHi[ideal_reg], typeToRegLo[ideal_reg]);
  %}

  // Use register pair for return value.
  // Location of compiled Java return values. Same as C
  return_value %{
    assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values");
    static int typeToRegLo[Op_RegL+1] = { 0, 0, Z_R2_num, Z_R2_num, Z_R2_num, Z_F0_num, Z_F0_num, Z_R2_num };
    static int typeToRegHi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, Z_R2_H_num, OptoReg::Bad, Z_F0_H_num, Z_R2_H_num };
    return OptoRegPair(typeToRegHi[ideal_reg], typeToRegLo[ideal_reg]);
  %}
%}


//----------ATTRIBUTES---------------------------------------------------------

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

//----------Instruction Attributes---------------------------------------------

// Cost attribute. required.
ins_attrib ins_cost(DEFAULT_COST);

// Is this instruction a non-matching short branch variant of some
// long branch? Not required.
ins_attrib ins_short_branch(0);

// Indicates this is a trap based check node and final control-flow fixup
// must generate a proper fall through.
ins_attrib ins_is_TrapBasedCheckNode(true);

// Attribute of instruction to tell how many constants the instruction will generate.
// (optional attribute). Default: 0.
ins_attrib ins_num_consts(0);

// Required alignment attribute (must be a power of 2)
// specifies the alignment that some part of the instruction (not
// necessarily the start) requires. If > 1, a compute_padding()
// function must be provided for the instruction.
//
// WARNING: Don't use size(FIXED_SIZE) or size(VARIABLE_SIZE) in
// instructions which depend on the proper alignment, because the
// desired alignment isn't guaranteed for the call to "emit()" during
// the size computation.
ins_attrib ins_alignment(1);

// Enforce/prohibit rematerializations.
// - If an instruction is attributed with 'ins_cannot_rematerialize(true)'
//   then rematerialization of that instruction is prohibited and the
//   instruction's value will be spilled if necessary.
// - If an instruction is attributed with 'ins_should_rematerialize(true)'
//   then rematerialization is enforced and the instruction's value will
//   never get spilled. a copy of the instruction will be inserted if
//   necessary.
//   Note: this may result in rematerializations in front of every use.
// (optional attribute)
ins_attrib ins_cannot_rematerialize(false);
ins_attrib ins_should_rematerialize(false);

//----------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
// Please note:
// Formats are generated automatically for constants and base registers.

//----------------------------------------------
// SIGNED (shorter than INT) immediate operands
//----------------------------------------------

// Byte Immediate: constant 'int -1'
operand immB_minus1() %{
  //         sign-ext constant      zero-ext constant
  predicate((n->get_int() == -1) || ((n->get_int()&0x000000ff) == 0x000000ff));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Byte Immediate: constant, but not 'int 0' nor 'int -1'.
operand immB_n0m1() %{
  //                             sign-ext constant     zero-ext constant
  predicate(n->get_int() != 0 && n->get_int() != -1 && (n->get_int()&0x000000ff) != 0x000000ff);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Short Immediate: constant 'int -1'
operand immS_minus1() %{
  //         sign-ext constant      zero-ext constant
  predicate((n->get_int() == -1) || ((n->get_int()&0x0000ffff) == 0x0000ffff));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Short Immediate: constant, but not 'int 0' nor 'int -1'.
operand immS_n0m1() %{
  //                             sign-ext constant     zero-ext constant
  predicate(n->get_int() != 0 && n->get_int() != -1 && (n->get_int()&0x0000ffff) != 0x0000ffff);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//-----------------------------------------
//  SIGNED INT immediate operands
//-----------------------------------------

// Integer Immediate: 32-bit
operand immI() %{
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Int Immediate: 20-bit
operand immI20() %{
  predicate(Immediate::is_simm20(n->get_int()));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 16-bit
operand immI16() %{
  predicate(Immediate::is_simm16(n->get_int()));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 8-bit
operand immI8() %{
  predicate(Immediate::is_simm8(n->get_int()));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: constant 'int 0'
operand immI_0() %{
  predicate(n->get_int() == 0);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: constant 'int -1'
operand immI_minus1() %{
  predicate(n->get_int() == -1);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: constant, but not 'int 0' nor 'int -1'.
operand immI_n0m1() %{
  predicate(n->get_int() != 0 && n->get_int() != -1);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//-------------------------------------------
// UNSIGNED INT immediate operands
//-------------------------------------------

// Unsigned Integer Immediate: 32-bit
operand uimmI() %{
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Integer Immediate: 16-bit
operand uimmI16() %{
  predicate(Immediate::is_uimm16(n->get_int()));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Integer Immediate: 12-bit
operand uimmI12() %{
  predicate(Immediate::is_uimm12(n->get_int()));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Integer Immediate: 12-bit
operand uimmI8() %{
  predicate(Immediate::is_uimm8(n->get_int()));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 6-bit
operand uimmI6() %{
  predicate(Immediate::is_uimm(n->get_int(), 6));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: 5-bit
operand uimmI5() %{
  predicate(Immediate::is_uimm(n->get_int(), 5));
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Length for SS instructions, given in DWs,
//   possible range [1..512], i.e. [8..4096] Bytes
//   used     range [1..256], i.e. [8..2048] Bytes
//   operand type int
// Unsigned Integer Immediate: 9-bit
operand SSlenDW() %{
  predicate(Immediate::is_uimm8(n->get_long()-1));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//------------------------------------------
// (UN)SIGNED INT specific values
//------------------------------------------

// Integer Immediate: the value 1
operand immI_1() %{
  predicate(n->get_int() == 1);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: the value 16.
operand immI_16() %{
  predicate(n->get_int() == 16);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: the value 24.
operand immI_24() %{
  predicate(n->get_int() == 24);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Integer Immediate: the value 255
operand immI_255() %{
  predicate(n->get_int() == 255);
  match(ConI);
  op_cost(1);
  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(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Integer Immediate: LL-part, extended by 1s.
operand uimmI_LL1() %{
  predicate((n->get_int() & 0xFFFF0000) == 0xFFFF0000);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Integer Immediate: LH-part, extended by 1s.
operand uimmI_LH1() %{
  predicate((n->get_int() & 0xFFFF) == 0xFFFF);
  match(ConI);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//------------------------------------------
// SIGNED LONG immediate operands
//------------------------------------------

operand immL() %{
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: 32-bit
operand immL32() %{
  predicate(Immediate::is_simm32(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: 20-bit
operand immL20() %{
  predicate(Immediate::is_simm20(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: 16-bit
operand immL16() %{
  predicate(Immediate::is_simm16(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: 8-bit
operand immL8() %{
  predicate(Immediate::is_simm8(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//--------------------------------------------
// UNSIGNED LONG immediate operands
//--------------------------------------------

operand uimmL32() %{
  predicate(Immediate::is_uimm32(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: 16-bit
operand uimmL16() %{
  predicate(Immediate::is_uimm16(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: 12-bit
operand uimmL12() %{
  predicate(Immediate::is_uimm12(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: 8-bit
operand uimmL8() %{
  predicate(Immediate::is_uimm8(n->get_long()));
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//-------------------------------------------
// (UN)SIGNED LONG specific values
//-------------------------------------------

// Long Immediate: the value FF
operand immL_FF() %{
  predicate(n->get_long() == 0xFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: the value FFFF
operand immL_FFFF() %{
  predicate(n->get_long() == 0xFFFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: the value FFFFFFFF
operand immL_FFFFFFFF() %{
  predicate(n->get_long() == 0xFFFFFFFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

operand immL_0() %{
  predicate(n->get_long() == 0L);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: LL-part, extended by 1s.
operand uimmL_LL1() %{
  predicate((n->get_long() & 0xFFFFFFFFFFFF0000L) == 0xFFFFFFFFFFFF0000L);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: LH-part, extended by 1s.
operand uimmL_LH1() %{
  predicate((n->get_long() & 0xFFFFFFFF0000FFFFL) == 0xFFFFFFFF0000FFFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: HL-part, extended by 1s.
operand uimmL_HL1() %{
  predicate((n->get_long() & 0xFFFF0000FFFFFFFFL) == 0xFFFF0000FFFFFFFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Unsigned Long Immediate: HH-part, extended by 1s.
operand uimmL_HH1() %{
  predicate((n->get_long() & 0xFFFFFFFFFFFFL) == 0xFFFFFFFFFFFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Long Immediate: low 32-bit mask
operand immL_32bits() %{
  predicate(n->get_long() == 0xFFFFFFFFL);
  match(ConL);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//--------------------------------------
//  POINTER immediate operands
//--------------------------------------

// Pointer Immediate: 64-bit
operand immP() %{
  match(ConP);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Pointer Immediate: 32-bit
operand immP32() %{
  predicate(Immediate::is_uimm32(n->get_ptr()));
  match(ConP);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Pointer Immediate: 16-bit
operand immP16() %{
  predicate(Immediate::is_uimm16(n->get_ptr()));
  match(ConP);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Pointer Immediate: 8-bit
operand immP8() %{
  predicate(Immediate::is_uimm8(n->get_ptr()));
  match(ConP);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//-----------------------------------
// POINTER specific values
//-----------------------------------

// Pointer Immediate: NULL
operand immP0() %{
  predicate(n->get_ptr() == 0);
  match(ConP);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

//---------------------------------------------
// NARROW POINTER immediate operands
//---------------------------------------------

// Narrow Pointer Immediate
operand immN() %{
  match(ConN);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

operand immNKlass() %{
  match(ConNKlass);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Narrow Pointer Immediate
operand immN8() %{
  predicate(Immediate::is_uimm8(n->get_narrowcon()));
  match(ConN);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Narrow NULL Pointer Immediate
operand immN0() %{
  predicate(n->get_narrowcon() == 0);
  match(ConN);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// FLOAT and DOUBLE immediate operands

// Double Immediate
operand immD() %{
  match(ConD);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Double Immediate: +-0
operand immDpm0() %{
  predicate(n->getd() == 0);
  match(ConD);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Double Immediate: +0
operand immDp0() %{
  predicate(jlong_cast(n->getd()) == 0);
  match(ConD);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate
operand immF() %{
  match(ConF);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate: +-0
operand immFpm0() %{
  predicate(n->getf() == 0);
  match(ConF);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// Float Immediate: +0
operand immFp0() %{
  predicate(jint_cast(n->getf()) == 0);
  match(ConF);
  op_cost(1);
  format %{ %}
  interface(CONST_INTER);
%}

// End of Immediate Operands

// Integer Register Operands
// Integer Register
operand iRegI() %{
  constraint(ALLOC_IN_RC(z_int_reg));
  match(RegI);
  match(noArg_iRegI);
  match(rarg1RegI);
  match(rarg2RegI);
  match(rarg3RegI);
  match(rarg4RegI);
  match(rarg5RegI);
  match(noOdd_iRegI);
  match(revenRegI);
  match(roddRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand noArg_iRegI() %{
  constraint(ALLOC_IN_RC(z_no_arg_int_reg));
  match(RegI);
  format %{ %}
  interface(REG_INTER);
%}

// revenRegI and roddRegI constitute and even-odd-pair.
operand revenRegI() %{
  constraint(ALLOC_IN_RC(z_rarg3_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

// revenRegI and roddRegI constitute and even-odd-pair.
operand roddRegI() %{
  constraint(ALLOC_IN_RC(z_rarg4_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg1RegI() %{
  constraint(ALLOC_IN_RC(z_rarg1_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg2RegI() %{
  constraint(ALLOC_IN_RC(z_rarg2_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg3RegI() %{
  constraint(ALLOC_IN_RC(z_rarg3_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg4RegI() %{
  constraint(ALLOC_IN_RC(z_rarg4_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg5RegI() %{
  constraint(ALLOC_IN_RC(z_rarg5_int_reg));
  match(iRegI);
  format %{ %}
  interface(REG_INTER);
%}

operand noOdd_iRegI() %{
  constraint(ALLOC_IN_RC(z_no_odd_int_reg));
  match(RegI);
  match(revenRegI);
  format %{ %}
  interface(REG_INTER);
%}

// Pointer Register
operand iRegP() %{
  constraint(ALLOC_IN_RC(z_ptr_reg));
  match(RegP);
  match(noArg_iRegP);
  match(rarg1RegP);
  match(rarg2RegP);
  match(rarg3RegP);
  match(rarg4RegP);
  match(rarg5RegP);
  match(revenRegP);
  match(roddRegP);
  format %{ %}
  interface(REG_INTER);
%}

// thread operand
operand threadRegP() %{
  constraint(ALLOC_IN_RC(z_thread_ptr_reg));
  match(RegP);
  format %{ "Z_THREAD" %}
  interface(REG_INTER);
%}

operand noArg_iRegP() %{
  constraint(ALLOC_IN_RC(z_no_arg_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg1RegP() %{
  constraint(ALLOC_IN_RC(z_rarg1_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg2RegP() %{
  constraint(ALLOC_IN_RC(z_rarg2_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg3RegP() %{
  constraint(ALLOC_IN_RC(z_rarg3_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg4RegP() %{
  constraint(ALLOC_IN_RC(z_rarg4_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg5RegP() %{
  constraint(ALLOC_IN_RC(z_rarg5_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand memoryRegP() %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(RegP);
  match(iRegP);
  match(threadRegP);
  format %{ %}
  interface(REG_INTER);
%}

// revenRegP and roddRegP constitute and even-odd-pair.
operand revenRegP() %{
  constraint(ALLOC_IN_RC(z_rarg3_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

// revenRegP and roddRegP constitute and even-odd-pair.
operand roddRegP() %{
  constraint(ALLOC_IN_RC(z_rarg4_ptr_reg));
  match(iRegP);
  format %{ %}
  interface(REG_INTER);
%}

operand lock_ptr_RegP() %{
  constraint(ALLOC_IN_RC(z_lock_ptr_reg));
  match(RegP);
  format %{ %}
  interface(REG_INTER);
%}

operand rscratch2RegP() %{
  constraint(ALLOC_IN_RC(z_rscratch2_bits64_reg));
  match(RegP);
  format %{ %}
  interface(REG_INTER);
%}

operand iRegN() %{
  constraint(ALLOC_IN_RC(z_int_reg));
  match(RegN);
  match(noArg_iRegN);
  match(rarg1RegN);
  match(rarg2RegN);
  match(rarg3RegN);
  match(rarg4RegN);
  match(rarg5RegN);
  format %{ %}
  interface(REG_INTER);
%}

operand noArg_iRegN() %{
  constraint(ALLOC_IN_RC(z_no_arg_int_reg));
  match(iRegN);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg1RegN() %{
  constraint(ALLOC_IN_RC(z_rarg1_int_reg));
  match(iRegN);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg2RegN() %{
  constraint(ALLOC_IN_RC(z_rarg2_int_reg));
  match(iRegN);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg3RegN() %{
  constraint(ALLOC_IN_RC(z_rarg3_int_reg));
  match(iRegN);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg4RegN() %{
  constraint(ALLOC_IN_RC(z_rarg4_int_reg));
  match(iRegN);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg5RegN() %{
  constraint(ALLOC_IN_RC(z_rarg5_ptrN_reg));
  match(iRegN);
  format %{ %}
  interface(REG_INTER);
%}

// Long Register
operand iRegL() %{
  constraint(ALLOC_IN_RC(z_long_reg));
  match(RegL);
  match(revenRegL);
  match(roddRegL);
  match(allRoddRegL);
  match(rarg1RegL);
  match(rarg5RegL);
  format %{ %}
  interface(REG_INTER);
%}

// revenRegL and roddRegL constitute and even-odd-pair.
operand revenRegL() %{
  constraint(ALLOC_IN_RC(z_rarg3_long_reg));
  match(iRegL);
  format %{ %}
  interface(REG_INTER);
%}

// revenRegL and roddRegL constitute and even-odd-pair.
operand roddRegL() %{
  constraint(ALLOC_IN_RC(z_rarg4_long_reg));
  match(iRegL);
  format %{ %}
  interface(REG_INTER);
%}

// available odd registers for iRegL
operand allRoddRegL() %{
  constraint(ALLOC_IN_RC(z_long_odd_reg));
  match(iRegL);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg1RegL() %{
  constraint(ALLOC_IN_RC(z_rarg1_long_reg));
  match(iRegL);
  format %{ %}
  interface(REG_INTER);
%}

operand rarg5RegL() %{
  constraint(ALLOC_IN_RC(z_rarg5_long_reg));
  match(iRegL);
  format %{ %}
  interface(REG_INTER);
%}

// Condition Code Flag Registers
operand flagsReg() %{
  constraint(ALLOC_IN_RC(z_condition_reg));
  match(RegFlags);
  format %{ "CR" %}
  interface(REG_INTER);
%}

// Condition Code Flag Registers for rules with result tuples
operand TD_flagsReg() %{
  constraint(ALLOC_IN_RC(z_condition_reg));
  match(RegFlags);
  format %{ "CR" %}
  interface(REG_TUPLE_DEST_INTER);
%}

operand regD() %{
  constraint(ALLOC_IN_RC(z_dbl_reg));
  match(RegD);
  format %{ %}
  interface(REG_INTER);
%}

operand rscratchRegD() %{
  constraint(ALLOC_IN_RC(z_rscratch1_dbl_reg));
  match(RegD);
  format %{ %}
  interface(REG_INTER);
%}

operand regF() %{
  constraint(ALLOC_IN_RC(z_flt_reg));
  match(RegF);
  format %{ %}
  interface(REG_INTER);
%}

operand rscratchRegF() %{
  constraint(ALLOC_IN_RC(z_rscratch1_flt_reg));
  match(RegF);
  format %{ %}
  interface(REG_INTER);
%}

// Special Registers

// Method Register
operand inline_cache_regP(iRegP reg) %{
  constraint(ALLOC_IN_RC(z_r9_regP)); // inline_cache_reg
  match(reg);
  format %{ %}
  interface(REG_INTER);
%}

// Operands to remove register moves in unscaled mode.
// Match read/write registers with an EncodeP node if neither shift nor add are required.
operand iRegP2N(iRegP reg) %{
  predicate(CompressedOops::shift() == 0 && _leaf->as_EncodeP()->in(0) == NULL);
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(EncodeP reg);
  format %{ "$reg" %}
  interface(REG_INTER)
%}

operand iRegN2P(iRegN reg) %{
  predicate(CompressedOops::base() == NULL && CompressedOops::shift() == 0 &&
            _leaf->as_DecodeN()->in(0) == NULL);
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(DecodeN reg);
  format %{ "$reg" %}
  interface(REG_INTER)
%}


//----------Complex Operands---------------------------------------------------

// Indirect Memory Reference
operand indirect(memoryRegP base) %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(base);
  op_cost(1);
  format %{ "#0[,$base]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp(0x0);
  %}
%}

// Indirect with Offset (long)
operand indOffset20(memoryRegP base, immL20 offset) %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP base offset);
  op_cost(1);
  format %{ "$offset[,$base]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($offset);
  %}
%}

operand indOffset20Narrow(iRegN base, immL20 offset) %{
  predicate(Matcher::narrow_oop_use_complex_address());
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP (DecodeN base) offset);
  op_cost(1);
  format %{ "$offset[,$base]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($offset);
  %}
%}

// Indirect with Offset (short)
operand indOffset12(memoryRegP base, uimmL12 offset) %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP base offset);
  op_cost(1);
  format %{ "$offset[[,$base]]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($offset);
  %}
%}

operand indOffset12Narrow(iRegN base, uimmL12 offset) %{
  predicate(Matcher::narrow_oop_use_complex_address());
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP (DecodeN base) offset);
  op_cost(1);
  format %{ "$offset[[,$base]]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($offset);
  %}
%}

// Indirect with Register Index
operand indIndex(memoryRegP base, iRegL index) %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP base index);
  op_cost(1);
  format %{ "#0[($index,$base)]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index($index);
    scale(0x0);
    disp(0x0);
  %}
%}

// Indirect with Offset (long) and index
operand indOffset20index(memoryRegP base, immL20 offset, iRegL index) %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP (AddP base index) offset);
  op_cost(1);
  format %{ "$offset[($index,$base)]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index($index);
    scale(0x0);
    disp($offset);
  %}
%}

operand indOffset20indexNarrow(iRegN base, immL20 offset, iRegL index) %{
  predicate(Matcher::narrow_oop_use_complex_address());
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP (AddP (DecodeN base) index) offset);
  op_cost(1);
  format %{ "$offset[($index,$base)]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index($index);
    scale(0x0);
    disp($offset);
  %}
%}

// Indirect with Offset (short) and index
operand indOffset12index(memoryRegP base, uimmL12 offset, iRegL index) %{
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP (AddP base index) offset);
  op_cost(1);
  format %{ "$offset[[($index,$base)]]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index($index);
    scale(0x0);
    disp($offset);
  %}
%}

operand indOffset12indexNarrow(iRegN base, uimmL12 offset, iRegL index) %{
  predicate(Matcher::narrow_oop_use_complex_address());
  constraint(ALLOC_IN_RC(z_memory_ptr_reg));
  match(AddP (AddP (DecodeN base) index) offset);
  op_cost(1);
  format %{ "$offset[[($index,$base)]]" %}
  interface(MEMORY_INTER) %{
    base($base);
    index($index);
    scale(0x0);
    disp($offset);
  %}
%}

//----------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(1);
  format %{ "[$reg(stackSlotI)]" %}
  interface(MEMORY_INTER) %{
    base(0xf);   // Z_SP
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($reg);  // stack offset
  %}
%}

operand stackSlotP(sRegP reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(1);
  format %{ "[$reg(stackSlotP)]" %}
  interface(MEMORY_INTER) %{
    base(0xf);   // Z_SP
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotF(sRegF reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(1);
  format %{ "[$reg(stackSlotF)]" %}
  interface(MEMORY_INTER) %{
    base(0xf);   // Z_SP
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotD(sRegD reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(1);
  //match(RegD);
  format %{ "[$reg(stackSlotD)]" %}
  interface(MEMORY_INTER) %{
    base(0xf);   // Z_SP
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($reg);  // Stack Offset
  %}
%}

operand stackSlotL(sRegL reg) %{
  constraint(ALLOC_IN_RC(stack_slots));
  op_cost(1);  //match(RegL);
  format %{ "[$reg(stackSlotL)]" %}
  interface(MEMORY_INTER) %{
    base(0xf);   // Z_SP
    index(0xffffFFFF); // noreg
    scale(0x0);
    disp($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.

// INT cmpOps for CompareAndBranch and CompareAndTrap instructions should not
// have mask bit #3 set.
operand cmpOpT() %{
  match(Bool);
  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x8);         // Assembler::bcondEqual
    not_equal(0x6);     // Assembler::bcondNotEqual
    less(0x4);          // Assembler::bcondLow
    greater_equal(0xa); // Assembler::bcondNotLow
    less_equal(0xc);    // Assembler::bcondNotHigh
    greater(0x2);       // Assembler::bcondHigh
    overflow(0x1);      // Assembler::bcondOverflow
    no_overflow(0xe);   // Assembler::bcondNotOverflow
  %}
%}

// When used for floating point comparisons: unordered is treated as less.
operand cmpOpF() %{
  match(Bool);
  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x8);
    not_equal(0x7);     // Includes 'unordered'.
    less(0x5);          // Includes 'unordered'.
    greater_equal(0xa);
    less_equal(0xd);    // Includes 'unordered'.
    greater(0x2);
    overflow(0x0);      // Not meaningful on z/Architecture.
    no_overflow(0x0);   // leave unchanged (zero) therefore
  %}
%}

// "Regular" cmpOp for int comparisons, includes bit #3 (overflow).
operand cmpOp() %{
  match(Bool);
  format %{ "" %}
  interface(COND_INTER) %{
    equal(0x8);
    not_equal(0x7);     // Includes 'unordered'.
    less(0x5);          // Includes 'unordered'.
    greater_equal(0xa);
    less_equal(0xd);    // Includes 'unordered'.
    greater(0x2);
    overflow(0x1);      // Assembler::bcondOverflow
    no_overflow(0xe);   // Assembler::bcondNotOverflow
  %}
%}

//----------OPERAND CLASSES----------------------------------------------------
// Operand Classes are groups of operands that are used to simplify
// instruction definitions by not requiring the AD writer to specify
// seperate 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.
// Indirect is not included since its use is limited to Compare & Swap

// Most general memory operand, allows base, index, and long displacement.
opclass memory(indirect, indIndex, indOffset20, indOffset20Narrow, indOffset20index, indOffset20indexNarrow);
opclass memoryRXY(indirect, indIndex, indOffset20, indOffset20Narrow, indOffset20index, indOffset20indexNarrow);

// General memory operand, allows base, index, and short displacement.
opclass memoryRX(indirect, indIndex, indOffset12, indOffset12Narrow, indOffset12index, indOffset12indexNarrow);

// Memory operand, allows only base and long displacement.
opclass memoryRSY(indirect, indOffset20, indOffset20Narrow);

// Memory operand, allows only base and short displacement.
opclass memoryRS(indirect, indOffset12, indOffset12Narrow);

// Operand classes to match encode and decode.
opclass iRegN_P2N(iRegN);
opclass iRegP_N2P(iRegP);


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

//----------ATTRIBUTES---------------------------------------------------------
attributes %{
  // z/Architecture instructions are of length 2, 4, or 6 bytes.
  variable_size_instructions;
  instruction_unit_size = 2;

  // Meaningless on z/Architecture.
  max_instructions_per_bundle = 1;

  // The z/Architecture processor fetches 64 bytes...
  instruction_fetch_unit_size = 64;

  // ...in one line.
  instruction_fetch_units = 1
%}

//----------RESOURCES----------------------------------------------------------
// Resources are the functional units available to the machine.
resources(
   Z_BR,     // branch unit
   Z_CR,     // condition unit
   Z_FX1,    // integer arithmetic unit 1
   Z_FX2,    // integer arithmetic unit 2
   Z_LDST1,  // load/store unit 1
   Z_LDST2,  // load/store unit 2
   Z_FP1,    // float arithmetic unit 1
   Z_FP2,    // float arithmetic unit 2
   Z_LDST = Z_LDST1 | Z_LDST2,
   Z_FX   = Z_FX1 | Z_FX2,
   Z_FP   = Z_FP1 | Z_FP2
  );

//----------PIPELINE DESCRIPTION-----------------------------------------------
// Pipeline Description specifies the stages in the machine's pipeline.
pipe_desc(
   // TODO: adapt
   Z_IF,  // instruction fetch
   Z_IC,
   Z_D0,  // decode
   Z_D1,  // decode
   Z_D2,  // decode
   Z_D3,  // decode
   Z_Xfer1,
   Z_GD,  // group definition
   Z_MP,  // map
   Z_ISS, // issue
   Z_RF,  // resource fetch
   Z_EX1, // execute (all units)
   Z_EX2, // execute (FP, LDST)
   Z_EX3, // execute (FP, LDST)
   Z_EX4, // execute (FP)
   Z_EX5, // execute (FP)
   Z_EX6, // execute (FP)
   Z_WB,  // write back
   Z_Xfer2,
   Z_CP
  );

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

// Providing the `ins_pipe' declarations in the instruction
// specifications seems to be of little use. So we use
// `pipe_class_dummy' for all our instructions at present.
pipe_class pipe_class_dummy() %{
  single_instruction;
  fixed_latency(4);
%}

// SIGTRAP based implicit range checks in compiled code.
// Currently, no pipe classes are used on z/Architecture.
pipe_class pipe_class_trap() %{
  single_instruction;
%}

pipe_class pipe_class_fx_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
  single_instruction;
  dst  : Z_EX1(write);
  src1 : Z_RF(read);
  src2 : Z_RF(read);
  Z_FX : Z_RF;
%}

pipe_class pipe_class_ldst(iRegP dst, memory mem) %{
  single_instruction;
  mem : Z_RF(read);
  dst : Z_WB(write);
  Z_LDST : Z_RF;
%}

define %{
  MachNop = pipe_class_dummy;
%}

%}

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

//---------- 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);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "L       $dst,$src\t # stk reload int" %}
  opcode(L_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Store integer to stack slot.
instruct regI_to_stkI(stackSlotI dst, iRegI src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "ST      $src,$dst\t # stk spill int" %}
  opcode(ST_ZOPC);
  ins_encode(z_form_rt_mem(src, dst)); // rs=rt
  ins_pipe(pipe_class_dummy);
%}

// Load long from stack slot.
instruct stkL_to_regL(iRegL dst, stackSlotL src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LG      $dst,$src\t # stk reload long" %}
  opcode(LG_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Store long to stack slot.
instruct regL_to_stkL(stackSlotL dst, iRegL src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "STG     $src,$dst\t # stk spill long" %}
  opcode(STG_ZOPC);
  ins_encode(z_form_rt_mem(src, dst)); // rs=rt
  ins_pipe(pipe_class_dummy);
%}

// 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);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LG      $dst,$src\t # stk reload ptr" %}
  opcode(LG_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Store pointer to stack slot.
instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "STG     $src,$dst\t # stk spill ptr" %}
  opcode(STG_ZOPC);
  ins_encode(z_form_rt_mem(src, dst)); // rs=rt
  ins_pipe(pipe_class_dummy);
%}

//  Float types

// Load float value from stack slot.
instruct stkF_to_regF(regF dst, stackSlotF src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LE(Y)   $dst,$src\t # stk reload float" %}
  opcode(LE_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Store float value to stack slot.
instruct regF_to_stkF(stackSlotF dst, regF src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STE(Y)  $src,$dst\t # stk spill float" %}
  opcode(STE_ZOPC);
  ins_encode(z_form_rt_mem(src, dst));
  ins_pipe(pipe_class_dummy);
%}

// Load double value from stack slot.
instruct stkD_to_regD(regD dst, stackSlotD src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LD(Y)   $dst,$src\t # stk reload double" %}
  opcode(LD_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Store double value to stack slot.
instruct regD_to_stkD(stackSlotD dst, regD src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "STD(Y)  $src,$dst\t # stk spill double" %}
  opcode(STD_ZOPC);
  ins_encode(z_form_rt_mem(src, dst));
  ins_pipe(pipe_class_dummy);
%}

//----------Load/Store/Move Instructions---------------------------------------

//----------Load Instructions--------------------------------------------------

//------------------
//  MEMORY
//------------------

//  BYTE
// Load Byte (8bit signed)
instruct loadB(iRegI dst, memory mem) %{
  match(Set dst (LoadB mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LB      $dst, $mem\t # sign-extend byte to int" %}
  opcode(LB_ZOPC, LB_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Byte (8bit signed)
instruct loadB2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadB mem)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LGB     $dst, $mem\t # sign-extend byte to long" %}
  opcode(LGB_ZOPC, LGB_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// 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(Z_DISP3_SIZE);
  format %{ "LLGC    $dst,$mem\t # zero-extend byte to int" %}
  opcode(LLGC_ZOPC, LLGC_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// 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(Z_DISP3_SIZE);
  format %{ "LLGC    $dst,$mem\t # zero-extend byte to long" %}
  opcode(LLGC_ZOPC, LLGC_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// CHAR/SHORT

// Load Short (16bit signed)
instruct loadS(iRegI dst, memory mem) %{
  match(Set dst (LoadS mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "LH(Y)   $dst,$mem\t # sign-extend short to int" %}
  opcode(LHY_ZOPC, LH_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Short (16bit signed)
instruct loadS2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadS mem)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LGH     $dst,$mem\t # sign-extend short to long" %}
  opcode(LGH_ZOPC, LGH_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Char (16bit Unsigned)
instruct loadUS(iRegI dst, memory mem) %{
  match(Set dst (LoadUS mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LLGH    $dst,$mem\t # zero-extend short to int" %}
  opcode(LLGH_ZOPC, LLGH_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// 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(Z_DISP3_SIZE);
  format %{ "LLGH    $dst,$mem\t # zero-extend short to long" %}
  opcode(LLGH_ZOPC, LLGH_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// INT

// Load Integer
instruct loadI(iRegI dst, memory mem) %{
  match(Set dst (LoadI mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "L(Y)    $dst,$mem\t #" %}
  opcode(LY_ZOPC, L_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load and convert to long.
instruct loadI2L(iRegL dst, memory mem) %{
  match(Set dst (ConvI2L (LoadI mem)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LGF     $dst,$mem\t #" %}
  opcode(LGF_ZOPC, LGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Unsigned Integer into a Long Register
instruct loadUI2L(iRegL dst, memory mem, immL_FFFFFFFF mask) %{
  match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LLGF    $dst,$mem\t # zero-extend int to long" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// range = array length (=jint)
// Load Range
instruct loadRange(iRegI dst, memory mem) %{
  match(Set dst (LoadRange mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "L(Y)    $dst,$mem\t # range" %}
  opcode(LY_ZOPC, L_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// LONG

// Load Long - aligned
instruct loadL(iRegL dst, memory mem) %{
  match(Set dst (LoadL mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LG      $dst,$mem\t # long" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Long - UNaligned
instruct loadL_unaligned(iRegL dst, memory mem) %{
  match(Set dst (LoadL_unaligned mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LG      $dst,$mem\t # unaligned long" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}


// PTR

// Load Pointer
instruct loadP(iRegP dst, memory mem) %{
  match(Set dst (LoadP mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LG      $dst,$mem\t # ptr" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// LoadP + CastP2L
instruct castP2X_loadP(iRegL dst, memory mem) %{
  match(Set dst (CastP2X (LoadP mem)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LG      $dst,$mem\t # ptr + p2x" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Klass Pointer
instruct loadKlass(iRegP dst, memory mem) %{
  match(Set dst (LoadKlass mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LG      $dst,$mem\t # klass ptr" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

instruct loadTOC(iRegL dst) %{
  effect(DEF dst);
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  // TODO: check why this attribute causes many unnecessary rematerializations.
  //
  // The graphs I saw just had high register pressure. Further the
  // register TOC is loaded to is overwritten by the constant short
  // after. Here something as round robin register allocation might
  // help. But rematerializing seems not to hurt, jack even seems to
  // improve slightly.
  //
  // Without this flag we get spill-split recycle sanity check
  // failures in
  // spec.benchmarks._228_jack.NfaState::GenerateCode. This happens in
  // a block with three loadConP_dynTOC nodes and a tlsLoadP. The
  // tlsLoadP has a huge amount of outs and forces the TOC down to the
  // stack. Later tlsLoadP is rematerialized, leaving the register
  // allocator with TOC on the stack and a badly placed reload.
  ins_should_rematerialize(true);
  format %{ "LARL    $dst, &constant_pool\t; load dynTOC" %}
  ins_encode %{ __ load_toc($dst$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// FLOAT

// Load Float
instruct loadF(regF dst, memory mem) %{
  match(Set dst (LoadF mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "LE(Y)    $dst,$mem" %}
  opcode(LEY_ZOPC, LE_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// DOUBLE

// Load Double
instruct loadD(regD dst, memory mem) %{
  match(Set dst (LoadD mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "LD(Y)    $dst,$mem" %}
  opcode(LDY_ZOPC, LD_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load Double - UNaligned
instruct loadD_unaligned(regD dst, memory mem) %{
  match(Set dst (LoadD_unaligned mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "LD(Y)    $dst,$mem" %}
  opcode(LDY_ZOPC, LD_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}


//----------------------
//  IMMEDIATES
//----------------------

instruct loadConI(iRegI dst, immI src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LGFI    $dst,$src\t # (int)" %}
  ins_encode %{ __ z_lgfi($dst$$Register, $src$$constant); %}  // Sign-extend to 64 bit, it's at no cost.
  ins_pipe(pipe_class_dummy);
%}

instruct loadConI16(iRegI dst, immI16 src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LGHI    $dst,$src\t # (int)" %}
  ins_encode %{ __ z_lghi($dst$$Register, $src$$constant); %}  // Sign-extend to 64 bit, it's at no cost.
  ins_pipe(pipe_class_dummy);
%}

instruct loadConI_0(iRegI dst, immI_0 src, flagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "loadConI $dst,$src\t # (int) XGR because ZERO is loaded" %}
  opcode(XGR_ZOPC);
  ins_encode(z_rreform(dst, dst));
  ins_pipe(pipe_class_dummy);
%}

instruct loadConUI16(iRegI dst, uimmI16 src) %{
  match(Set dst src);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LLILL    $dst,$src" %}
  opcode(LLILL_ZOPC);
  ins_encode(z_riform_unsigned(dst, src) );
  ins_pipe(pipe_class_dummy);
%}

// Load long constant from TOC with pcrelative address.
instruct loadConL_pcrelTOC(iRegL dst, immL src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST_LO);
  size(6);
  format %{ "LGRL    $dst,[pcrelTOC]\t # load long $src from table" %}
  ins_encode %{
    address long_address = __ long_constant($src$$constant);
    if (long_address == NULL) {
      Compile::current()->env()->record_out_of_memory_failure();
      return;
    }
    __ load_long_pcrelative($dst$$Register, long_address);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct loadConL32(iRegL dst, immL32 src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LGFI     $dst,$src\t # (long)" %}
  ins_encode %{ __ z_lgfi($dst$$Register, $src$$constant); %}  // Sign-extend to 64 bit, it's at no cost.
  ins_pipe(pipe_class_dummy);
%}

instruct loadConL16(iRegL dst, immL16 src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LGHI     $dst,$src\t # (long)" %}
  ins_encode %{ __ z_lghi($dst$$Register, $src$$constant); %}  // Sign-extend to 64 bit, it's at no cost.
  ins_pipe(pipe_class_dummy);
%}

instruct loadConL_0(iRegL dst, immL_0 src, flagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  format %{ "LoadConL    $dst,$src\t # (long) XGR because ZERO is loaded" %}
  opcode(XGR_ZOPC);
  ins_encode(z_rreform(dst, dst));
  ins_pipe(pipe_class_dummy);
%}

// Load ptr constant from TOC with pc relative address.
// Special handling for oop constants required.
instruct loadConP_pcrelTOC(iRegP dst, immP src) %{
  match(Set dst src);
  ins_cost(MEMORY_REF_COST_LO);
  size(6);
  format %{ "LGRL    $dst,[pcrelTOC]\t # load ptr $src from table" %}
  ins_encode %{
    relocInfo::relocType constant_reloc = $src->constant_reloc();
    if (constant_reloc == relocInfo::oop_type) {
      AddressLiteral a = __ allocate_oop_address((jobject)$src$$constant);
      bool success = __ load_oop_from_toc($dst$$Register, a);
      if (!success) {
        Compile::current()->env()->record_out_of_memory_failure();
        return;
      }
    } else if (constant_reloc == relocInfo::metadata_type) {
      AddressLiteral a = __ constant_metadata_address((Metadata *)$src$$constant);
      address const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
      if (const_toc_addr == NULL) {
        Compile::current()->env()->record_out_of_memory_failure();
        return;
      }
      __ load_long_pcrelative($dst$$Register, const_toc_addr);
    } else {          // Non-oop pointers, e.g. card mark base, heap top.
      address long_address = __ long_constant((jlong)$src$$constant);
      if (long_address == NULL) {
        Compile::current()->env()->record_out_of_memory_failure();
        return;
      }
      __ load_long_pcrelative($dst$$Register, long_address);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// We don't use immP16 to avoid problems with oops.
instruct loadConP0(iRegP dst, immP0 src, flagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  size(4);
  format %{ "XGR     $dst,$dst\t # NULL ptr" %}
  opcode(XGR_ZOPC);
  ins_encode(z_rreform(dst, dst));
  ins_pipe(pipe_class_dummy);
%}

//----------Load Float Constant Instructions-------------------------------------------------

// We may not specify this instruction via an `expand' rule. If we do,
// code selection will forget that this instruction needs a floating
// point constant inserted into the code buffer. So `Shorten_branches'
// will fail.
instruct loadConF_dynTOC(regF dst, immF src, flagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(6);
  // If this instruction rematerializes, it prolongs the live range
  // of the toc node, causing illegal graphs.
  ins_cannot_rematerialize(true);
  format %{ "LE(Y)    $dst,$constantoffset[,$constanttablebase]\t # load FLOAT $src from table" %}
  ins_encode %{
    __ load_float_largeoffset($dst$$FloatRegister, $constantoffset($src), $constanttablebase, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// E may not specify this instruction via an `expand' rule. If we do,
// code selection will forget that this instruction needs a floating
// point constant inserted into the code buffer. So `Shorten_branches'
// will fail.
instruct loadConD_dynTOC(regD dst, immD src, flagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(6);
  // If this instruction rematerializes, it prolongs the live range
  // of the toc node, causing illegal graphs.
  ins_cannot_rematerialize(true);
  format %{ "LD(Y)    $dst,$constantoffset[,$constanttablebase]\t # load DOUBLE $src from table" %}
  ins_encode %{
    __ load_double_largeoffset($dst$$FloatRegister, $constantoffset($src), $constanttablebase, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Special case: Load Const 0.0F

// There's a special instr to clear a FP register.
instruct loadConF0(regF dst, immFp0 src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LZER     $dst,$src\t # clear to zero" %}
  opcode(LZER_ZOPC);
  ins_encode(z_rreform(dst, Z_F0));
  ins_pipe(pipe_class_dummy);
%}

// There's a special instr to clear a FP register.
instruct loadConD0(regD dst, immDp0 src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LZDR     $dst,$src\t # clear to zero" %}
  opcode(LZDR_ZOPC);
  ins_encode(z_rreform(dst, Z_F0));
  ins_pipe(pipe_class_dummy);
%}


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

// BYTE

// Store Byte
instruct storeB(memory mem, iRegI src) %{
  match(Set mem (StoreB mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "STC(Y)  $src,$mem\t # byte" %}
  opcode(STCY_ZOPC, STC_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

instruct storeCM(memory mem, immI_0 src) %{
  match(Set mem (StoreCM mem src));
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "STC(Y)  $src,$mem\t # CMS card-mark byte (must be 0!)" %}
  ins_encode %{
    guarantee($mem$$index$$Register != Z_R0, "content will not be used.");
    if ($mem$$index$$Register != noreg) {
      // Can't use clear_mem --> load const zero and store character.
      __ load_const_optimized(Z_R0_scratch, (long)0);
      if (Immediate::is_uimm12($mem$$disp)) {
        __ z_stc(Z_R0_scratch, $mem$$Address);
      } else {
        __ z_stcy(Z_R0_scratch, $mem$$Address);
      }
    } else {
      __ clear_mem(Address($mem$$Address), 1);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// CHAR/SHORT

// Store Char/Short
instruct storeC(memory mem, iRegI src) %{
  match(Set mem (StoreC mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "STH(Y)  $src,$mem\t # short" %}
  opcode(STHY_ZOPC, STH_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

// INT

// Store Integer
instruct storeI(memory mem, iRegI src) %{
  match(Set mem (StoreI mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "ST(Y)   $src,$mem\t # int" %}
  opcode(STY_ZOPC, ST_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

// LONG

// Store Long
instruct storeL(memory mem, iRegL src) %{
  match(Set mem (StoreL mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "STG     $src,$mem\t # long" %}
  opcode(STG_ZOPC, STG_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

// PTR

// Store Pointer
instruct storeP(memory dst, memoryRegP src) %{
  match(Set dst (StoreP dst src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "STG     $src,$dst\t # ptr" %}
  opcode(STG_ZOPC, STG_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, dst));
  ins_pipe(pipe_class_dummy);
%}

// FLOAT

// Store Float
instruct storeF(memory mem, regF src) %{
  match(Set mem (StoreF mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "STE(Y)   $src,$mem\t # float" %}
  opcode(STEY_ZOPC, STE_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

// DOUBLE

// Store Double
instruct storeD(memory mem, regD src) %{
  match(Set mem (StoreD mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "STD(Y)   $src,$mem\t # double" %}
  opcode(STDY_ZOPC, STD_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

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

// Should support match rule for PrefetchAllocation.
// Still needed after 8068977 for PrefetchAllocate.
instruct prefetchAlloc(memory mem) %{
  match(PrefetchAllocation mem);
  predicate(VM_Version::has_Prefetch());
  ins_cost(DEFAULT_COST);
  format %{ "PREFETCH 2, $mem\t # Prefetch allocation, z10 only" %}
  ins_encode %{ __ z_pfd(0x02, $mem$$Address); %}
  ins_pipe(pipe_class_dummy);
%}

//----------Memory init instructions------------------------------------------

// Move Immediate to 1-byte memory.
instruct memInitB(memoryRSY mem, immI8 src) %{
  match(Set mem (StoreB mem src));
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MVI     $mem,$src\t # direct mem init 1" %}
  ins_encode %{
    if (Immediate::is_uimm12((long)$mem$$disp)) {
      __ z_mvi($mem$$Address, $src$$constant);
    } else {
      __ z_mviy($mem$$Address, $src$$constant);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Move Immediate to 2-byte memory.
instruct memInitC(memoryRS mem, immI16 src) %{
  match(Set mem (StoreC mem src));
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "MVHHI   $mem,$src\t # direct mem init 2" %}
  opcode(MVHHI_ZOPC);
  ins_encode(z_silform(mem, src));
  ins_pipe(pipe_class_dummy);
%}

// Move Immediate to 4-byte memory.
instruct memInitI(memoryRS mem, immI16 src) %{
  match(Set mem (StoreI mem src));
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "MVHI    $mem,$src\t # direct mem init 4" %}
  opcode(MVHI_ZOPC);
  ins_encode(z_silform(mem, src));
  ins_pipe(pipe_class_dummy);
%}


// Move Immediate to 8-byte memory.
instruct memInitL(memoryRS mem, immL16 src) %{
  match(Set mem (StoreL mem src));
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "MVGHI   $mem,$src\t # direct mem init 8" %}
  opcode(MVGHI_ZOPC);
  ins_encode(z_silform(mem, src));
  ins_pipe(pipe_class_dummy);
%}

// Move Immediate to 8-byte memory.
instruct memInitP(memoryRS mem, immP16 src) %{
  match(Set mem (StoreP mem src));
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "MVGHI   $mem,$src\t # direct mem init 8" %}
  opcode(MVGHI_ZOPC);
  ins_encode(z_silform(mem, src));
  ins_pipe(pipe_class_dummy);
%}


//----------Instructions for compressed pointers (cOop and NKlass)-------------

// See cOop encoding classes for elaborate comment.

// Moved here because it is needed in expand rules for encode.
// Long negation.
instruct negL_reg_reg(iRegL dst, immL_0 zero, iRegL src, flagsReg cr) %{
  match(Set dst (SubL zero src));
  effect(KILL cr);
  size(4);
  format %{ "NEG     $dst, $src\t # long" %}
  ins_encode %{ __ z_lcgr($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Load Compressed Pointer

// Load narrow oop
instruct loadN(iRegN dst, memory mem) %{
  match(Set dst (LoadN mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LoadN   $dst,$mem\t # (cOop)" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load narrow Klass Pointer
instruct loadNKlass(iRegN dst, memory mem) %{
  match(Set dst (LoadNKlass mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LoadNKlass $dst,$mem\t # (klass cOop)" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Load constant Compressed Pointer

instruct loadConN(iRegN dst, immN src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "loadConN    $dst,$src\t # (cOop)" %}
  ins_encode %{
    AddressLiteral cOop = __ constant_oop_address((jobject)$src$$constant);
    __ relocate(cOop.rspec(), 1);
    __ load_narrow_oop($dst$$Register, (narrowOop)cOop.value());
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct loadConN0(iRegN dst, immN0 src, flagsReg cr) %{
  match(Set dst src);
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "loadConN    $dst,$src\t # (cOop) XGR because ZERO is loaded" %}
  opcode(XGR_ZOPC);
  ins_encode(z_rreform(dst, dst));
  ins_pipe(pipe_class_dummy);
%}

instruct loadConNKlass(iRegN dst, immNKlass src) %{
  match(Set dst src);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "loadConNKlass $dst,$src\t # (cKlass)" %}
  ins_encode %{
    AddressLiteral NKlass = __ constant_metadata_address((Metadata*)$src$$constant);
    __ relocate(NKlass.rspec(), 1);
    __ load_narrow_klass($dst$$Register, (Klass*)NKlass.value());
  %}
  ins_pipe(pipe_class_dummy);
%}

// Load and Decode Compressed Pointer
// optimized variants for Unscaled cOops

instruct decodeLoadN(iRegP dst, memory mem) %{
  match(Set dst (DecodeN (LoadN mem)));
  predicate(false && (CompressedOops::base()==NULL)&&(CompressedOops::shift()==0));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "DecodeLoadN  $dst,$mem\t # (cOop Load+Decode)" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

instruct decodeLoadNKlass(iRegP dst, memory mem) %{
  match(Set dst (DecodeNKlass (LoadNKlass mem)));
  predicate(false && (CompressedKlassPointers::base()==NULL)&&(CompressedKlassPointers::shift()==0));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "DecodeLoadNKlass  $dst,$mem\t # (load/decode NKlass)" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

instruct decodeLoadConNKlass(iRegP dst, immNKlass src) %{
  match(Set dst (DecodeNKlass src));
  ins_cost(3 * DEFAULT_COST);
  size(12);
  format %{ "DecodeLoadConNKlass  $dst,$src\t # decode(cKlass)" %}
  ins_encode %{
    AddressLiteral NKlass = __ constant_metadata_address((Metadata*)$src$$constant);
    __ relocate(NKlass.rspec(), 1);
    __ load_const($dst$$Register, (Klass*)NKlass.value());
  %}
  ins_pipe(pipe_class_dummy);
%}

// Decode Compressed Pointer

// General decoder
instruct decodeN(iRegP dst, iRegN src, flagsReg cr) %{
  match(Set dst (DecodeN src));
  effect(KILL cr);
  predicate(CompressedOops::base() == NULL || !ExpandLoadingBaseDecode);
  ins_cost(MEMORY_REF_COST+3 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "decodeN  $dst,$src\t # (decode cOop)" %}
  ins_encode %{  __ oop_decoder($dst$$Register, $src$$Register, true); %}
  ins_pipe(pipe_class_dummy);
%}

// General Klass decoder
instruct decodeKlass(iRegP dst, iRegN src, flagsReg cr) %{
  match(Set dst (DecodeNKlass src));
  effect(KILL cr);
  ins_cost(3 * DEFAULT_COST);
  format %{ "decode_klass $dst,$src" %}
  ins_encode %{ __ decode_klass_not_null($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// General decoder
instruct decodeN_NN(iRegP dst, iRegN src, flagsReg cr) %{
  match(Set dst (DecodeN src));
  effect(KILL cr);
  predicate((n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull ||
             n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant) &&
            (CompressedOops::base()== NULL || !ExpandLoadingBaseDecode_NN));
  ins_cost(MEMORY_REF_COST+2 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "decodeN  $dst,$src\t # (decode cOop NN)" %}
  ins_encode %{ __ oop_decoder($dst$$Register, $src$$Register, false); %}
  ins_pipe(pipe_class_dummy);
%}

  instruct loadBase(iRegL dst, immL baseImm) %{
    effect(DEF dst, USE baseImm);
    predicate(false);
    format %{ "llihl    $dst=$baseImm \t// load heap base" %}
    ins_encode %{ __ get_oop_base($dst$$Register, $baseImm$$constant); %}
    ins_pipe(pipe_class_dummy);
  %}

  // Decoder for heapbased mode peeling off loading the base.
  instruct decodeN_base(iRegP dst, iRegN src, iRegL base, flagsReg cr) %{
    match(Set dst (DecodeN src base));
    // Note: Effect TEMP dst was used with the intention to get
    // different regs for dst and base, but this has caused ADLC to
    // generate wrong code. Oop_decoder generates additional lgr when
    // dst==base.
    effect(KILL cr);
    predicate(false);
    // TODO: s390 port size(VARIABLE_SIZE);
    format %{ "decodeN  $dst = ($src == 0) ? NULL : ($src << 3) + $base + pow2_offset\t # (decode cOop)" %}
    ins_encode %{
      __ oop_decoder($dst$$Register, $src$$Register, true, $base$$Register,
                     (jlong)MacroAssembler::get_oop_base_pow2_offset((uint64_t)(intptr_t)CompressedOops::base()));
    %}
    ins_pipe(pipe_class_dummy);
  %}

  // Decoder for heapbased mode peeling off loading the base.
  instruct decodeN_NN_base(iRegP dst, iRegN src, iRegL base, flagsReg cr) %{
    match(Set dst (DecodeN src base));
    effect(KILL cr);
    predicate(false);
    // TODO: s390 port size(VARIABLE_SIZE);
    format %{ "decodeN  $dst = ($src << 3) + $base + pow2_offset\t # (decode cOop)" %}
    ins_encode %{
      __ oop_decoder($dst$$Register, $src$$Register, false, $base$$Register,
                     (jlong)MacroAssembler::get_oop_base_pow2_offset((uint64_t)(intptr_t)CompressedOops::base()));
    %}
    ins_pipe(pipe_class_dummy);
  %}

// Decoder for heapbased mode peeling off loading the base.
instruct decodeN_Ex(iRegP dst, iRegN src, flagsReg cr) %{
  match(Set dst (DecodeN src));
  predicate(CompressedOops::base() != NULL && ExpandLoadingBaseDecode);
  ins_cost(MEMORY_REF_COST+3 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  expand %{
    immL baseImm %{ (jlong)(intptr_t)CompressedOops::base() %}
    iRegL base;
    loadBase(base, baseImm);
    decodeN_base(dst, src, base, cr);
  %}
%}

// Decoder for heapbased mode peeling off loading the base.
instruct decodeN_NN_Ex(iRegP dst, iRegN src, flagsReg cr) %{
  match(Set dst (DecodeN src));
  predicate((n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull ||
             n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant) &&
            CompressedOops::base() != NULL && ExpandLoadingBaseDecode_NN);
  ins_cost(MEMORY_REF_COST+2 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  expand %{
    immL baseImm %{ (jlong)(intptr_t)CompressedOops::base() %}
    iRegL base;
    loadBase(base, baseImm);
    decodeN_NN_base(dst, src, base, cr);
  %}
%}

//  Encode Compressed Pointer

// General encoder
instruct encodeP(iRegN dst, iRegP src, flagsReg cr) %{
  match(Set dst (EncodeP src));
  effect(KILL cr);
  predicate((n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull) &&
            (CompressedOops::base() == 0 ||
             CompressedOops::base_disjoint() ||
             !ExpandLoadingBaseEncode));
  ins_cost(MEMORY_REF_COST+3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "encodeP  $dst,$src\t # (encode cOop)" %}
  ins_encode %{ __ oop_encoder($dst$$Register, $src$$Register, true, Z_R1_scratch, -1, all_outs_are_Stores(this)); %}
  ins_pipe(pipe_class_dummy);
%}

// General class encoder
instruct encodeKlass(iRegN dst, iRegP src, flagsReg cr) %{
  match(Set dst (EncodePKlass src));
  effect(KILL cr);
  format %{ "encode_klass $dst,$src" %}
  ins_encode %{ __ encode_klass_not_null($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct encodeP_NN(iRegN dst, iRegP src, flagsReg cr) %{
  match(Set dst (EncodeP src));
  effect(KILL cr);
  predicate((n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull) &&
            (CompressedOops::base() == 0 ||
             CompressedOops::base_disjoint() ||
             !ExpandLoadingBaseEncode_NN));
  ins_cost(MEMORY_REF_COST+3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "encodeP  $dst,$src\t # (encode cOop)" %}
  ins_encode %{ __ oop_encoder($dst$$Register, $src$$Register, false, Z_R1_scratch, -1, all_outs_are_Stores(this)); %}
  ins_pipe(pipe_class_dummy);
%}

  // Encoder for heapbased mode peeling off loading the base.
  instruct encodeP_base(iRegN dst, iRegP src, iRegL base) %{
    match(Set dst (EncodeP src (Binary base dst)));
    effect(TEMP_DEF dst);
    predicate(false);
    ins_cost(MEMORY_REF_COST+2 * DEFAULT_COST);
    // TODO: s390 port size(VARIABLE_SIZE);
    format %{ "encodeP  $dst = ($src>>3) +$base + pow2_offset\t # (encode cOop)" %}
    ins_encode %{
      jlong offset = -(jlong)MacroAssembler::get_oop_base_pow2_offset
        (((uint64_t)(intptr_t)CompressedOops::base()) >> CompressedOops::shift());
      __ oop_encoder($dst$$Register, $src$$Register, true, $base$$Register, offset);
    %}
    ins_pipe(pipe_class_dummy);
  %}

  // Encoder for heapbased mode peeling off loading the base.
  instruct encodeP_NN_base(iRegN dst, iRegP src, iRegL base, immL pow2_offset) %{
    match(Set dst (EncodeP src base));
    effect(USE pow2_offset);
    predicate(false);
    ins_cost(MEMORY_REF_COST+2 * DEFAULT_COST);
    // TODO: s390 port size(VARIABLE_SIZE);
    format %{ "encodeP  $dst = ($src>>3) +$base + $pow2_offset\t # (encode cOop)" %}
    ins_encode %{ __ oop_encoder($dst$$Register, $src$$Register, false, $base$$Register, $pow2_offset$$constant); %}
    ins_pipe(pipe_class_dummy);
  %}

// Encoder for heapbased mode peeling off loading the base.
instruct encodeP_Ex(iRegN dst, iRegP src, flagsReg cr) %{
  match(Set dst (EncodeP src));
  effect(KILL cr);
  predicate((n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull) &&
            (CompressedOops::base_overlaps() && ExpandLoadingBaseEncode));
  ins_cost(MEMORY_REF_COST+3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  expand %{
    immL baseImm %{ ((jlong)(intptr_t)CompressedOops::base()) >> CompressedOops::shift() %}
    immL_0 zero %{ (0) %}
    flagsReg ccr;
    iRegL base;
    iRegL negBase;
    loadBase(base, baseImm);
    negL_reg_reg(negBase, zero, base, ccr);
    encodeP_base(dst, src, negBase);
  %}
%}

// Encoder for heapbased mode peeling off loading the base.
instruct encodeP_NN_Ex(iRegN dst, iRegP src, flagsReg cr) %{
  match(Set dst (EncodeP src));
  effect(KILL cr);
  predicate((n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull) &&
            (CompressedOops::base_overlaps() && ExpandLoadingBaseEncode_NN));
  ins_cost(MEMORY_REF_COST+3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  expand %{
    immL baseImm %{ (jlong)(intptr_t)CompressedOops::base() %}
    immL pow2_offset %{ -(jlong)MacroAssembler::get_oop_base_pow2_offset(((uint64_t)(intptr_t)CompressedOops::base())) %}
    immL_0 zero %{ 0 %}
    flagsReg ccr;
    iRegL base;
    iRegL negBase;
    loadBase(base, baseImm);
    negL_reg_reg(negBase, zero, base, ccr);
    encodeP_NN_base(dst, src, negBase, pow2_offset);
  %}
%}

//  Store Compressed Pointer

// Store Compressed Pointer
instruct storeN(memory mem, iRegN_P2N src) %{
  match(Set mem (StoreN mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "ST      $src,$mem\t # (cOop)" %}
  opcode(STY_ZOPC, ST_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

// Store Compressed Klass pointer
instruct storeNKlass(memory mem, iRegN src) %{
  match(Set mem (StoreNKlass mem src));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP_SIZE);
  format %{ "ST      $src,$mem\t # (cKlass)" %}
  opcode(STY_ZOPC, ST_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

// Compare Compressed Pointers

instruct compN_iRegN(iRegN_P2N src1, iRegN_P2N src2, flagsReg cr) %{
  match(Set cr (CmpN src1 src2));
  ins_cost(DEFAULT_COST);
  size(2);
  format %{ "CLR     $src1,$src2\t # (cOop)" %}
  opcode(CLR_ZOPC);
  ins_encode(z_rrform(src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct compN_iRegN_immN(iRegN_P2N src1, immN src2, flagsReg cr) %{
  match(Set cr (CmpN src1 src2));
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "CLFI    $src1,$src2\t # (cOop) compare immediate narrow" %}
  ins_encode %{
    AddressLiteral cOop = __ constant_oop_address((jobject)$src2$$constant);
    __ relocate(cOop.rspec(), 1);
    __ compare_immediate_narrow_oop($src1$$Register, (narrowOop)cOop.value());
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct compNKlass_iRegN_immN(iRegN src1, immNKlass src2, flagsReg cr) %{
  match(Set cr (CmpN src1 src2));
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "CLFI    $src1,$src2\t # (NKlass) compare immediate narrow" %}
  ins_encode %{
    AddressLiteral NKlass = __ constant_metadata_address((Metadata*)$src2$$constant);
    __ relocate(NKlass.rspec(), 1);
    __ compare_immediate_narrow_klass($src1$$Register, (Klass*)NKlass.value());
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct compN_iRegN_immN0(iRegN_P2N src1, immN0 src2, flagsReg cr) %{
  match(Set cr (CmpN src1 src2));
  ins_cost(DEFAULT_COST);
  size(2);
  format %{ "LTR     $src1,$src2\t # (cOop) LTR because comparing against zero" %}
  opcode(LTR_ZOPC);
  ins_encode(z_rrform(src1, src1));
  ins_pipe(pipe_class_dummy);
%}


//----------MemBar Instructions-----------------------------------------------

// Memory barrier flavors

instruct membar_acquire() %{
  match(MemBarAcquire);
  match(LoadFence);
  ins_cost(4*MEMORY_REF_COST);
  size(0);
  format %{ "MEMBAR-acquire" %}
  ins_encode %{ __ z_acquire(); %}
  ins_pipe(pipe_class_dummy);
%}

instruct membar_acquire_lock() %{
  match(MemBarAcquireLock);
  ins_cost(0);
  size(0);
  format %{ "MEMBAR-acquire (CAS in prior FastLock so empty encoding)" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct membar_release() %{
  match(MemBarRelease);
  match(StoreFence);
  ins_cost(4 * MEMORY_REF_COST);
  size(0);
  format %{ "MEMBAR-release" %}
  ins_encode %{ __ z_release(); %}
  ins_pipe(pipe_class_dummy);
%}

instruct membar_release_lock() %{
  match(MemBarReleaseLock);
  ins_cost(0);
  size(0);
  format %{ "MEMBAR-release (CAS in succeeding FastUnlock so empty encoding)" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct membar_volatile() %{
  match(MemBarVolatile);
  ins_cost(4 * MEMORY_REF_COST);
  size(2);
  format %{ "MEMBAR-volatile" %}
  ins_encode %{ __ z_fence(); %}
  ins_pipe(pipe_class_dummy);
%}

instruct unnecessary_membar_volatile() %{
  match(MemBarVolatile);
  predicate(Matcher::post_store_load_barrier(n));
  ins_cost(0);
  size(0);
  format %{ "# MEMBAR-volatile (empty)" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct membar_CPUOrder() %{
  match(MemBarCPUOrder);
  ins_cost(0);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "MEMBAR-CPUOrder (empty)" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct membar_storestore() %{
  match(MemBarStoreStore);
  ins_cost(0);
  size(0);
  format %{ "MEMBAR-storestore (empty)" %}
  ins_encode();
  ins_pipe(pipe_class_dummy);
%}


//----------Register Move Instructions-----------------------------------------
instruct roundDouble_nop(regD dst) %{
  match(Set dst (RoundDouble dst));
  ins_cost(0);
  // TODO: s390 port size(FIXED_SIZE);
  // z/Architecture results are already "rounded" (i.e., normal-format IEEE).
  ins_encode();
  ins_pipe(pipe_class_dummy);
%}

instruct roundFloat_nop(regF dst) %{
  match(Set dst (RoundFloat dst));
  ins_cost(0);
  // TODO: s390 port size(FIXED_SIZE);
  // z/Architecture results are already "rounded" (i.e., normal-format IEEE).
  ins_encode();
  ins_pipe(pipe_class_dummy);
%}

// Cast Long to Pointer for unsafe natives.
instruct castX2P(iRegP dst, iRegL src) %{
  match(Set dst (CastX2P src));
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "LGR     $dst,$src\t # CastX2P" %}
  ins_encode %{ __ lgr_if_needed($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Cast Pointer to Long for unsafe natives.
instruct castP2X(iRegL dst, iRegP_N2P src) %{
  match(Set dst (CastP2X src));
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "LGR     $dst,$src\t # CastP2X" %}
  ins_encode %{ __ lgr_if_needed($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct stfSSD(stackSlotD stkSlot, regD src) %{
  // %%%% TODO: Tell the coalescer that this kind of node is a copy!
  match(Set stkSlot src);   // chain rule
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ " STD   $src,$stkSlot\t # stk" %}
  opcode(STD_ZOPC);
  ins_encode(z_form_rt_mem(src, stkSlot));
  ins_pipe(pipe_class_dummy);
%}

instruct stfSSF(stackSlotF stkSlot, regF src) %{
  // %%%% TODO: Tell the coalescer that this kind of node is a copy!
  match(Set stkSlot src);   // chain rule
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "STE   $src,$stkSlot\t # stk" %}
  opcode(STE_ZOPC);
  ins_encode(z_form_rt_mem(src, stkSlot));
  ins_pipe(pipe_class_dummy);
%}

//----------Conditional Move---------------------------------------------------

instruct cmovN_reg(cmpOp cmp, flagsReg cr, iRegN dst, iRegN_P2N src) %{
  match(Set dst (CMoveN (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveN,$cmp   $dst,$src" %}
  ins_encode(z_enc_cmov_reg(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovN_imm(cmpOp cmp, flagsReg cr, iRegN dst, immN0 src) %{
  match(Set dst (CMoveN (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveN,$cmp   $dst,$src" %}
  ins_encode(z_enc_cmov_imm(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovI_reg(cmpOp cmp, flagsReg cr, iRegI dst, iRegI src) %{
  match(Set dst (CMoveI (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveI,$cmp   $dst,$src" %}
  ins_encode(z_enc_cmov_reg(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovI_imm(cmpOp cmp, flagsReg cr, iRegI dst, immI16 src) %{
  match(Set dst (CMoveI (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveI,$cmp   $dst,$src" %}
  ins_encode(z_enc_cmov_imm(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovP_reg(cmpOp cmp, flagsReg cr, iRegP dst, iRegP_N2P src) %{
  match(Set dst (CMoveP (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveP,$cmp    $dst,$src" %}
  ins_encode(z_enc_cmov_reg(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovP_imm(cmpOp cmp, flagsReg cr, iRegP dst, immP0 src) %{
  match(Set dst (CMoveP (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveP,$cmp  $dst,$src" %}
  ins_encode(z_enc_cmov_imm(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovF_reg(cmpOpF cmp, flagsReg cr, regF dst, regF src) %{
  match(Set dst (CMoveF (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveF,$cmp   $dst,$src" %}
  ins_encode %{
    // Don't emit code if operands are identical (same register).
    if ($dst$$FloatRegister != $src$$FloatRegister) {
      Label done;
      __ z_brc(Assembler::inverse_float_condition((Assembler::branch_condition)$cmp$$cmpcode), done);
      __ z_ler($dst$$FloatRegister, $src$$FloatRegister);
      __ bind(done);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmovD_reg(cmpOpF cmp, flagsReg cr, regD dst, regD src) %{
  match(Set dst (CMoveD (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveD,$cmp   $dst,$src" %}
  ins_encode %{
    // Don't emit code if operands are identical (same register).
    if ($dst$$FloatRegister != $src$$FloatRegister) {
      Label done;
      __ z_brc(Assembler::inverse_float_condition((Assembler::branch_condition)$cmp$$cmpcode), done);
      __ z_ldr($dst$$FloatRegister, $src$$FloatRegister);
      __ bind(done);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmovL_reg(cmpOp cmp, flagsReg cr, iRegL dst, iRegL src) %{
  match(Set dst (CMoveL (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveL,$cmp  $dst,$src" %}
  ins_encode(z_enc_cmov_reg(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

instruct cmovL_imm(cmpOp cmp, flagsReg cr, iRegL dst, immL16 src) %{
  match(Set dst (CMoveL (Binary cmp cr) (Binary dst src)));
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CMoveL,$cmp  $dst,$src" %}
  ins_encode(z_enc_cmov_imm(cmp,dst,src));
  ins_pipe(pipe_class_dummy);
%}

//----------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(threadRegP dst) %{
  match(Set dst (ThreadLocal));
  ins_cost(0);
  size(0);
  ins_should_rematerialize(true);
  format %{ "# $dst=ThreadLocal" %}
  ins_encode(/* empty */);
  ins_pipe(pipe_class_dummy);
%}

instruct checkCastPP(iRegP dst) %{
  match(Set dst (CheckCastPP dst));
  size(0);
  format %{ "# checkcastPP of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct castPP(iRegP dst) %{
  match(Set dst (CastPP dst));
  size(0);
  format %{ "# castPP of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct castII(iRegI dst) %{
  match(Set dst (CastII dst));
  size(0);
  format %{ "# castII of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct castLL(iRegL dst) %{
  match(Set dst (CastLL dst));
  size(0);
  format %{ "# castLL of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct castFF(regF dst) %{
  match(Set dst (CastFF dst));
  size(0);
  format %{ "# castFF of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct castDD(regD dst) %{
  match(Set dst (CastDD dst));
  size(0);
  format %{ "# castDD of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

instruct castVV(iRegL dst) %{
  match(Set dst (CastVV dst));
  size(0);
  format %{ "# castVV of $dst" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

//----------Conditional_store--------------------------------------------------
// Conditional-store of the updated heap-top.
// Used during allocation of the shared heap.
// Sets flags (EQ) on success.

// Implement LoadPLocked. Must be ordered against changes of the memory location
// by storePConditional.
// Don't know whether this is ever used.
instruct loadPLocked(iRegP dst, memory mem) %{
  match(Set dst (LoadPLocked mem));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "LG      $dst,$mem\t # LoadPLocked" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// As compareAndSwapP, but return flag register instead of boolean value in
// int register.
// This instruction is matched if UseTLAB is off. Needed to pass
// option tests.  Mem_ptr must be a memory operand, else this node
// does not get Flag_needs_anti_dependence_check set by adlc. If this
// is not set this node can be rematerialized which leads to errors.
instruct storePConditional(indirect mem_ptr, rarg5RegP oldval, iRegP_N2P newval, flagsReg cr) %{
  match(Set cr (StorePConditional mem_ptr (Binary oldval newval)));
  effect(KILL oldval);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "storePConditional $oldval,$newval,$mem_ptr" %}
  ins_encode(z_enc_casL(oldval, newval, mem_ptr));
  ins_pipe(pipe_class_dummy);
%}

// As compareAndSwapL, but return flag register instead of boolean value in
// int register.
// Used by sun/misc/AtomicLongCSImpl.java. Mem_ptr must be a memory
// operand, else this node does not get
// Flag_needs_anti_dependence_check set by adlc. If this is not set
// this node can be rematerialized which leads to errors.
instruct storeLConditional(indirect mem_ptr, rarg5RegL oldval, iRegL newval, flagsReg cr) %{
  match(Set cr (StoreLConditional mem_ptr (Binary oldval newval)));
  effect(KILL oldval);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "storePConditional $oldval,$newval,$mem_ptr" %}
  ins_encode(z_enc_casL(oldval, newval, mem_ptr));
  ins_pipe(pipe_class_dummy);
%}

// No flag versions for CompareAndSwap{P,I,L,N} because matcher can't match them.

instruct compareAndSwapI_bool(iRegP mem_ptr, rarg5RegI oldval, iRegI newval, iRegI res, flagsReg cr) %{
  match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
  effect(USE mem_ptr, USE_KILL oldval, KILL cr);
  size(16);
  format %{ "$res = CompareAndSwapI $oldval,$newval,$mem_ptr" %}
  ins_encode(z_enc_casI(oldval, newval, mem_ptr),
             z_enc_cctobool(res));
  ins_pipe(pipe_class_dummy);
%}

instruct compareAndSwapL_bool(iRegP mem_ptr, rarg5RegL oldval, iRegL newval, iRegI res, flagsReg cr) %{
  match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
  effect(USE mem_ptr, USE_KILL oldval, KILL cr);
  size(18);
  format %{ "$res = CompareAndSwapL $oldval,$newval,$mem_ptr" %}
  ins_encode(z_enc_casL(oldval, newval, mem_ptr),
             z_enc_cctobool(res));
  ins_pipe(pipe_class_dummy);
%}

instruct compareAndSwapP_bool(iRegP mem_ptr, rarg5RegP oldval, iRegP_N2P newval, iRegI res, flagsReg cr) %{
  match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
  effect(USE mem_ptr, USE_KILL oldval, KILL cr);
  size(18);
  format %{ "$res = CompareAndSwapP $oldval,$newval,$mem_ptr" %}
  ins_encode(z_enc_casL(oldval, newval, mem_ptr),
             z_enc_cctobool(res));
  ins_pipe(pipe_class_dummy);
%}

instruct compareAndSwapN_bool(iRegP mem_ptr, rarg5RegN oldval, iRegN_P2N newval, iRegI res, flagsReg cr) %{
  match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
  effect(USE mem_ptr, USE_KILL oldval, KILL cr);
  size(16);
  format %{ "$res = CompareAndSwapN $oldval,$newval,$mem_ptr" %}
  ins_encode(z_enc_casI(oldval, newval, mem_ptr),
             z_enc_cctobool(res));
  ins_pipe(pipe_class_dummy);
%}

//----------Atomic operations on memory (GetAndSet*, GetAndAdd*)---------------

// Exploit: direct memory arithmetic
// Prereqs: - instructions available
//          - instructions guarantee atomicity
//          - immediate operand to be added
//          - immediate operand is small enough (8-bit signed).
//          - result of instruction is not used
instruct addI_mem_imm8_atomic_no_res(memoryRSY mem, Universe dummy, immI8 src, flagsReg cr) %{
  match(Set dummy (GetAndAddI mem src));
  effect(KILL cr);
  predicate(VM_Version::has_AtomicMemWithImmALUOps() && n->as_LoadStore()->result_not_used());
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "ASI     [$mem],$src\t # GetAndAddI (atomic)" %}
  opcode(ASI_ZOPC);
  ins_encode(z_siyform(mem, src));
  ins_pipe(pipe_class_dummy);
%}

// Fallback: direct memory arithmetic not available
// Disadvantages: - CS-Loop required, very expensive.
//                - more code generated (26 to xx bytes vs. 6 bytes)
instruct addI_mem_imm16_atomic(memoryRSY mem, iRegI dst, immI16 src, iRegI tmp, flagsReg cr) %{
  match(Set dst (GetAndAddI mem src));
  effect(KILL cr, TEMP_DEF dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST+100*DEFAULT_COST);
  format %{ "BEGIN ATOMIC {\n\t"
            "  LGF     $dst,[$mem]\n\t"
            "  AHIK    $tmp,$dst,$src\n\t"
            "  CSY     $dst,$tmp,$mem\n\t"
            "  retry if failed\n\t"
            "} END ATOMIC"
         %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    int      Isrc = $src$$constant;
    Label    retry;

    // Iterate until update with incremented value succeeds.
    __ z_lgf(Rdst, $mem$$Address);    // current contents
    __ bind(retry);
      // Calculate incremented value.
      if (VM_Version::has_DistinctOpnds()) {
        __ z_ahik(Rtmp, Rdst, Isrc);
      } else {
        __ z_lr(Rtmp, Rdst);
        __ z_ahi(Rtmp, Isrc);
      }
      // Swap into memory location.
      __ z_csy(Rdst, Rtmp, $mem$$Address); // Try to store new value.
    __ z_brne(retry);                      // Yikes, concurrent update, need to retry.
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct addI_mem_imm32_atomic(memoryRSY mem, iRegI dst, immI src, iRegI tmp, flagsReg cr) %{
  match(Set dst (GetAndAddI mem src));
  effect(KILL cr, TEMP_DEF dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST+200*DEFAULT_COST);
  format %{ "BEGIN ATOMIC {\n\t"
            "  LGF     $dst,[$mem]\n\t"
            "  LGR     $tmp,$dst\n\t"
            "  AFI     $tmp,$src\n\t"
            "  CSY     $dst,$tmp,$mem\n\t"
            "  retry if failed\n\t"
            "} END ATOMIC"
         %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    int      Isrc = $src$$constant;
    Label    retry;

    // Iterate until update with incremented value succeeds.
    __ z_lgf(Rdst, $mem$$Address);    // current contents
    __ bind(retry);
      // Calculate incremented value.
      __ z_lr(Rtmp, Rdst);
      __ z_afi(Rtmp, Isrc);
      // Swap into memory location.
      __ z_csy(Rdst, Rtmp, $mem$$Address); // Try to store new value.
    __ z_brne(retry);                      // Yikes, concurrent update, need to retry.
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct addI_mem_reg_atomic(memoryRSY mem, iRegI dst, iRegI src, iRegI tmp, flagsReg cr) %{
  match(Set dst (GetAndAddI mem src));
  effect(KILL cr, TEMP_DEF dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST+100*DEFAULT_COST);
  format %{ "BEGIN ATOMIC {\n\t"
            "  LGF     $dst,[$mem]\n\t"
            "  ARK     $tmp,$dst,$src\n\t"
            "  CSY     $dst,$tmp,$mem\n\t"
            "  retry if failed\n\t"
            "} END ATOMIC"
         %}
  ins_encode %{
    Register Rsrc = $src$$Register;
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    Label    retry;

    // Iterate until update with incremented value succeeds.
    __ z_lgf(Rdst, $mem$$Address);  // current contents
    __ bind(retry);
      // Calculate incremented value.
      if (VM_Version::has_DistinctOpnds()) {
        __ z_ark(Rtmp, Rdst, Rsrc);
      } else {
        __ z_lr(Rtmp, Rdst);
        __ z_ar(Rtmp, Rsrc);
      }
      __ z_csy(Rdst, Rtmp, $mem$$Address); // Try to store new value.
    __ z_brne(retry);                      // Yikes, concurrent update, need to retry.
  %}
  ins_pipe(pipe_class_dummy);
%}


// Exploit: direct memory arithmetic
// Prereqs: - instructions available
//          - instructions guarantee atomicity
//          - immediate operand to be added
//          - immediate operand is small enough (8-bit signed).
//          - result of instruction is not used
instruct addL_mem_imm8_atomic_no_res(memoryRSY mem, Universe dummy, immL8 src, flagsReg cr) %{
  match(Set dummy (GetAndAddL mem src));
  effect(KILL cr);
  predicate(VM_Version::has_AtomicMemWithImmALUOps() && n->as_LoadStore()->result_not_used());
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "AGSI    [$mem],$src\t # GetAndAddL (atomic)" %}
  opcode(AGSI_ZOPC);
  ins_encode(z_siyform(mem, src));
  ins_pipe(pipe_class_dummy);
%}

// Fallback: direct memory arithmetic not available
// Disadvantages: - CS-Loop required, very expensive.
//                - more code generated (26 to xx bytes vs. 6 bytes)
instruct addL_mem_imm16_atomic(memoryRSY mem, iRegL dst, immL16 src, iRegL tmp, flagsReg cr) %{
  match(Set dst (GetAndAddL mem src));
  effect(KILL cr, TEMP_DEF dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST+100*DEFAULT_COST);
  format %{ "BEGIN ATOMIC {\n\t"
            "  LG      $dst,[$mem]\n\t"
            "  AGHIK   $tmp,$dst,$src\n\t"
            "  CSG     $dst,$tmp,$mem\n\t"
            "  retry if failed\n\t"
            "} END ATOMIC"
         %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    int      Isrc = $src$$constant;
    Label    retry;

    // Iterate until update with incremented value succeeds.
    __ z_lg(Rdst, $mem$$Address);  // current contents
    __ bind(retry);
      // Calculate incremented value.
      if (VM_Version::has_DistinctOpnds()) {
        __ z_aghik(Rtmp, Rdst, Isrc);
      } else {
        __ z_lgr(Rtmp, Rdst);
        __ z_aghi(Rtmp, Isrc);
      }
      __ z_csg(Rdst, Rtmp, $mem$$Address); // Try to store new value.
    __ z_brne(retry);                      // Yikes, concurrent update, need to retry.
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct addL_mem_imm32_atomic(memoryRSY mem, iRegL dst, immL32 src, iRegL tmp, flagsReg cr) %{
  match(Set dst (GetAndAddL mem src));
  effect(KILL cr, TEMP_DEF dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST+100*DEFAULT_COST);
  format %{ "BEGIN ATOMIC {\n\t"
            "  LG      $dst,[$mem]\n\t"
            "  LGR     $tmp,$dst\n\t"
            "  AGFI    $tmp,$src\n\t"
            "  CSG     $dst,$tmp,$mem\n\t"
            "  retry if failed\n\t"
            "} END ATOMIC"
         %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    int      Isrc = $src$$constant;
    Label    retry;

    // Iterate until update with incremented value succeeds.
    __ z_lg(Rdst, $mem$$Address);  // current contents
    __ bind(retry);
      // Calculate incremented value.
      __ z_lgr(Rtmp, Rdst);
      __ z_agfi(Rtmp, Isrc);
      __ z_csg(Rdst, Rtmp, $mem$$Address); // Try to store new value.
    __ z_brne(retry);                      // Yikes, concurrent update, need to retry.
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct addL_mem_reg_atomic(memoryRSY mem, iRegL dst, iRegL src, iRegL tmp, flagsReg cr) %{
  match(Set dst (GetAndAddL mem src));
  effect(KILL cr, TEMP_DEF dst, TEMP tmp);
  ins_cost(MEMORY_REF_COST+100*DEFAULT_COST);
  format %{ "BEGIN ATOMIC {\n\t"
            "  LG      $dst,[$mem]\n\t"
            "  AGRK    $tmp,$dst,$src\n\t"
            "  CSG     $dst,$tmp,$mem\n\t"
            "  retry if failed\n\t"
            "} END ATOMIC"
         %}
  ins_encode %{
    Register Rsrc = $src$$Register;
    Register Rdst = $dst$$Register;
    Register Rtmp = $tmp$$Register;
    Label    retry;

    // Iterate until update with incremented value succeeds.
    __ z_lg(Rdst, $mem$$Address);  // current contents
    __ bind(retry);
      // Calculate incremented value.
      if (VM_Version::has_DistinctOpnds()) {
        __ z_agrk(Rtmp, Rdst, Rsrc);
      } else {
        __ z_lgr(Rtmp, Rdst);
        __ z_agr(Rtmp, Rsrc);
      }
      __ z_csg(Rdst, Rtmp, $mem$$Address); // Try to store new value.
    __ z_brne(retry);                      // Yikes, concurrent update, need to retry.
  %}
  ins_pipe(pipe_class_dummy);
%}

// Increment value in memory, save old value in dst.
instruct addI_mem_reg_atomic_z196(memoryRSY mem, iRegI dst, iRegI src) %{
  match(Set dst (GetAndAddI mem src));
  predicate(VM_Version::has_LoadAndALUAtomicV1());
  ins_cost(MEMORY_REF_COST + DEFAULT_COST);
  size(6);
  format %{ "LAA     $dst,$src,[$mem]" %}
  ins_encode %{ __ z_laa($dst$$Register, $src$$Register, $mem$$Address); %}
  ins_pipe(pipe_class_dummy);
%}

// Increment value in memory, save old value in dst.
instruct addL_mem_reg_atomic_z196(memoryRSY mem, iRegL dst, iRegL src) %{
  match(Set dst (GetAndAddL mem src));
  predicate(VM_Version::has_LoadAndALUAtomicV1());
  ins_cost(MEMORY_REF_COST + DEFAULT_COST);
  size(6);
  format %{ "LAAG    $dst,$src,[$mem]" %}
  ins_encode %{ __ z_laag($dst$$Register, $src$$Register, $mem$$Address); %}
  ins_pipe(pipe_class_dummy);
%}


instruct xchgI_reg_mem(memoryRSY mem, iRegI dst, iRegI tmp, flagsReg cr) %{
  match(Set dst (GetAndSetI mem dst));
  effect(KILL cr, TEMP tmp); // USE_DEF dst by match rule.
  format %{ "XCHGI   $dst,[$mem]\t # EXCHANGE (int, atomic), temp $tmp" %}
  ins_encode(z_enc_SwapI(mem, dst, tmp));
  ins_pipe(pipe_class_dummy);
%}

instruct xchgL_reg_mem(memoryRSY mem, iRegL dst, iRegL tmp, flagsReg cr) %{
  match(Set dst (GetAndSetL mem dst));
  effect(KILL cr, TEMP tmp); // USE_DEF dst by match rule.
  format %{ "XCHGL   $dst,[$mem]\t # EXCHANGE (long, atomic), temp $tmp" %}
  ins_encode(z_enc_SwapL(mem, dst, tmp));
  ins_pipe(pipe_class_dummy);
%}

instruct xchgN_reg_mem(memoryRSY mem, iRegN dst, iRegI tmp, flagsReg cr) %{
  match(Set dst (GetAndSetN mem dst));
  effect(KILL cr, TEMP tmp); // USE_DEF dst by match rule.
  format %{ "XCHGN   $dst,[$mem]\t # EXCHANGE (coop, atomic), temp $tmp" %}
  ins_encode(z_enc_SwapI(mem, dst, tmp));
  ins_pipe(pipe_class_dummy);
%}

instruct xchgP_reg_mem(memoryRSY mem, iRegP dst, iRegL tmp, flagsReg cr) %{
  match(Set dst (GetAndSetP mem dst));
  effect(KILL cr, TEMP tmp); // USE_DEF dst by match rule.
  format %{ "XCHGP   $dst,[$mem]\t # EXCHANGE (oop, atomic), temp $tmp" %}
  ins_encode(z_enc_SwapL(mem, dst, tmp));
  ins_pipe(pipe_class_dummy);
%}


//----------Arithmetic Instructions--------------------------------------------

// The rules are sorted by right operand type and operand length. Please keep
// it that way.
// Left operand type is always reg. Left operand len is I, L, P
// Right operand type is reg, imm, mem. Right operand len is S, I, L, P
// Special instruction formats, e.g. multi-operand, are inserted at the end.

// ADD

// REG = REG + REG

// Register Addition
instruct addI_reg_reg_CISC(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (AddI dst src));
  effect(KILL cr);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AR      $dst,$src\t # int  CISC ALU" %}
  opcode(AR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct addI_reg_reg_RISC(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (AddI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "ARK     $dst,$src1,$src2\t # int  RISC ALU" %}
  opcode(ARK_ZOPC);
  ins_encode(z_rrfform(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// REG = REG + IMM

// Avoid use of LA(Y) for general ALU operation.
// Immediate Addition
instruct addI_reg_imm16_CISC(iRegI dst, immI16 con, flagsReg cr) %{
  match(Set dst (AddI dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AHI     $dst,$con\t # int  CISC ALU" %}
  opcode(AHI_ZOPC);
  ins_encode(z_riform_signed(dst, con));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
// Immediate Addition
instruct addI_reg_imm16_RISC(iRegI dst, iRegI src, immI16 con, flagsReg cr) %{
  match(Set dst (AddI src con));
  effect(KILL cr);
  predicate( VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AHIK    $dst,$src,$con\t # int  RISC ALU" %}
  opcode(AHIK_ZOPC);
  ins_encode(z_rieform_d(dst, src, con));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Addition
instruct addI_reg_imm32(iRegI dst, immI src, flagsReg cr) %{
  match(Set dst (AddI dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  size(6);
  format %{ "AFI     $dst,$src" %}
  opcode(AFI_ZOPC);
  ins_encode(z_rilform_signed(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Addition
instruct addI_reg_imm12(iRegI dst, iRegI src, uimmI12 con) %{
  match(Set dst (AddI src con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,$con(,$src)\t # int d12(,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg(dst, con, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Addition
instruct addI_reg_imm20(iRegI dst, iRegI src, immI20 con) %{
  match(Set dst (AddI src con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LAY     $dst,$con(,$src)\t # int d20(,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg(dst, con, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addI_reg_reg_imm12(iRegI dst, iRegI src1, iRegI src2, uimmI12 con) %{
  match(Set dst (AddI (AddI src1 src2) con));
  predicate( PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,$con($src1,$src2)\t # int d12(x,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct addI_reg_reg_imm20(iRegI dst, iRegI src1, iRegI src2, immI20 con) %{
  match(Set dst (AddI (AddI src1 src2) con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LAY     $dst,$con($src1,$src2)\t # int d20(x,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// REG = REG + MEM

instruct addI_Reg_mem(iRegI dst, memory src, flagsReg cr)%{
  match(Set dst (AddI dst (LoadI src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "A(Y)    $dst, $src\t # int" %}
  opcode(AY_ZOPC, A_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// MEM = MEM + IMM

// Add Immediate to 4-byte memory operand and result
instruct addI_mem_imm(memoryRSY mem, immI8 src, flagsReg cr) %{
  match(Set mem (StoreI mem (AddI (LoadI mem) src)));
  effect(KILL cr);
  predicate(VM_Version::has_MemWithImmALUOps());
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "ASI     $mem,$src\t # direct mem add 4" %}
  opcode(ASI_ZOPC);
  ins_encode(z_siyform(mem, src));
  ins_pipe(pipe_class_dummy);
%}


//

// REG = REG + REG

instruct addL_reg_regI(iRegL dst, iRegI src, flagsReg cr) %{
  match(Set dst (AddL dst (ConvI2L src)));
  effect(KILL cr);
  size(4);
  format %{ "AGFR    $dst,$src\t # long<-int CISC ALU" %}
  opcode(AGFR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addL_reg_reg_CISC(iRegL dst, iRegL src, flagsReg cr) %{
  match(Set dst (AddL dst src));
  effect(KILL cr);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AGR     $dst, $src\t # long CISC ALU" %}
  opcode(AGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct addL_reg_reg_RISC(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{
  match(Set dst (AddL src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "AGRK    $dst,$src1,$src2\t # long RISC ALU" %}
  opcode(AGRK_ZOPC);
  ins_encode(z_rrfform(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// REG = REG + IMM

instruct addL_reg_imm12(iRegL dst, iRegL src, uimmL12 con) %{
  match(Set dst (AddL src con));
  predicate( PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,$con(,$src)\t # long d12(,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg(dst, con, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addL_reg_imm20(iRegL dst, iRegL src, immL20 con) %{
  match(Set dst (AddL src con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LAY     $dst,$con(,$src)\t # long d20(,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg(dst, con, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addL_reg_imm32(iRegL dst, immL32 con, flagsReg cr) %{
  match(Set dst (AddL dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  size(6);
  format %{ "AGFI    $dst,$con\t # long CISC ALU" %}
  opcode(AGFI_ZOPC);
  ins_encode(z_rilform_signed(dst, con));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct addL_reg_imm16_CISC(iRegL dst, immL16 con, flagsReg cr) %{
  match(Set dst (AddL dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AGHI    $dst,$con\t # long CISC ALU" %}
  opcode(AGHI_ZOPC);
  ins_encode(z_riform_signed(dst, con));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct addL_reg_imm16_RISC(iRegL dst, iRegL src, immL16 con, flagsReg cr) %{
  match(Set dst (AddL src con));
  effect(KILL cr);
  predicate( VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "AGHIK   $dst,$src,$con\t # long RISC ALU" %}
  opcode(AGHIK_ZOPC);
  ins_encode(z_rieform_d(dst, src, con));
  ins_pipe(pipe_class_dummy);
%}

// REG = REG + MEM

instruct addL_Reg_memI(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (AddL dst (ConvI2L (LoadI src))));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "AGF     $dst, $src\t # long/int" %}
  opcode(AGF_ZOPC, AGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addL_Reg_mem(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (AddL dst (LoadL src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "AG      $dst, $src\t # long" %}
  opcode(AG_ZOPC, AG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addL_reg_reg_imm12(iRegL dst, iRegL src1, iRegL src2, uimmL12 con) %{
  match(Set dst (AddL (AddL src1 src2) con));
  predicate( PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA     $dst,$con($src1,$src2)\t # long d12(x,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct addL_reg_reg_imm20(iRegL dst, iRegL src1, iRegL src2, immL20 con) %{
  match(Set dst (AddL (AddL src1 src2) con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LAY    $dst,$con($src1,$src2)\t # long d20(x,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// MEM = MEM + IMM

// Add Immediate to 8-byte memory operand and result.
instruct addL_mem_imm(memoryRSY mem, immL8 src, flagsReg cr) %{
  match(Set mem (StoreL mem (AddL (LoadL mem) src)));
  effect(KILL cr);
  predicate(VM_Version::has_MemWithImmALUOps());
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "AGSI    $mem,$src\t # direct mem add 8" %}
  opcode(AGSI_ZOPC);
  ins_encode(z_siyform(mem, src));
  ins_pipe(pipe_class_dummy);
%}


// REG = REG + REG

// Ptr Addition
instruct addP_reg_reg_LA(iRegP dst, iRegP_N2P src1, iRegL src2) %{
  match(Set dst (AddP src1 src2));
  predicate( PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "LA      $dst,#0($src1,$src2)\t # ptr 0(x,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg_reg(dst, 0x0, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// Ptr Addition
// Avoid use of LA(Y) for general ALU operation.
instruct addP_reg_reg_CISC(iRegP dst, iRegL src, flagsReg cr) %{
  match(Set dst (AddP dst src));
  effect(KILL cr);
  predicate(!PreferLAoverADD && !VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "ALGR    $dst,$src\t # ptr CICS ALU" %}
  opcode(ALGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Ptr Addition
// Avoid use of LA(Y) for general ALU operation.
instruct addP_reg_reg_RISC(iRegP dst, iRegP_N2P src1, iRegL src2, flagsReg cr) %{
  match(Set dst (AddP src1 src2));
  effect(KILL cr);
  predicate(!PreferLAoverADD && VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "ALGRK   $dst,$src1,$src2\t # ptr RISC ALU" %}
  opcode(ALGRK_ZOPC);
  ins_encode(z_rrfform(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// REG = REG + IMM

instruct addP_reg_imm12(iRegP dst, iRegP_N2P src, uimmL12 con) %{
  match(Set dst (AddP src con));
  predicate( PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,$con(,$src)\t # ptr d12(,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg(dst, con, src));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct addP_reg_imm16_CISC(iRegP dst, immL16 src, flagsReg cr) %{
  match(Set dst (AddP dst src));
  effect(KILL cr);
  predicate(!PreferLAoverADD && !VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AGHI    $dst,$src\t # ptr CISC ALU" %}
  opcode(AGHI_ZOPC);
  ins_encode(z_riform_signed(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct addP_reg_imm16_RISC(iRegP dst, iRegP_N2P src, immL16 con, flagsReg cr) %{
  match(Set dst (AddP src con));
  effect(KILL cr);
  predicate(!PreferLAoverADD && VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "ALGHSIK $dst,$src,$con\t # ptr RISC ALU" %}
  opcode(ALGHSIK_ZOPC);
  ins_encode(z_rieform_d(dst, src, con));
  ins_pipe(pipe_class_dummy);
%}

instruct addP_reg_imm20(iRegP dst, memoryRegP src, immL20 con) %{
  match(Set dst (AddP src con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "LAY     $dst,$con(,$src)\t # ptr d20(,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg(dst, con, src));
  ins_pipe(pipe_class_dummy);
%}

// Pointer Immediate Addition
instruct addP_reg_imm32(iRegP dst, immL32 src, flagsReg cr) %{
  match(Set dst (AddP dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AGFI    $dst,$src\t # ptr" %}
  opcode(AGFI_ZOPC);
  ins_encode(z_rilform_signed(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// REG = REG1 + REG2 + IMM

instruct addP_reg_reg_imm12(iRegP dst, memoryRegP src1, iRegL src2, uimmL12 con) %{
  match(Set dst (AddP (AddP src1 src2) con));
  predicate( PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,$con($src1,$src2)\t # ptr d12(x,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct addP_regN_reg_imm12(iRegP dst, iRegP_N2P src1, iRegL src2, uimmL12 con) %{
  match(Set dst (AddP (AddP src1 src2) con));
  predicate( PreferLAoverADD && CompressedOops::base() == NULL && CompressedOops::shift() == 0);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,$con($src1,$src2)\t # ptr d12(x,b)" %}
  opcode(LA_ZOPC);
  ins_encode(z_rxform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct addP_reg_reg_imm20(iRegP dst, memoryRegP src1, iRegL src2, immL20 con) %{
  match(Set dst (AddP (AddP src1 src2) con));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LAY     $dst,$con($src1,$src2)\t # ptr d20(x,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct addP_regN_reg_imm20(iRegP dst, iRegP_N2P src1, iRegL src2, immL20 con) %{
  match(Set dst (AddP (AddP src1 src2) con));
  predicate( PreferLAoverADD && CompressedOops::base() == NULL && CompressedOops::shift() == 0);
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LAY     $dst,$con($src1,$src2)\t # ptr d20(x,b)" %}
  opcode(LAY_ZOPC);
  ins_encode(z_rxyform_imm_reg_reg(dst, con, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// MEM = MEM + IMM

// Add Immediate to 8-byte memory operand and result
instruct addP_mem_imm(memoryRSY mem, immL8 src, flagsReg cr) %{
  match(Set mem (StoreP mem (AddP (LoadP mem) src)));
  effect(KILL cr);
  predicate(VM_Version::has_MemWithImmALUOps());
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "AGSI    $mem,$src\t # direct mem add 8 (ptr)" %}
  opcode(AGSI_ZOPC);
  ins_encode(z_siyform(mem, src));
  ins_pipe(pipe_class_dummy);
%}

// SUB

// Register Subtraction
instruct subI_reg_reg_CISC(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (SubI dst src));
  effect(KILL cr);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "SR      $dst,$src\t # int  CISC ALU" %}
  opcode(SR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct subI_reg_reg_RISC(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (SubI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "SRK     $dst,$src1,$src2\t # int  RISC ALU" %}
  opcode(SRK_ZOPC);
  ins_encode(z_rrfform(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct subI_Reg_mem(iRegI dst, memory src, flagsReg cr)%{
  match(Set dst (SubI dst (LoadI src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "S(Y)    $dst, $src\t # int" %}
  opcode(SY_ZOPC, S_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct subI_zero_reg(iRegI dst, immI_0 zero, iRegI src, flagsReg cr) %{
  match(Set dst (SubI zero src));
  effect(KILL cr);
  size(2);
  format %{ "NEG     $dst, $src" %}
  ins_encode %{ __ z_lcr($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

//

// Long subtraction
instruct subL_reg_reg_CISC(iRegL dst, iRegL src, flagsReg cr) %{
  match(Set dst (SubL dst src));
  effect(KILL cr);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "SGR     $dst,$src\t # int  CISC ALU" %}
  opcode(SGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Avoid use of LA(Y) for general ALU operation.
instruct subL_reg_reg_RISC(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{
  match(Set dst (SubL src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_DistinctOpnds());
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "SGRK    $dst,$src1,$src2\t # int  RISC ALU" %}
  opcode(SGRK_ZOPC);
  ins_encode(z_rrfform(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct subL_reg_regI_CISC(iRegL dst, iRegI src, flagsReg cr) %{
  match(Set dst (SubL dst (ConvI2L src)));
  effect(KILL cr);
  size(4);
  format %{ "SGFR    $dst, $src\t # int  CISC ALU" %}
  opcode(SGFR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct subL_Reg_memI(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (SubL dst (ConvI2L (LoadI src))));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "SGF     $dst, $src\t # long/int" %}
  opcode(SGF_ZOPC, SGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct subL_Reg_mem(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (SubL dst (LoadL src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "SG      $dst, $src\t # long" %}
  opcode(SG_ZOPC, SG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Moved declaration of negL_reg_reg before encode nodes, where it is used.

//  MUL

// Register Multiplication
instruct mulI_reg_reg(iRegI dst, iRegI src) %{
  match(Set dst (MulI dst src));
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "MSR     $dst, $src" %}
  opcode(MSR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Multiplication
instruct mulI_reg_imm16(iRegI dst, immI16 con) %{
  match(Set dst (MulI dst con));
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "MHI     $dst,$con" %}
  opcode(MHI_ZOPC);
  ins_encode(z_riform_signed(dst,con));
  ins_pipe(pipe_class_dummy);
%}

// Immediate (32bit) Multiplication
instruct mulI_reg_imm32(iRegI dst, immI con) %{
  match(Set dst (MulI dst con));
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "MSFI    $dst,$con" %}
  opcode(MSFI_ZOPC);
  ins_encode(z_rilform_signed(dst,con));
  ins_pipe(pipe_class_dummy);
%}

instruct mulI_Reg_mem(iRegI dst, memory src)%{
  match(Set dst (MulI dst (LoadI src)));
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MS(Y)   $dst, $src\t # int" %}
  opcode(MSY_ZOPC, MS_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//

instruct mulL_reg_regI(iRegL dst, iRegI src) %{
  match(Set dst (MulL dst (ConvI2L src)));
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "MSGFR   $dst $src\t # long/int" %}
  opcode(MSGFR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct mulL_reg_reg(iRegL dst, iRegL src) %{
  match(Set dst (MulL dst src));
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "MSGR    $dst $src\t # long" %}
  opcode(MSGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Multiplication
instruct mulL_reg_imm16(iRegL dst, immL16 src) %{
  match(Set dst (MulL dst src));
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "MGHI    $dst,$src\t # long" %}
  opcode(MGHI_ZOPC);
  ins_encode(z_riform_signed(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate (32bit) Multiplication
instruct mulL_reg_imm32(iRegL dst, immL32 con) %{
  match(Set dst (MulL dst con));
  ins_cost(DEFAULT_COST);
  size(6);
  format %{ "MSGFI   $dst,$con" %}
  opcode(MSGFI_ZOPC);
  ins_encode(z_rilform_signed(dst,con));
  ins_pipe(pipe_class_dummy);
%}

instruct mulL_Reg_memI(iRegL dst, memory src)%{
  match(Set dst (MulL dst (ConvI2L (LoadI src))));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "MSGF    $dst, $src\t # long" %}
  opcode(MSGF_ZOPC, MSGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct mulL_Reg_mem(iRegL dst, memory src)%{
  match(Set dst (MulL dst (LoadL src)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "MSG     $dst, $src\t # long" %}
  opcode(MSG_ZOPC, MSG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct mulHiL_reg_reg(revenRegL Rdst, roddRegL Rsrc1, iRegL Rsrc2, iRegL Rtmp1, flagsReg cr)%{
  match(Set Rdst (MulHiL Rsrc1 Rsrc2));
  effect(TEMP_DEF Rdst, USE_KILL Rsrc1, TEMP Rtmp1, KILL cr);
  ins_cost(7*DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MulHiL  $Rdst, $Rsrc1, $Rsrc2\t # Multiply High Long" %}
  ins_encode%{
    Register dst  = $Rdst$$Register;
    Register src1 = $Rsrc1$$Register;
    Register src2 = $Rsrc2$$Register;
    Register tmp1 = $Rtmp1$$Register;
    Register tmp2 = $Rdst$$Register;
    // z/Architecture has only unsigned multiply (64 * 64 -> 128).
    // implementing mulhs(a,b) = mulhu(a,b) – (a & (b>>63)) – (b & (a>>63))
    __ z_srag(tmp2, src1, 63);  // a>>63
    __ z_srag(tmp1, src2, 63);  // b>>63
    __ z_ngr(tmp2, src2);       // b & (a>>63)
    __ z_ngr(tmp1, src1);       // a & (b>>63)
    __ z_agr(tmp1, tmp2);       // ((a & (b>>63)) + (b & (a>>63)))
    __ z_mlgr(dst, src2);       // tricky: 128-bit product is written to even/odd pair (dst,src1),
                                //         multiplicand is taken from oddReg (src1), multiplier in src2.
    __ z_sgr(dst, tmp1);
  %}
  ins_pipe(pipe_class_dummy);
%}

//  DIV

// Integer DIVMOD with Register, both quotient and mod results
instruct divModI_reg_divmod(roddRegI dst1src1, revenRegI dst2, noOdd_iRegI src2, flagsReg cr) %{
  match(DivModI dst1src1 src2);
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size((VM_Version::has_CompareBranch() ? 24 : 26));
  format %{ "DIVMODI ($dst1src1, $dst2) $src2" %}
  ins_encode %{
    Register d1s1 = $dst1src1$$Register;
    Register d2   = $dst2$$Register;
    Register s2   = $src2$$Register;

    assert_different_registers(d1s1, s2);

    Label do_div, done_div;
    if (VM_Version::has_CompareBranch()) {
      __ z_cij(s2, -1, Assembler::bcondNotEqual, do_div);
    } else {
      __ z_chi(s2, -1);
      __ z_brne(do_div);
    }
    __ z_lcr(d1s1, d1s1);
    __ clear_reg(d2, false, false);
    __ z_bru(done_div);
    __ bind(do_div);
    __ z_lgfr(d1s1, d1s1);
    __ z_dsgfr(d2, s2);
    __ bind(done_div);
  %}
  ins_pipe(pipe_class_dummy);
%}


// Register Division
instruct divI_reg_reg(roddRegI dst, iRegI src1, noOdd_iRegI src2, revenRegI tmp, flagsReg cr) %{
  match(Set dst (DivI src1 src2));
  effect(KILL tmp, KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size((VM_Version::has_CompareBranch() ? 20 : 22));
  format %{ "DIV_checked $dst, $src1,$src2\t # treats special case 0x80../-1" %}
  ins_encode %{
    Register a = $src1$$Register;
    Register b = $src2$$Register;
    Register t = $dst$$Register;

    assert_different_registers(t, b);

    Label do_div, done_div;
    if (VM_Version::has_CompareBranch()) {
      __ z_cij(b, -1, Assembler::bcondNotEqual, do_div);
    } else {
      __ z_chi(b, -1);
      __ z_brne(do_div);
    }
    __ z_lcr(t, a);
    __ z_bru(done_div);
    __ bind(do_div);
    __ z_lgfr(t, a);
    __ z_dsgfr(t->predecessor()/* t is odd part of a register pair. */, b);
    __ bind(done_div);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Immediate Division
instruct divI_reg_imm16(roddRegI dst, iRegI src1, immI16 src2, revenRegI tmp, flagsReg cr) %{
  match(Set dst (DivI src1 src2));
  effect(KILL tmp, KILL cr);  // R0 is killed, too.
  ins_cost(2 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "DIV_const  $dst,$src1,$src2" %}
  ins_encode %{
    // No sign extension of Rdividend needed here.
    if ($src2$$constant != -1) {
      __ z_lghi(Z_R0_scratch, $src2$$constant);
      __ z_lgfr($dst$$Register, $src1$$Register);
      __ z_dsgfr($dst$$Register->predecessor()/* Dst is odd part of a register pair. */, Z_R0_scratch);
    } else {
      __ z_lcr($dst$$Register, $src1$$Register);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Long DIVMOD with Register, both quotient and mod results
instruct divModL_reg_divmod(roddRegL dst1src1, revenRegL dst2, iRegL src2, flagsReg cr) %{
  match(DivModL dst1src1 src2);
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size((VM_Version::has_CompareBranch() ? 22 : 24));
  format %{ "DIVMODL ($dst1src1, $dst2) $src2" %}
  ins_encode %{
    Register d1s1 = $dst1src1$$Register;
    Register d2   = $dst2$$Register;
    Register s2   = $src2$$Register;

    Label do_div, done_div;
    if (VM_Version::has_CompareBranch()) {
      __ z_cgij(s2, -1, Assembler::bcondNotEqual, do_div);
    } else {
      __ z_cghi(s2, -1);
      __ z_brne(do_div);
    }
    __ z_lcgr(d1s1, d1s1);
    // indicate unused result
    (void) __ clear_reg(d2, true, false);
    __ z_bru(done_div);
    __ bind(do_div);
    __ z_dsgr(d2, s2);
    __ bind(done_div);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Long Division
instruct divL_reg_reg(roddRegL dst, iRegL src, revenRegL tmp, flagsReg cr) %{
  match(Set dst (DivL dst src));
  effect(KILL tmp, KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size((VM_Version::has_CompareBranch() ? 18 : 20));
  format %{ "DIVG_checked  $dst, $src\t # long, treats special case 0x80../-1" %}
  ins_encode %{
    Register b = $src$$Register;
    Register t = $dst$$Register;

    Label done_div;
    __ z_lcgr(t, t);    // Does no harm. divisor is in other register.
    if (VM_Version::has_CompareBranch()) {
      __ z_cgij(b, -1, Assembler::bcondEqual, done_div);
    } else {
      __ z_cghi(b, -1);
      __ z_bre(done_div);
    }
    __ z_lcgr(t, t);    // Restore sign.
    __ z_dsgr(t->predecessor()/* t is odd part of a register pair. */, b);
    __ bind(done_div);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Immediate Long Division
instruct divL_reg_imm16(roddRegL dst, iRegL src1, immL16 src2, revenRegL tmp, flagsReg cr) %{
  match(Set dst (DivL src1 src2));
  effect(KILL tmp, KILL cr);  // R0 is killed, too.
  ins_cost(2 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "DIVG_const  $dst,$src1,$src2\t # long" %}
  ins_encode %{
    if ($src2$$constant != -1) {
      __ z_lghi(Z_R0_scratch, $src2$$constant);
      __ lgr_if_needed($dst$$Register, $src1$$Register);
      __ z_dsgr($dst$$Register->predecessor()/* Dst is odd part of a register pair. */, Z_R0_scratch);
    } else {
      __ z_lcgr($dst$$Register, $src1$$Register);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// REM

// Integer Remainder
// Register Remainder
instruct modI_reg_reg(revenRegI dst, iRegI src1, noOdd_iRegI src2, roddRegI tmp, flagsReg cr) %{
  match(Set dst (ModI src1 src2));
  effect(KILL tmp, KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MOD_checked   $dst,$src1,$src2" %}
  ins_encode %{
    Register a = $src1$$Register;
    Register b = $src2$$Register;
    Register t = $dst$$Register;
    assert_different_registers(t->successor(), b);

    Label do_div, done_div;

    if ((t->encoding() != b->encoding()) && (t->encoding() != a->encoding())) {
      (void) __ clear_reg(t, true, false);  // Does no harm. Operands are in other regs.
      if (VM_Version::has_CompareBranch()) {
        __ z_cij(b, -1, Assembler::bcondEqual, done_div);
      } else {
        __ z_chi(b, -1);
        __ z_bre(done_div);
      }
      __ z_lgfr(t->successor(), a);
      __ z_dsgfr(t/* t is even part of a register pair. */, b);
    } else {
      if (VM_Version::has_CompareBranch()) {
        __ z_cij(b, -1, Assembler::bcondNotEqual, do_div);
      } else {
        __ z_chi(b, -1);
        __ z_brne(do_div);
      }
      __ clear_reg(t, true, false);
      __ z_bru(done_div);
      __ bind(do_div);
      __ z_lgfr(t->successor(), a);
      __ z_dsgfr(t/* t is even part of a register pair. */, b);
    }
    __ bind(done_div);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Immediate Remainder
instruct modI_reg_imm16(revenRegI dst, iRegI src1, immI16 src2, roddRegI tmp, flagsReg cr) %{
  match(Set dst (ModI src1 src2));
  effect(KILL tmp, KILL cr); // R0 is killed, too.
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MOD_const  $dst,src1,$src2" %}
  ins_encode %{
    assert_different_registers($dst$$Register, $src1$$Register);
    assert_different_registers($dst$$Register->successor(), $src1$$Register);
    int divisor = $src2$$constant;

    if (divisor != -1) {
      __ z_lghi(Z_R0_scratch, divisor);
      __ z_lgfr($dst$$Register->successor(), $src1$$Register);
      __ z_dsgfr($dst$$Register/* Dst is even part of a register pair. */, Z_R0_scratch); // Instruction kills tmp.
    } else {
      __ clear_reg($dst$$Register, true, false);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Long Remainder
instruct modL_reg_reg(revenRegL dst, roddRegL src1, iRegL src2, flagsReg cr) %{
  match(Set dst (ModL src1 src2));
  effect(KILL src1, KILL cr); // R0 is killed, too.
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MODG_checked   $dst,$src1,$src2" %}
  ins_encode %{
    Register a = $src1$$Register;
    Register b = $src2$$Register;
    Register t = $dst$$Register;
    assert(t->successor() == a, "(t,a) is an even-odd pair" );

    Label do_div, done_div;
    if (t->encoding() != b->encoding()) {
      (void) __ clear_reg(t, true, false); // Does no harm. Dividend is in successor.
      if (VM_Version::has_CompareBranch()) {
        __ z_cgij(b, -1, Assembler::bcondEqual, done_div);
      } else {
        __ z_cghi(b, -1);
        __ z_bre(done_div);
      }
      __ z_dsgr(t, b);
    } else {
      if (VM_Version::has_CompareBranch()) {
        __ z_cgij(b, -1, Assembler::bcondNotEqual, do_div);
      } else {
        __ z_cghi(b, -1);
        __ z_brne(do_div);
      }
      __ clear_reg(t, true, false);
      __ z_bru(done_div);
      __ bind(do_div);
      __ z_dsgr(t, b);
    }
    __ bind(done_div);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Long Remainder
instruct modL_reg_imm16(revenRegL dst, iRegL src1, immL16 src2, roddRegL tmp, flagsReg cr) %{
  match(Set dst (ModL src1 src2));
  effect(KILL tmp, KILL cr); // R0 is killed, too.
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MODG_const  $dst,src1,$src2\t # long" %}
  ins_encode %{
    int divisor = $src2$$constant;
    if (divisor != -1) {
      __ z_lghi(Z_R0_scratch, divisor);
      __ z_lgr($dst$$Register->successor(), $src1$$Register);
      __ z_dsgr($dst$$Register /* Dst is even part of a register pair. */, Z_R0_scratch);  // Instruction kills tmp.
    } else {
      __ clear_reg($dst$$Register, true, false);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// SHIFT

// Shift left logical

// Register Shift Left variable
instruct sllI_reg_reg(iRegI dst, iRegI src, iRegI nbits, flagsReg cr) %{
  match(Set dst (LShiftI src nbits));
  effect(KILL cr); // R1 is killed, too.
  ins_cost(3 * DEFAULT_COST);
  size(14);
  format %{ "SLL     $dst,$src,[$nbits] & 31\t # use RISC-like SLLG also for int" %}
  ins_encode %{
    __ z_lgr(Z_R1_scratch, $nbits$$Register);
    __ z_nill(Z_R1_scratch, BitsPerJavaInteger-1);
    __ z_sllg($dst$$Register, $src$$Register, 0, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Left Immediate
// Constant shift count is masked in ideal graph already.
instruct sllI_reg_imm(iRegI dst, iRegI src, immI nbits) %{
  match(Set dst (LShiftI src nbits));
  size(6);
  format %{ "SLL     $dst,$src,$nbits\t # use RISC-like SLLG also for int" %}
  ins_encode %{
    int Nbit = $nbits$$constant;
    assert((Nbit & (BitsPerJavaInteger - 1)) == Nbit, "Check shift mask in ideal graph");
    __ z_sllg($dst$$Register, $src$$Register, Nbit & (BitsPerJavaInteger - 1), Z_R0);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Left Immediate by 1bit
instruct sllI_reg_imm_1(iRegI dst, iRegI src, immI_1 nbits) %{
  match(Set dst (LShiftI src nbits));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,#0($src,$src)\t # SLL by 1 (int)" %}
  ins_encode %{ __ z_la($dst$$Register, 0, $src$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Left Long
instruct sllL_reg_reg(iRegL dst, iRegL src1, iRegI nbits) %{
  match(Set dst (LShiftL src1 nbits));
  size(6);
  format %{ "SLLG    $dst,$src1,[$nbits]" %}
  opcode(SLLG_ZOPC);
  ins_encode(z_rsyform_reg_reg(dst, src1, nbits));
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Left Long Immediate
instruct sllL_reg_imm(iRegL dst, iRegL src1, immI nbits) %{
  match(Set dst (LShiftL src1 nbits));
  size(6);
  format %{ "SLLG    $dst,$src1,$nbits" %}
  opcode(SLLG_ZOPC);
  ins_encode(z_rsyform_const(dst, src1, nbits));
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Left Long Immediate by 1bit
instruct sllL_reg_imm_1(iRegL dst, iRegL src1, immI_1 nbits) %{
  match(Set dst (LShiftL src1 nbits));
  predicate(PreferLAoverADD);
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LA      $dst,#0($src1,$src1)\t # SLLG by 1 (long)" %}
  ins_encode %{ __ z_la($dst$$Register, 0, $src1$$Register, $src1$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Shift right arithmetic

// Register Arithmetic Shift Right
instruct sraI_reg_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (RShiftI dst src));
  effect(KILL cr); // R1 is killed, too.
  ins_cost(3 * DEFAULT_COST);
  size(12);
  format %{ "SRA     $dst,[$src] & 31" %}
  ins_encode %{
    __ z_lgr(Z_R1_scratch, $src$$Register);
    __ z_nill(Z_R1_scratch, BitsPerJavaInteger-1);
    __ z_sra($dst$$Register, 0, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Arithmetic Shift Right Immediate
// Constant shift count is masked in ideal graph already.
instruct sraI_reg_imm(iRegI dst, immI src, flagsReg cr) %{
  match(Set dst (RShiftI dst src));
  effect(KILL cr);
  size(4);
  format %{ "SRA     $dst,$src" %}
  ins_encode %{
    int Nbit = $src$$constant;
    assert((Nbit & (BitsPerJavaInteger - 1)) == Nbit, "Check shift mask in ideal graph");
    __ z_sra($dst$$Register, Nbit & (BitsPerJavaInteger - 1), Z_R0);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Arithmetic Shift Right Long
instruct sraL_reg_reg(iRegL dst, iRegL src1, iRegI src2, flagsReg cr) %{
  match(Set dst (RShiftL src1 src2));
  effect(KILL cr);
  size(6);
  format %{ "SRAG    $dst,$src1,[$src2]" %}
  opcode(SRAG_ZOPC);
  ins_encode(z_rsyform_reg_reg(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// Register Arithmetic Shift Right Long Immediate
instruct sraL_reg_imm(iRegL dst, iRegL src1, immI src2, flagsReg cr) %{
  match(Set dst (RShiftL src1 src2));
  effect(KILL cr);
  size(6);
  format %{ "SRAG    $dst,$src1,$src2" %}
  opcode(SRAG_ZOPC);
  ins_encode(z_rsyform_const(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

//  Shift right logical

// Register Shift Right
instruct srlI_reg_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (URShiftI dst src));
  effect(KILL cr); // R1 is killed, too.
  ins_cost(3 * DEFAULT_COST);
  size(12);
  format %{ "SRL     $dst,[$src] & 31" %}
  ins_encode %{
    __ z_lgr(Z_R1_scratch, $src$$Register);
    __ z_nill(Z_R1_scratch, BitsPerJavaInteger-1);
    __ z_srl($dst$$Register, 0, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Right Immediate
// Constant shift count is masked in ideal graph already.
instruct srlI_reg_imm(iRegI dst, immI src) %{
  match(Set dst (URShiftI dst src));
  size(4);
  format %{ "SRL     $dst,$src" %}
  ins_encode %{
    int Nbit = $src$$constant;
    assert((Nbit & (BitsPerJavaInteger - 1)) == Nbit, "Check shift mask in ideal graph");
    __ z_srl($dst$$Register, Nbit & (BitsPerJavaInteger - 1), Z_R0);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Right Long
instruct srlL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
  match(Set dst (URShiftL src1 src2));
  size(6);
  format %{ "SRLG    $dst,$src1,[$src2]" %}
  opcode(SRLG_ZOPC);
  ins_encode(z_rsyform_reg_reg(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Right Long Immediate
instruct srlL_reg_imm(iRegL dst, iRegL src1, immI src2) %{
  match(Set dst (URShiftL src1 src2));
  size(6);
  format %{ "SRLG    $dst,$src1,$src2" %}
  opcode(SRLG_ZOPC);
  ins_encode(z_rsyform_const(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Right Immediate with a CastP2X
instruct srlP_reg_imm(iRegL dst, iRegP_N2P src1, immI src2) %{
  match(Set dst (URShiftL (CastP2X src1) src2));
  size(6);
  format %{ "SRLG    $dst,$src1,$src2\t # Cast ptr $src1 to long and shift" %}
  opcode(SRLG_ZOPC);
  ins_encode(z_rsyform_const(dst, src1, src2));
  ins_pipe(pipe_class_dummy);
%}

//----------Rotate Instructions------------------------------------------------

// Rotate left 32bit.
instruct rotlI_reg_immI8(iRegI dst, iRegI src, immI8 lshift, immI8 rshift) %{
  match(Set dst (OrI (LShiftI src lshift) (URShiftI src rshift)));
  predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
  size(6);
  format %{ "RLL     $dst,$src,$lshift\t # ROTL32" %}
  opcode(RLL_ZOPC);
  ins_encode(z_rsyform_const(dst, src, lshift));
  ins_pipe(pipe_class_dummy);
%}

// Rotate left 64bit.
instruct rotlL_reg_immI8(iRegL dst, iRegL src, immI8 lshift, immI8 rshift) %{
  match(Set dst (OrL (LShiftL src lshift) (URShiftL src rshift)));
  predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
  size(6);
  format %{ "RLLG    $dst,$src,$lshift\t # ROTL64" %}
  opcode(RLLG_ZOPC);
  ins_encode(z_rsyform_const(dst, src, lshift));
  ins_pipe(pipe_class_dummy);
%}

// Rotate right 32bit.
instruct rotrI_reg_immI8(iRegI dst, iRegI src, immI8 rshift, immI8 lshift) %{
  match(Set dst (OrI (URShiftI src rshift) (LShiftI src lshift)));
  predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "RLL     $dst,$src,$rshift\t # ROTR32" %}
  opcode(RLL_ZOPC);
  ins_encode(z_rsyform_const(dst, src, rshift));
  ins_pipe(pipe_class_dummy);
%}

// Rotate right 64bit.
instruct rotrL_reg_immI8(iRegL dst, iRegL src, immI8 rshift, immI8 lshift) %{
  match(Set dst (OrL (URShiftL src rshift) (LShiftL src lshift)));
  predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "RLLG    $dst,$src,$rshift\t # ROTR64" %}
  opcode(RLLG_ZOPC);
  ins_encode(z_rsyform_const(dst, src, rshift));
  ins_pipe(pipe_class_dummy);
%}


//----------Overflow Math Instructions-----------------------------------------

instruct overflowAddI_reg_reg(flagsReg cr, iRegI op1, iRegI op2) %{
  match(Set cr (OverflowAddI op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AR      $op1,$op2\t # overflow check int" %}
  ins_encode %{
    __ z_lr(Z_R0_scratch, $op1$$Register);
    __ z_ar(Z_R0_scratch, $op2$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowAddI_reg_imm(flagsReg cr, iRegI op1, immI op2) %{
  match(Set cr (OverflowAddI op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "AR      $op1,$op2\t # overflow check int" %}
  ins_encode %{
    __ load_const_optimized(Z_R0_scratch, $op2$$constant);
    __ z_ar(Z_R0_scratch, $op1$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowAddL_reg_reg(flagsReg cr, iRegL op1, iRegL op2) %{
  match(Set cr (OverflowAddL op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "AGR     $op1,$op2\t # overflow check long" %}
  ins_encode %{
    __ z_lgr(Z_R0_scratch, $op1$$Register);
    __ z_agr(Z_R0_scratch, $op2$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowAddL_reg_imm(flagsReg cr, iRegL op1, immL op2) %{
  match(Set cr (OverflowAddL op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "AGR     $op1,$op2\t # overflow check long" %}
  ins_encode %{
    __ load_const_optimized(Z_R0_scratch, $op2$$constant);
    __ z_agr(Z_R0_scratch, $op1$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowSubI_reg_reg(flagsReg cr, iRegI op1, iRegI op2) %{
  match(Set cr (OverflowSubI op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "SR      $op1,$op2\t # overflow check int" %}
  ins_encode %{
    __ z_lr(Z_R0_scratch, $op1$$Register);
    __ z_sr(Z_R0_scratch, $op2$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowSubI_reg_imm(flagsReg cr, iRegI op1, immI op2) %{
  match(Set cr (OverflowSubI op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "SR      $op1,$op2\t # overflow check int" %}
  ins_encode %{
    __ load_const_optimized(Z_R1_scratch, $op2$$constant);
    __ z_lr(Z_R0_scratch, $op1$$Register);
    __ z_sr(Z_R0_scratch, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowSubL_reg_reg(flagsReg cr, iRegL op1, iRegL op2) %{
  match(Set cr (OverflowSubL op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "SGR     $op1,$op2\t # overflow check long" %}
  ins_encode %{
    __ z_lgr(Z_R0_scratch, $op1$$Register);
    __ z_sgr(Z_R0_scratch, $op2$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowSubL_reg_imm(flagsReg cr, iRegL op1, immL op2) %{
  match(Set cr (OverflowSubL op1 op2));
  effect(DEF cr, USE op1, USE op2);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "SGR     $op1,$op2\t # overflow check long" %}
  ins_encode %{
    __ load_const_optimized(Z_R1_scratch, $op2$$constant);
    __ z_lgr(Z_R0_scratch, $op1$$Register);
    __ z_sgr(Z_R0_scratch, Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowNegI_rReg(flagsReg cr, immI_0 zero, iRegI op2) %{
  match(Set cr (OverflowSubI zero op2));
  effect(DEF cr, USE op2);
  format %{ "NEG    $op2\t # overflow check int" %}
  ins_encode %{
    __ clear_reg(Z_R0_scratch, false, false);
    __ z_sr(Z_R0_scratch, $op2$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct overflowNegL_rReg(flagsReg cr, immL_0 zero, iRegL op2) %{
  match(Set cr (OverflowSubL zero op2));
  effect(DEF cr, USE op2);
  format %{ "NEGG    $op2\t # overflow check long" %}
  ins_encode %{
    __ clear_reg(Z_R0_scratch, true, false);
    __ z_sgr(Z_R0_scratch, $op2$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

// No intrinsics for multiplication, since there is no easy way
// to check for overflow.


//----------Floating Point Arithmetic Instructions-----------------------------

//  ADD

//  Add float single precision
instruct addF_reg_reg(regF dst, regF src, flagsReg cr) %{
  match(Set dst (AddF dst src));
  effect(KILL cr);
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "AEBR     $dst,$src" %}
  opcode(AEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addF_reg_mem(regF dst, memoryRX src, flagsReg cr)%{
  match(Set dst (AddF dst (LoadF src)));
  effect(KILL cr);
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "AEB      $dst,$src\t # floatMemory" %}
  opcode(AEB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Add float double precision
instruct addD_reg_reg(regD dst, regD src, flagsReg cr) %{
  match(Set dst (AddD dst src));
  effect(KILL cr);
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "ADBR     $dst,$src" %}
  opcode(ADBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct addD_reg_mem(regD dst, memoryRX src, flagsReg cr)%{
  match(Set dst (AddD dst (LoadD src)));
  effect(KILL cr);
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "ADB      $dst,$src\t # doubleMemory" %}
  opcode(ADB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// SUB

// Sub float single precision
instruct subF_reg_reg(regF dst, regF src, flagsReg cr) %{
  match(Set dst (SubF dst src));
  effect(KILL cr);
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "SEBR     $dst,$src" %}
  opcode(SEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct subF_reg_mem(regF dst, memoryRX src, flagsReg cr)%{
  match(Set dst (SubF dst (LoadF src)));
  effect(KILL cr);
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "SEB      $dst,$src\t # floatMemory" %}
  opcode(SEB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//  Sub float double precision
instruct subD_reg_reg(regD dst, regD src, flagsReg cr) %{
  match(Set dst (SubD dst src));
  effect(KILL cr);
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "SDBR     $dst,$src" %}
  opcode(SDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct subD_reg_mem(regD dst, memoryRX src, flagsReg cr)%{
  match(Set dst (SubD dst (LoadD src)));
  effect(KILL cr);
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "SDB      $dst,$src\t # doubleMemory" %}
  opcode(SDB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// MUL

// Mul float single precision
instruct mulF_reg_reg(regF dst, regF src) %{
  match(Set dst (MulF dst src));
  // CC unchanged by MUL.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "MEEBR    $dst,$src" %}
  opcode(MEEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct mulF_reg_mem(regF dst, memoryRX src)%{
  match(Set dst (MulF dst (LoadF src)));
  // CC unchanged by MUL.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MEEB     $dst,$src\t # floatMemory" %}
  opcode(MEEB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//  Mul float double precision
instruct mulD_reg_reg(regD dst, regD src) %{
  match(Set dst (MulD dst src));
  // CC unchanged by MUL.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "MDBR     $dst,$src" %}
  opcode(MDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct mulD_reg_mem(regD dst, memoryRX src)%{
  match(Set dst (MulD dst (LoadD src)));
  // CC unchanged by MUL.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MDB      $dst,$src\t # doubleMemory" %}
  opcode(MDB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Multiply-Accumulate
// src1 * src2 + dst
instruct maddF_reg_reg(regF dst, regF src1, regF src2) %{
  match(Set dst (FmaF dst (Binary src1 src2)));
  // CC unchanged by MUL-ADD.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "MAEBR    $dst, $src1, $src2" %}
  ins_encode %{
    __ z_maebr($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 + dst
instruct maddD_reg_reg(regD dst, regD src1, regD src2) %{
  match(Set dst (FmaD dst (Binary src1 src2)));
  // CC unchanged by MUL-ADD.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "MADBR    $dst, $src1, $src2" %}
  ins_encode %{
    __ z_madbr($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 - dst
instruct msubF_reg_reg(regF dst, regF src1, regF src2) %{
  match(Set dst (FmaF (NegF dst) (Binary src1 src2)));
  // CC unchanged by MUL-SUB.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "MSEBR    $dst, $src1, $src2" %}
  ins_encode %{
    __ z_msebr($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 - dst
instruct msubD_reg_reg(regD dst, regD src1, regD src2) %{
  match(Set dst (FmaD (NegD dst) (Binary src1 src2)));
  // CC unchanged by MUL-SUB.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "MSDBR    $dst, $src1, $src2" %}
  ins_encode %{
    __ z_msdbr($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 + dst
instruct maddF_reg_mem(regF dst, regF src1, memoryRX src2) %{
  match(Set dst (FmaF dst (Binary src1 (LoadF src2))));
  // CC unchanged by MUL-ADD.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MAEB     $dst, $src1, $src2" %}
  ins_encode %{
    __ z_maeb($dst$$FloatRegister, $src1$$FloatRegister,
              Address(reg_to_register_object($src2$$base), $src2$$index$$Register, $src2$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 + dst
instruct maddD_reg_mem(regD dst, regD src1, memoryRX src2) %{
  match(Set dst (FmaD dst (Binary src1 (LoadD src2))));
  // CC unchanged by MUL-ADD.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MADB     $dst, $src1, $src2" %}
  ins_encode %{
    __ z_madb($dst$$FloatRegister, $src1$$FloatRegister,
              Address(reg_to_register_object($src2$$base), $src2$$index$$Register, $src2$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 - dst
instruct msubF_reg_mem(regF dst, regF src1, memoryRX src2) %{
  match(Set dst (FmaF (NegF dst) (Binary src1 (LoadF src2))));
  // CC unchanged by MUL-SUB.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MSEB     $dst, $src1, $src2" %}
  ins_encode %{
    __ z_mseb($dst$$FloatRegister, $src1$$FloatRegister,
              Address(reg_to_register_object($src2$$base), $src2$$index$$Register, $src2$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 - dst
instruct msubD_reg_mem(regD dst, regD src1, memoryRX src2) %{
  match(Set dst (FmaD (NegD dst) (Binary src1 (LoadD src2))));
  // CC unchanged by MUL-SUB.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MSDB    $dst, $src1, $src2" %}
  ins_encode %{
    __ z_msdb($dst$$FloatRegister, $src1$$FloatRegister,
              Address(reg_to_register_object($src2$$base), $src2$$index$$Register, $src2$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 + dst
instruct maddF_mem_reg(regF dst, memoryRX src1, regF src2) %{
  match(Set dst (FmaF dst (Binary (LoadF src1) src2)));
  // CC unchanged by MUL-ADD.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MAEB     $dst, $src1, $src2" %}
  ins_encode %{
    __ z_maeb($dst$$FloatRegister, $src2$$FloatRegister,
              Address(reg_to_register_object($src1$$base), $src1$$index$$Register, $src1$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 + dst
instruct maddD_mem_reg(regD dst, memoryRX src1, regD src2) %{
  match(Set dst (FmaD dst (Binary (LoadD src1) src2)));
  // CC unchanged by MUL-ADD.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MADB     $dst, $src1, $src2" %}
  ins_encode %{
    __ z_madb($dst$$FloatRegister, $src2$$FloatRegister,
              Address(reg_to_register_object($src1$$base), $src1$$index$$Register, $src1$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 - dst
instruct msubF_mem_reg(regF dst, memoryRX src1, regF src2) %{
  match(Set dst (FmaF (NegF dst) (Binary (LoadF src1) src2)));
  // CC unchanged by MUL-SUB.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MSEB     $dst, $src1, $src2" %}
  ins_encode %{
    __ z_mseb($dst$$FloatRegister, $src2$$FloatRegister,
              Address(reg_to_register_object($src1$$base), $src1$$index$$Register, $src1$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

// src1 * src2 - dst
instruct msubD_mem_reg(regD dst, memoryRX src1, regD src2) %{
  match(Set dst (FmaD (NegD dst) (Binary (LoadD src1) src2)));
  // CC unchanged by MUL-SUB.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "MSDB    $dst, $src1, $src2" %}
  ins_encode %{
    __ z_msdb($dst$$FloatRegister, $src2$$FloatRegister,
              Address(reg_to_register_object($src1$$base), $src1$$index$$Register, $src1$$disp));
  %}
  ins_pipe(pipe_class_dummy);
%}

//  DIV

//  Div float single precision
instruct divF_reg_reg(regF dst, regF src) %{
  match(Set dst (DivF dst src));
  // CC unchanged by DIV.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "DEBR     $dst,$src" %}
  opcode(DEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct divF_reg_mem(regF dst, memoryRX src)%{
  match(Set dst (DivF dst (LoadF src)));
  // CC unchanged by DIV.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "DEB      $dst,$src\t # floatMemory" %}
  opcode(DEB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//  Div float double precision
instruct divD_reg_reg(regD dst, regD src) %{
  match(Set dst (DivD dst src));
  // CC unchanged by DIV.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "DDBR     $dst,$src" %}
  opcode(DDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct divD_reg_mem(regD dst, memoryRX src)%{
  match(Set dst (DivD dst (LoadD src)));
  // CC unchanged by DIV.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "DDB      $dst,$src\t # doubleMemory" %}
  opcode(DDB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// ABS

// Absolute float single precision
instruct absF_reg(regF dst, regF src, flagsReg cr) %{
  match(Set dst (AbsF src));
  effect(KILL cr);
  size(4);
  format %{ "LPEBR    $dst,$src\t float" %}
  opcode(LPEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Absolute float double precision
instruct absD_reg(regD dst, regD src, flagsReg cr) %{
  match(Set dst (AbsD src));
  effect(KILL cr);
  size(4);
  format %{ "LPDBR    $dst,$src\t double" %}
  opcode(LPDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//  NEG(ABS)

// Negative absolute float single precision
instruct nabsF_reg(regF dst, regF src, flagsReg cr) %{
  match(Set dst (NegF (AbsF src)));
  effect(KILL cr);
  size(4);
  format %{ "LNEBR    $dst,$src\t float" %}
  opcode(LNEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Negative absolute float double precision
instruct nabsD_reg(regD dst, regD src, flagsReg cr) %{
  match(Set dst (NegD (AbsD src)));
  effect(KILL cr);
  size(4);
  format %{ "LNDBR    $dst,$src\t double" %}
  opcode(LNDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// NEG

instruct negF_reg(regF dst, regF src, flagsReg cr) %{
  match(Set dst (NegF src));
  effect(KILL cr);
  size(4);
  format %{ "NegF     $dst,$src\t float" %}
  ins_encode %{ __ z_lcebr($dst$$FloatRegister, $src$$FloatRegister); %}
  ins_pipe(pipe_class_dummy);
%}

instruct negD_reg(regD dst, regD src, flagsReg cr) %{
  match(Set dst (NegD src));
  effect(KILL cr);
  size(4);
  format %{ "NegD     $dst,$src\t double" %}
  ins_encode %{ __ z_lcdbr($dst$$FloatRegister, $src$$FloatRegister); %}
  ins_pipe(pipe_class_dummy);
%}

// SQRT

// Sqrt float precision
instruct sqrtF_reg(regF dst, regF src) %{
  match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
  // CC remains unchanged.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "SQEBR    $dst,$src" %}
  opcode(SQEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Sqrt double precision
instruct sqrtD_reg(regD dst, regD src) %{
  match(Set dst (SqrtD src));
  // CC remains unchanged.
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "SQDBR    $dst,$src" %}
  opcode(SQDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct sqrtF_mem(regF dst, memoryRX src) %{
  match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
  // CC remains unchanged.
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "SQEB     $dst,$src\t # floatMemory" %}
  opcode(SQEB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct sqrtD_mem(regD dst, memoryRX src) %{
  match(Set dst (SqrtD src));
  // CC remains unchanged.
  ins_cost(ALU_MEMORY_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "SQDB     $dst,$src\t # doubleMemory" %}
  opcode(SQDB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//----------Logical Instructions-----------------------------------------------

// Register And
instruct andI_reg_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (AndI dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  size(2);
  format %{ "NR      $dst,$src\t # int" %}
  opcode(NR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct andI_Reg_mem(iRegI dst, memory src, flagsReg cr)%{
  match(Set dst (AndI dst (LoadI src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "N(Y)    $dst, $src\t # int" %}
  opcode(NY_ZOPC, N_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate And
instruct andI_reg_uimm32(iRegI dst, uimmI src, flagsReg cr) %{
  match(Set dst (AndI dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  size(6);
  format %{ "NILF    $dst,$src" %}
  opcode(NILF_ZOPC);
  ins_encode(z_rilform_unsigned(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct andI_reg_uimmI_LH1(iRegI dst, uimmI_LH1 src, flagsReg cr) %{
  match(Set dst (AndI dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NILH    $dst,$src" %}
  ins_encode %{ __ z_nilh($dst$$Register, ($src$$constant >> 16) & 0xFFFF); %}
  ins_pipe(pipe_class_dummy);
%}

instruct andI_reg_uimmI_LL1(iRegI dst, uimmI_LL1 src, flagsReg cr) %{
  match(Set dst (AndI dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NILL    $dst,$src" %}
  ins_encode %{ __ z_nill($dst$$Register, $src$$constant & 0xFFFF); %}
  ins_pipe(pipe_class_dummy);
%}

// Register And Long
instruct andL_reg_reg(iRegL dst, iRegL src, flagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NGR     $dst,$src\t # long" %}
  opcode(NGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct andL_Reg_mem(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (AndL dst (LoadL src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "NG      $dst, $src\t # long" %}
  opcode(NG_ZOPC, NG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct andL_reg_uimmL_LL1(iRegL dst, uimmL_LL1 src, flagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NILL    $dst,$src\t # long" %}
  ins_encode %{ __ z_nill($dst$$Register, $src$$constant & 0xFFFF); %}
  ins_pipe(pipe_class_dummy);
%}

instruct andL_reg_uimmL_LH1(iRegL dst, uimmL_LH1 src, flagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NILH    $dst,$src\t # long" %}
  ins_encode %{ __ z_nilh($dst$$Register, ($src$$constant >> 16) & 0xFFFF); %}
  ins_pipe(pipe_class_dummy);
%}

instruct andL_reg_uimmL_HL1(iRegL dst, uimmL_HL1 src, flagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NIHL    $dst,$src\t # long" %}
  ins_encode %{ __ z_nihl($dst$$Register, ($src$$constant >> 32) & 0xFFFF); %}
  ins_pipe(pipe_class_dummy);
%}

instruct andL_reg_uimmL_HH1(iRegL dst, uimmL_HH1 src, flagsReg cr) %{
  match(Set dst (AndL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "NIHH    $dst,$src\t # long" %}
  ins_encode %{ __ z_nihh($dst$$Register, ($src$$constant >> 48) & 0xFFFF); %}
  ins_pipe(pipe_class_dummy);
%}

//  OR

// Or Instructions
// Register Or
instruct orI_reg_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (OrI dst src));
  effect(KILL cr);
  size(2);
  format %{ "OR      $dst,$src" %}
  opcode(OR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct orI_Reg_mem(iRegI dst, memory src, flagsReg cr)%{
  match(Set dst (OrI dst (LoadI src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "O(Y)    $dst, $src\t # int" %}
  opcode(OY_ZOPC, O_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Or
instruct orI_reg_uimm16(iRegI dst, uimmI16 con, flagsReg cr) %{
  match(Set dst (OrI dst con));
  effect(KILL cr);
  size(4);
  format %{ "OILL    $dst,$con" %}
  opcode(OILL_ZOPC);
  ins_encode(z_riform_unsigned(dst,con));
  ins_pipe(pipe_class_dummy);
%}

instruct orI_reg_uimm32(iRegI dst, uimmI con, flagsReg cr) %{
  match(Set dst (OrI dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  size(6);
  format %{ "OILF    $dst,$con" %}
  opcode(OILF_ZOPC);
  ins_encode(z_rilform_unsigned(dst,con));
  ins_pipe(pipe_class_dummy);
%}

// Register Or Long
instruct orL_reg_reg(iRegL dst, iRegL src, flagsReg cr) %{
  match(Set dst (OrL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "OGR      $dst,$src\t # long" %}
  opcode(OGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct orL_Reg_mem(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (OrL dst (LoadL src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "OG      $dst, $src\t # long" %}
  opcode(OG_ZOPC, OG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Or long
instruct orL_reg_uimm16(iRegL dst, uimmL16 con, flagsReg cr) %{
  match(Set dst (OrL dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "OILL    $dst,$con\t # long" %}
  opcode(OILL_ZOPC);
  ins_encode(z_riform_unsigned(dst,con));
  ins_pipe(pipe_class_dummy);
%}

instruct orL_reg_uimm32(iRegI dst, uimmL32 con, flagsReg cr) %{
  match(Set dst (OrI dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "OILF    $dst,$con\t # long" %}
  opcode(OILF_ZOPC);
  ins_encode(z_rilform_unsigned(dst,con));
  ins_pipe(pipe_class_dummy);
%}

// XOR

// Register Xor
instruct xorI_reg_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (XorI dst src));
  effect(KILL cr);
  size(2);
  format %{ "XR      $dst,$src" %}
  opcode(XR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct xorI_Reg_mem(iRegI dst, memory src, flagsReg cr)%{
  match(Set dst (XorI dst (LoadI src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "X(Y)    $dst, $src\t # int" %}
  opcode(XY_ZOPC, X_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Xor
instruct xorI_reg_uimm32(iRegI dst, uimmI src, flagsReg cr) %{
  match(Set dst (XorI dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  size(6);
  format %{ "XILF    $dst,$src" %}
  opcode(XILF_ZOPC);
  ins_encode(z_rilform_unsigned(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Register Xor Long
instruct xorL_reg_reg(iRegL dst, iRegL src, flagsReg cr) %{
  match(Set dst (XorL dst src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "XGR     $dst,$src\t # long" %}
  opcode(XGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct xorL_Reg_mem(iRegL dst, memory src, flagsReg cr)%{
  match(Set dst (XorL dst (LoadL src)));
  effect(KILL cr);
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "XG      $dst, $src\t # long" %}
  opcode(XG_ZOPC, XG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Immediate Xor Long
instruct xorL_reg_uimm32(iRegL dst, uimmL32 con, flagsReg cr) %{
  match(Set dst (XorL dst con));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_HIGH);
  size(6);
  format %{ "XILF    $dst,$con\t # long" %}
  opcode(XILF_ZOPC);
  ins_encode(z_rilform_unsigned(dst,con));
  ins_pipe(pipe_class_dummy);
%}

//----------Convert to Boolean-------------------------------------------------

// Convert integer to boolean.
instruct convI2B(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (Conv2B src));
  effect(KILL cr);
  ins_cost(3 * DEFAULT_COST);
  size(6);
  format %{ "convI2B $dst,$src" %}
  ins_encode %{
    __ z_lnr($dst$$Register, $src$$Register);  // Rdst := -|Rsrc|, i.e. Rdst == 0 <=> Rsrc == 0
    __ z_srl($dst$$Register, 31);              // Rdst := sign(Rdest)
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct convP2B(iRegI dst, iRegP_N2P src, flagsReg cr) %{
  match(Set dst (Conv2B src));
  effect(KILL cr);
  ins_cost(3 * DEFAULT_COST);
  size(10);
  format %{ "convP2B $dst,$src" %}
  ins_encode %{
    __ z_lngr($dst$$Register, $src$$Register);     // Rdst := -|Rsrc| i.e. Rdst == 0 <=> Rsrc == 0
    __ z_srlg($dst$$Register, $dst$$Register, 63); // Rdst := sign(Rdest)
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmpLTMask_reg_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (CmpLTMask dst src));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST);
  size(18);
  format %{ "Set $dst CmpLTMask $dst,$src" %}
  ins_encode %{
    // Avoid signed 32 bit overflow: Do sign extend and sub 64 bit.
    __ z_lgfr(Z_R0_scratch, $src$$Register);
    __ z_lgfr($dst$$Register, $dst$$Register);
    __ z_sgr($dst$$Register, Z_R0_scratch);
    __ z_srag($dst$$Register, $dst$$Register, 63);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmpLTMask_reg_zero(iRegI dst, immI_0 zero, flagsReg cr) %{
  match(Set dst (CmpLTMask dst zero));
  effect(KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "Set $dst CmpLTMask $dst,$zero" %}
  ins_encode %{ __ z_sra($dst$$Register, 31); %}
  ins_pipe(pipe_class_dummy);
%}


//----------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));
  // CC remains unchanged.
  size(4);
  format %{ "LEDBR   $dst,$src" %}
  opcode(LEDBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct convF2I_reg(iRegI dst, regF src, flagsReg cr) %{
  match(Set dst (ConvF2I src));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size(16);
  format %{ "convF2I  $dst,$src" %}
  ins_encode %{
    Label done;
    __ clear_reg($dst$$Register, false, false);  // Initialize with result for unordered: 0.
    __ z_cebr($src$$FloatRegister, $src$$FloatRegister);   // Round.
    __ z_brno(done);                             // Result is zero if unordered argument.
    __ z_cfebr($dst$$Register, $src$$FloatRegister, Assembler::to_zero);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct convD2I_reg(iRegI dst, regD src, flagsReg cr) %{
  match(Set dst (ConvD2I src));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size(16);
  format %{ "convD2I  $dst,$src" %}
  ins_encode %{
    Label done;
    __ clear_reg($dst$$Register, false, false);  // Initialize with result for unordered: 0.
    __ z_cdbr($src$$FloatRegister, $src$$FloatRegister);   // Round.
    __ z_brno(done);                             // Result is zero if unordered argument.
    __ z_cfdbr($dst$$Register, $src$$FloatRegister, Assembler::to_zero);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct convF2L_reg(iRegL dst, regF src, flagsReg cr) %{
  match(Set dst (ConvF2L src));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size(16);
  format %{ "convF2L  $dst,$src" %}
  ins_encode %{
    Label done;
    __ clear_reg($dst$$Register, true, false);  // Initialize with result for unordered: 0.
    __ z_cebr($src$$FloatRegister, $src$$FloatRegister);   // Round.
    __ z_brno(done);                             // Result is zero if unordered argument.
    __ z_cgebr($dst$$Register, $src$$FloatRegister, Assembler::to_zero);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct convD2L_reg(iRegL dst, regD src, flagsReg cr) %{
  match(Set dst (ConvD2L src));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  size(16);
  format %{ "convD2L  $dst,$src" %}
  ins_encode %{
    Label done;
    __ clear_reg($dst$$Register, true, false);  // Initialize with result for unordered: 0.
    __ z_cdbr($src$$FloatRegister, $src$$FloatRegister);   // Round.
    __ z_brno(done);                             // Result is zero if unordered argument.
    __ z_cgdbr($dst$$Register, $src$$FloatRegister, Assembler::to_zero);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct convF2D_reg(regD dst, regF src) %{
  match(Set dst (ConvF2D src));
  // CC remains unchanged.
  size(4);
  format %{ "LDEBR   $dst,$src" %}
  opcode(LDEBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct convF2D_mem(regD dst, memoryRX src) %{
  match(Set dst (ConvF2D src));
  // CC remains unchanged.
  size(6);
  format %{ "LDEB    $dst,$src" %}
  opcode(LDEB_ZOPC);
  ins_encode(z_form_rt_memFP(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct convI2D_reg(regD dst, iRegI src) %{
  match(Set dst (ConvI2D src));
  // CC remains unchanged.
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "CDFBR   $dst,$src" %}
  opcode(CDFBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Optimization that saves up to two memory operations for each conversion.
instruct convI2F_ireg(regF dst, iRegI src) %{
  match(Set dst (ConvI2F src));
  // CC remains unchanged.
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "CEFBR   $dst,$src\t # convert int to float" %}
  opcode(CEFBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct convI2L_reg(iRegL dst, iRegI src) %{
  match(Set dst (ConvI2L src));
  size(4);
  format %{ "LGFR    $dst,$src\t # int->long" %}
  opcode(LGFR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// 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 %{ "LLGFR   $dst, $src \t # zero-extend int to long" %}
  ins_encode %{ __ z_llgfr($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Zero-extend convert int to long.
instruct convI2L_mem_zex(iRegL dst, memory src, immL_32bits mask) %{
  match(Set dst (AndL (ConvI2L (LoadI src)) mask));
  // Uses load_const_optmized, so size can vary.
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "LLGF    $dst, $src \t # zero-extend int to long" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Zero-extend long
instruct zeroExtend_long(iRegL dst, iRegL src, immL_32bits mask) %{
  match(Set dst (AndL src mask));
  size(4);
  format %{ "LLGFR   $dst, $src \t # zero-extend long to long" %}
  ins_encode %{ __ z_llgfr($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct rShiftI16_lShiftI16_reg(iRegI dst, iRegI src, immI_16 amount) %{
  match(Set dst (RShiftI (LShiftI src amount) amount));
  size(4);
  format %{ "LHR     $dst,$src\t short->int" %}
  opcode(LHR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct rShiftI24_lShiftI24_reg(iRegI dst, iRegI src, immI_24 amount) %{
  match(Set dst (RShiftI (LShiftI src amount) amount));
  size(4);
  format %{ "LBR     $dst,$src\t byte->int" %}
  opcode(LBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct MoveF2I_stack_reg(iRegI dst, stackSlotF src) %{
  match(Set dst (MoveF2I src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "L       $dst,$src\t # MoveF2I" %}
  opcode(L_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// javax.imageio.stream.ImageInputStreamImpl.toFloats([B[FII)
instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
  match(Set dst (MoveI2F src));
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LE      $dst,$src\t # MoveI2F" %}
  opcode(LE_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct MoveD2L_stack_reg(iRegL dst, stackSlotD src) %{
  match(Set dst (MoveD2L src));
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "LG      $src,$dst\t # MoveD2L" %}
  opcode(LG_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
  match(Set dst (MoveL2D src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "LD      $dst,$src\t # MoveL2D" %}
  opcode(LD_ZOPC);
  ins_encode(z_form_rt_mem(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{
  match(Set dst (MoveI2F src));
  ins_cost(MEMORY_REF_COST);
  size(4);
  format %{ "ST      $src,$dst\t # MoveI2F" %}
  opcode(ST_ZOPC);
  ins_encode(z_form_rt_mem(src, dst));
  ins_pipe(pipe_class_dummy);
%}

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 %{ "STD     $src,$dst\t # MoveD2L" %}
  opcode(STD_ZOPC);
  ins_encode(z_form_rt_mem(src,dst));
  ins_pipe(pipe_class_dummy);
%}

instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{
  match(Set dst (MoveL2D src));
  ins_cost(MEMORY_REF_COST);
  size(6);
  format %{ "STG     $src,$dst\t # MoveL2D" %}
  opcode(STG_ZOPC);
  ins_encode(z_form_rt_mem(src,dst));
  ins_pipe(pipe_class_dummy);
%}

instruct convL2F_reg(regF dst, iRegL src) %{
  match(Set dst (ConvL2F src));
  // CC remains unchanged.
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "CEGBR   $dst,$src" %}
  opcode(CEGBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct convL2D_reg(regD dst, iRegL src) %{
  match(Set dst (ConvL2D src));
  // CC remains unchanged.
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "CDGBR   $dst,$src" %}
  opcode(CDGBR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct convL2I_reg(iRegI dst, iRegL src) %{
  match(Set dst (ConvL2I src));
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "LR      $dst,$src\t # long->int (if needed)" %}
  ins_encode %{ __ lr_if_needed($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Register Shift Right Immediate
instruct shrL_reg_imm6_L2I(iRegI dst, iRegL src, immI_32_63 cnt, flagsReg cr) %{
  match(Set dst (ConvL2I (RShiftL src cnt)));
  effect(KILL cr);
  size(6);
  format %{ "SRAG    $dst,$src,$cnt" %}
  opcode(SRAG_ZOPC);
  ins_encode(z_rsyform_const(dst, src, cnt));
  ins_pipe(pipe_class_dummy);
%}

//----------TRAP based zero checks and range checks----------------------------

// SIGTRAP based implicit range checks in compiled code.
// A range check in the ideal world has one of the following shapes:
//   - (If le (CmpU length index)), (IfTrue  throw exception)
//   - (If lt (CmpU index length)), (IfFalse throw exception)
//
// Match range check 'If le (CmpU length index)'
instruct rangeCheck_iReg_uimmI16(cmpOpT cmp, iRegI length, uimmI16 index, label labl) %{
  match(If cmp (CmpU length index));
  effect(USE labl);
  predicate(TrapBasedRangeChecks &&
            _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le &&
            PROB_UNLIKELY(_leaf->as_If ()->_prob) >= PROB_ALWAYS &&
            Matcher::branches_to_uncommon_trap(_leaf));
  ins_cost(1);
  // TODO: s390 port size(FIXED_SIZE);

  ins_is_TrapBasedCheckNode(true);

  format %{ "RangeCheck len=$length cmp=$cmp idx=$index => trap $labl" %}
  ins_encode %{ __ z_clfit($length$$Register, $index$$constant, $cmp$$cmpcode); %}
  ins_pipe(pipe_class_trap);
%}

// Match range check 'If lt (CmpU index length)'
instruct rangeCheck_iReg_iReg(cmpOpT cmp, iRegI index, iRegI length, label labl, flagsReg cr) %{
  match(If cmp (CmpU index length));
  effect(USE labl, KILL cr);
  predicate(TrapBasedRangeChecks &&
            _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt &&
            _leaf->as_If ()->_prob >= PROB_ALWAYS &&
            Matcher::branches_to_uncommon_trap(_leaf));
  ins_cost(1);
  // TODO: s390 port size(FIXED_SIZE);

  ins_is_TrapBasedCheckNode(true);

  format %{ "RangeCheck idx=$index cmp=$cmp len=$length => trap $labl" %}
  ins_encode %{ __ z_clrt($index$$Register, $length$$Register, $cmp$$cmpcode); %}
  ins_pipe(pipe_class_trap);
%}

// Match range check 'If lt (CmpU index length)'
instruct rangeCheck_uimmI16_iReg(cmpOpT cmp, iRegI index, uimmI16 length, label labl) %{
  match(If cmp (CmpU index length));
  effect(USE labl);
  predicate(TrapBasedRangeChecks &&
            _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt &&
            _leaf->as_If ()->_prob >= PROB_ALWAYS &&
            Matcher::branches_to_uncommon_trap(_leaf));
  ins_cost(1);
  // TODO: s390 port size(FIXED_SIZE);

  ins_is_TrapBasedCheckNode(true);

  format %{ "RangeCheck idx=$index cmp=$cmp len= $length => trap $labl" %}
  ins_encode %{ __ z_clfit($index$$Register, $length$$constant, $cmp$$cmpcode); %}
  ins_pipe(pipe_class_trap);
%}

// Implicit zero checks (more implicit null checks).
instruct zeroCheckP_iReg_imm0(cmpOpT cmp, iRegP_N2P value, immP0 zero, label labl) %{
  match(If cmp (CmpP value zero));
  effect(USE labl);
  predicate(TrapBasedNullChecks &&
            _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne &&
            _leaf->as_If ()->_prob >= PROB_LIKELY_MAG(4) &&
            Matcher::branches_to_uncommon_trap(_leaf));
  size(6);

  ins_is_TrapBasedCheckNode(true);

  format %{ "ZeroCheckP value=$value cmp=$cmp zero=$zero => trap $labl" %}
  ins_encode %{ __ z_cgit($value$$Register, 0, $cmp$$cmpcode); %}
  ins_pipe(pipe_class_trap);
%}

// Implicit zero checks (more implicit null checks).
instruct zeroCheckN_iReg_imm0(cmpOpT cmp, iRegN_P2N value, immN0 zero, label labl) %{
  match(If cmp (CmpN value zero));
  effect(USE labl);
  predicate(TrapBasedNullChecks &&
            _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne &&
            _leaf->as_If ()->_prob >= PROB_LIKELY_MAG(4) &&
            Matcher::branches_to_uncommon_trap(_leaf));
  size(6);

  ins_is_TrapBasedCheckNode(true);

  format %{ "ZeroCheckN value=$value cmp=$cmp zero=$zero => trap $labl" %}
  ins_encode %{ __ z_cit($value$$Register, 0, $cmp$$cmpcode); %}
  ins_pipe(pipe_class_trap);
%}

//----------Compare instructions-----------------------------------------------

// INT signed

// Compare Integers
instruct compI_reg_reg(flagsReg cr, iRegI op1, iRegI op2) %{
  match(Set cr (CmpI op1 op2));
  size(2);
  format %{ "CR      $op1,$op2" %}
  opcode(CR_ZOPC);
  ins_encode(z_rrform(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compI_reg_imm(flagsReg cr, iRegI op1, immI op2) %{
  match(Set cr (CmpI op1 op2));
  size(6);
  format %{ "CFI     $op1,$op2" %}
  opcode(CFI_ZOPC);
  ins_encode(z_rilform_signed(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compI_reg_imm16(flagsReg cr, iRegI op1, immI16 op2) %{
  match(Set cr (CmpI op1 op2));
  size(4);
  format %{ "CHI     $op1,$op2" %}
  opcode(CHI_ZOPC);
  ins_encode(z_riform_signed(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compI_reg_imm0(flagsReg cr, iRegI op1, immI_0 zero) %{
  match(Set cr (CmpI op1 zero));
  ins_cost(DEFAULT_COST_LOW);
  size(2);
  format %{ "LTR     $op1,$op1" %}
  opcode(LTR_ZOPC);
  ins_encode(z_rrform(op1, op1));
  ins_pipe(pipe_class_dummy);
%}

instruct compI_reg_mem(flagsReg cr, iRegI op1, memory op2)%{
  match(Set cr (CmpI op1 (LoadI op2)));
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "C(Y)    $op1, $op2\t # int" %}
  opcode(CY_ZOPC, C_ZOPC);
  ins_encode(z_form_rt_mem_opt(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

// INT unsigned

instruct compU_reg_reg(flagsReg cr, iRegI op1, iRegI op2) %{
  match(Set cr (CmpU op1 op2));
  size(2);
  format %{ "CLR     $op1,$op2\t # unsigned" %}
  opcode(CLR_ZOPC);
  ins_encode(z_rrform(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compU_reg_uimm(flagsReg cr, iRegI op1, uimmI op2) %{
  match(Set cr (CmpU op1 op2));
  size(6);
  format %{ "CLFI    $op1,$op2\t # unsigned" %}
  opcode(CLFI_ZOPC);
  ins_encode(z_rilform_unsigned(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compU_reg_mem(flagsReg cr, iRegI op1, memory op2)%{
  match(Set cr (CmpU op1 (LoadI op2)));
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CL(Y)   $op1, $op2\t # unsigned" %}
  opcode(CLY_ZOPC, CL_ZOPC);
  ins_encode(z_form_rt_mem_opt(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

// LONG signed

instruct compL_reg_reg(flagsReg cr, iRegL op1, iRegL op2) %{
  match(Set cr (CmpL op1 op2));
  size(4);
  format %{ "CGR     $op1,$op2\t # long" %}
  opcode(CGR_ZOPC);
  ins_encode(z_rreform(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_reg_regI(flagsReg cr, iRegL op1, iRegI op2) %{
  match(Set cr (CmpL op1 (ConvI2L op2)));
  size(4);
  format %{ "CGFR    $op1,$op2\t # long/int" %}
  opcode(CGFR_ZOPC);
  ins_encode(z_rreform(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_reg_imm32(flagsReg cr, iRegL op1, immL32 con) %{
  match(Set cr (CmpL op1 con));
  size(6);
  format %{ "CGFI    $op1,$con" %}
  opcode(CGFI_ZOPC);
  ins_encode(z_rilform_signed(op1, con));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_reg_imm16(flagsReg cr, iRegL op1, immL16 con) %{
  match(Set cr (CmpL op1 con));
  size(4);
  format %{ "CGHI    $op1,$con" %}
  opcode(CGHI_ZOPC);
  ins_encode(z_riform_signed(op1, con));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_reg_imm0(flagsReg cr, iRegL op1, immL_0 con) %{
  match(Set cr (CmpL op1 con));
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LTGR    $op1,$op1" %}
  opcode(LTGR_ZOPC);
  ins_encode(z_rreform(op1, op1));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_conv_reg_imm0(flagsReg cr, iRegI op1, immL_0 con) %{
  match(Set cr (CmpL (ConvI2L op1) con));
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LTGFR    $op1,$op1" %}
  opcode(LTGFR_ZOPC);
  ins_encode(z_rreform(op1, op1));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_reg_mem(iRegL dst, memory src, flagsReg cr)%{
  match(Set cr (CmpL dst (LoadL src)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "CG      $dst, $src\t # long" %}
  opcode(CG_ZOPC, CG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct compL_reg_memI(iRegL dst, memory src, flagsReg cr)%{
  match(Set cr (CmpL dst (ConvI2L (LoadI src))));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "CGF     $dst, $src\t # long/int" %}
  opcode(CGF_ZOPC, CGF_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//  LONG unsigned
// Added CmpUL for LoopPredicate.
instruct compUL_reg_reg(flagsReg cr, iRegL op1, iRegL op2) %{
  match(Set cr (CmpUL op1 op2));
  size(4);
  format %{ "CLGR    $op1,$op2\t # long" %}
  opcode(CLGR_ZOPC);
  ins_encode(z_rreform(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compUL_reg_imm32(flagsReg cr, iRegL op1, uimmL32 con) %{
  match(Set cr (CmpUL op1 con));
  size(6);
  format %{ "CLGFI   $op1,$con" %}
  opcode(CLGFI_ZOPC);
  ins_encode(z_rilform_unsigned(op1, con));
  ins_pipe(pipe_class_dummy);
%}

//  PTR unsigned

instruct compP_reg_reg(flagsReg cr, iRegP_N2P op1, iRegP_N2P op2) %{
  match(Set cr (CmpP op1 op2));
  size(4);
  format %{ "CLGR    $op1,$op2\t # ptr" %}
  opcode(CLGR_ZOPC);
  ins_encode(z_rreform(op1, op2));
  ins_pipe(pipe_class_dummy);
%}

instruct compP_reg_imm0(flagsReg cr, iRegP_N2P op1, immP0 op2) %{
  match(Set cr (CmpP op1 op2));
  ins_cost(DEFAULT_COST_LOW);
  size(4);
  format %{ "LTGR    $op1, $op1\t # ptr" %}
  opcode(LTGR_ZOPC);
  ins_encode(z_rreform(op1, op1));
  ins_pipe(pipe_class_dummy);
%}

// Don't use LTGFR which performs sign extend.
instruct compP_decode_reg_imm0(flagsReg cr, iRegN op1, immP0 op2) %{
  match(Set cr (CmpP (DecodeN op1) op2));
  predicate(CompressedOops::base() == NULL && CompressedOops::shift() == 0);
  ins_cost(DEFAULT_COST_LOW);
  size(2);
  format %{ "LTR    $op1, $op1\t # ptr" %}
  opcode(LTR_ZOPC);
  ins_encode(z_rrform(op1, op1));
  ins_pipe(pipe_class_dummy);
%}

instruct compP_reg_mem(iRegP dst, memory src, flagsReg cr)%{
  match(Set cr (CmpP dst (LoadP src)));
  ins_cost(MEMORY_REF_COST);
  size(Z_DISP3_SIZE);
  format %{ "CLG     $dst, $src\t # ptr" %}
  opcode(CLG_ZOPC, CLG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//----------Max and Min--------------------------------------------------------

// Max Register with Register
instruct z196_minI_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_LoadStoreConditional());
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI (z196 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    Register Rsrc2 = $src2$$Register;

    if (Rsrc1 == Rsrc2) {
      if (Rdst != Rsrc1) {
        __ z_lgfr(Rdst, Rsrc1);
      }
    } else if (Rdst == Rsrc1) {   // Rdst preset with src1.
      __ z_cr(Rsrc1, Rsrc2);      // Move src2 only if src1 is NotLow.
      __ z_locr(Rdst, Rsrc2, Assembler::bcondNotLow);
    } else if (Rdst == Rsrc2) {   // Rdst preset with src2.
      __ z_cr(Rsrc2, Rsrc1);      // Move src1 only if src2 is NotLow.
      __ z_locr(Rdst, Rsrc1, Assembler::bcondNotLow);
    } else {
      // Rdst is disjoint from operands, move in either case.
      __ z_cr(Rsrc1, Rsrc2);
      __ z_locr(Rdst, Rsrc2, Assembler::bcondNotLow);
      __ z_locr(Rdst, Rsrc1, Assembler::bcondLow);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Min Register with Register.
instruct z10_minI_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI (z10 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    Register Rsrc2 = $src2$$Register;
    Label done;

    if (Rsrc1 == Rsrc2) {
      if (Rdst != Rsrc1) {
        __ z_lgfr(Rdst, Rsrc1);
      }
    } else if (Rdst == Rsrc1) {
      __ z_crj(Rsrc1, Rsrc2, Assembler::bcondLow, done);
      __ z_lgfr(Rdst, Rsrc2);
    } else if (Rdst == Rsrc2) {
      __ z_crj(Rsrc2, Rsrc1, Assembler::bcondLow, done);
      __ z_lgfr(Rdst, Rsrc1);
    } else {
      __ z_lgfr(Rdst, Rsrc1);
      __ z_crj(Rsrc1, Rsrc2, Assembler::bcondLow, done);
      __ z_lgfr(Rdst, Rsrc2);
    }
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct minI_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  predicate(!VM_Version::has_CompareBranch());
  ins_cost(3 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    Register Rsrc2 = $src2$$Register;
    Label done;

    if (Rsrc1 == Rsrc2) {
      if (Rdst != Rsrc1) {
        __ z_lgfr(Rdst, Rsrc1);
      }
    } else if (Rdst == Rsrc1) {
      __ z_cr(Rsrc1, Rsrc2);
      __ z_brl(done);
      __ z_lgfr(Rdst, Rsrc2);
    } else if (Rdst == Rsrc2) {
      __ z_cr(Rsrc2, Rsrc1);
      __ z_brl(done);
      __ z_lgfr(Rdst, Rsrc1);
    } else {
      __ z_lgfr(Rdst, Rsrc1);
      __ z_cr(Rsrc1, Rsrc2);
      __ z_brl(done);
      __ z_lgfr(Rdst, Rsrc2);
    }
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct z196_minI_reg_imm32(iRegI dst, iRegI src1, immI src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_LoadStoreConditional());
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI const32 (z196 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    int      Isrc2 = $src2$$constant;

    if (Rdst == Rsrc1) {
      __ load_const_optimized(Z_R0_scratch, Isrc2);
      __ z_cfi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Z_R0_scratch, Assembler::bcondNotLow);
    } else {
      __ load_const_optimized(Rdst, Isrc2);
      __ z_cfi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Rsrc1, Assembler::bcondLow);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct minI_reg_imm32(iRegI dst, iRegI src1, immI src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI const32" %}
  ins_encode %{
    Label done;
    if ($dst$$Register != $src1$$Register) {
      __ z_lgfr($dst$$Register, $src1$$Register);
    }
    __ z_cfi($src1$$Register, $src2$$constant);
    __ z_brl(done);
    __ z_lgfi($dst$$Register, $src2$$constant);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct z196_minI_reg_imm16(iRegI dst, iRegI src1, immI16 src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_LoadStoreConditional());
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI const16 (z196 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    int      Isrc2 = $src2$$constant;

    if (Rdst == Rsrc1) {
      __ load_const_optimized(Z_R0_scratch, Isrc2);
      __ z_chi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Z_R0_scratch, Assembler::bcondNotLow);
    } else {
      __ load_const_optimized(Rdst, Isrc2);
      __ z_chi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Rsrc1, Assembler::bcondLow);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct minI_reg_imm16(iRegI dst, iRegI src1, immI16 src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI const16" %}
  ins_encode %{
    Label done;
    if ($dst$$Register != $src1$$Register) {
      __ z_lgfr($dst$$Register, $src1$$Register);
    }
    __ z_chi($src1$$Register, $src2$$constant);
    __ z_brl(done);
    __ z_lghi($dst$$Register, $src2$$constant);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct z10_minI_reg_imm8(iRegI dst, iRegI src1, immI8 src2, flagsReg cr) %{
  match(Set dst (MinI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MinI $dst $src1,$src2\t MinI const8 (z10 only)" %}
  ins_encode %{
    Label done;
    if ($dst$$Register != $src1$$Register) {
      __ z_lgfr($dst$$Register, $src1$$Register);
    }
    __ z_cij($src1$$Register, $src2$$constant, Assembler::bcondLow, done);
    __ z_lghi($dst$$Register, $src2$$constant);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Max Register with Register
instruct z196_maxI_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_LoadStoreConditional());
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI (z196 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    Register Rsrc2 = $src2$$Register;

    if (Rsrc1 == Rsrc2) {
      if (Rdst != Rsrc1) {
        __ z_lgfr(Rdst, Rsrc1);
      }
    } else if (Rdst == Rsrc1) { // Rdst preset with src1.
      __ z_cr(Rsrc1, Rsrc2);    // Move src2 only if src1 is NotHigh.
      __ z_locr(Rdst, Rsrc2, Assembler::bcondNotHigh);
    } else if (Rdst == Rsrc2) { // Rdst preset with src2.
      __ z_cr(Rsrc2, Rsrc1);    // Move src1 only if src2 is NotHigh.
      __ z_locr(Rdst, Rsrc1, Assembler::bcondNotHigh);
    } else {                    // Rdst is disjoint from operands, move in either case.
      __ z_cr(Rsrc1, Rsrc2);
      __ z_locr(Rdst, Rsrc2, Assembler::bcondNotHigh);
      __ z_locr(Rdst, Rsrc1, Assembler::bcondHigh);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Max Register with Register
instruct z10_maxI_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI (z10 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    Register Rsrc2 = $src2$$Register;
    Label done;

    if (Rsrc1 == Rsrc2) {
      if (Rdst != Rsrc1) {
        __ z_lgfr(Rdst, Rsrc1);
      }
    } else if (Rdst == Rsrc1) {
      __ z_crj(Rsrc1, Rsrc2, Assembler::bcondHigh, done);
      __ z_lgfr(Rdst, Rsrc2);
    } else if (Rdst == Rsrc2) {
      __ z_crj(Rsrc2, Rsrc1, Assembler::bcondHigh, done);
      __ z_lgfr(Rdst, Rsrc1);
    } else {
      __ z_lgfr(Rdst, Rsrc1);
      __ z_crj(Rsrc1, Rsrc2, Assembler::bcondHigh, done);
      __ z_lgfr(Rdst, Rsrc2);
    }
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct maxI_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  predicate(!VM_Version::has_CompareBranch());
  ins_cost(3 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    Register Rsrc2 = $src2$$Register;
    Label done;

    if (Rsrc1 == Rsrc2) {
      if (Rdst != Rsrc1) {
        __ z_lgfr(Rdst, Rsrc1);
      }
    } else if (Rdst == Rsrc1) {
      __ z_cr(Rsrc1, Rsrc2);
      __ z_brh(done);
      __ z_lgfr(Rdst, Rsrc2);
    } else if (Rdst == Rsrc2) {
      __ z_cr(Rsrc2, Rsrc1);
      __ z_brh(done);
      __ z_lgfr(Rdst, Rsrc1);
    } else {
      __ z_lgfr(Rdst, Rsrc1);
      __ z_cr(Rsrc1, Rsrc2);
      __ z_brh(done);
      __ z_lgfr(Rdst, Rsrc2);
    }

    __ bind(done);
  %}

  ins_pipe(pipe_class_dummy);
%}

instruct z196_maxI_reg_imm32(iRegI dst, iRegI src1, immI src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_LoadStoreConditional());
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI const32 (z196 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    int      Isrc2 = $src2$$constant;

    if (Rdst == Rsrc1) {
      __ load_const_optimized(Z_R0_scratch, Isrc2);
      __ z_cfi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Z_R0_scratch, Assembler::bcondNotHigh);
    } else {
      __ load_const_optimized(Rdst, Isrc2);
      __ z_cfi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Rsrc1, Assembler::bcondHigh);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct maxI_reg_imm32(iRegI dst, iRegI src1, immI src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI const32" %}
  ins_encode %{
    Label done;
    if ($dst$$Register != $src1$$Register) {
      __ z_lgfr($dst$$Register, $src1$$Register);
    }
    __ z_cfi($src1$$Register, $src2$$constant);
    __ z_brh(done);
    __ z_lgfi($dst$$Register, $src2$$constant);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct z196_maxI_reg_imm16(iRegI dst, iRegI src1, immI16 src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_LoadStoreConditional());
  ins_cost(3 * DEFAULT_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI const16 (z196 only)" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc1 = $src1$$Register;
    int      Isrc2 = $src2$$constant;
    if (Rdst == Rsrc1) {
      __ load_const_optimized(Z_R0_scratch, Isrc2);
      __ z_chi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Z_R0_scratch, Assembler::bcondNotHigh);
    } else {
      __ load_const_optimized(Rdst, Isrc2);
      __ z_chi(Rsrc1, Isrc2);
      __ z_locr(Rdst, Rsrc1, Assembler::bcondHigh);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct maxI_reg_imm16(iRegI dst, iRegI src1, immI16 src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  ins_cost(2 * DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI const16" %}
  ins_encode %{
    Label done;
    if ($dst$$Register != $src1$$Register) {
      __ z_lgfr($dst$$Register, $src1$$Register);
    }
    __ z_chi($src1$$Register, $src2$$constant);
    __ z_brh(done);
    __ z_lghi($dst$$Register, $src2$$constant);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct z10_maxI_reg_imm8(iRegI dst, iRegI src1, immI8 src2, flagsReg cr) %{
  match(Set dst (MaxI src1 src2));
  effect(KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(DEFAULT_COST + BRANCH_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "MaxI $dst $src1,$src2\t MaxI const8" %}
  ins_encode %{
    Label done;
    if ($dst$$Register != $src1$$Register) {
      __ z_lgfr($dst$$Register, $src1$$Register);
    }
    __ z_cij($src1$$Register, $src2$$constant, Assembler::bcondHigh, done);
    __ z_lghi($dst$$Register, $src2$$constant);
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

//----------Abs---------------------------------------------------------------

instruct absI_reg(iRegI dst, iRegI src, flagsReg cr) %{
  match(Set dst (AbsI src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LPR     $dst, $src" %}
  opcode(LPR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct absL_reg(iRegL dst, iRegL src, flagsReg cr) %{
  match(Set dst (AbsL src));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LPGR     $dst, $src" %}
  opcode(LPGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct negabsI_reg(iRegI dst, iRegI src, immI_0 zero, flagsReg cr) %{
  match(Set dst (SubI zero (AbsI src)));
  effect(KILL cr);
  ins_cost(DEFAULT_COST_LOW);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LNR     $dst, $src" %}
  opcode(LNR_ZOPC);
  ins_encode(z_rrform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

//----------Float Compares----------------------------------------------------

// Compare floating, generate condition code.
instruct cmpF_cc(flagsReg cr, regF src1, regF src2) %{
  match(Set cr (CmpF src1 src2));
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "FCMPcc   $src1,$src2\t # float" %}
  ins_encode %{ __ z_cebr($src1$$FloatRegister, $src2$$FloatRegister); %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmpD_cc(flagsReg cr, regD src1, regD src2) %{
  match(Set cr (CmpD src1 src2));
  ins_cost(ALU_REG_COST);
  size(4);
  format %{ "FCMPcc   $src1,$src2 \t # double" %}
  ins_encode %{ __ z_cdbr($src1$$FloatRegister, $src2$$FloatRegister); %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmpF_cc_mem(flagsReg cr, regF src1, memoryRX src2) %{
  match(Set cr (CmpF src1 (LoadF src2)));
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "FCMPcc_mem $src1,$src2\t # floatMemory" %}
  opcode(CEB_ZOPC);
  ins_encode(z_form_rt_memFP(src1, src2));
  ins_pipe(pipe_class_dummy);
%}

instruct cmpD_cc_mem(flagsReg cr, regD src1, memoryRX src2) %{
  match(Set cr (CmpD src1 (LoadD src2)));
  ins_cost(ALU_MEMORY_COST);
  size(6);
  format %{ "DCMPcc_mem $src1,$src2\t # doubleMemory" %}
  opcode(CDB_ZOPC);
  ins_encode(z_form_rt_memFP(src1, src2));
  ins_pipe(pipe_class_dummy);
%}

// Compare floating, generate condition code
instruct cmpF0_cc(flagsReg cr, regF src1, immFpm0 src2) %{
  match(Set cr (CmpF src1 src2));
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "LTEBR    $src1,$src1\t # float" %}
  opcode(LTEBR_ZOPC);
  ins_encode(z_rreform(src1, src1));
  ins_pipe(pipe_class_dummy);
%}

instruct cmpD0_cc(flagsReg cr, regD src1, immDpm0 src2) %{
  match(Set cr (CmpD src1 src2));
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "LTDBR    $src1,$src1 \t # double" %}
  opcode(LTDBR_ZOPC);
  ins_encode(z_rreform(src1, src1));
  ins_pipe(pipe_class_dummy);
%}

// Compare floating, generate -1,0,1
instruct cmpF_reg(iRegI dst, regF src1, regF src2, flagsReg cr) %{
  match(Set dst (CmpF3 src1 src2));
  effect(KILL cr);
  ins_cost(DEFAULT_COST * 5 + BRANCH_COST);
  size(24);
  format %{ "CmpF3    $dst,$src1,$src2" %}
  ins_encode %{
    // compare registers
    __ z_cebr($src1$$FloatRegister, $src2$$FloatRegister);
    // Convert condition code into -1,0,1, where
    // -1 means unordered or less
    //  0 means equal
    //  1 means greater.
    if (VM_Version::has_LoadStoreConditional()) {
      Register one       = Z_R0_scratch;
      Register minus_one = Z_R1_scratch;
      __ z_lghi(minus_one, -1);
      __ z_lghi(one, 1);
      __ z_lghi( $dst$$Register, 0);
      __ z_locgr($dst$$Register, one,       Assembler::bcondHigh);
      __ z_locgr($dst$$Register, minus_one, Assembler::bcondLowOrNotOrdered);
    } else {
      Label done;
      __ clear_reg($dst$$Register, true, false);
      __ z_bre(done);
      __ z_lhi($dst$$Register, 1);
      __ z_brh(done);
      __ z_lhi($dst$$Register, -1);
      __ bind(done);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmpD_reg(iRegI dst, regD src1, regD src2, flagsReg cr) %{
  match(Set dst (CmpD3 src1 src2));
  effect(KILL cr);
  ins_cost(DEFAULT_COST * 5 + BRANCH_COST);
  size(24);
  format %{ "CmpD3    $dst,$src1,$src2" %}
  ins_encode %{
    // compare registers
    __ z_cdbr($src1$$FloatRegister, $src2$$FloatRegister);
    // Convert condition code into -1,0,1, where
    // -1 means unordered or less
    //  0 means equal
    //  1 means greater.
    if (VM_Version::has_LoadStoreConditional()) {
      Register one       = Z_R0_scratch;
      Register minus_one = Z_R1_scratch;
      __ z_lghi(minus_one, -1);
      __ z_lghi(one, 1);
      __ z_lghi( $dst$$Register, 0);
      __ z_locgr($dst$$Register, one,       Assembler::bcondHigh);
      __ z_locgr($dst$$Register, minus_one, Assembler::bcondLowOrNotOrdered);
    } else {
      Label done;
      // indicate unused result
      (void) __ clear_reg($dst$$Register, true, false);
      __ z_bre(done);
      __ z_lhi($dst$$Register, 1);
      __ z_brh(done);
      __ z_lhi($dst$$Register, -1);
      __ bind(done);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

//----------Branches---------------------------------------------------------
// Jump

// Direct Branch.
instruct branch(label labl) %{
  match(Goto);
  effect(USE labl);
  ins_cost(BRANCH_COST);
  size(4);
  format %{ "BRU     $labl" %}
  ins_encode(z_enc_bru(labl));
  ins_pipe(pipe_class_dummy);
  // If set to 1 this indicates that the current instruction is a
  // short variant of a long branch. This avoids using this
  // instruction in first-pass matching. It will then only be used in
  // the `Shorten_branches' pass.
  ins_short_branch(1);
%}

// Direct Branch.
instruct branchFar(label labl) %{
  match(Goto);
  effect(USE labl);
  ins_cost(BRANCH_COST);
  size(6);
  format %{ "BRUL   $labl" %}
  ins_encode(z_enc_brul(labl));
  ins_pipe(pipe_class_dummy);
  // This is not a short variant of a branch, but the long variant.
  ins_short_branch(0);
%}

// Conditional Near Branch
instruct branchCon(cmpOp cmp, flagsReg cr, label lbl) %{
  // Same match rule as `branchConFar'.
  match(If cmp cr);
  effect(USE lbl);
  ins_cost(BRANCH_COST);
  size(4);
  format %{ "branch_con_short,$cmp   $lbl" %}
  ins_encode(z_enc_branch_con_short(cmp, lbl));
  ins_pipe(pipe_class_dummy);
  // If set to 1 this indicates that the current instruction is a
  // short variant of a long branch. This avoids using this
  // instruction in first-pass matching. It will then only be used in
  // the `Shorten_branches' pass.
  ins_short_branch(1);
%}

// This is for cases when the z/Architecture conditional branch instruction
// does not reach far enough. So we emit a far branch here, which is
// more expensive.
//
// Conditional Far Branch
instruct branchConFar(cmpOp cmp, flagsReg cr, label lbl) %{
  // Same match rule as `branchCon'.
  match(If cmp cr);
  effect(USE cr, USE lbl);
  // Make more expensive to prefer compare_and_branch over separate instructions.
  ins_cost(2 * BRANCH_COST);
  size(6);
  format %{ "branch_con_far,$cmp   $lbl" %}
  ins_encode(z_enc_branch_con_far(cmp, lbl));
  ins_pipe(pipe_class_dummy);
  // This is not a short variant of a branch, but the long variant..
  ins_short_branch(0);
%}

instruct branchLoopEnd(cmpOp cmp, flagsReg cr, label labl) %{
  match(CountedLoopEnd cmp cr);
  effect(USE labl);
  ins_cost(BRANCH_COST);
  size(4);
  format %{ "branch_con_short,$cmp   $labl\t # counted loop end" %}
  ins_encode(z_enc_branch_con_short(cmp, labl));
  ins_pipe(pipe_class_dummy);
  // If set to 1 this indicates that the current instruction is a
  // short variant of a long branch. This avoids using this
  // instruction in first-pass matching. It will then only be used in
  // the `Shorten_branches' pass.
  ins_short_branch(1);
%}

instruct branchLoopEndFar(cmpOp cmp, flagsReg cr, label labl) %{
  match(CountedLoopEnd cmp cr);
  effect(USE labl);
  ins_cost(BRANCH_COST);
  size(6);
  format %{ "branch_con_far,$cmp   $labl\t # counted loop end" %}
  ins_encode(z_enc_branch_con_far(cmp, labl));
  ins_pipe(pipe_class_dummy);
  // This is not a short variant of a branch, but the long variant.
  ins_short_branch(0);
%}

//----------Compare and Branch (short distance)------------------------------

// INT REG operands for loop counter processing.
instruct testAndBranchLoopEnd_Reg(cmpOpT boolnode, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
  match(CountedLoopEnd boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "test_and_branch_loop_end,$boolnode  $src1,$src2,$labl\t # counted loop end SHORT" %}
  opcode(CRJ_ZOPC);
  ins_encode(z_enc_cmpb_regreg(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// INT REG operands.
instruct cmpb_RegI(cmpOpT boolnode, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CRJ,$boolnode  $src1,$src2,$labl\t # SHORT" %}
  opcode(CRJ_ZOPC);
  ins_encode(z_enc_cmpb_regreg(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// Unsigned INT REG operands
instruct cmpbU_RegI(cmpOpT boolnode, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpU src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLRJ,$boolnode  $src1,$src2,$labl\t # SHORT" %}
  opcode(CLRJ_ZOPC);
  ins_encode(z_enc_cmpb_regreg(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// LONG REG operands
instruct cmpb_RegL(cmpOpT boolnode, iRegL src1, iRegL src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpL src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CGRJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CGRJ_ZOPC);
  ins_encode(z_enc_cmpb_regreg(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

//  PTR REG operands

// Separate rules for regular and narrow oops.  ADLC can't recognize
// rules with polymorphic operands to be sisters -> shorten_branches
// will not shorten.

instruct cmpb_RegPP(cmpOpT boolnode, iRegP src1, iRegP src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGRJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CLGRJ_ZOPC);
  ins_encode(z_enc_cmpb_regreg(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

instruct cmpb_RegNN(cmpOpT boolnode, iRegN src1, iRegN src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP (DecodeN src1) (DecodeN src2)));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGRJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CLGRJ_ZOPC);
  ins_encode(z_enc_cmpb_regreg(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// INT REG/IMM operands for loop counter processing
instruct testAndBranchLoopEnd_Imm(cmpOpT boolnode, iRegI src1, immI8 src2, label labl, flagsReg cr) %{
  match(CountedLoopEnd boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "test_and_branch_loop_end,$boolnode  $src1,$src2,$labl\t # counted loop end SHORT" %}
  opcode(CIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// INT REG/IMM operands
instruct cmpb_RegI_imm(cmpOpT boolnode, iRegI src1, immI8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CIJ,$boolnode  $src1,$src2,$labl\t # SHORT" %}
  opcode(CIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// INT REG/IMM operands
instruct cmpbU_RegI_imm(cmpOpT boolnode, iRegI src1, uimmI8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpU src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLIJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CLIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// LONG REG/IMM operands
instruct cmpb_RegL_imm(cmpOpT boolnode, iRegL src1, immL8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpL src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CGIJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CGIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// PTR REG-imm operands

// Separate rules for regular and narrow oops. ADLC can't recognize
// rules with polymorphic operands to be sisters -> shorten_branches
// will not shorten.

instruct cmpb_RegP_immP(cmpOpT boolnode, iRegP src1, immP8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGIJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CLGIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

// Compare against zero only, do not mix N and P oops (encode/decode required).
instruct cmpb_RegN_immP0(cmpOpT boolnode, iRegN src1, immP0 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP (DecodeN src1) src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGIJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CLGIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}

instruct cmpb_RegN_imm(cmpOpT boolnode, iRegN src1, immN8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP (DecodeN src1) (DecodeN src2)));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGIJ,$boolnode $src1,$src2,$labl\t # SHORT" %}
  opcode(CLGIJ_ZOPC);
  ins_encode(z_enc_cmpb_regimm(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(1);
%}


//----------Compare and Branch (far distance)------------------------------

// INT REG operands for loop counter processing
instruct testAndBranchLoopEnd_RegFar(cmpOpT boolnode, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
  match(CountedLoopEnd boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "test_and_branch_loop_end,$boolnode  $src1,$src2,$labl\t # counted loop end FAR" %}
  opcode(CR_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regregFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// INT REG operands
instruct cmpb_RegI_Far(cmpOpT boolnode, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CRJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CR_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regregFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// INT REG operands
instruct cmpbU_RegI_Far(cmpOpT boolnode, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpU src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLRJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLR_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regregFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// LONG REG operands
instruct cmpb_RegL_Far(cmpOpT boolnode, iRegL src1, iRegL src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpL src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CGRJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CGR_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regregFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// PTR REG operands

// Separate rules for regular and narrow oops. ADLC can't recognize
// rules with polymorphic operands to be sisters -> shorten_branches
// will not shorten.

instruct cmpb_RegPP_Far(cmpOpT boolnode, iRegP src1, iRegP src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGRJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLGR_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regregFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

instruct cmpb_RegNN_Far(cmpOpT boolnode, iRegN src1, iRegN src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP (DecodeN src1) (DecodeN src2)));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGRJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLGR_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regregFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// INT REG/IMM operands for loop counter processing
instruct testAndBranchLoopEnd_ImmFar(cmpOpT boolnode, iRegI src1, immI8 src2, label labl, flagsReg cr) %{
  match(CountedLoopEnd boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "test_and_branch_loop_end,$boolnode  $src1,$src2,$labl\t # counted loop end FAR" %}
  opcode(CHI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// INT REG/IMM operands
instruct cmpb_RegI_imm_Far(cmpOpT boolnode, iRegI src1, immI8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpI src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CIJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CHI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// INT REG/IMM operands
instruct cmpbU_RegI_imm_Far(cmpOpT boolnode, iRegI src1, uimmI8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpU src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLIJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLFI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// LONG REG/IMM operands
instruct cmpb_RegL_imm_Far(cmpOpT boolnode, iRegL src1, immL8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpL src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CGIJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CGHI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// PTR REG-imm operands

// Separate rules for regular and narrow oops. ADLC can't recognize
// rules with polymorphic operands to be sisters -> shorten_branches
// will not shorten.

instruct cmpb_RegP_immP_Far(cmpOpT boolnode, iRegP src1, immP8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP src1 src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGIJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLGFI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// Compare against zero only, do not mix N and P oops (encode/decode required).
instruct cmpb_RegN_immP0_Far(cmpOpT boolnode, iRegN src1, immP0 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP (DecodeN src1) src2));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGIJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLGFI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

instruct cmpb_RegN_immN_Far(cmpOpT boolnode, iRegN src1, immN8 src2, label labl, flagsReg cr) %{
  match(If boolnode (CmpP (DecodeN src1) (DecodeN src2)));
  effect(USE labl, KILL cr);
  predicate(VM_Version::has_CompareBranch());
  ins_cost(BRANCH_COST+DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CLGIJ,$boolnode   $src1,$src2,$labl\t # FAR(substituted)" %}
  opcode(CLGFI_ZOPC, BRCL_ZOPC);
  ins_encode(z_enc_cmpb_regimmFar(src1, src2, labl, boolnode));
  ins_pipe(pipe_class_dummy);
  ins_short_branch(0);
%}

// ============================================================================
// Long Compare

// 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 platforms which have few registers.

// 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 cr) %{
  match(Set dst (CmpL3 src1 src2));
  effect(KILL cr);
  ins_cost(DEFAULT_COST * 5 + BRANCH_COST);
  size(24);
  format %{ "CmpL3 $dst,$src1,$src2" %}
  ins_encode %{
    Label done;
    // compare registers
    __ z_cgr($src1$$Register, $src2$$Register);
    // Convert condition code into -1,0,1, where
    // -1 means less
    //  0 means equal
    //  1 means greater.
    if (VM_Version::has_LoadStoreConditional()) {
      Register one       = Z_R0_scratch;
      Register minus_one = Z_R1_scratch;
      __ z_lghi(minus_one, -1);
      __ z_lghi(one, 1);
      __ z_lghi( $dst$$Register, 0);
      __ z_locgr($dst$$Register, one,       Assembler::bcondHigh);
      __ z_locgr($dst$$Register, minus_one, Assembler::bcondLow);
    } else {
      __ clear_reg($dst$$Register, true, false);
      __ z_bre(done);
      __ z_lhi($dst$$Register, 1);
      __ z_brh(done);
      __ z_lhi($dst$$Register, -1);
    }
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

// ============================================================================
// Safepoint Instruction

instruct safePoint() %{
  match(SafePoint);
  predicate(false);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "UNIMPLEMENTED Safepoint_ " %}
  ins_encode(enc_unimplemented());
  ins_pipe(pipe_class_dummy);
%}

instruct safePoint_poll(iRegP poll, flagsReg cr) %{
  match(SafePoint poll);
  effect(USE poll, KILL cr); // R0 is killed, too.
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "TM      #0[,$poll],#111\t # Safepoint: poll for GC" %}
  ins_encode %{
    // Mark the code position where the load from the safepoint
    // polling page was emitted as relocInfo::poll_type.
    __ relocate(relocInfo::poll_type);
    __ load_from_polling_page($poll$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

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

// Call Instructions

// Call Java Static Instruction
instruct CallStaticJavaDirect_dynTOC(method meth) %{
  match(CallStaticJava);
  effect(USE meth);
  ins_cost(CALL_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CALL,static dynTOC $meth; ==> " %}
  ins_encode( z_enc_java_static_call(meth) );
  ins_pipe(pipe_class_dummy);
  ins_alignment(2);
%}

// Call Java Dynamic Instruction
instruct CallDynamicJavaDirect_dynTOC(method meth) %{
  match(CallDynamicJava);
  effect(USE meth);
  ins_cost(CALL_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "CALL,dynamic dynTOC $meth; ==> " %}
  ins_encode(z_enc_java_dynamic_call(meth));
  ins_pipe(pipe_class_dummy);
  ins_alignment(2);
%}

// Call Runtime Instruction
instruct CallRuntimeDirect(method meth) %{
  match(CallRuntime);
  effect(USE meth);
  ins_cost(CALL_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  ins_num_consts(1);
  ins_alignment(2);
  format %{ "CALL,runtime" %}
  ins_encode( z_enc_java_to_runtime_call(meth) );
  ins_pipe(pipe_class_dummy);
%}

// Call runtime without safepoint - same as CallRuntime
instruct CallLeafDirect(method meth) %{
  match(CallLeaf);
  effect(USE meth);
  ins_cost(CALL_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  ins_num_consts(1);
  ins_alignment(2);
  format %{ "CALL,runtime leaf $meth" %}
  ins_encode( z_enc_java_to_runtime_call(meth) );
  ins_pipe(pipe_class_dummy);
%}

// Call runtime without safepoint - same as CallLeaf
instruct CallLeafNoFPDirect(method meth) %{
  match(CallLeafNoFP);
  effect(USE meth);
  ins_cost(CALL_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  ins_num_consts(1);
  format %{ "CALL,runtime leaf nofp $meth" %}
  ins_encode( z_enc_java_to_runtime_call(meth) );
  ins_pipe(pipe_class_dummy);
  ins_alignment(2);
%}

// 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(iRegP jump_target, inline_cache_regP method_ptr) %{
  match(TailCall jump_target method_ptr);
  ins_cost(CALL_COST);
  size(2);
  format %{ "Jmp     $jump_target\t # $method_ptr holds method" %}
  ins_encode %{ __ z_br($jump_target$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// Return Instruction
instruct Ret() %{
  match(Return);
  size(2);
  format %{ "BR(Z_R14) // branch to link register" %}
  ins_encode %{ __ z_br(Z_R14); %}
  ins_pipe(pipe_class_dummy);
%}

// 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(iRegP jump_target, rarg1RegP ex_oop) %{
  match(TailJump jump_target ex_oop);
  ins_cost(CALL_COST);
  size(8);
  format %{ "TailJump $jump_target" %}
  ins_encode %{
    __ z_lg(Z_ARG2/* issuing pc */, _z_abi(return_pc), Z_SP);
    __ z_br($jump_target$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

// 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(rarg1RegP ex_oop) %{
  match(Set ex_oop (CreateEx));
  ins_cost(0);
  size(0);
  format %{ "# exception oop; no code emitted" %}
  ins_encode(/*empty*/);
  ins_pipe(pipe_class_dummy);
%}

// 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);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "Jmp    rethrow_stub" %}
  ins_encode %{
    cbuf.set_insts_mark();
    __ load_const_optimized(Z_R1_scratch, (address)OptoRuntime::rethrow_stub());
    __ z_br(Z_R1_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Die now.
instruct ShouldNotReachHere() %{
  match(Halt);
  ins_cost(CALL_COST);
  format %{ "ILLTRAP; ShouldNotReachHere" %}
  ins_encode %{
    if (is_reachable()) {
      __ stop(_halt_reason);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// ============================================================================
// 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(rarg1RegP index, rarg2RegP sub, rarg3RegP super, flagsReg pcc,
                             rarg4RegP scratch1, rarg5RegP scratch2) %{
  match(Set index (PartialSubtypeCheck sub super));
  effect(KILL pcc, KILL scratch1, KILL scratch2);
  ins_cost(10 * DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "  CALL   PartialSubtypeCheck\n" %}
  ins_encode %{
    AddressLiteral stub_address(StubRoutines::zarch::partial_subtype_check());
    __ load_const_optimized(Z_ARG4, stub_address);
    __ z_basr(Z_R14, Z_ARG4);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct partialSubtypeCheck_vs_zero(flagsReg pcc, rarg2RegP sub, rarg3RegP super, immP0 zero,
                                     rarg1RegP index, rarg4RegP scratch1, rarg5RegP scratch2) %{
  match(Set pcc (CmpI (PartialSubtypeCheck sub super) zero));
  effect(KILL scratch1, KILL scratch2, KILL index);
  ins_cost(10 * DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "CALL   PartialSubtypeCheck_vs_zero\n" %}
  ins_encode %{
    AddressLiteral stub_address(StubRoutines::zarch::partial_subtype_check());
    __ load_const_optimized(Z_ARG4, stub_address);
    __ z_basr(Z_R14, Z_ARG4);
  %}
  ins_pipe(pipe_class_dummy);
%}

// ============================================================================
// inlined locking and unlocking

instruct cmpFastLock(flagsReg pcc, iRegP_N2P oop, iRegP_N2P box, iRegP tmp1, iRegP tmp2) %{
  match(Set pcc (FastLock oop box));
  effect(TEMP tmp1, TEMP tmp2);
  ins_cost(100);
  // TODO: s390 port size(VARIABLE_SIZE); // Uses load_const_optimized.
  format %{ "FASTLOCK  $oop, $box; KILL Z_ARG4, Z_ARG5" %}
  ins_encode %{ __ compiler_fast_lock_object($oop$$Register, $box$$Register, $tmp1$$Register, $tmp2$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct cmpFastUnlock(flagsReg pcc, iRegP_N2P oop, iRegP_N2P box, iRegP tmp1, iRegP tmp2) %{
  match(Set pcc (FastUnlock oop box));
  effect(TEMP tmp1, TEMP tmp2);
  ins_cost(100);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "FASTUNLOCK  $oop, $box; KILL Z_ARG4, Z_ARG5" %}
  ins_encode %{ __ compiler_fast_unlock_object($oop$$Register, $box$$Register, $tmp1$$Register, $tmp2$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct inlineCallClearArrayConst(SSlenDW cnt, iRegP_N2P base, Universe dummy, flagsReg cr) %{
  match(Set dummy (ClearArray cnt base));
  effect(KILL cr);
  ins_cost(100);
  // TODO: s390 port size(VARIABLE_SIZE);       // Variable in size due to varying #instructions.
  format %{ "ClearArrayConst $cnt,$base" %}
  ins_encode %{ __ Clear_Array_Const($cnt$$constant, $base$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct inlineCallClearArrayConstBig(immL cnt, iRegP_N2P base, Universe dummy, allRoddRegL tmpL, flagsReg cr) %{
  match(Set dummy (ClearArray cnt base));
  effect(TEMP tmpL, KILL cr); // R0, R1 are killed, too.
  ins_cost(200);
  // TODO: s390 port size(VARIABLE_SIZE);       // Variable in size due to optimized constant loader.
  format %{ "ClearArrayConstBig $cnt,$base" %}
  ins_encode %{ __ Clear_Array_Const_Big($cnt$$constant, $base$$Register, $tmpL$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

instruct inlineCallClearArray(iRegL cnt, iRegP_N2P base, Universe dummy, allRoddRegL tmpL, flagsReg cr) %{
  match(Set dummy (ClearArray cnt base));
  effect(TEMP tmpL, KILL cr); // R0, R1 are killed, too.
  ins_cost(300);
  // TODO: s390 port size(FIXED_SIZE);  // z/Architecture: emitted code depends on PreferLAoverADD being on/off.
  format %{ "ClearArrayVar $cnt,$base" %}
  ins_encode %{ __ Clear_Array($cnt$$Register, $base$$Register, $tmpL$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// ============================================================================
// CompactStrings

// String equals
instruct string_equalsL(iRegP str1, iRegP str2, iRegI cnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrEquals (Binary str1 str2) cnt));
  effect(TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::LL);
  ins_cost(300);
  format %{ "String Equals byte[] $str1,$str2,$cnt -> $result" %}
  ins_encode %{
    __ array_equals(false, $str1$$Register, $str2$$Register,
                    $cnt$$Register, $oddReg$$Register, $evenReg$$Register,
                    $result$$Register, true /* byte */);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct string_equalsU(iRegP str1, iRegP str2, iRegI cnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrEquals (Binary str1 str2) cnt));
  effect(TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::UU || ((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::none);
  ins_cost(300);
  format %{ "String Equals char[] $str1,$str2,$cnt -> $result" %}
  ins_encode %{
    __ array_equals(false, $str1$$Register, $str2$$Register,
                    $cnt$$Register, $oddReg$$Register, $evenReg$$Register,
                    $result$$Register, false /* byte */);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct string_equals_imm(iRegP str1, iRegP str2, uimmI8 cnt, iRegI result, flagsReg cr) %{
  match(Set result (StrEquals (Binary str1 str2) cnt));
  effect(KILL cr); // R0 is killed, too.
  predicate(((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::LL || ((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::UU);
  ins_cost(100);
  format %{ "String Equals byte[] $str1,$str2,$cnt -> $result" %}
  ins_encode %{
    const int cnt_imm = $cnt$$constant;
    if (cnt_imm) { __ z_clc(0, cnt_imm - 1, $str1$$Register, 0, $str2$$Register); }
    __ z_lhi($result$$Register, 1);
    if (cnt_imm) {
      if (VM_Version::has_LoadStoreConditional()) {
        __ z_lhi(Z_R0_scratch, 0);
        __ z_locr($result$$Register, Z_R0_scratch, Assembler::bcondNotEqual);
      } else {
        Label Lskip;
        __ z_bre(Lskip);
        __ clear_reg($result$$Register);
        __ bind(Lskip);
      }
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct string_equalsC_imm(iRegP str1, iRegP str2, immI8 cnt, iRegI result, flagsReg cr) %{
  match(Set result (StrEquals (Binary str1 str2) cnt));
  effect(KILL cr); // R0 is killed, too.
  predicate(((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::none);
  ins_cost(100);
  format %{ "String Equals $str1,$str2,$cnt -> $result" %}
  ins_encode %{
    const int cnt_imm = $cnt$$constant; // positive immI8 (7 bits used)
    if (cnt_imm) { __ z_clc(0, (cnt_imm << 1) - 1, $str1$$Register, 0, $str2$$Register); }
    __ z_lhi($result$$Register, 1);
    if (cnt_imm) {
      if (VM_Version::has_LoadStoreConditional()) {
        __ z_lhi(Z_R0_scratch, 0);
        __ z_locr($result$$Register, Z_R0_scratch, Assembler::bcondNotEqual);
      } else {
        Label Lskip;
        __ z_bre(Lskip);
        __ clear_reg($result$$Register);
        __ bind(Lskip);
      }
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Array equals
instruct array_equalsB(iRegP ary1, iRegP ary2, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (AryEq ary1 ary2));
  effect(TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((AryEqNode*)n)->encoding() == StrIntrinsicNode::LL);
  ins_cost(300);
  format %{ "Array Equals $ary1,$ary2 -> $result" %}
  ins_encode %{
    __ array_equals(true, $ary1$$Register, $ary2$$Register,
                    noreg, $oddReg$$Register, $evenReg$$Register,
                    $result$$Register, true /* byte */);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct array_equalsC(iRegP ary1, iRegP ary2, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (AryEq ary1 ary2));
  effect(TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((AryEqNode*)n)->encoding() == StrIntrinsicNode::UU);
  ins_cost(300);
  format %{ "Array Equals $ary1,$ary2 -> $result" %}
  ins_encode %{
    __ array_equals(true, $ary1$$Register, $ary2$$Register,
                    noreg, $oddReg$$Register, $evenReg$$Register,
                    $result$$Register, false /* byte */);
  %}
  ins_pipe(pipe_class_dummy);
%}

// String CompareTo
instruct string_compareL(iRegP str1, iRegP str2, rarg2RegI cnt1, rarg5RegI cnt2, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(TEMP_DEF result, USE_KILL cnt1, USE_KILL cnt2, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::LL);
  ins_cost(300);
  format %{ "String Compare byte[] $str1,$cnt1,$str2,$cnt2 -> $result" %}
  ins_encode %{
    __ string_compare($str1$$Register, $str2$$Register,
                      $cnt1$$Register, $cnt2$$Register,
                      $oddReg$$Register, $evenReg$$Register,
                      $result$$Register, StrIntrinsicNode::LL);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct string_compareU(iRegP str1, iRegP str2, rarg2RegI cnt1, rarg5RegI cnt2, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(TEMP_DEF result, USE_KILL cnt1, USE_KILL cnt2, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::UU || ((StrCompNode*)n)->encoding() == StrIntrinsicNode::none);
  ins_cost(300);
  format %{ "String Compare char[] $str1,$cnt1,$str2,$cnt2 -> $result" %}
  ins_encode %{
    __ string_compare($str1$$Register, $str2$$Register,
                      $cnt1$$Register, $cnt2$$Register,
                      $oddReg$$Register, $evenReg$$Register,
                      $result$$Register, StrIntrinsicNode::UU);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct string_compareLU(iRegP str1, iRegP str2, rarg2RegI cnt1, rarg5RegI cnt2, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(TEMP_DEF result, USE_KILL cnt1, USE_KILL cnt2, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::LU);
  ins_cost(300);
  format %{ "String Compare byte[],char[] $str1,$cnt1,$str2,$cnt2 -> $result" %}
  ins_encode %{
    __ string_compare($str1$$Register, $str2$$Register,
                      $cnt1$$Register, $cnt2$$Register,
                      $oddReg$$Register, $evenReg$$Register,
                      $result$$Register, StrIntrinsicNode::LU);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct string_compareUL(iRegP str1, iRegP str2, rarg2RegI cnt1, rarg5RegI cnt2, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
  effect(TEMP_DEF result, USE_KILL cnt1, USE_KILL cnt2, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::UL);
  ins_cost(300);
  format %{ "String Compare char[],byte[] $str1,$cnt1,$str2,$cnt2 -> $result" %}
  ins_encode %{
    __ string_compare($str2$$Register, $str1$$Register,
                      $cnt2$$Register, $cnt1$$Register,
                      $oddReg$$Register, $evenReg$$Register,
                      $result$$Register, StrIntrinsicNode::UL);
  %}
  ins_pipe(pipe_class_dummy);
%}

// String IndexOfChar
instruct indexOfChar_U(iRegP haystack, iRegI haycnt, iRegI ch, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOfChar (Binary haystack haycnt) ch));
  effect(TEMP_DEF result, TEMP evenReg, TEMP oddReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfCharNode*)n)->encoding() == StrIntrinsicNode::U);
  ins_cost(200);
  format %{ "StringUTF16 IndexOfChar [0..$haycnt]($haystack), $ch -> $result" %}
  ins_encode %{
    __ string_indexof_char($result$$Register,
                           $haystack$$Register, $haycnt$$Register,
                           $ch$$Register, 0 /* unused, ch is in register */,
                           $oddReg$$Register, $evenReg$$Register, false /*is_byte*/);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOfChar_L(iRegP haystack, iRegI haycnt, iRegI ch, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOfChar (Binary haystack haycnt) ch));
  effect(TEMP_DEF result, TEMP evenReg, TEMP oddReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfCharNode*)n)->encoding() == StrIntrinsicNode::L);
  ins_cost(200);
  format %{ "StringLatin1 IndexOfChar [0..$haycnt]($haystack), $ch -> $result" %}
  ins_encode %{
    __ string_indexof_char($result$$Register,
                           $haystack$$Register, $haycnt$$Register,
                           $ch$$Register, 0 /* unused, ch is in register */,
                           $oddReg$$Register, $evenReg$$Register, true /*is_byte*/);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_imm1_U(iRegP haystack, iRegI haycnt, immP needle, immI_1 needlecnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
  effect(TEMP_DEF result, TEMP evenReg, TEMP oddReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU || ((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::none);
  ins_cost(200);
  format %{ "String IndexOf UL [0..$haycnt]($haystack), [0]($needle) -> $result" %}
  ins_encode %{
    immPOper *needleOper = (immPOper *)$needle;
    const TypeOopPtr *t = needleOper->type()->isa_oopptr();
    ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
    jchar chr;
#ifdef VM_LITTLE_ENDIAN
    Unimplemented();
#else
    chr = (((jchar)(unsigned char)needle_values->element_value(0).as_byte()) << 8) |
           ((jchar)(unsigned char)needle_values->element_value(1).as_byte());
#endif
    __ string_indexof_char($result$$Register,
                           $haystack$$Register, $haycnt$$Register,
                           noreg, chr,
                           $oddReg$$Register, $evenReg$$Register, false /*is_byte*/);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_imm1_L(iRegP haystack, iRegI haycnt, immP needle, immI_1 needlecnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
  effect(TEMP_DEF result, TEMP evenReg, TEMP oddReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
  ins_cost(200);
  format %{ "String IndexOf L [0..$haycnt]($haystack), [0]($needle) -> $result" %}
  ins_encode %{
    immPOper *needleOper = (immPOper *)$needle;
    const TypeOopPtr *t = needleOper->type()->isa_oopptr();
    ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
    jchar chr = (jchar)needle_values->element_value(0).as_byte();
    __ string_indexof_char($result$$Register,
                           $haystack$$Register, $haycnt$$Register,
                           noreg, chr,
                           $oddReg$$Register, $evenReg$$Register, true /*is_byte*/);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_imm1_UL(iRegP haystack, iRegI haycnt, immP needle, immI_1 needlecnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
  effect(TEMP_DEF result, TEMP evenReg, TEMP oddReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
  ins_cost(200);
  format %{ "String IndexOf UL [0..$haycnt]($haystack), [0]($needle) -> $result" %}
  ins_encode %{
    immPOper *needleOper = (immPOper *)$needle;
    const TypeOopPtr *t = needleOper->type()->isa_oopptr();
    ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
    jchar chr = (jchar)needle_values->element_value(0).as_byte();
    __ string_indexof_char($result$$Register,
                           $haystack$$Register, $haycnt$$Register,
                           noreg, chr,
                           $oddReg$$Register, $evenReg$$Register, false /*is_byte*/);
  %}
  ins_pipe(pipe_class_dummy);
%}

// String IndexOf
instruct indexOf_imm_U(iRegP haystack, rarg2RegI haycnt, iRegP needle, immI16 needlecntImm, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
  effect(TEMP_DEF result, USE_KILL haycnt, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU || ((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::none);
  ins_cost(250);
  format %{ "String IndexOf U [0..$needlecntImm]($needle) .in. [0..$haycnt]($haystack) -> $result" %}
  ins_encode %{
    __ string_indexof($result$$Register,
                      $haystack$$Register, $haycnt$$Register,
                      $needle$$Register, noreg, $needlecntImm$$constant,
                      $oddReg$$Register, $evenReg$$Register, StrIntrinsicNode::UU);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_imm_L(iRegP haystack, rarg2RegI haycnt, iRegP needle, immI16 needlecntImm, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
  effect(TEMP_DEF result, USE_KILL haycnt, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
  ins_cost(250);
  format %{ "String IndexOf L [0..$needlecntImm]($needle) .in. [0..$haycnt]($haystack) -> $result" %}
  ins_encode %{
    __ string_indexof($result$$Register,
                      $haystack$$Register, $haycnt$$Register,
                      $needle$$Register, noreg, $needlecntImm$$constant,
                      $oddReg$$Register, $evenReg$$Register, StrIntrinsicNode::LL);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_imm_UL(iRegP haystack, rarg2RegI haycnt, iRegP needle, immI16 needlecntImm, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
  effect(TEMP_DEF result, USE_KILL haycnt, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
  ins_cost(250);
  format %{ "String IndexOf UL [0..$needlecntImm]($needle) .in. [0..$haycnt]($haystack) -> $result" %}
  ins_encode %{
    __ string_indexof($result$$Register,
                      $haystack$$Register, $haycnt$$Register,
                      $needle$$Register, noreg, $needlecntImm$$constant,
                      $oddReg$$Register, $evenReg$$Register, StrIntrinsicNode::UL);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_U(iRegP haystack, rarg2RegI haycnt, iRegP needle, rarg5RegI needlecnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
  effect(TEMP_DEF result, USE_KILL haycnt, USE_KILL needlecnt, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU || ((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::none);
  ins_cost(300);
  format %{ "String IndexOf U [0..$needlecnt]($needle) .in. [0..$haycnt]($haystack) -> $result" %}
  ins_encode %{
    __ string_indexof($result$$Register,
                      $haystack$$Register, $haycnt$$Register,
                      $needle$$Register, $needlecnt$$Register, 0,
                      $oddReg$$Register, $evenReg$$Register, StrIntrinsicNode::UU);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_L(iRegP haystack, rarg2RegI haycnt, iRegP needle, rarg5RegI needlecnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
  effect(TEMP_DEF result, USE_KILL haycnt, USE_KILL needlecnt, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
  ins_cost(300);
  format %{ "String IndexOf L [0..$needlecnt]($needle) .in. [0..$haycnt]($haystack) -> $result" %}
  ins_encode %{
    __ string_indexof($result$$Register,
                      $haystack$$Register, $haycnt$$Register,
                      $needle$$Register, $needlecnt$$Register, 0,
                      $oddReg$$Register, $evenReg$$Register, StrIntrinsicNode::LL);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct indexOf_UL(iRegP haystack, rarg2RegI haycnt, iRegP needle, rarg5RegI needlecnt, iRegI result, roddRegL oddReg, revenRegL evenReg, flagsReg cr) %{
  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
  effect(TEMP_DEF result, USE_KILL haycnt, USE_KILL needlecnt, TEMP oddReg, TEMP evenReg, KILL cr); // R0, R1 are killed, too.
  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
  ins_cost(300);
  format %{ "String IndexOf UL [0..$needlecnt]($needle) .in. [0..$haycnt]($haystack) -> $result" %}
  ins_encode %{
    __ string_indexof($result$$Register,
                      $haystack$$Register, $haycnt$$Register,
                      $needle$$Register, $needlecnt$$Register, 0,
                      $oddReg$$Register, $evenReg$$Register, StrIntrinsicNode::UL);
  %}
  ins_pipe(pipe_class_dummy);
%}

// char[] to byte[] compression
instruct string_compress(iRegP src, iRegP dst, iRegI result, iRegI len, iRegI tmp, flagsReg cr) %{
  match(Set result (StrCompressedCopy src (Binary dst len)));
  effect(TEMP_DEF result, TEMP tmp, KILL cr); // R0, R1 are killed, too.
  ins_cost(300);
  format %{ "String Compress $src->$dst($len) -> $result" %}
  ins_encode %{
    __ string_compress($result$$Register, $src$$Register, $dst$$Register, $len$$Register,
                       $tmp$$Register, false);
  %}
  ins_pipe(pipe_class_dummy);
%}

// byte[] to char[] inflation. trot implementation is shorter, but slower than the unrolled icm(h) loop.
//instruct string_inflate_trot(Universe dummy, iRegP src, revenRegP dst, roddRegI len, iRegI tmp, flagsReg cr) %{
//  match(Set dummy (StrInflatedCopy src (Binary dst len)));
//  effect(USE_KILL dst, USE_KILL len, TEMP tmp, KILL cr); // R0, R1 are killed, too.
//  predicate(VM_Version::has_ETF2Enhancements());
//  ins_cost(300);
//  format %{ "String Inflate (trot) $dst,$src($len)" %}
//  ins_encode %{
//    __ string_inflate_trot($src$$Register, $dst$$Register, $len$$Register, $tmp$$Register);
//  %}
//  ins_pipe(pipe_class_dummy);
//%}

// byte[] to char[] inflation
instruct string_inflate(Universe dummy, iRegP src, iRegP dst, iRegI len, iRegI tmp, flagsReg cr) %{
  match(Set dummy (StrInflatedCopy src (Binary dst len)));
  effect(TEMP tmp, KILL cr); // R0, R1 are killed, too.
  ins_cost(300);
  format %{ "String Inflate $src->$dst($len)" %}
  ins_encode %{
    __ string_inflate($src$$Register, $dst$$Register, $len$$Register, $tmp$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

// byte[] to char[] inflation
instruct string_inflate_const(Universe dummy, iRegP src, iRegP dst, iRegI tmp, immI len, flagsReg cr) %{
  match(Set dummy (StrInflatedCopy src (Binary dst len)));
  effect(TEMP tmp, KILL cr); // R0, R1 are killed, too.
  ins_cost(300);
  format %{ "String Inflate (constLen) $src->$dst($len)" %}
  ins_encode %{
    __ string_inflate_const($src$$Register, $dst$$Register, $tmp$$Register, $len$$constant);
  %}
  ins_pipe(pipe_class_dummy);
%}

// StringCoding.java intrinsics
instruct has_negatives(rarg5RegP ary1, iRegI len, iRegI result, roddRegI oddReg, revenRegI evenReg, iRegI tmp, flagsReg cr) %{
  match(Set result (HasNegatives ary1 len));
  effect(TEMP_DEF result, USE_KILL ary1, TEMP oddReg, TEMP evenReg, TEMP tmp, KILL cr); // R0, R1 are killed, too.
  ins_cost(300);
  format %{ "has negatives byte[] $ary1($len) -> $result" %}
  ins_encode %{
    __ has_negatives($result$$Register, $ary1$$Register, $len$$Register,
                     $oddReg$$Register, $evenReg$$Register, $tmp$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

// encode char[] to byte[] in ISO_8859_1
instruct encode_iso_array(iRegP src, iRegP dst, iRegI result, iRegI len, iRegI tmp, flagsReg cr) %{
  predicate(!((EncodeISOArrayNode*)n)->is_ascii());
  match(Set result (EncodeISOArray src (Binary dst len)));
  effect(TEMP_DEF result, TEMP tmp, KILL cr); // R0, R1 are killed, too.
  ins_cost(300);
  format %{ "Encode array $src->$dst($len) -> $result" %}
  ins_encode %{
    __ string_compress($result$$Register, $src$$Register, $dst$$Register, $len$$Register,
                       $tmp$$Register, true);
  %}
  ins_pipe(pipe_class_dummy);
%}


//----------PEEPHOLE RULES-----------------------------------------------------
// These must follow all instruction definitions as they use the names
// defined in the instructions definitions.
//
// peepmatch (root_instr_name [preceeding_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 (leaI_eReg_immI(0.dst 1.src 0.src));
// %}
//
// Implementation no longer uses movX instructions since
// machine-independent system no longer uses CopyX nodes.
//
// peephole %{
//   peepmatch (incI_eReg movI);
//   peepconstraint (0.dst == 1.dst);
//   peepreplace (leaI_eReg_immI(0.dst 1.src 0.src));
// %}
//
// peephole %{
//   peepmatch (decI_eReg movI);
//   peepconstraint (0.dst == 1.dst);
//   peepreplace (leaI_eReg_immI(0.dst 1.src 0.src));
// %}
//
// peephole %{
//   peepmatch (addI_eReg_imm movI);
//   peepconstraint (0.dst == 1.dst);
//   peepreplace (leaI_eReg_immI(0.dst 1.src 0.src));
// %}
//
// peephole %{
//   peepmatch (addP_eReg_imm movP);
//   peepconstraint (0.dst == 1.dst);
//   peepreplace (leaP_eReg_immI(0.dst 1.src 0.src));
// %}


//  This peephole rule does not work, probably because ADLC can't handle two effects:
//  Effect 1 is defining 0.op1 and effect 2 is setting CC
// condense a load from memory and subsequent test for zero
// into a single, more efficient ICM instruction.
// peephole %{
//   peepmatch (compI_iReg_imm0 loadI);
//   peepconstraint (1.dst == 0.op1);
//   peepreplace (loadtest15_iReg_mem(0.op1 0.op1 1.mem));
// %}

// // 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));
%}

peephole %{
  peepmatch (loadL storeL);
  peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
  peepreplace (storeL(1.mem 1.mem 1.src));
%}

peephole %{
  peepmatch (loadP storeP);
  peepconstraint (1.src == 0.dst, 1.dst == 0.mem);
  peepreplace (storeP(1.dst 1.dst 1.src));
%}

//----------SUPERWORD RULES---------------------------------------------------

//  Expand rules for special cases

instruct expand_storeF(stackSlotF mem, regF src) %{
  // No match rule, false predicate, for expand only.
  effect(DEF mem, USE src);
  predicate(false);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "STE      $src,$mem\t # replicate(float2stack)" %}
  opcode(STE_ZOPC, STE_ZOPC);
  ins_encode(z_form_rt_mem(src, mem));
  ins_pipe(pipe_class_dummy);
%}

instruct expand_LoadLogical_I2L(iRegL dst, stackSlotF mem) %{
  // No match rule, false predicate, for expand only.
  effect(DEF dst, USE mem);
  predicate(false);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LLGF     $dst,$mem\t # replicate(stack2reg(unsigned))" %}
  opcode(LLGF_ZOPC, LLGF_ZOPC);
  ins_encode(z_form_rt_mem(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar int to packed int values (8 Bytes)
instruct expand_Repl2I_reg(iRegL dst, iRegL src) %{
  // Dummy match rule, false predicate, for expand only.
  match(Set dst (ConvI2L src));
  predicate(false);
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "REPLIC2F $dst,$src\t # replicate(pack2F)" %}
  ins_encode %{
    if ($dst$$Register == $src$$Register) {
      __ z_sllg(Z_R0_scratch, $src$$Register, 64-32);
      __ z_ogr($dst$$Register, Z_R0_scratch);
    }  else {
      __ z_sllg($dst$$Register, $src$$Register, 64-32);
      __ z_ogr( $dst$$Register, $src$$Register);
    }
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replication

// Exploit rotate_then_insert, if available
// Replicate scalar byte to packed byte values (8 Bytes).
instruct Repl8B_reg_risbg(iRegL dst, iRegI src, flagsReg cr) %{
  match(Set dst (ReplicateB src));
  effect(KILL cr);
  predicate((n->as_Vector()->length() == 8));
  format %{ "REPLIC8B $dst,$src\t # pack8B" %}
  ins_encode %{
    if ($dst$$Register != $src$$Register) {
      __ z_lgr($dst$$Register, $src$$Register);
    }
    __ rotate_then_insert($dst$$Register, $dst$$Register, 48, 55,  8, false);
    __ rotate_then_insert($dst$$Register, $dst$$Register, 32, 47, 16, false);
    __ rotate_then_insert($dst$$Register, $dst$$Register,  0, 31, 32, false);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar byte to packed byte values (8 Bytes).
instruct Repl8B_imm(iRegL dst, immB_n0m1 src) %{
  match(Set dst (ReplicateB src));
  predicate(n->as_Vector()->length() == 8);
  ins_should_rematerialize(true);
  format %{ "REPLIC8B $dst,$src\t # pack8B imm" %}
  ins_encode %{
    int64_t  Isrc8 = $src$$constant & 0x000000ff;
    int64_t Isrc16 =  Isrc8 <<  8 |  Isrc8;
    int64_t Isrc32 = Isrc16 << 16 | Isrc16;
    assert(Isrc8 != 0x000000ff && Isrc8 != 0, "should be handled by other match rules.");

    __ z_llilf($dst$$Register, Isrc32);
    __ z_iihf($dst$$Register, Isrc32);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar byte to packed byte values (8 Bytes).
instruct Repl8B_imm0(iRegL dst, immI_0 src) %{
  match(Set dst (ReplicateB src));
  predicate(n->as_Vector()->length() == 8);
  ins_should_rematerialize(true);
  format %{ "REPLIC8B $dst,$src\t # pack8B imm0" %}
  ins_encode %{ __ z_laz($dst$$Register, 0, Z_R0); %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar byte to packed byte values (8 Bytes).
instruct Repl8B_immm1(iRegL dst, immB_minus1 src) %{
  match(Set dst (ReplicateB src));
  predicate(n->as_Vector()->length() == 8);
  ins_should_rematerialize(true);
  format %{ "REPLIC8B $dst,$src\t # pack8B immm1" %}
  ins_encode %{ __ z_lghi($dst$$Register, -1); %}
  ins_pipe(pipe_class_dummy);
%}

// Exploit rotate_then_insert, if available
// Replicate scalar short to packed short values (8 Bytes).
instruct Repl4S_reg_risbg(iRegL dst, iRegI src, flagsReg cr) %{
  match(Set dst (ReplicateS src));
  effect(KILL cr);
  predicate((n->as_Vector()->length() == 4));
  format %{ "REPLIC4S $dst,$src\t # pack4S" %}
  ins_encode %{
    if ($dst$$Register != $src$$Register) {
      __ z_lgr($dst$$Register, $src$$Register);
    }
    __ rotate_then_insert($dst$$Register, $dst$$Register, 32, 47, 16, false);
    __ rotate_then_insert($dst$$Register, $dst$$Register,  0, 31, 32, false);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar short to packed short values (8 Bytes).
instruct Repl4S_imm(iRegL dst, immS_n0m1 src) %{
  match(Set dst (ReplicateS src));
  predicate(n->as_Vector()->length() == 4);
  ins_should_rematerialize(true);
  format %{ "REPLIC4S $dst,$src\t # pack4S imm" %}
  ins_encode %{
    int64_t Isrc16 = $src$$constant & 0x0000ffff;
    int64_t Isrc32 = Isrc16 << 16 | Isrc16;
    assert(Isrc16 != 0x0000ffff && Isrc16 != 0, "Repl4S_imm: (src == " INT64_FORMAT
           ") should be handled by other match rules.", $src$$constant);

    __ z_llilf($dst$$Register, Isrc32);
    __ z_iihf($dst$$Register, Isrc32);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar short to packed short values (8 Bytes).
instruct Repl4S_imm0(iRegL dst, immI_0 src) %{
  match(Set dst (ReplicateS src));
  predicate(n->as_Vector()->length() == 4);
  ins_should_rematerialize(true);
  format %{ "REPLIC4S $dst,$src\t # pack4S imm0" %}
  ins_encode %{ __ z_laz($dst$$Register, 0, Z_R0); %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar short to packed short values (8 Bytes).
instruct Repl4S_immm1(iRegL dst, immS_minus1 src) %{
  match(Set dst (ReplicateS src));
  predicate(n->as_Vector()->length() == 4);
  ins_should_rematerialize(true);
  format %{ "REPLIC4S $dst,$src\t # pack4S immm1" %}
  ins_encode %{ __ z_lghi($dst$$Register, -1); %}
  ins_pipe(pipe_class_dummy);
%}

// Exploit rotate_then_insert, if available.
// Replicate scalar int to packed int values (8 Bytes).
instruct Repl2I_reg_risbg(iRegL dst, iRegI src, flagsReg cr) %{
  match(Set dst (ReplicateI src));
  effect(KILL cr);
  predicate((n->as_Vector()->length() == 2));
  format %{ "REPLIC2I $dst,$src\t # pack2I" %}
  ins_encode %{
    if ($dst$$Register != $src$$Register) {
      __ z_lgr($dst$$Register, $src$$Register);
    }
    __ rotate_then_insert($dst$$Register, $dst$$Register, 0, 31, 32, false);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar int to packed int values (8 Bytes).
instruct Repl2I_imm(iRegL dst, immI_n0m1 src) %{
  match(Set dst (ReplicateI src));
  predicate(n->as_Vector()->length() == 2);
  ins_should_rematerialize(true);
  format %{ "REPLIC2I $dst,$src\t # pack2I imm" %}
  ins_encode %{
    int64_t Isrc32 = $src$$constant;
    assert(Isrc32 != -1 && Isrc32 != 0, "should be handled by other match rules.");

    __ z_llilf($dst$$Register, Isrc32);
    __ z_iihf($dst$$Register, Isrc32);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar int to packed int values (8 Bytes).
instruct Repl2I_imm0(iRegL dst, immI_0 src) %{
  match(Set dst (ReplicateI src));
  predicate(n->as_Vector()->length() == 2);
  ins_should_rematerialize(true);
  format %{ "REPLIC2I $dst,$src\t # pack2I imm0" %}
  ins_encode %{ __ z_laz($dst$$Register, 0, Z_R0); %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar int to packed int values (8 Bytes).
instruct Repl2I_immm1(iRegL dst, immI_minus1 src) %{
  match(Set dst (ReplicateI src));
  predicate(n->as_Vector()->length() == 2);
  ins_should_rematerialize(true);
  format %{ "REPLIC2I $dst,$src\t # pack2I immm1" %}
  ins_encode %{ __ z_lghi($dst$$Register, -1); %}
  ins_pipe(pipe_class_dummy);
%}

//

instruct Repl2F_reg_indirect(iRegL dst, regF src, flagsReg cr) %{
  match(Set dst (ReplicateF src));
  effect(KILL cr);
  predicate(!VM_Version::has_FPSupportEnhancements() && n->as_Vector()->length() == 2);
  format %{ "REPLIC2F $dst,$src\t # pack2F indirect" %}
  expand %{
    stackSlotF tmp;
    iRegL      tmp2;
    expand_storeF(tmp, src);
    expand_LoadLogical_I2L(tmp2, tmp);
    expand_Repl2I_reg(dst, tmp2);
  %}
%}

// Replicate scalar float to packed float values in GREG (8 Bytes).
instruct Repl2F_reg_direct(iRegL dst, regF src, flagsReg cr) %{
  match(Set dst (ReplicateF src));
  effect(KILL cr);
  predicate(VM_Version::has_FPSupportEnhancements() && n->as_Vector()->length() == 2);
  format %{ "REPLIC2F $dst,$src\t # pack2F direct" %}
  ins_encode %{
    assert(VM_Version::has_FPSupportEnhancements(), "encoder should never be called on old H/W");
    __ z_lgdr($dst$$Register, $src$$FloatRegister);

    __ z_srlg(Z_R0_scratch, $dst$$Register, 32);  // Floats are left-justified in 64bit reg.
    __ z_iilf($dst$$Register, 0);                 // Save a "result not ready" stall.
    __ z_ogr($dst$$Register, Z_R0_scratch);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar float immediate to packed float values in GREG (8 Bytes).
instruct Repl2F_imm(iRegL dst, immF src) %{
  match(Set dst (ReplicateF src));
  predicate(n->as_Vector()->length() == 2);
  ins_should_rematerialize(true);
  format %{ "REPLIC2F $dst,$src\t # pack2F imm" %}
  ins_encode %{
    union {
      int   Isrc32;
      float Fsrc32;
    };
    Fsrc32 = $src$$constant;
    __ z_llilf($dst$$Register, Isrc32);
    __ z_iihf($dst$$Register, Isrc32);
  %}
  ins_pipe(pipe_class_dummy);
%}

// Replicate scalar float immediate zeroes to packed float values in GREG (8 Bytes).
// Do this only for 'real' zeroes, especially don't loose sign of negative zeroes.
instruct Repl2F_imm0(iRegL dst, immFp0 src) %{
  match(Set dst (ReplicateF src));
  predicate(n->as_Vector()->length() == 2);
  ins_should_rematerialize(true);
  format %{ "REPLIC2F $dst,$src\t # pack2F imm0" %}
  ins_encode %{ __ z_laz($dst$$Register, 0, Z_R0); %}
  ins_pipe(pipe_class_dummy);
%}

// Load/Store vector

// Store Aligned Packed Byte register to memory (8 Bytes).
instruct storeA8B(memory mem, iRegL src) %{
  match(Set mem (StoreVector mem src));
  predicate(n->as_StoreVector()->memory_size() == 8);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "STG     $src,$mem\t # ST(packed8B)" %}
  opcode(STG_ZOPC, STG_ZOPC);
  ins_encode(z_form_rt_mem_opt(src, mem));
  ins_pipe(pipe_class_dummy);
%}

instruct loadV8(iRegL dst, memory mem) %{
  match(Set dst (LoadVector mem));
  predicate(n->as_LoadVector()->memory_size() == 8);
  ins_cost(MEMORY_REF_COST);
  // TODO: s390 port size(VARIABLE_SIZE);
  format %{ "LG      $dst,$mem\t # L(packed8B)" %}
  opcode(LG_ZOPC, LG_ZOPC);
  ins_encode(z_form_rt_mem_opt(dst, mem));
  ins_pipe(pipe_class_dummy);
%}

// Reinterpret: only one vector size used
instruct reinterpret(iRegL dst) %{
  match(Set dst (VectorReinterpret dst));
  ins_cost(0);
  format %{ "reinterpret $dst" %}
  ins_encode( /*empty*/ );
  ins_pipe(pipe_class_dummy);
%}

//----------POPULATION COUNT RULES--------------------------------------------

// Byte reverse

instruct bytes_reverse_int(iRegI dst, iRegI src) %{
  match(Set dst (ReverseBytesI src));
  predicate(UseByteReverseInstruction);  // See Matcher::match_rule_supported
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "LRVR    $dst,$src\t # byte reverse int" %}
  opcode(LRVR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

instruct bytes_reverse_long(iRegL dst, iRegL src) %{
  match(Set dst (ReverseBytesL src));
  predicate(UseByteReverseInstruction);  // See Matcher::match_rule_supported
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "LRVGR   $dst,$src\t # byte reverse long" %}
  opcode(LRVGR_ZOPC);
  ins_encode(z_rreform(dst, src));
  ins_pipe(pipe_class_dummy);
%}

// Leading zeroes

// The instruction FLOGR (Find Leftmost One in Grande (64bit) Register)
// returns the bit position of the leftmost 1 in the 64bit source register.
// As the bits are numbered from left to right (0..63), the returned
// position index is equivalent to the number of leading zeroes.
// If no 1-bit is found (i.e. the regsiter contains zero), the instruction
// returns position 64. That's exactly what we need.

instruct countLeadingZerosI(revenRegI dst, iRegI src, roddRegI tmp, flagsReg cr) %{
  match(Set dst (CountLeadingZerosI src));
  effect(KILL tmp, KILL cr);
  ins_cost(3 * DEFAULT_COST);
  size(14);
  format %{ "SLLG    $dst,$src,32\t # no need to always count 32 zeroes first\n\t"
            "IILH    $dst,0x8000 \t # insert \"stop bit\" to force result 32 for zero src.\n\t"
            "FLOGR   $dst,$dst"
         %}
  ins_encode %{
    // Performance experiments indicate that "FLOGR" is using some kind of
    // iteration to find the leftmost "1" bit.
    //
    // The prior implementation zero-extended the 32-bit argument to 64 bit,
    // thus forcing "FLOGR" to count 32 bits of which we know they are zero.
    // We could gain measurable speedup in micro benchmark:
    //
    //               leading   trailing
    //   z10:   int     2.04       1.68
    //         long     1.00       1.02
    //   z196:  int     0.99       1.23
    //         long     1.00       1.11
    //
    // By shifting the argument into the high-word instead of zero-extending it.
    // The add'l branch on condition (taken for a zero argument, very infrequent,
    // good prediction) is well compensated for by the savings.
    //
    // We leave the previous implementation in for some time in the future when
    // the "FLOGR" instruction may become less iterative.

    // Version 2: shows 62%(z9), 204%(z10), -1%(z196) improvement over original
    __ z_sllg($dst$$Register, $src$$Register, 32); // No need to always count 32 zeroes first.
    __ z_iilh($dst$$Register, 0x8000);   // Insert "stop bit" to force result 32 for zero src.
    __ z_flogr($dst$$Register, $dst$$Register);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct countLeadingZerosL(revenRegI dst, iRegL src, roddRegI tmp, flagsReg cr) %{
  match(Set dst (CountLeadingZerosL src));
  effect(KILL tmp, KILL cr);
  ins_cost(DEFAULT_COST);
  size(4);
  format %{ "FLOGR   $dst,$src \t # count leading zeros (long)\n\t" %}
  ins_encode %{ __ z_flogr($dst$$Register, $src$$Register); %}
  ins_pipe(pipe_class_dummy);
%}

// trailing zeroes

// We transform the trailing zeroes problem to a leading zeroes problem
// such that can use the FLOGR instruction to our advantage.

// With
//   tmp1 = src - 1
// we flip all trailing zeroes to ones and the rightmost one to zero.
// All other bits remain unchanged.
// With the complement
//   tmp2 = ~src
// we get all ones in the trailing zeroes positions. Thus,
//   tmp3 = tmp1 & tmp2
// yields ones in the trailing zeroes positions and zeroes elsewhere.
// Now we can apply FLOGR and get 64-(trailing zeroes).
instruct countTrailingZerosI(revenRegI dst, iRegI src, roddRegI tmp, flagsReg cr) %{
  match(Set dst (CountTrailingZerosI src));
  effect(TEMP_DEF dst, TEMP tmp, KILL cr);
  ins_cost(8 * DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);  // Emitted code depends on PreferLAoverADD being on/off.
  format %{ "LLGFR   $dst,$src  \t # clear upper 32 bits (we are dealing with int)\n\t"
            "LCGFR   $tmp,$src  \t # load 2's complement (32->64 bit)\n\t"
            "AGHI    $dst,-1    \t # tmp1 = src-1\n\t"
            "AGHI    $tmp,-1    \t # tmp2 = -src-1 = ~src\n\t"
            "NGR     $dst,$tmp  \t # tmp3 = tmp1&tmp2\n\t"
            "FLOGR   $dst,$dst  \t # count trailing zeros (int)\n\t"
            "AHI     $dst,-64   \t # tmp4 = 64-(trailing zeroes)-64\n\t"
            "LCR     $dst,$dst  \t # res = -tmp4"
         %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    // Rtmp only needed for for zero-argument shortcut. With kill effect in
    // match rule Rsrc = roddReg would be possible, saving one register.
    Register Rtmp = $tmp$$Register;

    assert_different_registers(Rdst, Rsrc, Rtmp);

    // Algorithm:
    // - Isolate the least significant (rightmost) set bit using (src & (-src)).
    //   All other bits in the result are zero.
    // - Find the "leftmost one" bit position in the single-bit result from previous step.
    // - 63-("leftmost one" bit position) gives the # of trailing zeros.

    // Version 2: shows 79%(z9), 68%(z10), 23%(z196) improvement over original.
    Label done;
    __ load_const_optimized(Rdst, 32); // Prepare for shortcut (zero argument), result will be 32.
    __ z_lcgfr(Rtmp, Rsrc);
    __ z_bre(done);                    // Taken very infrequently, good prediction, no BHT entry.

    __ z_nr(Rtmp, Rsrc);               // (src) & (-src) leaves nothing but least significant bit.
    __ z_ahi(Rtmp,  -1);               // Subtract one to fill all trailing zero positions with ones.
                                       // Use 32bit op to prevent borrow propagation (case Rdst = 0x80000000)
                                       // into upper half of reg. Not relevant with sllg below.
    __ z_sllg(Rdst, Rtmp, 32);         // Shift interesting contents to upper half of register.
    __ z_bre(done);                    // Shortcut for argument = 1, result will be 0.
                                       // Depends on CC set by ahi above.
                                       // Taken very infrequently, good prediction, no BHT entry.
                                       // Branch delayed to have Rdst set correctly (Rtmp == 0(32bit)
                                       // after SLLG Rdst == 0(64bit)).
    __ z_flogr(Rdst, Rdst);            // Kills tmp which is the oddReg for dst.
    __ add2reg(Rdst,  -32);            // 32-pos(leftmost1) is #trailing zeros
    __ z_lcgfr(Rdst, Rdst);            // Provide 64bit result at no cost.
    __ bind(done);
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct countTrailingZerosL(revenRegI dst, iRegL src, roddRegL tmp, flagsReg cr) %{
  match(Set dst (CountTrailingZerosL src));
  effect(TEMP_DEF dst, KILL tmp, KILL cr);
  ins_cost(8 * DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);  // Emitted code depends on PreferLAoverADD being on/off.
  format %{ "LCGR    $dst,$src  \t # preserve src\n\t"
            "NGR     $dst,$src  \t #\n\t"
            "AGHI    $dst,-1    \t # tmp1 = src-1\n\t"
            "FLOGR   $dst,$dst  \t # count trailing zeros (long), kill $tmp\n\t"
            "AHI     $dst,-64   \t # tmp4 = 64-(trailing zeroes)-64\n\t"
            "LCR     $dst,$dst  \t #"
         %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    assert_different_registers(Rdst, Rsrc); // Rtmp == Rsrc allowed.

    // New version: shows 5%(z9), 2%(z10), 11%(z196) improvement over original.
    __ z_lcgr(Rdst, Rsrc);
    __ z_ngr(Rdst, Rsrc);
    __ add2reg(Rdst,   -1);
    __ z_flogr(Rdst, Rdst); // Kills tmp which is the oddReg for dst.
    __ add2reg(Rdst,  -64);
    __ z_lcgfr(Rdst, Rdst); // Provide 64bit result at no cost.
  %}
  ins_pipe(pipe_class_dummy);
%}


// bit count

instruct popCountI(iRegI dst, iRegI src, iRegI tmp, flagsReg cr) %{
  match(Set dst (PopCountI src));
  effect(TEMP_DEF dst, TEMP tmp, KILL cr);
  predicate(UsePopCountInstruction && VM_Version::has_PopCount());
  ins_cost(DEFAULT_COST);
  size(24);
  format %{ "POPCNT  $dst,$src\t # pop count int" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Register Rtmp = $tmp$$Register;

    // Prefer compile-time assertion over run-time SIGILL.
    assert(VM_Version::has_PopCount(), "bad predicate for countLeadingZerosI");
    assert_different_registers(Rdst, Rtmp);

    // Version 2: shows 10%(z196) improvement over original.
    __ z_popcnt(Rdst, Rsrc);
    __ z_srlg(Rtmp, Rdst, 16); // calc  byte4+byte6 and byte5+byte7
    __ z_alr(Rdst, Rtmp);      //   into byte6 and byte7
    __ z_srlg(Rtmp, Rdst,  8); // calc (byte4+byte6) + (byte5+byte7)
    __ z_alr(Rdst, Rtmp);      //   into byte7
    __ z_llgcr(Rdst, Rdst);    // zero-extend sum
  %}
  ins_pipe(pipe_class_dummy);
%}

instruct popCountL(iRegI dst, iRegL src, iRegL tmp, flagsReg cr) %{
  match(Set dst (PopCountL src));
  effect(TEMP_DEF dst, TEMP tmp, KILL cr);
  predicate(UsePopCountInstruction && VM_Version::has_PopCount());
  ins_cost(DEFAULT_COST);
  // TODO: s390 port size(FIXED_SIZE);
  format %{ "POPCNT  $dst,$src\t # pop count long" %}
  ins_encode %{
    Register Rdst = $dst$$Register;
    Register Rsrc = $src$$Register;
    Register Rtmp = $tmp$$Register;

    // Prefer compile-time assertion over run-time SIGILL.
    assert(VM_Version::has_PopCount(), "bad predicate for countLeadingZerosI");
    assert_different_registers(Rdst, Rtmp);

    // Original version. Using LA instead of algr seems to be a really bad idea (-35%).
    __ z_popcnt(Rdst, Rsrc);
    __ z_ahhlr(Rdst, Rdst, Rdst);
    __ z_sllg(Rtmp, Rdst, 16);
    __ z_algr(Rdst, Rtmp);
    __ z_sllg(Rtmp, Rdst,  8);
    __ z_algr(Rdst, Rtmp);
    __ z_srlg(Rdst, Rdst, 56);
  %}
  ins_pipe(pipe_class_dummy);
%}

//----------SMARTSPILL RULES---------------------------------------------------
// These must follow all instruction definitions as they use the names
// defined in the instructions definitions.

// ============================================================================
// TYPE PROFILING RULES