< prev index next >

src/hotspot/cpu/x86/macroAssembler_x86.cpp

Print this page

   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "precompiled.hpp"
   26 #include "asm/assembler.hpp"
   27 #include "asm/assembler.inline.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"

   31 #include "crc32c.h"
   32 #include "gc/shared/barrierSet.hpp"
   33 #include "gc/shared/barrierSetAssembler.hpp"
   34 #include "gc/shared/collectedHeap.inline.hpp"
   35 #include "gc/shared/tlab_globals.hpp"
   36 #include "interpreter/bytecodeHistogram.hpp"
   37 #include "interpreter/interpreter.hpp"
   38 #include "jvm.h"
   39 #include "memory/resourceArea.hpp"
   40 #include "memory/universe.hpp"
   41 #include "oops/accessDecorators.hpp"
   42 #include "oops/compressedKlass.inline.hpp"
   43 #include "oops/compressedOops.inline.hpp"
   44 #include "oops/klass.inline.hpp"

   45 #include "prims/methodHandles.hpp"
   46 #include "runtime/continuation.hpp"
   47 #include "runtime/interfaceSupport.inline.hpp"
   48 #include "runtime/javaThread.hpp"
   49 #include "runtime/jniHandles.hpp"
   50 #include "runtime/objectMonitor.hpp"
   51 #include "runtime/os.hpp"
   52 #include "runtime/safepoint.hpp"
   53 #include "runtime/safepointMechanism.hpp"
   54 #include "runtime/sharedRuntime.hpp"

   55 #include "runtime/stubRoutines.hpp"
   56 #include "utilities/checkedCast.hpp"
   57 #include "utilities/macros.hpp"




   58 
   59 #ifdef PRODUCT
   60 #define BLOCK_COMMENT(str) /* nothing */
   61 #define STOP(error) stop(error)
   62 #else
   63 #define BLOCK_COMMENT(str) block_comment(str)
   64 #define STOP(error) block_comment(error); stop(error)
   65 #endif
   66 
   67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   68 
   69 #ifdef ASSERT
   70 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   71 #endif
   72 
   73 static const Assembler::Condition reverse[] = {
   74     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   75     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   76     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   77     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,

 1701 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1702   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2));
 1703   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1704   pass_arg2(this, arg_2);
 1705   pass_arg1(this, arg_1);
 1706   pass_arg0(this, arg_0);
 1707   call_VM_leaf(entry_point, 3);
 1708 }
 1709 
 1710 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1711   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3));
 1712   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1713   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1714   pass_arg3(this, arg_3);
 1715   pass_arg2(this, arg_2);
 1716   pass_arg1(this, arg_1);
 1717   pass_arg0(this, arg_0);
 1718   call_VM_leaf(entry_point, 3);
 1719 }
 1720 




 1721 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1722   pass_arg0(this, arg_0);
 1723   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1724 }
 1725 
 1726 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1727   LP64_ONLY(assert_different_registers(arg_0, c_rarg1));
 1728   pass_arg1(this, arg_1);
 1729   pass_arg0(this, arg_0);
 1730   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1731 }
 1732 
 1733 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1734   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2));
 1735   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1736   pass_arg2(this, arg_2);
 1737   pass_arg1(this, arg_1);
 1738   pass_arg0(this, arg_0);
 1739   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1740 }

 2885     lea(rscratch, src);
 2886     Assembler::mulss(dst, Address(rscratch, 0));
 2887   }
 2888 }
 2889 
 2890 void MacroAssembler::null_check(Register reg, int offset) {
 2891   if (needs_explicit_null_check(offset)) {
 2892     // provoke OS null exception if reg is null by
 2893     // accessing M[reg] w/o changing any (non-CC) registers
 2894     // NOTE: cmpl is plenty here to provoke a segv
 2895     cmpptr(rax, Address(reg, 0));
 2896     // Note: should probably use testl(rax, Address(reg, 0));
 2897     //       may be shorter code (however, this version of
 2898     //       testl needs to be implemented first)
 2899   } else {
 2900     // nothing to do, (later) access of M[reg + offset]
 2901     // will provoke OS null exception if reg is null
 2902   }
 2903 }
 2904 





























































































































 2905 void MacroAssembler::os_breakpoint() {
 2906   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2907   // (e.g., MSVC can't call ps() otherwise)
 2908   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2909 }
 2910 
 2911 void MacroAssembler::unimplemented(const char* what) {
 2912   const char* buf = nullptr;
 2913   {
 2914     ResourceMark rm;
 2915     stringStream ss;
 2916     ss.print("unimplemented: %s", what);
 2917     buf = code_string(ss.as_string());
 2918   }
 2919   stop(buf);
 2920 }
 2921 
 2922 #ifdef _LP64
 2923 #define XSTATE_BV 0x200
 2924 #endif

 4049 }
 4050 
 4051 // C++ bool manipulation
 4052 void MacroAssembler::testbool(Register dst) {
 4053   if(sizeof(bool) == 1)
 4054     testb(dst, 0xff);
 4055   else if(sizeof(bool) == 2) {
 4056     // testw implementation needed for two byte bools
 4057     ShouldNotReachHere();
 4058   } else if(sizeof(bool) == 4)
 4059     testl(dst, dst);
 4060   else
 4061     // unsupported
 4062     ShouldNotReachHere();
 4063 }
 4064 
 4065 void MacroAssembler::testptr(Register dst, Register src) {
 4066   LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
 4067 }
 4068 


















































































































 4069 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 4070 void MacroAssembler::tlab_allocate(Register thread, Register obj,
 4071                                    Register var_size_in_bytes,
 4072                                    int con_size_in_bytes,
 4073                                    Register t1,
 4074                                    Register t2,
 4075                                    Label& slow_case) {
 4076   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 4077   bs->tlab_allocate(this, thread, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 4078 }
 4079 
 4080 RegSet MacroAssembler::call_clobbered_gp_registers() {
 4081   RegSet regs;
 4082 #ifdef _LP64
 4083   regs += RegSet::of(rax, rcx, rdx);
 4084 #ifndef WINDOWS
 4085   regs += RegSet::of(rsi, rdi);
 4086 #endif
 4087   regs += RegSet::range(r8, r11);
 4088 #else

 4307     // clear topmost word (no jump would be needed if conditional assignment worked here)
 4308     movptr(Address(address, index, Address::times_8, offset_in_bytes - 0*BytesPerWord), temp);
 4309     // index could be 0 now, must check again
 4310     jcc(Assembler::zero, done);
 4311     bind(even);
 4312   }
 4313 #endif // !_LP64
 4314   // initialize remaining object fields: index is a multiple of 2 now
 4315   {
 4316     Label loop;
 4317     bind(loop);
 4318     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 4319     NOT_LP64(movptr(Address(address, index, Address::times_8, offset_in_bytes - 2*BytesPerWord), temp);)
 4320     decrement(index);
 4321     jcc(Assembler::notZero, loop);
 4322   }
 4323 
 4324   bind(done);
 4325 }
 4326 































































 4327 // Look up the method for a megamorphic invokeinterface call.
 4328 // The target method is determined by <intf_klass, itable_index>.
 4329 // The receiver klass is in recv_klass.
 4330 // On success, the result will be in method_result, and execution falls through.
 4331 // On failure, execution transfers to the given label.
 4332 void MacroAssembler::lookup_interface_method(Register recv_klass,
 4333                                              Register intf_klass,
 4334                                              RegisterOrConstant itable_index,
 4335                                              Register method_result,
 4336                                              Register scan_temp,
 4337                                              Label& L_no_such_interface,
 4338                                              bool return_method) {
 4339   assert_different_registers(recv_klass, intf_klass, scan_temp);
 4340   assert_different_registers(method_result, intf_klass, scan_temp);
 4341   assert(recv_klass != method_result || !return_method,
 4342          "recv_klass can be destroyed when method isn't needed");
 4343 
 4344   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 4345          "caller must use same register for non-constant itable index as for method");
 4346 

 5108   } else {
 5109     Label L;
 5110     jccb(negate_condition(cc), L);
 5111     movl(dst, src);
 5112     bind(L);
 5113   }
 5114 }
 5115 
 5116 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 5117   if (VM_Version::supports_cmov()) {
 5118     cmovl(cc, dst, src);
 5119   } else {
 5120     Label L;
 5121     jccb(negate_condition(cc), L);
 5122     movl(dst, src);
 5123     bind(L);
 5124   }
 5125 }
 5126 
 5127 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 5128   if (!VerifyOops) return;




 5129 
 5130   BLOCK_COMMENT("verify_oop {");
 5131 #ifdef _LP64
 5132   push(rscratch1);
 5133 #endif
 5134   push(rax);                          // save rax
 5135   push(reg);                          // pass register argument
 5136 
 5137   // Pass register number to verify_oop_subroutine
 5138   const char* b = nullptr;
 5139   {
 5140     ResourceMark rm;
 5141     stringStream ss;
 5142     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 5143     b = code_string(ss.as_string());
 5144   }
 5145   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5146   pushptr(buffer.addr(), rscratch1);
 5147 
 5148   // call indirectly to solve generation ordering problem

 5170   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 5171   int stackElementSize = Interpreter::stackElementSize;
 5172   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 5173 #ifdef ASSERT
 5174   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 5175   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 5176 #endif
 5177   Register             scale_reg    = noreg;
 5178   Address::ScaleFactor scale_factor = Address::no_scale;
 5179   if (arg_slot.is_constant()) {
 5180     offset += arg_slot.as_constant() * stackElementSize;
 5181   } else {
 5182     scale_reg    = arg_slot.as_register();
 5183     scale_factor = Address::times(stackElementSize);
 5184   }
 5185   offset += wordSize;           // return PC is on stack
 5186   return Address(rsp, scale_reg, scale_factor, offset);
 5187 }
 5188 
 5189 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5190   if (!VerifyOops) return;




 5191 
 5192 #ifdef _LP64
 5193   push(rscratch1);
 5194 #endif
 5195   push(rax); // save rax,
 5196   // addr may contain rsp so we will have to adjust it based on the push
 5197   // we just did (and on 64 bit we do two pushes)
 5198   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5199   // stores rax into addr which is backwards of what was intended.
 5200   if (addr.uses(rsp)) {
 5201     lea(rax, addr);
 5202     pushptr(Address(rax, LP64_ONLY(2 *) BytesPerWord));
 5203   } else {
 5204     pushptr(addr);
 5205   }
 5206 
 5207   // Pass register number to verify_oop_subroutine
 5208   const char* b = nullptr;
 5209   {
 5210     ResourceMark rm;

 5657 
 5658 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5659   // get mirror
 5660   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5661   load_method_holder(mirror, method);
 5662   movptr(mirror, Address(mirror, mirror_offset));
 5663   resolve_oop_handle(mirror, tmp);
 5664 }
 5665 
 5666 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5667   load_method_holder(rresult, rmethod);
 5668   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5669 }
 5670 
 5671 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5672   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5673   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5674   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5675 }
 5676 








 5677 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5678   assert_different_registers(src, tmp);
 5679   assert_different_registers(dst, tmp);
 5680 #ifdef _LP64
 5681   if (UseCompressedClassPointers) {
 5682     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5683     decode_klass_not_null(dst, tmp);
 5684   } else
 5685 #endif
 5686     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));





 5687 }
 5688 
 5689 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5690   assert_different_registers(src, tmp);
 5691   assert_different_registers(dst, tmp);
 5692 #ifdef _LP64
 5693   if (UseCompressedClassPointers) {
 5694     encode_klass_not_null(src, tmp);
 5695     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5696   } else
 5697 #endif
 5698     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5699 }
 5700 
 5701 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5702                                     Register tmp1, Register thread_tmp) {
 5703   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5704   decorators = AccessInternal::decorator_fixup(decorators, type);
 5705   bool as_raw = (decorators & AS_RAW) != 0;
 5706   if (as_raw) {
 5707     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
 5708   } else {
 5709     bs->load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
 5710   }
 5711 }
 5712 
 5713 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5714                                      Register tmp1, Register tmp2, Register tmp3) {
 5715   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5716   decorators = AccessInternal::decorator_fixup(decorators, type);
 5717   bool as_raw = (decorators & AS_RAW) != 0;
 5718   if (as_raw) {
 5719     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5720   } else {
 5721     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5722   }
 5723 }
 5724 














































 5725 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
 5726                                    Register thread_tmp, DecoratorSet decorators) {
 5727   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, thread_tmp);
 5728 }
 5729 
 5730 // Doesn't do verification, generates fixed size code
 5731 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 5732                                             Register thread_tmp, DecoratorSet decorators) {
 5733   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, thread_tmp);
 5734 }
 5735 
 5736 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5737                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5738   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5739 }
 5740 
 5741 // Used for storing nulls.
 5742 void MacroAssembler::store_heap_oop_null(Address dst) {
 5743   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5744 }

 6044 
 6045 void MacroAssembler::reinit_heapbase() {
 6046   if (UseCompressedOops) {
 6047     if (Universe::heap() != nullptr) {
 6048       if (CompressedOops::base() == nullptr) {
 6049         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6050       } else {
 6051         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6052       }
 6053     } else {
 6054       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6055     }
 6056   }
 6057 }
 6058 
 6059 #endif // _LP64
 6060 
 6061 #if COMPILER2_OR_JVMCI
 6062 
 6063 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6064 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6065   // cnt - number of qwords (8-byte words).
 6066   // base - start address, qword aligned.
 6067   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6068   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 6069   if (use64byteVector) {
 6070     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 6071   } else if (MaxVectorSize >= 32) {
 6072     vpxor(xtmp, xtmp, xtmp, AVX_256bit);


 6073   } else {
 6074     pxor(xtmp, xtmp);

 6075   }
 6076   jmp(L_zero_64_bytes);
 6077 
 6078   BIND(L_loop);
 6079   if (MaxVectorSize >= 32) {
 6080     fill64(base, 0, xtmp, use64byteVector);
 6081   } else {
 6082     movdqu(Address(base,  0), xtmp);
 6083     movdqu(Address(base, 16), xtmp);
 6084     movdqu(Address(base, 32), xtmp);
 6085     movdqu(Address(base, 48), xtmp);
 6086   }
 6087   addptr(base, 64);
 6088 
 6089   BIND(L_zero_64_bytes);
 6090   subptr(cnt, 8);
 6091   jccb(Assembler::greaterEqual, L_loop);
 6092 
 6093   // Copy trailing 64 bytes
 6094   if (use64byteVector) {
 6095     addptr(cnt, 8);
 6096     jccb(Assembler::equal, L_end);
 6097     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 6098     jmp(L_end);
 6099   } else {
 6100     addptr(cnt, 4);
 6101     jccb(Assembler::less, L_tail);
 6102     if (MaxVectorSize >= 32) {
 6103       vmovdqu(Address(base, 0), xtmp);
 6104     } else {
 6105       movdqu(Address(base,  0), xtmp);
 6106       movdqu(Address(base, 16), xtmp);
 6107     }
 6108   }
 6109   addptr(base, 32);
 6110   subptr(cnt, 4);
 6111 
 6112   BIND(L_tail);
 6113   addptr(cnt, 4);
 6114   jccb(Assembler::lessEqual, L_end);
 6115   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6116     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 6117   } else {
 6118     decrement(cnt);
 6119 
 6120     BIND(L_sloop);
 6121     movq(Address(base, 0), xtmp);
 6122     addptr(base, 8);
 6123     decrement(cnt);
 6124     jccb(Assembler::greaterEqual, L_sloop);
 6125   }
 6126   BIND(L_end);
 6127 }
 6128 
















































































































































































































































































































































































































 6129 // Clearing constant sized memory using YMM/ZMM registers.
 6130 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6131   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6132   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 6133 
 6134   int vector64_count = (cnt & (~0x7)) >> 3;
 6135   cnt = cnt & 0x7;
 6136   const int fill64_per_loop = 4;
 6137   const int max_unrolled_fill64 = 8;
 6138 
 6139   // 64 byte initialization loop.
 6140   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6141   int start64 = 0;
 6142   if (vector64_count > max_unrolled_fill64) {
 6143     Label LOOP;
 6144     Register index = rtmp;
 6145 
 6146     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6147 
 6148     movl(index, 0);

 6198         break;
 6199       case 7:
 6200         if (use64byteVector) {
 6201           movl(rtmp, 0x7F);
 6202           kmovwl(mask, rtmp);
 6203           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6204         } else {
 6205           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6206           movl(rtmp, 0x7);
 6207           kmovwl(mask, rtmp);
 6208           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6209         }
 6210         break;
 6211       default:
 6212         fatal("Unexpected length : %d\n",cnt);
 6213         break;
 6214     }
 6215   }
 6216 }
 6217 
 6218 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 6219                                bool is_large, KRegister mask) {
 6220   // cnt      - number of qwords (8-byte words).
 6221   // base     - start address, qword aligned.
 6222   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6223   assert(base==rdi, "base register must be edi for rep stos");
 6224   assert(tmp==rax,   "tmp register must be eax for rep stos");
 6225   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6226   assert(InitArrayShortSize % BytesPerLong == 0,
 6227     "InitArrayShortSize should be the multiple of BytesPerLong");
 6228 
 6229   Label DONE;
 6230   if (!is_large || !UseXMMForObjInit) {
 6231     xorptr(tmp, tmp);
 6232   }
 6233 
 6234   if (!is_large) {
 6235     Label LOOP, LONG;
 6236     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6237     jccb(Assembler::greater, LONG);
 6238 
 6239     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
 6240 
 6241     decrement(cnt);
 6242     jccb(Assembler::negative, DONE); // Zero length
 6243 
 6244     // Use individual pointer-sized stores for small counts:
 6245     BIND(LOOP);
 6246     movptr(Address(base, cnt, Address::times_ptr), tmp);
 6247     decrement(cnt);
 6248     jccb(Assembler::greaterEqual, LOOP);
 6249     jmpb(DONE);
 6250 
 6251     BIND(LONG);
 6252   }
 6253 
 6254   // Use longer rep-prefixed ops for non-small counts:
 6255   if (UseFastStosb) {
 6256     shlptr(cnt, 3); // convert to number of bytes
 6257     rep_stosb();
 6258   } else if (UseXMMForObjInit) {
 6259     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6260   } else {
 6261     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
 6262     rep_stos();
 6263   }
 6264 
 6265   BIND(DONE);
 6266 }
 6267 
 6268 #endif //COMPILER2_OR_JVMCI
 6269 
 6270 
 6271 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6272                                    Register to, Register value, Register count,
 6273                                    Register rtmp, XMMRegister xtmp) {
 6274   ShortBranchVerifier sbv(this);
 6275   assert_different_registers(to, value, count, rtmp);
 6276   Label L_exit;
 6277   Label L_fill_2_bytes, L_fill_4_bytes;
 6278 
 6279 #if defined(COMPILER2) && defined(_LP64)

10294 
10295   // Load top.
10296   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10297 
10298   // Check if the lock-stack is full.
10299   cmpl(top, LockStack::end_offset());
10300   jcc(Assembler::greaterEqual, slow);
10301 
10302   // Check for recursion.
10303   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10304   jcc(Assembler::equal, push);
10305 
10306   // Check header for monitor (0b10).
10307   testptr(reg_rax, markWord::monitor_value);
10308   jcc(Assembler::notZero, slow);
10309 
10310   // Try to lock. Transition lock bits 0b01 => 0b00
10311   movptr(tmp, reg_rax);
10312   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10313   orptr(reg_rax, markWord::unlocked_value);




10314   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10315   jcc(Assembler::notEqual, slow);
10316 
10317   // Restore top, CAS clobbers register.
10318   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10319 
10320   bind(push);
10321   // After successful lock, push object on lock-stack.
10322   movptr(Address(thread, top), obj);
10323   incrementl(top, oopSize);
10324   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10325 }
10326 
10327 // Implements lightweight-unlocking.
10328 //
10329 // obj: the object to be unlocked
10330 // reg_rax: rax
10331 // thread: the thread
10332 // tmp: a temporary register
10333 void MacroAssembler::lightweight_unlock(Register obj, Register reg_rax, Register thread, Register tmp, Label& slow) {

   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "precompiled.hpp"
   26 #include "asm/assembler.hpp"
   27 #include "asm/assembler.inline.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"
   31 #include "ci/ciInlineKlass.hpp"
   32 #include "crc32c.h"
   33 #include "gc/shared/barrierSet.hpp"
   34 #include "gc/shared/barrierSetAssembler.hpp"
   35 #include "gc/shared/collectedHeap.inline.hpp"
   36 #include "gc/shared/tlab_globals.hpp"
   37 #include "interpreter/bytecodeHistogram.hpp"
   38 #include "interpreter/interpreter.hpp"
   39 #include "jvm.h"
   40 #include "memory/resourceArea.hpp"
   41 #include "memory/universe.hpp"
   42 #include "oops/accessDecorators.hpp"
   43 #include "oops/compressedKlass.inline.hpp"
   44 #include "oops/compressedOops.inline.hpp"
   45 #include "oops/klass.inline.hpp"
   46 #include "oops/resolvedFieldEntry.hpp"
   47 #include "prims/methodHandles.hpp"
   48 #include "runtime/continuation.hpp"
   49 #include "runtime/interfaceSupport.inline.hpp"
   50 #include "runtime/javaThread.hpp"
   51 #include "runtime/jniHandles.hpp"
   52 #include "runtime/objectMonitor.hpp"
   53 #include "runtime/os.hpp"
   54 #include "runtime/safepoint.hpp"
   55 #include "runtime/safepointMechanism.hpp"
   56 #include "runtime/sharedRuntime.hpp"
   57 #include "runtime/signature_cc.hpp"
   58 #include "runtime/stubRoutines.hpp"
   59 #include "utilities/checkedCast.hpp"
   60 #include "utilities/macros.hpp"
   61 #include "vmreg_x86.inline.hpp"
   62 #ifdef COMPILER2
   63 #include "opto/output.hpp"
   64 #endif
   65 
   66 #ifdef PRODUCT
   67 #define BLOCK_COMMENT(str) /* nothing */
   68 #define STOP(error) stop(error)
   69 #else
   70 #define BLOCK_COMMENT(str) block_comment(str)
   71 #define STOP(error) block_comment(error); stop(error)
   72 #endif
   73 
   74 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   75 
   76 #ifdef ASSERT
   77 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   78 #endif
   79 
   80 static const Assembler::Condition reverse[] = {
   81     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   82     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   83     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   84     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,

 1708 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1709   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2));
 1710   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1711   pass_arg2(this, arg_2);
 1712   pass_arg1(this, arg_1);
 1713   pass_arg0(this, arg_0);
 1714   call_VM_leaf(entry_point, 3);
 1715 }
 1716 
 1717 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1718   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3));
 1719   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1720   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1721   pass_arg3(this, arg_3);
 1722   pass_arg2(this, arg_2);
 1723   pass_arg1(this, arg_1);
 1724   pass_arg0(this, arg_0);
 1725   call_VM_leaf(entry_point, 3);
 1726 }
 1727 
 1728 void MacroAssembler::super_call_VM_leaf(address entry_point) {
 1729   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1730 }
 1731 
 1732 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1733   pass_arg0(this, arg_0);
 1734   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1735 }
 1736 
 1737 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1738   LP64_ONLY(assert_different_registers(arg_0, c_rarg1));
 1739   pass_arg1(this, arg_1);
 1740   pass_arg0(this, arg_0);
 1741   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1742 }
 1743 
 1744 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1745   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2));
 1746   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1747   pass_arg2(this, arg_2);
 1748   pass_arg1(this, arg_1);
 1749   pass_arg0(this, arg_0);
 1750   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1751 }

 2896     lea(rscratch, src);
 2897     Assembler::mulss(dst, Address(rscratch, 0));
 2898   }
 2899 }
 2900 
 2901 void MacroAssembler::null_check(Register reg, int offset) {
 2902   if (needs_explicit_null_check(offset)) {
 2903     // provoke OS null exception if reg is null by
 2904     // accessing M[reg] w/o changing any (non-CC) registers
 2905     // NOTE: cmpl is plenty here to provoke a segv
 2906     cmpptr(rax, Address(reg, 0));
 2907     // Note: should probably use testl(rax, Address(reg, 0));
 2908     //       may be shorter code (however, this version of
 2909     //       testl needs to be implemented first)
 2910   } else {
 2911     // nothing to do, (later) access of M[reg + offset]
 2912     // will provoke OS null exception if reg is null
 2913   }
 2914 }
 2915 
 2916 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
 2917   andptr(markword, markWord::inline_type_mask_in_place);
 2918   cmpptr(markword, markWord::inline_type_pattern);
 2919   jcc(Assembler::equal, is_inline_type);
 2920 }
 2921 
 2922 void MacroAssembler::test_klass_is_inline_type(Register klass, Register temp_reg, Label& is_inline_type) {
 2923   movl(temp_reg, Address(klass, Klass::access_flags_offset()));
 2924   testl(temp_reg, JVM_ACC_IDENTITY);
 2925   jcc(Assembler::zero, is_inline_type);
 2926 }
 2927 
 2928 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type) {
 2929   testptr(object, object);
 2930   jcc(Assembler::zero, not_inline_type);
 2931   const int is_inline_type_mask = markWord::inline_type_pattern;
 2932   movptr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
 2933   andptr(tmp, is_inline_type_mask);
 2934   cmpptr(tmp, is_inline_type_mask);
 2935   jcc(Assembler::notEqual, not_inline_type);
 2936 }
 2937 
 2938 void MacroAssembler::test_klass_is_empty_inline_type(Register klass, Register temp_reg, Label& is_empty_inline_type) {
 2939 #ifdef ASSERT
 2940   {
 2941     Label done_check;
 2942     test_klass_is_inline_type(klass, temp_reg, done_check);
 2943     stop("test_klass_is_empty_inline_type with non inline type klass");
 2944     bind(done_check);
 2945   }
 2946 #endif
 2947   movl(temp_reg, Address(klass, InstanceKlass::misc_flags_offset()));
 2948   testl(temp_reg, InstanceKlassFlags::is_empty_inline_type_value());
 2949   jcc(Assembler::notZero, is_empty_inline_type);
 2950 }
 2951 
 2952 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
 2953   movl(temp_reg, flags);
 2954   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2955   jcc(Assembler::notEqual, is_null_free_inline_type);
 2956 }
 2957 
 2958 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
 2959   movl(temp_reg, flags);
 2960   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2961   jcc(Assembler::equal, not_null_free_inline_type);
 2962 }
 2963 
 2964 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
 2965   movl(temp_reg, flags);
 2966   testl(temp_reg, 1 << ResolvedFieldEntry::is_flat_shift);
 2967   jcc(Assembler::notEqual, is_flat);
 2968 }
 2969 
 2970 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
 2971   movl(temp_reg, flags);
 2972   testl(temp_reg, 1 << ResolvedFieldEntry::has_null_marker_shift);
 2973   jcc(Assembler::notEqual, has_null_marker);
 2974 }
 2975 
 2976 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
 2977   Label test_mark_word;
 2978   // load mark word
 2979   movptr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
 2980   // check displaced
 2981   testl(temp_reg, markWord::unlocked_value);
 2982   jccb(Assembler::notZero, test_mark_word);
 2983   // slow path use klass prototype
 2984   push(rscratch1);
 2985   load_prototype_header(temp_reg, oop, rscratch1);
 2986   pop(rscratch1);
 2987 
 2988   bind(test_mark_word);
 2989   testl(temp_reg, test_bit);
 2990   jcc((jmp_set) ? Assembler::notZero : Assembler::zero, jmp_label);
 2991 }
 2992 
 2993 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg,
 2994                                          Label& is_flat_array) {
 2995 #ifdef _LP64
 2996   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2997 #else
 2998   load_klass(temp_reg, oop, noreg);
 2999   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 3000   test_flat_array_layout(temp_reg, is_flat_array);
 3001 #endif
 3002 }
 3003 
 3004 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 3005                                              Label& is_non_flat_array) {
 3006 #ifdef _LP64
 3007   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 3008 #else
 3009   load_klass(temp_reg, oop, noreg);
 3010   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 3011   test_non_flat_array_layout(temp_reg, is_non_flat_array);
 3012 #endif
 3013 }
 3014 
 3015 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 3016 #ifdef _LP64
 3017   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 3018 #else
 3019   Unimplemented();
 3020 #endif
 3021 }
 3022 
 3023 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 3024 #ifdef _LP64
 3025   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 3026 #else
 3027   Unimplemented();
 3028 #endif
 3029 }
 3030 
 3031 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 3032   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 3033   jcc(Assembler::notZero, is_flat_array);
 3034 }
 3035 
 3036 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
 3037   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 3038   jcc(Assembler::zero, is_non_flat_array);
 3039 }
 3040 
 3041 void MacroAssembler::os_breakpoint() {
 3042   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 3043   // (e.g., MSVC can't call ps() otherwise)
 3044   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 3045 }
 3046 
 3047 void MacroAssembler::unimplemented(const char* what) {
 3048   const char* buf = nullptr;
 3049   {
 3050     ResourceMark rm;
 3051     stringStream ss;
 3052     ss.print("unimplemented: %s", what);
 3053     buf = code_string(ss.as_string());
 3054   }
 3055   stop(buf);
 3056 }
 3057 
 3058 #ifdef _LP64
 3059 #define XSTATE_BV 0x200
 3060 #endif

 4185 }
 4186 
 4187 // C++ bool manipulation
 4188 void MacroAssembler::testbool(Register dst) {
 4189   if(sizeof(bool) == 1)
 4190     testb(dst, 0xff);
 4191   else if(sizeof(bool) == 2) {
 4192     // testw implementation needed for two byte bools
 4193     ShouldNotReachHere();
 4194   } else if(sizeof(bool) == 4)
 4195     testl(dst, dst);
 4196   else
 4197     // unsupported
 4198     ShouldNotReachHere();
 4199 }
 4200 
 4201 void MacroAssembler::testptr(Register dst, Register src) {
 4202   LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
 4203 }
 4204 
 4205 // Object / value buffer allocation...
 4206 //
 4207 // Kills klass and rsi on LP64
 4208 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
 4209                                        Register t1, Register t2,
 4210                                        bool clear_fields, Label& alloc_failed)
 4211 {
 4212   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
 4213   Register layout_size = t1;
 4214   assert(new_obj == rax, "needs to be rax");
 4215   assert_different_registers(klass, new_obj, t1, t2);
 4216 
 4217   // get instance_size in InstanceKlass (scaled to a count of bytes)
 4218   movl(layout_size, Address(klass, Klass::layout_helper_offset()));
 4219   // test to see if it is malformed in some way
 4220   testl(layout_size, Klass::_lh_instance_slow_path_bit);
 4221   jcc(Assembler::notZero, slow_case_no_pop);
 4222 
 4223   // Allocate the instance:
 4224   //  If TLAB is enabled:
 4225   //    Try to allocate in the TLAB.
 4226   //    If fails, go to the slow path.
 4227   //  Else If inline contiguous allocations are enabled:
 4228   //    Try to allocate in eden.
 4229   //    If fails due to heap end, go to slow path.
 4230   //
 4231   //  If TLAB is enabled OR inline contiguous is enabled:
 4232   //    Initialize the allocation.
 4233   //    Exit.
 4234   //
 4235   //  Go to slow path.
 4236 
 4237   push(klass);
 4238   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(klass);
 4239 #ifndef _LP64
 4240   if (UseTLAB) {
 4241     get_thread(thread);
 4242   }
 4243 #endif // _LP64
 4244 
 4245   if (UseTLAB) {
 4246     tlab_allocate(thread, new_obj, layout_size, 0, klass, t2, slow_case);
 4247     if (ZeroTLAB || (!clear_fields)) {
 4248       // the fields have been already cleared
 4249       jmp(initialize_header);
 4250     } else {
 4251       // initialize both the header and fields
 4252       jmp(initialize_object);
 4253     }
 4254   } else {
 4255     jmp(slow_case);
 4256   }
 4257 
 4258   // If UseTLAB is true, the object is created above and there is an initialize need.
 4259   // Otherwise, skip and go to the slow path.
 4260   if (UseTLAB) {
 4261     if (clear_fields) {
 4262       // The object is initialized before the header.  If the object size is
 4263       // zero, go directly to the header initialization.
 4264       bind(initialize_object);
 4265       decrement(layout_size, sizeof(oopDesc));
 4266       jcc(Assembler::zero, initialize_header);
 4267 
 4268       // Initialize topmost object field, divide size by 8, check if odd and
 4269       // test if zero.
 4270       Register zero = klass;
 4271       xorl(zero, zero);    // use zero reg to clear memory (shorter code)
 4272       shrl(layout_size, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
 4273 
 4274   #ifdef ASSERT
 4275       // make sure instance_size was multiple of 8
 4276       Label L;
 4277       // Ignore partial flag stall after shrl() since it is debug VM
 4278       jcc(Assembler::carryClear, L);
 4279       stop("object size is not multiple of 2 - adjust this code");
 4280       bind(L);
 4281       // must be > 0, no extra check needed here
 4282   #endif
 4283 
 4284       // initialize remaining object fields: instance_size was a multiple of 8
 4285       {
 4286         Label loop;
 4287         bind(loop);
 4288         movptr(Address(new_obj, layout_size, Address::times_8, sizeof(oopDesc) - 1*oopSize), zero);
 4289         NOT_LP64(movptr(Address(new_obj, layout_size, Address::times_8, sizeof(oopDesc) - 2*oopSize), zero));
 4290         decrement(layout_size);
 4291         jcc(Assembler::notZero, loop);
 4292       }
 4293     } // clear_fields
 4294 
 4295     // initialize object header only.
 4296     bind(initialize_header);
 4297     pop(klass);
 4298     Register mark_word = t2;
 4299     movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 4300     movptr(Address(new_obj, oopDesc::mark_offset_in_bytes ()), mark_word);
 4301 #ifdef _LP64
 4302     xorl(rsi, rsi);                 // use zero reg to clear memory (shorter code)
 4303     store_klass_gap(new_obj, rsi);  // zero klass gap for compressed oops
 4304 #endif
 4305     movptr(t2, klass);         // preserve klass
 4306     store_klass(new_obj, t2, rscratch1);  // src klass reg is potentially compressed
 4307 
 4308     jmp(done);
 4309   }
 4310 
 4311   bind(slow_case);
 4312   pop(klass);
 4313   bind(slow_case_no_pop);
 4314   jmp(alloc_failed);
 4315 
 4316   bind(done);
 4317 }
 4318 
 4319 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 4320 void MacroAssembler::tlab_allocate(Register thread, Register obj,
 4321                                    Register var_size_in_bytes,
 4322                                    int con_size_in_bytes,
 4323                                    Register t1,
 4324                                    Register t2,
 4325                                    Label& slow_case) {
 4326   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 4327   bs->tlab_allocate(this, thread, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 4328 }
 4329 
 4330 RegSet MacroAssembler::call_clobbered_gp_registers() {
 4331   RegSet regs;
 4332 #ifdef _LP64
 4333   regs += RegSet::of(rax, rcx, rdx);
 4334 #ifndef WINDOWS
 4335   regs += RegSet::of(rsi, rdi);
 4336 #endif
 4337   regs += RegSet::range(r8, r11);
 4338 #else

 4557     // clear topmost word (no jump would be needed if conditional assignment worked here)
 4558     movptr(Address(address, index, Address::times_8, offset_in_bytes - 0*BytesPerWord), temp);
 4559     // index could be 0 now, must check again
 4560     jcc(Assembler::zero, done);
 4561     bind(even);
 4562   }
 4563 #endif // !_LP64
 4564   // initialize remaining object fields: index is a multiple of 2 now
 4565   {
 4566     Label loop;
 4567     bind(loop);
 4568     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 4569     NOT_LP64(movptr(Address(address, index, Address::times_8, offset_in_bytes - 2*BytesPerWord), temp);)
 4570     decrement(index);
 4571     jcc(Assembler::notZero, loop);
 4572   }
 4573 
 4574   bind(done);
 4575 }
 4576 
 4577 void MacroAssembler::get_inline_type_field_klass(Register holder_klass, Register index, Register inline_klass) {
 4578   inline_layout_info(holder_klass, index, inline_klass);
 4579   movptr(inline_klass, Address(inline_klass, InlineLayoutInfo::klass_offset()));
 4580 }
 4581 
 4582 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 4583   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 4584 #ifdef ASSERT
 4585   {
 4586     Label done;
 4587     cmpptr(layout_info, 0);
 4588     jcc(Assembler::notEqual, done);
 4589     stop("inline_layout_info_array is null");
 4590     bind(done);
 4591   }
 4592 #endif
 4593 
 4594   InlineLayoutInfo array[2];
 4595   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 4596   if (is_power_of_2(size)) {
 4597     shll(index, log2i_exact(size)); // Scale index by power of 2
 4598   } else {
 4599     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 4600   }
 4601   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 4602 }
 4603 
 4604 void MacroAssembler::get_default_value_oop(Register inline_klass, Register temp_reg, Register obj) {
 4605 #ifdef ASSERT
 4606   {
 4607     Label done_check;
 4608     test_klass_is_inline_type(inline_klass, temp_reg, done_check);
 4609     stop("get_default_value_oop from non inline type klass");
 4610     bind(done_check);
 4611   }
 4612 #endif
 4613   Register offset = temp_reg;
 4614   // Getting the offset of the pre-allocated default value
 4615   movptr(offset, Address(inline_klass, in_bytes(InstanceKlass::adr_inlineklass_fixed_block_offset())));
 4616   movl(offset, Address(offset, in_bytes(InlineKlass::default_value_offset_offset())));
 4617 
 4618   // Getting the mirror
 4619   movptr(obj, Address(inline_klass, in_bytes(Klass::java_mirror_offset())));
 4620   resolve_oop_handle(obj, inline_klass);
 4621 
 4622   // Getting the pre-allocated default value from the mirror
 4623   Address field(obj, offset, Address::times_1);
 4624   load_heap_oop(obj, field);
 4625 }
 4626 
 4627 void MacroAssembler::get_empty_inline_type_oop(Register inline_klass, Register temp_reg, Register obj) {
 4628 #ifdef ASSERT
 4629   {
 4630     Label done_check;
 4631     test_klass_is_empty_inline_type(inline_klass, temp_reg, done_check);
 4632     stop("get_empty_value from non-empty inline klass");
 4633     bind(done_check);
 4634   }
 4635 #endif
 4636   get_default_value_oop(inline_klass, temp_reg, obj);
 4637 }
 4638 
 4639 
 4640 // Look up the method for a megamorphic invokeinterface call.
 4641 // The target method is determined by <intf_klass, itable_index>.
 4642 // The receiver klass is in recv_klass.
 4643 // On success, the result will be in method_result, and execution falls through.
 4644 // On failure, execution transfers to the given label.
 4645 void MacroAssembler::lookup_interface_method(Register recv_klass,
 4646                                              Register intf_klass,
 4647                                              RegisterOrConstant itable_index,
 4648                                              Register method_result,
 4649                                              Register scan_temp,
 4650                                              Label& L_no_such_interface,
 4651                                              bool return_method) {
 4652   assert_different_registers(recv_klass, intf_klass, scan_temp);
 4653   assert_different_registers(method_result, intf_klass, scan_temp);
 4654   assert(recv_klass != method_result || !return_method,
 4655          "recv_klass can be destroyed when method isn't needed");
 4656 
 4657   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 4658          "caller must use same register for non-constant itable index as for method");
 4659 

 5421   } else {
 5422     Label L;
 5423     jccb(negate_condition(cc), L);
 5424     movl(dst, src);
 5425     bind(L);
 5426   }
 5427 }
 5428 
 5429 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 5430   if (VM_Version::supports_cmov()) {
 5431     cmovl(cc, dst, src);
 5432   } else {
 5433     Label L;
 5434     jccb(negate_condition(cc), L);
 5435     movl(dst, src);
 5436     bind(L);
 5437   }
 5438 }
 5439 
 5440 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 5441   if (!VerifyOops || VerifyAdapterSharing) {
 5442     // Below address of the code string confuses VerifyAdapterSharing
 5443     // because it may differ between otherwise equivalent adapters.
 5444     return;
 5445   }
 5446 
 5447   BLOCK_COMMENT("verify_oop {");
 5448 #ifdef _LP64
 5449   push(rscratch1);
 5450 #endif
 5451   push(rax);                          // save rax
 5452   push(reg);                          // pass register argument
 5453 
 5454   // Pass register number to verify_oop_subroutine
 5455   const char* b = nullptr;
 5456   {
 5457     ResourceMark rm;
 5458     stringStream ss;
 5459     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 5460     b = code_string(ss.as_string());
 5461   }
 5462   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5463   pushptr(buffer.addr(), rscratch1);
 5464 
 5465   // call indirectly to solve generation ordering problem

 5487   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 5488   int stackElementSize = Interpreter::stackElementSize;
 5489   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 5490 #ifdef ASSERT
 5491   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 5492   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 5493 #endif
 5494   Register             scale_reg    = noreg;
 5495   Address::ScaleFactor scale_factor = Address::no_scale;
 5496   if (arg_slot.is_constant()) {
 5497     offset += arg_slot.as_constant() * stackElementSize;
 5498   } else {
 5499     scale_reg    = arg_slot.as_register();
 5500     scale_factor = Address::times(stackElementSize);
 5501   }
 5502   offset += wordSize;           // return PC is on stack
 5503   return Address(rsp, scale_reg, scale_factor, offset);
 5504 }
 5505 
 5506 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5507   if (!VerifyOops || VerifyAdapterSharing) {
 5508     // Below address of the code string confuses VerifyAdapterSharing
 5509     // because it may differ between otherwise equivalent adapters.
 5510     return;
 5511   }
 5512 
 5513 #ifdef _LP64
 5514   push(rscratch1);
 5515 #endif
 5516   push(rax); // save rax,
 5517   // addr may contain rsp so we will have to adjust it based on the push
 5518   // we just did (and on 64 bit we do two pushes)
 5519   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5520   // stores rax into addr which is backwards of what was intended.
 5521   if (addr.uses(rsp)) {
 5522     lea(rax, addr);
 5523     pushptr(Address(rax, LP64_ONLY(2 *) BytesPerWord));
 5524   } else {
 5525     pushptr(addr);
 5526   }
 5527 
 5528   // Pass register number to verify_oop_subroutine
 5529   const char* b = nullptr;
 5530   {
 5531     ResourceMark rm;

 5978 
 5979 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5980   // get mirror
 5981   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5982   load_method_holder(mirror, method);
 5983   movptr(mirror, Address(mirror, mirror_offset));
 5984   resolve_oop_handle(mirror, tmp);
 5985 }
 5986 
 5987 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5988   load_method_holder(rresult, rmethod);
 5989   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5990 }
 5991 
 5992 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5993   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5994   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5995   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5996 }
 5997 
 5998 void MacroAssembler::load_metadata(Register dst, Register src) {
 5999   if (UseCompressedClassPointers) {
 6000     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 6001   } else {
 6002     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 6003   }
 6004 }
 6005 
 6006 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 6007   assert_different_registers(src, tmp);
 6008   assert_different_registers(dst, tmp);
 6009 #ifdef _LP64
 6010   if (UseCompressedClassPointers) {
 6011     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 6012     decode_klass_not_null(dst, tmp);
 6013   } else
 6014 #endif
 6015   movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 6016 }
 6017 
 6018 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 6019   load_klass(dst, src, tmp);
 6020   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 6021 }
 6022 
 6023 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 6024   assert_different_registers(src, tmp);
 6025   assert_different_registers(dst, tmp);
 6026 #ifdef _LP64
 6027   if (UseCompressedClassPointers) {
 6028     encode_klass_not_null(src, tmp);
 6029     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 6030   } else
 6031 #endif
 6032     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 6033 }
 6034 
 6035 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 6036                                     Register tmp1, Register thread_tmp) {
 6037   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 6038   decorators = AccessInternal::decorator_fixup(decorators, type);
 6039   bool as_raw = (decorators & AS_RAW) != 0;
 6040   if (as_raw) {
 6041     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
 6042   } else {
 6043     bs->load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
 6044   }
 6045 }
 6046 
 6047 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 6048                                      Register tmp1, Register tmp2, Register tmp3) {
 6049   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 6050   decorators = AccessInternal::decorator_fixup(decorators, type);
 6051   bool as_raw = (decorators & AS_RAW) != 0;
 6052   if (as_raw) {
 6053     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 6054   } else {
 6055     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 6056   }
 6057 }
 6058 
 6059 void MacroAssembler::access_value_copy(DecoratorSet decorators, Register src, Register dst,
 6060                                        Register inline_klass) {
 6061   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 6062   bs->value_copy(this, decorators, src, dst, inline_klass);
 6063 }
 6064 
 6065 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 6066                                      Register inline_layout_info) {
 6067   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 6068   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 6069 }
 6070 
 6071 void MacroAssembler::first_field_offset(Register inline_klass, Register offset) {
 6072   movptr(offset, Address(inline_klass, InstanceKlass::adr_inlineklass_fixed_block_offset()));
 6073   movl(offset, Address(offset, InlineKlass::first_field_offset_offset()));
 6074 }
 6075 
 6076 void MacroAssembler::data_for_oop(Register oop, Register data, Register inline_klass) {
 6077   // ((address) (void*) o) + vk->first_field_offset();
 6078   Register offset = (data == oop) ? rscratch1 : data;
 6079   first_field_offset(inline_klass, offset);
 6080   if (data == oop) {
 6081     addptr(data, offset);
 6082   } else {
 6083     lea(data, Address(oop, offset));
 6084   }
 6085 }
 6086 
 6087 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
 6088                                                 Register index, Register data) {
 6089   assert(index != rcx, "index needs to shift by rcx");
 6090   assert_different_registers(array, array_klass, index);
 6091   assert_different_registers(rcx, array, index);
 6092 
 6093   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
 6094   movl(rcx, Address(array_klass, Klass::layout_helper_offset()));
 6095 
 6096   // Klass::layout_helper_log2_element_size(lh)
 6097   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
 6098   shrl(rcx, Klass::_lh_log2_element_size_shift);
 6099   andl(rcx, Klass::_lh_log2_element_size_mask);
 6100   shlptr(index); // index << rcx
 6101 
 6102   lea(data, Address(array, index, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_PRIMITIVE_OBJECT)));
 6103 }
 6104 
 6105 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
 6106                                    Register thread_tmp, DecoratorSet decorators) {
 6107   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, thread_tmp);
 6108 }
 6109 
 6110 // Doesn't do verification, generates fixed size code
 6111 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 6112                                             Register thread_tmp, DecoratorSet decorators) {
 6113   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, thread_tmp);
 6114 }
 6115 
 6116 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 6117                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 6118   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 6119 }
 6120 
 6121 // Used for storing nulls.
 6122 void MacroAssembler::store_heap_oop_null(Address dst) {
 6123   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 6124 }

 6424 
 6425 void MacroAssembler::reinit_heapbase() {
 6426   if (UseCompressedOops) {
 6427     if (Universe::heap() != nullptr) {
 6428       if (CompressedOops::base() == nullptr) {
 6429         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6430       } else {
 6431         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6432       }
 6433     } else {
 6434       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6435     }
 6436   }
 6437 }
 6438 
 6439 #endif // _LP64
 6440 
 6441 #if COMPILER2_OR_JVMCI
 6442 
 6443 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6444 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6445   // cnt - number of qwords (8-byte words).
 6446   // base - start address, qword aligned.
 6447   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6448   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 6449   if (use64byteVector) {
 6450     evpbroadcastq(xtmp, val, AVX_512bit);
 6451   } else if (MaxVectorSize >= 32) {
 6452     movdq(xtmp, val);
 6453     punpcklqdq(xtmp, xtmp);
 6454     vinserti128_high(xtmp, xtmp);
 6455   } else {
 6456     movdq(xtmp, val);
 6457     punpcklqdq(xtmp, xtmp);
 6458   }
 6459   jmp(L_zero_64_bytes);
 6460 
 6461   BIND(L_loop);
 6462   if (MaxVectorSize >= 32) {
 6463     fill64(base, 0, xtmp, use64byteVector);
 6464   } else {
 6465     movdqu(Address(base,  0), xtmp);
 6466     movdqu(Address(base, 16), xtmp);
 6467     movdqu(Address(base, 32), xtmp);
 6468     movdqu(Address(base, 48), xtmp);
 6469   }
 6470   addptr(base, 64);
 6471 
 6472   BIND(L_zero_64_bytes);
 6473   subptr(cnt, 8);
 6474   jccb(Assembler::greaterEqual, L_loop);
 6475 
 6476   // Copy trailing 64 bytes
 6477   if (use64byteVector) {
 6478     addptr(cnt, 8);
 6479     jccb(Assembler::equal, L_end);
 6480     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6481     jmp(L_end);
 6482   } else {
 6483     addptr(cnt, 4);
 6484     jccb(Assembler::less, L_tail);
 6485     if (MaxVectorSize >= 32) {
 6486       vmovdqu(Address(base, 0), xtmp);
 6487     } else {
 6488       movdqu(Address(base,  0), xtmp);
 6489       movdqu(Address(base, 16), xtmp);
 6490     }
 6491   }
 6492   addptr(base, 32);
 6493   subptr(cnt, 4);
 6494 
 6495   BIND(L_tail);
 6496   addptr(cnt, 4);
 6497   jccb(Assembler::lessEqual, L_end);
 6498   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6499     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6500   } else {
 6501     decrement(cnt);
 6502 
 6503     BIND(L_sloop);
 6504     movq(Address(base, 0), xtmp);
 6505     addptr(base, 8);
 6506     decrement(cnt);
 6507     jccb(Assembler::greaterEqual, L_sloop);
 6508   }
 6509   BIND(L_end);
 6510 }
 6511 
 6512 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6513   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6514   // An inline type might be returned. If fields are in registers we
 6515   // need to allocate an inline type instance and initialize it with
 6516   // the value of the fields.
 6517   Label skip;
 6518   // We only need a new buffered inline type if a new one is not returned
 6519   testptr(rax, 1);
 6520   jcc(Assembler::zero, skip);
 6521   int call_offset = -1;
 6522 
 6523 #ifdef _LP64
 6524   // The following code is similar to allocate_instance but has some slight differences,
 6525   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6526   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
 6527   Label slow_case;
 6528   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6529   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6530   if (vk != nullptr) {
 6531     // Called from C1, where the return type is statically known.
 6532     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6533     jint lh = vk->layout_helper();
 6534     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6535     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6536       tlab_allocate(r15_thread, rax, noreg, lh, r13, r14, slow_case);
 6537     } else {
 6538       jmp(slow_case);
 6539     }
 6540   } else {
 6541     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6542     mov(rbx, rax);
 6543     andptr(rbx, -2);
 6544     if (UseTLAB) {
 6545       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6546       testl(r14, Klass::_lh_instance_slow_path_bit);
 6547       jcc(Assembler::notZero, slow_case);
 6548       tlab_allocate(r15_thread, rax, r14, 0, r13, r14, slow_case);
 6549     } else {
 6550       jmp(slow_case);
 6551     }
 6552   }
 6553   if (UseTLAB) {
 6554     // 2. Initialize buffered inline instance header
 6555     Register buffer_obj = rax;
 6556     movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6557     xorl(r13, r13);
 6558     store_klass_gap(buffer_obj, r13);
 6559     if (vk == nullptr) {
 6560       // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6561       mov(r13, rbx);
 6562     }
 6563     store_klass(buffer_obj, rbx, rscratch1);
 6564     // 3. Initialize its fields with an inline class specific handler
 6565     if (vk != nullptr) {
 6566       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6567     } else {
 6568       movptr(rbx, Address(r13, InstanceKlass::adr_inlineklass_fixed_block_offset()));
 6569       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6570       call(rbx);
 6571     }
 6572     jmp(skip);
 6573   }
 6574   bind(slow_case);
 6575   // We failed to allocate a new inline type, fall back to a runtime
 6576   // call. Some oop field may be live in some registers but we can't
 6577   // tell. That runtime call will take care of preserving them
 6578   // across a GC if there's one.
 6579   mov(rax, rscratch1);
 6580 #endif
 6581 
 6582   if (from_interpreter) {
 6583     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6584   } else {
 6585     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6586     call_offset = offset();
 6587   }
 6588 
 6589   bind(skip);
 6590   return call_offset;
 6591 }
 6592 
 6593 // Move a value between registers/stack slots and update the reg_state
 6594 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6595   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6596   if (reg_state[to->value()] == reg_written) {
 6597     return true; // Already written
 6598   }
 6599   if (from != to && bt != T_VOID) {
 6600     if (reg_state[to->value()] == reg_readonly) {
 6601       return false; // Not yet writable
 6602     }
 6603     if (from->is_reg()) {
 6604       if (to->is_reg()) {
 6605         if (from->is_XMMRegister()) {
 6606           if (bt == T_DOUBLE) {
 6607             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6608           } else {
 6609             assert(bt == T_FLOAT, "must be float");
 6610             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6611           }
 6612         } else {
 6613           movq(to->as_Register(), from->as_Register());
 6614         }
 6615       } else {
 6616         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6617         Address to_addr = Address(rsp, st_off);
 6618         if (from->is_XMMRegister()) {
 6619           if (bt == T_DOUBLE) {
 6620             movdbl(to_addr, from->as_XMMRegister());
 6621           } else {
 6622             assert(bt == T_FLOAT, "must be float");
 6623             movflt(to_addr, from->as_XMMRegister());
 6624           }
 6625         } else {
 6626           movq(to_addr, from->as_Register());
 6627         }
 6628       }
 6629     } else {
 6630       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6631       if (to->is_reg()) {
 6632         if (to->is_XMMRegister()) {
 6633           if (bt == T_DOUBLE) {
 6634             movdbl(to->as_XMMRegister(), from_addr);
 6635           } else {
 6636             assert(bt == T_FLOAT, "must be float");
 6637             movflt(to->as_XMMRegister(), from_addr);
 6638           }
 6639         } else {
 6640           movq(to->as_Register(), from_addr);
 6641         }
 6642       } else {
 6643         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6644         movq(r13, from_addr);
 6645         movq(Address(rsp, st_off), r13);
 6646       }
 6647     }
 6648   }
 6649   // Update register states
 6650   reg_state[from->value()] = reg_writable;
 6651   reg_state[to->value()] = reg_written;
 6652   return true;
 6653 }
 6654 
 6655 // Calculate the extra stack space required for packing or unpacking inline
 6656 // args and adjust the stack pointer
 6657 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6658   // Two additional slots to account for return address
 6659   int sp_inc = (args_on_stack + 2) * VMRegImpl::stack_slot_size;
 6660   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6661   // Save the return address, adjust the stack (make sure it is properly
 6662   // 16-byte aligned) and copy the return address to the new top of the stack.
 6663   // The stack will be repaired on return (see MacroAssembler::remove_frame).
 6664   assert(sp_inc > 0, "sanity");
 6665   pop(r13);
 6666   subptr(rsp, sp_inc);
 6667   push(r13);
 6668   return sp_inc;
 6669 }
 6670 
 6671 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6672 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6673                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6674                                           RegState reg_state[]) {
 6675   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6676   assert(from->is_valid(), "source must be valid");
 6677   bool progress = false;
 6678 #ifdef ASSERT
 6679   const int start_offset = offset();
 6680 #endif
 6681 
 6682   Label L_null, L_notNull;
 6683   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6684   Register tmp1 = r10;
 6685   Register tmp2 = r13;
 6686   Register fromReg = noreg;
 6687   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, -1);
 6688   bool done = true;
 6689   bool mark_done = true;
 6690   VMReg toReg;
 6691   BasicType bt;
 6692   // Check if argument requires a null check
 6693   bool null_check = false;
 6694   VMReg nullCheckReg;
 6695   while (stream.next(nullCheckReg, bt)) {
 6696     if (sig->at(stream.sig_index())._offset == -1) {
 6697       null_check = true;
 6698       break;
 6699     }
 6700   }
 6701   stream.reset(sig_index, to_index);
 6702   while (stream.next(toReg, bt)) {
 6703     assert(toReg->is_valid(), "destination must be valid");
 6704     int idx = (int)toReg->value();
 6705     if (reg_state[idx] == reg_readonly) {
 6706       if (idx != from->value()) {
 6707         mark_done = false;
 6708       }
 6709       done = false;
 6710       continue;
 6711     } else if (reg_state[idx] == reg_written) {
 6712       continue;
 6713     }
 6714     assert(reg_state[idx] == reg_writable, "must be writable");
 6715     reg_state[idx] = reg_written;
 6716     progress = true;
 6717 
 6718     if (fromReg == noreg) {
 6719       if (from->is_reg()) {
 6720         fromReg = from->as_Register();
 6721       } else {
 6722         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6723         movq(tmp1, Address(rsp, st_off));
 6724         fromReg = tmp1;
 6725       }
 6726       if (null_check) {
 6727         // Nullable inline type argument, emit null check
 6728         testptr(fromReg, fromReg);
 6729         jcc(Assembler::zero, L_null);
 6730       }
 6731     }
 6732     int off = sig->at(stream.sig_index())._offset;
 6733     if (off == -1) {
 6734       assert(null_check, "Missing null check at");
 6735       if (toReg->is_stack()) {
 6736         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6737         movq(Address(rsp, st_off), 1);
 6738       } else {
 6739         movq(toReg->as_Register(), 1);
 6740       }
 6741       continue;
 6742     }
 6743     assert(off > 0, "offset in object should be positive");
 6744     Address fromAddr = Address(fromReg, off);
 6745     if (!toReg->is_XMMRegister()) {
 6746       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6747       if (is_reference_type(bt)) {
 6748         load_heap_oop(dst, fromAddr);
 6749       } else {
 6750         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6751         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6752       }
 6753       if (toReg->is_stack()) {
 6754         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6755         movq(Address(rsp, st_off), dst);
 6756       }
 6757     } else if (bt == T_DOUBLE) {
 6758       movdbl(toReg->as_XMMRegister(), fromAddr);
 6759     } else {
 6760       assert(bt == T_FLOAT, "must be float");
 6761       movflt(toReg->as_XMMRegister(), fromAddr);
 6762     }
 6763   }
 6764   if (progress && null_check) {
 6765     if (done) {
 6766       jmp(L_notNull);
 6767       bind(L_null);
 6768       // Set IsInit field to zero to signal that the argument is null.
 6769       // Also set all oop fields to zero to make the GC happy.
 6770       stream.reset(sig_index, to_index);
 6771       while (stream.next(toReg, bt)) {
 6772         if (sig->at(stream.sig_index())._offset == -1 ||
 6773             bt == T_OBJECT || bt == T_ARRAY) {
 6774           if (toReg->is_stack()) {
 6775             int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6776             movq(Address(rsp, st_off), 0);
 6777           } else {
 6778             xorq(toReg->as_Register(), toReg->as_Register());
 6779           }
 6780         }
 6781       }
 6782       bind(L_notNull);
 6783     } else {
 6784       bind(L_null);
 6785     }
 6786   }
 6787 
 6788   sig_index = stream.sig_index();
 6789   to_index = stream.regs_index();
 6790 
 6791   if (mark_done && reg_state[from->value()] != reg_written) {
 6792     // This is okay because no one else will write to that slot
 6793     reg_state[from->value()] = reg_writable;
 6794   }
 6795   from_index--;
 6796   assert(progress || (start_offset == offset()), "should not emit code");
 6797   return done;
 6798 }
 6799 
 6800 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6801                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6802                                         RegState reg_state[], Register val_array) {
 6803   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6804   assert(to->is_valid(), "destination must be valid");
 6805 
 6806   if (reg_state[to->value()] == reg_written) {
 6807     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6808     return true; // Already written
 6809   }
 6810 
 6811   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
 6812   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6813   Register val_obj_tmp = r11;
 6814   Register from_reg_tmp = r14;
 6815   Register tmp1 = r10;
 6816   Register tmp2 = r13;
 6817   Register tmp3 = rbx;
 6818   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6819 
 6820   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6821 
 6822   if (reg_state[to->value()] == reg_readonly) {
 6823     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6824       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6825       return false; // Not yet writable
 6826     }
 6827     val_obj = val_obj_tmp;
 6828   }
 6829 
 6830   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6831   load_heap_oop(val_obj, Address(val_array, index));
 6832 
 6833   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6834   VMReg fromReg;
 6835   BasicType bt;
 6836   Label L_null;
 6837   while (stream.next(fromReg, bt)) {
 6838     assert(fromReg->is_valid(), "source must be valid");
 6839     reg_state[fromReg->value()] = reg_writable;
 6840 
 6841     int off = sig->at(stream.sig_index())._offset;
 6842     if (off == -1) {
 6843       // Nullable inline type argument, emit null check
 6844       Label L_notNull;
 6845       if (fromReg->is_stack()) {
 6846         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6847         testb(Address(rsp, ld_off), 1);
 6848       } else {
 6849         testb(fromReg->as_Register(), 1);
 6850       }
 6851       jcc(Assembler::notZero, L_notNull);
 6852       movptr(val_obj, 0);
 6853       jmp(L_null);
 6854       bind(L_notNull);
 6855       continue;
 6856     }
 6857 
 6858     assert(off > 0, "offset in object should be positive");
 6859     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6860 
 6861     Address dst(val_obj, off);
 6862     if (!fromReg->is_XMMRegister()) {
 6863       Register src;
 6864       if (fromReg->is_stack()) {
 6865         src = from_reg_tmp;
 6866         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6867         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6868       } else {
 6869         src = fromReg->as_Register();
 6870       }
 6871       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6872       if (is_reference_type(bt)) {
 6873         store_heap_oop(dst, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6874       } else {
 6875         store_sized_value(dst, src, size_in_bytes);
 6876       }
 6877     } else if (bt == T_DOUBLE) {
 6878       movdbl(dst, fromReg->as_XMMRegister());
 6879     } else {
 6880       assert(bt == T_FLOAT, "must be float");
 6881       movflt(dst, fromReg->as_XMMRegister());
 6882     }
 6883   }
 6884   bind(L_null);
 6885   sig_index = stream.sig_index();
 6886   from_index = stream.regs_index();
 6887 
 6888   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6889   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6890   assert(success, "to register must be writeable");
 6891   return true;
 6892 }
 6893 
 6894 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6895   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6896 }
 6897 
 6898 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6899   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6900   if (needs_stack_repair) {
 6901     movq(rbp, Address(rsp, initial_framesize));
 6902     // The stack increment resides just below the saved rbp
 6903     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6904   } else {
 6905     if (initial_framesize > 0) {
 6906       addq(rsp, initial_framesize);
 6907     }
 6908     pop(rbp);
 6909   }
 6910 }
 6911 
 6912 // Clearing constant sized memory using YMM/ZMM registers.
 6913 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6914   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6915   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 6916 
 6917   int vector64_count = (cnt & (~0x7)) >> 3;
 6918   cnt = cnt & 0x7;
 6919   const int fill64_per_loop = 4;
 6920   const int max_unrolled_fill64 = 8;
 6921 
 6922   // 64 byte initialization loop.
 6923   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6924   int start64 = 0;
 6925   if (vector64_count > max_unrolled_fill64) {
 6926     Label LOOP;
 6927     Register index = rtmp;
 6928 
 6929     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6930 
 6931     movl(index, 0);

 6981         break;
 6982       case 7:
 6983         if (use64byteVector) {
 6984           movl(rtmp, 0x7F);
 6985           kmovwl(mask, rtmp);
 6986           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6987         } else {
 6988           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6989           movl(rtmp, 0x7);
 6990           kmovwl(mask, rtmp);
 6991           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6992         }
 6993         break;
 6994       default:
 6995         fatal("Unexpected length : %d\n",cnt);
 6996         break;
 6997     }
 6998   }
 6999 }
 7000 
 7001 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 7002                                bool is_large, bool word_copy_only, KRegister mask) {
 7003   // cnt      - number of qwords (8-byte words).
 7004   // base     - start address, qword aligned.
 7005   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 7006   assert(base==rdi, "base register must be edi for rep stos");
 7007   assert(val==rax,   "val register must be eax for rep stos");
 7008   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 7009   assert(InitArrayShortSize % BytesPerLong == 0,
 7010     "InitArrayShortSize should be the multiple of BytesPerLong");
 7011 
 7012   Label DONE;



 7013 
 7014   if (!is_large) {
 7015     Label LOOP, LONG;
 7016     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 7017     jccb(Assembler::greater, LONG);
 7018 
 7019     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
 7020 
 7021     decrement(cnt);
 7022     jccb(Assembler::negative, DONE); // Zero length
 7023 
 7024     // Use individual pointer-sized stores for small counts:
 7025     BIND(LOOP);
 7026     movptr(Address(base, cnt, Address::times_ptr), val);
 7027     decrement(cnt);
 7028     jccb(Assembler::greaterEqual, LOOP);
 7029     jmpb(DONE);
 7030 
 7031     BIND(LONG);
 7032   }
 7033 
 7034   // Use longer rep-prefixed ops for non-small counts:
 7035   if (UseFastStosb && !word_copy_only) {
 7036     shlptr(cnt, 3); // convert to number of bytes
 7037     rep_stosb();
 7038   } else if (UseXMMForObjInit) {
 7039     xmm_clear_mem(base, cnt, val, xtmp, mask);
 7040   } else {
 7041     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
 7042     rep_stos();
 7043   }
 7044 
 7045   BIND(DONE);
 7046 }
 7047 
 7048 #endif //COMPILER2_OR_JVMCI
 7049 
 7050 
 7051 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 7052                                    Register to, Register value, Register count,
 7053                                    Register rtmp, XMMRegister xtmp) {
 7054   ShortBranchVerifier sbv(this);
 7055   assert_different_registers(to, value, count, rtmp);
 7056   Label L_exit;
 7057   Label L_fill_2_bytes, L_fill_4_bytes;
 7058 
 7059 #if defined(COMPILER2) && defined(_LP64)

11074 
11075   // Load top.
11076   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
11077 
11078   // Check if the lock-stack is full.
11079   cmpl(top, LockStack::end_offset());
11080   jcc(Assembler::greaterEqual, slow);
11081 
11082   // Check for recursion.
11083   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
11084   jcc(Assembler::equal, push);
11085 
11086   // Check header for monitor (0b10).
11087   testptr(reg_rax, markWord::monitor_value);
11088   jcc(Assembler::notZero, slow);
11089 
11090   // Try to lock. Transition lock bits 0b01 => 0b00
11091   movptr(tmp, reg_rax);
11092   andptr(tmp, ~(int32_t)markWord::unlocked_value);
11093   orptr(reg_rax, markWord::unlocked_value);
11094   if (EnableValhalla) {
11095     // Mask inline_type bit such that we go to the slow path if object is an inline type
11096     andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
11097   }
11098   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
11099   jcc(Assembler::notEqual, slow);
11100 
11101   // Restore top, CAS clobbers register.
11102   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
11103 
11104   bind(push);
11105   // After successful lock, push object on lock-stack.
11106   movptr(Address(thread, top), obj);
11107   incrementl(top, oopSize);
11108   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
11109 }
11110 
11111 // Implements lightweight-unlocking.
11112 //
11113 // obj: the object to be unlocked
11114 // reg_rax: rax
11115 // thread: the thread
11116 // tmp: a temporary register
11117 void MacroAssembler::lightweight_unlock(Register obj, Register reg_rax, Register thread, Register tmp, Label& slow) {
< prev index next >