/*

 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.

 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.

 *

 * This code is free software; you can redistribute it and/or modify it

 * under the terms of the GNU General Public License version 2 only, as

 * published by the Free Software Foundation.

 *

 * This code is distributed in the hope that it will be useful, but WITHOUT

 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or

 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License

 * version 2 for more details (a copy is included in the LICENSE file that

 * accompanied this code).

 *

 * You should have received a copy of the GNU General Public License version

 * 2 along with this work; if not, write to the Free Software Foundation,

 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.

 *

 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA

 * or visit www.oracle.com if you need additional information or have any

 * questions.

 *

 */

#include "incls/_precompiled.incl"

#include "incls/_c1_Runtime1_sparc.cpp.incl"

// Implementation of StubAssembler

int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry_point, int number_of_arguments) {

  // for sparc changing the number of arguments doesn't change

  // anything about the frame size so we'll always lie and claim that

  // we are only passing 1 argument.

  set_num_rt_args(1);

  assert_not_delayed();

  // bang stack before going to runtime

  set(-os::vm_page_size() + STACK_BIAS, G3_scratch);

  st(G0, SP, G3_scratch);

  // debugging support

  assert(number_of_arguments >= 0   , "cannot have negative number of arguments");

  set_last_Java_frame(SP, noreg);

  if (VerifyThread)  mov(G2_thread, O0); // about to be smashed; pass early

  save_thread(L7_thread_cache);

  // do the call

  call(entry_point, relocInfo::runtime_call_type);

  if (!VerifyThread) {

    delayed()->mov(G2_thread, O0);  // pass thread as first argument

  } else {

    delayed()->nop();             // (thread already passed)

  }

  int call_offset = offset();  // offset of return address

  restore_thread(L7_thread_cache);

  reset_last_Java_frame();

  // check for pending exceptions

  { Label L;

    Address exception_addr(G2_thread, Thread::pending_exception_offset());

    ld_ptr(exception_addr, Gtemp);

    br_null(Gtemp, false, pt, L);

    delayed()->nop();

    Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());

    st_ptr(G0, vm_result_addr);

    Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());

    st_ptr(G0, vm_result_addr_2);

    if (frame_size() == no_frame_size) {

      // we use O7 linkage so that forward_exception_entry has the issuing PC

      call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);

      delayed()->restore();

    } else if (_stub_id == Runtime1::forward_exception_id) {

      should_not_reach_here();

    } else {

      AddressLiteral exc(Runtime1::entry_for(Runtime1::forward_exception_id));

      jump_to(exc, G4);

      delayed()->nop();

    }

    bind(L);

  }

  // get oop result if there is one and reset the value in the thread

  if (oop_result1->is_valid()) {                    // get oop result if there is one and reset it in the thread

    get_vm_result  (oop_result1);

  } else {

    // be a little paranoid and clear the result

    Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());

    st_ptr(G0, vm_result_addr);

  }

  if (oop_result2->is_valid()) {

    get_vm_result_2(oop_result2);

  } else {

    // be a little paranoid and clear the result

    Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());

    st_ptr(G0, vm_result_addr_2);

  }

  return call_offset;

}

int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {

  // O0 is reserved for the thread

  mov(arg1, O1);

  return call_RT(oop_result1, oop_result2, entry, 1);

}

int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {

  // O0 is reserved for the thread

  mov(arg1, O1);

  mov(arg2, O2); assert(arg2 != O1, "smashed argument");

  return call_RT(oop_result1, oop_result2, entry, 2);

}

int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {

  // O0 is reserved for the thread

  mov(arg1, O1);

  mov(arg2, O2); assert(arg2 != O1,               "smashed argument");

  mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");

  return call_RT(oop_result1, oop_result2, entry, 3);

}

// Implementation of Runtime1

#define __ sasm->

static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];

static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];

static int reg_save_size_in_words;

static int frame_size_in_bytes = -1;

static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {

  assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),

         " mismatch in calculation");

  sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);

  int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);

  OopMap* oop_map = new OopMap(frame_size_in_slots, 0);

  int i;

  for (i = 0; i < FrameMap::nof_cpu_regs; i++) {

    Register r = as_Register(i);

    if (r == G1 || r == G3 || r == G4 || r == G5) {

      int sp_offset = cpu_reg_save_offsets[i];

      oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),

                                r->as_VMReg());

    }

  }

  if (save_fpu_registers) {

    for (i = 0; i < FrameMap::nof_fpu_regs; i++) {

      FloatRegister r = as_FloatRegister(i);

      int sp_offset = fpu_reg_save_offsets[i];

      oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),

                                r->as_VMReg());

    }

  }

  return oop_map;

}

static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {

  assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),

         " mismatch in calculation");

  __ save_frame_c1(frame_size_in_bytes);

  sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);

  // Record volatile registers as callee-save values in an OopMap so their save locations will be

  // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for

  // deoptimization; see compiledVFrame::create_stack_value).  The caller's I, L and O registers

  // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame

  // (as the stub's I's) when the runtime routine called by the stub creates its frame.

  // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))

  int i;

  for (i = 0; i < FrameMap::nof_cpu_regs; i++) {

    Register r = as_Register(i);

    if (r == G1 || r == G3 || r == G4 || r == G5) {

      int sp_offset = cpu_reg_save_offsets[i];

      __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);

    }

  }

  if (save_fpu_registers) {

    for (i = 0; i < FrameMap::nof_fpu_regs; i++) {

      FloatRegister r = as_FloatRegister(i);

      int sp_offset = fpu_reg_save_offsets[i];

      __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);

    }

  }

  return generate_oop_map(sasm, save_fpu_registers);

}

static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {

  for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {

    Register r = as_Register(i);

    if (r == G1 || r == G3 || r == G4 || r == G5) {

      __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);

    }

  }

  if (restore_fpu_registers) {

    for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {

      FloatRegister r = as_FloatRegister(i);

      __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);

    }

  }

}

void Runtime1::initialize_pd() {

  // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines

  //

  // A stub routine will have a frame that is at least large enough to hold

  // a register window save area (obviously) and the volatile g registers

  // and floating registers. A user of save_live_registers can have a frame

  // that has more scratch area in it (although typically they will use L-regs).

  // in that case the frame will look like this (stack growing down)

  //

  // FP -> |             |

  //       | scratch mem |

  //       |   "      "  |

  //       --------------

  //       | float regs  |

  //       |   "    "    |

  //       ---------------

  //       | G regs      |

  //       | "  "        |

  //       ---------------

  //       | abi reg.    |

  //       | window save |

  //       | area        |

  // SP -> ---------------

  //

  int i;

  int sp_offset = round_to(frame::register_save_words, 2); //  start doubleword aligned

  // only G int registers are saved explicitly; others are found in register windows

  for (i = 0; i < FrameMap::nof_cpu_regs; i++) {

    Register r = as_Register(i);

    if (r == G1 || r == G3 || r == G4 || r == G5) {

      cpu_reg_save_offsets[i] = sp_offset;

      sp_offset++;

    }

  }

  // all float registers are saved explicitly

  assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");

  for (i = 0; i < FrameMap::nof_fpu_regs; i++) {

    fpu_reg_save_offsets[i] = sp_offset;

    sp_offset++;

  }

  reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;

  // this should match assembler::total_frame_size_in_bytes, which

  // isn't callable from this context.  It's checked by an assert when

  // it's used though.

  frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);

}

OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {

  // make a frame and preserve the caller's caller-save registers

  OopMap* oop_map = save_live_registers(sasm);

  int call_offset;

  if (!has_argument) {

    call_offset = __ call_RT(noreg, noreg, target);

  } else {

    call_offset = __ call_RT(noreg, noreg, target, G4);

  }

  OopMapSet* oop_maps = new OopMapSet();

  oop_maps->add_gc_map(call_offset, oop_map);

  __ should_not_reach_here();

  return oop_maps;

}

OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,

                                        Register arg1, Register arg2, Register arg3) {

  // make a frame and preserve the caller's caller-save registers

  OopMap* oop_map = save_live_registers(sasm);

  int call_offset;

  if (arg1 == noreg) {

    call_offset = __ call_RT(result, noreg, target);

  } else if (arg2 == noreg) {

    call_offset = __ call_RT(result, noreg, target, arg1);

  } else if (arg3 == noreg) {

    call_offset = __ call_RT(result, noreg, target, arg1, arg2);

  } else {

    call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);

  }

  OopMapSet* oop_maps = NULL;

  oop_maps = new OopMapSet();

  oop_maps->add_gc_map(call_offset, oop_map);

  restore_live_registers(sasm);

  __ ret();

  __ delayed()->restore();

  return oop_maps;

}

OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {

  // make a frame and preserve the caller's caller-save registers

  OopMap* oop_map = save_live_registers(sasm);

  // call the runtime patching routine, returns non-zero if nmethod got deopted.

  int call_offset = __ call_RT(noreg, noreg, target);

  OopMapSet* oop_maps = new OopMapSet();

  oop_maps->add_gc_map(call_offset, oop_map);

  // re-execute the patched instruction or, if the nmethod was deoptmized, return to the

  // deoptimization handler entry that will cause re-execution of the current bytecode

  DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();

  assert(deopt_blob != NULL, "deoptimization blob must have been created");

  Label no_deopt;

  __ tst(O0);

  __ brx(Assembler::equal, false, Assembler::pt, no_deopt);

  __ delayed()->nop();

  // return to the deoptimization handler entry for unpacking and rexecute

  // if we simply returned the we'd deopt as if any call we patched had just

  // returned.

  restore_live_registers(sasm);

  __ restore();

  __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);

  __ delayed()->nop();

  __ bind(no_deopt);

  restore_live_registers(sasm);

  __ ret();

  __ delayed()->restore();

  return oop_maps;

}

OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {

  OopMapSet* oop_maps = NULL;

  // for better readability

  const bool must_gc_arguments = true;

  const bool dont_gc_arguments = false;

  // stub code & info for the different stubs

  switch (id) {

    case forward_exception_id:

      {

        // we're handling an exception in the context of a compiled

        // frame.  The registers have been saved in the standard

        // places.  Perform an exception lookup in the caller and

        // dispatch to the handler if found.  Otherwise unwind and

        // dispatch to the callers exception handler.

        oop_maps = new OopMapSet();

        OopMap* oop_map = generate_oop_map(sasm, true);

        // transfer the pending exception to the exception_oop

        __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);

        __ ld_ptr(Oexception, 0, G0);

        __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));

        __ add(I7, frame::pc_return_offset, Oissuing_pc);

        generate_handle_exception(sasm, oop_maps, oop_map);

        __ should_not_reach_here();

      }

      break;

    case new_instance_id:

    case fast_new_instance_id:

    case fast_new_instance_init_check_id:

      {

        Register G5_klass = G5; // Incoming

        Register O0_obj   = O0; // Outgoing

        if (id == new_instance_id) {

          __ set_info("new_instance", dont_gc_arguments);

        } else if (id == fast_new_instance_id) {

          __ set_info("fast new_instance", dont_gc_arguments);

        } else {

          assert(id == fast_new_instance_init_check_id, "bad StubID");

          __ set_info("fast new_instance init check", dont_gc_arguments);

        }

        if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&

            UseTLAB && FastTLABRefill) {

          Label slow_path;

          Register G1_obj_size = G1;

          Register G3_t1 = G3;

          Register G4_t2 = G4;

          assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);

          // Push a frame since we may do dtrace notification for the

          // allocation which requires calling out and we don't want

          // to stomp the real return address.

          __ save_frame(0);

          if (id == fast_new_instance_init_check_id) {

            // make sure the klass is initialized

            __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1);

            __ cmp(G3_t1, instanceKlass::fully_initialized);

            __ br(Assembler::notEqual, false, Assembler::pn, slow_path);

            __ delayed()->nop();

          }

#ifdef ASSERT

          // assert object can be fast path allocated

          {

            Label ok, not_ok;

          __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size);

          __ cmp(G1_obj_size, 0);  // make sure it's an instance (LH > 0)

          __ br(Assembler::lessEqual, false, Assembler::pn, not_ok);

          __ delayed()->nop();

          __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);

          __ br(Assembler::zero, false, Assembler::pn, ok);

          __ delayed()->nop();

          __ bind(not_ok);

          __ stop("assert(can be fast path allocated)");

          __ should_not_reach_here();

          __ bind(ok);

          }

#endif // ASSERT

          // if we got here then the TLAB allocation failed, so try

          // refilling the TLAB or allocating directly from eden.

          Label retry_tlab, try_eden;

          __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass

          __ bind(retry_tlab);

          // get the instance size

          __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);

          __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);

          __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);

          __ verify_oop(O0_obj);

          __ mov(O0, I0);

          __ ret();

          __ delayed()->restore();

          __ bind(try_eden);

          // get the instance size

          __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);

          __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);

          __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);

          __ verify_oop(O0_obj);

          __ mov(O0, I0);

          __ ret();

          __ delayed()->restore();

          __ bind(slow_path);

          // pop this frame so generate_stub_call can push it's own

          __ restore();

        }

        oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);

        // I0->O0: new instance

      }

      break;

#ifdef TIERED

    case counter_overflow_id:

        // G4 contains bci

      oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4);

      break;

#endif // TIERED

    case new_type_array_id:

    case new_object_array_id:

      {

        Register G5_klass = G5; // Incoming

        Register G4_length = G4; // Incoming

        Register O0_obj   = O0; // Outgoing

        Address klass_lh(G5_klass, ((klassOopDesc::header_size() * HeapWordSize)

                                    + Klass::layout_helper_offset_in_bytes()));

        assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");

        assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");

        // Use this offset to pick out an individual byte of the layout_helper:

        const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}

                                                 - Klass::_lh_header_size_shift / BitsPerByte);

        if (id == new_type_array_id) {

          __ set_info("new_type_array", dont_gc_arguments);

        } else {

          __ set_info("new_object_array", dont_gc_arguments);

        }

#ifdef ASSERT

        // assert object type is really an array of the proper kind

        {

          Label ok;

          Register G3_t1 = G3;

          __ ld(klass_lh, G3_t1);

          __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);

          int tag = ((id == new_type_array_id)

                     ? Klass::_lh_array_tag_type_value

                     : Klass::_lh_array_tag_obj_value);

          __ cmp(G3_t1, tag);

          __ brx(Assembler::equal, false, Assembler::pt, ok);

          __ delayed()->nop();

          __ stop("assert(is an array klass)");

          __ should_not_reach_here();

          __ bind(ok);

        }

#endif // ASSERT

        if (UseTLAB && FastTLABRefill) {

          Label slow_path;

          Register G1_arr_size = G1;

          Register G3_t1 = G3;

          Register O1_t2 = O1;

          assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);

          // check that array length is small enough for fast path

          __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);

          __ cmp(G4_length, G3_t1);

          __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);

          __ delayed()->nop();