/*

 * Copyright (c) 2002, 2019, Oracle and/or its affiliates. All rights reserved.

 * Copyright (c) 2012, 2019, 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.

 *

 */

#ifndef CPU_PPC_MACROASSEMBLER_PPC_HPP

#define CPU_PPC_MACROASSEMBLER_PPC_HPP

#include "asm/assembler.hpp"

#include "oops/accessDecorators.hpp"

#include "runtime/rtmLocking.hpp"

#include "utilities/macros.hpp"

// MacroAssembler extends Assembler by a few frequently used macros.

class ciTypeArray;

class MacroAssembler: public Assembler {

 public:

  MacroAssembler(CodeBuffer* code) : Assembler(code) {}

  //

  // Optimized instruction emitters

  //

  inline static int largeoffset_si16_si16_hi(int si31) { return (si31 + (1<<15)) >> 16; }

  inline static int largeoffset_si16_si16_lo(int si31) { return si31 - (((si31 + (1<<15)) >> 16) << 16); }

  // load d = *[a+si31]

  // Emits several instructions if the offset is not encodable in one instruction.

  void ld_largeoffset_unchecked(Register d, int si31, Register a, int emit_filler_nop);

  void ld_largeoffset          (Register d, int si31, Register a, int emit_filler_nop);

  inline static bool is_ld_largeoffset(address a);

  inline static int get_ld_largeoffset_offset(address a);

  inline void round_to(Register r, int modulus);

  // Load/store with type given by parameter.

  void load_sized_value( Register dst, RegisterOrConstant offs, Register base, size_t size_in_bytes, bool is_signed);

  void store_sized_value(Register dst, RegisterOrConstant offs, Register base, size_t size_in_bytes);

  // Move register if destination register and target register are different

  inline void mr_if_needed(Register rd, Register rs);

  inline void fmr_if_needed(FloatRegister rd, FloatRegister rs);

  // This is dedicated for emitting scheduled mach nodes. For better

  // readability of the ad file I put it here.

  // Endgroups are not needed if

  //  - the scheduler is off

  //  - the scheduler found that there is a natural group end, in that

  //    case it reduced the size of the instruction used in the test

  //    yielding 'needed'.

  inline void endgroup_if_needed(bool needed);

  // Memory barriers.

  inline void membar(int bits);

  inline void release();

  inline void acquire();

  inline void fence();

  // nop padding

  void align(int modulus, int max = 252, int rem = 0);

  //

  // Constants, loading constants, TOC support

  //

  // Address of the global TOC.

  inline static address global_toc();

  // Offset of given address to the global TOC.

  inline static int offset_to_global_toc(const address addr);

  // Address of TOC of the current method.

  inline address method_toc();

  // Offset of given address to TOC of the current method.

  inline int offset_to_method_toc(const address addr);

  // Global TOC.

  void calculate_address_from_global_toc(Register dst, address addr,

                                         bool hi16 = true, bool lo16 = true,

                                         bool add_relocation = true, bool emit_dummy_addr = false);

  inline void calculate_address_from_global_toc_hi16only(Register dst, address addr) {

    calculate_address_from_global_toc(dst, addr, true, false);

  };

  inline void calculate_address_from_global_toc_lo16only(Register dst, address addr) {

    calculate_address_from_global_toc(dst, addr, false, true);

  };

  inline static bool is_calculate_address_from_global_toc_at(address a, address bound);

  // Returns address of first instruction in sequence.

  static address patch_calculate_address_from_global_toc_at(address a, address bound, address addr);

  static address get_address_of_calculate_address_from_global_toc_at(address a, address addr);

#ifdef _LP64

  // Patch narrow oop constant.

  inline static bool is_set_narrow_oop(address a, address bound);

  // Returns address of first instruction in sequence.

  static address patch_set_narrow_oop(address a, address bound, narrowOop data);

  static narrowOop get_narrow_oop(address a, address bound);

#endif

  inline static bool is_load_const_at(address a);

  // Emits an oop const to the constant pool, loads the constant, and

  // sets a relocation info with address current_pc.

  // Returns true if successful.

  bool load_const_from_method_toc(Register dst, AddressLiteral& a, Register toc, bool fixed_size = false);

  static bool is_load_const_from_method_toc_at(address a);

  static int get_offset_of_load_const_from_method_toc_at(address a);

  // Get the 64 bit constant from a `load_const' sequence.

  static long get_const(address load_const);

  // Patch the 64 bit constant of a `load_const' sequence. This is a

  // low level procedure. It neither flushes the instruction cache nor

  // is it atomic.

  static void patch_const(address load_const, long x);

  // Metadata in code that we have to keep track of.

  AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index

  AddressLiteral constant_metadata_address(Metadata* obj); // find_index

  // Oops used directly in compiled code are stored in the constant pool,

  // and loaded from there.

  // Allocate new entry for oop in constant pool. Generate relocation.

  AddressLiteral allocate_oop_address(jobject obj);

  // Find oop obj in constant pool. Return relocation with it's index.

  AddressLiteral constant_oop_address(jobject obj);

  // Find oop in constant pool and emit instructions to load it.

  // Uses constant_oop_address.

  inline void set_oop_constant(jobject obj, Register d);

  // Same as load_address.

  inline void set_oop         (AddressLiteral obj_addr, Register d);

  // Read runtime constant:  Issue load if constant not yet established,

  // else use real constant.

  virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,

                                                Register tmp,

                                                int offset);

  //

  // branch, jump

  //

  inline void pd_patch_instruction(address branch, address target, const char* file, int line);

  NOT_PRODUCT(static void pd_print_patched_instruction(address branch);)

  // Conditional far branch for destinations encodable in 24+2 bits.

  // Same interface as bc, e.g. no inverse boint-field.

  enum {

    bc_far_optimize_not         = 0,

    bc_far_optimize_on_relocate = 1

  };

  // optimize: flag for telling the conditional far branch to optimize

  //           itself when relocated.

  void bc_far(int boint, int biint, Label& dest, int optimize);

  void bc_far_optimized(int boint, int biint, Label& dest); // 1 or 2 instructions

  // Relocation of conditional far branches.

  static bool    is_bc_far_at(address instruction_addr);

  static address get_dest_of_bc_far_at(address instruction_addr);

  static void    set_dest_of_bc_far_at(address instruction_addr, address dest);

 private:

  static bool inline is_bc_far_variant1_at(address instruction_addr);

  static bool inline is_bc_far_variant2_at(address instruction_addr);

  static bool inline is_bc_far_variant3_at(address instruction_addr);

 public:

  // Convenience bc_far versions.

  inline void blt_far(ConditionRegister crx, Label& L, int optimize);

  inline void bgt_far(ConditionRegister crx, Label& L, int optimize);

  inline void beq_far(ConditionRegister crx, Label& L, int optimize);

  inline void bso_far(ConditionRegister crx, Label& L, int optimize);

  inline void bge_far(ConditionRegister crx, Label& L, int optimize);

  inline void ble_far(ConditionRegister crx, Label& L, int optimize);

  inline void bne_far(ConditionRegister crx, Label& L, int optimize);

  inline void bns_far(ConditionRegister crx, Label& L, int optimize);

  // Emit, identify and patch a NOT mt-safe patchable 64 bit absolute call/jump.

 private:

  enum {

    bxx64_patchable_instruction_count = (2/*load_codecache_const*/ + 3/*5load_const*/ + 1/*mtctr*/ + 1/*bctrl*/),

    bxx64_patchable_size              = bxx64_patchable_instruction_count * BytesPerInstWord,

    bxx64_patchable_ret_addr_offset   = bxx64_patchable_size

  };

  void bxx64_patchable(address target, relocInfo::relocType rt, bool link);

  static bool is_bxx64_patchable_at(            address instruction_addr, bool link);

  // Does the instruction use a pc-relative encoding of the destination?

  static bool is_bxx64_patchable_pcrelative_at( address instruction_addr, bool link);

  static bool is_bxx64_patchable_variant1_at(   address instruction_addr, bool link);

  // Load destination relative to global toc.

  static bool is_bxx64_patchable_variant1b_at(  address instruction_addr, bool link);

  static bool is_bxx64_patchable_variant2_at(   address instruction_addr, bool link);

  static void set_dest_of_bxx64_patchable_at(   address instruction_addr, address target, bool link);

  static address get_dest_of_bxx64_patchable_at(address instruction_addr, bool link);

 public:

  // call

  enum {

    bl64_patchable_instruction_count = bxx64_patchable_instruction_count,

    bl64_patchable_size              = bxx64_patchable_size,

    bl64_patchable_ret_addr_offset   = bxx64_patchable_ret_addr_offset

  };

  inline void bl64_patchable(address target, relocInfo::relocType rt) {

    bxx64_patchable(target, rt, /*link=*/true);

  }

  inline static bool is_bl64_patchable_at(address instruction_addr) {

    return is_bxx64_patchable_at(instruction_addr, /*link=*/true);

  }

  inline static bool is_bl64_patchable_pcrelative_at(address instruction_addr) {

    return is_bxx64_patchable_pcrelative_at(instruction_addr, /*link=*/true);

  }

  inline static void set_dest_of_bl64_patchable_at(address instruction_addr, address target) {

    set_dest_of_bxx64_patchable_at(instruction_addr, target, /*link=*/true);

  }

  inline static address get_dest_of_bl64_patchable_at(address instruction_addr) {

    return get_dest_of_bxx64_patchable_at(instruction_addr, /*link=*/true);

  }

  // jump

  enum {

    b64_patchable_instruction_count = bxx64_patchable_instruction_count,

    b64_patchable_size              = bxx64_patchable_size,

  };

  inline void b64_patchable(address target, relocInfo::relocType rt) {

    bxx64_patchable(target, rt, /*link=*/false);

  }

  inline static bool is_b64_patchable_at(address instruction_addr) {

    return is_bxx64_patchable_at(instruction_addr, /*link=*/false);

  }

  inline static bool is_b64_patchable_pcrelative_at(address instruction_addr) {

    return is_bxx64_patchable_pcrelative_at(instruction_addr, /*link=*/false);

  }

  inline static void set_dest_of_b64_patchable_at(address instruction_addr, address target) {

    set_dest_of_bxx64_patchable_at(instruction_addr, target, /*link=*/false);

  }

  inline static address get_dest_of_b64_patchable_at(address instruction_addr) {

    return get_dest_of_bxx64_patchable_at(instruction_addr, /*link=*/false);

  }

  //

  // Support for frame handling

  //

  // some ABI-related functions

  void save_nonvolatile_gprs(   Register dst_base, int offset);

  void restore_nonvolatile_gprs(Register src_base, int offset);

  enum { num_volatile_regs = 11 + 14 }; // GPR + FPR

  void save_volatile_gprs(   Register dst_base, int offset);

  void restore_volatile_gprs(Register src_base, int offset);

  void save_LR_CR(   Register tmp);     // tmp contains LR on return.

  void restore_LR_CR(Register tmp);

  // Get current PC using bl-next-instruction trick.

  address get_PC_trash_LR(Register result);

  // Resize current frame either relatively wrt to current SP or absolute.

  void resize_frame(Register offset, Register tmp);

  void resize_frame(int      offset, Register tmp);

  void resize_frame_absolute(Register addr, Register tmp1, Register tmp2);

  // Push a frame of size bytes.

  void push_frame(Register bytes, Register tmp);

  // Push a frame of size `bytes'. No abi space provided.

  void push_frame(unsigned int bytes, Register tmp);

  // Push a frame of size `bytes' plus abi_reg_args on top.

  void push_frame_reg_args(unsigned int bytes, Register tmp);

  // Setup up a new C frame with a spill area for non-volatile GPRs and additional

  // space for local variables

  void push_frame_reg_args_nonvolatiles(unsigned int bytes, Register tmp);

  // pop current C frame

  void pop_frame();

  //

  // Calls

  //

 private:

  address _last_calls_return_pc;

#if defined(ABI_ELFv2)

  // Generic version of a call to C function.

  // Updates and returns _last_calls_return_pc.

  address branch_to(Register function_entry, bool and_link);

#else

  // Generic version of a call to C function via a function descriptor

  // with variable support for C calling conventions (TOC, ENV, etc.).

  // updates and returns _last_calls_return_pc.

  address branch_to(Register function_descriptor, bool and_link, bool save_toc_before_call,

                    bool restore_toc_after_call, bool load_toc_of_callee, bool load_env_of_callee);

#endif

 public:

  // Get the pc where the last call will return to. returns _last_calls_return_pc.

  inline address last_calls_return_pc();

#if defined(ABI_ELFv2)

  // Call a C function via a function descriptor and use full C

  // calling conventions. Updates and returns _last_calls_return_pc.

  address call_c(Register function_entry);

  // For tail calls: only branch, don't link, so callee returns to caller of this function.

  address call_c_and_return_to_caller(Register function_entry);

  address call_c(address function_entry, relocInfo::relocType rt);

#else

  // Call a C function via a function descriptor and use full C

  // calling conventions. Updates and returns _last_calls_return_pc.

  address call_c(Register function_descriptor);

  // For tail calls: only branch, don't link, so callee returns to caller of this function.

  address call_c_and_return_to_caller(Register function_descriptor);

  address call_c(const FunctionDescriptor* function_descriptor, relocInfo::relocType rt);

  address call_c_using_toc(const FunctionDescriptor* function_descriptor, relocInfo::relocType rt,

                           Register toc);

#endif

 protected:

  // It is imperative that all calls into the VM are handled via the

  // call_VM macros. They make sure that the stack linkage is setup

  // correctly. call_VM's correspond to ENTRY/ENTRY_X entry points

  // while call_VM_leaf's correspond to LEAF entry points.

  //

  // This is the base routine called by the different versions of

  // call_VM. The interpreter may customize this version by overriding

  // it for its purposes (e.g., to save/restore additional registers

  // when doing a VM call).

  //

  // If no last_java_sp is specified (noreg) then SP will be used instead.

  virtual void call_VM_base(

     // where an oop-result ends up if any; use noreg otherwise

    Register        oop_result,

    // to set up last_Java_frame in stubs; use noreg otherwise

    Register        last_java_sp,

    // the entry point

    address         entry_point,

    // flag which indicates if exception should be checked

    bool            check_exception = true

  );

  // Support for VM calls. This is the base routine called by the

  // different versions of call_VM_leaf. The interpreter may customize

  // this version by overriding it for its purposes (e.g., to

  // save/restore additional registers when doing a VM call).

  void call_VM_leaf_base(address entry_point);

 public:

  // Call into the VM.

  // Passes the thread pointer (in R3_ARG1) as a prepended argument.

  // Makes sure oop return values are visible to the GC.

  void call_VM(Register oop_result, address entry_point, bool check_exceptions = true);

  void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);

  void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);

  void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg3, bool check_exceptions = true);

  void call_VM_leaf(address entry_point);

  void call_VM_leaf(address entry_point, Register arg_1);

  void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);

  void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);

  // Call a stub function via a function descriptor, but don't save

  // TOC before call, don't setup TOC and ENV for call, and don't

  // restore TOC after call. Updates and returns _last_calls_return_pc.

  inline address call_stub(Register function_entry);

  inline void call_stub_and_return_to(Register function_entry, Register return_pc);

  //

  // Java utilities

  //

  // Read from the polling page, its address is already in a register.

  inline void load_from_polling_page(Register polling_page_address, int offset = 0);

  // Check whether instruction is a read access to the polling page

  // which was emitted by load_from_polling_page(..).

  static bool is_load_from_polling_page(int instruction, void* ucontext/*may be NULL*/,

                                        address* polling_address_ptr = NULL);

  // Support for NULL-checks

  //

  // Generates code that causes a NULL OS exception if the content of reg is NULL.

  // If the accessed location is M[reg + offset] and the offset is known, provide the

  // offset. No explicit code generation is needed if the offset is within a certain

  // range (0 <= offset <= page_size).

  // Stack overflow checking

  void bang_stack_with_offset(int offset);

  // If instruction is a stack bang of the form ld, stdu, or

  // stdux, return the banged address. Otherwise, return 0.

  static address get_stack_bang_address(int instruction, void* ucontext);

  // Check for reserved stack access in method being exited. If the reserved

  // stack area was accessed, protect it again and throw StackOverflowError.

  void reserved_stack_check(Register return_pc);

  // Atomics

  // CmpxchgX sets condition register to cmpX(current, compare).

  // (flag == ne) => (dest_current_value != compare_value), (!swapped)

  // (flag == eq) => (dest_current_value == compare_value), ( swapped)

  static inline bool cmpxchgx_hint_acquire_lock()  { return true; }

  // The stxcx will probably not be succeeded by a releasing store.

  static inline bool cmpxchgx_hint_release_lock()  { return false; }

  static inline bool cmpxchgx_hint_atomic_update() { return false; }

  // Cmpxchg semantics

  enum {

    MemBarNone = 0,

    MemBarRel  = 1,

    MemBarAcq  = 2,

    MemBarFenceAfter = 4 // use powers of 2

  };

 private:

  // Helper functions for word/sub-word atomics.

  void atomic_get_and_modify_generic(Register dest_current_value, Register exchange_value,

                                     Register addr_base, Register tmp1, Register tmp2, Register tmp3,

                                     bool cmpxchgx_hint, bool is_add, int size);

  void cmpxchg_loop_body(ConditionRegister flag, Register dest_current_value,

                         Register compare_value, Register exchange_value,

                         Register addr_base, Register tmp1, Register tmp2,

                         Label &retry, Label &failed, bool cmpxchgx_hint, int size);

  void cmpxchg_generic(ConditionRegister flag,

                       Register dest_current_value, Register compare_value, Register exchange_value, Register addr_base,

                       Register tmp1, Register tmp2,

                       int semantics, bool cmpxchgx_hint, Register int_flag_success, bool contention_hint, bool weak, int size);

 public:

  // Temps and addr_base are killed if processor does not support Power 8 instructions.

  // Result will be sign extended.

  void getandsetb(Register dest_current_value, Register exchange_value, Register addr_base,

                  Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {

    atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, false, 1);

  }

  // Temps and addr_base are killed if processor does not support Power 8 instructions.

  // Result will be sign extended.

  void getandseth(Register dest_current_value, Register exchange_value, Register addr_base,

                  Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {

    atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, false, 2);

  }

  void getandsetw(Register dest_current_value, Register exchange_value, Register addr_base,

                  bool cmpxchgx_hint) {

    atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, noreg, noreg, noreg, cmpxchgx_hint, false, 4);

  }

  void getandsetd(Register dest_current_value, Register exchange_value, Register addr_base,

                  bool cmpxchgx_hint);

  // tmp2/3 and addr_base are killed if processor does not support Power 8 instructions (tmp1 is always needed).

  // Result will be sign extended.

  void getandaddb(Register dest_current_value, Register inc_value, Register addr_base,

                  Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {

    atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, true, 1);

  }

  // tmp2/3 and addr_base are killed if processor does not support Power 8 instructions (tmp1 is always needed).

  // Result will be sign extended.

  void getandaddh(Register dest_current_value, Register inc_value, Register addr_base,

                  Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {

    atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, true, 2);

  }

  void getandaddw(Register dest_current_value, Register inc_value, Register addr_base,

                  Register tmp1, bool cmpxchgx_hint) {

    atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, noreg, noreg, cmpxchgx_hint, true, 4);

  }

  void getandaddd(Register dest_current_value, Register exchange_value, Register addr_base,

                  Register tmp, bool cmpxchgx_hint);

  // Temps, addr_base and exchange_value are killed if processor does not support Power 8 instructions.

  // compare_value must be at least 32 bit sign extended. Result will be sign extended.

  void cmpxchgb(ConditionRegister flag,

                Register dest_current_value, Register compare_value, Register exchange_value, Register addr_base,

                Register tmp1, Register tmp2, int semantics, bool cmpxchgx_hint = false,

                Register int_flag_success = noreg, bool contention_hint = false, bool weak = false) {

    cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base, tmp1, tmp2,

                    semantics, cmpxchgx_hint, int_flag_success, contention_hint, weak, 1);

  }

  // Temps, addr_base and exchange_value are killed if processor does not support Power 8 instructions.

  // compare_value must be at least 32 bit sign extended. Result will be sign extended.

  void cmpxchgh(ConditionRegister flag,

                Register dest_current_value, Register compare_value, Register exchange_value, Register addr_base,

                Register tmp1, Register tmp2, int semantics, bool cmpxchgx_hint = false,

                Register int_flag_success = noreg, bool contention_hint = false, bool weak = false) {

    cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base, tmp1, tmp2,

                    semantics, cmpxchgx_hint, int_flag_success, contention_hint, weak, 2);

  }

  void cmpxchgw(ConditionRegister flag,

                Register dest_current_value, Register compare_value, Register exchange_value, Register addr_base,

                int semantics, bool cmpxchgx_hint = false,

                Register int_flag_success = noreg, bool contention_hint = false, bool weak = false) {

    cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base, noreg, noreg,

                    semantics, cmpxchgx_hint, int_flag_success, contention_hint, weak, 4);

  }

  void cmpxchgd(ConditionRegister flag,

                Register dest_current_value, RegisterOrConstant compare_value, Register exchange_value,

                Register addr_base, int semantics, bool cmpxchgx_hint = false,

                Register int_flag_success = noreg, Label* failed = NULL, bool contention_hint = false, bool weak = false);

  // interface method calling

  void lookup_interface_method(Register recv_klass,

                               Register intf_klass,

                               RegisterOrConstant itable_index,

                               Register method_result,